summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/amd/amdkfd
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
-rw-r--r--drivers/gpu/drm/amd/amdkfd/Kconfig2
-rw-r--r--drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c7
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_chardev.c4
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_events.c1
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c2
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c23
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_svm.c54
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_svm.h11
-rw-r--r--drivers/gpu/drm/amd/amdkfd/kfd_topology.c6
9 files changed, 47 insertions, 63 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/Kconfig b/drivers/gpu/drm/amd/amdkfd/Kconfig
index d3c3d3ab7225..62e88e5362e9 100644
--- a/drivers/gpu/drm/amd/amdkfd/Kconfig
+++ b/drivers/gpu/drm/amd/amdkfd/Kconfig
@@ -5,7 +5,7 @@
config HSA_AMD
bool "HSA kernel driver for AMD GPU devices"
- depends on DRM_AMDGPU && (X86_64 || ARM64 || PPC64)
+ depends on DRM_AMDGPU && (X86_64 || ARM64 || PPC64 || (RISCV && 64BIT))
select HMM_MIRROR
select MMU_NOTIFIER
select DRM_AMDGPU_USERPTR
diff --git a/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c b/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c
index 981d9adcc5e1..73acbe0b7c21 100644
--- a/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c
+++ b/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c
@@ -91,7 +91,6 @@ static void cik_event_interrupt_wq(struct kfd_node *dev,
const struct cik_ih_ring_entry *ihre =
(const struct cik_ih_ring_entry *)ih_ring_entry;
uint32_t context_id = ihre->data & 0xfffffff;
- unsigned int vmid = (ihre->ring_id & 0x0000ff00) >> 8;
u32 pasid = (ihre->ring_id & 0xffff0000) >> 16;
if (pasid == 0)
@@ -125,11 +124,7 @@ static void cik_event_interrupt_wq(struct kfd_node *dev,
return;
}
- if (info.vmid == vmid)
- kfd_signal_vm_fault_event(pdd, &info, NULL);
- else
- kfd_signal_vm_fault_event(pdd, &info, NULL);
-
+ kfd_signal_vm_fault_event(pdd, &info, NULL);
kfd_unref_process(p);
}
}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 1e9dd00620bf..a2149afa5803 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -2039,9 +2039,7 @@ static int criu_get_process_object_info(struct kfd_process *p,
num_events = kfd_get_num_events(p);
- ret = svm_range_get_info(p, &num_svm_ranges, &svm_priv_data_size);
- if (ret)
- return ret;
+ svm_range_get_info(p, &num_svm_ranges, &svm_priv_data_size);
*num_objects = num_queues + num_events + num_svm_ranges;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
index e54e708ed82d..2b294ada3ec0 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
@@ -1350,6 +1350,7 @@ void kfd_signal_poison_consumed_event(struct kfd_node *dev, u32 pasid)
user_gpu_id = kfd_process_get_user_gpu_id(p, dev->id);
if (unlikely(user_gpu_id == -EINVAL)) {
WARN_ONCE(1, "Could not get user_gpu_id from dev->id:%x\n", dev->id);
+ kfd_unref_process(p);
return;
}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
index 8fa6489b6f5d..505036968a77 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c
@@ -240,7 +240,7 @@ static int pm_map_queues_v9(struct packet_manager *pm, uint32_t *buffer,
packet->bitfields2.engine_sel =
engine_sel__mes_map_queues__compute_vi;
- packet->bitfields2.gws_control_queue = q->gws ? 1 : 0;
+ packet->bitfields2.gws_control_queue = q->properties.is_gws ? 1 : 0;
packet->bitfields2.extended_engine_sel =
extended_engine_sel__mes_map_queues__legacy_engine_sel;
packet->bitfields2.queue_type =
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 6d5fa57d4a23..c643e0ccec52 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
@@ -279,20 +279,17 @@ static int init_user_queue(struct process_queue_manager *pqm,
/* 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->adev->mes.sched_version & AMDGPU_MES_API_VERSION_MASK)
- >> AMDGPU_MES_API_VERSION_SHIFT) >= 2) {
- if (dev->adev != amdgpu_ttm_adev(q_properties->wptr_bo->tbo.bdev)) {
- pr_err("Queue memory allocated to wrong device\n");
- retval = -EINVAL;
