summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c')
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c47
1 files changed, 23 insertions, 24 deletions
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 625ea8ab7a74..804851632c4c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -137,9 +137,10 @@ static void init_sdma_bitmaps(struct device_queue_manager *dqm)
bitmap_set(dqm->xgmi_sdma_bitmap, 0, get_num_xgmi_sdma_queues(dqm));
/* Mask out the reserved queues */
- bitmap_andnot(dqm->sdma_bitmap, dqm->sdma_bitmap,
- dqm->dev->kfd->device_info.reserved_sdma_queues_bitmap,
- KFD_MAX_SDMA_QUEUES);
+ bitmap_clear(dqm->sdma_bitmap, 0, kfd_get_num_sdma_engines(dqm->dev) *
+ dqm->dev->kfd->device_info.num_reserved_sdma_queues_per_engine);
+ bitmap_clear(dqm->xgmi_sdma_bitmap, 0, kfd_get_num_xgmi_sdma_engines(dqm->dev) *
+ dqm->dev->kfd->device_info.num_reserved_sdma_queues_per_engine);
}
void program_sh_mem_settings(struct device_queue_manager *dqm,
@@ -254,6 +255,9 @@ static int add_queue_mes(struct device_queue_manager *dqm, struct queue *q,
queue_input.queue_type = (uint32_t)queue_type;
queue_input.exclusively_scheduled = q->properties.is_gws;
+ queue_input.sh_mem_config_data = qpd->sh_mem_config;
+ queue_input.vm_cntx_cntl = qpd->vm_cntx_cntl;
+ queue_input.xcc_id = ffs(dqm->dev->xcc_mask) - 1;
amdgpu_mes_lock(&adev->mes);
r = adev->mes.funcs->add_hw_queue(&adev->mes, &queue_input);
@@ -284,6 +288,7 @@ static int remove_queue_mes(struct device_queue_manager *dqm, struct queue *q,
memset(&queue_input, 0x0, sizeof(struct mes_remove_queue_input));
queue_input.doorbell_offset = q->properties.doorbell_off;
queue_input.gang_context_addr = q->gang_ctx_gpu_addr;
+ queue_input.xcc_id = ffs(dqm->dev->xcc_mask) - 1;
amdgpu_mes_lock(&adev->mes);
r = adev->mes.funcs->remove_hw_queue(&adev->mes, &queue_input);
@@ -399,8 +404,7 @@ static void increment_queue_count(struct device_queue_manager *dqm,
struct queue *q)
{
dqm->active_queue_count++;
- if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE ||
- q->properties.type == KFD_QUEUE_TYPE_DIQ)
+ if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE)
dqm->active_cp_queue_count++;
if (q->properties.is_gws) {
@@ -414,8 +418,7 @@ static void decrement_queue_count(struct device_queue_manager *dqm,
struct queue *q)
{
dqm->active_queue_count--;
- if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE ||
- q->properties.type == KFD_QUEUE_TYPE_DIQ)
+ if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE)
dqm->active_cp_queue_count--;
if (q->properties.is_gws) {
@@ -673,7 +676,7 @@ static int create_queue_nocpsch(struct device_queue_manager *dqm,
/* Temporarily release dqm lock to avoid a circular lock dependency */
dqm_unlock(dqm);
- q->mqd_mem_obj = mqd_mgr->allocate_mqd(mqd_mgr->dev, &q->properties);
+ q->mqd_mem_obj = mqd_mgr->allocate_mqd(mqd_mgr, &q->properties);
dqm_lock(dqm);
if (!q->mqd_mem_obj) {
@@ -856,8 +859,7 @@ static int destroy_queue_nocpsch_locked(struct device_queue_manager *dqm,
int retval;
struct mqd_manager *mqd_mgr;
- mqd_mgr = dqm->mqd_mgrs[get_mqd_type_from_queue_type(
- q->properties.type)];
+ mqd_mgr = dqm->mqd_mgrs[get_mqd_type_from_queue_type(q->properties.type)];
if (q->properties.type == KFD_QUEUE_TYPE_COMPUTE)
deallocate_hqd(dqm, q);
@@ -1433,13 +1435,12 @@ static int register_process(struct device_queue_manager *dqm,
