[PATCH 09/35] drm/amdkfd: validate svm range system memory

Felix Kuehling Felix.Kuehling at amd.com
Thu Jan 7 03:01:01 UTC 2021


From: Philip Yang <Philip.Yang at amd.com>

Use HMM to get system memory pages address, which will be used to
map to GPUs or migrate to vram.

Signed-off-by: Philip Yang <Philip.Yang at amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling at amd.com>
---
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h |  1 +
 drivers/gpu/drm/amd/amdkfd/kfd_svm.c  | 88 +++++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdkfd/kfd_svm.h  |  2 +
 3 files changed, 91 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index cbb2bae1982d..97cf267b6f51 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -735,6 +735,7 @@ struct svm_range_list {
 	struct work_struct		srcu_free_work;
 	struct list_head		free_list;
 	struct mutex			free_list_lock;
+	struct mmu_interval_notifier	notifier;
 };
 
 /* Process data */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 017e77e9ae1e..02918faa70d5 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -135,6 +135,65 @@ svm_get_supported_dev_by_id(struct kfd_process *p, uint32_t gpu_id,
 	return dev;
 }
 
+/**
+ * svm_range_validate_ram - get system memory pages of svm range
+ *
+ * @mm: the mm_struct of process
+ * @prange: the range struct
+ *
+ * After mapping system memory to GPU, system memory maybe invalidated anytime
+ * during application running, we use HMM callback to sync GPU with CPU page
+ * table update, so we don't need use lock to prevent CPU invalidation and check
+ * hmm_range_get_pages_done return value.
+ *
+ * Return:
+ * 0 - OK, otherwise error code
+ */
+static int
+svm_range_validate_ram(struct mm_struct *mm, struct svm_range *prange)
+{
+	uint64_t i;
+	int r;
+
+	if (!prange->pages_addr) {
+		prange->pages_addr = kvmalloc_array(prange->npages,
+						sizeof(*prange->pages_addr),
+						GFP_KERNEL | __GFP_ZERO);
+		if (!prange->pages_addr)
+			return -ENOMEM;
+	}
+
+	r = amdgpu_hmm_range_get_pages(&prange->svms->notifier, mm, NULL,
+				       prange->it_node.start << PAGE_SHIFT,
+				       prange->npages, &prange->hmm_range,
+				       false, true);
+	if (r) {
+		pr_debug("failed %d to get svm range pages\n", r);
+		return r;
+	}
+
+	for (i = 0; i < prange->npages; i++)
+		prange->pages_addr[i] =
+			PFN_PHYS(prange->hmm_range->hmm_pfns[i]);
+
+	amdgpu_hmm_range_get_pages_done(prange->hmm_range);
+	prange->hmm_range = NULL;
+
+	return 0;
+}
+
+static int
+svm_range_validate(struct mm_struct *mm, struct svm_range *prange)
+{
+	int r = 0;
+
+	pr_debug("actual loc 0x%x\n", prange->actual_loc);
+
+	r = svm_range_validate_ram(mm, prange);
+
+	return r;
+}
+
 static int
 svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange,
 		      uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs)
@@ -349,10 +408,28 @@ static void svm_range_srcu_free_work(struct work_struct *work_struct)
 	mutex_unlock(&svms->free_list_lock);
 }
 
+/**
+ * svm_range_cpu_invalidate_pagetables - interval notifier callback
+ *
+ */
+static bool
+svm_range_cpu_invalidate_pagetables(struct mmu_interval_notifier *mni,
+				    const struct mmu_notifier_range *range,
+				    unsigned long cur_seq)
+{
+	return true;
+}
+
+static const struct mmu_interval_notifier_ops svm_range_mn_ops = {
+	.invalidate = svm_range_cpu_invalidate_pagetables,
+};
+
 void svm_range_list_fini(struct kfd_process *p)
 {
 	pr_debug("pasid 0x%x svms 0x%p\n", p->pasid, &p->svms);
 
+	mmu_interval_notifier_remove(&p->svms.notifier);
+
 	/* Ensure srcu free work is finished before process is destroyed */
 	flush_work(&p->svms.srcu_free_work);
 	cleanup_srcu_struct(&p->svms.srcu);
@@ -375,6 +452,8 @@ int svm_range_list_init(struct kfd_process *p)
 	INIT_WORK(&svms->srcu_free_work, svm_range_srcu_free_work);
 	INIT_LIST_HEAD(&svms->free_list);
 	mutex_init(&svms->free_list_lock);
+	mmu_interval_notifier_insert(&svms->notifier, current->mm, 0, ~1ULL,
+				     &svm_range_mn_ops);
 
 	return 0;
 }
@@ -531,6 +610,15 @@ svm_range_set_attr(struct kfd_process *p, uint64_t start, uint64_t size,
 		r = svm_range_apply_attrs(p, prange, nattr, attrs);
 		if (r) {
 			pr_debug("failed %d to apply attrs\n", r);
+			goto out_unlock;
+		}
+
+		r = svm_range_validate(mm, prange);
+		if (r)
+			pr_debug("failed %d to validate svm range\n", r);
+
+out_unlock:
+		if (r) {
 			mmap_read_unlock(mm);
 			srcu_read_unlock(&prange->svms->srcu, srcu_idx);
 			goto out_remove;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
index c7c54fb73dfb..4d394f72eefc 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
@@ -41,6 +41,7 @@
  * @list:       link list node, used to scan all ranges of svms
  * @update_list:link list node used to add to update_list
  * @remove_list:link list node used to add to remove list
+ * @hmm_range:  hmm range structure used by hmm_range_fault to get system pages
  * @npages:     number of pages
  * @pages_addr: list of system memory physical page address
  * @flags:      flags defined as KFD_IOCTL_SVM_FLAG_*
@@ -61,6 +62,7 @@ struct svm_range {
 	struct list_head		list;
 	struct list_head		update_list;
 	struct list_head		remove_list;
+	struct hmm_range		*hmm_range;
 	uint64_t			npages;
 	dma_addr_t			*pages_addr;
 	uint32_t			flags;
-- 
2.29.2



More information about the amd-gfx mailing list