After a nine year hiatus, a new version of the X Input Protocol is out. Credit for the work goes to Povilas Kanapickas, who also implemented support for XI 2.4 in the various pieces of the stack [0]. So let's have a look.

X has had touch events since XI 2.2 (2012) but those were only really useful for direct touch devices (read: touchscreens). There were accommodations for indirect touch devices like touchpads but they were never used. The synaptics driver set the required bits for a while but it was dropped in 2015 because ... it was complicated to make use of and no-one seemed to actually use it anyway. Meanwhile, the rest of the world moved on and touchpad gestures are now prevalent. They've been standard in MacOS for ages, in Windows for almost ages and - with recent GNOME releases - now feature prominently on the Linux desktop as well. They have been part of libinput and the Wayland protocol for years (and even recently gained a new set of "hold" gestures). Meanwhile, X was left behind in the dust or mud, depending on your local climate.

XI 2.4 fixes this, it adds pinch and swipe gestures to the XI2 protocol and makes those available to supporting clients [2]. Notably here is that the interpretation of gestures is left to the driver [1]. The server takes the gestures and does the required state handling but otherwise has no decision into what constitutes a gesture. This is of course no different to e.g. 2-finger scrolling on a touchpad where the server just receives scroll events and passes them on accordingly.

XI 2.4 gesture events are quite similar to touch events in that they are processed as a sequence of begin/update/end with both types having their own event types. So the events you will receive are e.g. XIGesturePinchBegin or XIGestureSwipeUpdate. As with touch events, a client must select for all three (begin/update/end) on a window. Only one gesture can exist at any time, so if you are a multi-tasking octopus prepare to be disappointed.

Because gestures are tied to an indirect-touch device, the location they apply at is wherever the cursor is currently positioned. In that, they work similar to button presses, and passive grabs apply as expected too. So long-term the window manager will likely want a passive grab on the root window for swipe gestures while applications will implement pinch-to-zoom as you'd expect.

In terms of API there are no suprises. libXi 1.8 is the version to implement the new features and there we have a new XIGestureClassInfo returned by XIQueryDevice and of course the two events: XIGesturePinchEvent and XIGestureSwipeEvent. Grabbing is done via e.g. XIGrabSwipeGestureBegin, so for those of you with XI2 experience this will all look familiar. For those of you without - it's probably no longer worth investing time into becoming an XI2 expert.

Overall, it's a nice addition to the protocol and it will help getting the X server slightly closer to Wayland for a widely-used feature. Once GTK, mutter and all the other pieces in the stack are in place, it will just work for any (GTK) application that supports gestures under Wayland already. The same will be true for Qt I expect.

X server 21.1 will be out in a few weeks, xf86-input-libinput 1.2.0 is already out and so are xorgproto 2021.5 and libXi 1.8.

[0] In addition to taking on the Xorg release, so clearly there are no limits here
[1] More specifically: it's done by libinput since neither xf86-input-evdev nor xf86-input-synaptics will ever see gestures being implemented
[2] Hold gestures missed out on the various deadlines

Xorg is about to released.

And it's a release without Xwayland.

Heads up: if you are familiar with X, the below is simplified to the point it hurts. Sorry about that, but as an X developer you're probably good at coping with pain.

Let's go back to the 1980s, when fashion was weird and there were still reasons to be optimistic about the future. Because this is a thought exercise, we go back with full hindsight 20/20 vision and, ideally, the winning Lotto numbers in case we have some time for some self-indulgence.

If we were to implement an X server from scratch, we'd come away with a set of components. libxprotocol that handles the actual protocol wire format parsing and provides a C api to access that (quite like libxcb, actually). That one will just be the protocol-to-code conversion layer.

We'd have a libxserver component which handles all the state management required for an X server to actually behave like an X server (nothing in the X protocol require an X server to display anything). That library has a few entry points for abstract input events (pointer and keyboard, because this is the 80s after all) and a few exit points for rendered output.

libxserver uses libxprotocol but that's an implementation detail, we can ignore the protocol for the rest of the post.

Let's create a github organisation and host those two libraries. We now have: http://github.com/x/libxserver and http://github.com/x/libxprotocol [1].

Now, to actually implement a working functional X server, our new project would link against libxserver hook into this library's API points. For input, you'd use libinput and pass those events through, for output you'd use the modesetting driver that knows how to scream at the hardware until something finally shows up. This is somewhere between outrageously simplified and unacceptably wrong but it'll do for this post.

Your X server has to handle a lot of the hardware-specifics but other than that it's a wrapper around libxserver which does the work of ... well, being an X server.

Our stack looks like this:

+------------------------+
|  xserver   [libxserver]|--------[ X client ]
|                        |
|[libinput] [modesetting]|
+------------------------+
|       kernel           |
+------------------------+

Now, let's say instead of physical display devices, we want to render into an framebuffer, and we have no input devices.

+------------------------+
|  xserver   [libxserver]|--------[ X client ]
|                        |
|            [write()]   |
+------------------------+
|       some buffer      |
+------------------------+

Now, let's say those buffers are allocated elsewhere and we're just rendering to them. And those buffer are passed to us via an IPC protocol, like... Wayland!

+------------------------+
|  xserver   [libxserver]|--------[ X client ]
|                        |
|input events    [render]|
+------------------------+
      |            |
+------------------------+
|   Wayland compositor   |
+------------------------+

Fun fact: the Wayland compositor doesn't need to run on the hardware, you can play display server matryoshka until you run out of turtles.

In our glorious revisioned past all these are distinct projects, re-using libxserver and some external libraries where needed. Depending on the projects things may be very simple or get very complex, it depends on how we render things.

But in the end, we have several independent projects all providing us with an X server process - the specific X bits are done in libxserver though. We can release Xwayland without having to release Xorg or Xvfb.

libxserver won't need a lot of releases, the behaviour is largely specified by the protocol requirements and once you're done implementing it, it'll be quite a slow-moving project.

Ok, now, fast forward to 2021, lose some hindsight, hope, and attitude and - oh, we have exactly the above structure. Except that it's not spread across multiple independent repos on github, it's all sitting in the same git directory: our Xorg, Xwayland, Xvfb, etc. are all sitting in hw/$name, and libxserver is basically the rest of the repo.

A traditional X server release was a tag in that git directory. An XWayland-only release is basically an rm -rf hw/*-but-not-xwayland followed by a tag, an Xorg-only release is basically an rm -rf hw/*-but-not-xfree86 [2].

In theory, we could've moved all these out into separate projects a while ago but the benefits are small and no-one has the time for that anyway.

So there you have it - you can have Xorg-only or XWayland-only releases without the world coming to an end.

Now, for the "Xorg is dead" claims - it's very likely that the current release will be the last Xorg release. [3] There is little interest in an X server that runs on hardware, or rather: there's little interest in the effort required to push out releases. Povilas did a great job in getting this one out but again, it's likely this is the last release. [4]

Xwayland - very different, it'll hang around for a long time because it's "just" a protocol translation layer. And of course the interest is there, so we have volunteers to do the releases.

So basically: expecting Xwayland releases, be surprised (but not confused) by Xorg releases.

[1] Github of course doesn't exist yet because we're in the 80s. Time-travelling is complicated.
[2] Historical directory name, just accept it.
[3] Just like the previous release...
[4] At least until the next volunteer steps ups. Turns out the problem "no-one wants to work on this" is easily fixed by "me! me! I want to work on this". A concept that is apparently quite hard to understand in the peanut gallery.

The Strange State of Authenticated Boot and Disk Encryption on Generic Linux Distributions

TL;DR: Linux has been supporting Full Disk Encryption (FDE) and technologies such as UEFI SecureBoot and TPMs for a long time. However, the way they are set up by most distributions is not as secure as they should be, and in some ways quite frankly weird. In fact, right now, your data is probably more secure if stored on current ChromeOS, Android, Windows or MacOS devices, than it is on typical Linux distributions.

Generic Linux distributions (i.e. Debian, Fedora, Ubuntu, …) adopted Full Disk Encryption (FDE) more than 15 years ago, with the LUKS/cryptsetup infrastructure. It was a big step forward to a more secure environment. Almost ten years ago the big distributions started adding UEFI SecureBoot to their boot process. Support for Trusted Platform Modules (TPMs) has been added to the distributions a long time ago as well — but even though many PCs/laptops these days have TPM chips on-board it's generally not used in the default setup of generic Linux distributions.

How these technologies currently fit together on generic Linux distributions doesn't really make too much sense to me — and falls short of what they could actually deliver. In this story I'd like to have a closer look at why I think that, and what I propose to do about it.

The Basic Technologies

Let's have a closer look what these technologies actually deliver:

1. LUKS/dm-crypt/cryptsetup provide disk encryption, and optionally data authentication. Disk encryption means that reading the data in clear-text form is only possible if you possess a secret of some form, usually a password/passphrase. Data authentication means that no one can make changes to the data on disk unless they possess a secret of some form. Most distributions only enable the former though — the latter is a more recent addition to LUKS/cryptsetup, and is not used by default on most distributions (though it probably should be). Closely related to LUKS/dm-crypt is dm-verity (which can authenticate immutable volumes) and dm-integrity (which can authenticate writable volumes, among other things).

2. UEFI SecureBoot provides mechanisms for authenticating boot loaders and other pre-OS binaries before they are invoked. If those boot loaders then authenticate the next step of booting in a similar fashion there's a chain of trust which can ensure that only code that has some level of trust associated with it will run on the system. Authentication of boot loaders is done via cryptographic signatures: the OS/boot loader vendors cryptographically sign their boot loader binaries. The cryptographic certificates that may be used to validate these signatures are then signed by Microsoft, and since Microsoft's certificates are basically built into all of today's PCs and laptops this will provide some basic trust chain: if you want to modify the boot loader of a system you must have access to the private key used to sign the code (or to the private keys further up the certificate chain).

3. TPMs do many things. For this text we'll focus one facet: they can be used to protect secrets (for example for use in disk encryption, see above), that are released only if the code that booted the host can be authenticated in some form. This works roughly like this: every component that is used during the boot process (i.e. code, certificates, configuration, …) is hashed with a cryptographic hash function before it is used. The resulting hash is written to some small volatile memory the TPM maintains that is write-only (the so called Platform Configuration Registers, "PCRs"): each step of the boot process will write hashes of the resources needed by the next part of the boot process into these PCRs. The PCRs cannot be written freely: the hashes written are combined with what is already stored in the PCRs — also through hashing and the result of that then replaces the previous value. Effectively this means: only if every component involved in the boot matches expectations the hash values exposed in the TPM PCRs match the expected values too. And if you then use those values to unlock the secrets you want to protect you can guarantee that the key is only released to the OS if the expected OS and configuration is booted. The process of hashing the components of the boot process and writing that to the TPM PCRs is called "measuring". What's also important to mention is that the secrets are not only protected by these PCR values but encrypted with a "seed key" that is generated on the TPM chip itself, and cannot leave the TPM (at least so goes the theory). The idea is that you cannot read out a TPM's seed key, and thus you cannot duplicate the chip: unless you possess the original, physical chip you cannot retrieve the secret it might be able to unlock for you. Finally, TPMs can enforce a limit on unlock attempts per time ("anti-hammering"): this makes it hard to brute force things: if you can only execute a certain number of unlock attempts within some specific time then brute forcing will be prohibitively slow.

How Linux Distributions use these Technologies

As mentioned already, Linux distributions adopted the first two of these technologies widely, the third one not so much.

So typically, here's how the boot process of Linux distributions works these days:

1. The UEFI firmware invokes a piece of code called "shim" (which is stored in the EFI System Partition — the "ESP" — of your system), that more or less is just a list of certificates compiled into code form. The shim is signed with the aforementioned Microsoft key, that is built into all PCs/laptops. This list of certificates then can be used to validate the next step of the boot process. The shim is measured by the firmware into the TPM. (Well, the shim can do a bit more than what I describe here, but this is outside of the focus of this article.)

2. The shim then invokes a boot loader (often Grub) that is signed by a private key owned by the distribution vendor. The boot loader is stored in the ESP as well, plus some other places (i.e. possibly a separate boot partition). The corresponding certificate is included in the list of certificates built into the shim. The boot loader components are also measured into the TPM.

3. The boot loader then invokes the kernel and passes it an initial RAM disk image (initrd), which contains initial userspace code. The kernel itself is signed by the distribution vendor too. It's also validated via the shim. The initrd is not validated, though (!). The kernel is measured into the TPM, the initrd sometimes too.

4. The kernel unpacks the initrd image, and invokes what is contained in it. Typically, the initrd then asks the user for a password for the encrypted root file system. The initrd then uses that to set up the encrypted volume. No code authentication or TPM measurements take place.

5. The initrd then transitions into the root file system. No code authentication or TPM measurements take place.

6. When the OS itself is up the user is prompted for their user name, and their password. If correct, this will unlock the user account: the system is now ready to use. At this point no code authentication, no TPM measurements take place. Moreover, the user's password is not used to unlock any data, it's used only to allow or deny the login attempt — the user's data has already been decrypted a long time ago, by the initrd, as mentioned above.

What you'll notice here of course is that code validation happens for the shim, the boot loader and the kernel, but not for the initrd or the main OS code anymore. TPM measurements might go one step further: the initrd is measured sometimes too, if you are lucky. Moreover, you might notice that the disk encryption password and the user password are inquired by code that is not validated, and is thus not safe from external manipulation. You might also notice that even though TPM measurements of boot loader/OS components are done nothing actually ever makes use of the resulting PCRs in the typical setup.

Attack Scenarios

Of course, before determining whether the setup described above makes sense or not, one should have an idea what one actually intends to protect against.

The most basic attack scenario to focus on is probably that you want to be reasonably sure that if someone steals your laptop that contains all your data then this data remains confidential. The model described above probably delivers that to some degree: the full disk encryption when used with a reasonably strong password should make it hard for the laptop thief to access the data. The data is as secure as the password used is strong. The attacker might attempt to brute force the password, thus if the password is not chosen carefully the attacker might be successful.

Two more interesting attack scenarios go something like this:

1. Instead of stealing your laptop the attacker takes the harddisk from your laptop while you aren't watching (e.g. while you went for a walk and left it at home or in your hotel room), makes a copy of it, and then puts it back. You'll never notice they did that. The attacker then analyzes the data in their lab, maybe trying to brute force the password. In this scenario you won't even know that your data is at risk, because for you nothing changed — unlike in the basic scenario above. If the attacker manages to break your password they have full access to the data included on it, i.e. everything you so far stored on it, but not necessarily on what you are going to store on it later. This scenario is worse than the basic one mentioned above, for the simple fact that you won't know that you might be attacked. (This scenario could be extended further: maybe the attacker has a chance to watch you type in your password or so, effectively lowering the password strength.)

2. Instead of stealing your laptop the attacker takes the harddisk from your laptop while you aren't watching, inserts backdoor code on it, and puts it back. In this scenario you won't know your data is at risk, because physically everything is as before. What's really bad though is that the attacker gets access to anything you do on your laptop, both the data already on it, and whatever you will do in the future.

I think in particular this backdoor attack scenario is something we should be concerned about. We know for a fact that attacks like that happen all the time (Pegasus, industry espionage, …), hence we should make them hard.

Are we Safe?

So, does the scheme so far implemented by generic Linux distributions protect us against the latter two scenarios? Unfortunately not at all. Because distributions set up disk encryption the way they do, and only bind it to a user password, an attacker can easily duplicate the disk, and then attempt to brute force your password. What's worse: since code authentication ends at the kernel — and the initrd is not authenticated anymore —, backdooring is trivially easy: an attacker can change the initrd any way they want, without having to fight any kind of protections. And given that FDE unlocking is implemented in the initrd, and it's the initrd that asks for the encryption password things are just too easy: an attacker could trivially easily insert some code that picks up the FDE password as you type it in and send it wherever they want. And not just that: since once they are in they are in, they can do anything they like for the rest of the system's lifecycle, with full privileges — including installing backdoors for versions of the OS or kernel that are installed on the device in the future, so that their backdoor remains open for as long as they like.

That is sad of course. It's particular sad given that the other popular OSes all address this much better. ChromeOS, Android, Windows and MacOS all have way better built-in protections against attacks like this. And it's why one can certainly claim that your data is probably better protected right now if you store it on those OSes then it is on generic Linux distributions.

(Yeah, I know that there are some niche distros which do this better, and some hackers hack their own. But I care about general purpose distros here, i.e. the big ones, that most people base their work on.)

Note that there are more problems with the current setup. For example, it's really weird that during boot the user is queried for an FDE password which actually protects their data, and then once the system is up they are queried again – now asking for a username, and another password. And the weird thing is that this second authentication that appears to be user-focused doesn't really protect the user's data anymore — at that moment the data is already unlocked and accessible. The username/password query is supposed to be useful in multi-user scenarios of course, but how does that make any sense, given that these multiple users would all have to know a disk encryption password that unlocks the whole thing during the FDE step, and thus they have access to every user's data anyway if they make an offline copy of the harddisk?

Can we do better?

Of course we can, and that is what this story is actually supposed to be about.

Let's first figure out what the minimal issues we should fix are (at least in my humble opinion):

1. The initrd must be authenticated before being booted into. (And measured unconditionally.)

2. The OS binary resources (i.e. /usr/) must be authenticated before being booted into. (But don't need to be encrypted, since everyone has the same anyway, there's nothing to hide here.)

3. The OS configuration and state (i.e. /etc/ and /var/) must be encrypted, and authenticated before they are used. The encryption key should be bound to the TPM device; i.e system data should be locked to a security concept belonging to the system, not the user.

4. The user's home directory (i.e. /home/lennart/ and similar) must be encrypted and authenticated. The unlocking key should be bound to a user password or user security token (FIDO2 or PKCS#11 token); i.e. user data should be locked to a security concept belonging to the user, not the system.

Or to summarize this differently:

1. Every single component of the boot process and OS needs to be authenticated, i.e. all of shim (done), boot loader (done), kernel (done), initrd (missing so far), OS binary resources (missing so far), OS configuration and state (missing so far), the user's home (missing so far).

2. Encryption is necessary for the OS configuration and state (bound to TPM), and for the user's home directory (bound to a user password or user security token).

In Detail

Let's see how we can achieve the above in more detail.

How to Authenticate the initrd

At the moment initrds are generated on the installed host via scripts (dracut and similar) that try to figure out a minimal set of binaries and configuration data to build an initrd that contains just enough to be able to find and set up the root file system. What is included in the initrd hence depends highly on the individual installation and its configuration. Pretty likely no two initrds generated that way will be fully identical due to this. This model clearly has benefits: the initrds generated this way are very small and minimal, and support exactly what is necessary for the system to boot, and not less or more. It comes with serious drawbacks too though: the generation process is fragile and sometimes more akin to black magic than following clear rules: the generator script natively has to understand a myriad of storage stacks to determine what needs to be included and what not. It also means that authenticating the image is hard: given that each individual host gets a different specialized initrd, it means we cannot just sign the initrd with the vendor key like we sign the kernel. If we want to keep this design we'd have to figure out some other mechanism (e.g. a per-host signature key – that is generated locally; or by authenticating it with a message authentication code bound to the TPM). While these approaches are certainly thinkable, I am not convinced they actually are a good idea though: locally and dynamically generated per-host initrds is something we probably should move away from.

If we move away from locally generated initrds, things become a lot simpler. If the distribution vendor generates the initrds on their build systems then it can be attached to the kernel image itself, and thus be signed and measured along with the kernel image, without any further work. This simplicity is simply lovely. Besides robustness and reproducibility this gives us an easy route to authenticated initrds.

But of course, nothing is really that simple: working with vendor-generated initrds means that we can't adjust them anymore to the specifics of the individual host: if we pre-build the initrds and include them in the kernel image in immutable fashion then it becomes harder to support complex, more exotic storage or to parameterize it with local network server information, credentials, passwords, and so on. Now, for my simple laptop use-case these things don't matter, there's no need to extend/parameterize things, laptops and their setups are not that wildly different. But what to do about the cases where we want both: extensibility to cover for less common storage subsystems (iscsi, LVM, multipath, drivers for exotic hardware…) and parameterization?

Here's a proposal how to achieve that: let's build a basic initrd into the kernel as suggested, but then do two things to make this scheme both extensible and parameterizable, without compromising security.

1. Let's define a way how the basic initrd can be extended with additional files, which are stored in separate "extension images". The basic initrd should be able to discover these extension images, authenticate them and then activate them, thus extending the initrd with additional resources on-the-fly.

2. Let's define a way how we can safely pass additional parameters to the kernel/initrd (and actually the rest of the OS, too) in an authenticated (and possibly encrypted) fashion. Parameters in this context can be anything specific to the local installation, i.e. server information, security credentials, certificates, SSH server keys, or even just the root password that shall be able to unlock the root account in the initrd …

In such a scheme we should be able to deliver everything we are looking for:

1. We'll have a full trust chain for the code: the boot loader will authenticate and measure the kernel and basic initrd. The initrd extension images will then be authenticated by the basic initrd image.

2. We'll have authentication for all the parameters passed to the initrd.

This so far sounds very unspecific? Let's make it more specific by looking closer at the components I'd suggest to be used for this logic:

1. The systemd suite since a few months contains a subsystem implementing system extensions (v248). System extensions are ultimately just disk images (for example a squashfs file system in a GPT envelope) that can extend an underlying OS tree. Extending in this regard means they simply add additional files and directories into the OS tree, i.e. below /usr/. For a longer explanation see systemd-sysext(8). When a system extension is activated it is simply mounted and then merged into the main /usr/ tree via a read-only overlayfs mount. Now what's particularly nice about them in this context we are talking about here is that the extension images may carry dm-verity authentication data, and PKCS#7 signatures (once this is merged, that is, i.e. v250).

2. The systemd suite also contains a concept called service "credentials". These are small pieces of information passed to services in a secure way. One key feature of these credentials is that they can be encrypted and authenticated in a very simple way with a key bound to the TPM (v250). See LoadCredentialEncrypted= and systemd-creds(1) for details. They are great for safely storing SSL private keys and similar on your system, but they also come handy for parameterizing initrds: an encrypted credential is just a file that can only be decoded if the right TPM is around with the right PCR values set.

3. The systemd suite contains a component called systemd-stub(7). It's an EFI stub, i.e. a small piece of code that is attached to a kernel image, and turns the kernel image into a regular EFI binary that can be directly executed by the firmware (or a boot loader). This stub has a number of nice features (for example, it can show a boot splash before invoking the Linux kernel itself and such). Once this work is merged (v250) the stub will support one more feature: it will automatically search for system extension image files and credential files next to the kernel image file, measure them and pass them on to the main initrd of the host.

Putting this together we have nice way to provide fully authenticated kernel images, initrd images and initrd extension images; as well as encrypted and authenticated parameters via the credentials logic.

How would a distribution actually make us of this? A distribution vendor would pre-build the basic initrd, and glue it into the kernel image, and sign that as a whole. Then, for each supposed extension of the basic initrd (e.g. one for iscsi support, one for LVM, one for multipath, …), the vendor would use a tool such as mkosi to build an extension image, i.e. a GPT disk image containing the files in squashfs format, a Verity partition that authenticates it, plus a PKCS#7 signature partition that validates the root hash for the dm-verity partition, and that can be checked against a key provided by the boot loader or main initrd. Then, any parameters for the initrd will be encrypted using systemd-creds encrypt -T. The resulting encrypted credentials and the initrd extension images are then simply placed next to the kernel image in the ESP (or boot partition). Done.

This checks all boxes: everything is authenticated and measured, the credentials also encrypted. Things remain extensible and modular, can be pre-built by the vendor, and installation is as simple as dropping in one file for each extension and/or credential.

How to Authenticate the Binary OS Resources

Let's now have a look how to authenticate the Binary OS resources, i.e. the stuff you find in /usr/, i.e. the stuff traditionally shipped to the user's system via RPMs or DEBs.

I think there are three relevant ways how to authenticate this:

1. Make /usr/ a dm-verity volume. dm-verity is a concept implemented in the Linux kernel that provides authenticity to read-only block devices: every read access is cryptographically verified against a top-level hash value. This top-level hash is typically a 256bit value that you can either encode in the kernel image you are using, or cryptographically sign (which is particularly nice once this is merged). I think this is actually the best approach since it makes the /usr/ tree entirely immutable in a very simple way. However, this also means that the whole of /usr/ needs to be updated as once, i.e. the traditional rpm/apt based update logic cannot work in this mode.

2. Make /usr/ a dm-integrity volume. dm-integrity is a concept provided by the Linux kernel that offers integrity guarantees to writable block devices, i.e. in some ways it can be considered to be a bit like dm-verity while permitting write access. It can be used in three ways, one of which I think is particularly relevant here. The first way is with a simple hash function in "stand-alone" mode: this is not too interesting here, it just provides greater data safety for file systems that don't hash check their files' data on their own. The second way is in combination with dm-crypt, i.e. with disk encryption. In this case it adds authenticity to confidentiality: only if you know the right secret you can read and make changes to the data, and any attempt to make changes without knowing this secret key will be detected as IO error on next read by those in possession of the secret (more about this below). The third way is the one I think is most interesting here: in "stand-alone" mode, but with a keyed hash function (e.g. HMAC). What's this good for? This provides authenticity without encryption: if you make changes to the disk without knowing the secret this will be noticed on the next read attempt of the data and result in IO errors. This mode provides what we want (authenticity) and doesn't do what we don't need (encryption). Of course, the secret key for the HMAC must be provided somehow, I think ideally by the TPM.

3. Make /usr/ a dm-crypt (LUKS) + dm-integrity volume. This provides both authenticity and encryption. The latter isn't typically needed for /usr/ given that it generally contains no secret data: anyone can download the binaries off the Internet anyway, and the sources too. By encrypting this you'll waste CPU cycles, but beyond that it doesn't hurt much. (Admittedly, some people might want to hide the precise set of packages they have installed, since it of course does reveal a bit of information about you: i.e. what you are working on, maybe what your job is – think: if you are a hacker you have hacking tools installed – and similar). Going this way might simplify things in some cases, as it means you don't have to distinguish "OS binary resources" (i.e /usr/) and "OS configuration and state" (i.e. /etc/ + /var/, see below), and just make it the same volume. Here too, the secret key must be provided somehow, I think ideally by the TPM.

All three approach are valid. The first approach has my primary sympathies, but for distributions not willing to abandon client-side updates via RPM/dpkg this is not an option, in which case I would propose the other two approaches for these cases.

The LUKS encryption key (and in case of dm-integrity standalone mode the key for the keyed hash function) should be bound to the TPM. Why the TPM for this? You could also use a user password, a FIDO2 or PKCS#11 security token — but I think TPM is the right choice: why that? To reduce the requirement for repeated authentication, i.e. that you first have to provide the disk encryption password, and then you have to login, providing another password. It should be possible that the system boots up unattended and then only one authentication prompt is needed to unlock the user's data properly. The TPM provides a way to do this in a reasonably safe and fully unattended way. Also, when we stop considering just the laptop use-case for a moment: on servers interactive disk encryption prompts don't make much sense — the fact that TPMs can provide secrets without this requiring user interaction and thus the ability to work in entirely unattended environments is quite desirable. Note that crypttab(5) as implemented by systemd (v248) provides native support for authentication via password, via TPM2, via PKCS#11 or via FIDO2, so the choice is ultimately all yours.

How to Encrypt/Authenticate OS Configuration and State

Let's now look at the OS configuration and state, i.e. the stuff in /etc/ and /var/. It probably makes sense to not consider these two hierarchies independently but instead just consider this to be the root file system. If the OS binary resources are in a separate file system it is then mounted onto the /usr/ sub-directory of the root file system.

The OS configuration and state (or: root file system) should be both encrypted and authenticated: it might contain secret keys, user passwords, privileged logs and similar. This data matters and contains plenty data that should remain confidential.

The encryption of choice here is dm-crypt (LUKS) + dm-integrity similar as discussed above, again with the key bound to the TPM.

If the OS binary resources are protected the same way it is safe to merge these two volumes and have a single partition for both (see above)

How to Encrypt/Authenticate the User's Home Directory

The data in the user's home directory should be encrypted, and bound to the user's preferred token of authentication (i.e. a password or FIDO2/PKCS#11 security token). As mentioned, in the traditional mode of operation the user's home directory is not individually encrypted, but only encrypted because FDE is in use. The encryption key for that is a system wide key though, not a per-user key. And I think that's problem, as mentioned (and probably not even generally understood by our users). We should correct that and ensure that the user's password is what unlocks the user's data.

In the systemd suite we provide a service systemd-homed(8) (v245) that implements this in a safe way: each user gets its own LUKS volume stored in a loopback file in /home/, and this is enough to synthesize a user account. The encryption password for this volume is the user's account password, thus it's really the password provided at login time that unlocks the user's data. systemd-homed also supports other mechanisms of authentication, in particular PKCS#11/FIDO2 security tokens. It also provides support for other storage back-ends (such as fscrypt), but I'd always suggest to use the LUKS back-end since it's the only one providing the comprehensive confidentiality guarantees one wants for a UNIX-style home directory.