- goto free_gang_ctx_bo;
- }
+ if (dev->adev != amdgpu_ttm_adev(q_properties->wptr_bo->tbo.bdev)) {
+ pr_err("Queue memory allocated to wrong device\n");
+ retval = -EINVAL;
+ goto free_gang_ctx_bo;
+ }
- retval = amdgpu_amdkfd_map_gtt_bo_to_gart(q_properties->wptr_bo,
- &(*q)->wptr_bo_gart);
- if (retval) {
- pr_err("Failed to map wptr bo to GART\n");
- goto free_gang_ctx_bo;
- }
+ retval = amdgpu_amdkfd_map_gtt_bo_to_gart(q_properties->wptr_bo,
+ &(*q)->wptr_bo_gart);
+ if (retval) {
+ pr_err("Failed to map wptr bo to GART\n");
+ goto free_gang_ctx_bo;
}
}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 72be6e152e88..a0f22ea6d15a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -1171,13 +1171,12 @@ svm_range_split_head(struct svm_range *prange, uint64_t new_start,
}
static void
-svm_range_add_child(struct svm_range *prange, struct mm_struct *mm,
- struct svm_range *pchild, enum svm_work_list_ops op)
+svm_range_add_child(struct svm_range *prange, struct svm_range *pchild, enum svm_work_list_ops op)
{
pr_debug("add child 0x%p [0x%lx 0x%lx] to prange 0x%p child list %d\n",
pchild, pchild->start, pchild->last, prange, op);
- pchild->work_item.mm = mm;
+ pchild->work_item.mm = NULL;
pchild->work_item.op = op;
list_add_tail(&pchild->child_list, &prange->child_list);
}
@@ -1278,7 +1277,7 @@ svm_range_get_pte_flags(struct kfd_node *node,
mapping_flags |= ext_coherent ? AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC;
/* system memory accessed by the dGPU */
} else {
- if (gc_ip_version < IP_VERSION(9, 5, 0))
+ if (gc_ip_version < IP_VERSION(9, 5, 0) || ext_coherent)
mapping_flags |= AMDGPU_VM_MTYPE_UC;
else
mapping_flags |= AMDGPU_VM_MTYPE_NC;
@@ -2394,15 +2393,17 @@ svm_range_add_list_work(struct svm_range_list *svms, struct svm_range *prange,
prange->work_item.op != SVM_OP_UNMAP_RANGE)
prange->work_item.op = op;
} else {
- prange->work_item.op = op;
-
- /* Pairs with mmput in deferred_list_work */
- mmget(mm);
- prange->work_item.mm = mm;
- list_add_tail(&prange->deferred_list,
- &prange->svms->deferred_range_list);
- pr_debug("add prange 0x%p [0x%lx 0x%lx] to work list op %d\n",
- prange, prange->start, prange->last, op);
+ /* Pairs with mmput in deferred_list_work.
+ * If process is exiting and mm is gone, don't update mmu notifier.
+ */
+ if (mmget_not_zero(mm)) {
+ prange->work_item.mm = mm;
+ prange->work_item.op = op;
+ list_add_tail(&prange->deferred_list,
+ &prange->svms->deferred_range_list);
+ pr_debug("add prange 0x%p [0x%lx 0x%lx] to work list op %d\n",
+ prange, prange->start, prange->last, op);
+ }
}
spin_unlock(&svms->deferred_list_lock);
}
@@ -2416,8 +2417,7 @@ void schedule_deferred_list_work(struct svm_range_list *svms)
}
static void
-svm_range_unmap_split(struct mm_struct *mm, struct svm_range *parent,
- struct svm_range *prange, unsigned long start,
+svm_range_unmap_split(struct svm_range *parent, struct svm_range *prange, unsigned long start,
unsigned long last)
{
struct svm_range *head;
@@ -2438,12 +2438,12 @@ svm_range_unmap_split(struct mm_struct *mm, struct svm_range *parent,
svm_range_split(tail, last + 1, tail->last, &head);
if (head != prange && tail != prange) {
- svm_range_add_child(parent, mm, head, SVM_OP_UNMAP_RANGE);
- svm_range_add_child(parent, mm, tail, SVM_OP_ADD_RANGE);
+ svm_range_add_child(parent, head, SVM_OP_UNMAP_RANGE);
+ svm_range_add_child(parent, tail, SVM_OP_ADD_RANGE);
} else if (tail != prange) {
- svm_range_add_child(parent, mm, tail, SVM_OP_UNMAP_RANGE);
+ svm_range_add_child(parent, tail, SVM_OP_UNMAP_RANGE);
} else if (head != prange) {
- svm_range_add_child(parent, mm, head, SVM_OP_UNMAP_RANGE);
+ svm_range_add_child(parent, head, SVM_OP_UNMAP_RANGE);
} else if (parent != prange) {
prange->work_item.op = SVM_OP_UNMAP_RANGE;
}
@@ -2520,14 +2520,14 @@ svm_range_unmap_from_cpu(struct mm_struct *mm, struct svm_range *prange,
l = min(last, pchild->last);
if (l >= s)
svm_range_unmap_from_gpus(pchild, s, l, trigger);
- svm_range_unmap_split(mm, prange, pchild, start, last);
+ svm_range_unmap_split(prange, pchild, start, last);
