drm/amdkfd: Set svm range max pages

This will be used to split giant svm range into smaller ranges, to
support VRAM overcommitment by giant range and improve GPU retry fault
recover on giant range.

Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
index eecb262..373e5bf 100644 (file)
--- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c
@@ -1019,6 +1019,8 @@ int svm_migrate_init(struct amdgpu_device *adev)
 
        amdgpu_amdkfd_reserve_system_mem(SVM_HMM_PAGE_STRUCT_SIZE(size));
 
+       svm_range_set_max_pages(adev);
+
        pr_info("HMM registered %ldMB device memory\n", size >> 20);
 
        return 0;

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index b0c1c71..041342f 100644 (file)
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -46,6 +46,12 @@
  */
 #define AMDGPU_SVM_RANGE_RETRY_FAULT_PENDING   (2UL * NSEC_PER_MSEC)
 
+/* Giant svm range split into smaller ranges based on this, it is decided using
+ * minimum of all dGPU/APU 1/32 VRAM size, between 2MB to 1GB and alignment to
+ * power of 2MB.
+ */
+static uint64_t max_svm_range_pages;
+
 struct criu_svm_metadata {
        struct list_head list;
        struct kfd_criu_svm_range_priv_data data;
@@ -1870,6 +1876,21 @@ static struct svm_range *svm_range_clone(struct svm_range *old)
        return new;
 }
 
+void svm_range_set_max_pages(struct amdgpu_device *adev)
+{
+       uint64_t max_pages;
+       uint64_t pages, _pages;
+
+       /* 1/32 VRAM size in pages */
+       pages = adev->gmc.real_vram_size >> 17;
+       pages = clamp(pages, 1ULL << 9, 1ULL << 18);
+       pages = rounddown_pow_of_two(pages);
+       do {
+               max_pages = READ_ONCE(max_svm_range_pages);
+               _pages = min_not_zero(max_pages, pages);
+       } while (cmpxchg(&max_svm_range_pages, max_pages, _pages) != max_pages);
+}
+
 /**
  * svm_range_add - add svm range and handle overlap
  * @p: the range add to this process svms

diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
index eab7f6d..9156b04 100644 (file)
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h
@@ -204,6 +204,9 @@ void svm_range_list_lock_and_flush_work(struct svm_range_list *svms, struct mm_s
 #define KFD_IS_SVM_API_SUPPORTED(dev) ((dev)->pgmap.type != 0)
 
 void svm_range_bo_unref_async(struct svm_range_bo *svm_bo);
+
+void svm_range_set_max_pages(struct amdgpu_device *adev);
+
 #else
 
 struct kfd_process;