Note that there's one special caveat here: if the user's home directory (e.g. /home/lennart/) is encrypted and authenticated, what about the file system this data is stored on, i.e. /home/ itself? If that dir is part of the the root file system this would result in double encryption: first the data is encrypted with the TPM root file system key, and then again with the per-user key. Such double encryption is a waste of resources, and unnecessary. I'd thus suggest to make /home/ its own dm-integrity volume with a HMAC, keyed by the TPM. This means the data stored directly in /home/ will be authenticated but not encrypted. That's good not only for performance, but also has practical benefits: it allows extracting the encrypted volume of the various users in case the TPM key is lost, as a way to recover from dead laptops or similar.

Why authenticate /home/, if it only contains per-user home directories that are authenticated on their own anyway? That's a valid question: it's because the kernel file system maintainers made clear that Linux file system code is not considered safe against rogue disk images, and is not tested for that; this means before you mount anything you need to establish trust in some way because otherwise there's a risk that the act of mounting might exploit your kernel.

Summary of Resources and their Protections

So, let's now put this all together. Here's a table showing the various resources we deal with, and how I think they should be protected (in my idealized world).

This should provide all the desired guarantees: everything is authenticated, and the individualized per-host or per-user data is also encrypted. No double encryption takes place. The encryption keys/verification certificates are stored/bound to the most appropriate infrastructure.

Does this address the three attack scenarios mentioned earlier? I think so, yes. The basic attack scenario I described is addressed by the fact that /var/, /etc/ and /home/*/ are encrypted. Brute forcing the former two is harder than in the status quo ante model, since a high entropy key is used instead of one derived from a user provided password. Moreover, the "anti-hammering" logic of the TPM will make brute forcing prohibitively slow. The home directories are protected by the user's password or ideally a personal FIDO2/PKCS#11 security token in this model. Of course, a password isn't better security-wise then the status quo ante. But given the FIDO2/PKCS#11 support built into systemd-homed it should be easier to lock down the home directories securely.

Binding encryption of /var/ and /etc/ to the TPM also addresses the first of the two more advanced attack scenarios: a copy of the harddisk is useless without the physical TPM chip, since the seed key is sealed into that. (And even if the attacker had the chance to watch you type in your password, it won't help unless they possess access to to the TPM chip.) For the home directory this attack is not addressed as long as a plain password is used. However, since binding home directories to FIDO2/PKCS#11 tokens is built into systemd-homed things should be safe here too — provided the user actually possesses and uses such a device.

The backdoor attack scenario is addressed by the fact that every resource in play now is authenticated: it's hard to backdoor the OS if there's no component that isn't verified by signature keys or TPM secrets the attacker hopefully doesn't know.

For general purpose distributions that focus on updating the OS per RPM/dpkg the idealized model above won't work out, since (as mentioned) this implies an immutable /usr/, and thus requires updating /usr/ via an atomic update operation. For such distros a setup like the following is probably more realistic, but see above.

This means there's only one root file system that contains all of /etc/, /var/ and /usr/.

Recovery Keys

When binding encryption to TPMs one problem that arises is what strategy to adopt if the TPM is lost, due to hardware failure: if I need the TPM to unlock my encrypted volume, what do I do if I need the data but lost the TPM?

The answer here is supporting recovery keys (this is similar to how other OSes approach this). Recovery keys are pretty much the same concept as passwords. The main difference being that they are computer generated rather than user-chosen. Because of that they typically have much higher entropy (which makes them more annoying to type in, i.e you want to use them only when you must, not day-to-day). By having higher entropy they are useful in combination with TPM, FIDO2 or PKCS#11 based unlocking: unlike a combination with passwords they do not compromise the higher strength of protection that TPM/FIDO2/PKCS#11 based unlocking is supposed to provide.

Current versions of systemd-cryptenroll(1) implement a recovery key concept in an attempt to address this problem. You may enroll any combination of TPM chips, PKCS#11 tokens, FIDO2 tokens, recovery keys and passwords on the same LUKS volume. When enrolling a recovery key it is generated and shown on screen both in text form and as QR code you can scan off screen if you like. The idea is write down/store this recovery key at a safe place so that you can use it when you need it. Note that such recovery keys can be entered wherever a LUKS password is requested, i.e. after generation they behave pretty much the same as a regular password.

TPM PCR Brittleness

Locking devices to TPMs and enforcing a PCR policy with this (i.e. configuring the TPM key to be unlockable only if certain PCRs match certain values, and thus requiring the OS to be in a certain state) brings a problem with it: TPM PCR brittleness. If the key you want to unlock with the TPM requires the OS to be in a specific state (i.e. that all OS components' hashes match certain expectations or similar) then doing OS updates might have the affect of making your key inaccessible: the OS updates will cause the code to change, and thus the hashes of the code, and thus certain PCRs. (Thankfully, you unrolled a recovery key, as described above, so this doesn't mean you lost your data, right?).

To address this I'd suggest three strategies:

1. Most importantly: don't actually use the TPM PCRs that contain code hashes. There are actually multiple PCRs defined, each containing measurements of different aspects of the boot process. My recommendation is to bind keys to PCR 7 only, a PCR that contains measurements of the UEFI SecureBoot certificate databases. Thus, the keys will remain accessible as long as these databases remain the same, and updates to code will not affect it (updates to the certificate databases will, and they do happen too, though hopefully much less frequent then code updates). Does this reduce security? Not much, no, because the code that's run is after all not just measured but also validated via code signatures, and those signatures are validated with the aforementioned certificate databases. Thus binding an encrypted TPM key to PCR 7 should enforce a similar level of trust in the boot/OS code as binding it to a PCR with hashes of specific versions of that code. i.e. using PCR 7 means you say "every code signed by these vendors is allowed to unlock my key" while using a PCR that contains code hashes means "only this exact version of my code may access my key".

2. Use LUKS key management to enroll multiple versions of the TPM keys in relevant volumes, to support multiple versions of the OS code (or multiple versions of the certificate database, as discussed above). Specifically: whenever an update is done that might result changing the relevant PCRs, pre-calculate the new PCRs, and enroll them in an additional LUKS slot on the relevant volumes. This means that the unlocking keys tied to the TPM remain accessible in both states of the system. Eventually, once rebooted after the update, remove the old slots.

3. If these two strategies didn't work out (maybe because the OS/firmware was updated outside of OS control, or the update mechanism was aborted at the wrong time) and the TPM PCRs changed unexpectedly, and the user now needs to use their recovery key to get access to the OS back, let's handle this gracefully and automatically reenroll the current TPM PCRs at boot, after the recovery key checked out, so that for future boots everything is in order again.

Other approaches can work too: for example, some OSes simply remove TPM PCR policy protection of disk encryption keys altogether immediately before OS or firmware updates, and then reenable it right after. Of course, this opens a time window where the key bound to the TPM is much less protected than people might assume. I'd try to avoid such a scheme if possible.

Anything Else?

So, given that we are talking about idealized systems: I personally actually think the ideal OS would be much simpler, and thus more secure than this:

I'd try to ditch the Shim, and instead focus on enrolling the distribution vendor keys directly in the UEFI firmware certificate list. This is actually supported by all firmwares too. This has various benefits: it's no longer necessary to bind everything to Microsoft's root key, you can just enroll your own stuff and thus make sure only what you want to trust is trusted and nothing else. To make an approach like this easier, we have been working on doing automatic enrollment of these keys from the systemd-boot boot loader, see this work in progress for details. This way the Firmware will authenticate the boot loader/kernel/initrd without any further component for this in place.

I'd also not bother with a separate boot partition, and just use the ESP for everything. The ESP is required anyway by the firmware, and is good enough for storing the few files we need.

FAQ

Can I implement all of this in my distribution today?

Probably not. While the big issues have mostly been addressed there's a lot of integration work still missing. As you might have seen I linked some PRs that haven't even been merged into our tree yet, and definitely not been released yet or even entered the distributions.

Will this show up in Fedora/Debian/Ubuntu soon?

I don't know. I am making a proposal how these things might work, and am working on getting various building blocks for this into shape. What the distributions do is up to them. But even if they don't follow the recommendations I make 100%, or don't want to use the building blocks I propose I think it's important they start thinking about this, and yes, I think they should be thinking about defaulting to setups like this.

Work for measuring/signing initrds on Fedora has been started, here's a slide deck with some information about it.

But isn't a TPM evil?

Some corners of the community tried (unfortunately successfully to some degree) to paint TPMs/Trusted Computing/SecureBoot as generally evil technologies that stop us from using our systems the way we want. That idea is rubbish though, I think. We should focus on what it can deliver for us (and that's a lot I think, see above), and appreciate the fact we can actually use it to kick out perceived evil empires from our devices instead of being subjected to them. Yes, the way SecureBoot/TPMs are defined puts you in the driver seat if you want — and you may enroll your own certificates to keep out everything you don't like.

What if my system doesn't have a TPM?

TPMs are becoming quite ubiquitous, in particular as the upcoming Windows versions will require them. In general I think we should focus on modern, fully equipped systems when designing all this, and then find fall-backs for more limited systems. Frankly it feels as if so far the design approach for all this was the other way round: try to make the new stuff work like the old rather than the old like the new (I mean, to me it appears this thinking is the main raison d'être for the Grub boot loader).

More specifically, on the systems where we have no TPM we ultimately cannot provide the same security guarantees as for those which have. So depending on the resource to protect we should fall back to different TPM-less mechanisms. For example, if we have no TPM then the root file system should probably be encrypted with a user provided password, typed in at boot as before. And for the encrypted boot credentials we probably should simply not encrypt them, and place them in the ESP unencrypted.

Effectively this means: without TPM you'll still get protection regarding the basic attack scenario, as before, but not the other two.

What if my system doesn't have UEFI?

Many of the mechanisms explained above taken individually do not require UEFI. But of course the chain of trust suggested above requires something like UEFI SecureBoot. If your system lacks UEFI it's probably best to find work-alikes to the technologies suggested above, but I doubt I'll be able to help you there.

rpm/dpkg already cryptographically validates all packages at installation time (gpg), why would I need more than that?

This type of package validation happens once: at the moment of installation (or update) of the package, but not anymore when the data installed is actually used. Thus when an attacker manages to modify the package data after installation and before use they can make any change they like without this ever being noticed. Such package download validation does address certain attack scenarios (i.e. man-in-the-middle attacks on network downloads), but it doesn't protect you from attackers with physical access, as described in the attack scenarios above.

Systems such as ostree aren't better than rpm/dpkg regarding this BTW, their data is not validated on use either, but only during download or when processing tree checkouts.

Key really here is that the scheme explained here provides offline protection for the data "at rest" — even someone with physical access to your device cannot easily make changes that aren't noticed on next use. rpm/dpkg/ostree provide online protection only: as long as the system remains up, and all OS changes are done through the intended program code-paths, and no one has physical access everything should be good. In today's world I am sure this is not good enough though. As mentioned most modern OSes provide offline protection for the data at rest in one way or another. Generic Linux distributions are terribly behind on this.

This is all so desktop/laptop focused, what about servers?

I am pretty sure servers should provide similar security guarantees as outlined above. In a way servers are a much simpler case: there are no users and no interactivity. Thus the discussion of /home/ and what it contains and of user passwords doesn't matter. However, the authenticated initrd and the unattended TPM-based encryption I think are very important for servers too, in a trusted data center environment. It provides security guarantees so far not given by Linux server OSes.

I'd like to help with this, or discuss/comment on this

Submit patches or reviews through GitHub. General discussion about this is best done on the systemd mailing list.

ES: But Why

I got a request recently to fix up the WebGL Aquarium demo. I’ve had this bookmarked for a while since it’s one of the only test cases for GL_EXT_multisampled_render_to_texture I’m aware of, at least when running Chrome in EGL mode.

Naturally, I decided to do both at once since this would be yet another extension that no native desktop driver in Mesa currently supports.

Transient Multisampling

The idea behind this extension is that on tilers, a single-sample image can be temporarily treated as multisampled without needing the extra memory bandwidth that multisampled images require. The multisampled rendering image is transient, meaning it isn’t loaded or written to, so it can be lazily allocated and then discarded.

Vulkan has similar mechanisms: loadOp and storeOp set to VK_ATTACHMENT_LOAD_OP_DONT_CARE with an image allocated using ZINK_HEAP_DEVICE_LOCAL_LAZY and VK_IMAGE_USAGE_TRANSIENT_ATTACHMENT_BIT, then adding a resolve attachment for writeout. Simple enough.

Unfortunately, this doesn’t quite work.

As Danylo Piliaiev, RenderDoc master and Finder Of Bad Pixels, was quick to point out, this approach will discard any previous data the texture had, resulting in bad pixels. Lots of bad pixels, in fact.

The solution, which I hate, is that the “transient” multisampled image now gets a full-texture draw before its “transient” renderpass to initialize it, then is loaded with VK_ATTACHMENT_LOAD_OP_LOAD, effectively making it not transient at all.

Sorry tilers.

No frames for you.

But it does work and give solid performance (-20% or so while recording smh screen recorders git gud), so I can’t be too mad.

I am back with another status update on raytracing in RADV. And the good news is that things are finally starting to come together. After ~9 months of on and off work we’re now having games working with raytracing. Working on first try after getting all the required functionality was Control:

After poking for a long time at CTS and demos it is really nice to see the fruits of ones work.

The piece that I added recently was copy/compaction and serialization of acceleration structures which was a bunch of shader writing, handling another type of queries and dealing with indirect dispatches. (Since of course the API doesn’t give the input size on the CPU. No idea how this API should be secured …)

What games?

I did try 5 games:

1. Quake 2 RTX (Vulkan): works. This was working already on my previous update.
2. Control (D3D): works. Pretty much just works. Runs at maybe 30-50% of RT performance on Windows.
3. Metro Exodus (Vulkan): works. Needs one workaround and is very finicky in WSI but otherwise works fine. Runs at 20-25% of RT performance on Windows.
4. Ghostrunner (D3D): Does not work. This really needs per shadergroup compilation instead of just mashing all the shaders together as I get shaders now with 1 million NIR instructions, which is a pain to debug.
5. Doom Eternal (Vulkan): Does not work. The raytracing option in the menu stays grayed out and at this point I’m at a loss what is required to make the game allow enabling RT.

If anybody could tell me how to get Doom Eternal to allow RT I’d appreciate it.

What is next?

Of course the support is far from done. Some things to still make progress on:

1. Upstreaming what I have. Samuel has been busy reviewing my MRs and I think there is a good chance that what I have now will make it into 21.3.
2. Improve the pipeline compilation model to hopefully make ghostrunner work.
3. Improved BVH building. The current BVH is really naive, which is likely one of the big performance factors.
4. Improve traversal.
5. Move on to stuff needed for DXR 1.1 like VK_KHR_ray_query.

P.S. If you haven’t seen it yet, Jason Ekstrand from Intel recently gave a talk about how Intel implements raytracing. Nice showcase of how you can provide some more involved hardware implementation than RDNA2 does.

Been some time since my last update, so I felt it was time to flex my blog writing muscles again and provide some updates of some of the things we are working on in Fedora in preparation for Fedora Workstation 35. This is not meant to be a comprehensive whats new article about Fedora Workstation 35, more of a listing of some of the things we are doing as part of the Red Hat desktop team.

NVidia support for Wayland
One thing we spent a lot of effort on for a long time now is getting full support for the NVidia binary driver under Wayland. It has been a recurring topic in our bi-weekly calls with the NVidia engineering team ever since we started looking at moving to Wayland. There has been basic binary driver support for some time, meaning you could run a native Wayland session on top of the binary driver, but the critical missing piece was that you could not get support for accelerated graphics when running applications through XWayland, our X.org compatibility layer. Which basically meant that any application requiring 3D support and which wasn’t a native Wayland application yet wouldn’t work. So over the last Months we been having a great collaboration with NVidia around closing this gap, with them working closely with us in fixing issues in their driver while we have been fixing bugs and missing pieces in the rest of the stack. We been reporting and discussing issues back and forth allowing us a very quickly turnaround on issues as we find them which of course all resulted in the NVidia 470.42.01 driver with XWayland support. I am sure we will find new corner cases that needs to be resolved in the coming Months, but I am equally sure we will be able to quickly resolve them due to the close collaboration we have now established with NVidia. And I know some people will wonder why we spent so much time working with NVidia around their binary driver, but the reality is that NVidia is the market leader, especially in the professional Linux workstation space, and there are lot of people who either would end up not using Linux or using Linux with X without it, including a lot of Red Hat customers and Fedora users. And that is what I and my team are here for at the end of the day, to make sure Red Hat customers are able to get their job done using their Linux systems.

Lightweight kiosk mode
One of the wonderful things about open source is the constant flow of code and innovation between all the different parts of the ecosystem. For instance one thing we on the RHEL side have often been asked about over the last few years is a lightweight and simple to use solution for people wanting to run single application setups, like information boards, ATM machines, cash registers, information kiosks and so on. For many use cases people felt that running a full GNOME 3 desktop underneath their application was either to resource hungry and or created a risk that people accidentally end up in the desktop session. At the same time from our viewpoint as a development team we didn’t want a completely separate stack for this use case as that would just increase our maintenance burden as we would end up having to do a lot of things twice. So to solve this problem Ray Strode spent some time writing what we call GNOME Kiosk mode which makes setting up a simple session running single application easy and without running things like the GNOME shell, tracker, evolution etc. This gives you a window manager with full support for the latest technologies such as compositing, libinput and Wayland, but coming in at about 18MB, which is about 71MB less than a minimal GNOME 3 desktop session. You can read more about the new Kiosk mode and how to use it in this great blog post from our savvy Edge Computing Product Manager Ben Breard. The kiosk mode session described in Ben’s article about RHEL will be available with Fedora Workstation 35.

high-definition mouse wheel support
A major part of what we do is making sure that Red Hat Enterprise Linux customers and Fedora users get hardware support on par with what you find on other operating systems. We try our best to work with our hardware partners, like Lenovo, to ensure that such hardware support comes day and date with when those features are enabled on other systems, but some things ends up taking longer time for various reasons. Support for high-definition mouse wheels was one of those. Peter Hutterer, our resident input expert, put together a great blog post explaining the history and status of high-definition mouse wheel support. As Peter points out in his blog post the feature is not yet fully supported under Wayland, but we hope to close that gap in time for Fedora Workstation 35.

Mouse with HiRes scroll wheel

PipeWire
I feel I can’t do one of these posts without talking about latest developments in PipeWire, our unified audio and video server. Wim Taymans keeps working with rapidly growing PipeWire community to fix issues as they are reported and add new features to PipeWire. Most recently Wims focus has been on implementing support for S/PDIF passthrough support over both S/PDIF and HDMI connections. This will allow us to send undecoded data over such connections which is critical for working well with surround sound systems and soundbars. Also the PipeWire community has been working hard on further improving the Bluetooth support with bluetooth battery status support for head-set profile and using Apple extensions. aptX-LL and FastStream codec support was also added. And of course a huge amount of bug fixes, it turns out that when you replace two different sound servers that has been around for close to two decades there are a lot of corner cases to cover :). Make sure to check out two latest release notes for 0.3.35 and for 0.3.36 for details.

EasyEffects is a great example of a cool new application built with PipeWire

Privacy screen
Another feature that we have been working on as a result of our Lenovo partnership is Privacy screen support. For those not familiar with this technology it is basically to allow you to reduce the readability of your screen when viewed from the side, so that if you are using your laptop at a coffee shop for instance then a person sitting close by will have a lot harder time trying to read what is on your screen. Hans de Goede has been shepherding the kernel side of this forward working with Marco Trevisan from Canonical on the userspace part of it (which also makes this a nice example of cross-company collaboration), allowing you to turn this feature on or off. This feature though is not likely to fully land in time for Fedora Workstation 35 so we are looking at if we will bring this in as an update to Fedora Workstation 35 or if it will be a Fedora Workstation 36 feature.

Penny

Zink inside the penny

Zink implements OpenGL in terms of Vulkan, as far as the drawing itself is concerned, but presenting that drawing to the rest of the system is currently system-specific (GLX). For hardware that already has a Mesa driver, we use GBM. On NVIDIA’s Vulkan (and probably any other binary stacks on Linux, and probably also like WSL or macOS + MoltenVK) we download the image from the GPU back to the CPU and then use the same software upload/display path as llvmpipe, which as you can imagine is Not Fast.

Penny aims to extend Zink by replacing both of those paths, and instead using the various Vulkan WSI extensions to manage presentation. Even for the GBM case this should enable higher performance since zink will have more information about the rendering pipeline (multisampling in particular is poorly handled atm). Future window system integration work can focus on Vulkan, with EGL and GLX getting features “for free” once they’re enabled in Vulkan.

3rd party software cleanup
Over time we have been working on adding more and more 3rd party software for easy consumption in Fedora Workstation. The problem we discovered though was that due to this being done over time, with changing requirements and expectations, the functionality was not behaving in a very intuitive way and there was also new questions that needed to be answered. So Allan Day and Owen Taylor spent some time this cycle to review all the bits and pieces of this functionality and worked to clean it up. So the goal is that when you enable third-party repositories in Fedora Workstation 35 it behaves in a much more predictable and understandable way and also includes a lot of applications from Flathub. Yes, that is correct you should be able to install a lot of applications from Flathub in Fedora Workstation 35 without having to first visit the Flathub website to enable it, instead they will show up once you turned the knob for general 3rd party application support.

Power profiles
Another item we spent quite a bit of time for Fedora Workstation 35 is making sure we integrate the Power Profiles work that Bastien Nocera has been working on as part of our collaboration with Lenovo. Power Profiles is basically a feature that allows your system to behave in a smarter way when it comes to power consumption and thus prolongs your battery life. So for instance when we notice you are getting low on battery we can offer you to go into a strong power saving mode to prolong how long you can use the system until you can recharge. More in-depth explanation of Power profiles in the official README.

Wayland
I usually also have ended up talking about Wayland in my posts, but I expect to be doing less going forward as we have now covered all the major gaps we saw between Wayland and X.org. Jonas Ådahl got the headless support merged which was one of our big missing pieces and as mentioned above Olivier Fourdan and Jonas and others worked with NVidia on getting the binary driver with XWayland support working with GNOME Shell. Of course this being software we are never truly done, there will of course be new issues discovered, random bugs that needs to be fixed, and of course also new features that needs to be implemented. We already have our next big team focus in place, HDR support, which will need work from the graphics drivers, up through Mesa, into the window manager and the GUI toolkits and in the applications themselves. We been investigating and trying out some things for a while already, but we are now ready to make this a main focus for the team. In fact we will soon be posting a new job listing for a fulltime engineer to work on HDR vertically through the stack so keep an eye out for that if you are interested in working on this. The job will be open to candidates who which to work remotely, so as long as Red Hat has a business presence in the country you live we should be able to offer you the job if you are the right candidate for us. Update:Job listing is now online for our HDR engineer.

BTW, if you want to see future updates and keep on top of other happenings from Fedora and Red Hat in the desktop space, make sure to follow me on twitter.

Zink Is Over: This Time I’m Serious.

Look.

I know what you’re gonna say, and maybe I did just say zink was done a week or two ago.

I’m not saying I didn’t.

But that was practically last year at the speed with which zink’s codebase moves and its developer community sits in my office eating cookies between Mesa builds, and it was also before I set off on my journey to make the rest of those zany Phoronix benchmark games run instead of crashing or whatever.

What do we got left on that list anyway?

Metro: Last Light Redux

Oh you want some Metro? We got Metro at home.

HITMAN

Agent 47, I’m gonna pretend I didn’t see that. Pull yourself together.

Basemark: High Settings

It’s uh… Mangohud’s slowing me down.

Bioshock Infinite

I bet you’re wondering where this one is, huh.

Warhammer 40,000: Dawn of War

Easy as that, ju—Wait, what?

This game requires ARB_bindless_texture just to run? Is this a joke? Even fucking DOOM 2016, the final boss of OpenGL, doesn’t require bindless textures.

Fine.

Totally fine.

Not at all a problem, and I’m sure it’ll be easy to do.

Definitely no reason why only two Mesa drivers total implement it other than it being some trivial switch that everyone forgot to flip, right?

Probably just a config value here, or maybe a couple lines of code there…

Ignore all the validation errors because descriptor indexing isn’t accurately supported…

Add some null checks…

Fire up ASAN to fix a random stack explosion…

File a piglit ticket because two of the eighty unit tests for the extension are bugged and these are quite literally the only unit tests available…

Kapow, first try, expect it in zink-wip later today-ish.

It’s just that easy.

If you disagree, you are nitpicking and biased.

I've been working on portals recently and one of the issues for me was that the documentation just didn't quite hit the sweet spot. At least the bits I found were either too high-level or too implementation-specific. So here's a set of notes on how a portal works, in the hope that this is actually correct.

First, Portals are supposed to be a way for sandboxed applications (flatpaks) to trigger functionality they don't have direct access too. The prime example: opening a file without the application having access to $HOME. This is done by the applications talking to portals instead of doing the functionality themselves.

There is really only one portal process: /usr/libexec/xdg-desktop-portal, started as a systemd user service. That process owns a DBus bus name (org.freedesktop.portal.Desktop) and an object on that name (/org/freedesktop/portal/desktop). You can see that bus name and object with D-Feet, from DBus' POV there's nothing special about it. What makes it the portal is simply that the application running inside the sandbox can talk to that DBus name and thus call the various methods. Obviously the xdg-desktop-portal needs to run outside the sandbox to do its things.

There are multiple portal interfaces, all available on that one object. Those interfaces have names like org.freedesktop.portal.FileChooser (to open/save files). The xdg-desktop-portal implements those interfaces and thus handles any method calls on those interfaces. So where an application is sandboxed, it doesn't implement the functionality itself, it instead calls e.g. the OpenFile() method on the org.freedesktop.portal.FileChooser interface. Then it gets an fd back and can read the content of that file without needing full access to the file system.

Some interfaces are fully handled within xdg-desktop-portal. For example, the Camera portal checks a few things internally, pops up a dialog for the user to confirm access to if needed [1] but otherwise there's nothing else involved with this specific method call.

Other interfaces have a backend "implementation" DBus interface. For example, the org.freedesktop.portal.FileChooser interface has a org.freedesktop.impl.portal.FileChooser (notice the "impl") counterpart. xdg-desktop-portal does not implement those impl.portals. xdg-desktop-portal instead routes the DBus calls to the respective "impl.portal". Your sandboxed application calls OpenFile(), xdg-desktop-portal now calls OpenFile() on org.freedesktop.impl.portal.FileChooser. That interface returns a value, xdg-desktop-portal extracts it and returns it back to the application in respones to the original OpenFile() call.

What provides those impl.portals doesn't matter to xdg-desktop-portal, and this is where things are hot-swappable. GTK and Qt both provide (some of) those impl portals, There are GTK and Qt-specific portals with xdg-desktop-portal-gtk and xdg-desktop-portal-kde but another one is provided by GNOME Shell directly. You can check the files in /usr/share/xdg-desktop-portal/portals/ and see which impl portal is provided on which bus name. The reason those impl.portals exist is so they can be native to the desktop environment - regardless what application you're running and with a generic xdg-desktop-portal, you see the native file chooser dialog for your desktop environment.

So the full call sequence is:

• At startup, xdg-desktop-portal parses the /usr/libexec/xdg-desktop-portal/*.portal files to know which impl.portal interface is provided on which bus name
• The application calls OpenFile() on the org.freedesktop.portal.FileChooser interface on the object path /org/freedesktop/portal/desktop. It can do so because the bus name this object sits on is not restricted by the sandbox
• xdg-desktop-portal receives that call. This is portal with an impl.portal so xdg-desktop-portal calls OpenFile() on the bus name that provides the org.freedesktop.impl.portal.FileChooser interface (as previously established by reading the *.portal files)
• Assuming xdg-desktop-portal-gtk provides that portal at the moment, that process now pops up a GTK FileChooser dialog that runs outside the sandbox. User selects a file
• xdg-desktop-portal-gtk sends back the fd for the file to the xdg-desktop-portal, and the impl.portal parts are done
• xdg-desktop-portal receives that fd and sends it back as reply to the OpenFile() method in the normal portal
• The application receives the fd and can read the file now

Finally: because of DBus restrictions, the various methods in the portal interfaces don't just reply with values. Instead, the xdg-desktop-portal creates a new org.freedesktop.portal.Request object and returns the object path for that. Once that's done the method is complete from DBus' POV. When the actual return value arrives (e.g. the fd), that value is passed via a signal on that Request object, which is then destroyed. This roundabout way is done for purely technical reasons, regular DBus methods would time out while the user picks a file path.

Anyway. Maybe this helps someone understanding how the portal bits fit together.

