drm/amdgpu: Account for shadow PTs in mapping update IB size.
When amdgpu_vm_frag_ptes calls amdgpu_vm_update_ptes and the pt
has a shadow PT we mirror all the write to the shadow PT too, which
results in twice the commands.
Christian König [Mon, 4 Sep 2017 18:58:45 +0000 (20:58 +0200)]
drm/ttm: fix ttm_bo_cleanup_refs_or_queue once more
With shared reservation objects __ttm_bo_reserve() can easily fail even on
destroyed BOs. This prevents correct handling when we need to individualize
the reservation object.
Fix this by individualizing the object before even trying to reserve it.
The patch doesn't work at all:
1. The CS can still be blocked because of amdgpu_ctx_add_fence().
2. The order of submission isn't correct any more.
3. We could end up using freed up memory because we now drop the
ctx reference to early.
This needs to be fixed cleanly by doing the context handling after the BO
handling, but this is a larger task just avoid the obvious crashes for now.
Rex Zhu [Tue, 29 Aug 2017 08:08:56 +0000 (16:08 +0800)]
drm/amd/powerplay: add UMD P-state in powerplay.
This feature is for UMD to run benchmark in a
power state that is as steady as possible. kmd
need to fix the power state as stable as possible.
now, kmd support four level:
profile_standard,peak,min_sclk,min_mclk
The header comment in include/trace/define_trace.h specifies that the
TRACE_INCLUDE_PATH needs to be relative to the define_trace.h header
rather than the trace file including it. Most instances get that wrong
and work around it by adding the $(src) directory to the include path.
While this works, it is preferable to refer to the correct path to the
trace file in the first place and avoid any workaround.
The header comment in include/trace/define_trace.h specifies that the
TRACE_INCLUDE_PATH needs to be relative to the define_trace.h header
rather than the trace file including it. Most instances get that wrong
and work around it by adding the $(src) directory to the include path.
While this works, it is preferable to refer to the correct path to the
trace file in the first place and avoid any workaround.
Alex Deucher [Sat, 19 Aug 2017 03:52:37 +0000 (23:52 -0400)]
drm/amdgpu/gfx9: adjust mqd allocation size
To allocate additional space for the dynamic cu masks.
Confirmed with the hw team that we only need 1 dword
for the mask. The mask is the same for each SE so
you only need 1 dword.
Kent Russell [Tue, 22 Aug 2017 16:31:43 +0000 (12:31 -0400)]
drm/amdgpu: Add debugfs file for VBIOS and version
Add 2 debugfs files, one that contains the VBIOS version, and one that
contains the VBIOS itself. These won't change after initialization,
so we can add the VBIOS version when we parse the atombios information.
This ensures that we can find out the VBIOS version, even when the dmesg
buffer fills up, and makes it easier to associate which VBIOS version is
for which GPU on mGPU configurations. Set the size to 20 characters in
case of some weird VBIOS version that exceeds the expected 17 character
format (3-8-3\0). The VBIOS dump also allows for easy debugging
Tom St Denis [Wed, 23 Aug 2017 19:33:40 +0000 (15:33 -0400)]
drm/amd/amdgpu: Remove AMDGPU tracepoint and use new TTM tracepoint (v2)
Switches the AMDGPU driver over to the TTM tracepoint and removes
our old one. Now you can enable traces before loading the module
and trace all mappings.
Tom St Denis [Wed, 23 Aug 2017 19:17:47 +0000 (15:17 -0400)]
drm/ttm: Add DMA map/unmap tracepoint (v3)
Also exports two functions that vendor drivers can call
to trace DMA mappings. This is meant to help translate
IOMMU mappings of bus addresses back to physical pages.
Used by the umr amdgpu debugger for instance.
Signed-off-by: Tom St Denis <[email protected]> Reviewed-by: Christian König <[email protected]> Signed-off-by: Alex Deucher <[email protected]>
(v2): Use dev_name() to get PCI path instead.
(v3): Use correct types for dma/phys addresses
Evan Quan [Fri, 18 Aug 2017 08:46:47 +0000 (16:46 +0800)]
drm/amdgpu: support polaris10/11/12 new cp firmwares
Newer versions of the CP firmware require changes in how the driver
initializes the hw block.
