On Thu, 24 Jun 2021 at 02:20, Daniel Vetter daniel.vetter@ffwll.ch wrote:

Docs for struct dma_resv are fairly clear:

"A reservation object can have attached one exclusive fence (normally associated with write operations) or N shared fences (read operations)."

https://dri.freedesktop.org/docs/drm/driver-api/dma-buf.html#reservation-obj...

Furthermore a review across all of upstream.

First of render drivers and how they set implicit fences:

• nouveau follows this contract, see in validate_fini_no_ticket()

                    nouveau_bo_fence(nvbo, fence, !!b->write_domains);
  
  

  and that last boolean controls whether the exclusive or shared fence slot is used.

• radeon follows this contract by setting

            p->relocs[i].tv.num_shared = !r->write_domain;
  
  

  in radeon_cs_parser_relocs(), which ensures that the call to ttm_eu_fence_buffer_objects() in radeon_cs_parser_fini() will do the right thing.

• vmwgfx seems to follow this contract with the shotgun approach of always setting ttm_val_buf->num_shared = 0, which means ttm_eu_fence_buffer_objects() will only use the exclusive slot.

• etnaviv follows this contract, as can be trivially seen by looking at submit_attach_object_fences()

• i915 is a bit a convoluted maze with multiple paths leading to i915_vma_move_to_active(). Which sets the exclusive flag if EXEC_OBJECT_WRITE is set. This can either come as a buffer flag for softpin mode, or through the write_domain when using relocations. It follows this contract.

• lima follows this contract, see lima_gem_submit() which sets the exclusive fence when the LIMA_SUBMIT_BO_WRITE flag is set for that bo

• msm follows this contract, see msm_gpu_submit() which sets the exclusive flag when the MSM_SUBMIT_BO_WRITE is set for that buffer

• panfrost follows this contract with the shotgun approach of just always setting the exclusive fence, see panfrost_attach_object_fences(). Benefits of a single engine I guess

• v3d follows this contract with the same shotgun approach in v3d_attach_fences_and_unlock_reservation(), but it has at least an XXX comment that maybe this should be improved

• v4c uses the same shotgun approach of always setting an exclusive fence, see vc4_update_bo_seqnos()

• vgem also follows this contract, see vgem_fence_attach_ioctl() and the VGEM_FENCE_WRITE. This is used in some igts to validate prime sharing with i915.ko without the need of a 2nd gpu

• vritio follows this contract again with the shotgun approach of always setting an exclusive fence, see virtio_gpu_array_add_fence()

This covers the setting of the exclusive fences when writing.

Synchronizing against the exclusive fence is a lot more tricky, and I only spot checked a few:

• i915 does it, with the optional EXEC_OBJECT_ASYNC to skip all implicit dependencies (which is used by vulkan)

• etnaviv does this. Implicit dependencies are collected in submit_fence_sync(), again with an opt-out flag ETNA_SUBMIT_NO_IMPLICIT. These are then picked up in etnaviv_sched_dependency which is the drm_sched_backend_ops->dependency callback.

• v4c seems to not do much here, maybe gets away with it by not having a scheduler and only a single engine. Since all newer broadcom chips than the OG vc4 use v3d for rendering, which follows this contract, the impact of this issue is fairly small.

• v3d does this using the drm_gem_fence_array_add_implicit() helper, which then it's drm_sched_backend_ops->dependency callback v3d_job_dependency() picks up.

• panfrost is nice here and tracks the implicit fences in panfrost_job->implicit_fences, which again the drm_sched_backend_ops->dependency callback panfrost_job_dependency() picks up. It is mildly questionable though since it only picks up exclusive fences in panfrost_acquire_object_fences(), but not buggy in practice because it also always sets the exclusive fence. It should pick up both sets of fences, just in case there's ever going to be a 2nd gpu in a SoC with a mali gpu. Or maybe a mali SoC with a pcie port and a real gpu, which might actually happen eventually. A bug, but easy to fix. Should probably use the drm_gem_fence_array_add_implicit() helper.

• lima is nice an easy, uses drm_gem_fence_array_add_implicit() and the same schema as v3d.

• msm is mildly entertaining. It also supports MSM_SUBMIT_NO_IMPLICIT, but because it doesn't use the drm/scheduler it handles fences from the wrong context with a synchronous dma_fence_wait. See submit_fence_sync() leading to msm_gem_sync_object(). Investing into a scheduler might be a good idea.

• all the remaining drivers are ttm based, where I hope they do appropriately obey implicit fences already. I didn't do the full audit there because a) not follow the contract would confuse ttm quite well and b) reading non-standard scheduler and submit code which isn't based on drm/scheduler is a pain.

Onwards to the display side.

