[ Dropping stable@kernel.org from CC, it'll get picked up for stable once it hits mainline ]
On Don, 2011-10-13 at 16:38 -0400, j.glisse@gmail.com wrote:
From: Jerome Glisse jglisse@redhat.com
After GPU lockup VRAM gart table is unpinned and thus its pointer becomes unvalid. This patch move the unpin code to a common helper function and set pointer to NULL so that page update code can check if it should update GPU page table or not. That way bo still bound to GART can be unbound (pci_unmap_page for all there page) properly while there is no need to update the GPU page table.
Signed-off-by: Jerome Glisse jglisse@redhat.com
[...]
rdev->gart.pages_addr[p] = rdev->dummy_page.addr; page_base = rdev->gart.pages_addr[p]; for (j = 0; j < (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); j++, t++) {
radeon_gart_set_page(rdev, t, page_base);
if (rdev->gart.ptr) {radeon_gart_set_page(rdev, t, page_base);} page_base += RADEON_GPU_PAGE_SIZE; }@@ -201,7 +211,9 @@ int radeon_gart_bind(struct radeon_device *rdev, unsigned offset, rdev->gart.pages[p] = pagelist[i]; page_base = rdev->gart.pages_addr[p]; for (j = 0; j < (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); j++, t++) {
radeon_gart_set_page(rdev, t, page_base);
if (rdev->gart.ptr) {radeon_gart_set_page(rdev, t, page_base);} page_base += RADEON_GPU_PAGE_SIZE;
Looks like these two loops could be completely guarded by if (rdev->gart.ptr).
@@ -218,7 +230,9 @@ void radeon_gart_restore(struct radeon_device *rdev) for (i = 0, t = 0; i < rdev->gart.num_cpu_pages; i++) { page_base = rdev->gart.pages_addr[i]; for (j = 0; j < (PAGE_SIZE / RADEON_GPU_PAGE_SIZE); j++, t++) {
radeon_gart_set_page(rdev, t, page_base);
if (rdev->gart.ptr) {radeon_gart_set_page(rdev, t, page_base);} page_base += RADEON_GPU_PAGE_SIZE;
I think this should rather BUG_ON(!rdev->gart.ptr).
Looks good to me otherwise.