Re: [PATCH 25/33] drm/amdkfd: add debug wave launch mode operation
Am 2023-05-25 um 13:27 schrieb Jonathan Kim: Allow the debugger to set wave behaviour on to either normally operate, halt at launch, trap on every instruction, terminate immediately or stall on allocation. v2: fixup with new kfd_node struct reference for mes check Signed-off-by: Jonathan Kim Reviewed-by: Felix Kuehling --- .../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c | 12 +++ .../drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c | 1 + .../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c| 25 + .../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h| 3 ++ .../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c | 3 +- .../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c| 14 +++- .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c | 25 + .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h | 3 ++ drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 3 ++ drivers/gpu/drm/amd/amdkfd/kfd_debug.c| 36 ++- drivers/gpu/drm/amd/amdkfd/kfd_debug.h| 2 ++ 11 files changed, 124 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c index d7881bbd828d..774ecfc3451a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c @@ -107,6 +107,17 @@ static uint32_t kgd_aldebaran_set_wave_launch_trap_override(struct amdgpu_device return data; } +static uint32_t kgd_aldebaran_set_wave_launch_mode(struct amdgpu_device *adev, + uint8_t wave_launch_mode, + uint32_t vmid) +{ + uint32_t data = 0; + + data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, LAUNCH_MODE, wave_launch_mode); + + return data; +} + const struct kfd2kgd_calls aldebaran_kfd2kgd = { .program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings, .set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping, @@ -129,6 +140,7 @@ const struct kfd2kgd_calls aldebaran_kfd2kgd = { .disable_debug_trap = kgd_aldebaran_disable_debug_trap, .validate_trap_override_request = kgd_aldebaran_validate_trap_override_request, .set_wave_launch_trap_override = kgd_aldebaran_set_wave_launch_trap_override, + .set_wave_launch_mode = kgd_aldebaran_set_wave_launch_mode, .get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times, .build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info, .program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c index ec2587664001..fbdc1b7b1e42 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c @@ -412,6 +412,7 @@ const struct kfd2kgd_calls arcturus_kfd2kgd = { .disable_debug_trap = kgd_arcturus_disable_debug_trap, .validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request, .set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override, + .set_wave_launch_mode = kgd_gfx_v9_set_wave_launch_mode, .get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times, .build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info, .get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c index 7ea0362dcab3..a7a6edda557f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c @@ -856,6 +856,30 @@ uint32_t kgd_gfx_v10_set_wave_launch_trap_override(struct amdgpu_device *adev, return 0; } +uint32_t kgd_gfx_v10_set_wave_launch_mode(struct amdgpu_device *adev, + uint8_t wave_launch_mode, + uint32_t vmid) +{ + uint32_t data = 0; + bool is_mode_set = !!wave_launch_mode; + + mutex_lock(&adev->grbm_idx_mutex); + + kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true); + + data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2, + VMID_MASK, is_mode_set ? 1 << vmid : 0); + data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2, + MODE, is_mode_set ? wave_launch_mode : 0); + WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL2), data); + + kgd_gfx_v10_set_wave_launch_stall(adev, vmid, false); + + mutex_unlock(&adev->grbm_idx_mutex); + + return 0; +} + /* kgd_gfx_v10_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values * The values read are: * ib_offload_wait_time -- Wait Count for Indirect Buffer Offloads. @@ -944,6 +968,7 @@ const struct kfd2kgd_calls gfx_v10_kfd2kgd = { .disable_debug_trap = kgd_gfx_v10_disable_debug_trap,
[PATCH 25/33] drm/amdkfd: add debug wave launch mode operation
