diff options
| author | Yong Zhao <Yong.Zhao@amd.com> | 2019-09-25 23:57:30 -0400 |
|---|---|---|
| committer | Alex Deucher <alexander.deucher@amd.com> | 2019-10-03 09:11:04 -0500 |
| commit | 56fc40aba4d9bc22ccbb8ce88a1c41f666d04f05 (patch) | |
| tree | 5047bac876d499703715286c831c39f2420c0784 /drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c | |
| parent | 3fe023d42ea82be7d32ad26ab9409d970c1d361f (diff) | |
| download | linux-stable-56fc40aba4d9bc22ccbb8ce88a1c41f666d04f05.tar.gz linux-stable-56fc40aba4d9bc22ccbb8ce88a1c41f666d04f05.tar.bz2 linux-stable-56fc40aba4d9bc22ccbb8ce88a1c41f666d04f05.zip | |
drm/amdkfd: Eliminate get_atc_vmid_pasid_mapping_valid
get_atc_vmid_pasid_mapping_valid() is very similar to
get_atc_vmid_pasid_mapping_pasid(), so they can be merged into a new
function get_atc_vmid_pasid_mapping_info() to reduce register access
times. More importantly, getting the PASID and the valid bit atomically
with a single read fixes some potential race conditions where the
mapping changes between the two reads.
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c | 28 |
1 files changed, 9 insertions, 19 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c index 8c97be0ff396..f77ddf7dba2b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c @@ -133,9 +133,8 @@ static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd, unsigned int watch_point_id, unsigned int reg_offset); -static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd, uint8_t vmid); -static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd, - uint8_t vmid); +static bool get_atc_vmid_pasid_mapping_info(struct kgd_dev *kgd, + uint8_t vmid, uint16_t *p_pasid); static void set_scratch_backing_va(struct kgd_dev *kgd, uint64_t va, uint32_t vmid); @@ -186,8 +185,7 @@ static const struct kfd2kgd_calls kfd2kgd = { .address_watch_execute = kgd_address_watch_execute, .wave_control_execute = kgd_wave_control_execute, .address_watch_get_offset = kgd_address_watch_get_offset, - .get_atc_vmid_pasid_mapping_pasid = get_atc_vmid_pasid_mapping_pasid, - .get_atc_vmid_pasid_mapping_valid = get_atc_vmid_pasid_mapping_valid, + .get_atc_vmid_pasid_mapping_info = get_atc_vmid_pasid_mapping_info, .set_scratch_backing_va = set_scratch_backing_va, .get_tile_config = get_tile_config, .set_vm_context_page_table_base = set_vm_context_page_table_base, @@ -753,24 +751,16 @@ static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd, return watchRegs[watch_point_id * ADDRESS_WATCH_REG_MAX + reg_offset]; } -static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd, - uint8_t vmid) +static bool get_atc_vmid_pasid_mapping_info(struct kgd_dev *kgd, + uint8_t vmid, uint16_t *p_pasid) { - uint32_t reg; + uint32_t value; struct amdgpu_device *adev = (struct amdgpu_device *) kgd; - reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid); - return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK; -} - -static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd, - uint8_t vmid) -{ - uint32_t reg; - struct amdgpu_device *adev = (struct amdgpu_device *) kgd; + value = RREG32(mmATC_VMID0_PASID_MAPPING + vmid); + *p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK; - reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid); - return reg & ATC_VMID0_PASID_MAPPING__PASID_MASK; + return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK); } static void set_scratch_backing_va(struct kgd_dev *kgd, |