drm/amdgpu/vcn2.0: stall DPG when WPTR/RPTR reset
Subsystems (2)
PATCH 1/5 drm/amdgpu/vcn2.0: stall DPG when WPTR/RPTR reset
Add vcn dpg harware synchronization to fix race condition issue between vcn driver and hardware.
--- drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+)
drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +16 -0
@@ -876,6 +876,10 @@ static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect) tmp =REG_SET_FIELD (tmp,UVD_RBC_RB_CNTL ,RB_RPTR_WR_EN ,1 );WREG32_SOC15 (UVD ,0 , mmUVD_RBC_RB_CNTL, tmp); +/* Stall DPG before WPTR/RPTR reset */ +WREG32_P (SOC15_REG_OFFSET (UVD ,0 , mmUVD_POWER_STATUS), +UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK , + ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK );/* set the write pointer delay */ WREG32_SOC15 (UVD ,0 , mmUVD_RBC_RB_WPTR_CNTL,0 );
@@ -898,6 +902,9 @@ static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)WREG32_SOC15 (UVD ,0 , mmUVD_RBC_RB_WPTR,lower_32_bits (ring->wptr)); +/* Unstall DPG */ +WREG32_P (SOC15_REG_OFFSET (UVD ,0 , mmUVD_POWER_STATUS), +0 , ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK );return 0 ; }
@@ -1193,8 +1200,13 @@ static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK , UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, ret_code); +/* Stall DPG before WPTR/RPTR reset */ +WREG32_P (SOC15_REG_OFFSET (UVD ,0 , mmUVD_POWER_STATUS), +UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK , + ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK );/* Restore */ ring = &adev->vcn.inst->ring_enc[0 ]; + ring->wptr =0 ;WREG32_SOC15 (UVD ,0 , mmUVD_RB_BASE_LO, ring->gpu_addr);WREG32_SOC15 (UVD ,0 , mmUVD_RB_BASE_HI,upper_32_bits (ring->gpu_addr));WREG32_SOC15 (UVD ,0 , mmUVD_RB_SIZE, ring->ring_size /4 );
@@ -1202,6 +1214,7 @@ static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,WREG32_SOC15 (UVD ,0 , mmUVD_RB_WPTR,lower_32_bits (ring->wptr)); ring = &adev->vcn.inst->ring_enc[1 ]; + ring->wptr =0 ;WREG32_SOC15 (UVD ,0 , mmUVD_RB_BASE_LO2, ring->gpu_addr);WREG32_SOC15 (UVD ,0 , mmUVD_RB_BASE_HI2,upper_32_bits (ring->gpu_addr));WREG32_SOC15 (UVD ,0 , mmUVD_RB_SIZE2, ring->ring_size /4 );
@@ -1210,6 +1223,9 @@ static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,WREG32_SOC15 (UVD ,0 , mmUVD_RBC_RB_WPTR,RREG32_SOC15 (UVD ,0 , mmUVD_SCRATCH2) &0x7FFFFFFF ); +/* Unstall DPG */ +WREG32_P (SOC15_REG_OFFSET (UVD ,0 , mmUVD_POWER_STATUS), +0 , ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK ); SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON,
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 2/5 drm/amdgpu/vcn2.5: stall DPG when WPTR/RPTR reset
Add vcn dpg harware synchronization to fix race condition issue between vcn driver and hardware.
--- drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+)
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +21 -0
@@ -855,6 +855,12 @@ static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, boo tmp =REG_SET_FIELD (tmp,UVD_RBC_RB_CNTL ,RB_RPTR_WR_EN ,1 );WREG32_SOC15 (UVD , inst_idx, mmUVD_RBC_RB_CNTL, tmp); +/* Stall DPG before WPTR/RPTR reset */ +WREG32_P (SOC15_REG_OFFSET (UVD , inst_idx, mmUVD_POWER_STATUS), +UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK , + ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK ); + fw_shared->multi_queue.decode_queue_mode |= fw_queue_ring_reset; +/* set the write pointer delay */ WREG32_SOC15 (UVD , inst_idx, mmUVD_RBC_RB_WPTR_CNTL,0 );
@@ -877,6 +883,10 @@ static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, booWREG32_SOC15 (UVD , inst_idx, mmUVD_RBC_RB_WPTR,lower_32_bits (ring->wptr)); +/* Unstall DPG */ +WREG32_P (SOC15_REG_OFFSET (UVD , inst_idx, mmUVD_POWER_STATUS), +0 , ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK ); +return 0 ; }
@@ -1389,8 +1399,14 @@ static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK , UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, ret_code); +/* Stall DPG before WPTR/RPTR reset */ +WREG32_P (SOC15_REG_OFFSET (UVD , inst_idx, mmUVD_POWER_STATUS), +UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK , + ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK ); +/* Restore */ ring = &adev->vcn.inst[inst_idx].ring_enc[0 ]; + ring->wptr =0 ;WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_BASE_HI,upper_32_bits (ring->gpu_addr));WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_SIZE, ring->ring_size /4 );
@@ -1398,6 +1414,7 @@ static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_WPTR,lower_32_bits (ring->wptr)); ring = &adev->vcn.inst[inst_idx].ring_enc[1 ]; + ring->wptr =0 ;WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_BASE_HI2,upper_32_bits (ring->gpu_addr));WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_SIZE2, ring->ring_size /4 );
@@ -1407,6 +1424,10 @@ static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,WREG32_SOC15 (UVD , inst_idx, mmUVD_RBC_RB_WPTR,RREG32_SOC15 (UVD , inst_idx, mmUVD_SCRATCH2) &0x7FFFFFFF ); +/* Unstall DPG */ +WREG32_P (SOC15_REG_OFFSET (UVD , inst_idx, mmUVD_POWER_STATUS), +0 , ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK ); +SOC15_WAIT_ON_RREG (UVD , inst_idx, mmUVD_POWER_STATUS,UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON ,UVD_POWER_STATUS__UVD_POWER_STATUS_MASK , ret_code); }
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 3/5 drm/amdgpu/vcn: Add firmware share memory support
In reply to James ZhuAdded firmware share memory support for VCN. Current multiple queue mode is enabled only.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 13 +++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 24 ++++++++++++++++++++++++ 2 files changed, 37 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +13 -0
@@ -182,6 +182,14 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev)return r; } } + + r =amdgpu_bo_create_kernel (adev,AMDGPU_GPU_PAGE_ALIGN (sizeof (struct amdgpu_fw_shared )), +PAGE_SIZE ,AMDGPU_GEM_DOMAIN_VRAM , &adev->vcn.inst[i].fw_shared_bo, + &adev->vcn.inst[i].fw_shared_gpu_addr, &adev->vcn.inst[i].fw_shared_cpu_addr); +if (r) { +dev_err (adev->dev,"VCN %d (%d) failed to allocate fimware shared bo\n" , i, r); +return r; + } }return 0 ;
@@ -196,6 +204,11 @@ int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)for (j =0 ; j < adev->vcn.num_vcn_inst; ++j) {if (adev->vcn.harvest_config & (1 << j))continue ; + +amdgpu_bo_free_kernel (&adev->vcn.inst[j].fw_shared_bo, + &adev->vcn.inst[j].fw_shared_gpu_addr, + (void **)&adev->vcn.inst[j].fw_shared_cpu_addr); +if (adev->vcn.indirect_sram) { amdgpu_bo_free_kernel(&adev->vcn.inst[j].dpg_sram_bo, &adev->vcn.inst[j].dpg_sram_gpu_addr,
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +24 -0
@@ -179,11 +179,14 @@ struct amdgpu_vcn_inst {struct amdgpu_irq_src irq;struct amdgpu_vcn_reg external;struct amdgpu_bo *dpg_sram_bo; +struct amdgpu_bo *fw_shared_bo;struct dpg_pause_state pause_state;void *dpg_sram_cpu_addr;uint64_t dpg_sram_gpu_addr;uint32_t *dpg_sram_curr_addr;atomic_t dpg_enc_submission_cnt; +void *fw_shared_cpu_addr; +uint64_t fw_shared_gpu_addr; };struct amdgpu_vcn {
@@ -209,6 +212,27 @@ struct amdgpu_vcn { int inst_idx, struct dpg_pause_state *new_state); }; +#define AMDGPU_VCN_MULTI_QUEUE_FLAG (1 << 8) + +enum fw_queue_mode { + fw_queue_ring_reset =1 , + fw_queue_dpg_hold_off =2 , +}; + +struct amdgpu_fw_shared_multi_queue { +uint8_t decode_queue_mode; +uint8_t encode_generalpurpose_queue_mode; +uint8_t encode_lowlatency_queue_mode; +uint8_t encode_realtime_queue_mode; +uint8_t padding[4 ]; +}; + +struct amdgpu_fw_shared { +uint32_t present_flag_0; +uint8_t pad[53 ]; +struct amdgpu_fw_shared_multi_queue multi_queue; +} __attribute__((__packed__ )); +int amdgpu_vcn_sw_init (struct amdgpu_device *adev);int amdgpu_vcn_sw_fini (struct amdgpu_device *adev);int amdgpu_vcn_suspend (struct amdgpu_device *adev);
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Re: [PATCH 3/5] drm/amdgpu/vcn: Add firmware share memory support
Quoted 60 lines · James Zhu
Added firmware share memory support for VCN. Current multiple queue mode is enabled only.
