drm/amdkfd: Enable GFX11 usermode queue oversubscription
Subsystems (3)
PATCH v4 2/3 drm/amdkfd: Enable GFX11 usermode queue oversubscription
Starting with GFX11, MES requires wptr BOs to be GTT allocated/mapped to GART for usermode queues in order to support oversubscription. In the case that work is submitted to an unmapped queue, MES must have a GART wptr address to determine whether the queue should be mapped.
This change is accompanied with changes in MES and is applicable for MES_API_VERSION >= 2.
v3:
- Use amdgpu_vm_bo_lookup_mapping for wptr_bo mapping lookup
- Move wptr_bo refcount increment to amdgpu_amdkfd_map_gtt_bo_to_gart
- Remove list_del_init from amdgpu_amdkfd_map_gtt_bo_to_gart
- Cleanup/fix create_queue wptr_bo error handling
v4:
- Add MES version shift/mask defines to amdgpu_mes.h
- Change version check from MES_VERSION to MES_API_VERSION
- Add check in kfd_ioctl_create_queue before wptr bo pin/GART map to ensure bo is a single page.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 2 + .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 48 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h | 7 +++ drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 45 ++++++++++++++++- .../drm/amd/amdkfd/kfd_device_queue_manager.c | 9 +++- .../gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c | 2 + drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 ++ .../amd/amdkfd/kfd_process_queue_manager.c | 20 +++++--- 8 files changed, 127 insertions(+), 9 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +2 -0
@@ -286,6 +286,8 @@ int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct kgd_mem *mem, void **kptr, uint64_t *size);void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel (struct kgd_mem *mem); +int amdgpu_amdkfd_map_gtt_bo_to_gart (struct amdgpu_device *adev,struct amdgpu_bo *bo); +int amdgpu_amdkfd_gpuvm_restore_process_bos (void *process_info,struct dma_fence **ef); int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev,
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +48 -0
54 lines not on this page · raw message
drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h +7 -0
@@ -32,6 +32,13 @@#define AMDGPU_MES_MAX_GFX_PIPES 2#define AMDGPU_MES_MAX_SDMA_PIPES 2 +#define AMDGPU_MES_API_VERSION_SHIFT 12 +#define AMDGPU_MES_FEAT_VERSION_SHIFT 24 + +#define AMDGPU_MES_VERSION_MASK 0x00000fff +#define AMDGPU_MES_API_VERSION_MASK 0x00fff000 +#define AMDGPU_MES_FEAT_VERSION_MASK 0xff000000 +enum amdgpu_mes_priority_level {AMDGPU_MES_PRIORITY_LEVEL_LOW =0 ,AMDGPU_MES_PRIORITY_LEVEL_NORMAL =1 ,
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +43 -2
67 lines not on this page · raw message
drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +8 -1
@@ -177,6 +177,7 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q,struct kfd_process_device *pdd =qpd_to_pdd (qpd);struct mes_add_queue_input queue_input;int r, queue_type; +uint64_t wptr_addr_off;if (dqm->is_hws_hang)return -EIO ;
@@ -196,7 +197,13 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q,AMDGPU_MES_PRIORITY_LEVEL_NORMAL ; queue_input.doorbell_offset = q->properties.doorbell_off; queue_input.mqd_addr = q->gart_mqd_addr; - queue_input.wptr_addr = (uint64_t )q->properties.write_ptr; + +if (q->wptr_bo) { + wptr_addr_off = (uint64_t )q->properties.write_ptr - (uint64_t )q->wptr_bo->kfd_bo->va; + queue_input.wptr_addr = ((uint64_t )q->wptr_bo->tbo.resource->start <<PAGE_SHIFT ) + wptr_addr_off; + }else + queue_input.wptr_addr = (uint64_t )q->properties.write_ptr; + queue_input.paging = false; queue_input.tba_addr = qpd->tba_addr; queue_input.tma_addr = qpd->tma_addr;
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c +2 -0
@@ -377,6 +377,8 @@ static void update_mqd_sdma(struct mqd_manager *mm, void *mqd, m->sdmax_rlcx_rb_base_hi =upper_32_bits (q->queue_address >>8 ); m->sdmax_rlcx_rb_rptr_addr_lo =lower_32_bits ((uint64_t )q->read_ptr); m->sdmax_rlcx_rb_rptr_addr_hi =upper_32_bits ((uint64_t )q->read_ptr); + m->sdmax_rlcx_rb_wptr_poll_addr_lo =lower_32_bits ((uint64_t )q->write_ptr); + m->sdmax_rlcx_rb_wptr_poll_addr_hi =upper_32_bits ((uint64_t )q->write_ptr); m->sdmax_rlcx_doorbell_offset = q->doorbell_off <<SDMA0_QUEUE0_DOORBELL_OFFSET__OFFSET__SHIFT ;
drivers/gpu/drm/amd/amdkfd/kfd_priv.h +3 -0
@@ -570,6 +570,8 @@ struct queue {void *gang_ctx_bo;uint64_t gang_ctx_gpu_addr;void *gang_ctx_cpu_ptr; + +struct amdgpu_bo *wptr_bo; };enum KFD_MQD_TYPE {
@@ -1205,6 +1207,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,struct file *f, struct queue_properties *properties, unsigned int *qid, + struct amdgpu_bo *wptr_bo, const struct kfd_criu_queue_priv_data *q_data, const void *restore_mqd, const void *restore_ctl_stack,
drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +14 -6
@@ -180,7 +180,8 @@ void pqm_uninit(struct process_queue_manager *pqm)static int init_user_queue (struct process_queue_manager *pqm,struct kfd_dev *dev,struct queue **q,struct queue_properties *q_properties, -struct file *f,unsigned int qid) +struct file *f,struct amdgpu_bo *wptr_bo, +unsigned int qid) {int retval;
