The initialization is not necessary. amd-kfd-staging and ROCm
releases have worked without it for two years.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
It turns out that idr_for_each_entry is really slow compared to just
iterating over the slots. Based on measurements the difference is
estimated to be about a factor 64. That means using idr_for_each_entry
is only worth it with very few allocated events.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
The problem happens on Raven and Carrizo. The context save handler
should not clear the high bits of PC_HI before extracting the bits
of IB_STS.
The bug is not relevant to VEGA10 until we enable demand paging.
Signed-off-by: Jay Cornwall <Jay.Cornwall@amd.com>
Signed-off-by: Yong Zhao <yong.zhao@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Since the assembly code is inside "#if 0", it is ineffective. Despite that,
during debugging, we need to change the assembly code, extract it into
a separate file and compile the new file into hex values using sp3.
That process also requires us to remove "#if 0" and modify lines starting
with "#", so that sp3 can successfully compile the new file.
With this change, all the above chore is no longer needed, and
cwsr_trap_handler_gfx*.asm can be directly used by sp3 to generate its
hex values.
Signed-off-by: Yong Zhao <yong.zhao@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
MTYPE_NC_NV (0) marks scalar/vector L1 cache lines as non-volatile.
Cache lines loaded through these apertures are intended to be
invalidated before (and sometimes during) a dispatch. The non-volatile
qualifier prevents these cache lines from being distinguished from
those loaded through the private aperture.
Use MTYPE_NC (1) instead on both Gfx7 and Gfx8. This allows the
compiler to use the BUFFER_WBINVL1_VOL instruction and is a precursor
to automatic per-dispatch scalar/vector L1 volatile invalidation.
Signed-off-by: Jay Cornwall <Jay.Cornwall@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Synchronization between context-saving wavefronts is achieved by
sending a SAVEWAVE message to the SPI and then spin-waiting for a
response. These spin-waiting wavefronts may inhibit the progress
of other wavefronts in the context save handler, leading to the
synchronization condition never being achieved.
Before spin-waiting reduce the priority of each wavefront to
guarantee foward progress in the others.
Signed-off-by: Jay Cornwall <Jay.Cornwall@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
args->n_devices is a u32 that comes from the user. The multiplication
could overflow on 32 bit systems possibly leading to privilege
escalation.
Fixes: 5ec7e02854 ("drm/amdkfd: Add ioctls for GPUVM memory management")
Signed-off-by: Dan Carpenter dan.carpenter@oracle.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
* Report 64-bit doorbells as HSA_CAP_DOORBELL_TYPE_2_0 in topology
* Report cache information in topology (duplicates GFXv8 info for now)
* Add device info for Vega10 support in KFD
Raven is not enabled at this time as it needs additional changes in
DQM to work with a single SDMA engine.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Report failure to enable atomics only on GPUs that require them.
This allows GPUs that don't require atomics to function, but can
benefit if they are available. This is the case for Vega10, which
doesn't use atomics for basic functioning of the MEC, AQL and HWS
microcode. So it can work without atomics. But shader programs can
still use atomic instructions on systems that support PCIe atomics.
Signed-off-by: welu <Wei.Lu2@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
This condition was missed in a previous commit with the same title.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
kq->queue->properties.write_ptr is a GPU address which can'd be
derefenced in the kernel. Use kq->wptr_kernel instead, which is the
kernel CPU address of the same buffer.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Missed a spot in previous cleanup commit:
Remove gotos that do not feature any common cleanup, and use gotos
instead of repeating cleanup commands.
According to kernel.org: "The goto statement comes in handy when a
function exits from multiple locations and some common work such as
cleanup has to be done. If there is no cleanup needed then just return
directly."
Signed-off-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
This is in preparation for GFXv9 (Vega10) which uses incompatible PM4
packet formats from previous ASIC generations.
Signed-off-by: Shaoyun Liu <Shaoyun.Liu@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Allocate doorbells according to the doorbell routing information on
SOC15 ASICs (Vega10 and later). On older ASICs we continue to use the
queue_id as the doorbell ID to maintain compatibility with the Thunk.
Signed-off-by: Shaoyun Liu <Shaoyun.Liu@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Use bit-rotate for better clarity and remove _MASK from the #defines as
these represent mmap types.