Allow the debugger to set wave behaviour on to either normally operate, halt at launch, trap on every instruction, terminate immediately or stall on allocation. v2: fixup with new kfd_node struct reference for mes check Signed-off-by: Jonathan Kim --- .../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c | 12 +++ .../drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c | 1 + .../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c| 25 + .../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.h| 3 ++ .../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10_3.c | 3 +- .../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c| 14 +++- .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c | 25 + .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h | 3 ++ drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 3 ++ drivers/gpu/drm/amd/amdkfd/kfd_debug.c| 36 ++- drivers/gpu/drm/amd/amdkfd/kfd_debug.h| 2 ++ 11 files changed, 124 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c index d7881bbd828d..774ecfc3451a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c @@ -107,6 +107,17 @@ static uint32_t kgd_aldebaran_set_wave_launch_trap_override(struct amdgpu_device return data; } +static uint32_t kgd_aldebaran_set_wave_launch_mode(struct amdgpu_device *adev, + uint8_t wave_launch_mode, + uint32_t vmid) +{ + uint32_t data = 0; + + data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, LAUNCH_MODE, wave_launch_mode); + + return data; +} + const struct kfd2kgd_calls aldebaran_kfd2kgd = { .program_sh_mem_settings = kgd_gfx_v9_program_sh_mem_settings, .set_pasid_vmid_mapping = kgd_gfx_v9_set_pasid_vmid_mapping, @@ -129,6 +140,7 @@ const struct kfd2kgd_calls aldebaran_kfd2kgd = { .disable_debug_trap = kgd_aldebaran_disable_debug_trap, .validate_trap_override_request = kgd_aldebaran_validate_trap_override_request, .set_wave_launch_trap_override = kgd_aldebaran_set_wave_launch_trap_override, + .set_wave_launch_mode = kgd_aldebaran_set_wave_launch_mode, .get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times, .build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info, .program_trap_handler_settings = kgd_gfx_v9_program_trap_handler_settings, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c index ec2587664001..fbdc1b7b1e42 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c @@ -412,6 +412,7 @@ const struct kfd2kgd_calls arcturus_kfd2kgd = { .disable_debug_trap = kgd_arcturus_disable_debug_trap, .validate_trap_override_request = kgd_gfx_v9_validate_trap_override_request, .set_wave_launch_trap_override = kgd_gfx_v9_set_wave_launch_trap_override, + .set_wave_launch_mode = kgd_gfx_v9_set_wave_launch_mode, .get_iq_wait_times = kgd_gfx_v9_get_iq_wait_times, .build_grace_period_packet_info = kgd_gfx_v9_build_grace_period_packet_info, .get_cu_occupancy = kgd_gfx_v9_get_cu_occupancy, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c index 7ea0362dcab3..a7a6edda557f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c @@ -856,6 +856,30 @@ uint32_t kgd_gfx_v10_set_wave_launch_trap_override(struct amdgpu_device *adev, return 0; } +uint32_t kgd_gfx_v10_set_wave_launch_mode(struct amdgpu_device *adev, + uint8_t wave_launch_mode, + uint32_t vmid) +{ + uint32_t data = 0; + bool is_mode_set = !!wave_launch_mode; + + mutex_lock(&adev->grbm_idx_mutex); + + kgd_gfx_v10_set_wave_launch_stall(adev, vmid, true); + + data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2, + VMID_MASK, is_mode_set ? 1 << vmid : 0); + data = REG_SET_FIELD(data, SPI_GDBG_WAVE_CNTL2, + MODE, is_mode_set ? wave_launch_mode : 0); + WREG32(SOC15_REG_OFFSET(GC, 0, mmSPI_GDBG_WAVE_CNTL2), data); + + kgd_gfx_v10_set_wave_launch_stall(adev, vmid, false); + + mutex_unlock(&adev->grbm_idx_mutex); + + return 0; +} + /* kgd_gfx_v10_get_iq_wait_times: Returns the mmCP_IQ_WAIT_TIME1/2 values * The values read are: * ib_offload_wait_time -- Wait Count for Indirect Buffer Offloads. @@ -944,6 +968,7 @@ const struct kfd2kgd_calls gfx_v10_kfd2kgd = { .disable_debug_trap = kgd_gfx_v10_disable_debug_trap, .validate_trap_override_request = kgd_gfx_v10_validate_trap_override_request, .set_wav