Change the firmware name for new firmware to maintain compatibility with
older kernels.
Check memory allocation failure and return -ENOMEM in such a case.
'num_post_dep_syncobjs' still has to be set to 0 before the test in order
to have it initialized if 'amdgpu_cs_parser_fini()' is called to free
resources.
The calling graph would be, in such a case!
failure in amdgpu_cs_process_syncobj_out_dep()
---> error code returned by amdgpu_cs_dependencies()
--> amdgpu_cs_parser_fini() is called
Alex Deucher [Tue, 22 Aug 2017 20:39:30 +0000 (16:39 -0400)]
drm/amdgpu: set sched_hw_submission higher for KIQ (v3)
KIQ doesn't really use the GPU scheduler. The base
drivers generally use the KIQ ring directly rather than
submitting IBs. However, amdgpu_sched_hw_submission
(which defaults to 2) limits the number of outstanding
fences to 2. KFD uses the KIQ for TLB flushes and the
2 fence limit hurts performance when there are several KFD
processes running.
v2: move some expressions to one line
change KIQ sched_hw_submission to at least 16
v3: bump to 256
Christian König [Fri, 18 Aug 2017 13:50:17 +0000 (15:50 +0200)]
drm/amdgpu: fix and cleanup shadow handling
Set the shadow flag on the shadow and not the parent, always bind shadow BOs
during allocation instead of manually, use the reservation_object wrappers
to grab the lock.
This fixes a couple of issues with binding the shadow BOs as well as correctly
evicting them when memory becomes tight.
Alex Deucher [Mon, 21 Aug 2017 15:58:25 +0000 (11:58 -0400)]
drm/amdgpu: add automatic per asic settings for gart_size
We need a larger gart for asics that do not support GPUVM on all
engines (e.g., MM) to make sure we have enough space for all
gtt buffers in physical mode. Change the default size based on
the asic type.
Emily Deng [Mon, 21 Aug 2017 01:51:10 +0000 (09:51 +0800)]
drm/amdgpu/virtual_dce: Virtual display doesn't support disable vblank immediately
For virtual display, it uses software timer to emulate the vsync interrupt,
it doesn't have high precision, so doesn't support disable vblank immediately.
Dave Airlie [Tue, 29 Aug 2017 00:38:14 +0000 (10:38 +1000)]
Merge branch 'drm-vmwgfx-next' of git://people.freedesktop.org/~syeh/repos_linux into drm-next
vmwgfx add fence fd support.
* 'drm-vmwgfx-next' of git://people.freedesktop.org/~syeh/repos_linux:
drm/vmwgfx: Bump the version for fence FD support
drm/vmwgfx: Add export fence to file descriptor support
drm/vmwgfx: Add support for imported Fence File Descriptor
drm/vmwgfx: Prepare to support fence fd
drm/vmwgfx: Fix incorrect command header offset at restart
drm/vmwgfx: Support the NOP_ERROR command
drm/vmwgfx: Restart command buffers after errors
drm/vmwgfx: Move irq bottom half processing to threads
drm/vmwgfx: Don't use drm_irq_[un]install
Dave Airlie [Tue, 29 Aug 2017 00:37:36 +0000 (10:37 +1000)]
Merge tag 'exynos-drm-next-for-v4.14' of git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos into drm-next
Summary:
- Provide NV12MT pixel format support of Mixer driver in generic way.