@@ -210,6 +211,7 @@ static int init_user_queue(struct process_queue_manager *pqm, goto cleanup; }memset ((*q)->gang_ctx_cpu_ptr,0 ,AMDGPU_MES_GANG_CTX_SIZE ); + (*q)->wptr_bo = wptr_bo; }pr_debug ("PQM After init queue" );
@@ -226,6 +228,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,struct file *f, struct queue_properties *properties, unsigned int *qid, + struct amdgpu_bo *wptr_bo, const struct kfd_criu_queue_priv_data *q_data, const void *restore_mqd, const void *restore_ctl_stack,
@@ -288,7 +291,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,* allocate_sdma_queue() in create_queue() has the * corresponding check logic. */ - retval =init_user_queue (pqm, dev, &q, properties, f, *qid); + retval =init_user_queue (pqm, dev, &q, properties, f, wptr_bo, *qid);if (retval !=0 ) goto err_create_queue; pqn->q = q;
@@ -309,7 +312,7 @@ int pqm_create_queue(struct process_queue_manager *pqm, goto err_create_queue; } - retval =init_user_queue (pqm, dev, &q, properties, f, *qid); + retval =init_user_queue (pqm, dev, &q, properties, f, wptr_bo, *qid);if (retval !=0 ) goto err_create_queue; pqn->q = q;
@@ -435,10 +438,15 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid) pdd->qpd.num_gws =0 ; } -if (dev->shared_resources.enable_mes) +if (dev->shared_resources.enable_mes) {amdgpu_amdkfd_free_gtt_mem (dev->adev, pqn->q->gang_ctx_bo); -kfd_procfs_del_queue (pqn->q); +if (pqn->q->wptr_bo) +amdgpu_amdkfd_free_gtt_mem (dev->adev, pqn->q->wptr_bo); + +kfd_procfs_del_queue (pqn->q); + + }uninit_queue (pqn->q); }
@@ -844,7 +852,7 @@ int kfd_criu_restore_queue(struct kfd_process *p,print_queue_properties (&qp); - ret =pqm_create_queue (&p->pqm, pdd->dev,NULL , &qp, &queue_id, q_data, mqd, ctl_stack, + ret =pqm_create_queue (&p->pqm, pdd->dev,NULL , &qp, &queue_id,NULL , q_data, mqd, ctl_stack,NULL );if (ret) {pr_err ("Failed to create new queue err:%d\n" , ret);
Signature · 1 line
2.25.1
On 2022-06-22 11:36, Graham Sider wrote:
Starting with GFX11, MES requires wptr BOs to be GTT allocated/mapped to GART for usermode queues in order to support oversubscription. In the case that work is submitted to an unmapped queue, MES must have a GART wptr address to determine whether the queue should be mapped. This change is accompanied with changes in MES and is applicable for MES_API_VERSION >= 2. v3: - Use amdgpu_vm_bo_lookup_mapping for wptr_bo mapping lookup - Move wptr_bo refcount increment to amdgpu_amdkfd_map_gtt_bo_to_gart - Remove list_del_init from amdgpu_amdkfd_map_gtt_bo_to_gart - Cleanup/fix create_queue wptr_bo error handling v4: - Add MES version shift/mask defines to amdgpu_mes.h - Change version check from MES_VERSION to MES_API_VERSION - Add check in kfd_ioctl_create_queue before wptr bo pin/GART map to ensure bo is a single page. Signed-off-by: Graham Sider <[email protected]> Acked-by: Alex Deucher <[email protected]> Reviewed-by: Philip Yang <[email protected]> --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 2 + .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 48 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h | 7 +++ drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 45 ++++++++++++++++- .../drm/amd/amdkfd/kfd_device_queue_manager.c | 9 +++- .../gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c | 2 + drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 ++ .../amd/amdkfd/kfd_process_queue_manager.c | 20 +++++--- 8 files changed, 127 insertions(+), 9 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h index 648c031942e9..b25b41f50213 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h @@ -286,6 +286,8 @@ int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct kgd_mem *mem, void **kptr, uint64_t *size); void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(struct kgd_mem *mem); +int amdgpu_amdkfd_map_gtt_bo_to_gart(struct amdgpu_device *adev, struct amdgpu_bo *bo); + int amdgpu_amdkfd_gpuvm_restore_process_bos(void *process_info, struct dma_fence **ef); int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index afd6e6923189..615ac2895d62 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -2148,6 +2148,54 @@ int amdgpu_amdkfd_gpuvm_sync_memory( return ret; } +/** + * amdgpu_amdkfd_map_gtt_bo_to_gart - Map BO to GART and increment reference count + * @adev: Device to which allocated BO belongs + * @bo: Buffer object to be mapped + * + * Before return, bo reference count is incremented. To release the reference and unpin/ + * unmap the BO, call amdgpu_amdkfd_free_gtt_mem. + */ +int amdgpu_amdkfd_map_gtt_bo_to_gart(struct amdgpu_device *adev, struct amdgpu_bo *bo) +{ + int ret; + + ret = amdgpu_bo_reserve(bo, true); + if (ret) { + pr_err("Failed to reserve bo. ret %d\n", ret); + goto err_reserve_bo_failed; + } + + ret = amdgpu_bo_pin(bo, AMDGPU_GEM_DOMAIN_GTT); + if (ret) { + pr_err("Failed to pin bo. ret %d\n", ret); + goto err_pin_bo_failed; + } + + ret = amdgpu_ttm_alloc_gart(&bo->tbo); + if (ret) { + pr_err("Failed to bind bo to GART. ret %d\n", ret); + goto