[1] it does so using another portal but let's ignore that
[2] not really hot-swappable though. You need to restart xdg-desktop-portal but not your host. So luke-warm-swappable only

Edit Sep 01: clarify that it's not GTK/Qt providing the portals, but xdg-desktop-portal-gtk and -kde

Gut Ding braucht Weile. Almost three years ago, we added high-resolution wheel scrolling to the kernel (v5.0). The desktop stack however was first lagging and eventually left behind (except for an update a year ago or so, see here). However, I'm happy to announce that thanks to José Expósito's efforts, we now pushed it across the line. So - in a socially distanced manner and masked up to your eyebrows - gather round children, for it is storytime.

Historical History

In the beginning, there was the wheel detent. Or rather there were 24 of them, dividing a 360 degree [1] movement of a wheel into a neat set of 15 clicks. libinput exposed those wheel clicks as part of the "pointer axis" namespace and you could get the click count with libinput_event_pointer_get_axis_discrete() (announced here). The degree value is exposed as libinput_event_pointer_get_axis_value(). Other scroll backends (finger-scrolling or button-based scrolling) expose the pixel-precise value via that same function.

In a "recent" Microsoft Windows version (Vista!), MS added the ability for wheels to trigger more than 24 clicks per rotation. The MS Windows API now treats one "traditional" wheel click as a value of 120, anything finer-grained will be a fraction thereof. You may have a mouse that triggers quarter-wheel clicks, each sending a value of 30. This makes for smoother scrolling and is supported(-ish) by a lot of mice introduced in the last 10 years [2]. Obviously, three small scrolls are nicer than one large scroll, so the UX is less bad than before.

Now it's time for libinput to catch up with Windows Vista! For $reasons, the existing pointer axis API could get changed to accommodate for the high-res values, so a new API was added for scroll events. Read on for the details, you will believe what happens next.

Out with the old, in with the new

As of libinput 1.19, libinput has three new events: LIBINPUT_EVENT_POINTER_SCROLL_WHEEL, LIBINPUT_EVENT_POINTER_SCROLL_FINGER, and LIBINPUT_EVENT_POINTER_SCROLL_CONTINUOUS. These events reflect, perhaps unsuprisingly, scroll movements of a wheel, a finger or along a continuous axis (e.g. button scrolling). And they replace the old event LIBINPUT_EVENT_POINTER_AXIS. Those familiar with libinput will notice that the new event names encode the scroll source in the event name now. This makes them slightly more flexible and saves callers an extra call.

In terms of actual API, the new events have two new functions: libinput_event_pointer_get_scroll_value(). For the FINGER and CONTINUOUS events, the value returned is in "pixels" [3]. For the new WHEEL events, the value is in degrees. IOW this is a drop-in replacement for the old libinput_event_pointer_get_axis_value() function. The second call is libinput_event_pointer_get_scroll_value_v120() which, for WHEEL events, returns the 120-based logical units the kernel uses as well. libinput_event_pointer_has_axis() returns true if the given axis has a value, just as before. With those three calls you now get the data for the new events.

Backwards compatibility

To ensure backwards compatibility, libinput generates both old and new events so the rule for callers is: if you want to support the new events, just ignore the old ones completely. libinput also guarantees new events even on pre-5.0 kernels. This makes the old and new code easy to ifdef out, and once you get past the immediate event handling the code paths are virtually identical.

When, oh when?

These changes have been merged into the libinput main branch and will be part of libinput 1.19. Which is due to be released over the next month or so, so feel free to work backwards from that for your favourite distribution.

Having said that, libinput is merely the lowest block in the Jenga tower that is the desktop stack. José linked to the various MRs in the upstream libinput MR, so if you're on your seat's edge waiting for e.g. GTK to get this, well, there's an MR for that.

[1] That's degrees of an angle, not Fahrenheit
[2] As usual, on a significant number of those you'll need to know whatever proprietary protocol the vendor deemed to be important IP. Older MS mice stand out here because they use straight HID.
[3] libinput doesn't really have a concept of pixels, but it has a normalized pixel that movements are defined as. Most callers take that as real pixels except for the high-resolution displays where it's appropriately scaled.

Zink Is Over

A while ago I blogged about finishing up ES 3.2. Then I didn’t mention it again because…well, I suppose I’ve only blogged four times since then, but I’m going to pretend this was part of my master plan to make everyone forget so I could build hype again.

The hype is here: Zink can now* run ES 3.2 apps.

• “now” is a variable unit of time subject to CI not trying to drown itself at the nearest pub, literally melt itself to slag, or hurl itself off a cliff the instant daniels takes his eyes off it

What Does This Mean For The Future?

I know I’ve said this a few times previously, and we all had a good laugh, but this time I mean it.

Zink is done.

The final boss has been beaten, there’s no more versions to support, no extensions left on my todo list, definitely no bugs remaining, and performance can’t possibly improve further.

If you think you’ve found a zink bug, report it to whoever wrote the test or app you’re running, because the only thing I plan on doing for the rest of 2021 is playing Cyberpunk 2077 on Lavapipe.

Right after it finishes loading.

The Struggle Continues

Everyone’s seen the Phoronix benchmark numbers by now, and though there’s a lot of confusion over how to calculate the percentage increase between “game does did not run a year ago” and “game runs”, it seems like a couple people out there at Big Triangle are starting to take us seriously.

With that said, even my parents are asking me what the deal is with this one result in particular:

Performance isn’t supposed to go down. Everyone knows this. The version numbers go up and so does the performance as long as it’s not Javascript-based.

Enraged, I sprinted to my computer and searched for tesseract game, which gave me the entirely wrong result, but I eventually did manage to find the right one. I fired up zink-wip, certain that this would end up being some bug I’d already fixed.

Unfortunately, this was not the case.

I vowed not to sleep, rebase, leave my office, or even run another application until this was resolved, so you can imagine how pleased I am to be writing this post after spending way too much time getting to the bottom of everything.

Speculation Interlude

Full disclosure: I didn’t actually check why performance went down. I’m pretty sure it’s just the result of having improved buffer mapping to be better in most cases, which ended up hurting this case.

But Why

…is the performance so bad?

A quick profiling revealed that this was down to a Gallium component called vbuf, used for translating vertex buffers and attributes from the ones specified by the application to ones that drivers can actually support. The component itself is fine, the problem was that, ideally, it’s not something you ever want to be hitting when you want performance.

Consider the usual sequence of drawing a frame:

• generate and upload vertex data
• bind some descriptors
• maybe throw in a query or two if you need some spice
• draw
• repeat until frame is done

This is all great and normal, but what would happen—just hypothetically of course—if instead it looked like this:

• generate and upload vertex data
• stall and read vertex data
• rewrite vertex data in another format and reupload
• bind some descriptors
• maybe throw in a query or two if you need some spice
• draw
• repeat until frame is done

Suddenly the driver is now stalling multiple times per frame on top of doing lots of CPU work!

Incidentally, this is (almost certainly) why performance appeared to have regressed: the vertex buffer is now device-local and can’t be mapped directly, so it has to be copied to a new buffer before it can be read, which is even slower.

Just AMD Problems

DISCLAIMER: We’re going deep into meme territory now, so let’s all dial down the seriousness about a thousand notches before posting about how much I hate AMD or whatever.

Unlike cool hardware, AMD opts to not support features which might be less performant. I assume this is in the hopes that developers will Make The Right Choice and not use those features, but obviously developers are gonna develop, and so it is that Tesseract-The-Game-But-Not-The-One-On-Steam uses 3-component vertex attributes that aren’t supported by AMD hardware, necessitating the use of vbuf to translate them to 4-component attributes that can be safely used.

Decomposition

The vertex buffer format at work here was R8G8B8_SNORM, which is a perfectly cromulent format as long as you hate yourself. A shader would read this as a vec4, which, by the power of buffer robustness, gets translated to vec4(x, y, z, 1.0) because the w component is missing.

The approach I took to solving this was to decompose the vertex attribute into three separate R8_SNORM attributes, as this single-component format is wimpy enough for AMD to handle. Thus, a vertex input state containing three separate attributes including this one would now contain five, as the original R8G8B8_SNORM one is split into three, each reading a single component at an offset to simulate the original attribute.

The tricky part to this is that it requires a vertex shader prolog and variant in order to successfully split the shader’s input in such a way that the read value is the same. It also requires a NIR pass. Let’s check out the NIR pass since this blog has gone for way too long without seeing any real work:

struct decompose_state {
  nir_variable **split;
  bool needs_w;
};

static bool
decompose_attribs(nir_shader *nir, uint32_t decomposed_attrs, uint32_t decomposed_attrs_without_w)
{
   uint32_t bits = 0;
   nir_foreach_variable_with_modes(var, nir, nir_var_shader_in)
      bits |= BITFIELD_BIT(var->data.driver_location);
   bits = ~bits;
   u_foreach_bit(location, decomposed_attrs | decomposed_attrs_without_w) {
      nir_variable *split[5];
      struct decompose_state state;
      state.split = split;
      nir_variable *var = nir_find_variable_with_driver_location(nir, nir_var_shader_in, location);
      assert(var);
      split[0] = var;
      bits |= BITFIELD_BIT(var->data.driver_location);
      const struct glsl_type *new_type = glsl_type_is_scalar(var->type) ? var->type : glsl_get_array_element(var->type);
      unsigned num_components = glsl_get_vector_elements(var->type);
      state.needs_w = (decomposed_attrs_without_w & BITFIELD_BIT(location)) != 0 && num_components == 4;
      for (unsigned i = 0; i < (state.needs_w ? num_components - 1 : num_components); i++) {
         split[i+1] = nir_variable_clone(var, nir);
         split[i+1]->name = ralloc_asprintf(nir, "%s_split%u", var->name, i);
         if (decomposed_attrs_without_w & BITFIELD_BIT(location))
            split[i+1]->type = !i && num_components == 4 ? var->type : new_type;
         else
            split[i+1]->type = new_type;
         split[i+1]->data.driver_location = ffs(bits) - 1;
         bits &= ~BITFIELD_BIT(split[i+1]->data.driver_location);
         nir_shader_add_variable(nir, split[i+1]);
      }
      var->data.mode = nir_var_shader_temp;
      nir_shader_instructions_pass(nir, lower_attrib, nir_metadata_dominance, &state);
   }
   nir_fixup_deref_modes(nir);
   NIR_PASS_V(nir, nir_remove_dead_variables, nir_var_shader_temp, NULL);
   optimize_nir(nir);
   return true;
}

First, the base of the pass; two masks are provided, one for attributes that are being fully split (i.e., four components) and one for attributes that have fewer than four components and thus need to have a w component added, as in the Tesseract case. Each variable in the mask is split into four, with slightly different behavior for the ones needing a w and the ones that don’t.

The new variables are all given new driver locations matching the ones given to the split attributes for the vertex input pipeline state, and the decompose_state is passed along to the per-instruction part of the pass:

static bool
lower_attrib(nir_builder *b, nir_instr *instr, void *data)
{
   struct decompose_state *state = data;
   nir_variable **split = state->split;
   if (instr->type != nir_instr_type_intrinsic)
      return false;
   nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
   if (intr->intrinsic != nir_intrinsic_load_deref)
      return false;
   nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
   nir_variable *var = nir_deref_instr_get_variable(deref);
   if (var != split[0])
      return false;
   unsigned num_components = glsl_get_vector_elements(split[0]->type);
   b->cursor = nir_after_instr(instr);
   nir_ssa_def *loads[4];
   for (unsigned i = 0; i < (state->needs_w ? num_components - 1 : num_components); i++)
      loads[i] = nir_load_deref(b, nir_build_deref_var(b, split[i+1]));
   if (state->needs_w) {
      loads[3] = nir_channel(b, loads[0], 3);
      loads[0] = nir_channel(b, loads[0], 0);
   }
   nir_ssa_def *new_load = nir_vec(b, loads, num_components);
   nir_ssa_def_rewrite_uses(&intr->dest.ssa, new_load);
   nir_instr_remove_v(instr);
   return true;
}

The existing variable is passed along with the new variable array. Where the original is loaded, instead the new variables are all loaded in sequence and assembled into a vec matching the length of the original one. For attributes needing a w component, the first new variable is loaded as a vec4 so that the w component can be reused naturally. Then the original load instruction is removed, and with it, the original variable and its brokenness.

Immediate Results

Sort of.

The frames were definitely there, but the graphics…

Occlusion Queries

It turns out there’s almost zero coverage for occlusion queries in Vulkan’s CTS. There’s surprisingly little coverage for most query-related things, in fact, which means it wasn’t too surprising when it turned out that there were RADV query bugs at play. What was surprising was how they manifested, but that was about par for anything that reads garbage memory.

A simple one-liner later (just kidding, this fucken thing took like 4 days to find) and, magically, things were happening:

We Did It.

A big thanks to Bas Nieuwenhuizen for consulting along the way even despite being so busy getting a RADV raytracing MR up and, as always, preparing his next blog post.

A year ago, I first announced libei - a library to support emulated input. After an initial spurt of development, it was left mostly untouched until a few weeks ago. Since then, another flurry of changes have been added, including some initial integration into GNOME's mutter. So, let's see what has changed.

A Recap

First, a short recap of what libei is: it's a transport layer for emulated input events to allow for any application to control the pointer, type, etc. But, unlike the XTEST extension in X, libei allows the compositor to be in control over clients, the devices they can emulate and the input events as well. So it's safer than XTEST but also a lot more flexible. libei already supports touch and smooth scrolling events, something XTest doesn't have or is struggling with.

Terminology refresher: libei is the client library (used by an application wanting to emulate input), EIS is the Emulated Input Server, i.e. the part that typically runs in the compositor.

Server-side Devices

So what has changed recently: first, the whole approach has flipped on its head - now a libei client connects to the EIS implementation and "binds" to the seats the EIS implementation provides. The EIS implementation then provides input devices to the client. In the simplest case, that's just a relative pointer but we have capabilities for absolute pointers, keyboards and touch as well. Plans for the future is to add gestures and tablet support too. Possibly joysticks, but I haven't really thought about that in detail yet.

So basically, the initial conversation with an EIS implementation goes like this:

• Client: Hello, I am $NAME
• Server: Hello, I have "seat0" and "seat1"
• Client: Bind to "seat0" for pointer, keyboard and touch
• Server: Here is a pointer device
• Server: Here is a keyboard device
• Client: Send relative motion event 10/2 through the pointer device

One of the design choices for libei is that devices are effectively static. If something changes on the EIS side, the device is removed and a new device is created with the new data. This applies for example to regions and keymaps (see below), so libei clients need to be able to re-create their internal states whenever the screen or the keymap changes.

Device Regions

Devices can now have regions attached to them, also provided by the EIS implementation. These regions define areas reachable by the device and are required for clients such as Barrier. On a dual-monitor setup you may have one device with two regions or two devices with one region (representing one monitor), it depends on the EIS implementation. But either way, as libei client you will know that there is an area and you will know how to reach any given pixel on that area. Since the EIS implementation decides the regions, it's possible to have areas that are unreachable by emulated input (though I'm struggling a bit for a real-world use-case).

So basically, the conversation with an EIS implementation goes like this:

• Client: Hello, I am $NAME
• Server: Hello, I have "seat0" and "seat1"
• Client: Bind to "seat0" for absolute pointer
• Server: Here is an abs pointer device with regions 1920x1080@0,0, 1080x1920@1920,0
• Server: Here is an abs pointer device with regions 1920x1080@0,0
• Server: Here is an abs pointer device with regions 1080x1920@1920,0
• Client: Send abs position 100/100 through the second device

Perhaps unsurprisingly, the use of regions make libei clients windowing-system independent. The Barrier EI support WIP no longer has any Wayland-specific code in it. In theory, we could implement EIS in the X server and libei clients would work against that unmodified.

Keymap handling

The keymap handling has been changed so the keymap too is provided by the EIS implementation now, effectively in the same way as the Wayland compositor provides the keymap to Wayland clients. This means a client knows what keycodes to send, it can handle the state to keep track of things, etc. Using Barrier as an example again - if you want to generate an "a", you need to look up the keymap to figure out which keycode generates an A, then you can send that through libei to actually press the key.

Admittedly, this is quite messy. XKB (and specifically libxkbcommon) does not make it easy to go from a keysym to a key code. The existing Barrier X code is full of corner-cases with XKB already, I espect those to be necessary for the EI support as well.

Scrolling

Scroll events have four types: pixel-based scrolling, discrete scrolling, and scroll stop/cancel events. The first should be obvious, discrete scrolling is for mouse wheels. It uses the same 120-based API that Windows (and the kernel) use, so it's compatible with high-resolution wheel mice. The scroll stop event notifies an EIS implementation that the scroll interaction has stopped (e.g. lifting fingers off) which in turn may start kinetic scrolling - just like the libinput/Wayland scroll stop events. The scroll cancel event notifies the EIS implementation that scrolling really has stopped and no kinetic scrolling should be triggered. There's no equivalent in libinput/Wayland for this yet but it helps to get the hook in place.

Emulation "Transactions"

This has fairly little functional effect, but interactions with an EIS server are now sandwiched in a start/stop emulating pair. While this doesn't matter for one-shot tools like xdotool, it does matter for things like Barrier which can send the start emulating event when the pointer enters the local window. This again allows the EIS implementation to provide some visual feedback to the user. To correct the example from above, the sequence is actually:

• ...
• Server: Here is a pointer device
• Client: Start emulating
• Client: Send relative motion event 10/2 through the pointer device
• Client: Send relative motion event 1/4 through the pointer device
• Client: Stop emulating

Properties

Finally, there is now a generic property API, something copied from PipeWire. Properties are simple key/value string pairs and cover those things that aren't in the immediate API. One example here: the portal can set things like "ei.application.appid" to the Flatpak's appid. Properties can be locked down and only libei itself can set properties before the initial connection. This makes them reliable enough for the EIS implementation to make decisions based on their values. Just like with PipeWire, the list of useful properties will grow over time. it's too early to tell what is really needed.

Repositories

Now, for the actual demo bits: I've added enough support to Barrier, XWayland, Mutter and GNOME Shell that I can control a GNOME on Wayland session through Barrier (note: the controlling host still needs to run X since we don't have the ability to capture input events under Wayland yet). The keymap handling in Barrier is nasty but it's enough to show that it can work.

GNOME Shell has a rudimentary UI, again just to show what works:

Note how xdotool is listed in this screenshot: that tool is unmodified, it's the XWayland libei implementation that allows it to work and show up correctly

The various repositories are in the "wip/ei" branch of:

• XWayland
• Barrier
• Mutter
• GNOME Shell

Where to go from here? The last weeks were driven by rapid development, so there's plenty of test cases to be written to make sure the new code actually works as intended. That's easy enough. Looking at the Flatpak integration is another big ticket item, once the portal details are sorted all the pieces are (at least theoretically) in place. That aside, improving the integrations into the various systems above is obviously what's needed to get this working OOTB on the various distributions. Right now it's all very much in alpha stage and I could use help with all of those (unless you're happy to wait another year or so...). Do ping me if you're interested to work on any of this.

We Back

Just a quick update today while I dip my toes back into the blogosphere to remind myself that it’s not so scary.

Remember when I blogged about how nice it would be to have a suballocator all those months ago?

Now it’s landed, and it’s nice indeed to have a suballocator.

Remember when everyone wanted GL 4.6 compatibility contexts so they could play Feral ports of their favorite games? zink-wip did that 6 months ago.

What this all means is that it’s more or less open testing season on zink. I’ve already got a sizable number of tickets open for various Steam games based on zink-wip testing, but this is hardly conclusive.

What games work for you?

What games don’t work?

I’m not gonna test them all myself, so get out there and start crashing.

Here I’m playing “Spelunky 2” on my laptop and simultaneously replaying the same Vulkan calls on an ARM board with Adreno GPU running the open source Turnip Vulkan driver. Hint: it’s an x64 Windows game that doesn’t run on ARM.

The bottom right is the game I’m playing on my laptop, the top left is GFXReconstruct immediately replaying Vulkan calls from the game on ARM board.

How is it done? And why would it be useful for debugging? Read below!

Debugging issues a driver faces with real-world applications requires the ability to capture and replay graphics API calls. However, for mobile GPUs it becomes even more challenging since for Vulkan driver the main “source” of real-world workload are x86-64 apps that run via Wine + DXVK, mainly games which were made for desktop x86-64 Windows and do not run on ARM. Efforts are being made to run these apps on ARM but it is still work-in-progress. And we want to test the drivers NOW.

The obvious solution would be to run those applications on an x86-64 machine capturing all Vulkan calls. Then replaying those calls on a second machine where we cannot run the app. This way it would be possible to test the driver even without running the application directly on it.

The main trouble is that Vulkan calls made on one GPU + Driver combo are not generally compatible with other GPU + Driver combo, sometimes even for one GPU vendor. There are different memory capabilities (VkPhysicalDeviceMemoryProperties), different memory requirements for buffer and images, different extensions available, and different optional features supported. It is easier with OpenGL but there are also some incompatibilities there.

There are two open-source vendor-agnostic tools for capturing Vulkan calls: RenderDoc (captures single frame) and GFXReconstruct (captures multiple frames). RenderDoc at the moment isn’t suitable for the task of capturing applications on desktop GPUs and replaying on mobile because it doesn’t translate memory type and requirements (see issue #814). GFXReconstruct on the other hand has the necessary features for this.

I’ll show a couple of tricks with GFXReconstruct I’m using to test things on Turnip.

Capturing with GFXReconstruct

At this point you either have the application itself or, if it doesn’t use Vulkan, a trace of its calls that could be translated to Vulkan. There is a detailed instruction on how to use GFXReconstruct to capture a trace on desktop OS. However there is no clear instruction of how to do this on Android (see issue #534), fortunately there is one in Android’s documentation:

For Android 9 you should copy layers to the application which will be traced
For Android 10+ it's easier to copy them to com.lunarg.gfxreconstruct.replay
You should have userdebug build of Android or probably rooted Android

# Push GFXReconstruct layer to the device
adb push libVkLayer_gfxreconstruct.so /sdcard/

# Since there is to APK for capture layer,
# copy the layer to e.g. folder of com.lunarg.gfxreconstruct.replay
adb shell run-as com.lunarg.gfxreconstruct.replay cp /sdcard/libVkLayer_gfxreconstruct.so .

# Enable layers
adb shell settings put global enable_gpu_debug_layers 1

# Specify target application
adb shell settings put global gpu_debug_app <package_name>

# Specify layer list (from top to bottom)
adb shell settings put global gpu_debug_layers VK_LAYER_LUNARG_gfxreconstruct

# Specify packages to search for layers
adb shell settings put global gpu_debug_layer_app com.lunarg.gfxreconstruct.replay

adb shell "setprop debug.gfxrecon.capture_file '/data/data/<target_app_folder>/files/'"

However, when trying to replay the trace you captured on another GPU - most likely it will result in an error:

[gfxrecon] FATAL - API call vkCreateDevice returned error value VK_ERROR_EXTENSION_NOT_PRESENT that does not match the result from the capture file: VK_SUCCESS.  Replay cannot continue.
Replay has encountered a fatal error and cannot continue: the specified extension does not exist

Or other errors/crashes. Fortunately we could limit the capabilities of desktop GPU with VK_LAYER_LUNARG_device_simulation

VK_LAYER_LUNARG_device_simulation when simulating another GPU should be told to intersect the capabilities of both GPUs, making the capture compatible with both of them. This could be achieved by recently added environment variables:

VK_DEVSIM_MODIFY_EXTENSION_LIST=whitelist
VK_DEVSIM_MODIFY_FORMAT_LIST=whitelist
VK_DEVSIM_MODIFY_FORMAT_PROPERTIES=whitelist

whitelist name is rather confusing because it’s essentially means “intersection”.

One would also need to get a json file which describes target GPU capabilities, this should be done by running:

vulkaninfo -j &> <device_name>.json

The final command to capture a trace would be:

VK_LAYER_PATH=<path/to/device-simulation-layer>:<path/to/gfxreconstruct-layer> \
VK_INSTANCE_LAYERS=VK_LAYER_LUNARG_gfxreconstruct:VK_LAYER_LUNARG_device_simulation \
VK_DEVSIM_FILENAME=<device_name>.json \
VK_DEVSIM_MODIFY_EXTENSION_LIST=whitelist \
VK_DEVSIM_MODIFY_FORMAT_LIST=whitelist \
VK_DEVSIM_MODIFY_FORMAT_PROPERTIES=whitelist \
<the_app>

Replaying with GFXReconstruct

gfxrecon-replay -m rebind --skip-failed-allocations <trace_name>.gfxr

• -m Enable memory translation for replay on GPUs with memory types that are not compatible with the capture GPU’s
  • rebind Change memory allocation behavior based on resource usage and replay memory properties. Resources may be bound to different allocations with different offsets.
• --skip-failed-allocations skip vkAllocateMemory, vkAllocateCommandBuffers, and vkAllocateDescriptorSets calls that failed during capture

Without these options replay would fail.

Now you could easily test any app/game on your ARM board, if you have enough RAM =) I even successfully ran a capture of “Metro Exodus” on Turnip.

But what if I want to test something that requires interactivity?

Or you don’t want to save a huge trace on disk, which could grow tens of gigabytes if application is running for considerable amount of time.

During the recording GFXReconstruct just appends calls to a file, there are no additional post-processing steps. Given that the next logical step is to just skip writing to a disk and send Vulkan calls over the network!

This would allow us to interact with the application and immediately see the results on another device with different GPU. And so I hacked together a crude support of over-the-network replay.

The only difference with ordinary tracing is that now instead of file we have to specify a network address of the target device:

VK_LAYER_PATH=<path/to/device-simulation-layer>:<path/to/gfxreconstruct-layer> \
    ...
GFXRECON_CAPTURE_FILE="<ip>:<port>" \
<the_app>

And on the target device:

while true; do gfxrecon-replay -m rebind --sfa ":<port>"; done

Why while true? It is common for DXVK to call vkCreateInstance several times leading to the creation of several traces. When replaying over the network we therefor want gfxrecon-replay to immediately restart when one trace ends to be ready for another.

You may want to bring the FPS down to match the capabilities of lower power GPU in order to prevent constant hiccups. It could be done either with libstrangle or with mangohud:

• stranglevk -f 10
• MANGOHUD_CONFIG=fps_limit=10 mangohud

You have seen the result at the start of the post.

I’ve been silent here for quite some time, so here is a quick summary of some of the new functionality we have been exposing in V3DV, the Vulkan driver for Raspberry PI 4, over the last few months:

• VK_KHR_bind_memory2
• VK_KHR_copy_commands2
• VK_KHR_dedicated_allocation
• VK_KHR_descriptor_update_template
• VK_KHR_device_group
• VK_KHR_device_group_creation
• VK_KHR_external_fence
• VK_KHR_external_fence_capabilities
• VK_KHR_external_fence_fd
• VK_KHR_external_semaphore
• VK_KHR_external_semaphore_capabilities
• VK_KHR_external_semaphore_fd
• VK_KHR_get_display_properties2
• VK_KHR_get_memory_requirements2
• VK_KHR_get_surface_capabilities2
• VK_KHR_image_format_list
• VK_KHR_incremental_present
• VK_KHR_maintenance2
• VK_KHR_maintenance3
• VK_KHR_multiview
• VK_KHR_relaxed_block_layout
• VK_KHR_sampler_mirror_clamp_to_edge
• VK_KHR_storage_buffer_storage_class
• VK_KHR_uniform_buffer_standard_layout
• VK_KHR_variable_pointers
• VK_EXT_custom_border_color
• VK_EXT_external_memory_dma_buf
• VK_EXT_index_type_uint8
• VK_EXT_physical_device_drm

Besides that list of extensions, we have also added basic support for Vulkan subgroups (this is a Vulkan 1.1 feature) and Geometry Shaders (we use this to implement multiview).

I think we now meet most (if not all) of the Vulkan 1.1 mandatory feature requirements, but we still need to check this properly and we also need to start doing Vulkan 1.1 CTS runs and fix test failures. In any case, the bottom line is that Vulkan 1.1 should be fairly close now.

Just about a year after the original announcement, I think it's time to see the progress on power-profiles-daemon.

Note that I would still recommend you read the up-to-date project README if you have questions about why this project was necessary, and why a new project was started rather than building on an existing one.

 The project was born out of the need to make a firmware feature available to end-users for a number of lines of Lenovo laptops for them to be fully usable on Fedora. For that, I worked with Mark Pearson from Lenovo, who wrote the initial kernel support for the feature and served as our link to the Lenovo firmware team, and Hans de Goede, who worked on making the kernel interfaces more generic.

More generic, but in a good way

 With the initial kernel support written for (select) Lenovo laptops, Hans implemented a more generic interface called platform_profile. This interface is now the one that power-profiles-daemon will integrate with, and means that it also supports a number of Microsoft Surface, HP, Lenovo's own Ideapad laptops, and maybe Razer laptops soon.

 The next item to make more generic is Lenovo's "lap detection" which still relies on a custom driver interface. This should be soon transformed into a generic proximity sensor, which will mean I get to work some more on iio-sensor-proxy.

