All of lore.kernel.org
 help / color / mirror / Atom feed
From: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
To: intel-gfx@lists.freedesktop.org, dri-devel@lists.freedesktop.org
Cc: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>,
	matthew.auld@intel.com
Subject: [PATCH v6 0/4] drm/i915: Prepare error capture for asynchronous migration
Date: Mon,  8 Nov 2021 18:45:43 +0100	[thread overview]
Message-ID: <20211108174547.979714-1-thomas.hellstrom@linux.intel.com> (raw)

This patch series prepares error capture for asynchronous migration,
where the vma pages may not reflect the pages the GPU is currently
executing from but may be several migrations ahead.

The first patch gets rid of some contigous memory allocations that may
blow up due to the change of allocation mode introduced in patch 2/4

The third patch introduces vma state snapshots that record the vma state
at request submission time.
It also takes additional measures to make sure that
the capture list and request is not disappearing from under us while
capturing. The latter may otherwise happen if a heartbeat triggered parallel
capture is running during a manual reset which retires the request.

Finally the last patch is more of a POC patch and not strictly needed yet,
but will be (or at least something very similar) soon for async unbinding.
It will make sure that unbinding doesn't complete or signal completion
before capture is done. Async reuse of memory can't happen until unbinding
signals complete and without waiting for capture done, we might capture
contents of reused memory.
Before the last patch the vma active is instead still keeping the vma alive,
but that will not work with async unbinding anymore, and also it is still
not clear how we guarantee keeping the vma alive long enough to even
grab an active reference during capture.

v2:
- Mostly Fixes for selftests and rebinding. See patch 3. 
v3:
- Honor the unbind fence also when evicting for suspend on gen6.
- Minor cleanups on patch 3.
v4:
- Break out patch 2 from patch 1.
v5:
- Ditch a patch from the since it's already commited.
- Use __GFP_KSWAPD_RECLAIM rather than GFP_NOWAIT in patch 2.
v6:
- Reorder patches and introduce patch 1/4 which gets rid of some
  contiguous allocations that are likely to fail after introduction of
  patch 2/4.
- Use #if IS_ENABLED() rather than #ifdef similar to the rest of the
  driver

Thomas Hellström (4):
  drm/i915: Avoid allocating a page array for the gpu coredump
  drm/i915: Use __GFP_KSWAPD_RECLAIM in the capture code
  drm/i915: Update error capture code to avoid using the current vma
    state
  drm/i915: Initial introduction of vma resources

 drivers/gpu/drm/i915/Makefile                 |   1 +
 .../gpu/drm/i915/gem/i915_gem_execbuffer.c    | 137 +++++++++--
 drivers/gpu/drm/i915/gt/intel_engine_cs.c     |   8 +-
 .../drm/i915/gt/intel_execlists_submission.c  |   3 +-
 drivers/gpu/drm/i915/i915_gpu_error.c         | 228 ++++++++++++------
 drivers/gpu/drm/i915/i915_gpu_error.h         |   4 +-
 drivers/gpu/drm/i915/i915_request.c           |  63 +++--
 drivers/gpu/drm/i915/i915_request.h           |  20 +-
 drivers/gpu/drm/i915/i915_vma.c               | 206 +++++++++++++++-
 drivers/gpu/drm/i915/i915_vma.h               |  20 +-
 drivers/gpu/drm/i915/i915_vma_snapshot.c      | 131 ++++++++++
 drivers/gpu/drm/i915/i915_vma_snapshot.h      | 112 +++++++++
 drivers/gpu/drm/i915/i915_vma_types.h         |   5 +
 drivers/gpu/drm/i915/selftests/i915_gem_gtt.c |  98 +++++---
 14 files changed, 865 insertions(+), 171 deletions(-)
 create mode 100644 drivers/gpu/drm/i915/i915_vma_snapshot.c
 create mode 100644 drivers/gpu/drm/i915/i915_vma_snapshot.h

-- 
2.31.1


WARNING: multiple messages have this Message-ID (diff)
From: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
To: intel-gfx@lists.freedesktop.org, dri-devel@lists.freedesktop.org
Cc: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>,
	matthew.auld@intel.com
Subject: [Intel-gfx] [PATCH v6 0/4] drm/i915: Prepare error capture for asynchronous migration
Date: Mon,  8 Nov 2021 18:45:43 +0100	[thread overview]
Message-ID: <20211108174547.979714-1-thomas.hellstrom@linux.intel.com> (raw)

This patch series prepares error capture for asynchronous migration,
where the vma pages may not reflect the pages the GPU is currently
executing from but may be several migrations ahead.

The first patch gets rid of some contigous memory allocations that may
blow up due to the change of allocation mode introduced in patch 2/4

The third patch introduces vma state snapshots that record the vma state
at request submission time.
It also takes additional measures to make sure that
the capture list and request is not disappearing from under us while
capturing. The latter may otherwise happen if a heartbeat triggered parallel
capture is running during a manual reset which retires the request.

