Patchset to enable soft reset
Subsystems (4)
cover Patchset to enable soft reset
This patch set is to enable soft reset on Sienna Cichlid on sdma and vcn. Sequence is enabled to satisfy the request of a faster and soft solution for multi-container environment. In order to get a more accurate engine hang detection, get hang from timeout job instead of register status. In order to maintain the previous reset logic, add a module parameter to control soft reset.
Victor Zhao (4):
drm/amdgpu: add param soft_reset_enable
drm/amdgpu: pass job to check soft reset
drm/amdgpu: add sdma v5_2 soft reset
drm/amdgpu: add vcn v3_0 soft reset
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 3 +- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 18 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 9 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 2 +- drivers/gpu/drm/amd/amdgpu/dce_v10_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 2 +- drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c | 2 +- drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c | 2 +- drivers/gpu/drm/amd/amdgpu/nv.c | 5 +- drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c | 79 ++++++++++++++++- drivers/gpu/drm/amd/amdgpu/tonga_ih.c | 2 +- drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c | 87 ++++++++++++++++++- .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 2 +- drivers/gpu/drm/amd/include/amd_shared.h | 2 +- 21 files changed, 204 insertions(+), 27 deletions(-)
Signature · 1 line
2.25.1
PATCH 1/4 drm/amdgpu: add param soft_reset_enable
add parameter soft_reset_enable to control the enablement of soft reset
--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 9 +++++++++ drivers/gpu/drm/amd/amdgpu/nv.c | 5 ++++- 3 files changed, 14 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -0
@@ -238,6 +238,7 @@ extern int amdgpu_si_support;extern int amdgpu_cik_support;#endif extern int amdgpu_num_kcq; +extern int amdgpu_soft_reset_enable;#define AMDGPU_VM_MAX_NUM_CTX 4096#define AMDGPU_SG_THRESHOLD (256*1024*1024)
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +9 -0
@@ -201,6 +201,7 @@ struct amdgpu_watchdog_timer amdgpu_watchdog_timer = { .timeout_fatal_disable = false, .period = 0x0, /* default to 0x0 (timeout disable) */ }; +int amdgpu_soft_reset_enable =-1 ;/** * DOC: vramlimit (int)
@@ -931,6 +932,14 @@ MODULE_PARM_DESC(smu_pptable_id, "specify pptable id to be used (-1 = auto(default) value, 0 = use pptable from vbios, > 0 = soft pptable id)");module_param_named (smu_pptable_id, amdgpu_smu_pptable_id, int,0444 ); +/** +* DOC: soft_reset (int) +* Used to enable grbm soft reset +*/ +MODULE_PARM_DESC (soft_reset_enable, +"enable soft reset (-1 = auto(default) value, 0 = disable, > 0 = enable)" ); +module_param_named (soft_reset_enable, amdgpu_soft_reset_enable, int,0444 ); +/* These devices are not supported by amdgpu. * They are supported by the mach64, r128, radeon drivers */
drivers/gpu/drm/amd/amdgpu/nv.c +4 -1
@@ -620,7 +620,10 @@ static uint32_t nv_get_rev_id(struct amdgpu_device *adev)static bool nv_need_full_reset (struct amdgpu_device *adev) { -return true; +if (amdgpu_soft_reset_enable <=0 ) +return true; +else +return false; }static bool nv_need_reset_on_init (struct amdgpu_device *adev)
Signature · 1 line
2.25.1
Christian König
NAKRe: [PATCH 1/4] drm/amdgpu: add param soft_reset_enable
Am 10.03.22 um 07:11 schrieb Victor Zhao:
Quoted 2 lines
add parameter
soft_reset_enableto control the enablement of soft reset
NAK, proper working soft recovery is something mandatory and shouldn't be controlled by the end user.
Marek and I have also decided to completely disable job re-submission on bare metal in the near future.
Regards,
Christian.
Quoted 63 lines
Signed-off-by Victor Zhao <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 9 +++++++++ drivers/gpu/drm/amd/amdgpu/nv.c | 5 ++++- 3 files changed, 14 insertions(+), 1 deletion(-)drivers/gpu/drm/amd/amdgpu/amdgpu.h +1
-0@@ -238,6 +238,7 @@ extern int amdgpu_si_support;extern int amdgpu_cik_support;#endif extern int amdgpu_num_kcq; +extern int amdgpu_soft_reset_enable;#define AMDGPU_VM_MAX_NUM_CTX 4096#define AMDGPU_SG_THRESHOLD (256*1024*1024)drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +9
-0@@ -201,6 +201,7 @@ struct amdgpu_watchdog_timer amdgpu_watchdog_timer = { .timeout_fatal_disable = false, .period = 0x0, /* default to 0x0 (timeout disable) */ }; +int amdgpu_soft_reset_enable =-1 ;/** * DOC: vramlimit (int) @@ -931,6 +932,14 @@ MODULE_PARM_DESC(smu_pptable_id, "specify pptable id to be used (-1 = auto(default) value, 0 = use pptable from vbios, > 0 = soft pptable id)");module_param_named (smu_pptable_id, amdgpu_smu_pptable_id, int,0444 ); +/** +* DOC: soft_reset (int) +* Used to enable grbm soft reset +*/ +MODULE_PARM_DESC (soft_reset_enable, +"enable soft reset (-1 = auto(default) value, 0 = disable, > 0 = enable)" ); +module_param_named (soft_reset_enable, amdgpu_soft_reset_enable, int,0444 ); +/* These devices are not supported by amdgpu. * They are supported by the mach64, r128, radeon drivers */ drivers/gpu/drm/amd/amdgpu/nv.c +4
-1@@ -620,7 +620,10 @@ static uint32_t nv_get_rev_id(struct amdgpu_device *adev)static bool nv_need_full_reset (struct amdgpu_device *adev) { -return true; +if (amdgpu_soft_reset_enable <=0 ) +return true; +else +return false; }static bool nv_need_reset_on_init (struct amdgpu_device *adev)
PATCH 2/4 drm/amdgpu: pass job to check soft reset
In reply to Victor ZhaoIn order to get more accurate engine hang detection, pass the hang job to check_soft_reset to find the hang engine instead of check register status.
