Skip to content

K
Kernel

openeuler / Kernel
1 年多 前同步成功

通知 8
Star 0
Fork 0
K
Kernel
切换分支/标签
  1. 01 11月, 2019 2 次提交
  3. 27 10月, 2019 2 次提交
  5. 24 10月, 2019 1 次提交
  7. 22 10月, 2019 2 次提交
  9. 19 10月, 2019 1 次提交
  11. 18 10月, 2019 1 次提交
  13. 17 10月, 2019 1 次提交
  15. 06 10月, 2019 2 次提交
  17. 04 10月, 2019 2 次提交
    • C
      drm/i915: Move context management under GEM · a4e7ccda
      由 Chris Wilson 提交于 
      Keep track of the GEM contexts underneath i915->gem.contexts and assign
them their own lock for the purposes of list management.

v2: Focus on lock tracking; ctx->vm is protected by ctx->mutex
v3: Correct split with removal of logical HW ID
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-15-chris@chris-wilson.co.uk
      a4e7ccda
    • C
      drm/i915: Pull i915_vma_pin under the vm->mutex · 2850748e
      由 Chris Wilson 提交于 
      Replace the struct_mutex requirement for pinning the i915_vma with the
local vm->mutex instead. Note that the vm->mutex is tainted by the
shrinker (we require unbinding from inside fs-reclaim) and so we cannot
allocate while holding that mutex. Instead we have to preallocate
workers to do allocate and apply the PTE updates after we have we
reserved their slot in the drm_mm (using fences to order the PTE writes
with the GPU work and with later unbind).

In adding the asynchronous vma binding, one subtle requirement is to
avoid coupling the binding fence into the backing object->resv. That is
the asynchronous binding only applies to the vma timeline itself and not
to the pages as that is a more global timeline (the binding of one vma
does not need to be ordered with another vma, nor does the implicit GEM
fencing depend on a vma, only on writes to the backing store). Keeping
the vma binding distinct from the backing store timelines is verified by
a number of async gem_exec_fence and gem_exec_schedule tests. The way we
do this is quite simple, we keep the fence for the vma binding separate
and only wait on it as required, and never add it to the obj->resv
itself.

Another consequence in reducing the locking around the vma is the
destruction of the vma is no longer globally serialised by struct_mutex.
A natural solution would be to add a kref to i915_vma, but that requires
decoupling the reference cycles, possibly by introducing a new
i915_mm_pages object that is own by both obj->mm and vma->pages.
However, we have not taken that route due to the overshadowing lmem/ttm
discussions, and instead play a series of complicated games with
trylocks to (hopefully) ensure that only one destruction path is called!

v2: Add some commentary, and some helpers to reduce patch churn.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk
      2850748e
  19. 02 10月, 2019 1 次提交
  21. 28 9月, 2019 1 次提交
  23. 23 9月, 2019 4 次提交
  25. 16 9月, 2019 1 次提交
  27. 12 9月, 2019 1 次提交
  29. 11 9月, 2019 1 次提交
  31. 10 9月, 2019 1 次提交
  33. 07 9月, 2019 1 次提交
  35. 27 8月, 2019 1 次提交
    • L
      drm/i915: Call dma_set_max_seg_size() in i915_driver_hw_probe() · 32f0a982
      由 Lyude Paul 提交于 
      Currently, we don't call dma_set_max_seg_size() for i915 because we
intentionally do not limit the segment length that the device supports.
However, this results in a warning being emitted if we try to map
anything larger than SZ_64K on a kernel with CONFIG_DMA_API_DEBUG_SG
enabled:

[    7.751926] DMA-API: i915 0000:00:02.0: mapping sg segment longer
than device claims to support [len=98304] [max=65536]
[    7.751934] WARNING: CPU: 5 PID: 474 at kernel/dma/debug.c:1220
debug_dma_map_sg+0x20f/0x340

This was originally brought up on
https://bugs.freedesktop.org/show_bug.cgi?id=108517 , and the consensus
there was it wasn't really useful to set a limit (and that dma-debug
isn't really all that useful for i915 in the first place). Unfortunately
though, CONFIG_DMA_API_DEBUG_SG is enabled in the debug configs for
various distro kernels. Since a WARN_ON() will disable automatic problem
reporting (and cause any CI with said option enabled to start
complaining), we really should just fix the problem.