Working those interactions

 power-profiles-dameon landed in a number of distributions, sometimes enabled by default, sometimes not enabled by default (sigh, the less said about that the better), which fortunately meant that we had some early feedback available.

 The goal was always to have the user in control, but we still needed to think carefully about how the UI would look and how users would interact with it when a profile was temporarily unavailable, or the system started a "power saver" mode because battery was running out.

 The latter is something that David Redondo's work on the "HoldProfile" API made possible. Software can programmatically switch to the power-saver or performance profile for the duration of a command. This is useful to switch to the Performance profile when running a compilation (eg. powerprofilesctl jhbuild --no-interact build gnome-shell), or for gnome-settings-daemon to set the power-saver profile when low on battery.

 The aforementioned David Redondo and Kai Uwe Broulik also worked on the KDE interface to power-profiles-daemon, as Florian Müllner implemented the gnome-shell equivalent.

 I took this opportunity to update the Power panel in Settings, which shows off the temporary switch to the performance mode, and the setting to automatically switch to power-saver when low on battery.

 I've been chasing a crocus misrendering bug show in a qt trace.

The bottom image is crocus vs 965 on top. This only happened on Gen4->5, so Ironlake and GM45 were my test machines. I burned a lot of time trying to work this out. I trimmed the traces down, dumped a stupendous amount of batchbuffers, turned off UBO push constants, dump all the index and vertex buffers, tried some RGBx changes, but nothing was rushing to hit me, except that the vertex shaders produced were different.

However they were different for many reasons, due to the optimization pipelines the mesa state tracker runs vs the 965 driver. Inputs and UBO loads were in different places so there was a lot of noise in the shaders.

I ported the trace to a piglit GL application so I could easier hack on the shaders and GL, with that I trimmed it down even further (even if I did burn some time on a misplace */+ typo).

Using the ported app, I removed all uniform buffer loads and then split the vertex shader in half (it was quite large, but had two chunks). I finally then could spot the difference in the NIR shaders.

What stood out was the 965 shader had an if which the crocus shader has converted to a bcsel. This is part of peephole optimization and the mesa/st calls it, and sure enough removing that call fixed the rendering, but why? it is a valid optimization.

In a parallel thread on another part of the planet, Ian Romanick filed a MR to mesa https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12191 fixing a bug in the gen4/5 fs backend with conditional selects. This was something he noticed while debugging elsewhere. However his fix was for the fragment shader backend, and my bug was in the vec4 vertex shader backend. I tracked down where the same changes were needed in the vec4 backend and tested a fix on top of his branch, and the misrendering disappeared.

It's a strange coincidence we both started hitting the same bug in different backends in the same week via different tests, but he's definitely saved me a lot of pain in working this out! Hopefully we can combine them and get it merged this week.

Also thanks to Angelo on the initial MR for testing crocus with some real workloads.

Deeper Into Software

I don’t feel like blogging about zink today, so here’s more about everyone’s favorite software implementation of Vulkan.

The existing LLVMpipe architecture works like this from a top-down view:

• mesa / st - this is the GL/Gallium state tracker
• llvmpipe - this is the Gallium driver
• gallivm - this is the LLVM program compiler
• llvm - this is where the fragment shader runs

In short, everything is for the purpose of compiling LLVM programs which will draw/compute the desired result.

Lavapipe makes a slight change:

• lavapipe - this is the Vulkan state tracker
• llvmpipe - this is the Gallium driver
• gallivm - this is the LLVM program compiler
• llvm - this is where the fragment shader runs

It’s that simple.

Thus, any time a new feature is added to Lavapipe, what’s actually being done is plumbing that Vulkan feature through some number of layers to change how LLVM is executed. Some features, like samplerAnisotropy, require significant work at the gallivm layer just to toggle a boolean flag at the lavapipe level.

Other changes, like KHR_timeline_semaphores are entirely contained in Lavapipe.

What Are Timeline Semaphores?

Vulkan has a number of mechanisms for synchronization including fences, events, and binary semaphores, all of which serve a specific purpose. For more conrete on all of them, please read the blog of an actual expert.

The best and most awesome (don’t @ me, it’s not debatable) of these synchronization methods, however is the timeline semaphore.

A timeline semaphore is an object that can be used to signal and wait on specific integer-assigned points in command execution, also known as timelines. Each queue submission can be accompanied by an array of timeline semaphores to wait on and an array to signal; command buffers in a given submission will wait before executing, then signal after they’re done. This enables parallel code design where one thread can assemble command buffers and submit them, and the GPU can be made to pause at certain points for buffers/images referenced to become populated by another thread before continuing with execution.

Typically, semaphores are managed through signals which pass through the kernel and hardware, meaning that “waiting” on a timeline is really just waiting on an ioctl (DRM_IOCTL_SYNCOBJ_TIMELINE_WAIT) to signal that the specified timeline id has occurred, which requires no additional host-side synchronization. Things get a bit trickier in software, however, as the kernel is not involved, so everything must be managed in the driver.

Lavapipe And Timelines

This was a todo item sitting on the list for a while because it was tricky to handle. The most visible problems here were:

• connecting timeline identifiers with queue submissions; timelines only need to be monotonic, not sequential, meaning that using something like a sliding array wouldn’t be very efficient
• the actual synchronization when threads are involved

After some thought and deliberation about my life choices up to this point, I decided to tackle this implementation. The methodology I selected was to add a monotonic counter to the internal command buffer submission and then create a series of per-object timeline “links” which would serve to match the counter to the timeline identifier. This would enable each timeline semaphore to maintain a singly-linked list of links each time they were submitted, and the list could then be pruned at any given time—referenced against the internal counter—to update the “current” timeline id and then evaluate whether a specified wait condition had passed. In the case where the condition had not passed, the timeline link could also store a handle to the fence from the llvmpipe queue submission that could be waited on directly.

Did it work?

Almost on the first try, actually.

But then I ran into a wall in CI while running piglit tests through zink.

It turns out that the CTS tests are considerably less aggressive than the piglit ones for things like this: specifically, there don’t appear to be any cases where a single timeline has 16 threads all trying to wait on it at different values, iterating thousands of times over the course of a couple seconds.

Oops.

But that’s now taken care of, and conformance never felt so good.

The road to Vulkan 1.2 continues!

This is a title

I’m back.

Where did I go?

My birthday passed recently, so I gifted myself a couple weeks off from blogging. Feels good.

For today, this is a Lavapipe blog.

What’s New With Lavapipe?

Lots.

Let’s check out what conformant features were added just in July:

• EXT_line_rasterization
• EXT_vertex_input_dynamic_state
• EXT_extended_dynamic_state2
• EXT_color_write_enable
• features.strictLines
• features.shaderStorageImageExtendedFormats
• features.shaderStorageImageReadWithoutFormat
• features.samplerAnisotropy
• KHR_timeline_semaphores

Also under the hood now is a new 2D rasterizer from VMWare which yields “a 2x to 3x performance improvement for 2D workloads”.

Why Aren’t You Using Lavapipe Yet?

Have a big Vulkan-using project? Do you constantly have to worry about breakages from all manner of patches being merged without testing? Can’t afford or too lazy to set up and maintain actual hardware for testing?

Why not Lavapipe?

Seriously, why not? If there’s features missing that you need for your project, open tickets so we know what to work on.

Part 1, Part 2, Part 3

After getting thouroughly nerd-sniped a few weeks back, we now have FreeBSD support through qemu in the freedesktop.org ci-templates. This is possible through the qemu image generation we have had for quite a while now. So let's see how we can easily add a FreeBSD VM (or other distributions) to our gitlab CI pipeline:

.freebsd:
  variables:
     FDO_DISTRIBUTION_VERSION: '13.0'
     FDO_DISTRIBUTION_TAG: 'freebsd.0' # some value for humans to read
     
build-image:
  extends:
     - .freebsd
     - .fdo.qemu-build@freebsd
  variables:
    FDO_DISTRIBUTION_PACKAGES: "curl wget"

Because of the nesting, we need to handle this accordingly in our script: tag for the actual test job - we need to start the image and make sure our jobs are actually built within. The templates set up an ssh alias "vm" for this and the vmctl script helps to do things on the vm:

test-build:
  extends:
    - .freebsd
    - .fdo.distribution-image@freebsd
  script:
    # start our QEMU image
    - /app/vmctl start
    
    # copy our current working directory to the VM
    # (this is a yaml multiline command to work around the colon)
    - |
      scp -r $PWD vm:
      
    # Run the build commands on the VM and if they succeed, create a .success file
    - /app/vmctl exec "cd $CI_PROJECT_NAME; meson builddir; ninja -C builddir" && touch .success || true
    
    # Copy results back to our run container so we can include them in artifacts:
    - |
      scp -r vm:$CI_PROJECT_NAME/builddir .
      
    # kill the VM
    - /app/vmctl stop
    
    # Now that we have cleaned up: if our build job before
    # failed, exit with an error
    - [[ -e .success ]] || exit 1

Obviously, if you want to build any other distribution you just swap the freebsd out for fedora or whatever - the process is the same. libinput has been using fedora qemu images for ages now.

Thanks to the work done by Josè Expòsito, libinput 1.19 will ship with a new type of gesture: Hold Gestures. So far libinput supported swipe (moving multiple fingers in the same direction) and pinch (moving fingers towards each other or away from each other). These gestures are well-known, commonly used, and familiar to most users. For example, GNOME 40 recently has increased its use of touchpad gestures to switch between workspaces, etc. Swipe and pinch gestures require movement, it was not possible (for callers) to detect fingers on the touchpad that don't move.

This gap is now filled by Hold gestures. These are triggered when a user puts fingers down on the touchpad, without moving the fingers. This allows for some new interactions and we had two specific ones in mind: hold-to-click, a common interaction on older touchscreen interfaces where holding a finger in place eventually triggers the context menu. On a touchpad, a three-finger hold could zoom in, or do dictionary lookups, or kill a kitten. Whatever matches your user interface most, I guess.

The second interaction was the ability to stop kinetic scrolling. libinput does not actually provide kinetic scrolling, it merely provides the information needed in the client to do it there: specifically, it tells the caller when a finger was lifted off a touchpad at the end of a scroll movement. It's up to the caller (usually: the toolkit) to implement the kinetic scrolling effects. One missing piece was that while libinput provided information about lifting the fingers, it didn't provide information about putting fingers down again later - a common way to stop scrolling on other systems.

Hold gestures are intended to address this: a hold gesture triggered after a flick with two fingers can now be used by callers (read: toolkits) to stop scrolling.

Now, one important thing about hold gestures is that they will generate a lot of false positives, so be careful how you implement them. The vast majority of interactions with the touchpad will trigger some movement - once that movement hits a certain threshold the hold gesture will be cancelled and libinput sends out the movement events. Those events may be tiny (depending on touchpad sensitivity) so getting the balance right for the aforementioned hold-to-click gesture is up to the caller.

As usual, the required bits to get hold gestures into the wayland protocol are either in the works, mid-flight or merge-ready so expect this to hit the various repositories over the medium-term future.

I have not talked about raytracing in RADV for a while, but after some procrastination being focused on some other things I recently got back to it and achieved my next milestone.

In particular I have been hacking away at CTS and got to a point where CTS on dEQP-VK.ray_tracing.* runs to completion without crashes or hangs. Furthermore, I got the passrate to 90% of non-skiped tests. So we’re finally getting somewhere close to usable.

As further show that it is usable my fixes for CTS also fixed the corruption issues in Quake 2 RTX (Github version), delivering this image:

Of course not everything is perfect yet. Besides the not 100% CTS passrate it has like half the Windows performance at 4k right now and we still have some feature gaps to make it really usable for most games.

Why is it slow?

TL;DR Because I haven’t optimized it yet and implemented every shortcut imaginable.

AMD raytracing primer

Raytracing with Vulkan works with two steps:

1. You built a giant acceleration structure that contains all your geometry. Typically this ends up being some kind of tree, typically a Bounding Volume Hierarchy (BVH).
2. Then you trace rays using some traversal shader through the acceleration structure you just built.

With RDNA2 AMD started accelerating this by adding an instruction that allowed doing intersection tests between a ray and a single BVH node, where the BVH node can either be

• A triangle
• A box node specifying 4 AABB boxes

Of course this isn’t quite enough to deal with all geometry types in Vulkan so we also add two more:

• an AABB box
• an instance of another BVH combined with a transformation matrix

Building the BVH

With a search tree like a BVH it is very possibly to make trees that are very useless. As an example consider a binary search tree that is very unbalanced. We can have similarly bad things with a BVH including making it unbalanced or having overlapping bounding volumes.

And my implementation is the simplest thing possible: the input geometry becomes the leaves in exactly the same order and then internal nodes are created just as you’d draw them. That is probably decently fast in building the BVH but surely results in a terrible BVH to actually use.

BVH traversal

After we built a BVH we can start tracing some rays. In rough pseudocode the current implementation is

stack = empty
insert root node into stack
while stack is not empty:

   node = pop a node from the stack

   if we left the bottom level BVH:
      reset ray origin/direction to initial origin/direction

   result = amd_intersect(ray, node)
   switch node type:
      triangle:
         if result is a hit:
            load some node data
            process hit
      box node:
         for each box hit:
            push child node on stack
      custom node 1 (instance):
         load node data
         push the root node of the bottom BVH on the stack
         apply transformation matrix to ray origin/direction
      custom node 2 (AABB geometry):
         load node data
         process hit

We already knew there were inherently going to be some difficulties:

• We have a poor BVH so we’re going to do way more iterations than needed.
• Calling shaders as a result of hits is going to result in some divergence.

Furthermore this also clearly shows some difficulties with how we approached the intersection instruction. Some advantages of the intersection instruction are that it avoids divergence in computing collisions if we have different node types in a subgroup and to be cheaper when there are only a few lanes active. (A single CU can process one ray/node intersection per cycle, modulo memory latency, while it can process an ALU instruction on 64 lanes per cycle).

However even if it avoids the divergence in the collision computation we still introduce a ton of divergence in the processing of the results of the intersection. So we are still doing pretty bad here.

A fast GPU traversal stack needs some work too

Another thing to be noted is our traversal stack size. According to the Vulkan specification a bottom level acceleration structure should support 2^24 -1 triangles and a top level acceleration structure should support 2^24 - 1 bottom level structures. Combined with a tree with 4 children in each internal node we can end up with a tree depth of about 24 levels.

In each internal node iteration of our loop we pop one element and push up to 4 elements, so at the deepest level of traversal we could end up with a 72 entry stack. Assuming these are 32-bit node identifiers, that ends up with 288 bytes of stack per lane, or ~18 KiB per 64 lane workgroup (the minimum which could possibly keep a CU busy with an ALU only workload). Given that we have 64 KiB of LDS (yes I am using LDS since there is no divergent dynamic register addressing) per CU that leaves only 3 workgroups per CU, leaving very little options for parallelism between different hardware execution units (e.g. the ALU and the texture unit that executes the ray intersections) or latency hiding of memory operations.

So ideally we get this stack size down significantly.

Where do we go next?

First step is to get CTS passing and getting an initial merge request into upstream Mesa. As a follow on to that I’d like to get a minimal prototype going for some DXR 1.0 games with vkd3d-proton just to make sure we have the right feature coverage.

After that we’ll have to do all the traversal optimizations. I’ll probably implement a bunch of instrumentation so I actually have a clue on what to optimize. This is where having some runnable games really helps get the right idea about performance bottlenecks.

Finally, with some luck better shaders to build a BVH will materialize as well.

If you want to write an X application, you need to use some library that speaks the X11 protocol. For a long time this meant libX11, often called xlib, which - like most things about X - is a fantastic bit of engineering that is very much a product of its time with some confusing baroque bits. Overall it does a very nice job of hiding the icky details of the protocol from the application developer.

One of the details it hides has to do with how resource IDs are allocated in X. A resource ID (an XID, in the jargon) is a 32 29-bit integer that names a resource - window, colormap, what have you. Those 29 bits are split up netmask/hostmask style, where the top 8 or so uniquely identify the client, and the rest identify the resource belonging to that client. When you create a window in X, what you really tell the server is "I want a window that's initially this size, this background color (etc.) and from now on when I say (my client id + 17) I mean that window." This is great for performance because it means resource allocation is assumed to succeed and you don't have to wait for a reply from the server.

Key to all this is that in xlib the XID is the return value from the call that issues the resource creation request. Internally the request gets queued into the protocol's write buffer, but the client can march ahead and issue the next few commands as if creation had succeeded - because it probably did, and if it didn't you're probably going to crash anyway.

So to allocate XIDs the client just marches forward through its XID range. What happens when you hit the end of the range? Before X11R4, you'd crash, because xlib doesn't keep track of which XIDs it's allocated, just the lowest one it hasn't allocated yet. Starting in R4 the server added an extension called XC-MISC that lets the client ask the server for a list of unused XIDs, so when xlib hits the end of the range it can request a new range from the server.

But. UI programming tends to want threads, and xlib is perhaps not the most thread-friendly. So XCB was invented, which sacrifices some of xlib's ease of use for a more direct binding to the protocol and (in theory) an explicitly thread-safe design. We then modified xlib and XCB to coexist in the same process, using the same I/O buffers, reply and event management, etc.

This literal reflection of the protocol into the API has consequences. In XCB, unlike xlib, XID generation is an explicit step. The client first calls into XCB to allocate the XID, and then passes that XID to the creation request in order to give the resource a name.

Which... sorta ruins that whole thread-safety thing.

Let's say you call xcb_generate_id in thread A and the XID it returns is the last one in your range. Then thread B schedules in and tries to allocate another XID. You'll ask the server for a new range, but since thread A hasn't called its resource creation request yet, from the server's perspective that "allocated" XID looks like it's still free! So now, whichever thread issues their resource creation request second will get BadIDChoice thrown at them if the other thread's resource hasn't been destroyed in the interim.

A library that was supposed to be about thread safety baked a thread safety hazard into the API. Good work, team.

How do you fix this without changing the API? Maybe you could keep a bitmap on the client side that tracks XID allocation, that's only like 256KB worst case, you can grow it dynamically and most clients don't create more than a few dozen resources anyway. Make xcb_generate_id consult that bitmap for the first unallocated ID, and mark it used when it returns. Then track every resource destruction request and zero it back out of the bitmap. You'd only need XC-MISC if some other client destroyed one of your resources and you were completely out of XIDs otherwise.

And you can implement this, except. One, XCB has zero idea what a resource destruction request is, that's simply not in the protocol description. Not a big deal, you can fix that, there's only like forty destructors you'd need to annotate. But then two, that would only catch resource destruction calls that flow through XCB's protocol binding API, which xlib does not, xlib instead pushes raw data through xcb_writev. So now you need to modify every client library (libXext, libGL, ...) to inform XCB about resource destruction.

Which is doable. Tedious. But doable.

I think.

I feel a little weird writing about this because: surely I can't be the first person to notice this.

In order to expose OpenGL 4.6 the last missing feature in llvmpipe is anisotropic texture filtering. Adding support for this also allows lavapipe expose the Vulkan samplerAnisotropy feature.

I started writing anisotropic support > 6 months ago. At the time we were trying to deprecate the classic swrast driver, and someone pointed out it had support for anisotropic filtering. This support had also been ported to the softpipe driver, but never to llvmpipe.

I had also considered porting swiftshaders anisotropic support, but since I was told the softpipe code was functional and had users I based my llvmpipe port on that.

Porting the code to llvmpipe means rewriting it to generate LLVM IR using the llvmpipe vector processing code. This is a lot messier than just writing linear processing code, and when I thought I had it working it passes GL CTS, but failed the VK CTS. The results also to my eye looked worse than I'd have thought was acceptable, and softpipe seemed to be as bad.

Once I swung back around to this I decided to port the VK CTS test to GL and run it on softpipe and llvmpipe code. Initially llvmpipe had some more bugs to solve esp where the mipmap levels were being chosen, but once I'd finished aligning softpipe and llvmpipe I started digging into why the softpipe code wasn't as nice as I expected.

The softpipe code was based on an implementation of an Elliptical Weighted Average Filter (EWA). The paper "Creating Raster Omnimax Images from Multiple Perspective Views Using the Elliptical Weighted Average Filter" described this. I sat down with the paper and softpipe code and eventually found the one line where they diverged.[1] This turned out to be a bug introduced in a refactoring 5 years ago, and nobody had noticed or tracked it down.

I then ported the same fix to my llvmpipe code, and VK CTS passes. I also optimized the llvmpipe code a bit to avoid doing pointless sampling and cleaned things up. This code landed in [2] today.

For GL4.6 there are still some fixes in other areas.


[1] https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11917

[2] https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8804


After a month of reverse-engineering, we’re excited to release documentation on the Valhall instruction set, available as a PDF. The findings are summarized in an XML architecture description for machine consumption. In tandem with the documentation, we’ve developed a Valhall assembler and disassembler as a reverse-engineering aid.

Valhall is the fourth Arm® Mali™ architecture and the fifth Mali instruction set. It is implemented in the Arm® Mali™-G78, the most recently released Mali hardware, and Valhall will continue to be implemented in Mali products yet to come.

Each architecture represents a paradigm shift from the last. Midgard generalizes the Utgard pixel processor to support compute shaders by unifying the shader stages, adding general purpose memory access, and supporting integers of various bit sizes. Bifrost scalarizes Midgard, transitioning away from the fixed 4-channel vector (vec4) architecture of Utgard and Midgard to instead rely on warp-based execution for parallelism, better using the hardware on modern workloads. Valhall linearizes Bifrost, removing the Very Long Instruction Word mechanisms of its predecessors. Valhall replaces the compiler’s static scheduling with hardware dynamic scheduling, trading additional control hardware for higher average performance. That means padding with “no operation” instructions is no longer required, which may decrease code size, promising better instruction cache use.

All information in this post and the linked PDF and XML is published in good faith and for general information purpose only. We do not make any warranties about the completeness, reliability and accuracy of this information. Any action you take upon the information you find here, is strictly at your own risk. We are not be liable for any losses and/or damages in connection with the use of this information.

While we strive to make the information as accurate as possible, we make no claims, promises, or guarantees about its accuracy, completeness, or adequacy. We expressly disclaim liability for content, errors and omissions in this information.

Let’s dig in.

Getting started

In June, Collabora procured an International edition of the Samsung Galaxy S21 phone, powered by a system-on-chip with Mali G78. Although Arm announced Valhall with the Mali G77 in May 2019, roll out has been slow due to the COVID-19 pandemic. At the time of writing, there are not yet Linux friendly devices with a Valhall chip, forcing use of a locked down Android device. There’s a silver lining: we have a head start on the reverse-engineering, so by the time hacker-friendly devices arrive with Valhall GPUs, we can have open source drivers ready.

Android complicates reverse-engineering (though not as much as macOS). On Linux, we can compile a library on the device to intercept data sent to the GPU. On Android, we must cross-compile from a desktop with the Android Native Development Kit, ironically software that doesn’t run on Arm processors. Further, where on Linux we can track the standard system calls, Android device drivers replace the standard open() system call with a complicated Android-only “binder” interface. Adapting the library to support binder would be gnarly, but do we have to? We could sprinkle in one little hack anywhere we see a file descriptor without the file name.

#define MALI0 "/dev/mali0"

bool is_mali(int fd)
{
    char in[128] = { 0 }, out[128] = { 0 };
    snprintf(in, sizeof(in), "/proc/self/fd/%d", fd);

    int count = readlink(in, out, sizeof(out) - 1);
    return count == strlen(MALI0) && strncmp(out, MALI0, count) == 0;
}

Now we can hook the Mali ioctl() calls without tracing binder and easily dump graphics memory.

We’re interested in the new instruction set, so we’re looking for the compiled shader binaries in memory. There’s a chicken-and-egg problem: we need to find the shaders to reverse-engineer them, but we need to reverse-engineer the shaders to know what to look for. Fortunately, there’s an escape hatch. The proprietary Mali drivers allow an OpenGL application to query the compiled binary with the ARM_mali_program_binary extension, returning a file in the Mali Binary Shader format. That format was reverse-engineered years ago by Connor Abbott for earlier Mali architectures, and the basic structure is unchanged in Valhall. Our task is simple: compile a test shader, dump both GPU memory and the Mali Binary Shader, and find the common section. Searching for the common bytes produces an address in executable graphics memory, in this case 0x7f0002de00. Searching for that address in turn finds the “shader program descriptor” which references it.

18 00 00 80 00 10 00 00 00 DE 02 00 7F 00 00 00

Another search shows this descriptor’s address in the payload of an index-driven vertex shading job for graphics or a compute job for OpenCL. Those jobs contain the Job Manager header introduced a decade ago for Midgard, so we understand them well: they form a linked list of jobs, and only the first job is passed to the kernel. The kernel interface has a “job chain” parameter on the submit system call taking a GPU address. We understand the kernel interface well as it is open source due to kernel licensing requirements.

With each layer identified, we teach the wrapper library to chase the pointers and dump every shader executed, enabling us to reverse-engineer the new instruction set and develop a disassembler.

Instruction set reconnaissance

Reverse-engineering in the dark is possible, but it’s easier to have some light. While waiting for the Valhall phone to arrive, I read everything Arm made public about the instruction set, particularly this article from Anandtech. Without lifting a finger, that article tells us Valhall is…

• Warp-based, like Bifrost, but with 16 threads per warp instead of Bifrost’s 4/8.
• Isomorphic to Bifrost on the instruction level (“operational equivalence”).
• Regularly encoded.
• Flat, lacking Bifrost’s clause and tuple packaging.

It also says that Valhall has a 16KB instruction cache, holding 2048 instructions. Since Valhall has a regular encoding, we divide 16384 bytes by 2048 instructions to find a Valhall instruction is 8 bytes. Our first attempt at a “disassembler” can print hex dumps of every 8 bytes on a line; our calculation ensures that is the correct segmentation.

From here on, reverse-engineering is iterative. We have a baseline level of knowledge, and we want to grow that knowledge. To do so, we input test programs into the proprietary driver to observe the output, then perturbe the input program to see how the output changes.

As we discover new facts about the architecture, we update our disassembler, demonstrating new knowledge and separating the known from the unknown. Ideally, we encode these facts in a machine-readable file forming a single reference for the architecture. From this file, we can generate a disassembler, an assembler, an instruction encoder, and documentation. For Valhall, I use an XML file, resembling Bifrost’s equivalent XML.

Filling out this file is usually straightforward though tedious. Modern APIs are large, so there is a great deal of effort required to map the API requirements to the hardware features.

However, some hardware features do not map to any API. Here are subtler tales from reversing Valhall.

Dependency slots

Arithmetic is faster than memory access, so modern processors execute arithmetic in parallel with pending memory accesses. Modern GPU architectures require the compiler to manage this mechanism by analyzing the program and instructing the hardware to wait for the results before they’re needed.

For this purpose, Bifrost uses an explicit scoreboarding system. Bifrost groups up to 16 instructions together in a clause, and each clause has a fixed header. The compiler assigns a “dependency slot” between 0 and 7 to each clause, specified in the header. Each clause can wait on any set of slots, specified with another 8-bits in the clause header. Specifying dependencies per-clause is a compromise between precision and code size.

We expect Valhall to feature a similar scheme, but Valhall doesn’t have clauses or clause headers, so where does it specify this info?

Studying compiled shaders, we see the last byte of every instruction is usually zero. But when the result of a memory access is first read, the previous instruction has a bit set in the last byte. Which bit is set depends on the number of memory accesses in flight, so it seems the last byte encodes a dependency wait. The memory access instructions themselves are often zero in their last bytes, so it doesn’t look like the last byte is used to encode the dependency slot – but executing many memory access instructions at once and comparing the bits, we see a single 2-bit field stands out as differing. The dependency slot is specified inside the instruction, not in the metadata.

What makes this design practical? Two factors.

One, only the waits need to be specified in general. Arithmetic instructions don’t need a dependency slot, since they complete immediately. The longest message passing instructions is shorter than the longer arithmetic instruction, so there is space in the instruction itself to specify only when needed.

Two, the performance gain from adding extra slots levels off quickly. Valhall cuts back on Bifrost’s 8 slots (6 general purpose). Instead it has 4 or 5 slots, with only 3 general purpose, saving 4-bits for every instruction.

This story exemplifies a general pattern: Valhall is a flattening of Bifrost. Alternatively, Bifrost is “Valhall with clauses”, although that description is an anachronism. Why does Bifrost have clauses, and why does Valhall remove them? The pattern in this story of dependency waits generalizes to answer the question: grouping many instructions into Bifrost clauses allows the hardware to amortize operations like dependency waits and reduce the hardware gate count of the shader core. However, clauses add substantial encoding overhead, compiler complexity, and imprecision. Bifrost optimizes for die space; Valhall optimizes for performance.