Signed-off-by James Zhu <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 13 +++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 24 ++++++++++++++++++++++++ 2 files changed, 37 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +13
-0@@ -182,6 +182,14 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev)return r; } } + + r =amdgpu_bo_create_kernel (adev,AMDGPU_GPU_PAGE_ALIGN (sizeof (struct amdgpu_fw_shared )), +PAGE_SIZE ,AMDGPU_GEM_DOMAIN_VRAM , &adev->vcn.inst[i].fw_shared_bo, + &adev->vcn.inst[i].fw_shared_gpu_addr, &adev->vcn.inst[i].fw_shared_cpu_addr); +if (r) { +dev_err (adev->dev,"VCN %d (%d) failed to allocate fimware shared bo\n" , i, r); +return r; + } }return 0 ;@@ -196,6 +204,11 @@ int amdgpu_vcn_sw_fini(struct amdgpu_device *adev)for (j =0 ; j < adev->vcn.num_vcn_inst; ++j) {if (adev->vcn.harvest_config & (1 << j))continue ; + +amdgpu_bo_free_kernel (&adev->vcn.inst[j].fw_shared_bo, + &adev->vcn.inst[j].fw_shared_gpu_addr, + (void **)&adev->vcn.inst[j].fw_shared_cpu_addr); +if (adev->vcn.indirect_sram) { amdgpu_bo_free_kernel(&adev->vcn.inst[j].dpg_sram_bo, &adev->vcn.inst[j].dpg_sram_gpu_addr,drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +3
-0@@ -179,11 +179,14 @@ struct amdgpu_vcn_inst {struct amdgpu_irq_src irq;struct amdgpu_vcn_reg external;struct amdgpu_bo *dpg_sram_bo; +struct amdgpu_bo *fw_shared_bo;struct dpg_pause_state pause_state;void *dpg_sram_cpu_addr;uint64_t dpg_sram_gpu_addr;uint32_t *dpg_sram_curr_addr;atomic_t dpg_enc_submission_cnt; +void *fw_shared_cpu_addr; +uint64_t fw_shared_gpu_addr; };struct amdgpu_vcn {
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h struct amdgpu_vcn { · line 220
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +6
-0@@ -209,6 +212,27 @@ struct amdgpu_vcn { int inst_idx, struct dpg_pause_state *new_state); }; +#define AMDGPU_VCN_MULTI_QUEUE_FLAG (1 << 8) + +enum fw_queue_mode { + fw_queue_ring_reset =1 , + fw_queue_dpg_hold_off =2 , +};
Please move the define and enum to the top as others. With that fixed, the series are
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +8
-0+ +struct amdgpu_fw_shared_multi_queue { +uint8_t decode_queue_mode; +uint8_t encode_generalpurpose_queue_mode; +uint8_t encode_lowlatency_queue_mode; +uint8_t encode_realtime_queue_mode; +uint8_t padding[4 ]; +};
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h struct amdgpu_vcn { · line 238
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +7
-0+ +struct amdgpu_fw_shared { +uint32_t present_flag_0; +uint8_t pad[53 ]; +struct amdgpu_fw_shared_multi_queue multi_queue; +} __attribute__((__packed__ )); +int amdgpu_vcn_sw_init (struct amdgpu_device *adev);int amdgpu_vcn_sw_fini (struct amdgpu_device *adev);int amdgpu_vcn_suspend (struct amdgpu_device *adev);
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
Re: [PATCH 3/5] drm/amdgpu/vcn: Add firmware share memory support
Quoted 66 lines · Leo Liu
Quoted 64 lines · James Zhu
Added firmware share memory support for VCN. Current multiple queue mode is enabled only.