• Any driver using the drm_gem_plane_helper_prepare_fb() helper will correctly. Overwhelmingly most drivers get this right, except a few totally dont. I'll follow up with a patch to make this the default and avoid a bunch of bugs.

• I didn't audit the ttm drivers, but given that dma_resv started there I hope they get this right.

In conclusion this IS the contract, both as documented and overwhelmingly implemented, specically as implemented by all render drivers except amdgpu.

Amdgpu tried to fix this already in

commit 049aca4363d8af87cab8d53de5401602db3b9999 Author: Christian König christian.koenig@amd.com Date: Wed Sep 19 16:54:35 2018 +0200

drm/amdgpu: fix using shared fence for exported BOs v2

but this fix falls short on a number of areas:

• It's racy, by the time the buffer is shared it might be too late. To make sure there's definitely never a problem we need to set the fences correctly for any buffer that's potentially exportable.

• It's breaking uapi, dma-buf fds support poll() and differentitiate between, which was introduced in

    commit 9b495a5887994a6d74d5c261d012083a92b94738
    Author: Maarten Lankhorst <maarten.lankhorst@canonical.com>
    Date:   Tue Jul 1 12:57:43 2014 +0200
  
        dma-buf: add poll support, v3
  
  

• Christian König wants to nack new uapi building further on this dma_resv contract because it breaks amdgpu, quoting

  "Yeah, and that is exactly the reason why I will NAK this uAPI change.

  "This doesn't works for amdgpu at all for the reasons outlined above."

  https://lore.kernel.org/dri-devel/f2eb6751-2f82-9b23-f57e-548de5b729de@gmail...

  Rejecting new development because your own driver is broken and violates established cross driver contracts and uapi is really not how upstream works.

Now this patch will have a severe performance impact on anything that runs on multiple engines. So we can't just merge it outright, but need a bit a plan:

• amdgpu needs a proper uapi for handling implicit fencing. The funny thing is that to do it correctly, implicit fencing must be treated as a very strange IPC mechanism for transporting fences, where both setting the fence and dependency intercepts must be handled explicitly. Current best practices is a per-bo flag to indicate writes, and a per-bo flag to to skip implicit fencing in the CS ioctl as a new chunk.

• Since amdgpu has been shipping with broken behaviour we need an opt-out flag from the butchered implicit fencing model to enable the proper explicit implicit fencing model.

• for kernel memory fences due to bo moves at least the i915 idea is to use ttm_bo->moving. amdgpu probably needs the same.

• since the current p2p dma-buf interface assumes the kernel memory fence is in the exclusive dma_resv fence slot we need to add a new fence slot for kernel fences, which must never be ignored. Since currently only amdgpu supports this there's no real problem here yet, until amdgpu gains a NO_IMPLICIT CS flag.

• New userspace needs to ship in enough desktop distros so that users wont notice the perf impact. I think we can ignore LTS distros who upgrade their kernels but not their mesa3d snapshot.

• Then when this is all in place we can merge this patch here.

What is not a solution to this problem here is trying to make the dma_resv rules in the kernel more clever. The fundamental issue here is that the amdgpu CS uapi is the least expressive one across all drivers (only equalled by panfrost, which has an actual excuse) by not allowing any userspace control over how implicit sync is conducted.

Until this is fixed it's completely pointless to make the kernel more clever to improve amdgpu, because all we're doing is papering over this uapi design issue. amdgpu needs to attain the status quo established by other drivers first, once that's achieved we can tackle the remaining issues in a consistent way across drivers.

v2: Bas pointed me at AMDGPU_GEM_CREATE_EXPLICIT_SYNC, which I entirely missed.

This is great because it means the amdgpu specific piece for proper implicit fence handling exists already, and that since a while. The only thing that's now missing is

• fishing the implicit fences out of a shared object at the right time
• setting the exclusive implicit fence slot at the right time.

Jason has a patch series to fill that gap with a bunch of generic ioctl on the dma-buf fd:

https://lore.kernel.org/dri-devel/20210520190007.534046-1-jason@jlekstrand.n...

v3: Since Christian has fixed amdgpu now in

commit 8c505bdc9c8b955223b054e34a0be9c3d841cd20 (drm-misc/drm-misc-next) Author: Christian König christian.koenig@amd.com Date: Wed Jun 9 13:51:36 2021 +0200

drm/amdgpu: rework dma_resv handling v3

Use the audit covered in this commit message as the excuse to update the dma-buf docs around dma_buf.resv usage across drivers.

Since dynamic importers have different rules also hammer these in again while we're at it.

v4:

• Add the missing "through the device" in the dynamic section that I overlooked.
• Fix a kerneldoc markup mistake, the link didn't connect

This is pretty epic commit msg, thanks for the investment, the commit msg should be required reading.

Reviewed-by: Dave Airlie airlied@redhat.com

Dave.