Note that as me and Chris Wilson discussed, the other solution for this
would be to make DMA-API not make such assumptions when a driver hasn't
explicitly set a maximum segment size. But, taking a look at the commit
which originally introduced this behavior, commit 78c47830
("dma-debug: check scatterlist segments"), there is an explicit mention
of this assumption and how it applies to devices with no segment size:

	Conversely, devices which are less limited than the rather
	conservative defaults, or indeed have no limitations at all
	(e.g. GPUs with their own internal MMU), should be encouraged to
	set appropriate dma_parms, as they may get more efficient DMA
	mapping performance out of it.

So unless there's any concerns (I'm open to discussion!), let's just
follow suite and call dma_set_max_seg_size() with UINT_MAX as our limit
to silence any warnings.

Changes since v3:
* Drop patch for enabling CONFIG_DMA_API_DEBUG_SG in CI. It looks like
  just turning it on causes the kernel to spit out bogus WARN_ONs()
  during some igt tests which would otherwise require teaching igt to
  disable the various DMA-API debugging options causing this. This is
  too much work to be worth it, since DMA-API debugging is useless for
  us. So, we'll just settle with this single patch to squelch WARN_ONs()
  during driver load for users that have CONFIG_DMA_API_DEBUG_SG turned
  on for some reason.
* Move dma_set_max_seg_size() call into i915_driver_hw_probe() - Chris
  Wilson
Signed-off-by: NLyude Paul <lyude@redhat.com>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: <stable@vger.kernel.org> # v4.18+
Link: https://patchwork.freedesktop.org/patch/msgid/20190823205251.14298-1-lyude@redhat.com
(cherry picked from commit acd674af)
Signed-off-by: NJani Nikula <jani.nikula@intel.com>
      32f0a982
  37. 26 8月, 2019 1 次提交
    • L
      drm/i915: Call dma_set_max_seg_size() in i915_driver_hw_probe() · acd674af
      由 Lyude Paul 提交于 
      Currently, we don't call dma_set_max_seg_size() for i915 because we
intentionally do not limit the segment length that the device supports.
However, this results in a warning being emitted if we try to map
anything larger than SZ_64K on a kernel with CONFIG_DMA_API_DEBUG_SG
enabled:

[    7.751926] DMA-API: i915 0000:00:02.0: mapping sg segment longer
than device claims to support [len=98304] [max=65536]
[    7.751934] WARNING: CPU: 5 PID: 474 at kernel/dma/debug.c:1220
debug_dma_map_sg+0x20f/0x340

This was originally brought up on
https://bugs.freedesktop.org/show_bug.cgi?id=108517 , and the consensus
there was it wasn't really useful to set a limit (and that dma-debug
isn't really all that useful for i915 in the first place). Unfortunately
though, CONFIG_DMA_API_DEBUG_SG is enabled in the debug configs for
various distro kernels. Since a WARN_ON() will disable automatic problem
reporting (and cause any CI with said option enabled to start
complaining), we really should just fix the problem.

Note that as me and Chris Wilson discussed, the other solution for this
would be to make DMA-API not make such assumptions when a driver hasn't
explicitly set a maximum segment size. But, taking a look at the commit
which originally introduced this behavior, commit 78c47830
("dma-debug: check scatterlist segments"), there is an explicit mention
of this assumption and how it applies to devices with no segment size:

	Conversely, devices which are less limited than the rather
	conservative defaults, or indeed have no limitations at all
	(e.g. GPUs with their own internal MMU), should be encouraged to
	set appropriate dma_parms, as they may get more efficient DMA
	mapping performance out of it.

So unless there's any concerns (I'm open to discussion!), let's just
follow suite and call dma_set_max_seg_size() with UINT_MAX as our limit
to silence any warnings.

