diff options
| author | Honglei Huang <honghuan@amd.com> | 2026-07-01 14:27:58 +0800 |
|---|---|---|
| committer | Matthew Brost <matthew.brost@intel.com> | 2026-07-14 09:27:30 -0700 |
| commit | 0bc7c196f3ab855da0879283ba633c4c51ddfd81 (patch) | |
| tree | 09442977a2341569dc893f473575bab3c633277c /drivers/gpu | |
| parent | d1b894c5bbb3fee0012bd14356286dc2384e8213 (diff) | |
drm/gpusvm: free the whole IOVA reservation on unmap
dma_iova_try_alloc() reserves IOVA for the entire range, but in a mixed
range only the system pages are linked (their total size is state_offset)
while device pages never touch the IOVA state. dma_iova_destroy() with
state_offset only frees the linked part, permanently leaking the IOVA
reserved for the device pages and eventually exhausting the IOVA space.
Unlink the linked system-page portion and free the whole reserved IOVA
instead. On the get_pages() error path state_offset is 0 (no page linked,
dma_addr[0] unpopulated), so skip the unlink and just free the reservation;
this also avoids reading the uninitialized dma_addr[0].dir there.
Allocate the dma_addr array with the zeroing kvzalloc_objs() so every entry
has a well-defined value.
This issue was found by Sashiko AI review.
Fixes: 37ad039fb367 ("drm/gpusvm: Use dma-map IOVA alloc, link, and sync API in GPU SVM")
Cc: stable@vger.kernel.org
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Honglei Huang <honghuan@amd.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patch.msgid.link/20260701062800.409248-2-honghuan@amd.com
Diffstat (limited to 'drivers/gpu')
| -rw-r--r-- | drivers/gpu/drm/drm_gpusvm.c | 19 |
1 files changed, 14 insertions, 5 deletions
diff --git a/drivers/gpu/drm/drm_gpusvm.c b/drivers/gpu/drm/drm_gpusvm.c index 958cb605aedd..3145d55cd860 100644 --- a/drivers/gpu/drm/drm_gpusvm.c +++ b/drivers/gpu/drm/drm_gpusvm.c @@ -1146,10 +1146,19 @@ static void __drm_gpusvm_unmap_pages(struct drm_gpusvm *gpusvm, }; bool use_iova = dma_use_iova(&svm_pages->state); - if (use_iova) - dma_iova_destroy(dev, &svm_pages->state, - svm_pages->state_offset, - svm_pages->dma_addr[0].dir, 0); + /* + * IOVA is reserved for the whole range but only the linked + * system pages (state_offset bytes) need unlinking; free the + * entire reservation to avoid leaking the device-page part. + * On the error path state_offset is 0, so just free it. + */ + if (use_iova) { + if (svm_pages->state_offset) + dma_iova_unlink(dev, &svm_pages->state, 0, + svm_pages->state_offset, + svm_pages->dma_addr[0].dir, 0); + dma_iova_free(dev, &svm_pages->state); + } for (i = 0, j = 0; i < npages; j++) { struct drm_pagemap_addr *addr = &svm_pages->dma_addr[j]; @@ -1486,7 +1495,7 @@ map_pages: /* Unlock and restart mapping to allocate memory. */ drm_gpusvm_notifier_unlock(gpusvm); svm_pages->dma_addr = - kvmalloc_objs(*svm_pages->dma_addr, npages); + kvzalloc_objs(*svm_pages->dma_addr, npages); if (!svm_pages->dma_addr) { err = -ENOMEM; goto err_free; |