Intel-GFX Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
To: Natalie Vock <natalie.vock@gmx.de>,
	Maarten Lankhorst	 <maarten.lankhorst@linux.intel.com>,
	Maxime Ripard <mripard@kernel.org>,
	 Thomas Zimmermann <tzimmermann@suse.de>,
	David Airlie <airlied@gmail.com>, Simona Vetter <simona@ffwll.ch>,
	Jani Nikula	 <jani.nikula@linux.intel.com>,
	Joonas Lahtinen <joonas.lahtinen@linux.intel.com>,
	 Rodrigo Vivi <rodrigo.vivi@intel.com>,
	Tvrtko Ursulin <tursulin@ursulin.net>,
	Christian Koenig	 <christian.koenig@amd.com>,
	Huang Rui <ray.huang@amd.com>,
	Matthew Auld	 <matthew.auld@intel.com>,
	Matthew Brost <matthew.brost@intel.com>,
	Alex Deucher	 <alexander.deucher@amd.com>
Cc: dri-devel@lists.freedesktop.org, linux-kernel@vger.kernel.org,
	 intel-gfx@lists.freedesktop.org, intel-xe@lists.freedesktop.org,
	 amd-gfx@lists.freedesktop.org
Subject: Re: [PATCH 09/10] drm/ttm: support using drm_exec during eviction v4
Date: Tue, 07 Jul 2026 14:48:50 +0200	[thread overview]
Message-ID: <944807b30e00606dc3f4a8ad57c3e8e70ddd2d10.camel@linux.intel.com> (raw)
In-Reply-To: <20260703-ttm_2_drm_exec-v1-9-43685ac1286b@gmx.de>

On Fri, 2026-07-03 at 18:31 +0200, Natalie Vock wrote:
> From: Christian König <christian.koenig@amd.com>
> 
> Allow specifying a drm_exec object in TTMs operation context which is
> used to lock objects during eviction.
> 
> This allows to handle deadlocks much more gracefully and with that
> avoid returning -ENOMEM on heavily contended domains.
> 
> v2: rebased on top of Thomas work
> v3: rebased again
> v4: rebased, fixed locks of already-reserved buffers being dropped
> 
> Signed-off-by: Christian König <christian.koenig@amd.com>
> Signed-off-by: Natalie Vock <natalie.vock@gmx.de>
> ---
>  drivers/gpu/drm/ttm/ttm_bo_util.c | 30 ++++++++++++++++++++++-------
> -
>  include/drm/ttm/ttm_bo.h          |  5 +++++
>  2 files changed, 27 insertions(+), 8 deletions(-)
> 
> diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c
> b/drivers/gpu/drm/ttm/ttm_bo_util.c
> index a53b25e8c2967..96699532817c2 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo_util.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
> @@ -38,6 +38,7 @@
>  #include <drm/ttm/ttm_tt.h>
>  
>  #include <drm/drm_cache.h>
> +#include <drm/drm_exec.h>
>  
>  #include "ttm_bo_internal.h"
>  
> @@ -837,6 +838,8 @@ static bool ttm_lru_walk_trylock(struct
> ttm_bo_lru_cursor *curs,
>  	struct ttm_operation_ctx *ctx = curs->arg->ctx;
>  
>  	curs->needs_unlock = false;
> +	if (ctx->exec)
> +		return false;

The first passes of eviction typically trylocks, and no_wait_gpu also
skips sleeping locks since they can be held why waiting for fences.
Otoh we could check that no_wait_gpu and exec is mutually exclusive.

But I think trylocking should be tried before exec locking to also
avoid costly rollbacks, in line with mutex lock stealing.

