drm/amdkfd: Remove cu mask from struct queue_properties(v2)

Actually, cu_mask has been copied to mqd memory and
does't have to persist in queue_properties. Remove it
from queue_properties.

And use struct mqd_update_info to store such properties,
then pass it to update queue operation.

v2:
* Rename pqm_update_queue to pqm_update_queue_properties.
* Rename struct queue_update_info to struct mqd_update_info.
* Rename pqm_set_cu_mask to pqm_update_mqd.
Suggested-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Lang Yu <lang.yu@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdkfd/kfd_chardev.c

@@ -405,7 +405,7 @@ static int kfd_ioctl_update_queue(struct file *filp, struct kfd_process *p,
 
 	mutex_lock(&p->mutex);
 
-	retval = pqm_update_queue(&p->pqm, args->queue_id, &properties);
+	retval = pqm_update_queue_properties(&p->pqm, args->queue_id, &properties);
 
 	mutex_unlock(&p->mutex);
 
@@ -418,7 +418,7 @@ static int kfd_ioctl_set_cu_mask(struct file *filp, struct kfd_process *p,
 	int retval;
 	const int max_num_cus = 1024;
 	struct kfd_ioctl_set_cu_mask_args *args = data;
-	struct queue_properties properties;
+	struct mqd_update_info minfo = {0};
 	uint32_t __user *cu_mask_ptr = (uint32_t __user *)args->cu_mask_ptr;
 	size_t cu_mask_size = sizeof(uint32_t) * (args->num_cu_mask / 32);
 
@@ -428,8 +428,8 @@ static int kfd_ioctl_set_cu_mask(struct file *filp, struct kfd_process *p,
 		return -EINVAL;
 	}
 
-	properties.cu_mask_count = args->num_cu_mask;
-	if (properties.cu_mask_count == 0) {
+	minfo.cu_mask.count = args->num_cu_mask;
+	if (minfo.cu_mask.count == 0) {
 		pr_debug("CU mask cannot be 0");
 		return -EINVAL;
 	}
@@ -438,32 +438,33 @@ static int kfd_ioctl_set_cu_mask(struct file *filp, struct kfd_process *p,
 	 * limit of max_num_cus bits.  We can then just drop any CU mask bits
 	 * past max_num_cus bits and just use the first max_num_cus bits.
 	 */
-	if (properties.cu_mask_count > max_num_cus) {
+	if (minfo.cu_mask.count > max_num_cus) {
 		pr_debug("CU mask cannot be greater than 1024 bits");
-		properties.cu_mask_count = max_num_cus;
+		minfo.cu_mask.count = max_num_cus;
 		cu_mask_size = sizeof(uint32_t) * (max_num_cus/32);
 	}
 
-	properties.cu_mask = kzalloc(cu_mask_size, GFP_KERNEL);
-	if (!properties.cu_mask)
+	minfo.cu_mask.ptr = kzalloc(cu_mask_size, GFP_KERNEL);
+	if (!minfo.cu_mask.ptr)
 		return -ENOMEM;
 
-	retval = copy_from_user(properties.cu_mask, cu_mask_ptr, cu_mask_size);
+	retval = copy_from_user(minfo.cu_mask.ptr, cu_mask_ptr, cu_mask_size);
 	if (retval) {
 		pr_debug("Could not copy CU mask from userspace");
-		kfree(properties.cu_mask);
-		return -EFAULT;
+		retval = -EFAULT;
+		goto out;
 	}
 
+	minfo.update_flag = UPDATE_FLAG_CU_MASK;
+
 	mutex_lock(&p->mutex);
 
-	retval = pqm_set_cu_mask(&p->pqm, args->queue_id, &properties);
+	retval = pqm_update_mqd(&p->pqm, args->queue_id, &minfo);
 
 	mutex_unlock(&p->mutex);
 
-	if (retval)
-		kfree(properties.cu_mask);
-
+out:
+	kfree(minfo.cu_mask.ptr);
 	return retval;
 }
 

drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c

@@ -136,7 +136,6 @@ static bool kq_initialize(struct kernel_queue *kq, struct kfd_dev *dev,
 	prop.write_ptr = (uint32_t *) kq->wptr_gpu_addr;
 	prop.eop_ring_buffer_address = kq->eop_gpu_addr;
 	prop.eop_ring_buffer_size = PAGE_SIZE;
-	prop.cu_mask = NULL;
 
 	if (init_queue(&kq->queue, &prop) != 0)
 		goto err_init_queue;

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c

@@ -42,16 +42,17 @@ static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
 }
 
 static void update_cu_mask(struct mqd_manager *mm, void *mqd,
-			struct queue_properties *q)
+			struct mqd_update_info *minfo)
 {
 	struct cik_mqd *m;
 	uint32_t se_mask[4] = {0}; /* 4 is the max # of SEs */
 