Signed-off-by James Zhu <[email protected]>---
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c| 13 +++++++++++++drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h| 24 ++++++++++++++++++++++++2 files changed, 37 insertions(+)drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +6
-0@@ -182,6 +182,14 @@ int amdgpu_vcn_sw_init(struct amdgpu_device *adev) return r; } } + + r = amdgpu_bo_create_kernel(adev,AMDGPU_GPU_PAGE_ALIGN (sizeof (struct amdgpu_fw_shared )), + PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, &adev->vcn.inst[i].fw_shared_bo, + &adev->vcn.inst[i].fw_shared_gpu_addr, &adev->vcn.inst[i].fw_shared_cpu_addr); + if (r) { + dev_err(adev->dev, "VCN %d (%d) failed to allocatefimware shared bo\n", i, r);
+ return r;
+ }
}
return 0;drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c +5
-0@@ -196,6 +204,11 @@ int amdgpu_vcn_sw_fini(struct amdgpu_device *adev) for (j = 0; j < adev->vcn.num_vcn_inst; ++j) { if (adev->vcn.harvest_config & (1 << j)) continue; + +amdgpu_bo_free_kernel (&adev->vcn.inst[j].fw_shared_bo, + &adev->vcn.inst[j].fw_shared_gpu_addr, + (void **)&adev->vcn.inst[j].fw_shared_cpu_addr); + if (adev->vcn.indirect_sram) { amdgpu_bo_free_kernel(&adev->vcn.inst[j].dpg_sram_bo, &adev->vcn.inst[j].dpg_sram_gpu_addr,drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +3
-0@@ -179,11 +179,14 @@ struct amdgpu_vcn_inst { struct amdgpu_irq_src irq; struct amdgpu_vcn_reg external; struct amdgpu_bo *dpg_sram_bo; + struct amdgpu_bo *fw_shared_bo; struct dpg_pause_state pause_state; void *dpg_sram_cpu_addr; uint64_t dpg_sram_gpu_addr; uint32_t *dpg_sram_curr_addr; atomic_t dpg_enc_submission_cnt; + void *fw_shared_cpu_addr; + uint64_t fw_shared_gpu_addr; }; struct amdgpu_vcn {@@ -209,6 +212,27 @@ struct amdgpu_vcn { int inst_idx, struct dpg_pause_state *new_state); };
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h struct amdgpu_vcn { · line 220
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +5
-0+#define AMDGPU_VCN_MULTI_QUEUE_FLAG (1 << 8) + +enum fw_queue_mode { + fw_queue_ring_reset = 1, + fw_queue_dpg_hold_off = 2, +};Please move the define and enum to the top as others. With that fixed, the series are
Sure. Can you review the patches also?
Thanks!
James
Quoted 15 lines
Reviewed-by Leo Liu <[email protected]>Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +10
-0+ +struct amdgpu_fw_shared_multi_queue { + uint8_t decode_queue_mode; + uint8_t encode_generalpurpose_queue_mode; + uint8_t encode_lowlatency_queue_mode; + uint8_t encode_realtime_queue_mode; + uint8_t padding[4]; +}; + +struct amdgpu_fw_shared {
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h struct amdgpu_vcn { · line 238
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +5
-0+ uint32_t present_flag_0; + uint8_t pad[53]; + struct amdgpu_fw_shared_multi_queue multi_queue; +} __attribute__((__packed__)); + int amdgpu_vcn_sw_init(struct amdgpu_device *adev); int amdgpu_vcn_sw_fini(struct amdgpu_device *adev); int amdgpu_vcn_suspend(struct amdgpu_device *adev);
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 4/5 drm/amdgpu/vcn2.0: Add firmware w/r ptr reset sync
In reply to James ZhuAdd firmware write/read point reset sync through shared memory
--- drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c | 42 ++++++++++++++++++++++++++++++++--- 1 file changed, 39 insertions(+), 3 deletions(-)
drivers/gpu/drm/amd/amdgpu/vcn_v2_0.c +39 -3
@@ -92,6 +92,7 @@ static int vcn_v2_0_sw_init(void *handle)struct amdgpu_ring *ring;int i, r;struct amdgpu_device *adev = (struct amdgpu_device *)handle; +volatile struct amdgpu_fw_shared *fw_shared;/* VCN DEC TRAP */ r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN,