--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 18 +++++++++++------- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 2 +- drivers/gpu/drm/amd/amdgpu/dce_v10_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 2 +- drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c | 2 +- drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c | 2 +- drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/tonga_ih.c | 2 +- drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 2 +- .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 2 +- drivers/gpu/drm/amd/include/amd_shared.h | 2 +- 17 files changed, 27 insertions(+), 23 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -1
@@ -1359,7 +1359,7 @@ int emu_soc_asic_init(struct amdgpu_device *adev);/* Common functions */ bool amdgpu_device_has_job_running (struct amdgpu_device *adev); -bool amdgpu_device_should_recover_gpu (struct amdgpu_device *adev); +bool amdgpu_device_should_recover_gpu (struct amdgpu_device *adev,struct amdgpu_job *job);int amdgpu_device_gpu_recover (struct amdgpu_device *adev,struct amdgpu_job * job);void amdgpu_device_pci_config_reset (struct amdgpu_device *adev);
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +1 -1
@@ -238,7 +238,7 @@ void amdgpu_amdkfd_gpu_reset(struct kgd_dev *kgd) {struct amdgpu_device *adev = (struct amdgpu_device *)kgd; -if (amdgpu_device_should_recover_gpu (adev)) +if (amdgpu_device_should_recover_gpu (adev,NULL ))amdgpu_device_gpu_recover (adev,NULL ); }
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +11 -7
@@ -4162,7 +4162,7 @@ int amdgpu_device_resume(struct drm_device *dev, bool fbcon)* if the asic is still hung or not. * Returns true if any of the IPs are still in a hung state, false if not. */ -static bool amdgpu_device_ip_check_soft_reset (struct amdgpu_device *adev) +static bool amdgpu_device_ip_check_soft_reset (struct amdgpu_device *adev,struct amdgpu_job *job) {int i;bool asic_hang = false;
@@ -4178,7 +4178,7 @@ static bool amdgpu_device_ip_check_soft_reset(struct amdgpu_device *adev)continue ;if (adev->ip_blocks[i].version->funcs->check_soft_reset) adev->ip_blocks[i].status.hang = - adev->ip_blocks[i].version->funcs->check_soft_reset (adev); + adev->ip_blocks[i].version->funcs->check_soft_reset (adev, job);if (adev->ip_blocks[i].status.hang) {dev_info (adev->dev,"IP block:%s is hung!\n" , adev->ip_blocks[i].version->funcs->name); asic_hang = true;
@@ -4466,11 +4466,15 @@ bool amdgpu_device_has_job_running(struct amdgpu_device *adev)* Check amdgpu_gpu_recovery and SRIOV status to see if we should try to recover * a hung GPU. */ -bool amdgpu_device_should_recover_gpu (struct amdgpu_device *adev) +bool amdgpu_device_should_recover_gpu (struct amdgpu_device *adev,struct amdgpu_job *job) { -if (!amdgpu_device_ip_check_soft_reset (adev)) { -dev_info (adev->dev,"Timeout, but no hardware hang detected.\n" ); -return false; +if (!amdgpu_device_ip_check_soft_reset (adev, job)) { +if (job) { +dev_info (adev->dev,"No matching soft reset found, do whole gpu recovery\n" ); + }else { +dev_info (adev->dev,"Timeout, but no hardware hang detected.\n" ); +return false; + } }if (amdgpu_gpu_recovery ==0 )
@@ -4616,7 +4620,7 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev,amdgpu_device_ip_pre_soft_reset (adev); r =amdgpu_device_ip_soft_reset (adev);amdgpu_device_ip_post_soft_reset (adev); -if (r ||amdgpu_device_ip_check_soft_reset (adev)) { +if (r ||amdgpu_device_ip_check_soft_reset (adev,NULL )) {dev_info (adev->dev,"soft reset failed, will fallback to full reset!\n" ); need_full_reset = true; }
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +1 -1
@@ -62,7 +62,7 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)DRM_ERROR ("Process information: process %s pid %d thread %s pid %d\n" , ti.process_name, ti.tgid, ti.task_name, ti.pid); -if (amdgpu_device_should_recover_gpu (ring->adev)) { +if (amdgpu_device_should_recover_gpu (ring->adev, job)) {amdgpu_device_gpu_recover (ring->adev, job); }else {drm_sched_suspend_timeout (&ring->sched);
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +1 -1
@@ -1871,7 +1871,7 @@ static void amdgpu_ras_do_recovery(struct work_struct *work)amdgpu_put_xgmi_hive (hive); } -if (amdgpu_device_should_recover_gpu (ras->adev)) +if (amdgpu_device_should_recover_gpu (ras->adev,NULL ))amdgpu_device_gpu_recover (ras->adev,NULL );atomic_set (&ras->in_recovery,0 ); }
drivers/gpu/drm/amd/amdgpu/dce_v10_0.c +1 -1
@@ -3033,7 +3033,7 @@ static int dce_v10_0_wait_for_idle(void *handle)return 0 ; } -static bool dce_v10_0_check_soft_reset (void *handle) +static bool dce_v10_0_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +1 -1
@@ -4974,7 +4974,7 @@ static int gfx_v8_0_resume(void *handle)return gfx_v8_0_hw_init (handle); } -static bool gfx_v8_0_check_soft_reset (void *handle) +static bool gfx_v8_0_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;u32 grbm_soft_reset =0 , srbm_soft_reset =0 ;
drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +1 -1
@@ -1311,7 +1311,7 @@ static int gmc_v8_0_wait_for_idle(void *handle) } -static bool gmc_v8_0_check_soft_reset (void *handle) +static bool gmc_v8_0_check_soft_reset (void *handle,struct amdgpu_job *job) {u32 srbm_soft_reset =0 ;struct amdgpu_device *adev = (struct amdgpu_device *)handle;
drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c +1 -1