Finally the last patch is more of a POC patch and not strictly needed yet,
but will be (or at least something very similar) soon for async unbinding.
It will make sure that unbinding doesn't complete or signal completion
before capture is done. Async reuse of memory can't happen until unbinding
signals complete and without waiting for capture done, we might capture
contents of reused memory.
Before the last patch the vma active is instead still keeping the vma alive,
but that will not work with async unbinding anymore, and also it is still
not clear how we guarantee keeping the vma alive long enough to even
grab an active reference during capture.

v2:
- Mostly Fixes for selftests and rebinding. See patch 3. 
v3:
- Honor the unbind fence also when evicting for suspend on gen6.
- Minor cleanups on patch 3.
v4:
- Break out patch 2 from patch 1.
v5:
- Ditch a patch from the since it's already commited.
- Use __GFP_KSWAPD_RECLAIM rather than GFP_NOWAIT in patch 2.
v6:
- Reorder patches and introduce patch 1/4 which gets rid of some
  contiguous allocations that are likely to fail after introduction of
  patch 2/4.
- Use #if IS_ENABLED() rather than #ifdef similar to the rest of the
  driver

Thomas Hellström (4):
  drm/i915: Avoid allocating a page array for the gpu coredump
  drm/i915: Use __GFP_KSWAPD_RECLAIM in the capture code
  drm/i915: Update error capture code to avoid using the current vma
    state
  drm/i915: Initial introduction of vma resources

 drivers/gpu/drm/i915/Makefile                 |   1 +
 .../gpu/drm/i915/gem/i915_gem_execbuffer.c    | 137 +++++++++--
 drivers/gpu/drm/i915/gt/intel_engine_cs.c     |   8 +-
 .../drm/i915/gt/intel_execlists_submission.c  |   3 +-
 drivers/gpu/drm/i915/i915_gpu_error.c         | 228 ++++++++++++------
 drivers/gpu/drm/i915/i915_gpu_error.h         |   4 +-
 drivers/gpu/drm/i915/i915_request.c           |  63 +++--
 drivers/gpu/drm/i915/i915_request.h           |  20 +-
 drivers/gpu/drm/i915/i915_vma.c               | 206 +++++++++++++++-
 drivers/gpu/drm/i915/i915_vma.h               |  20 +-
 drivers/gpu/drm/i915/i915_vma_snapshot.c      | 131 ++++++++++
 drivers/gpu/drm/i915/i915_vma_snapshot.h      | 112 +++++++++
 drivers/gpu/drm/i915/i915_vma_types.h         |   5 +
 drivers/gpu/drm/i915/selftests/i915_gem_gtt.c |  98 +++++---
 14 files changed, 865 insertions(+), 171 deletions(-)
 create mode 100644 drivers/gpu/drm/i915/i915_vma_snapshot.c
 create mode 100644 drivers/gpu/drm/i915/i915_vma_snapshot.h

-- 
2.31.1


             reply	other threads:[~2021-11-08 18:01 UTC|newest]

Thread overview: 18+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-11-08 17:45 Thomas Hellström [this message]
2021-11-08 17:45 ` [Intel-gfx] [PATCH v6 0/4] drm/i915: Prepare error capture for asynchronous migration Thomas Hellström
2021-11-08 17:45 ` [PATCH v6 1/4] drm/i915: Avoid allocating a page array for the gpu coredump Thomas Hellström
2021-11-08 17:45   ` [Intel-gfx] " Thomas Hellström
2021-11-22 15:08   ` Ramalingam C
2021-11-22 15:08     ` [Intel-gfx] " Ramalingam C
2021-11-08 17:45 ` [PATCH v6 2/4] drm/i915: Use __GFP_KSWAPD_RECLAIM in the capture code Thomas Hellström
2021-11-08 17:45   ` [Intel-gfx] " Thomas Hellström
2021-11-22 16:50   ` Ramalingam C
2021-11-08 17:45 ` [PATCH v6 3/4] drm/i915: Update error capture code to avoid using the current vma state Thomas Hellström
2021-11-08 17:45   ` [Intel-gfx] " Thomas Hellström
2021-11-29  9:13   ` Ramalingam C
2021-11-29  9:13     ` [Intel-gfx] " Ramalingam C
2021-11-08 17:45 ` [PATCH v6 4/4] drm/i915: Initial introduction of vma resources Thomas Hellström
2021-11-08 17:45   ` [Intel-gfx] " Thomas Hellström
2021-11-08 18:29 ` [Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for drm/i915: Prepare error capture for asynchronous migration (rev2) Patchwork
2021-11-08 19:02 ` [Intel-gfx] ✓ Fi.CI.BAT: success " Patchwork
2021-11-08 20:21 ` [Intel-gfx] ✓ Fi.CI.IGT: " 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=20211108174547.979714-1-thomas.hellstrom@linux.intel.com \
    --to=thomas.hellstrom@linux.intel.com \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=intel-gfx@lists.freedesktop.org \
    --cc=matthew.auld@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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.