err_map_bo_gart_failed; + } + + amdgpu_amdkfd_remove_eviction_fence( + bo, bo->kfd_bo->process_info->eviction_fence); + + amdgpu_bo_unreserve(bo); + + bo = amdgpu_bo_ref(bo); + + return 0; + +err_map_bo_gart_failed: + amdgpu_bo_unpin(bo); +err_pin_bo_failed: + amdgpu_bo_unreserve(bo); +err_reserve_bo_failed: + + return ret; +} + /** amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel() - Map a GTT BO for kernel CPU access * * @mem: Buffer object to be mapped for CPU access diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h index be4b51a5b5c7..137a2cc2e807 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h @@ -32,6 +32,13 @@ #define AMDGPU_MES_MAX_GFX_PIPES 2 #define AMDGPU_MES_MAX_SDMA_PIPES 2 +#define AMDGPU_MES_API_VERSION_SHIFT 12 +#define AMDGPU_MES_FEAT_VERSION_SHIFT 24 + +#define AMDGPU_MES_VERSION_MASK 0x00000fff +#define AMDGPU_MES_API_VERSION_MASK 0x00fff000 +#define AMDGPU_MES_FEAT_VERSION_MASK 0xff000000 + enum amdgpu_mes_priority_level { AMDGPU_MES_PRIORITY_LEVEL_LOW = 0, AMDGPU_MES_PRIORITY_LEVEL_NORMAL = 1, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index 625e837f0119..a0246b4bae6b 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -299,6 +299,7 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, struct kfd_process_device *pdd; struct queue_properties q_properties; uint32_t doorbell_offset_in_process = 0; + struct amdgpu_bo *wptr_bo = NULL; memset(&q_properties, 0, sizeof(struct queue_properties)); @@ -326,12 +327,49 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, goto err_bind_process; } + /* Starting with GFX11, wptr BOs must be mapped to GART for MES to determine work + * on unmapped queues for usermode queue oversubscription (no aggregated doorbell) + */ + if (dev->shared_resources.enable_mes && + ((dev->adev->mes.sched_version & AMDGPU_MES_API_VERSION_MASK) + >> AMDGPU_MES_API_VERSION_SHIFT) >= 2) { + struct amdgpu_bo_va_mapping *wptr_mapping; + struct amdgpu_vm *wptr_vm; + + wptr_vm = drm_priv_to_vm(pdd->drm_priv); + err = amdgpu_bo_reserve(wptr_vm->root.bo, false); + if (err) + goto err_wptr_map_gart; + + wptr_mapping = amdgpu_vm_bo_lookup_mapping( + wptr_vm, args->write_pointer_address >> PAGE_SHIFT); + amdgpu_bo_unreserve(wptr_vm->root.bo); + if (!wptr_mapping) { + pr_err("Failed to lookup wptr bo\n"); + err = -EINVAL; + goto err_wptr_map_gart; + } + + wptr_bo = wptr_mapping->bo_va->base.bo; + if (wptr_bo->tbo.base.size > PAGE_SIZE) { + pr_err("Requested GART mapping for wptr bo larger than one page\n"); + err = -EINVAL; + goto err_wptr_map_gart; + } + + err = amdgpu_amdkfd_map_gtt_bo_to_gart(dev->adev, wptr_bo); + if (err) { + pr_err("Failed to map wptr bo to GART\n"); + goto err_wptr_map_gart; + } + } + pr_debug("Creating queue for PASID 0x%x on gpu 0x%x\n", p->pasid, dev->id); - err = pqm_create_queue(&p->pqm, dev, filep, &q_properties, &queue_id, NULL, NULL, NULL, - &doorbell_offset_in_process); + err = pqm_create_queue(&p->pqm, dev, filep, &q_properties, &queue_id, wptr_bo, + NULL, NULL, NULL, &doorbell_offset_in_process); if (err != 0) goto err_create_queue; @@ -363,6 +401,9 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, return 0; err_create_queue: + if (wptr_bo) + amdgpu_amdkfd_free_gtt_mem(dev->adev, wptr_bo); +err_wptr_map_gart: err_bind_process: err_pdd: mutex_unlock(&p->mutex); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c index 213246a5b4e4..299927a4959b 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -177,6 +177,7 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q, struct kfd_process_device *pdd = qpd_to_pdd(qpd); struct mes_add_queue_input queue_input; int r, queue_type; + uint64_t wptr_addr_off; if (dqm->is_hws_hang) return -EIO; @@ -196,7 +197,13 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q, AMDGPU_MES_PRIORITY_LEVEL_NORMAL; queue_input.doorbell_offset = q->properties.doorbell_off; queue_input.mqd_addr = q->gart_mqd_addr; - queue_input.wptr_addr = (uint64_t)q->properties.write_ptr; + + if (q->wptr_bo) { + wptr_addr_off = (uint64_t)q->properties.write_ptr - (uint64_t)q->wptr_bo->kfd_bo->va; + queue_input.wptr_addr = ((uint64_t)q->wptr_bo->tbo.resource->start << PAGE_SHIFT) + wptr_addr_off; + } else + queue_input.wptr_addr = (uint64_t)q->properties.write_ptr; + queue_input.paging = false; queue_input.tba_addr = qpd->tba_addr; queue_input.tma_addr = qpd->tma_addr; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c index 4e0387f591be..b8e14c2cc295 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c @@ -377,6 +377,8 @@ static void update_mqd_sdma(struct mqd_manager *mm, void *mqd, m->sdmax_rlcx_rb_base_hi = upper_32_bits(q->queue_address >> 8); m->sdmax_rlcx_rb_rptr_addr_lo = lower_32_bits((uint64_t)q->read_ptr); m->sdmax_rlcx_rb_rptr_addr_hi = upper_32_bits((uint64_t)q->read_ptr); + m->sdmax_rlcx_rb_wptr_poll_addr_lo = lower_32_bits((uint64_t)q->write_ptr); + m->sdmax_rlcx_rb_wptr_poll_addr_hi = upper_32_bits((uint64_t)q->write_ptr); m->sdmax_rlcx_doorbell_offset = q->doorbell_off << SDMA0_QUEUE0_DOORBELL_OFFSET__OFFSET__SHIFT; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index 91e5fa56f0a2..59ba50ce54d3 