- Refactor Exynos KMS drivers
. Refactoring to panel detection way
. Refactoring to setting up possible_crtcs
. Refactoring to video and command mode support
- Some cleanups
* tag 'exynos-drm-next-for-v4.14' of git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos:
drm/exynos: simplify set_pixfmt() in DECON and FIMD drivers
drm/exynos: consistent use of cpp
drm/exynos: mixer: remove src offset from mixer_graph_buffer()
drm/exynos: mixer: simplify mixer_graph_buffer()
drm/exynos: mixer: simplify vp_video_buffer()
drm/exynos: mixer: enable NV12MT support for the video plane
drm/exynos: mixer: fix chroma comment in vp_video_buffer()
arm64: dts: exynos: remove i80-if-timings nodes
dt-bindings: exynos5433-decon: remove i80-if-timings property
drm/exynos/decon5433: use mode info stored in CRTC to detect i80 mode
drm/exynos: add mode_valid callback to exynos_drm
drm/exynos/decon5433: refactor irq requesting code
drm/exynos/mic: use mode info stored in CRTC to detect i80 mode
drm/exynos/dsi: propagate info about command mode from panel
drm/exynos/dsi: refactor panel detection logic
drm/exynos: use helper to set possible crtcs
drm/exynos/decon5433: use readl_poll_timeout helpers
Dave Airlie [Tue, 29 Aug 2017 00:36:06 +0000 (10:36 +1000)]
Merge tag 'drm-misc-next-fixes-2017-08-28' of git://anongit.freedesktop.org/git/drm-misc into drm-next
UAPI Changes:
- Rename u32 to __u32 in struct drm_format_modifier_blob (Lionel)
Cc: Lionel Landwerlin <[email protected]>
* tag 'drm-misc-next-fixes-2017-08-28' of git://anongit.freedesktop.org/git/drm-misc:
drm: rename u32 in __u32 in uapi
Jason Ekstrand [Mon, 28 Aug 2017 21:10:28 +0000 (14:10 -0700)]
drm/syncobj: Add a signal ioctl (v3)
This IOCTL provides a mechanism for userspace to trigger a sync object
directly. There are other ways that userspace can trigger a syncobj
such as submitting a dummy batch somewhere or hanging on to a triggered
sync_file and doing an import. This just provides an easy way to
manually trigger the sync object without weird hacks.
The motivation for this IOCTL is Vulkan fences. Vulkan lets you create
a fence already in the signaled state so that you can wait on it
immediatly without stalling. We could also handle this with a new
create flag to ask the driver to create a syncobj that is already
signaled but the IOCTL seemed a bit cleaner and more generic.
v2:
- Take an array of sync objects (Dave Airlie)
v3:
- Throw -EINVAL if pad != 0
Jason Ekstrand [Mon, 28 Aug 2017 21:10:27 +0000 (14:10 -0700)]
drm/syncobj: Add a reset ioctl (v3)
This just resets the dma_fence to NULL so it looks like it's never been
signaled. This will be useful once we add the new wait API for allowing
wait on "submit and signal" behavior.
v2:
- Take an array of sync objects (Dave Airlie)
v3:
- Throw -EINVAL if pad != 0
Jason Ekstrand [Fri, 25 Aug 2017 17:52:26 +0000 (10:52 -0700)]
drm/syncobj: Add a syncobj_array_find helper
The wait ioctl has a bunch of code to read an syncobj handle array from
userspace and turn it into an array of syncobj pointers. We're about to
add two new IOCTLs which will need to work with arrays of syncobj
handles so let's make some helpers.
Jason Ekstrand [Fri, 25 Aug 2017 17:52:24 +0000 (10:52 -0700)]
drm/syncobj: Allow wait for submit and signal behavior (v5)
Vulkan VkFence semantics require that the application be able to perform
a CPU wait on work which may not yet have been submitted. This is
perfectly safe because the CPU wait has a timeout which will get
triggered eventually if no work is ever submitted. This behavior is
advantageous for multi-threaded workloads because, so long as all of the
threads agree on what fences to use up-front, you don't have the extra
cross-thread synchronization cost of thread A telling thread B that it
has submitted its dependent work and thread B is now free to wait.
Within a single process, this can be implemented in the userspace driver
by doing exactly the same kind of tracking the app would have to do
using posix condition variables or similar. However, in order for this
to work cross-process (as is required by VK_KHR_external_fence), we need
to handle this in the kernel.
This commit adds a WAIT_FOR_SUBMIT flag to DRM_IOCTL_SYNCOBJ_WAIT which
instructs the IOCTL to wait for the syncobj to have a non-null fence and
then wait on the fence. Combined with DRM_IOCTL_SYNCOBJ_RESET, you can
easily get the Vulkan behavior.
v2:
- Fix a bug in the invalid syncobj error path
- Unify the wait-all and wait-any cases
v3:
- Unify the timeout == 0 case a bit with the timeout > 0 case
- Use wait_event_interruptible_timeout
v4:
- Use proxy fence
v5:
- Revert to a combination of v2 and v3
- Don't use proxy fences
- Don't use wait_event_interruptible_timeout because it just adds an
extra layer of callbacks
projects / linux.git / log