Instead of distingting between shared and exclusive fences specify
the fence usage while adding fences.
Rework all drivers to use this interface instead and deprecate the old one.
v2: some kerneldoc comments suggested by Daniel
v3: fix a missing case in radeon
v4: rebase on nouveau changes, fix lockdep and temporary disable warning
v5: more documentation updates
v6: separate internal dma_resv changes from this patch, avoids to
disable warning temporary, rebase on upstream changes
v7: fix missed case in lima driver, minimize changes to i915_gem_busy_ioctl
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20220407085946.744568-3-christian.koenig@amd.com
Audit all the users of dma_resv_add_excl_fence() and make sure they
reserve a shared slot also when only trying to add an exclusive fence.
This is the next step towards handling the exclusive fence like a
shared one.
v2: fix missed case in amdgpu
v3: and two more radeon, rename function
v4: add one more case to TTM, fix i915 after rebase
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20220406075132.3263-2-christian.koenig@amd.com
Instead of having a global lock for potentially less contention.
Signed-off-by: Christian König <christian.koenig@amd.com>
Tested-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/424010/
Call qxl_bo_unpin (which does a reservation) without holding the
release_mutex lock. Fixes lockdep (correctly) warning on a possible
deadlock.
Fixes: 65ffea3c6e ("drm/qxl: unpin release objects")
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Acked-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: http://patchwork.freedesktop.org/patch/msgid/20210217123213.2199186-5-kraxel@redhat.com
Allow to set priorities for buffer objects. Use priority 1 for surface
and cursor command releases. Use priority 0 for drawing command
releases. That way the short-living drawing commands are first in line
when it comes to eviction, making it *much* less likely that
ttm_bo_mem_force_space() picks something which can't be evicted and
throws an error after waiting a while without success.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Acked-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: http://patchwork.freedesktop.org/patch/msgid/20210217123213.2199186-4-kraxel@redhat.com
Move qxl_io_notify_oom() call into wait condition.
That way the driver will call it again if one call
wasn't enough.
Also allows to remove the extra dma_fence_is_signaled()
check and the goto.
Fixes: 5a838e5d58 ("drm/qxl: simplify qxl_fence_wait")
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Acked-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: http://patchwork.freedesktop.org/patch/msgid/20210217123213.2199186-3-kraxel@redhat.com
Reorganize qxl_device_fini() a bit.
Add missing unpin() calls.
Count releases. Add wait queue for releases. That way
qxl_device_fini() can easily wait until everything is
ready for proper shutdown.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Acked-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: http://patchwork.freedesktop.org/patch/msgid/20210204145712.1531203-9-kraxel@redhat.com
Rename ttm_bo_device to ttm_device.
Rename ttm_bo_driver to ttm_device_funcs.
Rename ttm_bo_global to ttm_global.
Move global and device related functions to ttm_device.[ch].
No functional change.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/415222/
We only completely delete the BO from the LRU on destruction.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Link: https://patchwork.freedesktop.org/patch/404618/
As the name says global memory and bo accounting is global. So it doesn't
make to much sense having pointers to global structures all around the code.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Thomas Hellström <thellstrom@vmware.com>
Link: https://patchwork.freedesktop.org/patch/332879/
This allows blocking for BOs to become available
in the memory management.
Amdgpu is doing this for quite a while now during CS. Now
apply the new behavior to all drivers using TTM.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Thomas Hellstrom <thellstrom@vmware.com>
Link: https://patchwork.freedesktop.org/patch/332878/
Be more consistent with the naming of the other DMA-buf objects.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/323401/
We are already doing this for DMA-buf imports and also for
amdgpu VM BOs for quite a while now.
If this doesn't run into any problems we are probably going
to stop removing BOs from the LRU altogether.
v2: drop BUG_ON from ttm_bo_add_to_lru
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Tested-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[airlied: make etnaviv build again]
amdgpu:
- DC trace support
- More DC documentation
- XGMI hive reset support
- Rework IH interaction with KFD
- Misc fixes and cleanups
- Powerplay updates for newer polaris variants
- Add cursor plane update fast path
- Enable gpu reset by default on CI parts
- Fix config with KFD/HSA not enabled
amdkfd:
- Limit vram overcommit
- dmabuf support
- Support for doorbell BOs
ttm:
- Support for simultaneous submissions to multiple engines
scheduler:
- Add helpers for hw with preemption support
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181207233119.16861-1-alexander.deucher@amd.com
Let's support simultaneous submissions to multiple engines.
