Feed aggregator

Neat drm/i915 Stuff for 3.20

Daniel Vetter -

Linux 3.19 was just released and my usual overview of what the next merge window will bring is more than overdue. The big thing overall is certainly all the work around atomic display updates, but read on for what else all has been done.

Let's first start with all the driver internal rework to support atomic. The big thing with atomic is that it requires a clean split between code that checks display updates and the code that commits a new display state to the hardware. The corallary from that is that any derived state that's computed in the validation code and needed int the commit code must be stored somewhere in the state object. Gustavo Padovan and Matt Roper have done all that work to support atomic plane updates. This is the code that's now in 3.20 as a tech preview. The big things missing for proper atomic plane updates is async commit support (which has already landed for 3.21) and support to adjust watermark settings on the fly. Patches for from Ville have been around for a long time, but need to be rebased, reviewed and extended for recently added platforms.

On the modeset side Ander Conselvan de Oliveira has done a lot of the necessary work already. Unfortunately converting the modeset code is much more involved for mainly two reaons: First there's a lot more derived state that needs to be handled, and the existing code already has structures and code for this. Conceptually the code has been prepared for an atomic world since the big display rewrite and the introduction of CRTC configuration structures. But converting the i915 modeset code to the new DRM atomic structures and interface is still a lot of work. Most of these refactorings have landed in 3.20. The other hold-up is shared resources and the software state to handle that. This is mostly for handling display PLLs, but also other shared resources like the display FIFO buffers. Patches to handle this are still in-flight.

Continuing with modeset work Jani Nikula has reworked the DSI support to use the shared DSI helpers from the DRM core. Jani also reworked the DSI to in preparation for dual-link DSI support, which Gaurav Singh implemented. Rodrigo Vivi and others provided a lot of patches to improve PSR support and enable it for Baytrail/Braswell. Unfortunately there's still issues with the automated testcase and so PSR unfortunately stays disabled by default for now. Rodrigo also wrote some nice DocBook documentation for fbc, another step towards fully documenting i915 driver internals.

Moving on to platform enabling there has been a lot of work from Ville on Cherryview: RPS/gpu turbo and pipe CRC support (used for automated display testing) are both improved. On Skylake almost all the basic enabling is merged now: PM patches, enabling mmio pageflips and fastboot support from Damien have landed. Tvrtko Ursulin also create the infrastructure for global GTT views. This will be used for some upcoming features on Skylake. And to simplify enabling future platforms Chris Wilson and Mika Kuoppala have completely rewritten the forcewake domains handling code.

Also really important for Skylake is that the execlist support for gen8+ command submission is finally in a good enough to be used by default - on Skylake that's the only support path, legacy ring submission has been deprecated. Around that feature and a few other closely related ones a lot of code was refactoring: John Harrison implemented the conversion from raw sequence numbers to request objects for gpu progress tracking. This as is also prep work for landing a gpu scheduler. Nick Hoath removed the special execlist request tracking structures, simplifying the code. The engine initialization code was also refactored for a cleaner split between software and hardware initialization, leading to robuster code for driver load and system resume. Dave Gordon has also reworked the code tracking and computing the ring free space. On top of that we've also enabled full ppgtt again, but for now only where execlists are available since there are still issues with the legacy ring-based pagetable loading.

For generic GEM work there's the really nice support for write-combine cpu memory mappings from Akash Goel and Chris Wilson. On Atom SoC platforms lacking the giant LLC bigger chips have this gives a really fast way to upload textures. And even with LLC it's useful for uploading to scanout targets since those are always uncached. But like the special-purpose uploads paths for LLC platforms the cpu mmap views do not detile textures, hence special-purpose fastpaths need to be written in mesa and other userspace to fully exploit this. In other GEM features the shadow batch copying code for the command parser has now also landed.

Finally there's the redesign from Imre Deak to use the component framework for the snd-hda/i915 interactions. Modern platforms need a lot of coordination between the graphics and sound driver side for audio over hdmi, and thus far this was done with some ad-hoc dynamic symbol lookups. Which results in a lot of headaches to get the ordering correctly for driver load or system suspend and resume. With the component framework this depency is now explicit, which means we will be able to get rid of a few hacks. It's also much easier to extend for the future - new platforms tend to integrate different components even more.

Update for Atomic Display Updates

Daniel Vetter -

Another kernel release is imminent and a lot of things happened since my last big blog post about atomic modeset. Read on for what new shiny things 3.20 will bring this area.