@@ -174,6 +175,8 @@ static int vcn_v2_0_sw_init(void *handle)if (r)return r; + fw_shared = adev->vcn.inst->fw_shared_cpu_addr; + fw_shared->present_flag_0 =cpu_to_le32 (AMDGPU_VCN_MULTI_QUEUE_FLAG );return 0 ; }
@@ -188,6 +191,9 @@ static int vcn_v2_0_sw_fini(void *handle) {int r;struct amdgpu_device *adev = (struct amdgpu_device *)handle; +volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr; + + fw_shared->present_flag_0 =0 ;amdgpu_virt_free_mm_table (adev);
@@ -363,6 +369,15 @@ static void vcn_v2_0_mc_resume(struct amdgpu_device *adev)WREG32_SOC15 (UVD ,0 , mmUVD_VCPU_CACHE_OFFSET2,0 );WREG32_SOC15 (UVD ,0 , mmUVD_VCPU_CACHE_SIZE2,AMDGPU_VCN_CONTEXT_SIZE ); +/* non-cache window */ +WREG32_SOC15 (UVD ,0 , mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW, +lower_32_bits (adev->vcn.inst->fw_shared_gpu_addr)); +WREG32_SOC15 (UVD ,0 , mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH, +upper_32_bits (adev->vcn.inst->fw_shared_gpu_addr)); +WREG32_SOC15 (UVD ,0 , mmUVD_VCPU_NONCACHE_OFFSET0,0 ); +WREG32_SOC15 (UVD ,0 , mmUVD_VCPU_NONCACHE_SIZE0, +AMDGPU_GPU_PAGE_ALIGN (sizeof (struct amdgpu_fw_shared ))); +WREG32_SOC15 (UVD ,0 , mmUVD_GFX10_ADDR_CONFIG, adev->gfx.config.gb_addr_config); }
@@ -446,13 +461,16 @@ static void vcn_v2_0_mc_resume_dpg_mode(struct amdgpu_device *adev, bool indirec/* non-cache window */ WREG32_SOC15_DPG_MODE_2_0 (0 ,SOC15_DPG_MODE_OFFSET_2_0 ( -UVD ,0 , mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),0 ,0 , indirect); +UVD ,0 , mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), +lower_32_bits (adev->vcn.inst->fw_shared_gpu_addr),0 , indirect);WREG32_SOC15_DPG_MODE_2_0 (0 ,SOC15_DPG_MODE_OFFSET_2_0 ( -UVD ,0 , mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),0 ,0 , indirect); +UVD ,0 , mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), +upper_32_bits (adev->vcn.inst->fw_shared_gpu_addr),0 , indirect);WREG32_SOC15_DPG_MODE_2_0 (0 ,SOC15_DPG_MODE_OFFSET_2_0 (UVD ,0 , mmUVD_VCPU_NONCACHE_OFFSET0),0 ,0 , indirect);WREG32_SOC15_DPG_MODE_2_0 (0 ,SOC15_DPG_MODE_OFFSET_2_0 ( -UVD ,0 , mmUVD_VCPU_NONCACHE_SIZE0),0 ,0 , indirect); +UVD ,0 , mmUVD_VCPU_NONCACHE_SIZE0), +AMDGPU_GPU_PAGE_ALIGN (sizeof (struct amdgpu_fw_shared )),0 , indirect);/* VCN global tiling registers */ WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
@@ -777,6 +795,7 @@ static void vcn_v2_0_enable_static_power_gating(struct amdgpu_device *adev)static int vcn_v2_0_start_dpg_mode (struct amdgpu_device *adev,bool indirect) { +volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr;struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;uint32_t rb_bufsz, tmp;
@@ -880,6 +899,8 @@ static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)WREG32_P (SOC15_REG_OFFSET (UVD ,0 , mmUVD_POWER_STATUS),UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK , ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK ); + fw_shared->multi_queue.decode_queue_mode |= fw_queue_ring_reset; +/* set the write pointer delay */ WREG32_SOC15 (UVD ,0 , mmUVD_RBC_RB_WPTR_CNTL,0 );
@@ -902,6 +923,7 @@ static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)WREG32_SOC15 (UVD ,0 , mmUVD_RBC_RB_WPTR,lower_32_bits (ring->wptr)); + fw_shared->multi_queue.decode_queue_mode &= ~fw_queue_ring_reset;/* Unstall DPG */ WREG32_P (SOC15_REG_OFFSET (UVD ,0 , mmUVD_POWER_STATUS),0 , ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK );
@@ -910,6 +932,7 @@ static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)static int vcn_v2_0_start (struct amdgpu_device *adev) { +volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr;struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;uint32_t rb_bufsz, tmp;uint32_t lmi_swap_cntl;
@@ -1044,6 +1067,7 @@ static int vcn_v2_0_start(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,UVD_RBC_RB_CNTL ,RB_RPTR_WR_EN ,1 );WREG32_SOC15 (UVD ,0 , mmUVD_RBC_RB_CNTL, tmp); + fw_shared->multi_queue.decode_queue_mode |= fw_queue_ring_reset;/* programm the RB_BASE for ring buffer */ WREG32_SOC15 (UVD ,0 , mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,lower_32_bits (ring->gpu_addr));