mutex_unlock(&pchild->lock);
}
s = max(start, prange->start);
l = min(last, prange->last);
if (l >= s)
svm_range_unmap_from_gpus(prange, s, l, trigger);
- svm_range_unmap_split(mm, prange, prange, start, last);
+ svm_range_unmap_split(prange, prange, start, last);
if (unmap_parent)
svm_range_add_list_work(svms, prange, mm, SVM_OP_UNMAP_RANGE);
@@ -2570,8 +2570,6 @@ svm_range_cpu_invalidate_pagetables(struct mmu_interval_notifier *mni,
if (range->event == MMU_NOTIFY_RELEASE)
return true;
- if (!mmget_not_zero(mni->mm))
- return true;
start = mni->interval_tree.start;
last = mni->interval_tree.last;
@@ -2598,7 +2596,6 @@ svm_range_cpu_invalidate_pagetables(struct mmu_interval_notifier *mni,
}
svm_range_unlock(prange);
- mmput(mni->mm);
return true;
}
@@ -4075,8 +4072,8 @@ exit:
return ret;
}
-int svm_range_get_info(struct kfd_process *p, uint32_t *num_svm_ranges,
- uint64_t *svm_priv_data_size)
+void svm_range_get_info(struct kfd_process *p, uint32_t *num_svm_ranges,
+ uint64_t *svm_priv_data_size)
{
uint64_t total_size, accessibility_size, common_attr_size;
int nattr_common = 4, nattr_accessibility = 1;
@@ -4088,8 +4085,6 @@ int svm_range_get_info(struct kfd_process *p, uint32_t *num_svm_ranges,
*svm_priv_data_size = 0;
svms = &p->svms;
- if (!svms)
- return -EINVAL;
mutex_lock(&svms->lock);
list_for_each_entry(prange, &svms->list, list) {
@@ -4131,7 +4126,6 @@ int svm_range_get_info(struct kfd_process *p, uint32_t *num_svm_ranges,
pr_debug("num_svm_ranges %u total_priv_size %llu\n", *num_svm_ranges,
*svm_priv_data_size);
- return 0;
}
int kfd_criu_checkpoint_svm(struct kfd_process *p,
@@ -4148,8 +4142,6 @@ int kfd_criu_checkpoint_svm(struct kfd_process *p,
struct mm_struct *mm;
svms = &p->svms;
- if (!svms)
- return -EINVAL;
mm = get_task_mm(p->lead_thread);
if (!mm) {
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
index 6ea23c78009c..01c7a4877904 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
@@ -184,8 +184,8 @@ void schedule_deferred_list_work(struct svm_range_list *svms);
void svm_range_dma_unmap_dev(struct device *dev, dma_addr_t *dma_addr,
unsigned long offset, unsigned long npages);
void svm_range_dma_unmap(struct svm_range *prange);
-int svm_range_get_info(struct kfd_process *p, uint32_t *num_svm_ranges,
- uint64_t *svm_priv_data_size);
+void svm_range_get_info(struct kfd_process *p, uint32_t *num_svm_ranges,
+ uint64_t *svm_priv_data_size);
int kfd_criu_checkpoint_svm(struct kfd_process *p,
uint8_t __user *user_priv_data,
uint64_t *priv_offset);
@@ -237,13 +237,12 @@ static inline int svm_range_schedule_evict_svm_bo(
return -EINVAL;
}
-static inline int svm_range_get_info(struct kfd_process *p,
- uint32_t *num_svm_ranges,
- uint64_t *svm_priv_data_size)
+static inline void svm_range_get_info(struct kfd_process *p,
+ uint32_t *num_svm_ranges,
+ uint64_t *svm_priv_data_size)
{
*num_svm_ranges = 0;
*svm_priv_data_size = 0;
- return 0;
}
static inline int kfd_criu_checkpoint_svm(struct kfd_process *p,
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index baa2374acdeb..4ec73f33535e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -510,6 +510,10 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr,
dev->node_props.capability |=
HSA_CAP_AQL_QUEUE_DOUBLE_MAP;
+ if (KFD_GC_VERSION(dev->gpu) < IP_VERSION(10, 0, 0) &&
+ (dev->gpu->adev->sdma.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE))
+ dev->node_props.capability2 |= HSA_CAP2_PER_SDMA_QUEUE_RESET_SUPPORTED;
+
sysfs_show_32bit_prop(buffer, offs, "max_engine_clk_fcompute",
dev->node_props.max_engine_clk_fcompute);
@@ -2008,8 +2012,6 @@ static void kfd_topology_set_capabilities(struct kfd_topology_device *dev)
if (!amdgpu_sriov_vf(dev->gpu->adev))
dev->node_props.capability |= HSA_CAP_PER_QUEUE_RESET_SUPPORTED;
- if (dev->gpu->adev->sdma.supported_reset & AMDGPU_RESET_TYPE_PER_QUEUE)
- dev->node_props.capability2 |= HSA_CAP2_PER_SDMA_QUEUE_RESET_SUPPORTED;
} else {
dev->node_props.debug_prop |= HSA_DBG_WATCH_ADDR_MASK_LO_BIT_GFX10 |
HSA_DBG_WATCH_ADDR_MASK_HI_BIT;
generated by cgit (git 2.34.1) at 2025-07-30 08:35:57 +0000