Atomic IOCTL and Properties
The big thing for sure is that the actual atomic IOCTL from Ville has finally landed for 3.20. That together with all the work from Rob Clark to add all the new atomic properties to make it functional (there's no IOCTL structs for any standard values, everything is a property now) means userspace can finally start using atomic. Well, it's still hidden behind the experimental module option drm.atomic=1 but otherwise it's all there. There's a few big differences compared to earlier iterations:
  • Core properties are now fully handled by the core, drivers can only decode their own properties. This should mean that the handling of standardized properties should be more uniform and also makes it easier to extend (e.g. with the standardized rotation property we already have) beyond the paramaters the legacy interfaces provided.
  • Atomic props&ioctl are opt-in per file_priv, userspace needs to explicitly ask for it. This is the same idea as with universal plane support and will make sure that old userspace doesn't get confused and fall over when it would see all the new atomic properties. In the same spirit some of the legacy-only properties (like DPMS) will be rejected in the atomic IOCTL paths.
  • Atomic modesets are currently not possible since the exact ABI for how to handle the mode property is still under discussion. The big missing thing is handling blob properties, which will also be needed to upload gamma table updates through the atomic interface.
Another missing piece that's now resolved is the DPMS support. On a high level this greatly reduces the complexity of the legacy DPMS settings into a  simple per-CRTC boolean. Contemporary userspace really wants nothing more and anything but a simple on/off is the only thing that current hardware can support. Furthermore all the bookkeeping is done in the helpers, which call down into drivers to enable or disable entire display pipelines as needed. Which means that drivers can rip out tons of code for the legacy DPMS support by just wiring up drm_atomic_helper_connector_dpms.

New Driver Conversions and Backend Hooks
Another big thing for 3.20 is that driver support has moved forward greatly: Tegra has now most of the basic bits ready, MSM is already converted. Both still lack conversion and testing for DPMS since that landed very late though. There's a lot of prep work for exynos, but unfortunately nothing yet towards atomic support. And i915 is in the process of being converted to the new structures and will have very preliminary atomic plane updates support in 3.20, hidden behind a module option for now.

And all that work resulted in piles of little fixes. Like making sure that the legacy cursor IOCTLs don't suddenly stall a lot more, breaking old userspace. But we also added a bunch more hooks for driver backends to simplifiy driver code:
  • For drivers there's often a very big difference between a plane update, and disabling a plane. And with atomic state objects it's a bit more work to figure out when exactly you're doing a on to off transition. Thierry Redding added a new optional ->atomic_plane_disable() hook for drivers which will take care of all those disdinctions.
  • Mostly for hysterical raisins going back to the original Xrandr implementations for userspace mode setting the callbacks to enable and disable encoders and CRTCs used by the various helper libraries have really confusing names. And with the legacy helpers all kinds of strange semantics. Since the atomic helpers massively simplify things in this area it made a lot of sense to add a new set of ->enable() and ->disable() hooks, which are preferred if they're set. All the other hooks (namely ->prepare(), ->commit() and ->dpms()) will eventually be deprecated for atomic drivers. Note that ->mode_set is already superseeded by ->mode_set_nofb due to the explicit primary plane handling with atomic updates.
Finally driver conversions showed that vblank handling requirements imposed by the atomic helpers are a lot stricter than what userspace tends to cope with. i915 has recently reworked all its vblank handling and improved the core helpers with the addition of drm_crtc_vblank_off() and drm_crtc_vblank_on(). If these are used in the CRTC disable and enable hooks the vblank code will automatically reject vblank waits when the pipe is off. Which is the behaviour both the atomic helpers and the transitional helpers expect.

One complication is that on load the driver must ensure manually that the vblank state matches up with the hardware and atomic software state with a manual call to these functions. In the simple case where drivers reset everything to off (which is what the reset implementations provided by the atomic helpers presume) this just means calling drm_crtc_vblank_off() somewhen in the driver load code. For drivers that read out the actual hardware state they need to call either _off() or _on() matching on the actual display pipe status.

Future Work
Of course there's still a few things left to do before atomic display updates can be rolled out to the masses. And a few things that would be rather nice to have, too:
  • Support for blob properties so that modesets and a bunch of other neat things can finally be done.
  • Testing, and lots of it, before we enable atomic by default.
  • Thanks to DP MST connectors can be hotplugged, and there are still a lot of lifetime issues surrounding connector handling all over the drm subsystem. Atomic display code is unfortunately no exception.
  • And finally try to make the vblank code a bit more generally useful and then implement generic async atomic commit on top of that.
It's promising to keep interesting!