v2: rename the field to num_shared and fix up all users
v3: rebased
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fixes gcc '-Wunused-but-set-variable' warning:
drivers/gpu/drm/qxl/qxl_release.c: In function 'qxl_release_fence_buffer_objects':
drivers/gpu/drm/qxl/qxl_release.c:431:17: warning:
variable 'qbo' set but not used [-Wunused-but-set-variable]
drivers/gpu/drm/qxl/qxl_release.c:430:24: warning:
variable 'driver' set but not used [-Wunused-but-set-variable]
'qbo' not used since commit f2c24b83ae ("drm/ttm: flip the switch, and convert
to dma_fence")
And 'driver' never used since introduction in
8002db6336 ("qxl: convert qxl driver to proper use for reservations")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1542029556-88107-1-git-send-email-yuehaibing@huawei.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Let's support simultaneous submissions to multiple engines.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Link: https://patchwork.kernel.org/patch/10626149/
qxl expects that list_first_entry(release->bos) returns the first
element qxl added to the list. ttm_eu_reserve_buffers() may reorder
the list though.
Add a release_bo field to struct qxl_release and use that instead.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20180418054257.15388-3-kraxel@redhat.com
Signed-off-by: Sean Paul <seanpaul@chromium.org>
s/PAGE_SIZE/PAGE_MASK/
Luckily release_offset is never larger than PAGE_SIZE, so the bug has no
bad side effects and managed to stay unnoticed for years that way ...
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20180418054257.15388-2-kraxel@redhat.com
Signed-off-by: Sean Paul <seanpaul@chromium.org>
The pointer is available as bo->bdev->glob as well.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Give moving a BO into place an operation context to work with.
v2: rebased
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Instead of having a confusing wrapper or call the underlying ww_mutex
function directly.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-and-Tested-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In function qxl_release_alloc(), when kmalloc() returns a NULL pointer,
it returns value 0 and parameter *ret is uninitialized. 0 means no error
to the callers of qxl_release_alloc(). The callers keep going and will
try to reference the uninitialized variable. This patch fixes the bug,
returning "-ENOMEM" when kmalloc() fails.
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=188911
Signed-off-by: Pan Bian <bianpan2016@163.com>
[seanpaul fixed up subject prefix]
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: http://patchwork.freedesktop.org/patch/msgid/1480777902-7648-1-git-send-email-bianpan2016@163.com
Fence contexts are created on the fly (for example) by the GPU scheduler used
in the amdgpu driver as a result of an userspace request. Because of this
userspace could in theory force a wrap around of the 32bit context number
if it doesn't behave well.
Avoid this by increasing the context number to 64bits. This way even when
userspace manages to allocate a billion contexts per second it takes more
than 500 years for the context number to wrap around.
v2: fix printf formats as well.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
Acked-by: Sumit Semwal <sumit.semwal@linaro.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1464786612-5010-2-git-send-email-deathsimple@vodafone.de
qxl_bo_unref calls drm_gem_object_unreference_unlocked which
locks dev->struct_mutex. However this lock could be already
locked if the call came from qxl_gem_object_free.
As we don't need to call qxl_bo_ref/qxl_bo_unref cause
qxl_release_list_add will hold a reference by itself avoid
to call them and the possible deadlock.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Enable format string checks for qxl_io_log and remove resulting warnings
which could lead to memory errors on different platform or just printing
wrong information.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Free resources correctly if function fails
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This function return handle to allocated release object which is an int.
Signed-off-by: Frediano Ziglio <fziglio@redhat.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This patch adds an optional list_head parameter to ttm_eu_reserve_buffers.
If specified duplicates in the execbuf list are no longer reported as errors,
but moved to this list instead.
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch adds a new flag to the ttm_validate_buffer list to
add the fence as shared to the reservation object.
Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This crash was already here before the conversion, but qxl never leaked
hard enough to hit this.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
The locking of release_lock was stupid; t should have been be called with
fence_lock_irq if it was legitimately used. Unfortunately it never protected
anything except the fence implementation correctly.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Final driver! \o/
This is not a proper dma_fence because the hardware may never signal
anything, so don't use dma-buf with qxl, ever.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
This reorders the list to keep track of what buffers are reserved,
so previous members are always unreserved.
This gets rid of some bookkeeping that's no longer needed,
while simplifying the code some.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
Alex Constantino