Centralize all the parsing of the mmap offset in kfd_mmap and add device
parameter to doorbell and reserved_mem map functions.
Encode gpu_id into upper bits of vm_pgoff. This frees up the lower bits
for encoding the the doorbell ID on Vega10.
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
This prepares for GFXv9 (Vega10), which has 64-bit doorbells.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
-----BEGIN PGP SIGNATURE-----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=DrA+
-----END PGP SIGNATURE-----
Merge tag 'drm-for-v4.17' of git://people.freedesktop.org/~airlied/linux
Pull drm updates from Dave Airlie:
"Cannonlake and Vega12 support are probably the two major things. This
pull lacks nouveau, Ben had some unforseen leave and a few other
blockers so we'll see how things look or maybe leave it for this merge
window.
core:
- Device links to handle sound/gpu pm dependency
- Color encoding/range properties
- Plane clipping into plane check helper
- Backlight helpers
- DP TP4 + HBR3 helper support
amdgpu:
- Vega12 support
- Enable DC by default on all supported GPUs
- Powerplay restructuring and cleanup
- DC bandwidth calc updates
- DC backlight on pre-DCE11
- TTM backing store dropping support
- SR-IOV fixes
- Adding "wattman" like functionality
- DC crc support
- Improved DC dual-link handling
amdkfd:
- GPUVM support for dGPU
- KFD events for dGPU
- Enable PCIe atomics for dGPUs
- HSA process eviction support
- Live-lock fixes for process eviction
- VM page table allocation fix for large-bar systems
panel:
- Raydium RM68200
- AUO G104SN02 V2
- KEO TX31D200VM0BAA
- ARM Versatile panels
i915:
- Cannonlake support enabled
- AUX-F port support added
- Icelake base enabling until internal milestone of forcewake support
- Query uAPI interface (used for GPU topology information currently)
- Compressed framebuffer support for sprites
- kmem cache shrinking when GPU is idle
- Avoid boosting GPU when waited item is being processed already
- Avoid retraining LSPCON link unnecessarily
- Decrease request signaling latency
- Deprecation of I915_SET_COLORKEY_NONE
- Kerneldoc and compiler warning cleanup for upcoming CI enforcements
- Full range ycbcr toggling
- HDCP support
i915/gvt:
- Big refactor for shadow ppgtt
- KBL context save/restore via LRI cmd (Weinan)
- Properly unmap dma for guest page (Changbin)
vmwgfx:
- Lots of various improvements
etnaviv:
- Use the drm gpu scheduler
- prep work for GC7000L support
vc4:
- fix alpha blending
- Expose perf counters to userspace
pl111:
- Bandwidth checking/limiting
- Versatile panel support
sun4i:
- A83T HDMI support
- A80 support
- YUV plane support
- H3/H5 HDMI support
omapdrm:
- HPD support for DVI connector
- remove lots of static variables
msm:
- DSI updates from 10nm / SDM845
- fix for race condition with a3xx/a4xx fence completion irq
- some refactoring/prep work for eventual a6xx support (ie. when we
have a userspace)
- a5xx debugfs enhancements
- some mdp5 fixes/cleanups to prepare for eventually merging
writeback
- support (ie. when we have a userspace)
tegra:
- mmap() fixes for fbdev devices
- Overlay plane for hw cursor fix
- dma-buf cache maintenance support
mali-dp:
- YUV->RGB conversion support
rockchip:
- rk3399/chromebook fixes and improvements
rcar-du:
- LVDS support move to drm bridge
- DT bindings for R8A77995
- Driver/DT support for R8A77970
tilcdc:
- DRM panel support"
* tag 'drm-for-v4.17' of git://people.freedesktop.org/~airlied/linux: (1646 commits)
drm/i915: Fix hibernation with ACPI S0 target state
drm/i915/execlists: Use a locked clear_bit() for synchronisation with interrupt
drm/i915: Specify which engines to reset following semaphore/event lockups
drm/i915/dp: Write to SET_POWER dpcd to enable MST hub.