@@ -1056,20 +1080,25 @@ static int vcn_v2_0_start(struct amdgpu_device *adev) ring->wptr =RREG32_SOC15 (UVD ,0 , mmUVD_RBC_RB_RPTR);WREG32_SOC15 (UVD ,0 , mmUVD_RBC_RB_WPTR,lower_32_bits (ring->wptr)); + fw_shared->multi_queue.decode_queue_mode &= ~fw_queue_ring_reset; + fw_shared->multi_queue.encode_generalpurpose_queue_mode |= fw_queue_ring_reset; ring = &adev->vcn.inst->ring_enc[0 ];WREG32_SOC15 (UVD ,0 , mmUVD_RB_RPTR,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD ,0 , mmUVD_RB_WPTR,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD ,0 , mmUVD_RB_BASE_LO, ring->gpu_addr);WREG32_SOC15 (UVD ,0 , mmUVD_RB_BASE_HI,upper_32_bits (ring->gpu_addr));WREG32_SOC15 (UVD ,0 , mmUVD_RB_SIZE, ring->ring_size /4 ); + fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~fw_queue_ring_reset; + fw_shared->multi_queue.encode_lowlatency_queue_mode |= fw_queue_ring_reset; ring = &adev->vcn.inst->ring_enc[1 ];WREG32_SOC15 (UVD ,0 , mmUVD_RB_RPTR2,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD ,0 , mmUVD_RB_WPTR2,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD ,0 , mmUVD_RB_BASE_LO2, ring->gpu_addr);WREG32_SOC15 (UVD ,0 , mmUVD_RB_BASE_HI2,upper_32_bits (ring->gpu_addr));WREG32_SOC15 (UVD ,0 , mmUVD_RB_SIZE2, ring->ring_size /4 ); + fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~fw_queue_ring_reset;return 0 ; }
@@ -1191,6 +1220,7 @@ static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);if (!ret_code) { +volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst->fw_shared_cpu_addr;/* pause DPG */ reg_data |=UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK ;WREG32_SOC15 (UVD ,0 , mmUVD_DPG_PAUSE, reg_data);
@@ -1205,6 +1235,7 @@ static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK , ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);/* Restore */ + fw_shared->multi_queue.encode_generalpurpose_queue_mode |= fw_queue_ring_reset; ring = &adev->vcn.inst->ring_enc[0 ]; ring->wptr =0 ;WREG32_SOC15 (UVD ,0 , mmUVD_RB_BASE_LO, ring->gpu_addr);
@@ -1212,7 +1243,9 @@ static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,WREG32_SOC15 (UVD ,0 , mmUVD_RB_SIZE, ring->ring_size /4 );WREG32_SOC15 (UVD ,0 , mmUVD_RB_RPTR,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD ,0 , mmUVD_RB_WPTR,lower_32_bits (ring->wptr)); + fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~fw_queue_ring_reset; + fw_shared->multi_queue.encode_lowlatency_queue_mode |= fw_queue_ring_reset; ring = &adev->vcn.inst->ring_enc[1 ]; ring->wptr =0 ;WREG32_SOC15 (UVD ,0 , mmUVD_RB_BASE_LO2, ring->gpu_addr);
@@ -1220,9 +1253,12 @@ static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,WREG32_SOC15 (UVD ,0 , mmUVD_RB_SIZE2, ring->ring_size /4 );WREG32_SOC15 (UVD ,0 , mmUVD_RB_RPTR2,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD ,0 , mmUVD_RB_WPTR2,lower_32_bits (ring->wptr)); + fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~fw_queue_ring_reset; + fw_shared->multi_queue.decode_queue_mode |= fw_queue_ring_reset;WREG32_SOC15 (UVD ,0 , mmUVD_RBC_RB_WPTR,RREG32_SOC15 (UVD ,0 , mmUVD_SCRATCH2) &0x7FFFFFFF ); + fw_shared->multi_queue.decode_queue_mode &= ~fw_queue_ring_reset;/* Unstall DPG */ WREG32_P (SOC15_REG_OFFSET (UVD ,0 , mmUVD_POWER_STATUS),0 , ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK );
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx
PATCH 5/5 drm/amdgpu/vcn2.5: Add firmware w/r ptr reset sync
In reply to James ZhuAdd firmware write/read point reset sync through shared memory
--- drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c | 51 ++++++++++++++++++++++++++++++++--- 1 file changed, 47 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/vcn_v2_5.c +47 -4
