<html><head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"> </head> <body> <div class="moz-cite-prefix">On 2025-01-02 19:06, Emily Deng wrote: </div> <blockquote type="cite" cite="mid:20250103000644.1398643-1-Emily.Deng@amd.com"> <pre class="moz-quote-pre" wrap="">For partial migrate from ram to vram, the migrate->cpages is not equal to migrate->npages, should use migrate->npages to check all needed migrate pages which could be copied or not. And only need to set those pages could be migrated to migrate->dst[i], or the migrate_vma_pages will migrate the wrong pages based on the migrate->dst[i]. Signed-off-by: Emily Deng <a class="moz-txt-link-rfc2396E" href="mailto:Emily.Deng@amd.com"><Emily.Deng@amd.com></a> --- drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index 4b275937d05e..5c96c2d425e3 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -278,7 +278,7 @@ svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, struct migrate_vma *migrate, struct dma_fence **mfence, dma_addr_t *scratch, uint64_t ttm_res_offset) { - uint64_t npages = migrate->cpages; + uint64_t npages = migrate->npages;</pre> </blockquote> As partial migration size is based on prange granularity, by default 2MB, maybe always migrate->cpages equal to migrate->npages, that's why we didn't trigger this bug. Wondering how do you catch this bug? This bug will leak svm_bo too, as svm_migrate_get_vram_page reference counter is incorrect. <blockquote type="cite" cite="mid:20250103000644.1398643-1-Emily.Deng@amd.com"> <pre class="moz-quote-pre" wrap=""> struct amdgpu_device *adev = node->adev; struct device *dev = adev->dev; struct amdgpu_res_cursor cursor; @@ -299,9 +299,6 @@ svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, struct page *spage; dst[i] = cursor.start + (j << PAGE_SHIFT); - migrate->dst[i] = svm_migrate_addr_to_pfn(adev, dst[i]); - svm_migrate_get_vram_page(prange, migrate->dst[i]); - migrate->dst[i] = migrate_pfn(migrate->dst[i]); spage = migrate_pfn_to_page(migrate->src[i]); if (spage && !is_zone_device_page(spage)) { @@ -345,6 +342,9 @@ svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, } else { j++; } + migrate->dst[i] = svm_migrate_addr_to_pfn(adev, dst[i]); + svm_migrate_get_vram_page(prange, migrate->dst[i]); + migrate->dst[i] = migrate_pfn(migrate->dst[i]);</pre> </blockquote> This should move forward, to handle the corner case to migrate 1 page to the last page of VRAM res cursor. Please check this change, to add mpages accounting to break the loop earlier. - uint64_t npages = migrate->cpages; + uint64_t npages = migrate->npages; struct amdgpu_device *adev = node->adev; struct device *dev = adev->dev; struct amdgpu_res_cursor cursor; + uint64_t mpages = 0; dma_addr_t *src; uint64_t *dst; uint64_t i, j; @@ -295,14 +296,9 @@ svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, amdgpu_res_first(prange->ttm_res, ttm_res_offset, npages << PAGE_SHIFT, &cursor); - for (i = j = 0; i < npages; i++) { + for (i = j = 0; i < npages && mpages < migrate->cpages; i++) { struct page *spage; - dst[i] = cursor.start + (j << PAGE_SHIFT); - migrate->dst[i] = svm_migrate_addr_to_pfn(adev, dst[i]); - svm_migrate_get_vram_page(prange, migrate->dst[i]); - migrate->dst[i] = migrate_pfn(migrate->dst[i]); - spage = migrate_pfn_to_page(migrate->src[i]); if (spage && !is_zone_device_page(spage)) { src[i] = dma_map_page(dev, spage, 0, PAGE_SIZE, @@ -322,6 +318,7 @@ svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, mfence); if (r) goto out_free_vram_pages; + mpages += j; amdgpu_res_next(&cursor, (j + 1) << PAGE_SHIFT); j = 0; } else { @@ -333,6 +330,11 @@ svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, pr_debug_ratelimited("dma mapping src to 0x%llx, pfn 0x%lx\n", src[i] >> PAGE_SHIFT, page_to_pfn(spage)); + dst[i] = cursor.start + (j << PAGE_SHIFT); + migrate->dst[i] = svm_migrate_addr_to_pfn(adev, dst[i]); + svm_migrate_get_vram_page(prange, migrate->dst[i]); + migrate->dst[i] = migrate_pfn(migrate->dst[i]); + if (j >= (cursor.size >> PAGE_SHIFT) - 1 && i < npages - 1) { r = svm_migrate_copy_memory_gart(adev, src + i - j, dst + i - j, j + 1, @@ -340,6 +342,7 @@ svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, mfence); if (r) goto out_free_vram_pages; + mpages += j + 1; amdgpu_res_next(&cursor, (j + 1) * PAGE_SIZE); j = 0; } else { (END) @@ -322,6 +318,7 @@ svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, mfence); if (r) goto out_free_vram_pages; + mpages += j; amdgpu_res_next(&cursor, (j + 1) << PAGE_SHIFT); j = 0; } else { @@ -333,6 +330,11 @@ svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, pr_debug_ratelimited("dma mapping src to 0x%llx, pfn 0x%lx\n", src[i] >> PAGE_SHIFT, page_to_pfn(spage)); + dst[i] = cursor.start + (j << PAGE_SHIFT); + migrate->dst[i] = svm_migrate_addr_to_pfn(adev, dst[i]); + svm_migrate_get_vram_page(prange, migrate->dst[i]); + migrate->dst[i] = migrate_pfn(migrate->dst[i]); + if (j >= (cursor.size >> PAGE_SHIFT) - 1 && i < npages - 1) { r = svm_migrate_copy_memory_gart(adev, src + i - j, dst + i - j, j + 1, @@ -340,6 +342,7 @@ svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, mfence); if (r) goto out_free_vram_pages; + mpages += j + 1; amdgpu_res_next(&cursor, (j + 1) * PAGE_SIZE); j = 0; } else { <blockquote type="cite" cite="mid:20250103000644.1398643-1-Emily.Deng@amd.com"> <pre class="moz-quote-pre" wrap=""> } r = svm_migrate_copy_memory_gart(adev, src + i - j, dst + i - j, j, </pre> </blockquote> </body> </html>