drm/amdkfd: Use ordered workqueue to restore processes
drm/amdgpu: Fix acquiring VM on large-BAR systems
drm/amd/pp: clean header file hwmgr.h
drm/amd/pp: use mlck_table.count for array loop index limit
drm: Fix uabi regression by allowing garbage mode->type from userspace
drm/amdgpu: Add an ATPX quirk for hybrid laptop
drm/amdgpu: fix spelling mistake: "asssert" -> "assert"
drm/amd/pp: Add new asic support in pp_psm.c
drm/amd/pp: Clean up powerplay code on Vega12
drm/amd/pp: Add smu irq handlers for legacy asics
drm/amd/pp: Fix set wrong temperature range on smu7
drm/amdgpu: Don't change preferred domian when fallback GTT v5
drm/vmwgfx: Bump version patchlevel and date
drm/vmwgfx: use monotonic event timestamps
drm/vmwgfx: Unpin the screen object backup buffer when not used
drm/vmwgfx: Stricter count of legacy surface device resources
...
These interfaces allow KGD to stop and resume all GPU user mode queue
access to a process address space. This is needed for handling MMU
notifiers of userptrs mapped for GPU access in KFD VMs.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
When an MMU notifier runs in memory reclaim context, it can deadlock
trying to take locks that are already held in the thread causing the
memory reclaim. The solution is to avoid memory reclaim while holding
locks that are taken in MMU notifiers by using GFP_NOIO.
This commit fixes memory allocations done while holding the dqm->lock
which is needed in the MMU notifier (dqm->ops.evict_process_queues).
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Restoring multiple processes concurrently can lead to live-locks
where each process prevents the other from validating all its BOs.
v2: fix duplicate check of same variable
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Deallocate SDMA queues during abnormal process termination and when
queue creation fails after the SDMA allocation.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Program sh_hidden_private_base_vmid correctly in the map-process
PM4 packet.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Simulate large-BAR system by exporting only visible memory. This
limits the amount of available VRAM to the size of the BAR, but
enables CPU access to VRAM.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
The events page must be accessible in user mode by the GPU and CPU
as well as in kernel mode by the CPU. On dGPUs user mode virtual
addresses are managed by the Thunk's GPU memory allocation code.
Therefore we can't allocate the memory in kernel mode like we do
on APUs. But KFD still needs to map the memory for kernel access.
To facilitate this, the Thunk provides the buffer handle of the
events page to KFD when creating the first event.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
v2:
* Fix error handling after kfd_bind_process_to_device in
kfd_ioctl_map_memory_to_gpu
v3:
* Add ioctl to acquire VM from a DRM FD
v4:
* Return number of successful map/unmap operations in failure cases
* Facilitate partial retry after failed map/unmap
* Added comments with parameter descriptions to new APIs
* Defined AMDKFD_IOC_FREE_MEMORY_OF_GPU write-only
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
On GFX7 the CP does not perform a TC flush when queues are unmapped.
To avoid TC eviction from accessing an invalid VMID, flush it
explicitly before releasing a VMID.
v2: Fix unnecessary list_for_each_entry_safe
v3: Moved allocation to kfd_process_device_init_vm
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Add helpers for allocating GPUVM memory in kernel mode and use them
to allocate memory for the CWSR trap handler.
v2: Use dev instead of pdd->dev in kfd_process_free_gpuvm
v3:
* Cleaned up and simplified kfd_process_alloc_gpuvm
* Moved allocation for dGPU to kfd_process_device_init_vm
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Also used for cleaning up on process termination.
v2: Refactored cleanup on process termination
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Set up the GPUVM aperture for SVM (shared virtual memory) that allows
sharing a part of virtual address space between GPUs and CPUs.
Report the size of the GPUVM aperture that is supported by KGD accurately.
The low part of the GPUVM aperture is reserved for kernel use. This is
for kernel-allocated buffers that are only accessed on the GPU:
- CWSR trap handler
- IB for submitting commands in user-mode context from kernel mode
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Currently the number of GPUs is limited by aperture placement options
available on GFX7 and GFX8 hardware. This limitation is not necessary.
Scratch and LDS represent per-work-item and per-work-group storage
respectively. Different work-items and work-groups use the same virtual
address to access their own data. Work running on different GPUs is by
definition in different work-groups (different dispatches, in fact).
That means the same virtual addresses can be used for these apertures
on different GPUs.
Add a new AMDKFD_IOC_GET_PROCESS_APERTURES_NEW ioctl that removes the
artificial limitation on the number of GPUs that can be supported. The
new ioctl allows user mode to query the number of GPUs to allocate
enough memory for all GPUs to be reported.