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -570,6 +570,8 @@ struct queue { void *gang_ctx_bo; uint64_t gang_ctx_gpu_addr; void *gang_ctx_cpu_ptr; + + struct amdgpu_bo *wptr_bo; }; enum KFD_MQD_TYPE { @@ -1205,6 +1207,7 @@ int pqm_create_queue(struct process_queue_manager *pqm, struct file *f, struct queue_properties *properties, unsigned int *qid, + struct amdgpu_bo *wptr_bo, const struct kfd_criu_queue_priv_data *q_data, const void *restore_mqd, const void *restore_ctl_stack, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c index 99f2a6412201..8db58348de98 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c @@ -180,7 +180,8 @@ void pqm_uninit(struct process_queue_manager *pqm) static int init_user_queue(struct process_queue_manager *pqm, struct kfd_dev *dev, struct queue **q, struct queue_properties *q_properties, - struct file *f, unsigned int qid) + struct file *f, struct amdgpu_bo *wptr_bo, + unsigned int qid) { int retval; @@ -210,6 +211,7 @@ static int init_user_queue(struct process_queue_manager *pqm, goto cleanup; } memset((*q)->gang_ctx_cpu_ptr, 0, AMDGPU_MES_GANG_CTX_SIZE); + (*q)->wptr_bo = wptr_bo; } pr_debug("PQM After init queue"); @@ -226,6 +228,7 @@ int pqm_create_queue(struct process_queue_manager *pqm, struct file *f, struct queue_properties *properties, unsigned int *qid, + struct amdgpu_bo *wptr_bo, const struct kfd_criu_queue_priv_data *q_data, const void *restore_mqd, const void *restore_ctl_stack, @@ -288,7 +291,7 @@ int pqm_create_queue(struct process_queue_manager *pqm, * allocate_sdma_queue() in create_queue() has the * corresponding check logic. */ - retval = init_user_queue(pqm, dev, &q, properties, f, *qid); + retval = init_user_queue(pqm, dev, &q, properties, f, wptr_bo, *qid); if (retval != 0) goto err_create_queue; pqn->q = q; @@ -309,7 +312,7 @@ int pqm_create_queue(struct process_queue_manager *pqm, goto err_create_queue; } - retval = init_user_queue(pqm, dev, &q, properties, f, *qid); + retval = init_user_queue(pqm, dev, &q, properties, f, wptr_bo, *qid); if (retval != 0) goto err_create_queue; pqn->q = q; @@ -435,10 +438,15 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid) pdd->qpd.num_gws = 0; } - if (dev->shared_resources.enable_mes) + if (dev->shared_resources.enable_mes) { amdgpu_amdkfd_free_gtt_mem(dev->adev, pqn->q->gang_ctx_bo); - kfd_procfs_del_queue(pqn->q); + if (pqn->q->wptr_bo) + amdgpu_amdkfd_free_gtt_mem(dev->adev, pqn->q->wptr_bo); + + kfd_procfs_del_queue(pqn->q); Seems rebase issue, kfd_procfs_del_queue(pqn->q) should be outside if (dev->shared_resources.enable_mes) { + + } uninit_queue(pqn->q); } @@ -844,7 +852,7 @@ int kfd_criu_restore_queue(struct kfd_process *p, print_queue_properties(&qp); - ret = pqm_create_queue(&p->pqm, pdd->dev, NULL, &qp, &queue_id, q_data, mqd, ctl_stack, + ret = pqm_create_queue(&p->pqm, pdd->dev, NULL, &qp, &queue_id, NULL, q_data, mqd, ctl_stack, CRIU restore user queues should create struct queue wptr_bo GART mapping based on MES API version.
Regards,
Philip
NULL); if (ret) { pr_err("Failed to create new queue err:%d\n", ret);
[AMD Official Use Only - General]
Quoted 311 lines
Quoted 311 lines · Graham Sider
Starting with GFX11, MES requires wptr BOs to be GTT allocated/mapped to GART for usermode queues in order to support oversubscription. In the case that work is submitted to an unmapped queue, MES must have a GART wptr address to determine whether the queue should be mapped.
This change is accompanied with changes in MES and is applicable for
MES_API_VERSION>= 2.v3:
- Useamdgpu_vm_bo_lookup_mappingforwptr_bomapping lookup
- Movewptr_borefcount increment toamdgpu_amdkfd_map_gtt_bo_to_gart
- Removelist_del_initfromamdgpu_amdkfd_map_gtt_bo_to_gart
- Cleanup/fixcreate_queuewptr_boerror handling
v4:
- Add MES version shift/mask defines toamdgpu_mes.h
- Change version check fromMES_VERSIONtoMES_API_VERSION
- Add check inkfd_ioctl_create_queuebefore wptr bo pin/GART map to ensure bo is a single page.Signed-off-by Graham Sider [email protected]<mailto:[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 2 + .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 48 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h | 7 +++ drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 45 ++++++++++++++++- .../drm/amd/amdkfd/kfd_device_queue_manager.c | 9 +++- .../gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c | 2 + drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 ++ .../amd/amdkfd/kfd_process_queue_manager.c | 20 +++++--- 8 files changed, 127 insertions(+), 9 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +2
-0@@ -286,6 +286,8 @@ int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct kgd_mem *mem, void **kptr, uint64_t *size);void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel (struct kgd_mem *mem); +int amdgpu_amdkfd_map_gtt_bo_to_gart (struct amdgpu_device *adev,struct amdgpu_bo *bo); +int amdgpu_amdkfd_gpuvm_restore_process_bos (void *process_info,struct dma_fence **ef); int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev,drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +48