@@ -165,6 +165,8 @@ static int vcn_v2_5_sw_init(void *handle)return r;for (j =0 ; j < adev->vcn.num_vcn_inst; j++) { +volatile struct amdgpu_fw_shared *fw_shared; +if (adev->vcn.harvest_config & (1 << j))continue ; adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
@@ -207,6 +209,9 @@ static int vcn_v2_5_sw_init(void *handle)if (r)return r; } + + fw_shared = adev->vcn.inst[j].fw_shared_cpu_addr; + fw_shared->present_flag_0 =cpu_to_le32 (AMDGPU_VCN_MULTI_QUEUE_FLAG ); }if (amdgpu_sriov_vf (adev)) {
@@ -230,8 +235,16 @@ static int vcn_v2_5_sw_init(void *handle)*/ static int vcn_v2_5_sw_fini (void *handle) { -int r; +int i, r;struct amdgpu_device *adev = (struct amdgpu_device *)handle; +volatile struct amdgpu_fw_shared *fw_shared; + +for (i =0 ; i < adev->vcn.num_vcn_inst; i++) { +if (adev->vcn.harvest_config & (1 << i)) +continue ; + fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr; + fw_shared->present_flag_0 =0 ; + }if (amdgpu_sriov_vf (adev))amdgpu_virt_free_mm_table (adev);
@@ -424,6 +437,15 @@ static void vcn_v2_5_mc_resume(struct amdgpu_device *adev) upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));WREG32_SOC15 (UVD , i, mmUVD_VCPU_CACHE_OFFSET2,0 );WREG32_SOC15 (UVD , i, mmUVD_VCPU_CACHE_SIZE2,AMDGPU_VCN_CONTEXT_SIZE ); + +/* non-cache window */ +WREG32_SOC15 (UVD , i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW, +lower_32_bits (adev->vcn.inst[i].fw_shared_gpu_addr)); +WREG32_SOC15 (UVD , i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH, +upper_32_bits (adev->vcn.inst[i].fw_shared_gpu_addr)); +WREG32_SOC15 (UVD , i, mmUVD_VCPU_NONCACHE_OFFSET0,0 ); +WREG32_SOC15 (UVD , i, mmUVD_VCPU_NONCACHE_SIZE0, +AMDGPU_GPU_PAGE_ALIGN (sizeof (struct amdgpu_fw_shared ))); } }
@@ -507,13 +529,16 @@ static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx/* non-cache window */ WREG32_SOC15_DPG_MODE_2_0 (inst_idx,SOC15_DPG_MODE_OFFSET_2_0 ( -UVD ,0 , mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),0 ,0 , indirect); +UVD ,0 , mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), +lower_32_bits (adev->vcn.inst[inst_idx].fw_shared_gpu_addr),0 , indirect);WREG32_SOC15_DPG_MODE_2_0 (inst_idx,SOC15_DPG_MODE_OFFSET_2_0 ( -UVD ,0 , mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),0 ,0 , indirect); +UVD ,0 , mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), +upper_32_bits (adev->vcn.inst[inst_idx].fw_shared_gpu_addr),0 , indirect);WREG32_SOC15_DPG_MODE_2_0 (inst_idx,SOC15_DPG_MODE_OFFSET_2_0 (UVD ,0 , mmUVD_VCPU_NONCACHE_OFFSET0),0 ,0 , indirect);WREG32_SOC15_DPG_MODE_2_0 (inst_idx,SOC15_DPG_MODE_OFFSET_2_0 ( -UVD ,0 , mmUVD_VCPU_NONCACHE_SIZE0),0 ,0 , indirect); +UVD ,0 , mmUVD_VCPU_NONCACHE_SIZE0), +AMDGPU_GPU_PAGE_ALIGN (sizeof (struct amdgpu_fw_shared )),0 , indirect);/* VCN global tiling registers */ WREG32_SOC15_DPG_MODE_2_0(inst_idx, SOC15_DPG_MODE_OFFSET_2_0(
@@ -750,6 +775,7 @@ static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)static int vcn_v2_5_start_dpg_mode (struct amdgpu_device *adev,int inst_idx,bool indirect) { +volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr;struct amdgpu_ring *ring;uint32_t rb_bufsz, tmp;
@@ -883,6 +909,7 @@ static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, booWREG32_SOC15 (UVD , inst_idx, mmUVD_RBC_RB_WPTR,lower_32_bits (ring->wptr)); + fw_shared->multi_queue.decode_queue_mode &= ~fw_queue_ring_reset;/* Unstall DPG */ WREG32_P (SOC15_REG_OFFSET (UVD , inst_idx, mmUVD_POWER_STATUS),0 , ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK );
@@ -972,6 +999,7 @@ static int vcn_v2_5_start(struct amdgpu_device *adev)vcn_v2_5_mc_resume (adev);for (i =0 ; i < adev->vcn.num_vcn_inst; ++i) { +volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared_cpu_addr;if (adev->vcn.harvest_config & (1 << i))continue ;/* VCN global tiling registers */