Changes since v3:
* Drop patch for enabling CONFIG_DMA_API_DEBUG_SG in CI. It looks like
  just turning it on causes the kernel to spit out bogus WARN_ONs()
  during some igt tests which would otherwise require teaching igt to
  disable the various DMA-API debugging options causing this. This is
  too much work to be worth it, since DMA-API debugging is useless for
  us. So, we'll just settle with this single patch to squelch WARN_ONs()
  during driver load for users that have CONFIG_DMA_API_DEBUG_SG turned
  on for some reason.
* Move dma_set_max_seg_size() call into i915_driver_hw_probe() - Chris
  Wilson
Signed-off-by: NLyude Paul <lyude@redhat.com>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: <stable@vger.kernel.org> # v4.18+
Link: https://patchwork.freedesktop.org/patch/msgid/20190823205251.14298-1-lyude@redhat.com
      acd674af
  39. 24 8月, 2019 1 次提交
    • C
      drm/i915: Keep drm_i915_file_private around under RCU · 77715906
      由 Chris Wilson 提交于 
      Ensure that the drm_i915_file_private continues to exist as we attempt
to remove a request from its list, which may race with the destruction
of the file.

<6> [38.380714] [IGT] gem_ctx_create: starting subtest basic-files
<0> [42.201329] BUG: spinlock bad magic on CPU#0, kworker/u16:0/7
<4> [42.201356] general protection fault: 0000 [#1] PREEMPT SMP PTI
<4> [42.201371] CPU: 0 PID: 7 Comm: kworker/u16:0 Tainted: G     U            5.3.0-rc5-CI-Patchwork_14169+ #1
<4> [42.201391] Hardware name: Dell Inc.                 OptiPlex 745                 /0GW726, BIOS 2.3.1  05/21/2007
<4> [42.201594] Workqueue: i915 retire_work_handler [i915]
<4> [42.201614] RIP: 0010:spin_dump+0x5a/0x90
<4> [42.201625] Code: 00 48 8d 88 c0 06 00 00 48 c7 c7 00 71 09 82 e8 35 ef 00 00 48 85 db 44 8b 4d 08 41 b8 ff ff ff ff 48 c7 c1 0b cd 0f 82 74 0e <44> 8b 83 e0 04 00 00 48 8d 8b c0 06 00 00 8b 55 04 48 89 ee 48 c7
<4> [42.201660] RSP: 0018:ffffc9000004bd80 EFLAGS: 00010202
<4> [42.201673] RAX: 0000000000000031 RBX: 6b6b6b6b6b6b6b6b RCX: ffffffff820fcd0b
<4> [42.201688] RDX: 0000000000000000 RSI: ffff88803de266f8 RDI: 00000000ffffffff
<4> [42.201703] RBP: ffff888038381ff8 R08: 00000000ffffffff R09: 000000006b6b6b6b
<4> [42.201718] R10: 0000000041cb0b89 R11: 646162206b636f6c R12: ffff88802a618500
<4> [42.201733] R13: ffff88802b32c288 R14: ffff888038381ff8 R15: ffff88802b32c250
<4> [42.201748] FS:  0000000000000000(0000) GS:ffff88803de00000(0000) knlGS:0000000000000000
<4> [42.201765] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
<4> [42.201778] CR2: 00007f2cefc6d180 CR3: 00000000381ee000 CR4: 00000000000006f0
<4> [42.201793] Call Trace:
<4> [42.201805]  do_raw_spin_lock+0x66/0xb0
<4> [42.201898]  i915_request_retire+0x548/0x7c0 [i915]
<4> [42.201989]  retire_requests+0x4d/0x60 [i915]
<4> [42.202078]  i915_retire_requests+0x144/0x2e0 [i915]
<4> [42.202169]  retire_work_handler+0x10/0x40 [i915]

Recently, in commit 44c22f3f ("drm/i915: Serialize insertion into the
file->mm.request_list"), we fixed a race on insertion. Now, it appears
we also have a race with destruction!
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: NMatthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190823181455.31910-1-chris@chris-wilson.co.uk
      77715906
  41. 23 8月, 2019 1 次提交
  43. 20 8月, 2019 1 次提交
  45. 15 8月, 2019 1 次提交
  47. 10 8月, 2019 1 次提交
  49. 09 8月, 2019 5 次提交
  51. 08 8月, 2019 1 次提交
  53. 07 8月, 2019 2 次提交