-0@@ -2148,6 +2148,54 @@ int amdgpu_amdkfd_gpuvm_sync_memory(return ret; } +/** +* amdgpu_amdkfd_map_gtt_bo_to_gart - Map BO to GART and increment reference count +* @adev: Device to which allocated BO belongs +* @bo: Buffer object to be mapped +* +* Before return, bo reference count is incremented. To release the reference and unpin/ +* unmap the BO, call amdgpu_amdkfd_free_gtt_mem. +*/ +int amdgpu_amdkfd_map_gtt_bo_to_gart (struct amdgpu_device *adev,struct amdgpu_bo *bo) +{ +int ret; + + ret =amdgpu_bo_reserve (bo, true); +if (ret) { +pr_err ("Failed to reserve bo. ret %d\n" , ret); + goto err_reserve_bo_failed; + } + + ret =amdgpu_bo_pin (bo,AMDGPU_GEM_DOMAIN_GTT ); +if (ret) { +pr_err ("Failed to pin bo. ret %d\n" , ret); + goto err_pin_bo_failed; + } + + ret =amdgpu_ttm_alloc_gart (&bo->tbo); +if (ret) { +pr_err ("Failed to bind bo to GART. ret %d\n" , ret); + goto err_map_bo_gart_failed; + } + +amdgpu_amdkfd_remove_eviction_fence ( + bo, bo->kfd_bo->process_info->eviction_fence); + +amdgpu_bo_unreserve (bo); + + bo =amdgpu_bo_ref (bo); + +return 0 ; + +err_map_bo_gart_failed: +amdgpu_bo_unpin (bo); +err_pin_bo_failed: +amdgpu_bo_unreserve (bo); +err_reserve_bo_failed: + +return ret; +} +/** amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel() - Map a GTT BO for kernel CPU access * * @mem: Buffer object to be mapped for CPU access drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h +7
-0@@ -32,6 +32,13 @@#define AMDGPU_MES_MAX_GFX_PIPES 2#define AMDGPU_MES_MAX_SDMA_PIPES 2 +#define AMDGPU_MES_API_VERSION_SHIFT 12 +#define AMDGPU_MES_FEAT_VERSION_SHIFT 24 + +#define AMDGPU_MES_VERSION_MASK 0x00000fff +#define AMDGPU_MES_API_VERSION_MASK 0x00fff000 +#define AMDGPU_MES_FEAT_VERSION_MASK 0xff000000 +enum amdgpu_mes_priority_level {AMDGPU_MES_PRIORITY_LEVEL_LOW =0 ,AMDGPU_MES_PRIORITY_LEVEL_NORMAL =1 ,drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +43
-2@@ -299,6 +299,7 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,struct kfd_process_device *pdd;struct queue_properties q_properties;uint32_t doorbell_offset_in_process =0 ; +struct amdgpu_bo *wptr_bo =NULL ;memset (&q_properties,0 ,sizeof (struct queue_properties ));@@ -326,12 +327,49 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, goto err_bind_process; } +/* Starting with GFX11, wptr BOs must be mapped to GART for MES to determine work +* on unmapped queues for usermode queue oversubscription (no aggregated doorbell) +*/ +if (dev->shared_resources.enable_mes && + ((dev->adev->mes.sched_version &AMDGPU_MES_API_VERSION_MASK ) + >>AMDGPU_MES_API_VERSION_SHIFT ) >=2 ) { +struct amdgpu_bo_va_mapping *wptr_mapping; +struct amdgpu_vm *wptr_vm; + + wptr_vm =drm_priv_to_vm (pdd->drm_priv); + err =amdgpu_bo_reserve (wptr_vm->root.bo, false); +if (err) + goto err_wptr_map_gart; + + wptr_mapping =amdgpu_vm_bo_lookup_mapping ( + wptr_vm, args->write_pointer_address >>PAGE_SHIFT ); +amdgpu_bo_unreserve (wptr_vm->root.bo); +if (!wptr_mapping) { +pr_err ("Failed to lookup wptr bo\n" ); + err = -EINVAL ; + goto err_wptr_map_gart; + } + + wptr_bo = wptr_mapping->bo_va->base.bo; +if (wptr_bo->tbo.base.size >PAGE_SIZE ) { +pr_err ("Requested GART mapping for wptr bo larger than one page\n" ); + err = -EINVAL ; + goto err_wptr_map_gart; + } + + err =amdgpu_amdkfd_map_gtt_bo_to_gart (dev->adev, wptr_bo); +if (err) { +pr_err ("Failed to map wptr bo to GART\n" ); + goto err_wptr_map_gart; + } + } +pr_debug ("Creating queue for PASID 0x%x on gpu 0x%x\n" , p->pasid, dev->id); - err =pqm_create_queue (&p->pqm, dev, filep, &q_properties, &queue_id,NULL ,NULL ,NULL , - &doorbell_offset_in_process); + err =pqm_create_queue (&p->pqm, dev, filep, &q_properties, &queue_id, wptr_bo, +NULL ,NULL ,NULL , &doorbell_offset_in_process);if (err !=0 ) goto err_create_queue;@@ -363,6 +401,9 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,return 0 ; err_create_queue: +if (wptr_bo) +amdgpu_amdkfd_free_gtt_mem (dev->adev, wptr_bo); +err_wptr_map_gart: err_bind_process: err_pdd:mutex_unlock (&p->mutex);drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +8
-1@@ -177,6 +177,7 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q,struct kfd_process_device *pdd =qpd_to_pdd (qpd);struct mes_add_queue_input queue_input;int r, queue_type; +uint64_t wptr_addr_off;if (dqm->is_hws_hang)return -EIO ;@@ -196,7 +197,13 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q,AMDGPU_MES_PRIORITY_LEVEL_NORMAL ; queue_input.doorbell_offset = q->properties.doorbell_off; queue_input.mqd_addr = q->gart_mqd_addr; - queue_input.wptr_addr = (uint64_t )q->properties.write_ptr; + +if (q->wptr_bo) { + wptr_addr_off = (uint64_t )q->properties.write_ptr - (uint64_t )q->wptr_bo->kfd_bo->va; + queue_input.wptr_addr = ((uint64_t )q->wptr_bo->tbo.resource->start <<PAGE_SHIFT ) + wptr_addr_off; + }else + queue_input.wptr_addr = (uint64_t )q->properties.write_ptr; + queue_input.paging = false; queue_input.tba_addr = qpd->tba_addr; queue_input.tma_addr = qpd->tma_addr;drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c +2
