Unify BO evicting functionality for possible memory
types in amdgpu_ttm.c.
Signed-off-by: Nirmoy Das <[email protected]>
Reviewed-by: Christian König <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
This reverts commit 728e7e0cd6.
Further discussion reveals that this feature is severely broken
and needs to be reverted ASAP.
GPU reset can never be delayed by userspace even for debugging or
otherwise we can run into in kernel deadlocks.
Signed-off-by: Christian König <[email protected]>
Acked-by: Alex Deucher <[email protected]>
Acked-by: Nirmoy Das <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Use debugfs_create_file_size API for creating ring debugfs, and as its a
NULL returning API, change the return type for amdgpu_debugfs_ring_init
API as well. Also cleanup surrounding code.
Signed-off-by: Nirmoy Das <[email protected]>
Reviewed-by: Shashank Sharma <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
This new debugfs interface uses an IOCTL interface in order to pass
along state information like SRBM and GRBM bank switching. This
new interface also allows a full 32-bit MMIO address range which
the previous didn't. With this new design we have room to grow
the flexibility of the file as need be.
(v2): Move read/write to .read/.write, fix style, add comment
for IOCTL data structure
(v3): C style comments
(v4): use u32 in struct and remove offset variable
(v5): Drop flag clearing in op function, use 0xFFFFFFFF for broadcast
instead of 0x3FF, use mutex for op/ioctl.
Signed-off-by: Tom St Denis <[email protected]>
Reviewed-by: Christian König <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Why: Previously hw fence is alloced separately with job.
It caused historical lifetime issues and corner cases.
The ideal situation is to take fence to manage both job
and fence's lifetime, and simplify the design of gpu-scheduler.
How:
We propose to embed hw_fence into amdgpu_job.
1. We cover the normal job submission by this method.
2. For ib_test, and submit without a parent job keep the
legacy way to create a hw fence separately.
v2:
use AMDGPU_FENCE_FLAG_EMBED_IN_JOB_BIT to show that the fence is
embedded in a job.
v3:
remove redundant variable ring in amdgpu_job
v4:
add tdr sequence support for this feature. Add a job_run_counter to
indicate whether this job is a resubmit job.
v5
add missing handling in amdgpu_fence_enable_signaling
Signed-off-by: Jingwen Chen <[email protected]>
Signed-off-by: Jack Zhang <[email protected]>
Reviewed-by: Andrey Grodzovsky <[email protected]>
Reviewed by: Monk Liu <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Page table entries are now in embedded in VM BO, so
we do not need struct amdgpu_vm_pt. This patch replaces
struct amdgpu_vm_pt with struct amdgpu_vm_bo_base.
Signed-off-by: Nirmoy Das <[email protected]>
Reviewed-by: Christian König <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
the register offset isn't needed division by 4 to pass RREG32_PCIE()
Signed-off-by: Kevin Wang <[email protected]>
Reviewed-by: Lijo Lazar <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Fix the following coccicheck warning:
./drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c:1589:0-23: WARNING:
fops_ib_preempt should be defined with DEFINE_DEBUGFS_ATTRIBUTE
./drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c:1592:0-23: WARNING:
fops_sclk_set should be defined with DEFINE_DEBUGFS_ATTRIBUTE
Reported-by: Abaci Robot <[email protected]>
Signed-off-by: Yang Li <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Use debugfs API directly instead of drm middle layer.
This also includes following debugfs file output changes:
1 amdgpu_evict_vram/amdgpu_evict_gtt output will not contain any braces.
e.g. (0) --> 0
2 amdgpu_gpu_recover output will print return value of
amdgpu_device_gpu_recover() instead of not so important "gpu recover"
message.
v2: * checkpatch.pl: use '0444' instead of S_IRUGO.
* remove S_IFREG from mode.
* remove mode variable.
Signed-off-by: Nirmoy Das <[email protected]>
Reviewed-by: Christian König <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Use debugfs API directly instead of drm middle layer.
v2: * checkpatch.pl: use '0444' instead of S_IRUGO.