>  
>  	if (dma_resv_trylock(bo->base.resv)) {
>  		curs->needs_unlock = true;
> @@ -857,7 +860,9 @@ static int ttm_lru_walk_ticketlock(struct
> ttm_bo_lru_cursor *curs,
>  	struct ttm_lru_walk_arg *arg = curs->arg;
>  	int ret;
>  
> -	if (arg->ctx->interruptible)
> +	if (arg->ctx->exec)
> +		ret = drm_exec_lock_obj_report_dup(arg->ctx->exec,
> &bo->base);
> +	else if (arg->ctx->interruptible)
>  		ret = dma_resv_lock_interruptible(bo->base.resv,
> arg->ticket);
>  	else
>  		ret = dma_resv_lock(bo->base.resv, arg->ticket);
> @@ -871,7 +876,11 @@ static int ttm_lru_walk_ticketlock(struct
> ttm_bo_lru_cursor *curs,
>  		 * trylocking for this walk.
>  		 */
>  		arg->ticket = NULL;
> -	} else if (ret == -EDEADLK) {
> +
> +	} else if (arg->ctx->exec && arg->ctx->allow_res_evict &&
> +		   ret == -EALREADY) {
> +		ret = 0;
> +	} else if (!arg->ctx->exec && ret == -EDEADLK) {
>  		/* Caller needs to exit the ww transaction. */
>  		ret = -ENOSPC;
>  	}
> @@ -937,12 +946,17 @@ static void ttm_bo_lru_cursor_cleanup_bo(struct
> ttm_bo_lru_cursor *curs)
>  {
>  	struct ttm_buffer_object *bo = curs->bo;
>  
> -	if (bo) {
> -		if (curs->needs_unlock)
> +	if (!bo)
> +		return;
> +
> +	if (curs->needs_unlock) {
> +		if (curs->arg->ctx->exec)
> +			drm_exec_unlock_obj(curs->arg->ctx->exec,
> &bo->base);

Unlocking after each locks sort of defeats the purpose of ww locking,
since it can't guarantee forward progress, but I guess will work as an
initial attempt.

Thanks,
Thomas


> +		else
>  			dma_resv_unlock(bo->base.resv);
> -		ttm_bo_put(bo);
> -		curs->bo = NULL;
>  	}
> +	ttm_bo_put(bo);
> +	curs->bo = NULL;
>  }
>  
>  /**
> @@ -1016,8 +1030,8 @@ __ttm_bo_lru_cursor_next(struct
> ttm_bo_lru_cursor *curs)
>  		if (ttm_lru_walk_trylock(curs, bo)) {
>  			bo_locked = true;
>  
> -		} else if (!arg->ticket || arg->ctx->no_wait_gpu ||
> -			   arg->trylock_only) {
> +		} else if ((!arg->ticket || arg->ctx->no_wait_gpu ||
> +			    arg->trylock_only) && !arg->ctx->exec) {
>  			spin_unlock(lru_lock);
>  			ttm_bo_put(bo);
>  			spin_lock(lru_lock);
> diff --git a/include/drm/ttm/ttm_bo.h b/include/drm/ttm/ttm_bo.h
> index a4060e44d23d0..156444b5e85d8 100644
> --- a/include/drm/ttm/ttm_bo.h
> +++ b/include/drm/ttm/ttm_bo.h
> @@ -187,6 +187,11 @@ struct ttm_operation_ctx {
>  	 * @bytes_moved: Statistics on how many bytes have been
> moved.
>  	 */
>  	uint64_t bytes_moved;
> +	/**
> +	 * @exec: optional drm_exec object to use for locking BOs
> and
> +	 * tracking which are locked.
> +	 */
> +	struct drm_exec *exec;
>  };
>  
>  /**

  reply	other threads:[~2026-07-07 12:48 UTC|newest]