The missing modifier

Hardware features that are unused by the proprietary driver are a perennial challenge for reverse-engineering. However, we have a complete Bifrost reference at our disposal, and Valhall instructions are usually equivalent to Bifrost. Special instructions and modes from Bifrost cast a shadow on Valhall, showing where there are gaps in our knowledge. Sometimes these gaps are impractical to close, short of brute-forcing the encoding space. Other times we can transfer knowledge and make good guesses.

Consider the Cross Lane PERmute instruction, CLPER, which takes a register and the index of another lane in the warp, and returns the value of the register in the specified lane. CLPER is a “subgroup operation”, required for Vulkan and used to implement screen-space derivatives in fragment shaders. On Bifrost, the CLPER instruction is defined as:

<ins name="+CLPER.i32" mask="0xfc000" exact="0x7c000">
  <src start="0" mask="0x7"/>
  <src start="3"/>
  <mod name="lane_op" start="6" size="2">
    <opt>none</opt>
    <opt>xor</opt>
    <opt>accumulate</opt>
    <opt>shift</opt>
  </mod>
  <mod name="subgroup" start="8" size="2">
    <opt>subgroup2</opt>
    <opt>subgroup4</opt>
    <opt>subgroup8</opt>
  </mod>
  <mod name="inactive_result" start="10" size="4">
    <opt>zero</opt>
    <opt>umax</opt>
    ....
    <opt>v2infn</opt>
    <opt>v2inf</opt>
  </mod>
</ins>

We expect a similar definition for Valhall. One modification is needed: Valhall warps contain 16 threads, so there should be a subgroup16 option after subgroup8, with the natural binary encoding 11. Looking at a binary Valhall CLPER instruction, we see a 11 pair corresponding to the subgroup field. Similarly experimenting with different subgroup operations in OpenCL lets us figure out the lane_op field. We end up with an instruction definition like:

<ins name="CLPER.u32" title="Cross-lane permute" dests="1" opcode="0xA0" opcode2="0xF">
  <src/>
  <src widen="true"/>
  <subgroup/>
  <lane_op/>
</ins>

Notice we do not specify the encoding in the Valhall XML, since Valhall encoding is regular. Also notice we lack the inactive_result modifier. On Bifrost, inactive_result specifies the value returned if the program attempts to access an inactive lane. We may guess Valhall has the same mechanism, but that modifier is not directly controllable by current APIs. How do we proceed?

If we can run code on the device, we can experiment with the instruction. Inactive lanes may be caused by divergent control flow, where one lane in the thread branches but another lane does not, forcing the hardware to execute only part of the warp. After reverse-engineering Valhall’s branch instructions, we can construct a situation where a single lane is active and the rest are inactive. Then we insert a CLPER instruction with extra bits set, store the result to main memory, and print the result. This assembly program does the trick:

# Elect a single lane
BRANCHZ.reconverge.id lane_id, offset:3

# Try to read a value from an inactive thread
CLPER.u32 r0, r0, 0x01000000.b3, inactive_result:VALUE

# Store the value
STORE.i32.slot0.reconverge @r0, u0, offset:0

# End shader
NOP.return

With the assembler we’re writing, we can assemble this compute kernel. How do we run it on the device without knowing the GPU data structures required to dispatch compute shaders? We make use of another classic reverse-engineering technique: instead of writing the initialization code ourselves, piggyback off the proprietary driver. Our wrapper library allows us to access graphics memory before the driver submits work to the hardware. We use this to read the memory, but we may also modify it. We already identified the shader program descriptor, so we can inject our own shaders. From here, we can jury-rig a script to execute arbitrary shader binaries on the device in the context of an OpenCL application running under the proprietary driver.

Putting it together, we find the inactive_result bits in the CLPER encoding and write one more script to dump all values.

for ((i = 0 ; i < 16 ; i++)); do
  sed -e "s/VALUE/$i/" shader.asm | python3 asm.py shader.bin
  adb push shader.bin /data/local/tmp/
  adb shell 'REPLACE=/data/local/tmp/shader.bin '\
    'LD_PRELOAD=/data/local/tmp/panwrap.so '\
    '/data/local/tmp/test-opencl'
done

The script’s output contains sixteen possibilities – and they line up perfectly with Bifrost’s sixteen options. Success.

Next steps

There’s more to learn about Valhall, but we’ve reverse-engineered enough to develop a Valhall compiler. As Valhall is a simplification of Bifrost, and we’ve already developed a free and open source compiler for Bifrost, this task is within reach. Indeed, adapting the Bifrost compiler to Valhall will require refactoring but little new development.

Mali G78 does bring changes beyond the instruction set. The data structures are changed to reduce Vulkan driver overhead. For example, the monolithic “Renderer State Descriptor” on Bifrost is split into a “Shader Program Descriptor” and a “Depth Stencil Descriptor”, so changes to the depth/stencil state no longer require the driver to re-emit shader state. True, the changes require more reverse-engineering. Fortunately, many data structures are adapted from Bifrost requiring few changes to the Mesa driver.

Overall, supporting Valhall in Mesa is within reach. If you’re designing a Linux-friendly device with Valhall and looking for open source drivers, please reach out!

Originally posted on Collabora’s blog

It Happened.

That’s right.

Zink(-wip) now fully supports GL_KHR_blend_equation_advanced, which means ES 3.2 is a go (once my local CI clears me to push today’s snapshot).

And all it took was one brief exchange with a top Mesa reviewer who is incidentally rumored to be undergoing training deep in the mountains to become an expert BBQ master on the extremely professional #zink channel on OFTC:

That's the thing. You can totally do it in Zink.

My mind was blown.

Why hadn’t I thought of that sooner?

I could just…do it? Just like that? And then it’d be done?

Truly the experts are on a different level from us mortals.

So now it’s done, and that means zink is finished. I don’t expect there will be any more work to do now that the final boss has been defeated. Don’t even bother trying to file bug reports.

You may not like it, but this is what peak Friday looks like.

The Unsung Heroes

This is going to be less of a technical post and more of a have you thought about post from me personally (usual disclaimer: this post represents only my views). With that said, I think this is more important than the average post here, meaning that expectations should be set somewhere between I need to stop everything else I’m doing until I finish reading and this is the most important event in my life.

Let’s talk about open source. No, Open Source. The idea of it.

How Does Open Source Work?

Those of you who are veterans are rolling your eyes. Another post about the glory of Open Source.

The thing about Open Source is that it’s sort of whatever you make of it. At its core, it’s about getting people together to solve a problem—community building. Whether that community is large or small, the goal is the same: write some quality software.

To that end, you’ve got your usual corporate powerpoint slide of community roles:

• maintainers
• developers
• reviewers
• whatever other buzzwords are currently relevant

In Mesa, the maintainer and developer roles are mostly the same among core contributors: these are the people who write the code that gets posted about on all the news sites.

The reviewer is a bit more mysterious though. Who are reviewers, and what separates them from the others?

WD-40

Reviewers are the grease that makes the project work. There’s really no other way of saying it.

Outside of a few components of Mesa that are effectively the wild west, without any form of oversight or approval needed for changes to be landed, every driver and utility in the tree requires that changes undergo review before they land. This means that each and every patch which affects code or build has to have a person stop everything else they’re doing and physically scroll through each patch, line-by-line, then add a Reviewed-by or Acked-by tag.

If you’re unclear as to the meanings of these tags, consider it like you’re going skydiving with someone you’ve never met before who has been in charge of preparing your parachute:

• Reviewed-by means “I triple-checked your parachute as well as your reserve, and I’m as certain as a human is capable of being that everything is how it should be”
• Acked-by means “Hey, I grabbed this already-packed parachute off the hanger and gave it a once-over; you’ll probably be fine”

It’s then up to the developer to decide whether to merge the code based on the feedback given to them by the reviewer.

This, of course, assumes they get feedback at all.

Balance

Too often on news sites (and in certain corporate metrics) you’ll see something like “Patches McCodesAlot, working for GreatCodingCompany, authored the most code changes for this release cycle (9001 patches), which is over 100x more than the next highest contributor.”

The manager at a company sees this and thinks “I’ll send this up the chain. We should poach Patches so we can have greater control over this project which underpins our entire business strategy. Also it’ll make my powerpoint pie charts look rad.”

The casual reader sees this and says “Wow, Patches is awesome! Without Patches, I probably couldn’t even play Fororantwatch on my Linux gaming desktop!”

But how do the patches that Patches writes get merged into the release? Unless Patches works exclusively in one of the undermaintained areas of the project, in which case it’s unlikely that their work is being widely used, the odds are that someone’s pulling a huge lift on the review side to enable all of those patches landing into the repository.

This is the job of the reviewer.

A Thanks

As this Mesa release cycle starts to wind down, I hope that readers of this blog and news sites can take a moment to look past Patches McCodesAlot and see the people who make it possible for Patches to land so many damn patches.

At the time of this post, this is what the top 10 reviewers managed to accomplish over the past few months:

Summed up, that’s over 1300 patches reviewed! For perspective, that’s around 30% of all the patches in this release, and it’s about 70% of the total number of patches that zink has received in the course of its existence.

Looking at it another way though, this is over 1300 patches that other people wrote which were able to land because these people took the time to look over the proposed changes—to triple-check the parachutes, as it were.

So thanks, Mesa reviewers. The project wouldn’t exist without all of you (and your generous employers, who should be blasting these metrics in the press when they talk about being good Open Source citizens).

But Also

I’d be remiss if I didn’t also mention the people working on Mesa CI. There’s no patch counts or review counts or anything to recognize everyone hard at work here, but CI is what keeps the triangles blasting out of your GPUs looking how they should.

Thanks, CI team. You’re awesome.

According to a recent metric, the Mesa CI infrastructure only had a 0.6% accidental failure rate. That’s pretty good considering how many thousands of jobs run every day.

This year, I decided to participate as speaker in esLibre 2021 conference. esLibre is a Spanish free software conference that covers a lot of different topics related to open-source projects: from the technical point of view to its social impact.

This year the conference had talks about game development with Godot, KDE, LibreOffice, Free Software in Universities among many others. Check out the program.

This is my first time participating in this conference and I enjoyed it a lot. Huge applause to the organization team for the huge work to organize this edition, for helping out the speakers with different testing days and for their kindness to reply any question from me and other attendees. They did a superb job!

My talk was an introduction to Mesa where I covered things like where is Mesa in the open-source graphics stack, a summary of what it does, the drivers implemented in Mesa, how our community is organized and how to contribute to it. If you know Spanish, you can check it out here (PDF). But in case you want an English version of it, this talk is very similar to the one I gave at Ubucon Europe 2018.

My esLibre talk was recorded as well! I’ll update this post with the link to the recording once it is publicly available.

Enjoy it!

BREAKING: THIS IS NO LONGER A ZINK BLOG

For today, at least.

Today, this blog is a Gallium blog. And it’s a momentous day indeed.

We all know what this is:

It’s a screenshot of Portal 2 with the Gallium HUD activated and VSync disabled.

But what driver is that underneath?

Well, for today’s blog it’s RadeonSI, the reference implementation of a Gallium driver.

And why is this, I can hear you all asking.

What if I told you that this screenshot with 10% higher FPS is also Portal 2 with VSync disabled on RadeonSI using one trick that graphics developers WON’T TELL YOU:

Interested?

Coming Soon (Maybe, And Also Maybe Requiring Some Early 2000s-era Elbow Grease From Anyone Wanting To Try): Native Linux Source Games On Gallium Nine

We did it.

By assembling an elite team of individuals with a few minutes to spare here and there over the past week, including:

• Josh Ashton, expert spammer of 🐸 emojis
• Axel Davy, expert struct packer
• Me, expert blogger
• Is Such A Thing Even Possible? Why Yes, Yes It Is.

it is now (technically) possible to run DXVK-compatible Source Engine games through Gallium’s Nine state tracker, providing a native D3D9 runtime.

Is your Portal 2 in-game FPS sad and barely even 500 like this screenshot?

Why not jack it up to more than TWICE THAT NUMBER* with riced out, GPU-fan-shredding technology that Mesa Gallium drivers have been shipping for years?

Disclaimer*

This post does not represent any form of official statement or address from Valve and is only a small project that was started out of boredom while I waited for CTS runs to finish.

This post also does not make any claims or statements regarding performance on other drivers, or performance comparisons using alternative graphics emulation layers, though whew, it sure would be interesting to see what those kinds of numbers look like!

We Did It

After months and months of the construction crews hammering away, VK_EXT_multi_draw has now been released for general use.

Will this suddenly make zink the fastest GPU driver in history?

Obviously.

Long-time readers will recall that I memed about this extension some time ago, and the numbers in a synthetic benchmark targeted at exactly this feature are phenomenal.

For more on the topic, we go to our Senior Multidraw Correspondent and my personal Khronos BFF, Ricardo Garcia, who has been following this story since the beginning.

Fast Friday

In short, an issue was filed recently about getting the Nine state tracker working with zink.

Was it the first? No..

Was it the first one this year? Yes.

Thus began a minutes-long, helter-skelter sequence of events to get Nine up and running, spread out over the course of a day or two. In need of a skilled finagler knowledgeable in the mysterium of Gallium state trackers, I contacted the only developer I know with a rockstar name, Axel Davy. We set out at dawn, and I strapped on my parachute. It was almost immediately that I heard a familiar call: there’s a build issue.

Next stop was crashing in unimplemented interface methods with a stopover in flailing about wildly in TGSI what even is this before I arrivated at my target:

Ah, glorious triangles.

I Said I Would

A long, long time ago in a month far, far away I said I was going to blog about some improvements I’d been working on for zink. I blogged about some of them, but one was conspicuously absent from the original list:

• make zink usable for gaming

There’s a lot that goes into this item. The post you’re reading now isn’t about to go so far as to claim that zink(-wip) is usable for gaming. No, that day is still far, far away. But this post is going to be the first step.

To begin with, a riddle: what change was made to zink between these two screenshots?

That’s right, I put the punchline in the title.

A suballocator.

What Is A Suballocator?

A suballocator is a mechanism by which small blocks of memory can be suballocated out of larger one. For example, if I want to allocate an 64byte chunk of memory, I could allocate it directly and get my block, or I could allocate a 4096byte chunk of memory and then take 64bytes out of it.

When performance is involved, it’s important to consider the time-cost of allocations, and so yes, it’s useful to have already allocated another 63 instances of 64bytes when I need a second one, but there’s another, deeper issue that’s also necessary to address, especially as it relates to gaming: 32bit environments.

In a 32bit process, the amount of address space available is limited to 4GB, regardless of how much actual memory is physically present, some of which is dedicated to system resources and unavailable for general use. Any time a buffer or image is mapped by the driver in a process, this uses up address space in order to create an addressable region of memory that can be read or written to. Once all the address space has been used up, no other resources can be mapped, and it becomes impossible to continue normal operations.

In short, the game crashes.

In Vulkan, and just generally in driver work, it’s important to keep allocation sizes aligned to the preference of the hardware for a given usage; this amounts to minMemoryMapAlignment, which is 4096bytes on many drivers. Similarly, vkGetBufferMemoryRequirements and vkGetImageMemoryRequirements return aligned memory sizes, so even if only 64bytes are needed, 4096bytes must still be allocated—4032 bytes unused. This ends up wasting tons of memory when an app is allocating lots of smaller regions, and it’s further wasting address space since Vulkan prohibits memory from being mapped multiple times, meaning that each 64byte buffer is wasting an additional 4032bytes of address space.

While 4k of memory may seem like a small amount, and why would anyone ever need more than 256kb memory anyway, these allocations all add up, fast enough that zink runs out of address space in a 32bit game like Tomb Raider within a couple minutes.

Playable?

Probably not.

The Solution, As Always

If you’re working in Mesa, you basically have two options when you come across a new problem: delete some code or copy some code. It’s not often that I come across an issue which can’t be resolved by one of the two.

In this case, I had known for quite a while that the solution was going to be copying some code. Thus I entered the realm of Gallium’s awesome auxilliary/pipebuffer, a fearsome component that had only been leveraged by one driver.

Yup, it was time to throw more galaxybrain.jpg code into the blender and see what came out. Ultimately, I was able to repurpose a lot of the core calculation code for sizing allocations, which saved me from having to do any kind of thinking or maffs. This let me cut down my suballocator implementation to a little under 700 lines, leaving much, much, much more space for bugsactivities.

At a high level, here’s an overview of aux/pb:

• call pb_cache_init to set up a memory cache
• initialize slab allocators with pb_slabs_init
• when allocating a new resource, determine if it can be slab allocated; if yes, use pb_slab_alloc to reuse/reclaim a slab allocation, otherwise manually allocate new memory
• when destroying a resource, use pb_reference_with_winsys

There’s more under the hood, but it mostly boils down to filling in the interface functions to manage detecting whether resources are busy or can be reclaimed for reuse. The actual caching/reclaiming/reusing are all handled by aux/pb, meaning I was free to go about breaking everything with all the leftover time that I had.

Cultured users of zink-wip can now enjoy massively improved performance (and have already been enjoying it for the past month) in many apps. The rest of you get to sit around and watch while I bang my head against CI while ajax showers me with memes.

There’s a lot that has happened in the world of Zink since my last update, so let’s see if I can bring you up to date on the most important stuff.

Upstream development

Gosh, when I last blogged about Zink, it hadn’t even landed upstream in Mesa yet! Well, by now it’s been upstream for quite a while, and most development has moved there.

At the time of writing, we have merged 606 merge-requests labeled “zink”. The current tip of mesa’s main branch is totaling 1717 commits touching the src/gallium/drivers/zink/ sub-folder, written by 42 different contributors. That’s pretty awesome in my eyes, Zink has truly become a community project!

Another noteworthy change is that Mike Blumenkrantz has come aboard the project, and has churned out an incredible amount of improvements to Zink! He got hired by Valve to work on Zink (among other things), and is now the most prolific contributor, with more than twice the amount of commits than I have written.

If you want a job in Open Source graphics, Zink has a proven track-record as a job-creator! :smile:

In addition to Mike, there’s some other awesome people who have been helping out lately.

Half-Life 2 running with Zink.

OpenGL 4.6 support

Thanks to a lot of hard work by Mike assisted by Dave Airlie and Adam Jackson, both of RedHat, Zink is now able to expose the OpenGL 4.6 (Core Profile) feature set, given enough Vulkan features! :tada:

Please note that this doesn’t mean that Zink is yet a conformant implementation, there’s some details left to be ironed out before we can claim that. In particular, we need to pass the conformance tests, and submit a conformance report to Khronos. We’re not there yet.

I’m also happy to see that Zink is currently at the top of MesaMatrix (together with LLVMpipe, i965 and RadeonSI), reporting a total of 160 OpenGL extensions at the time of writing!

In theory, that means you can run any OpenGL application you can think of on top of Zink. Mike is hard at work testing the entire Steam game library, and things are working pretty well.

Is this the end of the line for Zink? Are we done now? Not at all! :laughing:

OpenGL compabibility profile

We’re still stuck at OpenGL 3.0 for compatibility contexts, mainly due to lack of testing. There’s a lot of features that need to work together in relatively complicated ways for this to work for us.

Note that this only matters for applications that rely on legacy OpenGL features. Modern OpenGL programs gets OpenGL 4.6 support, as mentioned previously.

I don’t think this is going to be a big deal to enable, but I haven’t spent time on it.

OpenGL ES 3.1 support

Similar to the OpenGL 4.6 support, we’re now able to expose the OpenGL ES 3.1 feature set. This is again thanks to a lot of hard work by Mike and the gang.

Why not OpenGL ES 3.2? This comes down to the GL_KHR_blend_equation_advanced feature. Mike blogged about the issue a while ago.

Lavapipe and continuous integration

To prevent regressions, we’ve started testing Zink on the Mesa CI system for every change. This is made possible thanks to Lavapipe, a Vulkan software implementation in Mesa that reuse the rasterizer from LLVMpipe.

This means we can run tests on virtual cloud machines without having to depend on unreliable hardware. :robot:

At the time of writing, we’re only exposing OpenGL 4.1 on top of Lavapipe, due to some lacking features. But we have patches in the works to bring this up to OpenGL 4.5, and OpenGL 4.6 probably won’t be far off when that lands.

Windows support

Basic support for Zink on Microsoft Windows has landed. This isn’t particularly useful at the moment, because we need better window-system integration to get anywhere near reasonable performance. But it’s there.

macOS support

Thanks to work by Duncan Hopkins of The Foundry, there’s also some support for macOS. This uses MoltenVK as the Vulkan implementation, meaning that we also support the Vulkan Portability Extension to some degree.

This support isn’t quite as drop-in as on other platforms, because it’s completely lacking window-system integration. But it seems to work for the use-cases they have at The Foundry, so it’s worth mentioning as well.

Driver support

Beyond this, Igalia has brought up Zink on the V3DV driver, and I’ve heard some whispers that there’s some people running Zink on top of Turnip, an open-source Vulkan driver for recent Qualcomm Adreno GPUs.

I’ve heard some people have some success getting things running on NVIDIA, but there’s a few obvious problems in the way there due to the lack of proper DRI support… Which brings us to:

Window System Integration

Another awesome new development is that Adam is working on Penny. So, what’s Penny?

Penny is another way of bringing up Zink, on systems without DRI support. It works as a dedicated GLX integration that uses the VK_KHR_swapchain extension to integrate properly with the native Vulkan driver’s window-system integration instead of Mesa baking its own.

This solves a lot of small, nasty issues in the DRI code-path. I’ll say the magic “implicit synchronization” word, and hope that scares away anyone wondering what it’s about.

Performance

A lot more has happened on the performance front as well, again all thanks to Mike. However, much of this is still out-of-tree, and waiting in Mike’s zink-wip branch.

So instead, I suggest you check out Mike’s blog for the latest performance information (and much more up-to-date info on Zink). There’s been a lot going on, and I’m sure there’s even more to come!

Closing words

I think this should cover the most interesting bits of development.

On a personal note, I recently became a dad for the first time, and as a result I’ll be away for a while on paternity leave, starting early this fall. Luckily, Zink is in good hands with Mike and the rest of the upstream community taking care of things.

I would like to again plug Mike’s blog as a great source of Zink-related news, if you’re not already following it. He posts a lot more frequent than I do, and he’s also an epic meme master, so it’s all great fun!

(Added a section on load/store pairs on June 14th)

This question probably seems absurd. An unoptimized memcpy is a simple loop that copies bytes. How hard can that be? Well...

There's a fascinating thread on llvm-dev started by George Mitenkov proposing a new family of "byte" types. I found the proposal and discussion difficult to follow. In my humble opinion, this is because the proposal touches some rather subtle and underspecified aspects of LLVM IR semantics, and rather than address those fundamentals systematically, it jumps right into the minutiae of the instruction set. I look forward to seeing how the proposal evolves. In the meantime, this article is a byproduct of me attempting to digest the problem space.

Here is a fairly natural way to (attempt to) implement memcpy in LLVM IR:

define void @memcpy(i8* %dst, i8* %src, i64 %n) {
entry:
  %dst.end = getelementptr i8, i8* %dst, i64 %n
  %isempty = icmp eq i64 %n, 0
  br i1 %isempty, label %out, label %loop

loop:
  %src.loop = phi i8* [ %src, %entry ], [ %src.next, %loop ]
  %dst.loop = phi i8* [ %dst, %entry ], [ %dst.next, %loop ]
  %ch = load i8, i8* %src.loop
  store i8 %ch, i8* %dst.loop
  %src.next = getelementptr i8, i8* %src.loop, i64 1
  %dst.next = getelementptr i8, i8* %dst.loop, i64 1
  %done = icmp eq i8* %dst.next, %dst.end
  br i1 %done, label %out, label %loop

out:
  ret void
}

Unfortunately, the copy that is written to the destination is not a perfect copy of the source.

Hold on, I hear you think, each byte of memory holds one of 256 possible bit patterns, and this bit pattern is perfectly copied by the `load`/`store` sequence! The catch is that in LLVM's model of execution, a byte of memory can in fact hold more than just one of those 256 values. For example, a byte of memory can be poison, which means that there are at least 257 possible values. Poison is forwarded perfectly by the code above, so that's fine. The trouble starts because of pointer provenance.

What and why is pointer provenance?

From a machine perspective, a pointer is just an integer that is interpreted as a memory address.

For the compiler, alias analysis -- that is, the ability to prove that different pointers point at different memory addresses -- is crucial for optimization. One basic tool in the alias analysis toolbox is to recognize that if pointers point into different "memory objects" -- different stack or heap allocations -- then they cannot alias.

Unfortunately, many pointers are obtained via getelementptr (GEP) using dynamic (non-constant) indices. These dynamic indices could be such that the resulting pointer points into a different memory object than the base pointer. This makes it nearly impossible to determine at compile time whether two pointers point into the same memory object or not.

Which is why there is a rule which says (among other things) that if a pointer P obtained via GEP ends up going out-of-bounds and pointing into a different memory object than the pointer on which the GEP was based, then dereferencing P is undefined behavior even though the pointer's memory address is valid from the machine perspective.

As a corollary, a situation is possible in which there are two pointers whose underlying memory address is identical but whose provenance is different. In that case, it's possible that one of them can be dereferenced while dereferencing the other is undefined behavior.

This only makes sense if, in the formal semantics of LLVM IR, pointer values carry more information than just an integer interpreted as a memory address. They also carry provenance information, which is essentially the set of memory objects that can be accessed via this pointer and any pointers derived from it.

Bytes in memory carry provenance information

What is the provenance of a pointer that results from a load instruction? In a clean operational semantics, the load must derive this provenance from the values stored in memory.

If bytes of memory can only hold one of 256 bit patterns (or poison), that doesn't give us much to work with. We could say that the provenance of the pointer is "empty", meaning the pointer cannot be used to access any memory objects -- but that's clearly useless. Or we could say that the provenance of the pointer is "all", meaning the pointer (or pointers derived from it) can be freely used to access all memory objects, assuming the underlying address is adjusted appropriately. That isn't much better.[0]

Instead, we must say that -- as far as LLVM IR semantics are concerned -- each byte of memory holds pointer provenance information in addition to its i8 content. The provenance information in memory is written by pointer store, and pointer load uses it to reconstruct the original provenance of the loaded pointer.

What happens to provenance information in non-pointer load/store? A load can simply ignore the additional information in memory. For store, I see 3 possible choices:

1. Leave the provenance information that already happens to be in memory unmodified.
2. Set the provenance to "empty".
3. Set the provenance to "all".

Looking back at our attempt to implement memcpy, there is no choice which results in a perfect copy. All of the choices lose provenance information.

Without major changes to LLVM IR, only the last choice is potentially viable because it is the only choice that allows dereferencing pointers that are loaded from the memcpy destination.

Should we care about losing provenance information?

Without major changes to LLVM IR, we can only implement a memcpy that loses provenance information during the copy.

So what? Alias analysis around memcpy and code like it ends up being conservative, but reasonable people can argue that this doesn't matter. The burden of evidence lies on whoever wants to make a large change here in order to improve alias analysis.

That said, we cannot just call it a day and go (or stay) home either, because there are related correctness issues in LLVM today, e.g. bug 37469 mentioned in the initial email of that llvm-dev thread.

Here's a simpler example of a correctness issue using our hand-coded memcpy:

define i32 @sample(i32** %pp) {
  %tmp = alloca i32*
  %pp.8 = bitcast i32** %pp to i8*
  %tmp.8 = bitcast i32** %tmp to i8*
  call void @memcpy(i8* %tmp.8, i8* %pp.8, i64 8)
  %p = load i32*, i32** %tmp
  %x = load i32, i32* %p
  ret i32 %x
}

A transform that should be possible is to eliminate the memcpy and temporary allocation:

define i32 @sample(i32** %pp) {
  %p = load i32*, i32** %pp
  %x = load i32, i32* %p
  ret i32 %x
}

This transform is incorrect because it introduces undefined behavior.

To see why, remember that this is the world where we agree that integer stores write an "all" provenance to memory, so %p in the original program has "all" provenance. In the transformed program, this may no longer be the case. If @sample is called with a pointer that was obtained through an out-of-bounds GEP whose resulting address just happens to fall into a different memory object, then the transformed program has undefined behavior where the original program didn't.

We could fix this correctness issue by introducing an unrestrict instruction which elevates a pointer's provenance to the "all" provenance:

define i32 @sample(i32** %pp) {
  %p = load i32*, i32** %pp
  %q = unrestrict i32* %p
  %x = load i32, i32* %q
  ret i32 %x
}

Here, %q has "all" provenance and therefore no undefined behavior is introduced.

I believe that (at least for address spaces that are well-behaved?) it would be correct to fold inttoptr(ptrtoint(x)) to unrestrict(x). The two are really the same.

For that reason, unrestrict could also be used to fix the above-mentioned bug 37469. Several folks in the bug's discussion stated the opinion that the bug is caused by incorrect store forwarding that should be weakened via inttoptr(ptrtoint(x)). unrestrict(x) is simply a clearer spelling of the same idea.

A dead end: integers cannot have provenance information

A natural thought at this point is that the situation could be improved by adding provenance information to integers. This is technically correct: our hand-coded memcpy would then produce a perfect copy of the memory contents.