* remove S_IFREG from mode.
* remove mode variable.
Signed-off-by: Nirmoy Das <[email protected]>
Reviewed-by: Christian König <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Use debugfs API directly instead of drm middle layer.
v2: * checkpatch.pl: use '0444' instead of S_IRUGO.
* remove S_IFREG from mode.
* remove mode variable.
Signed-off-by: Nirmoy Das <[email protected]>
Reviewed-by: Christian König <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Cleanup unnecessary debugfs dentries and surrounding functions.
v3: remove return value check for debugfs_create_file()
v2: remove ttm_debugfs_entries array.
do not init variables.
Signed-off-by: Nirmoy Das <[email protected]>
Reviewed-by: Christian König <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
amd-drm-next-5.12-2021-01-20:
amdgpu:
- Fix non-x86 build
- W=1 fixes from Lee Jones
- Enable GPU reset on Navy Flounder
- Kernel doc fixes
- SMU workload profile fixes for APUs
- Display updates
- SR-IOV fixes
- Vangogh SMU feature enablment and bug fixes
- GPU reset support for Vangogh
- Misc cleanups
Conflicts:
drivers/gpu/drm/amd/display/dc/dcn10/dcn10_mpc.c
Resolve the conflict by picking the initialization value from amd from
f03e80d2e8 ("drm/amd/display: Initialize stack variable") over the
one Linus picked in 61d791365b ("drm/amd/display: avoid
uninitialized variable warning"). It shouldn't matter.
From: Alex Deucher <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
Signed-off-by: Daniel Vetter <[email protected]>
Required backmerge since we will be based on top of v5.11, and there
has been a request to backmerge already to upstream some features.
Signed-off-by: Maarten Lankhorst <[email protected]>
Rename "ring_mirror_list" to "pending_list",
to describe what something is, not what it does,
how it's used, or how the hardware implements it.
This also abstracts the actual hardware
implementation, i.e. how the low-level driver
communicates with the device it drives, ring, CAM,
etc., shouldn't be exposed to DRM.
The pending_list keeps jobs submitted, which are
out of our control. Usually this means they are
pending execution status in hardware, but the
latter definition is a more general (inclusive)
definition.
Signed-off-by: Luben Tuikov <[email protected]>
Acked-by: Christian König <[email protected]>
Link: https://patchwork.freedesktop.org/patch/405573/
Cc: Alexander Deucher <[email protected]>
Cc: Andrey Grodzovsky <[email protected]>
Cc: Christian König <[email protected]>
Cc: Daniel Vetter <[email protected]>
Signed-off-by: Christian König <[email protected]>
Fixes the following W=1 kernel build warning(s):
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c:308: warning: Function parameter or member 'f' not described in 'amdgpu_debugfs_regs_read'
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c:308: warning: Function parameter or member 'buf' not described in 'amdgpu_debugfs_regs_read'
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c:308: warning: Function parameter or member 'size' not described in 'amdgpu_debugfs_regs_read'
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c:308: warning: Function parameter or member 'pos' not described in 'amdgpu_debugfs_regs_read'
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c:317: warning: Function parameter or member 'f' not described in 'amdgpu_debugfs_regs_write'
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c:317: warning: Function parameter or member 'buf' not described in 'amdgpu_debugfs_regs_write'
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c:317: warning: Function parameter or member 'size' not described in 'amdgpu_debugfs_regs_write'
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c:317: warning: Function parameter or member 'pos' not described in 'amdgpu_debugfs_regs_write'
Cc: Alex Deucher <[email protected]>
Cc: "Christian König" <[email protected]>
Cc: David Airlie <[email protected]>
Cc: Daniel Vetter <[email protected]>
Cc: Sumit Semwal <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Signed-off-by: Lee Jones <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
amd-drm-next-5.11-2020-11-05:
amdgpu:
- Add initial support for Vangogh
- Add support for Green Sardine
- Add initial support for Dimgrey Cavefish
- Scatter/Gather display support for Renoir
- Updates for Sienna Cichlid
- Updates for Navy Flounder
- SMU7 power improvements
- Modifier support for gfx9+
- CI BACO fixes
- Arcturus SMU fixes
- Lots of code cleanups
- DC fixes
- Kernel doc fixes
- Add more GPU HW client information to page fault error logging
- MPO clock tuning for RV
- FP fixes for DCN3 on ARM and PPC
radeon:
- Expose voltage via hwmon on Sumo APUs
amdkfd:
- Fix unique id handling
- Misc fixes
From: Alex Deucher <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
General code indentation and alignment changes such as replace spaces
by tabs or align function arguments as per the coding style
guidelines. The patch corrects issues for various amdgpu_*.c files
for this driver. Issue reported by checkpatch script.
