From: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
To: Matthew Auld <matthew.auld@intel.com>,
Matthew Brost <matthew.brost@intel.com>,
Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>,
Stuart Summers <stuart.summers@intel.com>
Cc: intel-xe@lists.freedesktop.org
Subject: Re: [PATCH 2/5] drm/xe: Requeue the rebind worker instead of retrying in-line on -EAGAIN
Date: Thu, 19 Nov 2026 13:51:33 +0100 [thread overview]
Message-ID: <d2c8112c02a1324473e3808aec2df793f60b9397.camel@linux.intel.com> (raw)
In-Reply-To: <20261119-userptr-repin-v1-2-af2e327833a9@intel.com>
On Thu, 2026-11-19 at 11:14 +0000, Matthew Auld wrote:
> From: Thomas Hellström <thomas.hellstrom@linux.intel.com>
>
> When the preempt rebind worker needs to retry, it currently jumps
> back
> and keeps holding vm->lock in write mode across the retry. Only VFs
> with migration support drop the lock and requeue the worker.
>
> Holding the vm lock across retries needlessly blocks other users of
> the
> lock and other work on the ordered workqueue, and makes the retry
> unfit
> for conditions that may last for a while, like repeatedly invalidated
> userptrs.
>
> Always drop the vm lock and requeue the worker on -EAGAIN. This also
> fixes a leak of the preempt fences allocated in the failed attempt on
> the existing VF path.
Do we need a Fixes: here?
Also needs a separate reviewer.
>
> The userptr invalidation injection state that forces one retry per
> rebind sequence was a local variable that would be reset by a
> requeue.
> Move it to the vm.
>
> Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Signed-off-by: Matthew Auld <matthew.auld@intel.com>
> ---
> drivers/gpu/drm/xe/xe_vm.c | 59 ++++++++++++++++++++++++++----
> ----------
> drivers/gpu/drm/xe/xe_vm_types.h | 8 ++++++
> 2 files changed, 46 insertions(+), 21 deletions(-)
>
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> index 75be5e5329bd..ff8d355728fc 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -542,6 +542,31 @@ void xe_vm_resume_rebind_worker(struct xe_vm
> *vm)
> queue_work(vm->xe->ordered_wq, &vm->preempt.rebind_work);
> }
>
> +#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
> +/*
> + * Force one retry per rebind sequence. The sequence spans requeues
> of the
> + * worker, so the state must live in the vm.
> + */
> +static bool xe_vm_rebind_worker_inject_retry(struct xe_vm *vm)
> +{
> + return !vm->preempt.inval_inject_tries++;
> +}
> +
> +static void xe_vm_rebind_worker_reset_inject(struct xe_vm *vm)
> +{
> + vm->preempt.inval_inject_tries = 0;
> +}
> +#else
> +static bool xe_vm_rebind_worker_inject_retry(struct xe_vm *vm)
> +{
> + return false;
> +}
> +
> +static void xe_vm_rebind_worker_reset_inject(struct xe_vm *vm)
> +{
> +}
> +#endif
> +
> static void preempt_rebind_work_func(struct work_struct *w)
> {
> struct xe_vm *vm = container_of(w, struct xe_vm,
> preempt.rebind_work);
> @@ -551,7 +576,6 @@ static void preempt_rebind_work_func(struct
> work_struct *w)
> LIST_HEAD(preempt_fences);
> int err = 0;
> long wait;
> - int __maybe_unused tries = 0;
>
> xe_assert(vm->xe, xe_vm_in_preempt_fence_mode(vm));
> trace_xe_vm_rebind_worker_enter(vm);
> @@ -564,7 +588,6 @@ static void preempt_rebind_work_func(struct
> work_struct *w)
> return;
> }
>
> -retry:
> if (!try_wait_for_completion(&vm->xe->pm_block) &&
> vm_suspend_rebind_worker(vm)) {
> up_write(&vm->lock);
> /* We don't actually block but don't make progress.
> */
> @@ -614,20 +637,14 @@ static void preempt_rebind_work_func(struct
> work_struct *w)
> goto out_unlock;
> }
>
> -#define retry_required(__tries, __vm) \
> - (IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT) ? \
> - (!(__tries)++ || __xe_vm_userptr_needs_repin(__vm)) : \
> - __xe_vm_userptr_needs_repin(__vm))
> -
> xe_svm_notifier_lock(vm);
> - if (retry_required(tries, vm)) {
> + if (xe_vm_rebind_worker_inject_retry(vm) ||
> + __xe_vm_userptr_needs_repin(vm)) {
> xe_svm_notifier_unlock(vm);
> err = -EAGAIN;
> goto out_unlock;
> }
>
> -#undef retry_required
> -
> xe_vm_bump_lru(vm, &exec);
>
> /* Point of no return. */
> @@ -642,20 +659,20 @@ static void preempt_rebind_work_func(struct
> work_struct *w)
> trace_xe_vm_rebind_worker_retry(vm);
>
> /*
> - * We can't block in workers on a VF which supports
> migration
> - * given this can block the VF post-migration
> workers from
> - * getting scheduled.