On Thu, Jun 24, 2021 at 1:08 PM Daniel Stone daniel@fooishbar.org wrote:

Hi,

On Wed, 23 Jun 2021 at 17:20, Daniel Vetter daniel.vetter@ffwll.ch wrote:

...

•    *
  

•    * IMPLICIT SYNCHRONIZATION RULES:
  

•    *
  

•    * Drivers which support implicit synchronization of buffer access as
  

•    * e.g. exposed in `Implicit Fence Poll Support`_ should follow the
  

•    * below rules.
  

'Should' ... ? Must.

Yeah I guess I can upgrade a bunch of them.

...

•    * - Drivers should add a shared fence through
  

•    *   dma_resv_add_shared_fence() for anything the userspace API
  

•    *   considers a read access. This highly depends upon the API and
  

•    *   window system: E.g. OpenGL is generally implicitly synchronized on
  

•    *   Linux, but explicitly synchronized on Android. Whereas Vulkan is
  

•    *   generally explicitly synchronized for everything, and window system
  

•    *   buffers have explicit API calls (which then need to make sure the
  

•    *   implicit fences store here in @resv are updated correctly).
  

•    *
  

•    * - [...]
  

Mmm, I think this is all right, but it could be worded much more clearly. Right now it's a bunch of points all smashed into one, and there's a lot of room for misinterpretation.

Here's a strawman, starting with most basic and restrictive, working through to when you're allowed to wriggle your way out:

Rule 1: Drivers must add a shared fence through dma_resv_add_shared_fence() for any read accesses against that buffer. This appends a fence to the shared array, ensuring that any future non-read access will be synchronised against this operation to only begin after it has completed.

Rule 2: Drivers must add an exclusive fence through dma_resv_add_excl_fence() for any write accesses against that buffer. This replaces the exclusive fence with the new operation, ensuring that all future access will be synchronised against this operation to only begin after it has completed.

Rule 3: Drivers must synchronise all accesses to buffers against existing implicit fences. Read accesses must synchronise against the exclusive fence (read-after-write), and write accesses must synchronise against both the exclusive (write-after-write) and shared (write-after-read) fences.

Note 1: Users like OpenGL and window systems on non-Android userspace are generally implicitly synchronised. An implicitly-synchronised userspace is unaware of fences from prior operations, so the kernel mediates scheduling to create the illusion that GPU work is FIFO. For example, an application will flush and schedule GPU write work to render its image, then immediately tell the window system to display that image; the window system may immediately flush and schedule GPU read work to display that image, with neither waiting for the write to have completed. The kernel provides coherence by synchronising the read access against the write fence in the exclusive slot, so that the image displayed is correct.

Note 2: Users like Vulkan and Android window system are generally explicitly synchronised. An explicitly-synchronised userspace is responsible for tracking its own read and write access and providing the kernel with synchronisation barriers. For instance, a Vulkan application rendering to a buffer and subsequently using it as a read texture, must annotate the read operation with a read-after-write synchronisation barrier.

Note 3: Implicit and explicit userspace can coexist. For instance, an explicitly-synchronised Vulkan application may be running as a client of an implicitly-synchronised window system which uses OpenGL for composition; an implicitly-synchronised OpenGL application may be running as a client of a window system which uses Vulkan for composition.

Note 4: Some subsystems, for example V4L2, do not pipeline operations, and instead only return to userspace when the scheduled work against a buffer has fully retired.

Exemption 1: Fully self-coherent userspace may skip implicit synchronisation barriers. For instance, accesses between two Vulkan-internal buffers allocated by a single application do not need to synchronise against each other's implicit fences, as the client is responsible for explicitly providing barriers for access. A self-contained OpenGL userspace also has no need to implicitly synchronise its access if the driver instead tracks all access and inserts the appropriate synchronisation barriers.

Exemption 2: When implicit and explicit userspace coexist, the explicit side may skip intermediate synchronisation, and only place synchronisation barriers at transition points. For example, a Vulkan compositor displaying a buffer from an OpenGL application would need to synchronise its first access against the fence placed in the exclusive implicit-synchronisation slot. Once this read has fully retired, the compositor has no need to participate in implicit synchronisation until it is ready to return the buffer to the application, at which point it must insert all its non-retired accesses into the shared slot, which the application will then synchronise future write accesses against.

So I think this is excellent, but maybe better suited in the uapi section as a sperate chapter? Essentially keep your rules in the driver-internal docs, but move the Note/excemptions into the uapi section under a "Implicit Sync Mode of Operation" or whatever heading?

The other thing to keep in mind is that this is very much incomplete: I'm silent on what drivers should do exactly with these fences. That's largely because I haven't fully completed that audit, and there's a pile of bugs there still. -Daniel