Signed-off-by: Deepak R Varma <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Create new debugfs entry to print memory info using VM buffer
objects.
V2: Added Common function for printing BO info.
Dump more VM lists for evicted, moved, relocated, invalidated.
Removed dumping VM mapped BOs.
V3: Fixed coding style comments, renamed print API and variables.
V4: Fixed coding style comments.
Signed-off-by: Mihir Bhogilal Patel <[email protected]>
Reviewed-by: Christian König <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Add a static inline adev_to_drm() to obtain
the DRM device pointer from an amdgpu_device pointer.
Signed-off-by: Luben Tuikov <[email protected]>
Reviewed-by: Alex Deucher <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Get the amdgpu_device from the DRM device by use
of an inline function, drm_to_adev(). The inline
function resolves a pointer to struct drm_device
to a pointer to struct amdgpu_device.
v2: Use a typed visible static inline function
instead of an invisible macro.
Signed-off-by: Luben Tuikov <[email protected]>
Reviewed-by: Alex Deucher <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
clients don't need reset-lock for synchronization when no
GPU recovery.
v2:
change to return the return value of down_read_killable.
v3:
if GPU recovery begin, VF ignore FLR notification.
Reviewed-by: Monk Liu <[email protected]>
Acked-by: Christian König <[email protected]>
Signed-off-by: Dennis Li <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
if other threads have holden the reset lock, recovery will
fail to try_lock. Therefore we introduce atomic hive->in_reset
and adev->in_gpu_reset, to avoid reentering GPU recovery.
v2:
drop "? true : false" in the definition of amdgpu_in_reset
Reviewed-by: Hawking Zhang <[email protected]>
Signed-off-by: Dennis Li <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
The whole approach wasn't thought through till the end.
We already had a reset lock like this in the past and it caused the same problems like this one.
Completely revert the patch for now and add individual trylock protection to the hardware access functions as necessary.
This reverts commit df9c8d1aa2.
Signed-off-by: Christian König <[email protected]>
Acked-by: Alex Deucher <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
After amdgpu driver loading successfully, we can use
RAP debugfs interface <debugfs_dir>/dri/xxx/rap_test
to trigger RAP test.
Currently only L0 validate test is supported.
v2: refine amdgpu_rap.h
Signed-off-by: Wenhui Sheng <[email protected]>
Reviewed-by: Guchun Chen <[email protected]>
Reviewed-by: Hawking Zhang <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
when GPU hang, driver has multi-paths to enter amdgpu_device_gpu_recover,
the atomic adev->in_gpu_reset and hive->in_reset are used to avoid
re-entering GPU recovery.
During GPU reset and resume, it is unsafe that other threads access GPU,
which maybe cause GPU reset failed. Therefore the new rw_semaphore
adev->reset_sem is introduced, which protect GPU from being accessed by
external threads during recovery.
v2:
1. add rwlock for some ioctls, debugfs and file-close function.
2. change to use dqm->is_resetting and dqm_lock for protection in kfd
driver.