-0@@ -377,6 +377,8 @@ static void update_mqd_sdma(struct mqd_manager *mm, void *mqd, m->sdmax_rlcx_rb_base_hi =upper_32_bits (q->queue_address >>8 ); m->sdmax_rlcx_rb_rptr_addr_lo =lower_32_bits ((uint64_t )q->read_ptr); m->sdmax_rlcx_rb_rptr_addr_hi =upper_32_bits ((uint64_t )q->read_ptr); + m->sdmax_rlcx_rb_wptr_poll_addr_lo =lower_32_bits ((uint64_t )q->write_ptr); + m->sdmax_rlcx_rb_wptr_poll_addr_hi =upper_32_bits ((uint64_t )q->write_ptr); m->sdmax_rlcx_doorbell_offset = q->doorbell_off <<SDMA0_QUEUE0_DOORBELL_OFFSET__OFFSET__SHIFT ;drivers/gpu/drm/amd/amdkfd/kfd_priv.h +3
-0@@ -570,6 +570,8 @@ struct queue {void *gang_ctx_bo;uint64_t gang_ctx_gpu_addr;void *gang_ctx_cpu_ptr; + +struct amdgpu_bo *wptr_bo; };enum KFD_MQD_TYPE {@@ -1205,6 +1207,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,struct file *f, struct queue_properties *properties, unsigned int *qid, + struct amdgpu_bo *wptr_bo, const struct kfd_criu_queue_priv_data *q_data, const void *restore_mqd, const void *restore_ctl_stack,drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +6
-3@@ -180,7 +180,8 @@ void pqm_uninit(struct process_queue_manager *pqm)static int init_user_queue (struct process_queue_manager *pqm,struct kfd_dev *dev,struct queue **q,struct queue_properties *q_properties, -struct file *f,unsigned int qid) +struct file *f,struct amdgpu_bo *wptr_bo, +unsigned int qid) {int retval;@@ -210,6 +211,7 @@ static int init_user_queue(struct process_queue_manager *pqm, goto cleanup; }memset ((*q)->gang_ctx_cpu_ptr,0 ,AMDGPU_MES_GANG_CTX_SIZE ); + (*q)->wptr_bo = wptr_bo; }pr_debug ("PQM After init queue" );@@ -226,6 +228,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,struct file *f, struct queue_properties *properties, unsigned int *qid, + struct amdgpu_bo *wptr_bo, const struct kfd_criu_queue_priv_data *q_data, const void *restore_mqd, const void *restore_ctl_stack,@@ -288,7 +291,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,* allocate_sdma_queue() in create_queue() has the * corresponding check logic. */ - retval =init_user_queue (pqm, dev, &q, properties, f, *qid); + retval =init_user_queue (pqm, dev, &q, properties, f, wptr_bo, *qid);if (retval !=0 ) goto err_create_queue; pqn->q = q;@@ -309,7 +312,7 @@ int pqm_create_queue(struct process_queue_manager *pqm, goto err_create_queue; } - retval =init_user_queue (pqm, dev, &q, properties, f, *qid); + retval =init_user_queue (pqm, dev, &q, properties, f, wptr_bo, *qid);if (retval !=0 ) goto err_create_queue; pqn->q = q;@@ -435,10 +438,15 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid) pdd->qpd.num_gws =0 ; }
Quoted 12 lines
drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid) · line 447
Quoted 9 lines
drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +5
-2-if (dev->shared_resources.enable_mes) +if (dev->shared_resources.enable_mes) {amdgpu_amdkfd_free_gtt_mem (dev->adev, pqn->q->gang_ctx_bo); -kfd_procfs_del_queue (pqn->q); +if (pqn->q->wptr_bo) +amdgpu_amdkfd_free_gtt_mem (dev->adev, pqn->q->wptr_bo); + +kfd_procfs_del_queue (pqn->q);Seems rebase issue,
kfd_procfs_del_queue(pqn->q)should be outside if (dev->shared_resources.enable_mes) {
Yeah this was a rebase issue. Fixed - thanks!
Quoted 7 lines
Quoted 7 lines
drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +2
-0+ + }uninit_queue (pqn->q); }@@ -844,7 +852,7 @@ int kfd_criu_restore_queue(struct kfd_process *p,
Quoted 9 lines
drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c int kfd_criu_restore_queue(struct kfd_process *p, · line 855
Quoted 4 lines
drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +1
-1print_queue_properties (&qp); - ret = pqm_create_queue(&p->pqm, pdd->dev, NULL, &qp, &queue_id, q_data, mqd, ctl_stack, + ret = pqm_create_queue(&p->pqm, pdd->dev, NULL, &qp, &queue_id, NULL, q_data, mqd, ctl_stack,CRIU restore user queues should create struct queue
wptr_boGART mapping based on MES API version. Regards,
Philip
As discussed offline since this is somewhat separated this will be modified in a future patch.
Best,
Graham
Quoted 4 lines
Quoted 4 lines
drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +0
-0NULL);if (ret) {pr_err ("Failed to create new queue err:%d\n" , ret);
Am 22.06.22 um 17:36 schrieb Graham Sider:
Quoted 19 lines
Starting with GFX11, MES requires wptr BOs to be GTT allocated/mapped to GART for usermode queues in order to support oversubscription. In the case that work is submitted to an unmapped queue, MES must have a GART wptr address to determine whether the queue should be mapped.
This change is accompanied with changes in MES and is applicable for
MES_API_VERSION>= 2.v3:
- Useamdgpu_vm_bo_lookup_mappingforwptr_bomapping lookup
- Movewptr_borefcount increment toamdgpu_amdkfd_map_gtt_bo_to_gart
- Removelist_del_initfromamdgpu_amdkfd_map_gtt_bo_to_gart
- Cleanup/fixcreate_queuewptr_boerror handling
v4:
- Add MES version shift/mask defines toamdgpu_mes.h
- Change version check fromMES_VERSIONtoMES_API_VERSION
- Add check inkfd_ioctl_create_queuebefore wptr bo pin/GART map to ensure bo is a single page.