This deprecates AMDKFD_IOC_GET_PROCESS_APERTURES.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Populate DRM render device minor in kfd topology
Signed-off-by: Oak Zeng <Oak.Zeng@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Instead of creating all VMs on process creation, create them when
a process is bound to a device. This will later allow registering
an existing VM from a DRM render node FD at runtime, before the
process is bound to the device. This way the render node VM can be
used for KFD instead of creating our own redundant VM.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
When CONFIG_ACPI is disabled, we never initialize the acpi_table
structure in kfd_create_crat_image_virtual:
drivers/gpu/drm/amd/amdkfd/kfd_crat.c: In function 'kfd_create_crat_image_virtual':
drivers/gpu/drm/amd/amdkfd/kfd_crat.c:888:40: error: 'acpi_table' may be used uninitialized in this function [-Werror=maybe-uninitialized]
The undefined behavior also happens for any other acpi_get_table()
failure, but then the compiler can't warn about it.
This adds an error check that prevents the structure from
being used in error, avoiding both the undefined behavior and
the warning about it.
Fixes: 520b8fb755 ("drm/amdkfd: Add topology support for CPUs")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
When the TTM memory manager in KGD evicts BOs, all user mode queues
potentially accessing these BOs must be evicted temporarily. Once
user mode queues are evicted, the eviction fence is signaled,
allowing the migration of the BO to proceed.
A delayed worker is scheduled to restore all the BOs belonging to
the evicted process and restart its queues.
During suspend/resume of the GPU we also evict all processes to allow
KGD to save BOs in system memory, since VRAM will be lost.
v2:
* Account for eviction when updating of q->is_active in MQD manager
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Create/destroy the GPUVM context during PDD creation/destruction.
Get VM page table base and program it during process registration
(HWS) or VMID allocation (non-HWS).
v2:
* Used dev instead of pdd->dev in kfd_flush_tlb
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Unaligned atomic operations can cause problems on some CPU
architectures. Use simpler bitmask operations instead. Atomic bit
manipulations are not necessary since dqm->lock is held during these
operations.
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
In case kfd_get_process_device_data returns null, there are some
null pointer dereferences in functions kfd_bind_processes_to_device
and kfd_unbind_processes_from_device.
Fix this by printing a WARN_ON for PDDs that aren't found and skip
them with continue statements.
Addresses-Coverity-ID: 1463794 ("Dereference null return value")
Addresses-Coverity-ID: 1463772 ("Dereference null return value")
Suggested-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Gustavo A. R. Silva <garsilva@embeddedor.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
When destroying an inactive queue, we don't need to call
execute_queues_cpsch.
Signed-off-by: Yong Zhao <yong.zhao@amd.com>
Reviewed-by: Oak Zeng <oak.zeng@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Some AMD motherboards without an APU have a broken CRAT table which
causes KFD initialization failures or incorrect information about
NUMA nodes, CPU cores or system memory. Ignore CRAT tables without
GPUs and rely on KFD's code to create a CRAT table for the CPU.
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Some systems have broken CRAT tables. Add a module option to ignore
a CRAT table.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
This is needed for enabling a user-mode workaround for an AQL queue
wrapping HW bug on Tonga.
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
* Wrong value for max_waves_per_simd
* Missing ATC capability bit
Signed-off-by: Philip Cox <Philip.Cox@amd.com>
Signed-off-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
For hardware blocks whose performance counters are accessed via MMIO
registers, KFD provides the support for those privileged blocks.
IOMMU is one of those privileged blocks. Most performance counter properties
required by Thunk are available at /sys/bus/event_source/devices/amd_iommu.
This patch adds properties to topology in KFD sysfs for information not
available in /sys/bus/event_source/devices/amd_iommu. They are shown at
/sys/devices/virtual/kfd/kfd/topology/nodes/0/perf/iommu/ formatted as
/sys/devices/virtual/kfd/kfd/topology/nodes/0/perf/<block>/<property>, i.e.
/sys/devices/virtual/kfd/kfd/topology/nodes/0/perf/iommu/max_concurrent.
For dGPUs, who don't have IOMMU, nothing appears under
/sys/devices/virtual/kfd/kfd/topology/nodes/0/perf.