3. remove try_lock and change adev->in_gpu_reset as atomic, to avoid
re-enter GPU recovery for the same GPU hang.
v3:
1. change back to use adev->reset_sem to protect kfd callback
functions, because dqm_lock couldn't protect all codes, for example:
free_mqd must be called outside of dqm_lock;
[ 1230.176199] Hardware name: Supermicro SYS-7049GP-TRT/X11DPG-QT, BIOS 3.1 05/23/2019
[ 1230.177221] Call Trace:
[ 1230.178249] dump_stack+0x98/0xd5
[ 1230.179443] amdgpu_virt_kiq_reg_write_reg_wait+0x181/0x190 [amdgpu]
[ 1230.180673] gmc_v9_0_flush_gpu_tlb+0xcc/0x310 [amdgpu]
[ 1230.181882] amdgpu_gart_unbind+0xa9/0xe0 [amdgpu]
[ 1230.183098] amdgpu_ttm_backend_unbind+0x46/0x180 [amdgpu]
[ 1230.184239] ? ttm_bo_put+0x171/0x5f0 [ttm]
[ 1230.185394] ttm_tt_unbind+0x21/0x40 [ttm]
[ 1230.186558] ttm_tt_destroy.part.12+0x12/0x60 [ttm]
[ 1230.187707] ttm_tt_destroy+0x13/0x20 [ttm]
[ 1230.188832] ttm_bo_cleanup_memtype_use+0x36/0x80 [ttm]
[ 1230.189979] ttm_bo_put+0x1be/0x5f0 [ttm]
[ 1230.191230] amdgpu_bo_unref+0x1e/0x30 [amdgpu]
[ 1230.192522] amdgpu_amdkfd_free_gtt_mem+0xaf/0x140 [amdgpu]
[ 1230.193833] free_mqd+0x25/0x40 [amdgpu]
[ 1230.195143] destroy_queue_cpsch+0x1a7/0x270 [amdgpu]
[ 1230.196475] pqm_destroy_queue+0x105/0x260 [amdgpu]
[ 1230.197819] kfd_ioctl_destroy_queue+0x37/0x70 [amdgpu]
[ 1230.199154] kfd_ioctl+0x277/0x500 [amdgpu]
[ 1230.200458] ? kfd_ioctl_get_clock_counters+0x60/0x60 [amdgpu]
[ 1230.201656] ? tomoyo_file_ioctl+0x19/0x20
[ 1230.202831] ksys_ioctl+0x98/0xb0
[ 1230.204004] __x64_sys_ioctl+0x1a/0x20
[ 1230.205174] do_syscall_64+0x5f/0x250
[ 1230.206339] entry_SYSCALL_64_after_hwframe+0x49/0xbe
2. remove try_lock and introduce atomic hive->in_reset, to avoid
re-enter GPU recovery.
v4:
1. remove an unnecessary whitespace change in kfd_chardev.c
2. remove comment codes in amdgpu_device.c
3. add more detailed comment in commit message
4. define a wrap function amdgpu_in_reset
v5:
1. Fix some style issues.
Reviewed-by: Hawking Zhang <[email protected]>
Suggested-by: Andrey Grodzovsky <[email protected]>
Suggested-by: Christian König <[email protected]>
Suggested-by: Felix Kuehling <[email protected]>
Suggested-by: Lijo Lazar <[email protected]>
Suggested-by: Luben Tukov <[email protected]>
Signed-off-by: Dennis Li <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Remove signaled jobs from job list and ensure the
job was indeed preempted.
Signed-off-by: Jack Xiao <[email protected]>
Reviewed-by: Hawking Zhang <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
This helps to maintain clear code layers and drop unnecessary
parameter.
Signed-off-by: Evan Quan <[email protected]>
Reviewed-by: Alex Deucher <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
This helps to maintain clear code layers and drop unnecessary
parameter.
Signed-off-by: Evan Quan <[email protected]>
Reviewed-by: Alex Deucher <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
The call to pm_runtime_get_sync increments the counter even in case of
failure, leading to incorrect ref count.
In case of failure, decrement the ref count before returning.
Acked-by: Evan Quan <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Fix memory leak in amdgpu_debugfs_gpr_read not freeing data when
amdgpu_virt_enable_access_debugfs failed.