@@ -1045,6 +1073,7 @@ static int vcn_v2_5_start(struct amdgpu_device *adev) tmp =REG_SET_FIELD (tmp,UVD_RBC_RB_CNTL ,RB_RPTR_WR_EN ,1 );WREG32_SOC15 (UVD , i, mmUVD_RBC_RB_CNTL, tmp); + fw_shared->multi_queue.decode_queue_mode |= fw_queue_ring_reset;/* programm the RB_BASE for ring buffer */ WREG32_SOC15 (UVD , i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,lower_32_bits (ring->gpu_addr));
@@ -1057,19 +1086,25 @@ static int vcn_v2_5_start(struct amdgpu_device *adev) ring->wptr =RREG32_SOC15 (UVD , i, mmUVD_RBC_RB_RPTR);WREG32_SOC15 (UVD , i, mmUVD_RBC_RB_WPTR,lower_32_bits (ring->wptr)); + fw_shared->multi_queue.decode_queue_mode &= ~fw_queue_ring_reset; + + fw_shared->multi_queue.encode_generalpurpose_queue_mode |= fw_queue_ring_reset; ring = &adev->vcn.inst[i].ring_enc[0 ];WREG32_SOC15 (UVD , i, mmUVD_RB_RPTR,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD , i, mmUVD_RB_WPTR,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD , i, mmUVD_RB_BASE_LO, ring->gpu_addr);WREG32_SOC15 (UVD , i, mmUVD_RB_BASE_HI,upper_32_bits (ring->gpu_addr));WREG32_SOC15 (UVD , i, mmUVD_RB_SIZE, ring->ring_size /4 ); + fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~fw_queue_ring_reset; + fw_shared->multi_queue.encode_lowlatency_queue_mode |= fw_queue_ring_reset; ring = &adev->vcn.inst[i].ring_enc[1 ];WREG32_SOC15 (UVD , i, mmUVD_RB_RPTR2,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD , i, mmUVD_RB_WPTR2,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD , i, mmUVD_RB_BASE_LO2, ring->gpu_addr);WREG32_SOC15 (UVD , i, mmUVD_RB_BASE_HI2,upper_32_bits (ring->gpu_addr));WREG32_SOC15 (UVD , i, mmUVD_RB_SIZE2, ring->ring_size /4 ); + fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~fw_queue_ring_reset; }return 0 ;
@@ -1390,6 +1425,8 @@ static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);if (!ret_code) { +volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared_cpu_addr; +/* pause DPG */ reg_data |=UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK ;WREG32_SOC15 (UVD , inst_idx, mmUVD_DPG_PAUSE, reg_data);
@@ -1405,6 +1442,7 @@ static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);/* Restore */ + fw_shared->multi_queue.encode_generalpurpose_queue_mode |= fw_queue_ring_reset; ring = &adev->vcn.inst[inst_idx].ring_enc[0 ]; ring->wptr =0 ;WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
@@ -1412,7 +1450,9 @@ static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_SIZE, ring->ring_size /4 );WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_RPTR,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_WPTR,lower_32_bits (ring->wptr)); + fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~fw_queue_ring_reset; + fw_shared->multi_queue.encode_lowlatency_queue_mode |= fw_queue_ring_reset; ring = &adev->vcn.inst[inst_idx].ring_enc[1 ]; ring->wptr =0 ;WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
@@ -1420,9 +1460,12 @@ static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_SIZE2, ring->ring_size /4 );WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_RPTR2,lower_32_bits (ring->wptr));WREG32_SOC15 (UVD , inst_idx, mmUVD_RB_WPTR2,lower_32_bits (ring->wptr)); + fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~fw_queue_ring_reset; + fw_shared->multi_queue.decode_queue_mode |= fw_queue_ring_reset;WREG32_SOC15 (UVD , inst_idx, mmUVD_RBC_RB_WPTR,RREG32_SOC15 (UVD , inst_idx, mmUVD_SCRATCH2) &0x7FFFFFFF ); + fw_shared->multi_queue.decode_queue_mode &= ~fw_queue_ring_reset;/* Unstall DPG */ WREG32_P(SOC15_REG_OFFSET(UVD, inst_idx, mmUVD_POWER_STATUS),
Signature · 6 lines
2.7.4 _______________________________________________ amd-gfx mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/amd-gfx