@@ -274,7 +274,7 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work)up_write (&adev->reset_sem);/* Trigger recovery for world switch failure if no TDR */ -if (amdgpu_device_should_recover_gpu (adev) +if (amdgpu_device_should_recover_gpu (adev,NULL ) && (!amdgpu_device_has_job_running (adev) || adev->sdma_timeout ==MAX_SCHEDULE_TIMEOUT ))amdgpu_device_gpu_recover (adev,NULL );
drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c +1 -1
@@ -303,7 +303,7 @@ static void xgpu_nv_mailbox_flr_work(struct work_struct *work)up_write (&adev->reset_sem);/* Trigger recovery for world switch failure if no TDR */ - if (amdgpu_device_should_recover_gpu(adev) + if (amdgpu_device_should_recover_gpu(adev, NULL) && (!amdgpu_device_has_job_running(adev) || adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT || adev->gfx_timeout == MAX_SCHEDULE_TIMEOUT ||
drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c +1 -1
@@ -520,7 +520,7 @@ static void xgpu_vi_mailbox_flr_work(struct work_struct *work) }/* Trigger recovery due to world switch failure */ -if (amdgpu_device_should_recover_gpu (adev)) +if (amdgpu_device_should_recover_gpu (adev,NULL ))amdgpu_device_gpu_recover (adev,NULL ); }
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +1 -1
@@ -1250,7 +1250,7 @@ static int sdma_v3_0_wait_for_idle(void *handle)return -ETIMEDOUT ; } -static bool sdma_v3_0_check_soft_reset (void *handle) +static bool sdma_v3_0_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;u32 srbm_soft_reset =0 ;
drivers/gpu/drm/amd/amdgpu/tonga_ih.c +1 -1
@@ -380,7 +380,7 @@ static int tonga_ih_wait_for_idle(void *handle)return -ETIMEDOUT ; } -static bool tonga_ih_check_soft_reset (void *handle) +static bool tonga_ih_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;u32 srbm_soft_reset =0 ;
drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c +1 -1
@@ -1159,7 +1159,7 @@ static int uvd_v6_0_wait_for_idle(void *handle) }#define AMDGPU_UVD_STATUS_BUSY_MASK 0xfd -static bool uvd_v6_0_check_soft_reset (void *handle) +static bool uvd_v6_0_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;u32 srbm_soft_reset =0 ;
drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +1 -1
@@ -629,7 +629,7 @@ static int vce_v3_0_wait_for_idle(void *handle)#define AMDGPU_VCE_STATUS_BUSY_MASK (VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK | \ VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK) -static bool vce_v3_0_check_soft_reset (void *handle) +static bool vce_v3_0_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;u32 srbm_soft_reset =0 ;
drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +1 -1
@@ -317,7 +317,7 @@ static int dm_wait_for_idle(void *handle)return 0 ; } -static bool dm_check_soft_reset (void *handle) +static bool dm_check_soft_reset (void *handle,struct amdgpu_job *job) {return false; }
drivers/gpu/drm/amd/include/amd_shared.h +1 -1
@@ -287,7 +287,7 @@ struct amd_ip_funcs {int (*resume)(void *handle);bool (*is_idle)(void *handle);int (*wait_for_idle)(void *handle); -bool (*check_soft_reset)(void *handle); +bool (*check_soft_reset)(void *handle,struct amdgpu_job *job);int (*pre_soft_reset)(void *handle);int (*soft_reset)(void *handle);int (*post_soft_reset)(void *handle);
Signature · 1 line
2.25.1
Christian König
NAKRe: [PATCH 2/4] drm/amdgpu: pass job to check soft reset
Am 10.03.22 um 07:11 schrieb Victor Zhao:
Quoted 3 lines
In order to get more accurate engine hang detection, pass the hang job to
check_soft_resetto find the hang engine instead of check register status.
NAK, the amdgpu_ring_soft_recovery() function already gets the hardware fence as parameter.
That should be sufficient to identify the hanging engine.
Regards,
Christian.
Quoted 281 lines
Signed-off-by Victor Zhao <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 18 +++++++++++------- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 2 +- drivers/gpu/drm/amd/amdgpu/dce_v10_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 2 +- drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c | 2 +- drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c | 2 +- drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/tonga_ih.c | 2 +- drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/vce_v3_0.c | 2 +- .../gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c | 2 +- drivers/gpu/drm/amd/include/amd_shared.h | 2 +- 17 files changed, 27 insertions(+), 23 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu.h +1
-1@@ -1359,7 +1359,7 @@ int emu_soc_asic_init(struct amdgpu_device *adev);/* Common functions */ bool amdgpu_device_has_job_running (struct amdgpu_device *adev); -bool amdgpu_device_should_recover_gpu (struct amdgpu_device *adev); +bool amdgpu_device_should_recover_gpu (struct amdgpu_device *adev,struct amdgpu_job *job);int amdgpu_device_gpu_recover (struct amdgpu_device *adev,struct amdgpu_job * job);void amdgpu_device_pci_config_reset (struct amdgpu_device *adev);drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +1
-1@@ -238,7 +238,7 @@ void amdgpu_amdkfd_gpu_reset(struct kgd_dev *kgd) {struct amdgpu_device *adev = (struct amdgpu_device *)kgd; -if (amdgpu_device_should_recover_gpu (adev)) +if (amdgpu_device_should_recover_gpu (adev,NULL ))amdgpu_device_gpu_recover (adev,NULL ); }drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +11
