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 1/5] drm/xe/vm: restore userptr invalidations on VMA op unwind
Date: Thu, 19 Nov 2026 13:48:42 +0100 [thread overview]
Message-ID: <905fc2549d7a7354d9486c8dafe092a2166ca4bd.camel@linux.intel.com> (raw)
In-Reply-To: <20261119-userptr-repin-v1-1-af2e327833a9@intel.com>
On Thu, 2026-11-19 at 11:14 +0000, Matthew Auld wrote:
> When xe_vma_op_commit() commits DRM_GPUVA_OP_UNMAP or
> DRM_GPUVA_OP_REMAP, prep_vma_destroy() sets XE_VMA_DESTROYED while
> the
> old VMA's GPU page tables remain live. If a subsequent step fails and
> unwinds (for example splitting a userptr VMA where new_vma() fails,
> or a
> later op in an array of binds failing), a concurrent MMU notifier
> invalidation is lost on non-fault-mode VMs:
>
> CPU0 (VM_BIND / MADVISE) CPU1 (MMU notifier)
> ------------------------ -------------------
> down_write(&vm->lock)
> xe_vma_op_commit()
> prep_vma_destroy(vma)
> flags |= XE_VMA_DESTROYED
> ...
> /* later step fails */
>
> xe_vma_userptr_invalidate_start()
> down_write(¬ifier_lock)
> mmu_interval_set_seq()
> /* XE_VMA_DESTROYED set:
> skips
> * vm->userptr.invalidated
> */
> drm_gpusvm_unmap_pages()
> up_write(¬ifier_lock)
> xe_vma_op_unwind()
> flags &= ~XE_VMA_DESTROYED
> xe_vm_insert_vma(vm, vma)
> up_write(&vm->lock)
>
> Once vm->lock is dropped, the VMA remains in the VM with live GPU
> PTEs
> pointing at unmapped pages, while neither exec nor the rebind worker
> knows to repin and rebind it.
>
> Add unprep_vma_destroy() to check xe_vma_userptr_check_repin() while
> clearing XE_VMA_DESTROYED under the notifier lock and queue
> invalidated
> non-fault-mode userptr VMAs onto vm->userptr.invalidated.
>
> Fixes: b06d47be7c83 ("drm/xe: Port Xe to GPUVA")
> Signed-off-by: Matthew Auld <matthew.auld@intel.com>
> Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> Cc: Matthew Brost <matthew.brost@intel.com>
> Cc: <stable@vger.kernel.org> # v6.12+
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> ---
> drivers/gpu/drm/xe/xe_vm.c | 35 ++++++++++++++++++++++++-----------
> 1 file changed, 24 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
> index dd855876f571..75be5e5329bd 100644
> --- a/drivers/gpu/drm/xe/xe_vm.c
> +++ b/drivers/gpu/drm/xe/xe_vm.c
> @@ -44,6 +44,7 @@
> #include "xe_tile.h"
> #include "xe_tlb_inval.h"
> #include "xe_trace_bo.h"
> +#include "xe_userptr.h"
> #include "xe_vm_madvise.h"
> #include "xe_wa.h"
>
> @@ -2401,6 +2402,26 @@ static void prep_vma_destroy(struct xe_vm *vm,
> struct xe_vma *vma,
> xe_vm_remove_vma(vm, vma);
> }
>
> +static void unprep_vma_destroy(struct xe_vm *vm, struct xe_vma *vma,
> + bool post_commit)
> +{
> + xe_svm_notifier_lock(vm);
> + vma->gpuva.flags &= ~XE_VMA_DESTROYED;
> + if (IS_ENABLED(CONFIG_DRM_GPUSVM) &&
> !xe_vm_in_fault_mode(vm) &&
> + xe_vma_is_userptr(vma) &&
> + xe_vma_userptr_check_repin(to_userptr_vma(vma))) {
> + struct xe_userptr_vma *uvma = to_userptr_vma(vma);
> +
> + spin_lock(&vm->userptr.invalidated_lock);
> + list_move_tail(&uvma->userptr.invalidate_link,
> + &vm->userptr.invalidated);
> + spin_unlock(&vm->userptr.invalidated_lock);
> + }
> + xe_svm_notifier_unlock(vm);
> + if (post_commit)
> + xe_vm_insert_vma(vm, vma);
> +}
> +
> #undef ULL
> #define ULL unsigned long long
>
> @@ -3093,13 +3114,8 @@ static void xe_vma_op_unwind(struct xe_vm *vm,
> struct xe_vma_op *op,
>
> xe_vm_assert_write_mode_or_garbage_collector(vm);
>
> - if (vma) {
> - xe_svm_notifier_lock(vm);
> - vma->gpuva.flags &= ~XE_VMA_DESTROYED;
> - xe_svm_notifier_unlock(vm);
> - if (post_commit)
> - xe_vm_insert_vma(vm, vma);
> - }
> + if (vma)
> + unprep_vma_destroy(vm, vma, post_commit);
> break;
> }
> case DRM_GPUVA_OP_REMAP:
> @@ -3117,9 +3133,6 @@ static void xe_vma_op_unwind(struct xe_vm *vm,
> struct xe_vma_op *op,
> xe_vma_destroy_unlocked(op->remap.next);
> }
> if (vma) {
> - xe_svm_notifier_lock(vm);
> - vma->gpuva.flags &= ~XE_VMA_DESTROYED;
> - xe_svm_notifier_unlock(vm);
> if (post_commit) {
> /*
> * Restore the old va range, in case
> of the
> @@ -3131,8 +3144,8 @@ static void xe_vma_op_unwind(struct xe_vm *vm,
> struct xe_vma_op *op,
> op->remap.old_start;
> op->base.remap.unmap->va->va.range =
> op->remap.old_range;
> - xe_vm_insert_vma(vm, vma);
> }
> + unprep_vma_destroy(vm, vma, post_commit);
> }
> break;
> }
next prev parent reply other threads:[~2026-11-19 12:48 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 [this message]
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
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=905fc2549d7a7354d9486c8dafe092a2166ca4bd.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