-	if (q->cu_mask_count == 0)
+	if (!minfo || (minfo->update_flag != UPDATE_FLAG_CU_MASK) ||
+	    !minfo->cu_mask.ptr)
 		return;
 
 	mqd_symmetrically_map_cu_mask(mm,
-		q->cu_mask, q->cu_mask_count, se_mask);
+		minfo->cu_mask.ptr, minfo->cu_mask.count, se_mask);
 
 	m = get_mqd(mqd);
 	m->compute_static_thread_mgmt_se0 = se_mask[0];
@@ -215,7 +216,7 @@ static void __update_mqd(struct mqd_manager *mm, void *mqd,
 	if (q->format == KFD_QUEUE_FORMAT_AQL)
 		m->cp_hqd_pq_control |= NO_UPDATE_RPTR;
 
-	update_cu_mask(mm, mqd, q);
+	update_cu_mask(mm, mqd, minfo);
 	set_priority(m, q);
 
 	q->is_active = QUEUE_IS_ACTIVE(*q);

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c

@@ -42,16 +42,17 @@ static inline struct v10_sdma_mqd *get_sdma_mqd(void *mqd)
 }
 
 static void update_cu_mask(struct mqd_manager *mm, void *mqd,
-			   struct queue_properties *q)
+			struct mqd_update_info *minfo)
 {
 	struct v10_compute_mqd *m;
 	uint32_t se_mask[4] = {0}; /* 4 is the max # of SEs */
 
-	if (q->cu_mask_count == 0)
+	if (!minfo || (minfo->update_flag != UPDATE_FLAG_CU_MASK) ||
+	    !minfo->cu_mask.ptr)
 		return;
 
 	mqd_symmetrically_map_cu_mask(mm,
-		q->cu_mask, q->cu_mask_count, se_mask);
+		minfo->cu_mask.ptr, minfo->cu_mask.count, se_mask);
 
 	m = get_mqd(mqd);
 	m->compute_static_thread_mgmt_se0 = se_mask[0];
@@ -219,7 +220,7 @@ static void update_mqd(struct mqd_manager *mm, void *mqd,
 	if (mm->dev->cwsr_enabled)
 		m->cp_hqd_ctx_save_control = 0;
 
-	update_cu_mask(mm, mqd, q);
+	update_cu_mask(mm, mqd, minfo);
 	set_priority(m, q);
 
 	q->is_active = QUEUE_IS_ACTIVE(*q);

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c

@@ -43,16 +43,17 @@ static inline struct v9_sdma_mqd *get_sdma_mqd(void *mqd)
 }
 
 static void update_cu_mask(struct mqd_manager *mm, void *mqd,
-			struct queue_properties *q)
+			struct mqd_update_info *minfo)
 {
 	struct v9_mqd *m;
 	uint32_t se_mask[KFD_MAX_NUM_SE] = {0};
 
-	if (q->cu_mask_count == 0)
+	if (!minfo || (minfo->update_flag != UPDATE_FLAG_CU_MASK) ||
+	    !minfo->cu_mask.ptr)
 		return;
 
 	mqd_symmetrically_map_cu_mask(mm,
-		q->cu_mask, q->cu_mask_count, se_mask);
+		minfo->cu_mask.ptr, minfo->cu_mask.count, se_mask);
 
 	m = get_mqd(mqd);
 	m->compute_static_thread_mgmt_se0 = se_mask[0];
@@ -270,7 +271,7 @@ static void update_mqd(struct mqd_manager *mm, void *mqd,
 	if (mm->dev->cwsr_enabled && q->ctx_save_restore_area_address)
 		m->cp_hqd_ctx_save_control = 0;
 
-	update_cu_mask(mm, mqd, q);
+	update_cu_mask(mm, mqd, minfo);
 	set_priority(m, q);
 
 	q->is_active = QUEUE_IS_ACTIVE(*q);

drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c

@@ -45,16 +45,17 @@ static inline struct vi_sdma_mqd *get_sdma_mqd(void *mqd)
 }
 
 static void update_cu_mask(struct mqd_manager *mm, void *mqd,
-			struct queue_properties *q)
+			struct mqd_update_info *minfo)
 {
 	struct vi_mqd *m;
 	uint32_t se_mask[4] = {0}; /* 4 is the max # of SEs */
 