Quoted 2 lines
Signed-off-by Graham Sider <[email protected]>Acked-by Alex Deucher <[email protected]>
Quoted 312 lines
Reviewed-by Philip Yang <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 2 + .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 48 +++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h | 7 +++ drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 45 ++++++++++++++++- .../drm/amd/amdkfd/kfd_device_queue_manager.c | 9 +++- .../gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c | 2 + drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 ++ .../amd/amdkfd/kfd_process_queue_manager.c | 20 +++++--- 8 files changed, 127 insertions(+), 9 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h +2
-0@@ -286,6 +286,8 @@ int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct kgd_mem *mem, void **kptr, uint64_t *size);void amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel (struct kgd_mem *mem); +int amdgpu_amdkfd_map_gtt_bo_to_gart (struct amdgpu_device *adev,struct amdgpu_bo *bo); +int amdgpu_amdkfd_gpuvm_restore_process_bos (void *process_info,struct dma_fence **ef); int amdgpu_amdkfd_gpuvm_get_vm_fault_info(struct amdgpu_device *adev,drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +48
-0@@ -2148,6 +2148,54 @@ int amdgpu_amdkfd_gpuvm_sync_memory(return ret; } +/** +* amdgpu_amdkfd_map_gtt_bo_to_gart - Map BO to GART and increment reference count +* @adev: Device to which allocated BO belongs +* @bo: Buffer object to be mapped +* +* Before return, bo reference count is incremented. To release the reference and unpin/ +* unmap the BO, call amdgpu_amdkfd_free_gtt_mem. +*/ +int amdgpu_amdkfd_map_gtt_bo_to_gart (struct amdgpu_device *adev,struct amdgpu_bo *bo) +{ +int ret; + + ret =amdgpu_bo_reserve (bo, true); +if (ret) { +pr_err ("Failed to reserve bo. ret %d\n" , ret); + goto err_reserve_bo_failed; + } + + ret =amdgpu_bo_pin (bo,AMDGPU_GEM_DOMAIN_GTT ); +if (ret) { +pr_err ("Failed to pin bo. ret %d\n" , ret); + goto err_pin_bo_failed; + } + + ret =amdgpu_ttm_alloc_gart (&bo->tbo); +if (ret) { +pr_err ("Failed to bind bo to GART. ret %d\n" , ret); + goto err_map_bo_gart_failed; + } + +amdgpu_amdkfd_remove_eviction_fence ( + bo, bo->kfd_bo->process_info->eviction_fence); + +amdgpu_bo_unreserve (bo); + + bo =amdgpu_bo_ref (bo); + +return 0 ; + +err_map_bo_gart_failed: +amdgpu_bo_unpin (bo); +err_pin_bo_failed: +amdgpu_bo_unreserve (bo); +err_reserve_bo_failed: + +return ret; +} +/** amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel() - Map a GTT BO for kernel CPU access * * @mem: Buffer object to be mapped for CPU access drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h +7
-0@@ -32,6 +32,13 @@#define AMDGPU_MES_MAX_GFX_PIPES 2#define AMDGPU_MES_MAX_SDMA_PIPES 2 +#define AMDGPU_MES_API_VERSION_SHIFT 12 +#define AMDGPU_MES_FEAT_VERSION_SHIFT 24 + +#define AMDGPU_MES_VERSION_MASK 0x00000fff +#define AMDGPU_MES_API_VERSION_MASK 0x00fff000 +#define AMDGPU_MES_FEAT_VERSION_MASK 0xff000000 +enum amdgpu_mes_priority_level {AMDGPU_MES_PRIORITY_LEVEL_LOW =0 ,AMDGPU_MES_PRIORITY_LEVEL_NORMAL =1 ,drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +43
-2@@ -299,6 +299,7 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,struct kfd_process_device *pdd;struct queue_properties q_properties;uint32_t doorbell_offset_in_process =0 ; +struct amdgpu_bo *wptr_bo =NULL ;memset (&q_properties,0 ,sizeof (struct queue_properties ));@@ -326,12 +327,49 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, goto err_bind_process; } +/* Starting with GFX11, wptr BOs must be mapped to GART for MES to determine work +* on unmapped queues for usermode queue oversubscription (no aggregated doorbell) +*/ +if (dev->shared_resources.enable_mes && + ((dev->adev->mes.sched_version &AMDGPU_MES_API_VERSION_MASK ) + >>AMDGPU_MES_API_VERSION_SHIFT ) >=2 ) { +struct amdgpu_bo_va_mapping *wptr_mapping; +struct amdgpu_vm *wptr_vm; + + wptr_vm =drm_priv_to_vm (pdd->drm_priv); + err =amdgpu_bo_reserve (wptr_vm->root.bo, false); +if (err) + goto err_wptr_map_gart; + + wptr_mapping =amdgpu_vm_bo_lookup_mapping ( + wptr_vm, args->write_pointer_address >>PAGE_SHIFT ); +amdgpu_bo_unreserve (wptr_vm->root.bo); +if (!wptr_mapping) { +pr_err ("Failed to lookup wptr bo\n" ); + err = -EINVAL ; + goto err_wptr_map_gart; + } + + wptr_bo = wptr_mapping->bo_va->base.bo; +if (wptr_bo->tbo.base.size >PAGE_SIZE ) { +pr_err ("Requested GART mapping for wptr bo larger than one page\n" ); + err = -EINVAL ; + goto err_wptr_map_gart; + } + + err =amdgpu_amdkfd_map_gtt_bo_to_gart (dev->adev, wptr_bo); +if (err) { +pr_err ("Failed to map wptr bo to GART\n" ); + goto err_wptr_map_gart; + } + } +pr_debug ("Creating queue for PASID 0x%x on gpu 0x%x\n" , p->pasid, dev->id); - err =pqm_create_queue (&p->pqm, dev, filep, &q_properties, &queue_id,NULL ,NULL ,NULL , - &doorbell_offset_in_process); + err =pqm_create_queue (&p->pqm, dev, filep, &q_properties, &queue_id, wptr_bo, +NULL ,NULL ,NULL , &doorbell_offset_in_process);if (err !=0 ) goto err_create_queue;@@ -363,6 +401,9 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,return 0 ; err_create_queue: +if (wptr_bo) +amdgpu_amdkfd_free_gtt_mem (dev->adev, wptr_bo); +err_wptr_map_gart: err_bind_process: err_pdd:mutex_unlock (&p->mutex);drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +8