-7@@ -4162,7 +4162,7 @@ int amdgpu_device_resume(struct drm_device *dev, bool fbcon)* if the asic is still hung or not. * Returns true if any of the IPs are still in a hung state, false if not. */ -static bool amdgpu_device_ip_check_soft_reset (struct amdgpu_device *adev) +static bool amdgpu_device_ip_check_soft_reset (struct amdgpu_device *adev,struct amdgpu_job *job) {int i;bool asic_hang = false;@@ -4178,7 +4178,7 @@ static bool amdgpu_device_ip_check_soft_reset(struct amdgpu_device *adev)continue ;if (adev->ip_blocks[i].version->funcs->check_soft_reset) adev->ip_blocks[i].status.hang = - adev->ip_blocks[i].version->funcs->check_soft_reset (adev); + adev->ip_blocks[i].version->funcs->check_soft_reset (adev, job);if (adev->ip_blocks[i].status.hang) {dev_info (adev->dev,"IP block:%s is hung!\n" , adev->ip_blocks[i].version->funcs->name); asic_hang = true;@@ -4466,11 +4466,15 @@ bool amdgpu_device_has_job_running(struct amdgpu_device *adev)* Check amdgpu_gpu_recovery and SRIOV status to see if we should try to recover * a hung GPU. */ -bool amdgpu_device_should_recover_gpu (struct amdgpu_device *adev) +bool amdgpu_device_should_recover_gpu (struct amdgpu_device *adev,struct amdgpu_job *job) { -if (!amdgpu_device_ip_check_soft_reset (adev)) { -dev_info (adev->dev,"Timeout, but no hardware hang detected.\n" ); -return false; +if (!amdgpu_device_ip_check_soft_reset (adev, job)) { +if (job) { +dev_info (adev->dev,"No matching soft reset found, do whole gpu recovery\n" ); + }else { +dev_info (adev->dev,"Timeout, but no hardware hang detected.\n" ); +return false; + } }if (amdgpu_gpu_recovery ==0 )@@ -4616,7 +4620,7 @@ int amdgpu_device_pre_asic_reset(struct amdgpu_device *adev,amdgpu_device_ip_pre_soft_reset (adev); r =amdgpu_device_ip_soft_reset (adev);amdgpu_device_ip_post_soft_reset (adev); -if (r ||amdgpu_device_ip_check_soft_reset (adev)) { +if (r ||amdgpu_device_ip_check_soft_reset (adev,NULL )) {dev_info (adev->dev,"soft reset failed, will fallback to full reset!\n" ); need_full_reset = true; }drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +1
-1@@ -62,7 +62,7 @@ static enum drm_gpu_sched_stat amdgpu_job_timedout(struct drm_sched_job *s_job)DRM_ERROR ("Process information: process %s pid %d thread %s pid %d\n" , ti.process_name, ti.tgid, ti.task_name, ti.pid); -if (amdgpu_device_should_recover_gpu (ring->adev)) { +if (amdgpu_device_should_recover_gpu (ring->adev, job)) {amdgpu_device_gpu_recover (ring->adev, job); }else {drm_sched_suspend_timeout (&ring->sched);drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +1
-1@@ -1871,7 +1871,7 @@ static void amdgpu_ras_do_recovery(struct work_struct *work)amdgpu_put_xgmi_hive (hive); } -if (amdgpu_device_should_recover_gpu (ras->adev)) +if (amdgpu_device_should_recover_gpu (ras->adev,NULL ))amdgpu_device_gpu_recover (ras->adev,NULL );atomic_set (&ras->in_recovery,0 ); }drivers/gpu/drm/amd/amdgpu/dce_v10_0.c +1
-1@@ -3033,7 +3033,7 @@ static int dce_v10_0_wait_for_idle(void *handle)return 0 ; } -static bool dce_v10_0_check_soft_reset (void *handle) +static bool dce_v10_0_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +1
-1@@ -4974,7 +4974,7 @@ static int gfx_v8_0_resume(void *handle)return gfx_v8_0_hw_init (handle); } -static bool gfx_v8_0_check_soft_reset (void *handle) +static bool gfx_v8_0_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;u32 grbm_soft_reset =0 , srbm_soft_reset =0 ;drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +1
-1@@ -1311,7 +1311,7 @@ static int gmc_v8_0_wait_for_idle(void *handle) } -static bool gmc_v8_0_check_soft_reset (void *handle) +static bool gmc_v8_0_check_soft_reset (void *handle,struct amdgpu_job *job) {u32 srbm_soft_reset =0 ;struct amdgpu_device *adev = (struct amdgpu_device *)handle;drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c +1
-1@@ -274,7 +274,7 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work)up_write (&adev->reset_sem);/* Trigger recovery for world switch failure if no TDR */ -if (amdgpu_device_should_recover_gpu (adev) +if (amdgpu_device_should_recover_gpu (adev,NULL ) && (!amdgpu_device_has_job_running (adev) || adev->sdma_timeout ==MAX_SCHEDULE_TIMEOUT ))amdgpu_device_gpu_recover (adev,NULL );drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c +1
-1@@ -303,7 +303,7 @@ static void xgpu_nv_mailbox_flr_work(struct work_struct *work)up_write (&adev->reset_sem);/* Trigger recovery for world switch failure if no TDR */ - if (amdgpu_device_should_recover_gpu(adev) + if (amdgpu_device_should_recover_gpu(adev, NULL) && (!amdgpu_device_has_job_running(adev) || adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT || adev->gfx_timeout == MAX_SCHEDULE_TIMEOUT ||drivers/gpu/drm/amd/amdgpu/mxgpu_vi.c +1
-1@@ -520,7 +520,7 @@ static void xgpu_vi_mailbox_flr_work(struct work_struct *work) }/* Trigger recovery due to world switch failure */ -if (amdgpu_device_should_recover_gpu (adev)) +if (amdgpu_device_should_recover_gpu (adev,NULL ))amdgpu_device_gpu_recover (adev,NULL ); }drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +1
-1@@ -1250,7 +1250,7 @@ static int sdma_v3_0_wait_for_idle(void *handle)return -ETIMEDOUT ; } -static bool sdma_v3_0_check_soft_reset (void *handle) +static bool sdma_v3_0_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;u32 srbm_soft_reset =0 ;drivers/gpu/drm/amd/amdgpu/tonga_ih.c +1
-1@@ -380,7 +380,7 @@ static int tonga_ih_wait_for_idle(void *handle)return -ETIMEDOUT ; } -static bool tonga_ih_check_soft_reset (void *handle) +static bool tonga_ih_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;u32 srbm_soft_reset =0 ;drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c +1