-	if (q->cu_mask_count == 0)
+	if (!minfo || (minfo->update_flag != UPDATE_FLAG_CU_MASK) ||
+	    !minfo->cu_mask.ptr)
 		return;
 
 	mqd_symmetrically_map_cu_mask(mm,
-		q->cu_mask, q->cu_mask_count, se_mask);
+		minfo->cu_mask.ptr, minfo->cu_mask.count, se_mask);
 
 	m = get_mqd(mqd);
 	m->compute_static_thread_mgmt_se0 = se_mask[0];
@@ -230,7 +231,7 @@ static void __update_mqd(struct mqd_manager *mm, void *mqd,
 			atc_bit << CP_HQD_CTX_SAVE_CONTROL__ATC__SHIFT |
 			mtype << CP_HQD_CTX_SAVE_CONTROL__MTYPE__SHIFT;
 
-	update_cu_mask(mm, mqd, q);
+	update_cu_mask(mm, mqd, minfo);
 	set_priority(m, q);
 
 	q->is_active = QUEUE_IS_ACTIVE(*q);

drivers/gpu/drm/amd/amdkfd/kfd_priv.h

@@ -472,9 +472,6 @@ struct queue_properties {
 	uint32_t ctl_stack_size;
 	uint64_t tba_addr;
 	uint64_t tma_addr;
-	/* Relevant for CU */
-	uint32_t cu_mask_count; /* Must be a multiple of 32 */
-	uint32_t *cu_mask;
 };
 
 #define QUEUE_IS_ACTIVE(q) ((q).queue_size > 0 &&	\
@@ -482,7 +479,19 @@ struct queue_properties {
 			    (q).queue_percent > 0 &&	\
 			    !(q).is_evicted)
 
-struct mqd_update_info;
+enum mqd_update_flag {
+	UPDATE_FLAG_CU_MASK = 0,
+};
+
+struct mqd_update_info {
+	union {
+		struct {
+			uint32_t count; /* Must be a multiple of 32 */
+			uint32_t *ptr;
+		} cu_mask;
+	};
+	enum mqd_update_flag update_flag;
+};
 
 /**
  * struct queue
@@ -1036,10 +1045,10 @@ int pqm_create_queue(struct process_queue_manager *pqm,
 			    unsigned int *qid,
 			    uint32_t *p_doorbell_offset_in_process);
 int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid);
-int pqm_update_queue(struct process_queue_manager *pqm, unsigned int qid,
-			struct queue_properties *p);
-int pqm_set_cu_mask(struct process_queue_manager *pqm, unsigned int qid,
+int pqm_update_queue_properties(struct process_queue_manager *pqm, unsigned int qid,
 			struct queue_properties *p);
+int pqm_update_mqd(struct process_queue_manager *pqm, unsigned int qid,
+			struct mqd_update_info *minfo);
 int pqm_set_gws(struct process_queue_manager *pqm, unsigned int qid,
 			void *gws);
 struct kernel_queue *pqm_get_kernel_queue(struct process_queue_manager *pqm,

drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c

@@ -394,8 +394,6 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid)
 			pdd->qpd.num_gws = 0;
 		}
 
-		kfree(pqn->q->properties.cu_mask);
-		pqn->q->properties.cu_mask = NULL;
 		uninit_queue(pqn->q);
 	}
 
@@ -411,8 +409,8 @@ int pqm_destroy_queue(struct process_queue_manager *pqm, unsigned int qid)
 	return retval;
 }
 
-int pqm_update_queue(struct process_queue_manager *pqm, unsigned int qid,
-			struct queue_properties *p)
+int pqm_update_queue_properties(struct process_queue_manager *pqm,
+				unsigned int qid, struct queue_properties *p)
 {
 	int retval;
 	struct process_queue_node *pqn;
@@ -436,8 +434,8 @@ int pqm_update_queue(struct process_queue_manager *pqm, unsigned int qid,
 	return 0;
 }
 
-int pqm_set_cu_mask(struct process_queue_manager *pqm, unsigned int qid,
-			struct queue_properties *p)
+int pqm_update_mqd(struct process_queue_manager *pqm,
+				unsigned int qid, struct mqd_update_info *minfo)
 {
 	int retval;
 	struct process_queue_node *pqn;
@@ -448,16 +446,8 @@ int pqm_set_cu_mask(struct process_queue_manager *pqm, unsigned int qid,
 		return -EFAULT;
 	}
 
-	/* Free the old CU mask memory if it is already allocated, then
-	 * allocate memory for the new CU mask.
-	 */
-	kfree(pqn->q->properties.cu_mask);
-
-	pqn->q->properties.cu_mask_count = p->cu_mask_count;
-	pqn->q->properties.cu_mask = p->cu_mask;
-
 	retval = pqn->q->device->dqm->ops.update_queue(pqn->q->device->dqm,
-							pqn->q, NULL);
+							pqn->q, minfo);
 	if (retval != 0)
 		return retval;