Fixes: 95a2f91738 ("drm/amdgpu: restrict debugfs register access under SR-IOV")
Signed-off-by: Chen Tao <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Fix memory leak in amdgpu_debugfs_gpr_read not freeing data when
pm_runtime_get_sync failed.
Fixes: a9ffe2a983 ("drm/amdgpu/debugfs: properly handle runtime pm")
Signed-off-by: Chen Tao <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
When GPU got timeout, it would notify an interested part
of an opportunity to dump info before actual GPU reset.
A usermode app would open 'autodump' node under debugfs system
and poll() for readable/writable. When a GPU reset is due,
amdgpu would notify usermode app through wait_queue_head and give
it 10 minutes to dump info.
After usermode app has done its work, this 'autodump' node is closed.
On node closure, amdgpu gets to know the dump is done through
the completion that is triggered in release().
There is no write or read callback because necessary info can be
obtained through dmesg and umr. Messages back and forth between
usermode app and amdgpu are unnecessary.
v2: (1) changed 'registered' to 'app_listening'
(2) add a mutex in open() to prevent race condition
v3 (chk): grab the reset lock to avoid race in autodump_open,
rename debugfs file to amdgpu_autodump,
provide autodump_read as well,
style and code cleanups
v4: add 'bool app_listening' to differentiate situations, so that
the node can be reopened; also, there is no need to wait for
completion when no app is waiting for a dump.
v5: change 'bool app_listening' to 'enum amdgpu_autodump_state'
add 'app_state_mutex' for race conditions:
(1)Only 1 user can open this file node
(2)wait_dump() can only take effect after poll() executed.
(3)eliminated the race condition between release() and
wait_dump()
v6: removed 'enum amdgpu_autodump_state' and 'app_state_mutex'
removed state checking in amdgpu_debugfs_wait_dump
Improve on top of version 3 so that the node can be reopened.
v7: move reinit_completion into open() so that only one user
can open it.
v8: remove complete_all() from amdgpu_debugfs_wait_dump().
Signed-off-by: Jiange Zhao <[email protected]>
Reviewed-by: Christian König <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Under bare metal, there is no more else to take
care of the GPU register access through MMIO.
Under Virtualization, to access GPU register is
implemented through KIQ during run-time due to
world-switch.
Therefore, under SR-IOV user can only access
debugfs to r/w GPU registers when meets all
three conditions below.
- amdgpu_gpu_recovery=0
- TDR happened
- in_gpu_reset=0
v2: merge amdgpu_virt_can_access_debugfs() into
amdgpu_virt_enable_access_debugfs()
v3: drop ret variable in amdgpu_virt_enable_access_debugfs()
and directly return result
Signed-off-by: Yintian Tao <[email protected]>
Acked-by: Alex Deucher <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
what changed:
1)provide new implementation interface for the rlcg access path
2)put SQ_CMD/SQ_IND_INDEX to GFX9 RLCG path to let debugfs's reg_op
function can access reg that need RLCG path help
now even debugfs's reg_op can used to dump wave.
tested-by: Monk Liu <[email protected]>
tested-by: Zhou pengju <[email protected]>
Signed-off-by: Zhou pengju <[email protected]>
Signed-off-by: Monk Liu <[email protected]>
Reviewed-by: Emily Deng <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
The offset into the array was specified in bytes but should
be in terms of 32-bit words. Also prevent large reads that
would also cause a buffer overread.
v2: Read from correct offset from internal storage buffer.
Signed-off-by: Tom St Denis <[email protected]>
Acked-by: Christian König <[email protected]>
Reviewed-by: Alex Deucher <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
and remove each ras IP's own debugfs creation
this is required to fix ras when the driver does not use the drm load
and unload callbacks due to ordering issues with the drm device node.
Signed-off-by: Tao Zhou <[email protected]>
Signed-off-by: Stanley.Yang <[email protected]>
Reviewed-by: Alex Deucher <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>