-1@@ -1159,7 +1159,7 @@ static int uvd_v6_0_wait_for_idle(void *handle) }#define AMDGPU_UVD_STATUS_BUSY_MASK 0xfd -static bool uvd_v6_0_check_soft_reset (void *handle) +static bool uvd_v6_0_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;u32 srbm_soft_reset =0 ;drivers/gpu/drm/amd/amdgpu/vce_v3_0.c +1
-1@@ -629,7 +629,7 @@ static int vce_v3_0_wait_for_idle(void *handle)#define AMDGPU_VCE_STATUS_BUSY_MASK (VCE_STATUS_VCPU_REPORT_AUTO_BUSY_MASK | \ VCE_STATUS_VCPU_REPORT_RB0_BUSY_MASK) -static bool vce_v3_0_check_soft_reset (void *handle) +static bool vce_v3_0_check_soft_reset (void *handle,struct amdgpu_job *job) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;u32 srbm_soft_reset =0 ;drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm.c +1
-1@@ -317,7 +317,7 @@ static int dm_wait_for_idle(void *handle)return 0 ; } -static bool dm_check_soft_reset (void *handle) +static bool dm_check_soft_reset (void *handle,struct amdgpu_job *job) {return false; }drivers/gpu/drm/amd/include/amd_shared.h +1
-1@@ -287,7 +287,7 @@ struct amd_ip_funcs {int (*resume)(void *handle);bool (*is_idle)(void *handle);int (*wait_for_idle)(void *handle); -bool (*check_soft_reset)(void *handle); +bool (*check_soft_reset)(void *handle,struct amdgpu_job *job);int (*pre_soft_reset)(void *handle);int (*soft_reset)(void *handle);int (*post_soft_reset)(void *handle);
PATCH 3/4 drm/amdgpu: add sdma v5_2 soft reset
In reply to Victor Zhaoenable sdma v5_2 soft reset
--- drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c | 79 +++++++++++++++++++++++++- 1 file changed, 78 insertions(+), 1 deletion(-)
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +78 -1
@@ -806,6 +806,80 @@ static int sdma_v5_2_load_microcode(struct amdgpu_device *adev)return 0 ; } +static bool sdma_v5_2_check_soft_reset (void *handle,struct amdgpu_job *job) +{ +struct amdgpu_device *adev = (struct amdgpu_device *)handle; + +if (job) { +if (!strncmp (job->base.sched->name,"sdma" ,4 )) +return true; +else +return false; + }else { +u32 tmp =RREG32_SOC15 (GC ,0 , mmGRBM_STATUS2); + +if (tmp ==0xffffffff ) +return false; + +if (tmp &GRBM_STATUS2__SDMA_BUSY_MASK ) +return true; +else +return false; + } +} + +static int sdma_v5_2_pre_soft_reset (void *handle) +{ +struct amdgpu_device *adev = (struct amdgpu_device *)handle; +int i, j; +uint32_t tmp, f32_cntl; + +for (i =0 ; i < adev->sdma.num_instances; i++) +WREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_FREEZE),0x10 ); + +for (i =0 ; i < adev->sdma.num_instances; i++) { +for (j =0 ; j <10 ; j++) { + tmp =RREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_FREEZE)); +DRM_DEBUG ("SDMA%d_FREEZE=0x%x" , i, tmp); +if (tmp &SDMA0_FREEZE__FROZEN_MASK ) { +WREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_FREEZE),0 ); +break ; + }else { +udelay (10 ); + } + } +if (j ==10 ) { +DRM_ERROR ("SDMA%d_FREEZE frozen not set" , i); +return -1 ; + } + } +for (i =0 ; i < adev->sdma.num_instances; i++) { + f32_cntl =RREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_F32_CNTL)); + f32_cntl =REG_SET_FIELD (f32_cntl,SDMA0_F32_CNTL ,HALT ,1 ); +WREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_F32_CNTL), f32_cntl); + } + +return 0 ; +} + +static int sdma_v5_2_post_soft_reset (void *handle) +{ +int i; +uint32_t f32_cntl; +struct amdgpu_device *adev = (struct amdgpu_device *)handle; + +for (i =0 ; i < adev->sdma.num_instances; i++) { + f32_cntl =RREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_F32_CNTL)); + f32_cntl =REG_SET_FIELD (f32_cntl,SDMA0_F32_CNTL ,HALT ,0 ); +WREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_F32_CNTL), f32_cntl); + } + +sdma_v5_2_gfx_resume (adev); +udelay (10 ); + +return 0 ; +} +static int sdma_v5_2_soft_reset (void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -831,7 +905,7 @@ static int sdma_v5_2_soft_reset(void *handle)WREG32_SOC15 (GC ,0 , mmGRBM_SOFT_RESET, tmp); tmp =RREG32_SOC15 (GC ,0 , mmGRBM_SOFT_RESET); -udelay (50 ); +mdelay (2 ); }return 0 ;
@@ -1655,6 +1729,9 @@ const struct amd_ip_funcs sdma_v5_2_ip_funcs = { .resume = sdma_v5_2_resume, .is_idle = sdma_v5_2_is_idle, .wait_for_idle = sdma_v5_2_wait_for_idle, + .pre_soft_reset = sdma_v5_2_pre_soft_reset, + .post_soft_reset = sdma_v5_2_post_soft_reset, + .check_soft_reset = sdma_v5_2_check_soft_reset, .soft_reset = sdma_v5_2_soft_reset, .set_clockgating_state = sdma_v5_2_set_clockgating_state, .set_powergating_state = sdma_v5_2_set_powergating_state,
Signature · 1 line
2.25.1
Re: [PATCH 3/4] drm/amdgpu: add sdma v5_2 soft reset
Am 10.03.22 um 07:11 schrieb Victor Zhao:
Quoted 12 lines
enable sdma
v5_2soft resetSigned-off-by Victor Zhao <[email protected]>--- drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c | 79 +++++++++++++++++++++++++- 1 file changed, 78 insertions(+), 1 deletion(-)drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +0
-0
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c static int sdma_v5_2_load_microcode(struct amdgpu_device *adev) · line 814
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +6
-0@@ -806,6 +806,80 @@ static int sdma_v5_2_load_microcode(struct amdgpu_device *adev)return 0 ; } +static bool sdma_v5_2_check_soft_reset (void *handle,struct amdgpu_job *job) +{ +struct amdgpu_device *adev = (struct amdgpu_device *)handle; + +if (job) { +if (!strncmp (job->base.sched->name,"sdma" ,4 ))
That's a really big NAK, please no string comparison in the kernel if it isn't absolutely necessary.