LCA 2015: Botching Up IOCTLs

Daniel Vetter -

So I'm stuck somewhere on an airport and jetlagged on my return trip from my very first LCA in Auckland - awesome conference, except for being on the wrong side of the globe: Very broad range of speakers, awesome people all around, great organization and since it's still a community conference none of the marketing nonsense and sales-pitch keynotes.

Also done a presentation about botching up ioctls. Compared to my blog post a bunch more details on technical issues and some overall comments on what's really important to avoid a v2 ioctl because v1 ended up being unsalvageable. Slides and video (curtesy the LCA video team).

2014Q4 Intel Graphics Stack has been released

Rodrigo Vivi -

The 2014Q4 contains many fixes and improvements, mainly for Broadwell. It also contains preparation for Skylake.

Xf86-video-intel has been updated with DRI3 disabled by default. During 2014Q3 validation many blocking bugs related to DRI3 was found so xf86-video-intel didn’t get updated because DRI3 was enabled by default at 2.99.915. Now with DRI3 disabled we can update xf86-video-intel again.

Components of this stack are:

Linux Kernel – 3.18.1 Mesa – 10.4 xf86-video-intel – 2.99.917 Libdrm – 2.4.58 Libva – 1.5.0 vaapi intel-driver – 1.5.0 Cairo – 1.12.16 Xserver Xorg – 1.16.0 Intel-gpu-tools – 1.9 For more details read the full release notes: https://01.org/linuxgraphics/downloads/2014/2014q4-intel-graphics-stack-release

Neat drm/i915 stuff for 3.19

Daniel Vetter -

So kernel version 3.18 is out the door and it's time for our regular look at what's in the next merge window.
First looking at new hardware the big item is basic Skylake support. There are still a few smalls things missing, but mostly it's there now. This has been contributed by Damien, Satheeshakrishna and a lot of other folks. Looking at other platforms there has also been a lot of changes for vlv/chv: Improved backlight code, completely refactored interrupt handling to bring it in line with other platforms, rewritten panel power sequencing code, all from Ville. Rodrigo contributed PSR support for vlv/chv together with a lot of other fixes for PSR. Unfortunately it's not yet again enabled by default.

Moving on to Broadwell and the render side of things, Mika and Arun provided patches to improve the render workaround code and bring the set of workarounds up to date. execlist (the new command submission support on Gen8+) is also being polished with the addition of on-demand pinning of context objects with patches from Thomas Daniel and Oscar Mateo. Finally the RPS/render-turbo code has seen a lot of polish from Imre with a few fixes from Tom O'Rourke.

Otherwise not a lot of really big things happened on the GEM side: Just a few patches to fix issues in ppgtt (unfortunately still not enabled by default anywhere due to fun with context switches). And there's a bit of prep work and reorg all over for new stuff landing hopefully soon.

Looking at overall infrastructure changes the big thing certainly is the preparations for atomic display updates. The drm core/driver interface for atomic and all the helper library code to convert drivers has landed in 3.19, and already some conversions. On the Intel side it's been just prep work under the hood thus far with patches from Ander to precompute display PLL state. The new code to use vblank evades for pagelips has also landed, which is needed for atomic plane updates. And prep patches from Gustavo Padovan started to split the low-level plane update functions into check and commit steps. Lots more patches from different people are in flight and some have been merged for 3.20 already.

Besides these driver internal changes for atomic there has been other work to improve the codebase: Imre reorganized our handlers for suspend, resume and thawing and freezing. Jani reworked the audio and eld code which is the gfx side of the puzzle needed to make audio over HDMI or DP work. Jesse provided patches to track infoframes more accurately, which is needed to correctly fastboot (i.e. without modesets if possible) on external screens.

For older machines Ville has spent a few spare cycles to make them more useful: GPU reset support for gen3/4 should mitigate some of the recent chromium crashes on mesa, and the modeset code on i830M might work correctly for the first time, ever.

And of course the usual pile of smaller fixes and improvements all over.

Not directly related to code or features is the start of documenting i915 driver internals: With this release we now have some of the interrupt handling, fifo underrun reporting, frontbuffer tracking and runtime pm support newly document. And there's lots more in-flight, so hopefully soonish this will be fairly useful.
Subscribe to Linux Graphics aggregator