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Signed-off-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Generate and parse VCRAT tables for dGPUs in kfd_topology_add_device.
Some information that isn't available in the CRAT table is patched
into the topology after parsing.
HSA_CAP_DOORBELL_TYPE_1_0 is dependent on the ASIC feature
CP_HQD_PQ_CONTROL.SLOT_BASED_WPTR, which was not introduced in VI
until Carrizo. Report HSA_CAP_DOORBELL_TYPE_PRE_1_0 on Tonga ASICs.
v2: Added #include <linux/pci.h> to kfd_crat.c to make it compile
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Signed-off-by: Jay Cornwall <Jay.Cornwall@amd.com>
Signed-off-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Currently, the KFD topology information is generated by parsing the CRAT
(ACPI) table. However, at present CRAT table is available only for AMD
APUs. To support CPUs on systems without a CRAT table, the KFD driver will
create a Virtual CRAT (VCRAT) table and then the existing code will parse
that table to generate topology.
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Change kfd_cache_properties.sibling_map[256] to
kfd_cache_properties.sibling_map[32]. Since, CRAT uses bitmap for
sibling_map, it is more efficient to use bitmap in the kfd structure
also.
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Only count memory banks in one place. Ignore redundant num_banks
entry in crat_subtype_computeunit.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Modify kfd_topology_enum_kfd_devices(..) function to support non-GPU
nodes. The function returned NULL when it encountered non-GPU (say CPU)
nodes. This caused kfd_ioctl_create_event and kfd_init_apertures to fail
for Intel + Tonga.
kfd_topology_enum_kfd_devices will now parse all the nodes and return
valid kfd_dev for nodes with GPU.
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Currently, CRAT parsing is intertwined with topology_device_list and
hence repeated calls to kfd_parse_crat_table() will fail. Decouple
kfd_parse_crat_table() and topology_device_list.
kfd_parse_crat_table() will parse CRAT and add topology devices to a
temporary list temp_topology_device_list and then
kfd_topology_update_device_list will move contents from temporary list to
master list.
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Signed-off-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Reorganize and rename kfd_topology_get_crat_acpi function. In this way
acpi_get_table(..) needs to be called only once. This will also aid in
dGPU topology implementation.
Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
Signed-off-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Take CRAT related functions out of kfd_topology.c and place them in
kfd_crat.c. This is the initial step of supporting more CRAT features,
i.e. creating virtual CRAT table for KFD devices without CRAT.
v2: Minor cleanup that was missed previously because code moved around
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Signed-off-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Kobject created using kobject_create_and_add() can be freed using
kobject_put() when there is no referenece any more. However,
kobject memory allocated with kzalloc() has to set up a release
callback in order to free it when the counter decreases to 0.
Otherwise it causes memory leak.
Signed-off-by: Yong Zhao <yong.zhao@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Overwrite the active simd_count from KGD at driver loading time. This is
based on assumption that register GC_USER_SHADER_ARRAY_CONFIG won’t get
changed.
V2: remove the incorrect simd_count reported at loading module.
Signed-off-by: Flora Cui <flora.cui@amd.com>
Reviewed by: Yair Shachar< yair.shachar@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
dGPUs work without IOMMUv2. Make IOMMUv2 initialization dependent on
ASIC information. Also allow building KFD without IOMMUv2 support.
This is still useful for dGPUs and prepares for enabling KFD on
architectures that don't support AMD IOMMUv2.
v2:
* Centralize IOMMUv2 code to avoid #ifdefs in too many places
v3:
* Imply AMD_IOMMU_V2 in Kconfig
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian Konig <christian.koenig@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
On dGPUs don't set ATC addressing bits and use MTYPE_UC for coherent
memory.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
GFXv7 and v8 dGPUs use a different addressing mode for KFD compared
to APUs (GPUVM64 vs HSA64). And dGPUs don't support MTYPE_CC. They
use MTYPE_UC instead for memory that requires coherency.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Some dGPUs don't support HWS. Allow them to use a per-device
sched_policy that may be different from the global default.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
This will be needed for most dGPUs.
CC: linux-pci@vger.kernel.org
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Use ARRAY_SIZE instead of dividing sizeof array with sizeof an element.
This issue was detected with the help of Coccinelle.