Regards,
Christian.
Quoted 90 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +72
-1+return true; + else + return false; + }else { +u32 tmp =RREG32_SOC15 (GC ,0 , mmGRBM_STATUS2); + +if (tmp ==0xffffffff ) +return false; + +if (tmp &GRBM_STATUS2__SDMA_BUSY_MASK ) +return true; +else +return false; + } +} + +static int sdma_v5_2_pre_soft_reset (void *handle) +{ +struct amdgpu_device *adev = (struct amdgpu_device *)handle; +int i, j; +uint32_t tmp, f32_cntl; + +for (i =0 ; i < adev->sdma.num_instances; i++) +WREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_FREEZE),0x10 ); + +for (i =0 ; i < adev->sdma.num_instances; i++) { +for (j =0 ; j <10 ; j++) { + tmp =RREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_FREEZE)); +DRM_DEBUG ("SDMA%d_FREEZE=0x%x" , i, tmp); +if (tmp &SDMA0_FREEZE__FROZEN_MASK ) { +WREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_FREEZE),0 ); +break ; + }else { +udelay (10 ); + } + } +if (j ==10 ) { +DRM_ERROR ("SDMA%d_FREEZE frozen not set" , i); +return -1 ; + } + } +for (i =0 ; i < adev->sdma.num_instances; i++) { + f32_cntl =RREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_F32_CNTL)); + f32_cntl =REG_SET_FIELD (f32_cntl,SDMA0_F32_CNTL ,HALT ,1 ); +WREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_F32_CNTL), f32_cntl); + } + +return 0 ; +} + +static int sdma_v5_2_post_soft_reset (void *handle) +{ +int i; +uint32_t f32_cntl; +struct amdgpu_device *adev = (struct amdgpu_device *)handle; + +for (i =0 ; i < adev->sdma.num_instances; i++) { + f32_cntl =RREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_F32_CNTL)); + f32_cntl =REG_SET_FIELD (f32_cntl,SDMA0_F32_CNTL ,HALT ,0 ); +WREG32 (sdma_v5_2_get_reg_offset (adev, i, mmSDMA0_F32_CNTL), f32_cntl); + } + +sdma_v5_2_gfx_resume (adev); +udelay (10 ); + +return 0 ; +} +static int sdma_v5_2_soft_reset (void *handle) {struct amdgpu_device *adev = (struct amdgpu_device *)handle;@@ -831,7 +905,7 @@ static int sdma_v5_2_soft_reset(void *handle)WREG32_SOC15 (GC ,0 , mmGRBM_SOFT_RESET, tmp); tmp =RREG32_SOC15 (GC ,0 , mmGRBM_SOFT_RESET); -udelay (50 ); +mdelay (2 ); }return 0 ;@@ -1655,6 +1729,9 @@ const struct amd_ip_funcs sdma_v5_2_ip_funcs = { .resume = sdma_v5_2_resume, .is_idle = sdma_v5_2_is_idle, .wait_for_idle = sdma_v5_2_wait_for_idle, + .pre_soft_reset = sdma_v5_2_pre_soft_reset, + .post_soft_reset = sdma_v5_2_post_soft_reset, + .check_soft_reset = sdma_v5_2_check_soft_reset, .soft_reset = sdma_v5_2_soft_reset, .set_clockgating_state = sdma_v5_2_set_clockgating_state, .set_powergating_state = sdma_v5_2_set_powergating_state,
Re: [PATCH 3/4] drm/amdgpu: add sdma v5_2 soft reset
Quoted 25 lines · Christian König
Am 10.03.22 um 07:11 schrieb Victor Zhao:
Quoted 12 lines
enable sdma
v5_2soft resetSigned-off-by Victor Zhao <[email protected]>---
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c| 79 +++++++++++++++++++++++++-1 file changed, 78 insertions(+), 1 deletion(-)drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +0
-0drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c static int sdma_v5_2_load_microcode(struct · line 814
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +5
-0@@ -806,6 +806,80 @@ static int sdma_v5_2_load_microcode(struct amdgpu_device *adev) return 0; } +static bool sdma_v5_2_check_soft_reset(void *handle, struct amdgpu_job *job) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (job) { + if (!strncmp(job->base.sched->name, "sdma", 4))That's a really big NAK, please no string comparison in the kernel if it isn't absolutely necessary.
Quoted 2 lines
Regards,
Christian.
And even if you would rely on this then at least you have to define a MACRO for scheduler name and use it everywhere because tomorrow someone will change from 'sdma' to something else in scheduler initialization and this code will break instantly.
Andrey
Quoted 97 lines
Quoted 96 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +31
-0+ return true; + else + return false; + } else { + u32 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2); + + if (tmp == 0xffffffff) + return false; + + if (tmp & GRBM_STATUS2__SDMA_BUSY_MASK) + return true; + else + return false; + } +} + +static int sdma_v5_2_pre_soft_reset (void *handle) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int i, j; + uint32_t tmp, f32_cntl; + + for (i = 0; i < adev->sdma.num_instances; i++) + WREG32(sdma_v5_2_get_reg_offset(adev, i, mmSDMA0_FREEZE),0x10 ); + + for (i = 0; i < adev->sdma.num_instances; i++) { + for (j = 0; j < 10; j++) { + tmp = RREG32(sdma_v5_2_get_reg_offset(adev, i, mmSDMA0_FREEZE)); + DRM_DEBUG("SDMA%d_FREEZE=0x%x", i, tmp); + if (tmp & SDMA0_FREEZE__FROZEN_MASK) { + WREG32(sdma_v5_2_get_reg_offset(adev, i,
mmSDMA0_FREEZE), 0);
+ break;
+ } else {
+ udelay(10);
+ }
+ }
+ if (j == 10) {
+DRM_ERROR("SDMA%d_FREEZE frozen not set", i); + return -1;
+ }
+ }
+ for (i = 0; i <adev->sdma.num_instances; i++) {
+f32_cntl=RREG32(sdma_v5_2_get_reg_offset(adev, i, mmSDMA0_F32_CNTL));
+f32_cntl=REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 1); +WREG32(sdma_v5_2_get_reg_offset(adev, i, mmSDMA0_F32_CNTL), f32_cntl);
+ }drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +12
-0+ + return 0; +} + +static int sdma_v5_2_post_soft_reset (void *handle) +{ + int i; + uint32_t f32_cntl; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + for (i = 0; i < adev->sdma.num_instances; i++) { + f32_cntl = RREG32(sdma_v5_2_get_reg_offset(adev, i,
mmSDMA0_F32_CNTL));
+f32_cntl=REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 0); +WREG32(sdma_v5_2_get_reg_offset(adev, i, mmSDMA0_F32_CNTL), f32_cntl);
+ }drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +7
-0+ + sdma_v5_2_gfx_resume(adev); + udelay(10); + + return 0; +} +static int
sdma_v5_2_soft_reset(void *handle)
{
structamdgpu_device*adev = (structamdgpu_device*)handle;drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +4
-0@@ -831,7 +905,7 @@ static int sdma_v5_2_soft_reset(void *handle) WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp); tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET); - udelay(50); + mdelay(2); } return 0;@@ -1655,6 +1729,9 @@ const struct amd_ip_funcs sdma_v5_2_ip_funcs = { .resume = sdma_v5_2_resume, .is_idle = sdma_v5_2_is_idle, .wait_for_idle = sdma_v5_2_wait_for_idle, + .pre_soft_reset = sdma_v5_2_pre_soft_reset, + .post_soft_reset = sdma_v5_2_post_soft_reset, + .check_soft_reset = sdma_v5_2_check_soft_reset, .soft_reset = sdma_v5_2_soft_reset, .set_clockgating_state = sdma_v5_2_set_clockgating_state, .set_powergating_state = sdma_v5_2_set_powergating_state,
Re: [PATCH 3/4] drm/amdgpu: add sdma v5_2 soft reset
Quoted 27 lines · Andrey Grodzovsky
Quoted 25 lines · Christian König
Am 10.03.22 um 07:11 schrieb Victor Zhao:
Quoted 12 lines
enable sdma
v5_2soft resetSigned-off-by Victor Zhao <[email protected]>---
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c| 79 +++++++++++++++++++++++++-1 file changed, 78 insertions(+), 1 deletion(-)drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +0
-0drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c static int sdma_v5_2_load_microcode(struct · line 814
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +5
-0@@ -806,6 +806,80 @@ static int sdma_v5_2_load_microcode(struct amdgpu_device *adev) return 0; } +static bool sdma_v5_2_check_soft_reset(void *handle, struct amdgpu_job *job) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + if (job) { + if (!strncmp(job->base.sched->name, "sdma", 4))That's a really big NAK, please no string comparison in the kernel if it isn't absolutely necessary.
Quoted 8 lines
Quoted 2 lines
Regards,
Christian.And even if you would rely on this then at least you have to define a MACRO for scheduler name and use it everywhere because tomorrow someone will change from 'sdma' to something else in scheduler initialization and this code will break instantly.
Andrey
Maybe, check if any of the sdma rings is associated to the job scheduler?
Thanks,
Lijo
Quoted 99 lines
Quoted 99 lines
Quoted 98 lines
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +31
-0+ return true; + else + return false; + } else { + u32 tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2); + + if (tmp == 0xffffffff) + return false; + + if (tmp & GRBM_STATUS2__SDMA_BUSY_MASK) + return true; + else + return false; + } +} + +static int sdma_v5_2_pre_soft_reset (void *handle) +{ + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + int i, j; + uint32_t tmp, f32_cntl; + + for (i = 0; i < adev->sdma.num_instances; i++) + WREG32(sdma_v5_2_get_reg_offset(adev, i, mmSDMA0_FREEZE),0x10 ); + + for (i = 0; i < adev->sdma.num_instances; i++) { + for (j = 0; j < 10; j++) { + tmp = RREG32(sdma_v5_2_get_reg_offset(adev, i, mmSDMA0_FREEZE)); + DRM_DEBUG("SDMA%d_FREEZE=0x%x", i, tmp); + if (tmp & SDMA0_FREEZE__FROZEN_MASK) { + WREG32(sdma_v5_2_get_reg_offset(adev, i,
mmSDMA0_FREEZE), 0);
+ break;