Thread overview: 45+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-07-03 16:31 [PATCH 00/10] Use drm_exec to lock TTM buffers, respin Natalie Vock
2026-07-03 16:31 ` [PATCH 01/10] drm/exec: Add helper to bypass IGNORE_DUPLICATES flag Natalie Vock
2026-07-06  8:43   ` Christian König
2026-07-07 12:09   ` Thomas Hellström
2026-07-07 12:28     ` Christian König
2026-07-07 12:41       ` Thomas Hellström
2026-07-07 12:54         ` Christian König
2026-07-07 13:12           ` Natalie Vock
2026-07-03 16:31 ` [PATCH 02/10] drm/ttm: replace TTMs refcount with the DRM refcount v4 Natalie Vock
2026-07-06 13:14   ` Thomas Hellström
2026-07-06 14:49     ` Christian König
2026-07-06 17:01       ` Thomas Hellström
2026-07-06 17:51         ` Matthew Brost
2026-07-06 17:53           ` Thomas Hellström
2026-07-06 18:03             ` Matthew Brost
2026-07-06 18:05               ` Matthew Brost
2026-07-07  6:30                 ` Thomas Hellström
2026-07-06 18:03           ` Thomas Hellström
2026-07-06 18:23         ` Christian König
2026-07-06 22:26           ` Dave Airlie
2026-07-07  6:56           ` Thomas Hellström
2026-07-07  7:35             ` Natalie Vock
2026-07-07  8:48             ` Christian König
2026-07-07  9:53               ` Thomas Hellström
2026-07-07 12:52                 ` Christian König
2026-07-07 14:07                   ` Thomas Hellström
2026-07-03 16:31 ` [PATCH 03/10] drm/ttm: remove ttm_lru_walk_ops Natalie Vock
2026-07-06 12:34   ` Thomas Hellström
2026-07-06 13:05     ` Christian König
2026-07-06 16:31       ` Matthew Brost
2026-07-06 13:08     ` Natalie Vock
2026-07-03 16:31 ` [PATCH 04/10] drm/ttm: grab BO reference before locking it Natalie Vock
2026-07-07 12:12   ` Thomas Hellström
2026-07-03 16:31 ` [PATCH 05/10] drm/ttm: switch to ttm_bo_lru_for_each_reserved_guarded for swapout Natalie Vock
2026-07-07 12:17   ` Thomas Hellström
2026-07-03 16:31 ` [PATCH 06/10] drm/ttm: move zombie handling into ttm_bo_evict Natalie Vock
2026-07-07 12:24   ` Thomas Hellström
2026-07-03 16:31 ` [PATCH 07/10] drm/ttm: use ttm_bo_lru_for_each_reserved_guarded in evict_all Natalie Vock
2026-07-07 12:35   ` Thomas Hellström
2026-07-03 16:31 ` [PATCH 08/10] drm/xe: remove workaround for TTM internals Natalie Vock
2026-07-07 11:52   ` Thomas Hellström
2026-07-03 16:31 ` [PATCH 09/10] drm/ttm: support using drm_exec during eviction v4 Natalie Vock
2026-07-07 12:48   ` Thomas Hellström [this message]
2026-07-03 16:31 ` [PATCH 10/10] drm/amdgpu: use drm_exec during BO validation Natalie Vock
2026-07-03 17:47 ` ✗ LGCI.VerificationFailed: failure for Use drm_exec to lock TTM buffers, respin Patchwork

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=944807b30e00606dc3f4a8ad57c3e8e70ddd2d10.camel@linux.intel.com \
    --to=thomas.hellstrom@linux.intel.com \
    --cc=airlied@gmail.com \
    --cc=alexander.deucher@amd.com \
    --cc=amd-gfx@lists.freedesktop.org \
    --cc=christian.koenig@amd.com \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=intel-gfx@lists.freedesktop.org \
    --cc=intel-xe@lists.freedesktop.org \
    --cc=jani.nikula@linux.intel.com \
    --cc=joonas.lahtinen@linux.intel.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=maarten.lankhorst@linux.intel.com \
    --cc=matthew.auld@intel.com \
    --cc=matthew.brost@intel.com \
    --cc=mripard@kernel.org \
    --cc=natalie.vock@gmx.de \
    --cc=ray.huang@amd.com \
    --cc=rodrigo.vivi@intel.com \
    --cc=simona@ffwll.ch \
    --cc=tursulin@ursulin.net \
    --cc=tzimmermann@suse.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox