aboutsummaryrefslogtreecommitdiffstats
path: root/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1609-drm-amdgpu-GTT-bind-before-copy-mem.patch
diff options
context:
space:
mode:
Diffstat (limited to 'meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1609-drm-amdgpu-GTT-bind-before-copy-mem.patch')
-rw-r--r--meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1609-drm-amdgpu-GTT-bind-before-copy-mem.patch204
1 files changed, 0 insertions, 204 deletions
diff --git a/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1609-drm-amdgpu-GTT-bind-before-copy-mem.patch b/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1609-drm-amdgpu-GTT-bind-before-copy-mem.patch
deleted file mode 100644
index 99c18c33..00000000
--- a/meta-amd-bsp/recipes-kernel/linux/linux-yocto-4.14.71/1609-drm-amdgpu-GTT-bind-before-copy-mem.patch
+++ /dev/null
@@ -1,204 +0,0 @@
-From 2c91c9a0ea6c34250054d56758c8000b3bbc0080 Mon Sep 17 00:00:00 2001
-From: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
-Date: Tue, 21 Feb 2017 19:34:33 -0500
-Subject: [PATCH 1609/4131] drm/amdgpu: GTT bind before copy mem
-
-This change is required after merging with amd-staging-4.9, since GTT
-binding is done only on demand.
-
-Change-Id: I6ca60c1521ca53a6db0a5c30fae7b9f8b5ca069c
-Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com>
----
- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 131 ++++++++++++++++++-----
- 1 file changed, 106 insertions(+), 25 deletions(-)
-
-diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
-index 801c34e..28dae72 100644
---- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
-+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
-@@ -2279,11 +2279,15 @@ int amdgpu_amdkfd_copy_mem_to_mem(struct kgd_dev *kgd, struct kgd_mem *src_mem,
- {
- struct amdgpu_device *adev = NULL;
- struct ttm_mem_reg *src = NULL, *dst = NULL;
-+ struct ttm_buffer_object *src_ttm_bo, *dst_ttm_bo;
-+ struct drm_mm_node *src_mm, *dst_mm;
- struct amdgpu_ring *ring;
- struct ww_acquire_ctx ticket;
- struct list_head list;
- struct amdgpu_bo_list_entry *entry;
- uint64_t src_start, dst_start;
-+ uint64_t src_left, dst_left, cur_copy_size, total_copy_size = 0;
-+ struct fence *fence = NULL;
- int r;
-
- if (!kgd || !src_mem || !dst_mem)
-@@ -2293,8 +2297,12 @@ int amdgpu_amdkfd_copy_mem_to_mem(struct kgd_dev *kgd, struct kgd_mem *src_mem,
- *actual_size = 0;
-
- adev = get_amdgpu_device(kgd);
-- src = &src_mem->bo->tbo.mem;
-- dst = &dst_mem->bo->tbo.mem;
-+ src_ttm_bo = &src_mem->bo->tbo;
-+ dst_ttm_bo = &dst_mem->bo->tbo;
-+ src = &src_ttm_bo->mem;
-+ dst = &dst_ttm_bo->mem;
-+ src_mm = (struct drm_mm_node *)src->mm_node;
-+ dst_mm = (struct drm_mm_node *)dst->mm_node;
-
- ring = adev->mman.buffer_funcs_ring;
-
-@@ -2304,55 +2312,128 @@ int amdgpu_amdkfd_copy_mem_to_mem(struct kgd_dev *kgd, struct kgd_mem *src_mem,
- entry = &dst_mem->bo_list_entry;
- list_add_tail(&entry->tv.head, &list);
-
-+ if (!ring->ready) {
-+ pr_err("Trying to move memory with ring turned off.\n");
-+ return -EINVAL;
-+ }
-+
- r = ttm_eu_reserve_buffers(&ticket, &list, false, NULL);
- if (r) {
- pr_err("Copy buffer failed. Unable to reserve bo (%d)\n", r);
- return r;
- }
-
-- src_start = (src->start << PAGE_SHIFT) + src_offset;
-- dst_start = (dst->start << PAGE_SHIFT) + dst_offset;
--
- switch (src->mem_type) {
-- case TTM_PL_VRAM:
-- src_start += adev->mc.vram_start;
-- break;
- case TTM_PL_TT:
-- src_start += adev->mc.gtt_start;
-+ r = amdgpu_ttm_bind(src_ttm_bo, src);
-+ if (r) {
-+ DRM_ERROR("Copy failed. Cannot bind to gart\n");
-+ goto copy_fail;
-+ }
-+ break;
-+ case TTM_PL_VRAM:
-+ /* VRAM could be scattered. Find the node in which the offset
-+ * belongs to
-+ */
-+ while (src_offset >= (src_mm->size << PAGE_SHIFT)) {
-+ src_offset -= (src_mm->size << PAGE_SHIFT);
-+ ++src_mm;
-+ }
- break;
- default:
- DRM_ERROR("Unknown placement %d\n", src->mem_type);
- r = -EINVAL;
- goto copy_fail;
- }
-+ src_start = src_mm->start << PAGE_SHIFT;
-+ src_start += src_ttm_bo->bdev->man[src->mem_type].gpu_offset;
-+ src_start += src_offset;
-+ src_left = (src_mm->size << PAGE_SHIFT) - src_offset;
-+
- switch (dst->mem_type) {
-- case TTM_PL_VRAM:
-- dst_start += adev->mc.vram_start;
-- break;
- case TTM_PL_TT:
-- dst_start += adev->mc.gtt_start;
-+ r = amdgpu_ttm_bind(dst_ttm_bo, dst);
-+ if (r) {
-+ DRM_ERROR("Copy failed. Cannot bind to gart\n");
-+ goto copy_fail;
-+ }
-+ break;
-+ case TTM_PL_VRAM:
-+ while (dst_offset >= (dst_mm->size << PAGE_SHIFT)) {
-+ dst_offset -= (dst_mm->size << PAGE_SHIFT);
-+ ++dst_mm;
-+ }
- break;
- default:
- DRM_ERROR("Unknown placement %d\n", dst->mem_type);
- r = -EINVAL;
- goto copy_fail;
- }
-- if (!ring->ready) {
-- pr_err("Trying to move memory with ring turned off.\n");
-- r = -EINVAL;
-- goto copy_fail;
-- }
-+ dst_start = dst_mm->start << PAGE_SHIFT;
-+ dst_start += dst_ttm_bo->bdev->man[dst->mem_type].gpu_offset;
-+ dst_start += dst_offset;
-+ dst_left = (dst_mm->size << PAGE_SHIFT) - dst_offset;
-
-- r = amdgpu_copy_buffer(ring, src_start, dst_start,
-- size, NULL, f, false);
-- if (r)
-- goto copy_fail;
-+ do {
-+ struct fence *next;
-+
-+ /* src_left/dst_left: amount of space left in the current node
-+ * Copy minimum of (src_left, dst_left, amount of bytes left to
-+ * copy)
-+ */
-+ cur_copy_size = min3(src_left, dst_left,
-+ (size - total_copy_size));
-+
-+ r = amdgpu_copy_buffer(ring, src_start, dst_start,
-+ cur_copy_size, NULL, &next, false);
-+ if (r)
-+ break;
-
-+ /* Just keep the last fence */
-+ fence_put(fence);
-+ fence = next;
-+
-+ total_copy_size += cur_copy_size;
-+ /* Required amount of bytes copied. Done. */
-+ if (total_copy_size >= size)
-+ break;
-+
-+ /* If end of src or dst node is reached, move to next node */
-+ src_left -= cur_copy_size;
-+ if (!src_left) {
-+ ++src_mm;
-+ src_start = src_mm->start << PAGE_SHIFT;
-+ src_start +=
-+ src_ttm_bo->bdev->man[src->mem_type].gpu_offset;
-+ src_left = src_mm->size << PAGE_SHIFT;
-+ } else
-+ src_start += cur_copy_size;
-+
-+ dst_left -= cur_copy_size;
-+ if (!dst_left) {
-+ ++dst_mm;
-+ dst_start = dst_mm->start << PAGE_SHIFT;
-+ dst_start +=
-+ dst_ttm_bo->bdev->man[dst->mem_type].gpu_offset;
-+ dst_left = dst_mm->size << PAGE_SHIFT;
-+ } else
-+ dst_start += cur_copy_size;
-+
-+ } while (total_copy_size < size);
-+
-+ /* Failure could occur after partial copy. So fill in amount copied
-+ * and fence, still fill-in
-+ */
- if (actual_size)
-- *actual_size = size;
-+ *actual_size = total_copy_size;
-+
-+ if (fence) {
-+ amdgpu_bo_fence(src_mem->bo, fence, true);
-+ amdgpu_bo_fence(dst_mem->bo, fence, true);
-+ }
-
-- amdgpu_bo_fence(src_mem->bo, *f, true);
-- amdgpu_bo_fence(dst_mem->bo, *f, true);
-+ if (f)
-+ *f = fence;
-
- copy_fail:
- ttm_eu_backoff_reservation(&ticket, &list);
---
-2.7.4
-
generated by cgit 1.2.3-korg (git 2.39.0) at 2024-04-25 03:37:59 +0000