Signed-off-by: Gustavo A. R. Silva <garsilva@embeddedor.com>
Reviewed-by: Felix Kuehling<Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
If no matching process is found, return NULL instead of a pointer
to the last process in the kfd_processes_table.
Signed-off-by: Yong Zhao <yong.zhao@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Use a reference counter instead of a lock to prevent process
destruction while functions running out of process context are using
the kfd_process structure. In many cases these functions don't need
the structure to be locked. In the few cases that really do need the
process lock, take it explicitly.
This helps simplify lock dependencies between the process lock and
other locks, particularly amdgpu and mm_struct locks. This will be
important when amdgpu calls back to amdkfd for memory evictions.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
This will be used to elliminate the use of the process lock for
preventing concurrent process destruction. This will simplify lock
dependencies between KFD and KGD.
This also simplifies the process destruction in a few ways:
* Don't allocate work struct dynamically
* Remove unnecessary hack that increments mm reference counter
* Remove unnecessary process locking during destruction
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Increment the kfd_process.lead_thread's reference counter to make
it safe to dereference. This is needed for getting a safe reference
to the process' mm_struct.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
This commit adds several debugfs entries for kfd:
kfd/hqds: dumps all HQDs on all GPUs for KFD-controlled compute and
SDMA RLC queues
kfd/mqds: dumps all MQDs of all KFD processes on all GPUs
kfd/rls: dumps HWS runlists on all GPUs
Signed-off-by: Yong Zhao <yong.zhao@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Allow HWS to to execute multiple processes on the hardware
concurrently. The number of concurrent processes is limited by
the number of VMIDs allocated to the HWS.
A module parameter can be used for limiting this further or turn
it off altogether (mainly for debugging purposes).
Signed-off-by: Yong Zhao <yong.zhao@amd.com>
Signed-off-by: Jay Cornwall <Jay.Cornwall@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Just print a pointer instead of casting
v2: Remove the 0x prefix, since %p prints that automatically, and remove
it from one other spot as well
Signed-off-by: Kent Russell <kent.russell@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
After debugger is registered, the pqm_destroy_queue fails because is_debug
is true, the queue should not be removed from process_queue_list since
the count is not reduced.
Test application calls debugger unregister without register debugger, add
null pointer check protection to avoid crash for this case
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
A second-level user mode trap handler can be installed. The CWSR trap
handler jumps to the secondary trap handler conditionally for any
conditions not handled by it. This can be used e.g. for debugging or
catching math exceptions.
When CWSR is disabled, the user mode trap handler is installed as
first level trap handler.
Signed-off-by: Shaoyun.liu <shaoyun.liu@amd.com>
Signed-off-by: Jay Cornwall <Jay.Cornwall@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
This hardware feature allows the GPU to preempt shader execution in
the middle of a compute wave, save the state and restore it later
to resume execution.
Memory for saving the state is allocated per queue in user mode and
the address and size passed to the create_queue ioctl. The size
depends on the number of waves that can be in flight simultaneously
on a given ASIC.
Signed-off-by: Shaoyun.liu <shaoyun.liu@amd.com>
Signed-off-by: Yong Zhao <yong.zhao@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
The trap handler is like an interrupt handler running on the GPU
compute unit. It is needed for supporting CWSR (compute wave
save/restore).
This file defines an array with the pre-compiled GFXv8 shader ISA.
The assembly code is included for reference in #if 0 ... #endif.
Signed-off-by: Shaoyun.liu <shaoyun.liu@amd.com>
Signed-off-by: Jay Cornwall <Jay.Cornwall@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
The PQM doesn't change after process creation. So initialize it in
kfd_create_process_device_data.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Replace (ffs(size) - 1) with order_base_2(size) as a more straight
forward way to get log2 of buffer sizes.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Don't use sizeof(uint32_t) or similar types for hardware or firmware
DWORD size. The hardware and firmware don't care about Linux types.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Add wptr and mm parameters to hqd_sdma_load and pass these parameters
from device_queue_manager through the mqd_manager.
SDMA doesn't support polling while the engine believes it's idle. The
driver must update the wptr. The new parameters will be used for looking
up the updated value from the specified mm when SDMA queues are resumed
after being disabled.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Was missing license text.
Acked-by: Harry Wentland <harry.wentland@amd.com>
Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Felix Kuehling