static int unregister_process(struct device_queue_manager *dqm,
struct qcm_process_device *qpd)
{
- int retval;
+ int retval = 0;
struct device_process_node *cur, *next;
pr_debug("qpd->queues_list is %s\n",
list_empty(&qpd->queues_list) ? "empty" : "not empty");
- retval = 0;
dqm_lock(dqm);
list_for_each_entry_safe(cur, next, &dqm->queues, list) {
@@ -1469,7 +1470,7 @@ set_pasid_vmid_mapping(struct device_queue_manager *dqm, u32 pasid,
unsigned int vmid)
{
uint32_t xcc_mask = dqm->dev->xcc_mask;
- int xcc_id, ret;
+ int xcc_id, ret = 0;
for_each_inst(xcc_id, xcc_mask) {
ret = dqm->dev->kfd2kgd->set_pasid_vmid_mapping(
@@ -1821,8 +1822,6 @@ static int start_cpsch(struct device_queue_manager *dqm)
struct device *dev = dqm->dev->adev->dev;
int retval, num_hw_queue_slots;
- retval = 0;
-
dqm_lock(dqm);
if (!dqm->dev->kfd->shared_resources.enable_mes) {
@@ -2003,7 +2002,7 @@ static int create_queue_cpsch(struct device_queue_manager *dqm, struct queue *q,
dqm->asic_ops.init_sdma_vm(dqm, q, qpd);
q->properties.tba_addr = qpd->tba_addr;
q->properties.tma_addr = qpd->tma_addr;
- q->mqd_mem_obj = mqd_mgr->allocate_mqd(mqd_mgr->dev, &q->properties);
+ q->mqd_mem_obj = mqd_mgr->allocate_mqd(mqd_mgr, &q->properties);
if (!q->mqd_mem_obj) {
retval = -ENOMEM;
goto out_deallocate_doorbell;
@@ -2774,7 +2773,7 @@ dqm_unlock:
static int process_termination_cpsch(struct device_queue_manager *dqm,
struct qcm_process_device *qpd)
{
- int retval;
+ int retval = 0;
struct queue *q;
struct device *dev = dqm->dev->adev->dev;
struct kernel_queue *kq, *kq_next;
@@ -2784,8 +2783,6 @@ static int process_termination_cpsch(struct device_queue_manager *dqm,
KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES;
bool found = false;
- retval = 0;
-
dqm_lock(dqm);
/* Clean all kernel queues */
@@ -2902,8 +2899,9 @@ static int allocate_hiq_sdma_mqd(struct device_queue_manager *dqm)
(dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size *
NUM_XCC(dqm->dev->xcc_mask));
- retval = amdgpu_amdkfd_alloc_gtt_mem(dev->adev, size,
- &(mem_obj->gtt_mem), &(mem_obj->gpu_addr),
+ retval = amdgpu_amdkfd_alloc_kernel_mem(dev->adev, size,
+ AMDGPU_GEM_DOMAIN_GTT,
+ &(mem_obj->mem), &(mem_obj->gpu_addr),
(void *)&(mem_obj->cpu_ptr), false);
return retval;
@@ -2914,7 +2912,7 @@ static void deallocate_hiq_sdma_mqd(struct kfd_node *dev,
{
WARN(!mqd, "No hiq sdma mqd trunk to free");
- amdgpu_amdkfd_free_gtt_mem(dev->adev, &mqd->gtt_mem);
+ amdgpu_amdkfd_free_kernel_mem(dev->adev, &mqd->mem);
}
struct device_queue_manager *device_queue_manager_init(struct kfd_node *dev)
@@ -3012,7 +3010,9 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_node *dev)
break;
default:
- if (KFD_GC_VERSION(dev) >= IP_VERSION(12, 0, 0))
+ if (KFD_GC_VERSION(dev) >= IP_VERSION(12, 1, 0))
+ device_queue_manager_init_v12_1(&dqm->asic_ops);
+ else if (KFD_GC_VERSION(dev) >= IP_VERSION(12, 0, 0))
device_queue_manager_init_v12(&dqm->asic_ops);
else if (KFD_GC_VERSION(dev) >= IP_VERSION(11, 0, 0))
device_queue_manager_init_v11(&dqm->asic_ops);
@@ -3457,7 +3457,6 @@ int suspend_queues(struct kfd_process *p,
else
per_device_suspended++;
} else if (err != -EBUSY) {
- r = err;
queue_ids[q_idx] |= KFD_DBG_QUEUE_ERROR_MASK;
break;
}
generated by cgit v1.2.3 (git 2.25.1) at 2026-08-05 15:42:47 +0000