However, we would get into serious trouble elsewhere because global value numbering (GVN) and similar transforms become incorrect: two integers could compare equal using the icmp instruction, but still be different because of different provenance. Replacing one by the other could result in miscompilation.

GVN is important enough that adding provenance information to integers is a no-go.

I suspect that the unrestrict instruction would allow us to apply GVN to pointers, at the cost of making later alias analysis more conservative and sprinkling unrestrict instructions that may inhibit other transforms. I have no idea what the trade-off is on that.

The "byte" types: accurate representation of memory contents

With all the above in mind, I can see the first-principles appeal of the proposed "byte" types. They allow us to represent the contents of memory accurately in SSA values, and so they fill a real gap in the expressiveness of LLVM IR.

That said, the software development cost of adding a whole new family of types to LLVM is very high, so it better be justified by more than just aesthetics.

Our hand-coded memcpy can be turned into a perfect copier with straightforward replacement of i8 by b8:

define void @memcpy(b8* %dst, b8* %src, i64 %n) {
entry:
  %dst.end = getelementptr b8, b8* %dst, i64 %n
  %isempty = icmp eq i64 %n, 0
  br i1 %isempty, label %out, label %loop

loop:
  %src.loop = phi b8* [ %src, %entry ], [ %src.next, %loop ]
  %dst.loop = phi b8* [ %dst, %entry ], [ %dst.next, %loop ]
  %ch = load b8, b8* %src.loop
  store b8 %ch, b8* %dst.loop
  %src.next = getelementptr b8, b8* %src.loop, i64 1
  %dst.next = getelementptr b8, b8* %dst.loop, i64 1
  %done = icmp eq b8* %dst.next, %dst.end
  br i1 %done, label %out, label %loop

out:
  ret void
}

Looking at the concrete choices made in the proposal, I disagree with some of them.

Memory should not be typed. In the proposal, storing an integer always results in different memory contents than storing a pointer (regardless of its provenance), and implicitly trying to mix pointers and integers is declared to be undefined behavior. In other words, a sequence such as:

store i64 %x, i64* %p
%q = bitcast i64* %p to i8**
%y = load i8*, i8** %q

... is undefined behavior under the proposal instead of being effectively inttoptr(%x). That seems fine for C/C++, but is it going to be fine for other frontends?

The corresponding distinction between bytes-as-integers and bytes-as-pointers complicates the proposal overall, e.g. it forces them to add a bytecast instruction.

Conversely, the benefits of the distinction are unclear to me. One benefit appears to be guaranteed non-aliasing between pointer and non-pointer memory accesses, but that is a form of type-based alias analysis which in LLVM should idiomatically be done via TBAA metadata. (Update: see the addendum for another potential argument in favor of typed memory.)

So let's keep memory untyped, please.

Bitwise poison in byte values makes me really nervous due to the arbitrary deviation from how poison works in other types. I don't see any justification for it in the proposal. I can kind of see how one could be motivated by implementing memcpy with vector intrinsics operating on, for example, <8 x b32>, but a simpler solution would be to just use <32 x b8> instead. And if poison is indeed bitwise, then certainly pointer provenance would also have to be bitwise!

Finally, no design discussion is complete without a little bit of bike-shedding. I believe the name "byte" is inspired by C++'s std::byte, but given that types such as b256 are possible, this name would forever be a source of confusion. Naming is hard, and I think we should at least try to look for a better one. Let me kick off the brainstorming by suggesting we think of them as "memory content" values, because that's what they are. The types could be spelled m8, m32, etc. in IR assembly.

A variation: adding a pointer provenance type

In the llvm-dev thread, Jeroen Dobbelaere points out work being done to introduce explicit `ptr_provenance` operands on certain instructions, in service of C99's restrict keyword. I haven't properly digested this work, but it inspired the thoughts of this section.

Values of the proposed byte types have both a bit pattern and a pointer provenance. Do we really need to have both pieces of information in the same SSA value? We could instead split them up into an integer bit pattern value and a pointer provenance value with an explicit provenance type. Loads of integers could read out the provenance information stored in memory and provide it as a secondary result. Similarly, stores of integers could accept the desired provenance to be stored in memory as a secondary data operand. This would allow us to write a perfect memcpy by replacing the core load/store sequence with something like:

%ch, %provenance = load_with_provenance i8, i8* %src
store_with_provenance i8 %ch, provenance %provenance, i8* %dst

The syntax and instruction names in the example are very much straw men. Don't take them too seriously, especially because LLVM IR doesn't currently allow multiple result values.

Interestingly, this split allows the derivation of pointer provenance to follow a different path than the calculation of the pointer's bit pattern. This in turn allows us in principle to perform GVN on pointers without being conservative for alias analysis.

One of the steps in bug 37469 is not quite GVN, but morally similar. Simplifying a lot, the original program sequence:

%ch1 = load i8, i8* %p1
%ch2 = load i8, i8* %p2
%eq = icmp eq i8 %ch1, %ch2
%ch = select i1 %eq, i8 %ch1, i8 %ch2
store i8 %ch, i8* %p3

... is transformed into:

%ch2 = load i8, i8* %p2
store i8 %ch2, i8* %p3

This is correct for the bit patterns being loaded and stored, but the program also indirectly relies on pointer provenance of the data. Of course, there is no pointer provenance information being copied here because i8 only holds a bit pattern. However, with the "byte" proposal, all the i8s would be replaced by b8s, and then the transform becomes incorrect because it changes the provenance information.

If we split the proposed use of b8 into a use of i8 and explicit provenance values, the original program becomes:

%ch1, %prov1 = load_with_provenance i8, i8* %p1
%ch2, %prov2 = load_with_provenance i8, i8* %p2
%eq = icmp eq i8 %ch1, %ch2
%ch = select i1 %eq, i8 %ch1, i8 %ch2
%prov = select i1 %eq, provenance %prov1, provenance %prov2
store_with_provenance i8 %ch, provenance %prov, i8* %p3

This could be transformed into something like:

%prov1 = load_only_provenance i8* %p1
%ch2, %prov2 = load_with_provenance i8, i8* %p2
%prov = merge provenance %prov1, %prov2
store_with_provenance i8 %ch2, provenance %prov, i8* %p3

... which is just as good for code generation but loses only very little provenance information.

Aside: loop idioms

Without major changes to LLVM IR, a perfect memcpy cannot be implemented because pointer provenance information is lost.

Nevertheless, one could still define the @llvm.memcpy intrinsic to be a perfect copy. This helps memcpys in the original source program be less conservative in terms of alias analysis. However, it also makes it incorrect to replace a memcpy loop idiom with a use of @llvm.memcpy: without adding unrestrict instructions, the replacement may introduce undefined behavior; and there is no way to bound the locations where such unrestricts may be needed.

We could augment @llvm.memcpy with an immediate argument that selects its provenance behavior.

In any case, one can argue that bug 37469 is really a bug in the loop idiom recognizer. It boils down to the details of how everything is defined, and unfortunately, these weird corner cases are currently underspecified in the LangRef.

Conclusion

We started with the question of whether memcpy can be implemented in LLVM IR. The answer is a qualified Yes. It is possible, but the resulting copy is imperfect because pointer provenance information is lost. This has surprising implications which in turn happen to cause real miscompilation bugs -- although those bugs could be fixed even without a perfect memcpy.

The "byte" proposal has a certain aesthetic appeal because it fixes a real gap in the expressiveness of LLVM IR, but its software engineering cost is large and I object to some of its details. There are also alternatives to consider.

The miscompilation bugs obviously need to be fixed, but they can be fixed much less intrusively, albeit at the cost of more conservative alias analysis in the affected places. It is not clear to me whether improving alias analysis justifies the more complex solutions.

I would like to understand better how all of this interacts with the C99 restrict work. That work introduces mechanisms for explicitly talking about pointer provenance in the IR, which may allow us to kill two birds with one stone.

In any case, this is a fascinating topic and discussion, and I feel like we're only at the beginning.

Addendum: storing back previously loaded integers

(Added this section on June 14th)

Harald van Dijk on Phrabricator and Ralf Jung on llvm-dev, referring to a Rust issue, explicitly and implicitly point out a curious issue with loading and storing integers.

Here is Harald's example:

define i8* @f(i8* %p) {
  %buf = alloca i8*
  %buf.i32 = bitcast i8** %buf to i32*
  store i8* %p, i8** %buf
  %i = load i32, i32* %buf.i32
  store i32 %i, i32* %buf.i32
  %q = load i8*, i8** %buf
  ret i8* %q
}

There is a pair of load/store of i32 which is fully redundant from a machine perspective and so we'd like to optimize that away, after which it becomes obvious that the function really just returns %p -- at least as far as bit patterns are concerned.

However, in a world where memory is untyped but has provenance information, this optimization is incorrect because it can introduce undefined behavior: the load/store of i32 resets the provenance information in memory to "all", so that the original function returns an unrestricted version of %p. This is no longer the case after the optimization.

There are at least two possible ways of resolving this conflict.

We could define memory to be typed, in the sense that each byte of memory remembers whether it was most recently stored as a pointer or a non-pointer. A load with the wrong type returns poison. In that case, the example above returns poison before the optimization (because %i is guaranteed to be poison). After the optimization it returns non-poison, which is an acceptable refinement, so the optimization is correct.

The alternative is to keep memory untyped and say that directly eliminating the i32 store in the example is incorrect.

We are facing a tradeoff that depends on how important that optimization is for performance.

Two observations to that end. First, the more common case of dead store elimination is one where there are multiple stores to the same address in a row, and we remove all but the last one of them. That more common optimization is unaffected by provenance issues either way.

Second, we can still perform store forwarding / peephole optimization across such load/store pairs, as long as we are careful to introduce unrestrict where needed. The example above can be optimized via store forwarding to:

define i8* @f(i8* %p) {
  %buf = alloca i8*
  %buf.i32 = bitcast i8** %buf to i32*
  store i8* %p, i8** %buf
  %i = load i32, i32* %buf.i32
  store i32 %i, i32* %buf.i32
  %q = unrestrict i8* %p
  ret i8* %q
}

We can then dead-code eliminate the bulk of the function and obtain:

define i8* @f(i8* %p) {
  %q = unrestrict i8* %p
  ret i8* %q
}

... which is as good as it can possibly get.

So, there is a good chance that preventing this particular optimization is relatively cheap in terms of code quality, and the gain in overall design simplicity may well be worth it.

[0] We could also say that the loaded pointer's provenance is magically the memory object that happens to be at the referenced memory address. Either way, provenance would become a useless no-op in most cases. For example, mem2reg would have to insert unrestrict instructions (defined later) everywhere because pointers become effectively "unrestricted" when loaded from alloca'd memory.

In an earlier article I showed how reading from VRAM with the CPU can be very slow. It however turns out there there are ways to make it less slow.

The key to this are instructions with non-temporal hints, in particular VMOVNTDQA. The Intel Instruction Manual says the following about this instruction:

“MOVNTDQA loads a double quadword from the source operand (second operand) to the destination operand (first operand) using a non-temporal hint if the memory source is WC (write combining) memory type. For WC memory type, the nontemporal hint may be implemented by loading a temporary internal buffer with the equivalent of an aligned cache line without filling this data to the cache. Any memory-type aliased lines in the cache will be snooped and flushed. Subsequent MOVNTDQA reads to unread portions of the WC cache line will receive data from the temporary internal buffer if data is available. “ (Intel® 64 and IA-32 Architectures Software Developer’s Manual Volume 2)

This sounds perfect for our VRAM and WC System Memory buffers as we typically only read 16-bytes per instruction and this allows us to read entire cachelines at time.

It turns out that Mesa already implemented a streaming memcpy using these instructions so all we had to do was throw that into our benchmark and write a corresponding memcpy that does non-temporal stores to benchmark writing to these memory regions.

As a reminder, we look into three allocation types that are exposed by the amdgpu Linux kernel driver:

• VRAM. This lives on the GPU and is mapped with Uncacheable Speculative Write Combining (USWC) on the CPU. This means that accesses from the CPU are not cached, but writes can be write-combined.

• Cacheable system memory. This is system memory that has caching enabled on the CPU and there is cache snooping to ensure the memory is coherent between the CPU and GPU (up till the top level caches. The GPU caches do not participate in the coherence).

• USWC system memory. This is system memory that is mapped with Uncacheable Speculative Write Combining on the CPU. This can lead to slight performance benefits compared to cacheable system memory due to lack of cache snooping.

Furthermore this still uses a RX 6800 XT + a 2990WX with 4 channel 3200 MT/s RAM.

Using this memcpy implementation we get significantly better performance in uncached memory situations, 50x for VRAM and 26x for USWC system memory. If this is a significant bottleneck in your workload this can be a gamechanger. Or if you were using SDMA to avoid this hit, you might be able to do things at significantly lower latency. That said it is not at a level where it does not matter. For big copies using DMA can still be a significant win.

Note that I initially gave an explanation on why the non-temporal loads should be faster, but the increases in performance are significantly above what something that just fiddles with loading entire cachelines would achieve. I have not dug into the why of the performance increase.

DMA performance

I have been claiming DMA is faster for CPU readbacks of VRAM in both this article and the previous article on the topic. One might ask how fast DMA is then. To demonstrate this I benchmarked VRAM<->Cacheable System Memory copies using the SDMA hardware block on Radeon GPUs.

Note that there is a significant overhead per copy here due to submitting work to the GPU, so I will shows results vs copy size. The rate is measured while doing a wait after each individual copy and taking the wall clock time as these usecases tend to be latency sensitive and hence batching is not too interesting.

This shows that for reads DMA is faster than a normal memcpy at 4 KiB and faster than a streaming memcpy at 64 KiB. Of course one still needs to do their CPU access at that point, but at both these thresholds even with an additional CPU memcpy the total process should still be fast with DMA.

TL;DR: Tag your GPT partitions with the right, descriptive partition types, and the world will become a better place.

A number of years ago we started the Discoverable Partitions Specification which defines GPT partition type UUIDs and partition flags for the various partitions Linux systems typically deal with. Before the specification all Linux partitions usually just used the same type, basically saying "Hey, I am a Linux partition" and not much else. With this specification the GPT partition type, flags and label system becomes a lot more expressive, as it can tell you:

1. What kind of data a partition contains (i.e. is this swap data, a file system or Verity data?)
2. What the purpose/mount point of a partition is (i.e. is this a /home/ partition or a root file system?)
3. What CPU architecture a partition is intended for (i.e. is this a root partition for x86-64 or for aarch64?)
4. Shall this partition be mounted automatically? (i.e. without specifically be configured via /etc/fstab)
5. And if so, shall it be mounted read-only?
6. And if so, shall the file system be grown to its enclosing partition size, if smaller?
7. Which partition contains the newer version of the same data (i.e. multiple root file systems, with different versions)

By embedding all of this information inside the GPT partition table disk images become self-descriptive: without requiring any other source of information (such as /etc/fstab) if you look at a compliant GPT disk image it is clear how an image is put together and how it should be used and mounted. This self-descriptiveness in particular breaks one philosophical weirdness of traditional Linux installations: the original source of information which file system the root file system is, typically is embedded in the root file system itself, in /etc/fstab. Thus, in a way, in order to know what the root file system is you need to know what the root file system is. 🤯 🤯 🤯

(Of course, the way this recursion is traditionally broken up is by then copying the root file system information from /etc/fstab into the boot loader configuration, resulting in a situation where the primary source of information for this — i.e. /etc/fstab — is actually mostly irrelevant, and the secondary source — i.e. the copy in the boot loader — becomes the configuration that actually matters.)

Today, the GPT partition type UUIDs defined by the specification have been adopted quite widely, by distributions and their installers, as well as a variety of partitioning tools and other tools.

In this article I want to highlight how the various tools the systemd project provides make use of the concepts the specification introduces.

But before we start with that, let's underline why tagging partitions with these descriptive partition type UUIDs (and the associated partition flags) is a good thing, besides the philosophical points made above.

1. Simplicity: in particular OS installers become simpler — adjusting /etc/fstab as part of the installation is not necessary anymore, as the partitioning step already put all information into place for assembling the system properly at boot. i.e. installing doesn't mean that you always have to get fdisk and /etc/fstab into place, the former suffices entirely.

2. Robustness: since partition tables mostly remain static after installation the chance of corruption is much lower than if the data is stored in file systems (e.g. in /etc/fstab). Moreover by associating the metadata directly with the objects it describes the chance of things getting out of sync is reduced. (i.e. if you lose /etc/fstab, or forget to rerun your initrd builder you still know what a partition is supposed to be just by looking at it.)

3. Programmability: if partitions are self-descriptive it's much easier to automatically process them with various tools. In fact, this blog story is mostly about that: various systemd tools can naturally process disk images prepared like this.

4. Alternative entry points: on traditional disk images, the boot loader needs to be told which kernel command line option root= to use, which then provides access to the root file system, where /etc/fstab is then found which describes the rest of the file systems. Where precisely root= is configured for the boot loader highly depends on the boot loader and distribution used, and is typically encoded in a Turing complete programming language (Grub…). This makes it very hard to automatically determine the right root file system to use, to implement alternative entry points to the system. By alternative entry points I mean other ways to boot the disk image, specifically for running it as a systemd-nspawn container — but this extends to other mechanisms where the boot loader may be bypassed to boot up the system, for example qemu when configured without a boot loader.

5. User friendliness: it's simply a lot nicer for the user looking at a partition table if the partition table explains what is what, instead of just saying "Hey, this is a Linux partition!" and nothing else.

Uses for the concept

Now that we cleared up the Why?, lets have a closer look how this is currently used and exposed in systemd's various components.

Use #1: Running a disk image in a container

If a disk image follows the Discoverable Partition Specification then systemd-nspawn has all it needs to just boot it up. Specifically, if you have a GPT disk image in a file foobar.raw and you want to boot it up in a container, just run systemd-nspawn -i foobar.raw -b, and that's it (you can specify a block device like /dev/sdb too if you like). It becomes easy and natural to prepare disk images that can be booted either on a physical machine, inside a virtual machine manager or inside such a container manager: the necessary meta-information is included in the image, easily accessible before actually looking into its file systems.

Use #2: Booting an OS image on bare-metal without /etc/fstab or kernel command line root=

If a disk image follows the specification in many cases you can remove /etc/fstab (or never even install it) — as the basic information needed is already included in the partition table. The systemd-gpt-auto-generator logic implements automatic discovery of the root file system as well as all auxiliary file systems. (Note that the former requires an initrd that uses systemd, some more conservative distributions do not support that yet, unfortunately). Effectively this means you can boot up a kernel/initrd with an entirely empty kernel command line, and the initrd will automatically find the root file system (by looking for a suitably marked partition on the same drive the EFI System Partition was found on).

(Note, if /etc/fstab or root= exist and contain relevant information they always takes precedence over the automatic logic. This is in particular useful to tweaks thing by specifying additional mount options and such.)

Use #3: Mounting a complex disk image for introspection or manipulation

The systemd-dissect tool may be used to introspect and manipulate OS disk images that implement the specification. If you pass the path to a disk image (or block device) it will extract various bits of useful information from the image (e.g. what OS is this? what partitions to mount?) and display it.

With the --mount switch a disk image (or block device) can be mounted to some location. This is useful for looking what is inside it, or changing its contents. This will dissect the image and then automatically mount all contained file systems matching their GPT partition description to the right places, so that you subsequently could chroot into it. (But why chroot if you can just use systemd-nspawn? 😎)

Use #4: Copying files in and out of a disk image

The systemd-dissect tool also has two switches --copy-from and --copy-to which allow copying files out of or into a compliant disk image, taking all included file systems and the resulting mount hierarchy into account.

Use #5: Running services directly off a disk image

The RootImage= setting in service unit files accepts paths to compliant disk images (or block device nodes), and can mount them automatically, running service binaries directly off them (in chroot() style). In fact, this is the base for the Portable Service concept of systemd.

Use #6: Provisioning disk images

systemd provides various tools that can run operations provisioning disk images in an "offline" mode. Specifically:

systemd-tmpfiles

With the --image= switch systemd-tmpfiles can directly operate on a disk image, and for example create all directories and other inodes defined in its declarative configuration files included in the image. This can be useful for example to set up the /var/ or /etc/ tree according to such configuration before first boot.

systemd-sysusers

Similar, the --image= switch of systemd-sysusers tells the tool to read the declarative system user specifications included in the image and synthesizes system users from it, writing them to the /etc/passwd (and related) files in the image. This is useful for provisioning these users before the first boot, for example to ensure UID/GID numbers are pre-allocated, and such allocations not delayed until first boot.

systemd-machine-id-setup

The --image= switch of systemd-machine-id-setup may be used to provision a fresh machine ID into /etc/machine-id of a disk image, before first boot.

systemd-firstboot

The --image= switch of systemd-firstboot may be used to set various basic system setting (such as root password, locale information, hostname, …) on the specified disk image, before booting it up.

Use #7: Extracting log information

The journalctl switch --image= may be used to show the journal log data included in a disk image (or, as usual, the specified block device). This is very useful for analyzing failed systems offline, as it gives direct access to the logs without any further, manual analysis.

Use #8: Automatic repartitioning/growing of file systems

The systemd-repart tool may be used to repartition a disk or image in an declarative and additive way. One primary use-case for it is to run during boot on physical or VM systems to grow the root file system to the disk size, or to add in, format, encrypt, populate additional partitions at boot.

With its --image= switch it the tool may operate on compliant disk images in offline mode of operation: it will then read the partition definitions that shall be grown or created off the image itself, and then apply them to the image. This is particularly useful in combination with the --size= which allows growing disk images to the specified size.

Specifically, consider the following work-flow: you download a minimized disk image foobar.raw that contains only the minimized root file system (and maybe an ESP, if you want to boot it on bare-metal, too). You then run systemd-repart --image=foo.raw --size=15G to enlarge the image to the 15G, based on the declarative rules defined in the repart.d/ drop-in files included in the image (this means this can grow the root partition, and/or add in more partitions, for example for /srv or so, maybe encrypted with a locally generated key or so). Then, you proceed to boot it up with systemd-nspawn --image=foo.raw -b, making use of the full 15G.

Versioning + Multi-Arch

Disk images implementing this specifications can carry OS executables in one of three ways:

1. Only a root file system

2. Only a /usr/ file system (in which case the root file system is automatically picked as tmpfs).

3. Both a root and a /usr/file system (in which case the two are combined, the /usr/ file system mounted into the root file system, and the former possibly in read-only fashion`)

They may also contain OS executables for different architectures, permitting "multi-arch" disk images that can safely boot up on multiple CPU architectures. As the root and /usr/ partition type UUIDs are specific to architectures this is easily done by including one such partition for x86-64, and another for aarch64. If the image is now used on an x86-64 system automatically the former partition is used, on aarch64 the latter.

Moreover, these OS executables may be contained in different versions, to implement a simple versioning scheme: when tools such as systemd-nspawn or systemd-gpt-auto-generator dissect a disk image, and they find two or more root or /usr/ partitions of the same type UUID, they will automatically pick the one whose GPT partition label (a 36 character free-form string every GPT partition may have) is the newest according to strverscmp() (OK, truth be told, we don't use strverscmp() as-is, but a modified version with some more modern syntax and semantics, but conceptually identical).

This logic allows to implement a very simple and natural A/B update scheme: an updater can drop multiple versions of the OS into separate root or /usr/ partitions, always updating the partition label to the version included there-in once the download is complete. All of the tools described here will then honour this, and always automatically pick the newest version of the OS.

Verity

When building modern OS appliances, security is highly relevant. Specifically, offline security matters: an attacker with physical access should have a difficult time modifying the OS in a way that isn't noticed. i.e. think of a car or a cell network base station: these appliances are usually parked/deployed in environments attackers can get physical access to: it's essential that in this case the OS itself sufficiently protected, so that the attacker cannot just mount the OS file system image, make modifications (inserting a backdoor, spying software or similar) and the system otherwise continues to run without this being immediately detected.

A great way to implement offline security is via Linux' dm-verity subsystem: it allows to securely bind immutable disk IO to a single, short trusted hash value: if an attacker manages to offline modify the disk image the modified disk image won't match the trusted hash anymore, and will not be trusted anymore (depending on policy this then just result in IO errors being generated, or automatic reboot/power-off).

The Discoverable Partitions Specification declares how to include Verity validation data in disk images, and how to relate them to the file systems they protect, thus making if very easy to deploy and work with such protected images. For example systemd-nspawn supports a --root-hash= switch, which accepts the Verity root hash and then will automatically assemble dm-verity with this, automatically matching up the payload and verity partitions. (Alternatively, just place a .roothash file next to the image file).

Future

The above already is a powerful tool set for working with disk images. However, there are some more areas I'd like to extend this logic to:

bootctl

Similar to the other tools mentioned above, bootctl (which is a tool to interface with the boot loader, and install/update systemd's own EFI boot loader sd-boot) should learn a --image= switch, to make installation of the boot loader on disk images easy and natural. It would automatically find the ESP and other relevant partitions in the image, and copy the boot loader binaries into them (or update them).

coredumpctl

Similar to the existing journalctl --image= logic the coredumpctl tool should also gain an --image= switch for extracting coredumps from compliant disk images. The combination of journalctl --image= and coredumpctl --image= would make it exceptionally easy to work with OS disk images of appliances and extracting logging and debugging information from them after failures.

And that's all for now. Please refer to the specification and the man pages for further details. If your distribution's installer does not yet tag the GPT partition it creates with the right GPT type UUIDs, consider asking them to do so.

Thank you for your time.

Memes

We’ve all been there. No matter how 10x someone is or feels, everyone has had a moment where abruptly they say to themselves, HOW THE FUCK DO THREADS EVEN WORK?

This may be precipitated by any number of events, including, but not limited to:

• forgetting a lock
• forgetting to unlock
• missing an unlock at an early return
• forgetting to initialize a lock
• forgetting to spawn a thread
• forgetting to signal a conditional
• forgetting to initialize a conditional
• running the test case with the wrong driver

I’m not going to say that I’ve been there recently.

I’m not going to say that it was today, nor am I going to state, on the record, that at least one existing zink-wip snapshot may or may not be affected by an issue which may or may not be on the above list.

I’m not going to say any of these things.

What I am going to do is talk about a new oom handler I’ve been working on to handle the dreaded spec@!opengl 1.1@streaming-texture-leak case from piglit.

The Case

This test is annoying in that it is effectively a test of a driver’s ability to throttle itself when an app is generating and using $infinity textures without ever explicitly triggering a flush.

In short, it’s:

for (i = 0; i < 5000; i++) {
   glGenTextures(1, &texture);
   glBindTexture(GL_TEXTURE_2D, texture);
   glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, TEX_SIZE, TEX_SIZE, 0, GL_RGBA, GL_UNSIGNED_BYTE, tex_buffer);
   piglit_draw_rect_tex(0, 0, piglit_width, piglit_height, 0, 0, 1, 1);
   glDeleteTextures(1, &texture);
}

The textures are “deleted”, yes, but because they’re in use, the driver can’t actually delete them at this point of call, meaning that they can only truly be deleted once they are no longer in use by the GPU. At some iteration, this will begin to oom the GPU, and the driver will have to determine how to handle things.

The Zink Case

At present, mainline zink uses a hammer-and-nail methodology that I came up with last year: the total amount of GPU memory in use by resources in a given cmdbuf is tracked, and that amount is tracked per-context. If the in-use context memory exceeds a threshold of the total VRAM, the driver stalls, thereby freeing up all the resources that are in use so they can be recycled into new ones.

There’s a number of problems with this approach, but the biggest one is that it fails to account for cases like a AAA game that just uses as much memory as it can in order to optimize performance/resolution/graphics. I discovered such a case some time ago while running Tomb Raider, and then I set out to improve things since it was costing me about 10% of my perf on the title screen.

The annoying part of this problem is that the piglit test is a very uncommon case, and it’s tricky to handle it in a way that doesn’t also impact other cases which appear similar but need to not get memory-clamped. As a result, it’s tough to really do anything based on “overall” memory usage.

In the end, what I decided on was using the per-cmdbuf memory usage counter to trigger a check for completed cmdbufs on submit, iterating over all the pending ones to check whether they’ve completed, resetting them and freeing associated resources when possible. This yields good memory reclaiming behavior for problem cases while leaving games like Tomb Raider untouched and definitely not deadlocking or anything like that.

Remember When…

I said I’d be blogging every day about some changes? And that was a month ago or however long it’s been? And we all had a good chuckle at the idea that I could blog every day like how things used to be?

Yeah, I remember that too.

Anyway, Bas still hasn’t blogged, so let’s check the blogenda:

• handwaving about C++ draw templates
• some obscure vbuf thing
• shower
• make zink usable for gaming
• complain about construction
• improve shader caching
• this week’s queue rewrite
• some other stuff
• suballocator?