+ } else {
+ udelay(10);
+ }
+ }
+ if (j == 10) {
+DRM_ERROR("SDMA%d_FREEZE frozen not set", i); + return -1;
+ }
+ }
+ for (i = 0; i <adev->sdma.num_instances; i++) {
+f32_cntl=RREG32(sdma_v5_2_get_reg_offset(adev, i, mmSDMA0_F32_CNTL));
+f32_cntl=REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 1); +WREG32(sdma_v5_2_get_reg_offset(adev, i, mmSDMA0_F32_CNTL), f32_cntl);
+ }drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +12
-0+ + return 0; +} + +static int sdma_v5_2_post_soft_reset (void *handle) +{ + int i; + uint32_t f32_cntl; + struct amdgpu_device *adev = (struct amdgpu_device *)handle; + + for (i = 0; i < adev->sdma.num_instances; i++) { + f32_cntl = RREG32(sdma_v5_2_get_reg_offset(adev, i,
mmSDMA0_F32_CNTL));
+f32_cntl=REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 0); +WREG32(sdma_v5_2_get_reg_offset(adev, i, mmSDMA0_F32_CNTL), f32_cntl);
+ }drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +7
-0+ + sdma_v5_2_gfx_resume(adev); + udelay(10); + + return 0; +} +static int
sdma_v5_2_soft_reset(void *handle)
{
structamdgpu_device*adev = (structamdgpu_device*)handle;drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +4
-0@@ -831,7 +905,7 @@ static int sdma_v5_2_soft_reset(void *handle) WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp); tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET); - udelay(50); + mdelay(2); } return 0;@@ -1655,6 +1729,9 @@ const struct amd_ip_funcs sdma_v5_2_ip_funcs = { .resume = sdma_v5_2_resume, .is_idle = sdma_v5_2_is_idle, .wait_for_idle = sdma_v5_2_wait_for_idle, + .pre_soft_reset = sdma_v5_2_pre_soft_reset, + .post_soft_reset = sdma_v5_2_post_soft_reset, + .check_soft_reset = sdma_v5_2_check_soft_reset, .soft_reset = sdma_v5_2_soft_reset, .set_clockgating_state = sdma_v5_2_set_clockgating_state, .set_powergating_state = sdma_v5_2_set_powergating_state,
PATCH 4/4 drm/amdgpu: add vcn v3_0 soft reset
In reply to Victor Zhaoadd soft reset sequence for vcn v3_0
--- drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c | 87 ++++++++++++++++++++++++++- 1 file changed, 85 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/vcn_v3_0.c +85 -2
@@ -2247,6 +2247,89 @@ static void vcn_v3_0_set_irq_funcs(struct amdgpu_device *adev) } } +static bool vcn_v3_0_check_soft_reset (void *handle,struct amdgpu_job *job) +{ +struct amdgpu_device *adev = (struct amdgpu_device *)handle; +int i; + +if (job) { +if (!strncmp (job->base.sched->name,"vcn" ,3 )) +return true; +else +return false; + }else { +for (i =0 ; i < adev->vcn.num_vcn_inst; ++i) { +uint32_t tmp =RREG32_SOC15 (VCN , i, mmUVD_STATUS); + +if (tmp ==0xDEADBEEF ) +continue ; +else if (tmp &UVD_STATUS__BUSY ) +return true; + } +return false; + } +} + +static int vcn_v3_0_soft_reset (void *handle) +{ +struct amdgpu_device *adev = (struct amdgpu_device *)handle; +uint32_t tmp; +int i =0 ; + +for (i =0 ; i < adev->vcn.num_vcn_inst; ++i) { +/* disable LMI UMC channel */ + tmp =RREG32_SOC15 (VCN , i, mmUVD_LMI_CTRL2); + tmp |=UVD_LMI_CTRL2__STALL_ARB_UMC_MASK ; +WREG32_SOC15 (VCN , i, mmUVD_LMI_CTRL2, tmp); + +/* block VCPU register access */ + tmp =RREG32_SOC15 (VCN , i, mmUVD_RB_ARB_CTRL); + tmp |=UVD_RB_ARB_CTRL__VCPU_DIS_MASK ; +WREG32_SOC15 (VCN , i, mmUVD_RB_ARB_CTRL, tmp); + +/* reset VCPU */ + tmp =RREG32_SOC15 (VCN , i, mmUVD_VCPU_CNTL); + tmp |=UVD_VCPU_CNTL__BLK_RST_MASK ; +WREG32_SOC15 (VCN , i, mmUVD_VCPU_CNTL, tmp); + +/* disable VCPU clock */ + tmp =RREG32_SOC15 (VCN , i, mmUVD_VCPU_CNTL); + tmp &= ~(UVD_VCPU_CNTL__CLK_EN_MASK ); +WREG32_SOC15 (VCN , i, mmUVD_VCPU_CNTL, tmp); + +/* apply soft reset */ + tmp =RREG32_SOC15 (VCN , i, mmUVD_SOFT_RESET); + tmp |=UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK ; +WREG32_SOC15 (VCN , i, mmUVD_SOFT_RESET, tmp); + tmp =RREG32_SOC15 (VCN , i, mmUVD_SOFT_RESET); + tmp |=UVD_SOFT_RESET__LMI_SOFT_RESET_MASK ; +WREG32_SOC15 (VCN , i, mmUVD_SOFT_RESET, tmp); + + tmp =RREG32_SOC15 (VCN , i, mmUVD_VCPU_CNTL); + tmp |=UVD_VCPU_CNTL__BLK_RST_MASK ; +WREG32_SOC15 (VCN , i, mmUVD_VCPU_CNTL, tmp); + +/* clear status */ +WREG32_SOC15 (VCN , i, mmUVD_STATUS,0 ); + +/* apply HW clock gating */ +vcn_v3_0_enable_clock_gating (adev, i); + +/* enable VCN power gating */ +vcn_v3_0_enable_static_power_gating (adev, i); + } + +for (i =0 ; i < adev->vcn.num_vcn_inst; ++i) { +/* disable VCN power gating */ +vcn_v3_0_disable_static_power_gating (adev, i); + +/*SW clock gating */ +vcn_v3_0_disable_clock_gating (adev, i); + } + +return 0 ; +} +static const struct amd_ip_funcs vcn_v3_0_ip_funcs = { .name ="vcn_v3_0" , .early_init = vcn_v3_0_early_init,
@@ -2259,9 +2342,9 @@ static const struct amd_ip_funcs vcn_v3_0_ip_funcs = { .resume = vcn_v3_0_resume, .is_idle = vcn_v3_0_is_idle, .wait_for_idle = vcn_v3_0_wait_for_idle, - .check_soft_reset = NULL, + .check_soft_reset = vcn_v3_0_check_soft_reset, .pre_soft_reset = NULL, - .soft_reset = NULL, + .soft_reset = vcn_v3_0_soft_reset, .post_soft_reset = NULL, .set_clockgating_state = vcn_v3_0_set_clockgating_state, .set_powergating_state = vcn_v3_0_set_powergating_state,
Signature · 1 line
2.25.1