> + * Drop all locks and requeue rather than retrying
> in-line.
> + * This avoids blocking other work items, such as VF
> + * post-migration workers, and lets other users of
> the vm
> + * lock make progress between attempts.
> */
> - if (IS_SRIOV_VF(vm->xe) &&
> - xe_sriov_vf_migration_supported(vm->xe)) {
> - up_write(&vm->lock);
> - xe_vm_queue_rebind_worker(vm);
> - return;
> - }
> -
> - goto retry;
> + up_write(&vm->lock);
> + free_preempt_fences(&preempt_fences);
> + xe_vm_queue_rebind_worker(vm);
> + trace_xe_vm_rebind_worker_exit(vm);
> + return;
> }
>
> + xe_vm_rebind_worker_reset_inject(vm);
> +
> if (err) {
> drm_warn(&vm->xe->drm, "VM worker error: %d\n",
> err);
> xe_vm_kill(vm, true);
> diff --git a/drivers/gpu/drm/xe/xe_vm_types.h
> b/drivers/gpu/drm/xe/xe_vm_types.h
> index fb9305124679..756cf2dbee8a 100644
> --- a/drivers/gpu/drm/xe/xe_vm_types.h
> +++ b/drivers/gpu/drm/xe/xe_vm_types.h
> @@ -350,6 +350,14 @@ struct xe_vm {
> * kicked on resume.
> */
> struct list_head pm_activate_link;
> +#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
> + /**
> + * @inval_inject_tries: Number of rebind attempts in
> the
> + * current rebind sequence. Used to force a retry
> for testing.
> + * Protected by @lock.
> + */
> + unsigned int inval_inject_tries;
> +#endif
> } preempt;
>
> /** @exec_queues: Manages list of exec queues attached to
> this VM, protected by lock. */
next prev parent reply other threads:[~2026-11-19 12:51 UTC|newest]
Thread overview: 20+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-19 11:14 [PATCH 0/5] Non-fatal repin retries and NUMA-local rebind worker Matthew Auld
2026-11-19 11:14 ` [PATCH 1/5] drm/xe/vm: restore userptr invalidations on VMA op unwind Matthew Auld
2026-11-19 12:48 ` Thomas Hellström
2026-11-19 16:35 ` Himal prasad Ghimiray
2026-11-19 11:14 ` [PATCH 2/5] drm/xe: Requeue the rebind worker instead of retrying in-line on -EAGAIN Matthew Auld
2026-11-19 12:51 ` Thomas Hellström [this message]
2026-11-19 16:29 ` Matthew Auld
2026-11-19 17:26 ` Thomas Hellström
2026-11-19 17:38 ` Matthew Auld
2026-11-19 16:36 ` Himal prasad Ghimiray
2026-11-19 11:14 ` [PATCH 3/5] drm/xe: Add a userptr repin timeout sysfs attribute Matthew Auld
2026-11-19 16:38 ` Himal prasad Ghimiray
2026-11-19 11:14 ` [PATCH 4/5] drm/xe/userptr: Make the repin retry loop non-fatal Matthew Auld
2026-11-19 13:03 ` Thomas Hellström
2026-11-19 17:10 ` Himal prasad Ghimiray
2026-11-19 11:14 ` [PATCH 5/5] drm/xe: run compute rebind worker on device NUMA node Matthew Auld
2026-11-19 13:11 ` Thomas Hellström
2026-11-19 13:38 ` Matthew Auld
2026-11-19 13:16 ` ✓ CI.KUnit: success for Non-fatal repin retries and NUMA-local rebind worker Patchwork
2026-11-19 13:54 ` ✓ Xe.CI.BAT: " 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=d2c8112c02a1324473e3808aec2df793f60b9397.camel@linux.intel.com \
--to=thomas.hellstrom@linux.intel.com \
--cc=himal.prasad.ghimiray@intel.com \
--cc=intel-xe@lists.freedesktop.org \
--cc=matthew.auld@intel.com \
--cc=matthew.brost@intel.com \
--cc=stuart.summers@intel.com \
/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