I guess it’s that time of the week again because the schedule says it’s time to talk about this week’s (or whenever it was) major rewrite of zink’s queue handling. But first, only 90s kids will remember that time I blogged about a major queue rewrite and was excited to almost be hitting 70% of native performance.

Synchronization

A common use of GL for big games is using multiple GL contexts to parallelize work. There’s a lot of tricky restrictions for this, both on the app side and the driver side, but this is sort of the closest thing to multiple cmdbufs that GL provides.

We all recall how zink now uses a monotonic queue: upon commencing recording, each cmdbuf gets tagged with a 32bit integer id that doubles as a timeline semaphore id for fencing. The queue iterates, the cmdbuf counter increments, queue submission is done per-context in a thread, the GPU gets triangles, everyone is happy.

But how well does that work out with multiple contexts?

Pretty well, it turns out, as long as you’re using a Vulkan driver that doesn’t actually check to ensure you’re using monotonic ids for your timeline values. Let’s check out a totally hypothetical scenario that isn’t just Steam:

• Have contexts A and B
• Context A starts recording, gets id 1 (nonzero id club represent)
• Context B starts recording, gets id 2
• Context A finishes recording, submits cmdbuf
• Context B finishes recording, submits cmdbuf
• Timeline wait on id 1
• Timeline wait on id 2

So far so good. But then we get past the “Checking for updates” window:

• Context A starts recording, gets id 3
• Context B starts recording, gets id 4
• Context B finishes recording, submits cmdbuf
• Context A finishes recording, submits cmdbuf
• Timeline wait on id 3
• Timeline wait on id 4

So now context B’s submit thread is dumping cmdbuf 4’s triangles into the GPU, then context A’s submit thread is also trying to dump cmdbuf 3’s triangles into the GPU, but the wait order for the timeline is still A -> B, meaning that the values are not monotonic.

Will any drivers care?

Magic 8-ball says no, no drivers care about this and everything still works fine. That’s cool and interesting, but probably it’d be better to not do that.

This Time It’s Definitely Fixed

The problem here is two problems:

• the queue submission thread is context-based when it should be screen based
• cmdbufs get an id when they start recording, not when they get submitted

The first problem is easy to fix: just deduplicate the thread and move the struct member.

The second one is trickier because everything in zink relies on cmdbufs getting an id as soon as they become active. This is done so that any resources written to by a given cmdbuf can have their usage tracked for synchronization purposes, e.g., reading back a buffer only after all its writes have landed.

The problem is further complicated by zink not having a great API barrier between directly accessing the “usage” for a resource and the value itself, by which I mean parts of the codebase directly reading the integer value vs having an API to wrap it; the latter would enable replacing the mechanism with whatever I wanted, so I decided to start by creating such a wrapper based on this:

struct zink_batch_usage {
   uint32_t usage;
   bool unflushed;
};

This is the existing struct zink_batch_usage but now with a bool value indicating that this cmdbuf is yet to be flushed. Each cmdbuf batch now has this sub-struct inlined onto it, and resources in zink can take references (pointers) to a specific cmdbuf’s usage struct. Because batches are never destroyed, this means the wrapper API can always dereference the struct to determine how to synchronize the usage: if it’s unflushed, it can flush or sync the flush thread; if it’s real, pending usage, it can safely wait on that usage as a timeline value and guarantee monotonic ordering.

bool
zink_screen_usage_check_completion(struct zink_screen *screen, const struct zink_batch_usage *u)
{
   if (!zink_batch_usage_exists(u))
      return true;
   if (zink_batch_usage_is_unflushed(u))
      return false;

   return zink_screen_batch_id_wait(screen, u->usage, 0);
}

bool
zink_batch_usage_check_completion(struct zink_context *ctx, const struct zink_batch_usage *u)
{
   if (!zink_batch_usage_exists(u))
      return true;
   if (zink_batch_usage_is_unflushed(u))
      return false;
   return zink_check_batch_completion(ctx, u->usage);
}

void
zink_batch_usage_wait(struct zink_context *ctx, const struct zink_batch_usage *u)
{
   if (!zink_batch_usage_exists(u))
      return;
   if (zink_batch_usage_is_unflushed(u))
      zink_fence_wait(&ctx->base);
   else
      zink_wait_on_batch(ctx, u->usage);
}

Now things render exactly the same, but with a truly monotonic queue underneath that’s conformant to specifications.

As the President of the GNOME Foundation Board of Directors, I’m really pleased to see the number and breadth of candidates we have for this year’s election. Thank you to everyone who has submitted their candidacy and volunteered their time to support the Foundation. Allan has recently blogged about how the board has been evolving, and I wanted to follow that post by talking about where the GNOME Foundation is in terms of its strategy. This may be helpful as people consider which candidates might bring the best skills to shape the Foundation’s next steps.

Around three years ago, the Foundation received a number of generous donations, and Rosanna (Director of Operations) gave a presentation at GUADEC about her and Neil’s (Executive Director, essentially the CEO of the Foundation) plans to use these funds to transform the Foundation. We would grow our activities, increasing the pace of events, outreach, development and infrastructure that supported the GNOME project and the wider desktop ecosystem – and, crucially, would grow our funding to match this increased level of activity.

I think it’s fair to say that half of this has been a great success – we’ve got a larger staff team than GNOME has ever had before. We’ve widened the GNOME software ecosystem to include related apps and projects under the GNOME Circle banner, we’ve helped get GTK 4 out of the door, run a wider-reaching program in the Community Engagement Challenge, and consistently supported better infrastructure for both GNOME and the Linux app community in Flathub.

Aside from another grant from Endless (note: my employer), our fundraising hasn’t caught up with this pace of activities. As a result, the Board recently approved a budget for this financial year which will spend more funds from our reserves than we expect to raise in income. Due to our reserves policy, this is essentially the last time we can do this: over the next 6-12 months we need to either raise more money, or start spending less.

For clarity – the Foundation is fit and well from a financial perspective – we have a very healthy bank balance, and a very conservative “12 month run rate” reserve policy to handle fluctuations in income. If we do have to slow down some of our activities, we will return to a “steady state” where our regular individual donations and corporate contributions can support a smaller staff team that supports the events and infrastructure we’ve come to rely on.

However, this isn’t what the Board wants to do – the previous and current boards were unanimous in their support of the idea that we should be ambitious: try to do more in the world and bring the benefits of GNOME to more people. We want to take our message of trusted, affordable and accessible computing to the wider world.

Typically, a lot of the activities of the Foundation have been very inwards-facing – supporting and engaging with either the existing GNOME or Open Source communities. This is a very restricted audience in terms of fundraising – many corporate actors in our community already support GNOME hugely in terms of both financial and in-kind contributions, and many OSS users are already supporters either through volunteer contributions or donating to those nonprofits that they feel are most relevant and important to them.

To raise funds from new sources, the Foundation needs to take the message and ideals of GNOME and Open Source software to new, wider audiences that we can help. We’ve been developing themes such as affordability, privacy/trust and education as promising areas for new programs that broaden our impact. The goal is to find projects and funding that allow us to both invest in the GNOME community and find new ways for FOSS to benefit people who aren’t already in our community.

Bringing it back to the election, I’d like to make clear that I see this – reaching the outside world, and finding funding to support that – as the main priority and responsibility of the Board for the next term. GNOME Foundation elections are a slightly unusual process that “filters” our board nominees by being existing Foundation members, which means that candidates already work inside our community when they stand for election. If you’re a candidate and are already active in the community – THANK YOU – you’re doing great work, keep doing it! That said, you don’t need to be a Director to achieve things within our community or gain the support of the Foundation: being a community leader is already a fantastic and important role.

The Foundation really needs support from the Board to make a success of the next 12-18 months. We need to understand our financial situation and the trade-offs we have to make, and help to define the strategy with the Executive Director so that we can launch some new programs that will broaden our impact – and funding – for the future. As people cast their votes, I’d like people to think about what kind of skills – building partnerships, commercial background, familiarity with finances, experience in nonprofit / impact spaces, etc – will help the Board make the Foundation as successful as it can be during the next term.

I Hate Construction.

Specifically when it’s right outside my house and starts at 5:30AM with heavy machinery moving around.

With this said, I’m overdue for a post, and if I don’t set a good example by continuing to blog, why would anyone else? PS. BAS IT’S TIME.

Let’s see what’s on the agenda:

• handwaving about C++ draw templates
• some obscure vbuf thing
• shower
• make zink usable for gaming
• complain about construction
• improve shader caching

Looks like the next thing on the list is shader caching.

The Art Of The Cache

If you’re a long-time zink connoisseur, or if you’re just a casual reader of the blog, you know that zink has a shader cache.

But did you know that it doesn’t actually do anything at present?

Indeed, it was to my chagrin that, upon diving back into my slapdash pipeline cache implementation, I discovered that it was doing absolutely nothing. And this was a different nothing than that one time I didn’t actually pass the cache back to the vulkan driver! Yes, this was the nothing of I have a cache, why am I still compiling a hundred pipelines per frame? that the occasional lucky developer runs into every now and again.

But hwhy? Who would do such a thing?

Past recriminations aside, how does a shader/pipeline cache work, anyway? The gist of it in most Mesa drivers is this:

Thus a shader gets cached based on its text representation, enabling matching shaders across programs to use the same cache entry. After noting the success of Steam’s fossilize-based single file cache, I decided to use a single file for zink’s shader cache.

Oops.

The problem in this case was that I was just jamming all the pipelines into a single file, written once at program exit, and expecting the Vulkan driver to figure things out.

But what if the program didn’t exit cleanly? Or what if the write failed for some reason?

In short, the pipeline cache was mostly being written as a big block of garbage data. Not very useful.

Next-Level Caching Technique

Clearly I needed to reeducate myself in the ways of managing a cache, something that, in my former life as a GUI expert, I did routinely, but that I could no longer comprehend now that I only speak bitfields and command buffers.

I sought out the reclusive Timothy Arceri, a well-known sage in many esoteric, arcane arts, and, as I recall it, purveyor of great wisdom such as (paraphrased because the original text has been lost to the ages): We Both Know The GLSL Compiler Code For Uniform Blocks Is Unfathomable, Why Do You Insist On Attempting To Modify It?

The answers I received from my sojourn were swift and concise:

Stop that. Fossilize caching wasn’t meant to work that way.

My thoughts whirling, confidence badly shaken, I stumbled and fell from the summit of the mountain and dashed my heretical cache implementation against the solid foundation of git rebase -i.

What had I been thinking?

It was back to the charts for me, and this time I had a number of different goals:

• go back to multi-file caching (since it’s the only option)
• smaller caches
• more frequent updates
• fully async

Turns out this wasn’t actually as hard as expected?

More Flowcharts (Fulfilling Image Quote For Graphics Blog)

Because we’re all big Vulkan adults, we do big Vulkan pipeline caches instead of wimpy OpenGL shader caches like so:

This has the added benefit of providing all the state variants for a given shader pipeline, saving additional lookups and ensuring that all the potential compiled pipelines are available at once. Furthermore, because there’s a (very) short delay between knowing what shaders are grouped together and needing the actual compiled pipeline, I can dump this all into a thread and handle the lookup while I update descriptors #2021 ASYNC THREADS++++++ BAYBEEEEEE.

But also, also, the one thing to absolutely not ever forget or else it’ll be really embarrassing is to ensure that you add your driver’s sha1 hash to your disk cache lookup, otherwise the whole thing explodes and @tarceri will frown down upon you.

Stop The Optimizing

I had planned to write more posts about some optimizations and whatever other cool stuff I’ve been working on.

I had planned to make more zink-wip snapshots.

I did shower; stop spamming frog emotes at me.

But I also encountered a bug so bizarre, so infruating, so esoteric, that I need to take a bit of a victory lap now that I’ve successfully corralled it. So let’s get into a real, vintage SGC blog post like we used to have back when SGC was a good blog and dive into what it took to fix a bug that took me four full days to resolve.

The Problem

In the course of writing a suballocator, I ran zero tests, as is my way. When the coding fugue ended, I stumbled weakly to my local CI script and managed to hit the Enter key before collapsing into a restless sleep where I was chased by angry triangles. Things were different when I awoke; namely I now had a lot of failing tests.

But I fixed them, because that’s what driver developers do.

All except one, which I assumed was a flake after running it a few times and seeing no failures.

This was how I got to know the horror of dEQP-GLES3.functional.vertex_array_objects.all_attributes.

The test itself is awful to debug. It generates GL_MAX_VERTEX_ATTRIBS vertex attributes to use with the maximum number of vertex buffers and does a series of draws, verifying the results. Normal enough.

Except the attributes are completely randomized, even whether they’re enabled, so no two runs are the same.

And the bisect hit just right.

The Basics Of Problem Solving

When a new bug is found with a driver, the first question is usually “Did this used to work?” followed quickly by “When did it start?” if the first answer was yes. The reason for this is that determining exactly when a problem began and what caused it to manifest gives the developer some vague starting point for determining what is happening to cause a bug.

So it was that I embarked on a bisect to figure out why dEQP-GLES3.functional.vertex_array_objects.all_attributes was suddenly failing. But obviously I couldn’t just bisect for this test. No, no, that would be far too easy.

This test only fails if run in conjunction with a series of other tests. Thus my deqp caselist file:

dEQP-GLES3.functional.vertex_array_objects.all_attributes
dEQP-GLES3.functional.vertex_arrays.single_attribute.first.byte.first6_offset1_stride2_quads256
dEQP-GLES3.functional.vertex_arrays.single_attribute.output_types.int.components2_ivec4_quads256
dEQP-GLES3.functional.vertex_arrays.single_attribute.usages.static_copy.stride32_fixed_quads1

The problem test always runs last, so something was clearly going on over time that was causing the failure. Armed with this knowledge, and so sure that this would end up being some trivial one-liner that I could fix in a few minutes, I set up my startpoint and endpoint for the bisect and went to work.

Zink By Bisection

Generally speaking, I assume every bug I find is going to be a zink bug. Just by the numbers, that’s usually the case, but then also it’s just always the case. It was therefore no surprise that my bisect landed on a certain commit:

commit 6b13e7cede95504ce8309744d8b9d83c7dbab7c9
Author: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Date:   Mon May 17 08:44:02 2021 -0400

    try better map flags

diff --git a/src/gallium/drivers/zink/zink_resource.c b/src/gallium/drivers/zink/zink_resource.c
index 55f37380d9f..121f6f0076e 100644
--- a/src/gallium/drivers/zink/zink_resource.c
+++ b/src/gallium/drivers/zink/zink_resource.c
@@ -1201,7 +1201,7 @@ buffer_transfer_map(struct zink_context *ctx, struct zink_resource *res, unsigne
          /* At this point, the buffer is always idle (we checked it above). */
          usage |= PIPE_MAP_UNSYNCHRONIZED;
       }
-   } else if ((usage & PIPE_MAP_READ) && !(usage & PIPE_MAP_PERSISTENT)) {
+   } else if (((usage & PIPE_MAP_READ) && !(usage & PIPE_MAP_PERSISTENT)) || !res->obj->host_visible) {
       assert(!(usage & (TC_TRANSFER_MAP_THREADED_UNSYNC | PIPE_MAP_THREAD_SAFE)));
       if (usage & PIPE_MAP_DONTBLOCK) {
          /* sparse/device-local will always need to wait since it has to copy */
@@ -1209,7 +1209,7 @@ buffer_transfer_map(struct zink_context *ctx, struct zink_resource *res, unsigne
             return NULL;
          if (!zink_resource_usage_check_completion(ctx, res, ZINK_RESOURCE_ACCESS_WRITE))
             return NULL;
-      } else if (!res->obj->host_visible) {
+      } else if (!res->obj->host_visible || res->base.b.usage != PIPE_USAGE_STAGING) {
          trans->staging_res = pipe_buffer_create(&screen->base, PIPE_BIND_LINEAR, PIPE_USAGE_STAGING, box->x + box->width);
          if (!trans->staging_res)
             return NULL;
@@ -1218,8 +1218,12 @@ buffer_transfer_map(struct zink_context *ctx, struct zink_resource *res, unsigne
          zink_copy_buffer(ctx, NULL, staging_res, res, box->x, box->x, box->width);
          res = staging_res;
          zink_fence_wait(&ctx->base);
-      } else
-         zink_resource_usage_wait(ctx, res, ZINK_RESOURCE_ACCESS_WRITE);
+      } else {
+         if (!(usage & PIPE_MAP_WRITE))
+            zink_resource_usage_wait(ctx, res, ZINK_RESOURCE_ACCESS_WRITE);
+         else
+            zink_resource_usage_wait(ctx, res, ZINK_RESOURCE_ACCESS_RW);
+      }
    }
 
    if (!ptr) {

As clearly explained by my laconic commit log, this patch aims to improve non-persistent buffer mappings by forcing non-staging resources to use a snooped staging resource. For more details on why this is desirable, check out this encyclopedia of wisdom on the topic, written by RADV co-founder and Commander Of The Rays, Bas Nieuwenhuizen.

But somehow this small patch was breaking the test, so I set out to investigate.

Isolation

Once a problem area is identified, it’s usually helpful to try and isolate the exact hunks of a patch which cause the problem. In this case, I had three distinct and only vaguely-related hunks, so it was an ideal case for this strategy. The middle hunk ended up being the culprit:

@@ -1209,7 +1209,7 @@ buffer_transfer_map(struct zink_context *ctx, struct zink_resource *res, unsigne
             return NULL;
          if (!zink_resource_usage_check_completion(ctx, res, ZINK_RESOURCE_ACCESS_WRITE))
             return NULL;
-      } else if (!res->obj->host_visible) {
+      } else if (!res->obj->host_visible || res->base.b.usage != PIPE_USAGE_STAGING) {
          trans->staging_res = pipe_buffer_create(&screen->base, PIPE_BIND_LINEAR, PIPE_USAGE_STAGING, box->x + box->width);
          if (!trans->staging_res)
             return NULL;

It seemed a bit odd to me, but nothing that stood out as impossible; perhaps there was some manner of issue with buffer copying offsets for setting up the staging resource, or some synchronization issue, or whatever. There were options, and I now knew that the problem was caused by setting up a staging buffer. Further prinfs revealed that this conditional was only hit for read access, so it was now narrowed down even further.

Initial Testing

Was it a buffer offset problem with copying the data for the staging resource?

Well.

No.

As interesting as it would’ve been for that to have been the case, there’s zero chance that this one test case was invoking a magical offset that wasn’t also triggered in other cases. If the general buffer copying code here was broken, it was probably broken everywhere in zink, so there would’ve been many, many more failures. There was only this one case, however, and deeper investigation confirmed this, as I directly mapped both buffers and compared the data ranges, which matched.

Synchronization it was, then, and I can hear the disembodied voice of Dave Airlie now shouting “Barriers!” before vanishing off into the ether.

First, I tried adding more GPU stalls. Like, lots more. Like, so many that the test took minutes to complete. There was no change, however. Just for hahas, I even added some usleep calls around.

Still nothing.

At this point I was seriously stumped. By now I’d fully instrumented all of the buffer access codepaths with asserts to verify the mapped contents matched the real buffer contents in all cases, and none of the asserts were ever hit.

But if it wasn’t actually an issue with synchronizing the staging buffer, what could it be?

I decided to check the test with with ANV at this point, being the case that I always run CTS against lavapipe to avoid killing my session in case I’ve foolished and added some code which triggers hangs, and…

And the test passed with ANV.

This was a real thinker, so I went to get a second opinion from Bas and RADV. RADV told me that ANV didn’t know what it was talking about, and the test was definitely failing, so I went with that answer because it seemed more sane.

As a final idea, I did the truly unthinkable: I threw in a malloc call, allocated some host memory, and copied the map contents directly into that buffer.

And leaked it.

Yes, I know, I know, We Don’t Do That, but it was just this one time. Just a little bit. Just to see if I could valg—Of course valgrind crashes when running anything in lavapipe due to unimplemented instructions, so why did I bother?

Getting Deeper

There comes a time when saying We Need To Go Deeper isn’t just a meme. That time was now, and I was about to get, as they say in technical terms when such depth is approached, deep as fuck.

Continuing to experiment with my memory leaking, the conditional block in question had by now degenerated into spaghetti:

         trans->staging_res = pipe_buffer_create(&screen->base, PIPE_BIND_LINEAR, PIPE_USAGE_STAGING, box->x + box->width);
         if (!trans->staging_res)
            return NULL;
         struct zink_resource *staging_res = zink_resource(trans->staging_res);
         trans->offset = staging_res->obj->offset;
         uint8_t *p = map_resource(screen, res);
         trans->data = malloc(box->x + box->width);
         trans->data2 = malloc(box->x + box->width);
         memset(trans->data, 0, box->x + box->width);
         memset(trans->data2, 0, box->x + box->width);
         memcpy(trans->data, p, box->x + box->width);
         memcpy(trans->data2, p, box->x + box->width);
         printf("SIZE NEEDED %u\n", box->x + box->width);
      for (unsigned i = 0; i < box->x + box->width; i++) {
         uint8_t *map = res->obj->map;
         assert(trans->data[i] == trans->data2[i]);
         assert(map[i] == trans->data2[i]);
         printf("MAP[%u] = %u\n", i, trans->data2[i]);
      }
         //zink_copy_buffer(ctx, NULL, staging_res, res, box->x, box->x, MIN2(box->width + 4, res->base.b.width0-box->x));
         //zink_copy_buffer(ctx, NULL, staging_res, res, box->x, box->x, box->width);
         ptr = trans->data;

         res = staging_res;
         zink_fence_wait(&ctx->base);

Obviously I’m gonna double buffer my memory leak so I can verify that it’s not secretly being modified on unmap (it wasn’t), and then also verify that the data matches before returning the pointer. And print it all, of course, because if you can actually read your terminal when you reach this sort of depth in the course of a debugging session, probably you’re doing it wrong.

But the time had come to start applying hacks elsewhere: namely the test itself. Being a random test case made it impossible to figure out what was going on between runs, but I’d determined one thing of interest: no matter what, unless I returned the direct mapping for the buffer, the test failed.

Let’s see what Mr. Crowbar had to say about that though when I applied him to the CTS case:

diff --git a/modules/gles3/functional/es3fVertexArrayObjectTests.cpp b/modules/gles3/functional/es3fVertexArrayObjectTests.cpp
index 82578b1ce..e231c4b1a 100644
--- a/modules/gles3/functional/es3fVertexArrayObjectTests.cpp
+++ b/modules/gles3/functional/es3fVertexArrayObjectTests.cpp
@@ -765,14 +765,14 @@ void MultiVertexArrayObjectTest::init (void)
 		m_spec.buffers.push_back(shortCoordBuffer48);
 
 		m_spec.state.attributes.push_back(Attribute());
-		m_spec.state.attributes[attribNdx].enabled		= (m_random.getInt(0, 4) == 0) ? GL_FALSE : GL_TRUE;
-		m_spec.state.attributes[attribNdx].size			= m_random.getInt(2,4);
-		m_spec.state.attributes[attribNdx].stride		= 2*m_random.getInt(1, 3);
+		m_spec.state.attributes[attribNdx].enabled		= GL_TRUE;
+		m_spec.state.attributes[attribNdx].size			= (attribNdx % 2) + 2;
+		m_spec.state.attributes[attribNdx].stride	= 2 * ((attribNdx % 2) + 1);
 		m_spec.state.attributes[attribNdx].type			= GL_SHORT;
-		m_spec.state.attributes[attribNdx].integer		= m_random.getBool();
-		m_spec.state.attributes[attribNdx].divisor		= m_random.getInt(0, 1);
-		m_spec.state.attributes[attribNdx].offset		= 2*m_random.getInt(0, 2);
-		m_spec.state.attributes[attribNdx].normalized	= m_random.getBool();
+		m_spec.state.attributes[attribNdx].integer		= attribNdx % 3 == 1;
+		m_spec.state.attributes[attribNdx].divisor		= 0;
+		m_spec.state.attributes[attribNdx].offset		= attribNdx % 5;
+		m_spec.state.attributes[attribNdx].normalized	= attribNdx % 3 == 1;
 		m_spec.state.attributes[attribNdx].bufferNdx	= attribNdx+1;
 
 		if (attribNdx == 0)
@@ -783,14 +783,14 @@ void MultiVertexArrayObjectTest::init (void)
 		}
 
 		m_spec.vao.attributes.push_back(Attribute());
-		m_spec.vao.attributes[attribNdx].enabled		= (m_random.getInt(0, 4) == 0) ? GL_FALSE : GL_TRUE;
-		m_spec.vao.attributes[attribNdx].size			= m_random.getInt(2,4);
-		m_spec.vao.attributes[attribNdx].stride			= 2*m_random.getInt(1, 3);
+		m_spec.vao.attributes[attribNdx].enabled		= GL_TRUE;
+		m_spec.vao.attributes[attribNdx].size			= (attribNdx % 2) + 2;
+		m_spec.vao.attributes[attribNdx].stride			= 2 * ((attribNdx % 2) + 1);
 		m_spec.vao.attributes[attribNdx].type			= GL_SHORT;
-		m_spec.vao.attributes[attribNdx].integer		= m_random.getBool();
-		m_spec.vao.attributes[attribNdx].divisor		= m_random.getInt(0, 1);
-		m_spec.vao.attributes[attribNdx].offset			= 2*m_random.getInt(0, 2);
-		m_spec.vao.attributes[attribNdx].normalized		= m_random.getBool();
+		m_spec.vao.attributes[attribNdx].integer		= attribNdx % 3 == 1;
+		m_spec.vao.attributes[attribNdx].divisor		= 0;
+		m_spec.vao.attributes[attribNdx].offset			= attribNdx % 5;
+		m_spec.vao.attributes[attribNdx].normalized		= attribNdx % 3 == 1;
 		m_spec.vao.attributes[attribNdx].bufferNdx		= attribCount - attribNdx;
 
 		if (attribNdx == 0)

Now I had a consistently failing test (as long as I ran it with the other test cases so it didn’t feel too lonely and accidentally pass) with consistent data, and I was dumping it all to logs that I could compare if I returned the direct pointer for the map to legitimately pass the test.

Naturally the output data that I was printing matched. It’d be pretty weird if it didn’t considering all the asserts that I had in the code, right? Hah, yeah, that’d be… That’d be pretty weird, all right…

The Forgotten Depths

By this point I had determined that it was a specific range of buffer mappings causing the problem, specifically those sized between 50 and 100 bytes. I also knew that these buffers were being mapped by u_vbuf, also known colloquially as the hinterlands of Gallium, an obscure component used to handle translating unsupported vertex buffer formats.

Veteran Mesa developers reading along are going full sensiblechuckle.gif right now, but I’ll request that we continue our no spoiler policy.

If the buffer contents were the same as the mapped contents but the test was still failing, then there had to be a reason for that. I fumbled my way over to the vertex attribute translator and fingerpainted in a printf to dump the translated vertex attributes. This enabled me to diff between a good run and a bad run.

It was then that I made a bewildering discovery.

Any time I had a 96-byte buffer map, the attributes starting at offset 92 didn’t match in cases when the test failed.

This was another thinker, so I decided to enhance my memory leaks a bit to copy more buffer since this was all 4096-aligned and it wasn’t like I was going to be copying out of bounds. This was when things started to get really weird.

Returning a copy of the resquested 96 bytes of the buffer failed the test, but returning 100 bytes passed it.

Uh-oh.

Now that I took a closer look at those vertex attribs, I realized that the ones which were failing were the ones that were read from bytes 96 and 97 of the buffer. The buffer which only had 96 bytes mapped, meaning that only the range of [0..95] was valid…

At Last

Resolution. What I had tripped over was a buffer overrun, one that was undetectable through normal means because of reasons like:

• this is a GPU buffer, so tools which would normally catch buffer overruns wouldn’t detect it
• this is u_vbuf, which is code that’s generally known to work pretty well given that it’s 10+ years old and is widely used and tested
• RadeonSI is likely the only other driver which uses the same sorts of buffer mapping optimizations, and it doesn’t use u_vbuf

Iteration on various fixes finally yielded a patch that was upstreamable; the crux of the problem here was that the stride of vertex attributes was being used to calculate the size of the region to map, but the stride only determines the number of bytes between elements, not their size. For example, if the stride was 4 bytes but the element was 8 bytes, the overrun would be 4 bytes for the last element. The solution was to calculate the offset of the last element being mapped, then add the size of the element using the attribute’s format block size, which guarantees that the last attribute won’t be truncated.

Fuck that bug.

So we are looking to hire quite a few people into the Desktop team currently. First of all we are looking to hire two graphics engineers to help us work on Linux Graphics drivers. The first of those two jobs is now online on the Red Hat jobs site. This is a job in our core graphics team focusing on RHEL, Fedora and upstream around the Intel, AMD and NVidia open source drivers. This is an opportunity to join a team of incredibly talented engineers working on everything from the graphics system of the Linux kernel and on the userspace bits like Vulkan, OpenGL and Wayland.  The job is listed as Senior Engineer, but for the right candidate we have flexibility there. We also have flexibility for people who want to work remotely, so as long as there is a Red Hat office in your home country you can work remotely for us.  The second job, which we hope to have up soon, will be looking more at ARM graphics and be tied to our automotive effort, but we will be looking at the applications for either position in combination so feel free to apply for the already listed job even if you are more interested in the second one as we will discuss both jobs with potential candidates.

