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>
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>
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>
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>