-1@@ -177,6 +177,7 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q,struct kfd_process_device *pdd =qpd_to_pdd (qpd);struct mes_add_queue_input queue_input;int r, queue_type; +uint64_t wptr_addr_off;if (dqm->is_hws_hang)return -EIO ;@@ -196,7 +197,13 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q,AMDGPU_MES_PRIORITY_LEVEL_NORMAL ; queue_input.doorbell_offset = q->properties.doorbell_off; queue_input.mqd_addr = q->gart_mqd_addr; - queue_input.wptr_addr = (uint64_t )q->properties.write_ptr; + +if (q->wptr_bo) { + wptr_addr_off = (uint64_t )q->properties.write_ptr - (uint64_t )q->wptr_bo->kfd_bo->va; + queue_input.wptr_addr = ((uint64_t )q->wptr_bo->tbo.resource->start <<PAGE_SHIFT ) + wptr_addr_off; + }else + queue_input.wptr_addr = (uint64_t )q->properties.write_ptr; + queue_input.paging = false; queue_input.tba_addr = qpd->tba_addr; queue_input.tma_addr = qpd->tma_addr;drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v11.c +2
-0@@ -377,6 +377,8 @@ static void update_mqd_sdma(struct mqd_manager *mm, void *mqd, m->sdmax_rlcx_rb_base_hi =upper_32_bits (q->queue_address >>8 ); m->sdmax_rlcx_rb_rptr_addr_lo =lower_32_bits ((uint64_t )q->read_ptr); m->sdmax_rlcx_rb_rptr_addr_hi =upper_32_bits ((uint64_t )q->read_ptr); + m->sdmax_rlcx_rb_wptr_poll_addr_lo =lower_32_bits ((uint64_t )q->write_ptr); + m->sdmax_rlcx_rb_wptr_poll_addr_hi =upper_32_bits ((uint64_t )q->write_ptr); m->sdmax_rlcx_doorbell_offset = q->doorbell_off <<SDMA0_QUEUE0_DOORBELL_OFFSET__OFFSET__SHIFT ;drivers/gpu/drm/amd/amdkfd/kfd_priv.h +3
-0@@ -570,6 +570,8 @@ struct queue {void *gang_ctx_bo;uint64_t gang_ctx_gpu_addr;void *gang_ctx_cpu_ptr; + +struct amdgpu_bo *wptr_bo; };enum KFD_MQD_TYPE {@@ -1205,6 +1207,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,struct file *f, struct queue_properties *properties, unsigned int *qid, + struct amdgpu_bo *wptr_bo, const struct kfd_criu_queue_priv_data *q_data, const void *restore_mqd, const void *restore_ctl_stack,drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +14
-6@@ -180,7 +180,8 @@ void pqm_uninit(struct process_queue_manager *pqm)static int init_user_queue (struct process_queue_manager *pqm,struct kfd_dev *dev,struct queue **q,struct queue_properties *q_properties, -struct file *f,unsigned int qid) +struct file *f,struct amdgpu_bo *wptr_bo, +unsigned int qid) {int retval;@@ -210,6 +211,7 @@ static int init_user_queue(struct process_queue_manager *pqm, goto cleanup; }memset ((*q)->gang_ctx_cpu_ptr,0 ,AMDGPU_MES_GANG_CTX_SIZE ); + (*q)->wptr_bo = wptr_bo; }pr_debug ("PQM After init queue" );@@ -226,6 +228,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,struct file *f, struct queue_properties *properties, unsigned int *qid, + struct amdgpu_bo *wptr_bo, const struct kfd_criu_queue_priv_data *q_data, const void *restore_mqd, const void *restore_ctl_stack,@@ -288,7 +291,7 @@ int pqm_create_queue(struct process_queue_manager *pqm,* allocate_sdma_queue() in create_queue() has the * corresponding check logic. */ - retval =init_user_queue (pqm, dev, &q, properties, f, *qid); + retval =init_user_queue (pqm, dev, &q, properties, f, wptr_bo, *qid);if (retval !=0 ) goto err_create_queue; pqn->q = q;@@ -309,7 +312,7 @@ int pqm_create_queue(struct process_queue_manager *pqm, goto err_create_queue; } - retval =init_user_queue (pqm, dev, &q, properties, f, *qid); + retval =init_user_queue (pqm, dev, &q, properties, f, wptr_bo, *qid);if (retval !=0 ) goto err_create_queue; pqn->q = q;@@ -435,10 +438,15 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid) pdd->qpd.num_gws =0 ; } -if (dev->shared_resources.enable_mes) +if (dev->shared_resources.enable_mes) {amdgpu_amdkfd_free_gtt_mem (dev->adev, pqn->q->gang_ctx_bo); -kfd_procfs_del_queue (pqn->q); +if (pqn->q->wptr_bo) +amdgpu_amdkfd_free_gtt_mem (dev->adev, pqn->q->wptr_bo); + +kfd_procfs_del_queue (pqn->q); + + }uninit_queue (pqn->q); }@@ -844,7 +852,7 @@ int kfd_criu_restore_queue(struct kfd_process *p,print_queue_properties (&qp); - ret =pqm_create_queue (&p->pqm, pdd->dev,NULL , &qp, &queue_id, q_data, mqd, ctl_stack, + ret =pqm_create_queue (&p->pqm, pdd->dev,NULL , &qp, &queue_id,NULL , q_data, mqd, ctl_stack,NULL );if (ret) {pr_err ("Failed to create new queue err:%d\n" , ret);