The second job we have up is for – Software Engineer – GPU, Input and Multimedia which is also for joining our Graphics team. This job is targetted at our  office in Brno, Czechia and is a great entry level position if you are interested in the field of graphics. The job listing can be found here and outlines the kind of things we want you to look at, but do expect initially your job will be focused on helping the rest of the team manage their backlog and then grow from there.

The last job we have online now is for the automotive team, where we are looking for someone at the Senior/Principal level to join our Infotainment team, working with car makers around issues related to multimedia and help identifying needs and gaps and then work with upstream communities to figure out how we can resolve those issues. The job is targeted at Madrid, Spain as it is where we hope to center some of the infotainment effort and it makes things easier in terms of hardware access and similar, but for the right candidate we might be open to looking for candidates wanting to work remote or in another Red Hat office. You can find this job listing here.

We expect to be posting further jobs for the infotainment team within a week or two, so I will update once they are up.

Using Power For Evil

There’s no shortage of very smart people working on Mesa. One of those, aspiring benchmark-quadrupler Marek Olšák, had a novel idea some time ago: Could C++ function templates were used to optimize draw dispatch in driver?

The answer was yes, and so began what was probably five or ten minutes of furiously jamming brackets and braces into a C++ file in order to achieve the intended result. Let’s check out what’s going on here.

Setup

To start, the templates must be accessible from C, as this is what the driver is written in. The methodology here is simple: Generate the templates as an array of function pointers such that they can be accessed by indexing the arrays with the template values. Here’s what the code looks like:

template <chip_class GFX_VERSION, si_has_tess HAS_TESS, si_has_gs HAS_GS,
          si_has_ngg NGG, si_has_prim_discard_cs ALLOW_PRIM_DISCARD_CS>
static void si_init_draw_vbo(struct si_context *sctx)
{
   /* Prim discard CS is only useful on gfx7+ because gfx6 doesn't have async compute. */
   if (ALLOW_PRIM_DISCARD_CS && GFX_VERSION < GFX7)
      return;

   if (NGG && GFX_VERSION < GFX10)
      return;

   sctx->draw_vbo[GFX_VERSION - GFX6][HAS_TESS][HAS_GS][NGG][ALLOW_PRIM_DISCARD_CS] =
      si_draw_vbo<GFX_VERSION, HAS_TESS, HAS_GS, NGG, ALLOW_PRIM_DISCARD_CS>;
}

template <chip_class GFX_VERSION, si_has_tess HAS_TESS, si_has_gs HAS_GS>
static void si_init_draw_vbo_all_internal_options(struct si_context *sctx)
{
   si_init_draw_vbo<GFX_VERSION, HAS_TESS, HAS_GS, NGG_OFF, PRIM_DISCARD_CS_OFF>(sctx);
   si_init_draw_vbo<GFX_VERSION, HAS_TESS, HAS_GS, NGG_OFF, PRIM_DISCARD_CS_ON>(sctx);
   si_init_draw_vbo<GFX_VERSION, HAS_TESS, HAS_GS, NGG_ON, PRIM_DISCARD_CS_OFF>(sctx);
   si_init_draw_vbo<GFX_VERSION, HAS_TESS, HAS_GS, NGG_ON, PRIM_DISCARD_CS_ON>(sctx);
}

template <chip_class GFX_VERSION>
static void si_init_draw_vbo_all_pipeline_options(struct si_context *sctx)
{
   si_init_draw_vbo_all_internal_options<GFX_VERSION, TESS_OFF, GS_OFF>(sctx);
   si_init_draw_vbo_all_internal_options<GFX_VERSION, TESS_OFF, GS_ON>(sctx);
   si_init_draw_vbo_all_internal_options<GFX_VERSION, TESS_ON, GS_OFF>(sctx);
   si_init_draw_vbo_all_internal_options<GFX_VERSION, TESS_ON, GS_ON>(sctx);
}

static void si_init_draw_vbo_all_families(struct si_context *sctx)
{
   si_init_draw_vbo_all_pipeline_options<GFX6>(sctx);
   si_init_draw_vbo_all_pipeline_options<GFX7>(sctx);
   si_init_draw_vbo_all_pipeline_options<GFX8>(sctx);
   si_init_draw_vbo_all_pipeline_options<GFX9>(sctx);
   si_init_draw_vbo_all_pipeline_options<GFX10>(sctx);
   si_init_draw_vbo_all_pipeline_options<GFX10_3>(sctx);
}

static void si_invalid_draw_vbo(struct pipe_context *pipe,
                                const struct pipe_draw_info *info,
                                const struct pipe_draw_indirect_info *indirect,
                                const struct pipe_draw_start_count *draws,
                                unsigned num_draws)
{
   unreachable("vertex shader not bound");
}

extern "C"
void si_init_draw_functions(struct si_context *sctx)
{
   si_init_draw_vbo_all_families(sctx);

   /* Bind a fake draw_vbo, so that draw_vbo isn't NULL, which would skip
    * initialization of callbacks in upper layers (such as u_threaded_context).
    */
   sctx->b.draw_vbo = si_invalid_draw_vbo;
   sctx->blitter->draw_rectangle = si_draw_rectangle;

   si_init_ia_multi_vgt_param_table(sctx);
}

This calls through a series of functions, ultimately reaching si_init_draw_vbo where a template is set to a member of the function pointer array based on the template parameters. Specialized functions can then be generated based on hardware type, pipeline shader presence, and more.

Application

Once initialized, there’s an inline function used to set the current function pointer:

static inline void si_select_draw_vbo(struct si_context *sctx)
{
   sctx->b.draw_vbo = sctx->draw_vbo[sctx->chip_class - GFX6]
                                    [!!sctx->shader.tes.cso]
                                    [!!sctx->shader.gs.cso]
                                    [sctx->ngg]
                                    [si_compute_prim_discard_enabled(sctx)];
   assert(sctx->b.draw_vbo);
}

Thus the parameters are pulled directly from the context, and the function can be called whenever the draw function pointer needs to be updated, such as when new shaders are bound or primitive discard is enabled.

Result

The result is that now the draw dispatch can be fully optimized for the codepath required by the active hardware and graphics pipeline, reducing the CPU overhead and making the draw code the tiniest bit faster. For example, here’s just the top part of the templated function:

template <chip_class GFX_VERSION, si_has_tess HAS_TESS, si_has_gs HAS_GS, si_has_ngg NGG,
          si_has_prim_discard_cs ALLOW_PRIM_DISCARD_CS>
static void si_draw_vbo(struct pipe_context *ctx,
                        const struct pipe_draw_info *info,
                        unsigned drawid_offset,
                        const struct pipe_draw_indirect_info *indirect,
                        const struct pipe_draw_start_count_bias *draws,
                        unsigned num_draws)
{
   /* Keep code that uses the least number of local variables as close to the beginning
    * of this function as possible to minimize register pressure.
    *
    * It doesn't matter where we return due to invalid parameters because such cases
    * shouldn't occur in practice.
    */
   struct si_context *sctx = (struct si_context *)ctx;

   /* Recompute and re-emit the texture resource states if needed. */
   unsigned dirty_tex_counter = p_atomic_read(&sctx->screen->dirty_tex_counter);
   if (unlikely(dirty_tex_counter != sctx->last_dirty_tex_counter)) {
      sctx->last_dirty_tex_counter = dirty_tex_counter;
      sctx->framebuffer.dirty_cbufs |= ((1 << sctx->framebuffer.state.nr_cbufs) - 1);
      sctx->framebuffer.dirty_zsbuf = true;
      si_mark_atom_dirty(sctx, &sctx->atoms.s.framebuffer);
      si_update_all_texture_descriptors(sctx);
   }

   unsigned dirty_buf_counter = p_atomic_read(&sctx->screen->dirty_buf_counter);
   if (unlikely(dirty_buf_counter != sctx->last_dirty_buf_counter)) {
      sctx->last_dirty_buf_counter = dirty_buf_counter;
      /* Rebind all buffers unconditionally. */
      si_rebind_buffer(sctx, NULL);
   }

   si_decompress_textures(sctx, u_bit_consecutive(0, SI_NUM_GRAPHICS_SHADERS));
   si_need_gfx_cs_space(sctx, num_draws);

   /* If we're using a secure context, determine if cs must be secure or not */
   if (GFX_VERSION >= GFX9 && unlikely(radeon_uses_secure_bos(sctx->ws))) {
      bool secure = si_gfx_resources_check_encrypted(sctx);
      if (secure != sctx->ws->cs_is_secure(&sctx->gfx_cs)) {
         si_flush_gfx_cs(sctx, RADEON_FLUSH_ASYNC_START_NEXT_GFX_IB_NOW |
                               RADEON_FLUSH_TOGGLE_SECURE_SUBMISSION, NULL);
      }
   }

   if (HAS_TESS) {
      struct si_shader_selector *tcs = sctx->shader.tcs.cso;

      /* The rarely occuring tcs == NULL case is not optimized. */
      bool same_patch_vertices =
         GFX_VERSION >= GFX9 &&
         tcs && info->vertices_per_patch == tcs->info.base.tess.tcs_vertices_out;

      if (sctx->same_patch_vertices != same_patch_vertices) {
         sctx->same_patch_vertices = same_patch_vertices;
         sctx->do_update_shaders = true;
      }

      if (GFX_VERSION == GFX9 && sctx->screen->info.has_ls_vgpr_init_bug) {
         /* Determine whether the LS VGPR fix should be applied.
          *
          * It is only required when num input CPs > num output CPs,
          * which cannot happen with the fixed function TCS. We should
          * also update this bit when switching from TCS to fixed
          * function TCS.
          */
         bool ls_vgpr_fix =
            tcs && info->vertices_per_patch > tcs->info.base.tess.tcs_vertices_out;

         if (ls_vgpr_fix != sctx->ls_vgpr_fix) {
            sctx->ls_vgpr_fix = ls_vgpr_fix;
            sctx->do_update_shaders = true;
         }
      }

Note that the hardware version parts are templated, as is the HAS_TESS conditional, enabling it to be skipped entirely if there’s no tessellation shader active.

With techniques like this, it’s no surprise that RadeonSI is the driver to beat in performance and low overhead. The latest zink-wip snapshots include similar work, skipping considerable amounts of the draw dispatch when possible, and (hopefully) lowering the CPU overhead of the draw dispatch.

TL;DR: don't use select() + bump the RLIMIT_NOFILE soft limit to the hard limit in your modern programs.

The primary way to reference, allocate and pin runtime OS resources on Linux today are file descriptors ("fds"). Originally they were used to reference open files and directories and maybe a bit more, but today they may be used to reference almost any kind of runtime resource in Linux userspace, including open devices, memory (memfd_create(2)), timers (timefd_create(2)) and even processes (with the new pidfd_open(2) system call). In a way, the philosophically skewed UNIX concept of "everything is a file" through the proliferation of fds actually acquires a bit of sensible meaning: "everything has a file descriptor" is certainly a much better motto to adopt.

Because of this proliferation of fds, non-trivial modern programs tend to have to deal with substantially more fds at the same time than they traditionally did. Today, you'll often encounter real-life programs that have a few thousand fds open at the same time.

Like on most runtime resources on Linux limits are enforced on file descriptors: once you hit the resource limit configured via RLIMIT_NOFILE any attempt to allocate more is refused with the EMFILE error — until you close a couple of those you already have open.

Because fds weren't such a universal concept traditionally, the limit of RLIMIT_NOFILE used to be quite low. Specifically, when the Linux kernel first invokes userspace it still sets RLIMIT_NOFILE to a low value of 1024 (soft) and 4096 (hard). (Quick explanation: the soft limit is what matters and causes the EMFILE issues, the hard limit is a secondary limit that processes may bump their soft limit to — if they like — without requiring further privileges to do so. Bumping the limit further would require privileges however.). A limit of 1024 fds made fds a scarce resource: APIs tried to be careful with using fds, since you simply couldn't have that many of them at the same time. This resulted in some questionable coding decisions and concepts at various places: often secondary descriptors that are very similar to fds — but were not actually fds — were introduced (e.g. inotify watch descriptors), simply to avoid for them the low limits enforced on true fds. Or code tried to aggressively close fds when not absolutely needing them (e.g. ftw()/nftw()), losing the nice + stable "pinning" effect of open fds.

Worse though is that certain OS level APIs were designed having only the low limits in mind. The worst offender being the BSD/POSIX select(2) system call: it only works with fds in the numeric range of 0…1023 (aka FD_SETSIZE-1). If you have an fd outside of this range, tough luck: select() won't work, and only if you are lucky you'll detect that and can handle it somehow.

Linux fds are exposed as simple integers, and for most calls it is guaranteed that the lowest unused integer is allocated for new fds. Thus, as long as the RLIMIT_NOFILE soft limit is set to 1024 everything remains compatible with select(): the resulting fds will also be below 1024. Yay. If we'd bump the soft limit above this threshold though and at some point in time an fd higher than the threshold is allocated, this fd would not be compatible with select() anymore.

Because of that, indiscriminately increasing the soft RLIMIT_NOFILE resource limit today for every userspace process is problematic: as long as there's userspace code still using select() doing so will risk triggering hard-to-handle, hard-to-debug errors all over the place.

However, given the nowadays ubiquitous use of fds for all kinds of resources (did you know, an eBPF program is an fd? and a cgroup too? and attaching an eBPF program to cgroup is another fd? …), we'd really like to raise the limit anyway. 🤔

So before we continue thinking about this problem, let's make the problem more complex (…uh, I mean… "more exciting") first. Having just one hard and one soft per-process limit on fds is boring. Let's add more limits on fds to the mix. Specifically on Linux there are two system-wide sysctls: fs.nr_open and fs.file-max. (Don't ask me why one uses a dash and the other an underscore, or why there are two of them...) On today's kernels they kinda lost their relevance. They had some originally, because fds weren't accounted by any other counter. But today, the kernel tracks fds mostly as small pieces of memory allocated on userspace requests — because that's ultimately what they are —, and thus charges them to the memory accounting done anyway.

So now, we have four limits (actually: five if you count the memory accounting) on the same kind of resource, and all of them make a resource artificially scarce that we don't want to be scarce. So what to do?

Back in systemd v240 already (i.e. 2019) we decided to do something about it. Specifically:

• Automatically at boot we'll now bump the two sysctls to their maximum, making them effectively ineffective. This one was easy. We got rid of two pretty much redundant knobs. Nice!

• The RLIMIT_NOFILE hard limit is bumped substantially to 512K. Yay, cheap fds! You may have an fd, and you, and you as well, everyone may have an fd!

• But … we left the soft RLIMIT_NOFILE limit at 1024. We weren't quite ready to break all programs still using select() in 2019 yet. But it's not as bad as it might sound I think: given the hard limit is bumped every program can easily opt-in to a larger number of fds, by setting the soft limit to the hard limit early on — without requiring privileges.

So effectively, with this approach fds should be much less scarce (at least for programs that opt into that), and the limits should be much easier to configure, since there are only two knobs now one really needs to care about:

• Configure the RLIMIT_NOFILE hard limit to the maximum number of fds you actually want to allow a process.

• In the program code then either bump the soft to the hard limit, or not. If you do, you basically declare "I understood the problem, I promise to not use select(), drown me fds please!". If you don't then effectively everything remains as it always was.

Apparently this approach worked, since the negative feedback on change was even scarcer than fds traditionally were (ha, fun!). We got reports from pretty much only two projects that were bitten by the change (one being a JVM implementation): they already bumped their soft limit automatically to their hard limit during program initialization, and then allocated an array with one entry per possible fd. With the new high limit this resulted in one massive allocation that traditionally was just a few K, and this caused memory checks to be hit.

Anyway, here's the take away of this blog story:

• Don't use select() anymore in 2021. Use poll(), epoll, iouring, …, but for heaven's sake don't use select(). It might have been all the rage in the 1990s but it doesn't scale and is simply not designed for today's programs. I wished the man page of select() would make clearer how icky it is and that there are plenty of more preferably APIs.

• If you hack on a program that potentially uses a lot of fds, add some simple code somewhere to its start-up that bumps the RLIMIT_NOFILE soft limit to the hard limit. But if you do this, you have to make sure your code (and any code that you link to from it) refrains from using select(). (Note: there's at least one glibc NSS plugin using select() internally. Given that NSS modules can end up being loaded into pretty much any process such modules should probably be considered just buggy.)

• If said program you hack on forks off foreign programs, make sure to reset the RLIMIT_NOFILE soft limit back to 1024 for them. Just because your program might be fine with fds >= 1024 it doesn't mean that those foreign programs might. And unfortunately RLIMIT_NOFILE is inherited down the process tree unless explicitly set.

And that's all I have for today. I hope this was enlightening.

Click Play

It’s been a while.

I meant to blog. I meant to make new zink-wip snapshots. I meant to shower.

Look, none of us are perfect, and I’m just gonna get into some graphics so nobody remembers how this post started.

Boom, beautiful triangles. Look at that ultra smooth fps in mangohud. Protip: if you’re seeing weird flickering or misrenders in your app/game, try throwing mangohud in front of the zink bus to see if it fixes them.

So what has been going on for the past however since the last post?

In a word: lots.

Here’s the rundown.

The Rundown

The 20210517 zink-wip snapshot is the biggest one in history. I say this with no exaggeration.

Changes since the last snapshot include:

• an imperial units (and I measured this precisely) fuckton of general driver overhead reduction
• (yet another) queue/dispatch rewrite, this one more optimized for threaded and multi-context use
• an actually working disk cache implementation
• an entire suballocator

One way or another, this is going to feel like a new driver. Ideally I’ll be doing a post every day detailing one of the items on that list, but for now I’ll close the post by saying that zink should be 100%-1000% faster (not a typo) in most scenarios where it was previously much slower than native GL drivers.

Yeah, Big Triangle knows who we are now.

1. For some reason I decided that I did not have enough projects going on at the same time already and I started looking into improving support for the G15 family of keyboards.

2. I started studying the g15daemon code and did a build of it to check that it actually works. I believe making sure that you have a known-to-work codebase as a starting point, even if it is somewhat crufty and ugly, is important. This way you know you have code which speaks the right protocol and you can try to slowly morph it into what you want it to become (making small changes testing every step). Even if you decide to mostly re-implement from scratch, then you will likely use the old code as a protocol documentation and it is important to know it actually works.

3. There were number of things which I did not like about g15daemon:

3.1 It uses libusb, taking control of the entire USB-interface used for the gaming functionality away from the kernel. 

3.2 As part of this it was not just dealing with the LCD, it also was acting as a dispatches for G-key key-presses. IMHO the key-press handling clearly belonged in the kernel. These keys are just extra keys, all macro functionality is handled inside software on the host/PC side. So as a kernel dev I found that these really should be handled as normal keys and emit normal evdev event with KEY_FOO codes from a /dev/input/event# kernel node.

3.3 It is a daemon, rather then a library; and most other code which would deal with the LCD such as lcdproc was a daemon itself too, so now we would have lcdproc's LCDd talking to g15daemon to get to the LCD which felt rather roundabout.
So I set about tackling all of these

4. Kernel changes: I wrote a new drivers/hid/hid-lg-g15.c kernel driver:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/hid/hid-lg-g15.c
Which sends KEY_MACRO1 .. KEY_MACRO18, KEY_MACRO_PRESET1 .. KEY_MACRO_PRESET3, KEY_MACRO_RECORD_START, KEY_KBD_LCD_MENU1 .. KEY_KBD_LCD_MENU4 keypresses for all the special keys. Note this requires that the kernel HID driver is left attached to the USB interface, which required changes on the g15dameon/lcdproc side.

This kernel driver also offers a /sys/class/leds/g15::kbd_backlight/ interface which allows controller the backlight intensity through e.g. GNOME kbd-backlight slider in the power pane of the control-center. It also offers a bunch of other LED interfaces under /sys/class/leds/ for controlling things like the LEDs under the M1 - M3 preset selection buttons. The idea here being that the kernel abstract the USB protocol for these gaming-kbd away and that a single userspace daemon for doing gaming kbd macro functionality can be written which will rely only on the kernel interfaces and will thus work with any kbd which has kernel support.

5. lcdproc changes

5.1 lcdproc already did have a g15 driver, which talked to g15daemon. So at first I started testing with this (at this point all my kernel work would not work, since g15daemon would take full control of the USB interface unbinding my kernel driver). I did a couple of bug-fixes / cleanups in this setup to get the code to a starting point where I could run it and it would not show any visible rendering bugs in any of the standard lcdproc screens

5.2 I wrote a little lcdproc helper library, lib_hidraw, which can be used by lcdproc drivers to open a /dev/hidraw device for them. The main feature is, you give the lib_hidraw_open() helper a list of USB-ids you are interested in; and it will then find the right /dev/hidraw# device node (which may be different every boot) and open it for you.

5.3 The actual sending of the bitmap to the LCD screen is quite simple, but it does need to be a very specific format. The format rendering is done by libg15render. So now it was time to replace the code talking to g15daemon with code to directly talk to the LCD through a /dev/hidraw# interface. I kept the libg15render dependency since that was fine. After a bit of refactoring, the actual change over to directly sending the data to the LCD was not that big.

5.4 The change to stop using the g15daemon meant that the g15 driver also lost support for detecting presses on the 4 buttons directly under the LCD which are meant for controlling the menu on the LCD. But now that the code is using /dev/hidraw# the kernel driver I wrote would actually work and report KEY_KBD_LCD_MENU1 .. KEY_KBD_LCD_MENU4 keypresses. So I did a bunch of cleanup / refactoring of lcdproc's linux_input driver and made it take over the reporting of those button presses.

5.5 I wanted everything to just work out of the box, so I wrote some udev rules which automatically generate a lcdproc.conf file configuring the g15 + linux_input drivers (including the key-mapping for the linux_input driver) when a G15 keyboard gets plugged in and the lcdproc.conf file does not exist yet.

All this together means that users under Fedora, where I also packaged all this can now do "dnf install lcdproc", pluging their G15 keyboard and everything will just work.

In Turnips in the wild (Part 1) we walked through two issues, one in TauCeti Benchmark and the other in Genshin Impact. Today, I have an update about the one I didn’t have plan to fix, and a showcase of two remaining issues I met in Genshin Impact.

Genshin Impact

Gameplay – Disco Water

In the previous post I said that I’m not planning to fix the broken water effect since it relied on undefined behavior.

</figcaption></figure>

</figcaption></figure>

</figcaption></figure>

 [drm:a6xx_irq [msm]] *ERROR* gpu fault ring 0 fence ...

</figcaption></figure>

</figcaption></figure>

</figcaption></figure>

</figcaption></figure>

_243 = clamp(_243, 0.0, 1.0);
_279 = clamp(_279, 0.0, 1.0);
float _290;
if (_72.x) {
  _290 = _279;
} else {
  _290 = _243;
}

color0 = vec4(_290);
return;

mad.f32 r0.z, c0.y, r0.x, c6.w
sqrt r0.y, r0.y
mul.f r0.x, r1.y, c1.z
(ss)(nop2) mad.f32 r1.z, c6.x, r0.y, c6.y
(nop3) cmps.f.ge r0.y, r0.x, r1.w
(sat)(nop3) sel.b32 r0.w, r0.z, r0.y, r1.z

(sat)mad.f32 r0.z, c0.y, r0.x, c6.w
sqrt r0.y, r0.y
(ss)(sat)mad.f32 r1.z, c6.x, r0.y, c6.y
(nop2) mul.f r0.y, r1.y, c1.z
add.f r0.x, r0.z, r1.z
(nop3) cmps.f.ge r0.w, r0.y, r1.w
cov.u r1.w, r0.w
(rpt2)nop
(nop3) add.f r0.w, r0.x, r1.w

</figcaption></figure>

</figcaption></figure>

</figcaption></figure>

wew

After a traumatic, tearful goodbye to weird glxgears which took more out of me than expected, I’m back and blogging again.

It’s been…

Well, I guess it’s been a while since my last post, huh.

So what’s happened since then?

Lots? Yeah, definitely lots. It’s May now. And there’s even been a new Mesa release since then.

Quick zink roundup in case you missed any of these:

• shader clocks are in
• sparse buffer support is in
• 16bit ALU support is in
• GPU memory queries are in

Cool.

But What’s Really Been Going On?

The truth is that I’ve been taking some time off from zink in a completely futile attempt to make progress on something else while zink-wip continues to land. Inspired by this ticket describing issues getting CS:GO working, I decided to tackle part of Mesa that I haven’t worked on much and that hasn’t seen much work in a long time:

PBOs.

PBO TIME

Pixel Buffer Objects are used when an application needs to perform a transfer of an image from host memory to the GPU or vice versa. A PBO is said to be downloaded when it is copied from the GPU into a host memory buffer, and it is uploaded when it is copied from a memory buffer into GPU memory. PBO uploads are a common way to load assets from disk (e.g., textures), and PBO downloads are common for…ideally nothing performance related, but RPCS3 uses them in such a way.

Uploading of textures in Mesa can take a number of different codepaths depending on various factors using this priority chain:

• If the format and data type of the pixels is compatible with the format of the texture used for the upload, they can be directly copied by the driver. this is the fastest method
• If the driver supports shader images and the texture format is supported, Gallium will generate a fragment shader which binds the data-to-be-uploaded as a bufferview, the texture as a framebuffer attachment, and then samples to the framebuffer. this is pretty fast
• If the format of the host memory buffer (the data being uploaded) is supported by the driver, Gallium will create a staging texture, memcpy the pixel data into it, and then blit it to the target texture. now we’re slowing down a bit
• Finally, if all else fails, Mesa will just demand the driver maps the target texture so it can dump the pixel data in on the CPU, usually resulting in what looks like a frozen screen because of all the staging textures, flushing, and stalling going on in order to get the pixels where they need to go. this is the bad place

The reason CS:GO takes forever to start with zink is because it performs texture uploads of formats which zink does not support, namely alpha and luminance formats that cannot be emulated in Vulkan, during startup in order to load assets onto the GPU. This hits the CPU copy path 100% of the time, which is actually going to be slower than using something like llvmpipe because GPU-based graphics drivers are not intended to be doing everything on the CPU.

Set PBOs To Ludicrous Speed

I spent a while considering the problem at hand, then decided to start in the place where I could understand what was going on: namely texture downloads. In contrast to uploads, downloads have fewer and simpler codepaths to choose from:

• If the format and data type of the texture matches the requested pixel format, the data is copied directly with memcpy. this is the fastest method
• If the format and data type of the requested pixel format are supported by the driver and the texture format is also compatible with the requested format and type, Gallium creates a staging texture to blit to and then memcpy. this is still pretty okay
• Software fallback. oh no

So to effectively improve this with a new mechanism, all I had to do was meet two criteria:

• Ensure that all formats are supported so that the software fallback is not needed
• Ensure that all format conversions are supported so that the software fallback is not needed

There was also, of course, the additional checklist item of Don’t Be Slower Than The Existing Semi-Fastpath Implementation, but I figured I could get to that later.

The implementation I settled on, after much trial and error, was to set up a SSBO descriptor and then use the source texture as a sampler to texelFetch from in a compute shader. A small, vec4-sized constant buffer is passed to the shader, and then everything is done on the GPU to convert the image to the requested pixel format. The SSBO can then be copied to the output memory buffer, and boom, done.

It took some time to get right, but the results are already quite promising. Here’s some results from pbobench, a tool I’m working on to help measure PBO performance. At present, it populates a 1024x1024 pixel texture using GL_R32F data, then downloads (glGetTextureSubImage) it as fast as possible and measures the number of calls per second, iterating over a number of different formats and types for the download.

Here’s the latest results that I’ve run on IRIS using GL_PACK_ALIGNMENT==4 and GL_PACK_SWAP_BYTES==1 for variety.

32x32

The 32x32 size is not very favorable for the new implementation. Drivers are already extremely optimized for small PBO operations, so at best, my work here manages to keep up with the existing codebase for some cases.

Let’s go up a power of two to 64x64.

64x64 starts to show some interesting results, cases like GL_SRGB8 where the compute implementation has dominant performance.

By the 128x128 texture size, the massive performance lead that the existing glGetTextureSubImage handling had has dwindled to 20-50% for some cases, while the compute shader is now outperforming by a factor of ten or more for other cases.

256x256 is the point at which the difference starts to become even more pronounced. The cases where the compute implementation is definitively worse are few and far between, and many of the cases in which it was previously worse are now neck and neck.

512x512 is even better for the compute-based implementation, which remains extremely consistent across nearly all cases. It now exhibits a commanding lead of almost 1500% for some cases, while at worst, it maintains a consistent rate and achieves only 60% of baseline performance for a few cases.