[PATCH 4/5] drm/ttm: add bit flag TTM_OPT_FLAG_ALLOW_ALLOC_ANYWAY

Roger He Hongbo.He at amd.com
Tue Feb 6 09:04:33 UTC 2018


set TTM_OPT_FLAG_ALLOW_ALLOC_ANYWAY when we are servicing for page
fault routine.

for ttm_mem_global_reserve if in page fault routine, allow the gtt
pages reservation always. because page fault routing already grabbed
system memory and the allowance of this exception is harmless.
Otherwise, it will trigger OOM killer.

v2: keep original behavior except ttm bo with flag no_retry

Signed-off-by: Roger He <Hongbo.He at amd.com>
---
 drivers/gpu/drm/ttm/ttm_bo_vm.c          |  6 ++++--
 drivers/gpu/drm/ttm/ttm_memory.c         | 25 +++++++++++++++++--------
 drivers/gpu/drm/ttm/ttm_page_alloc_dma.c |  1 -
 include/drm/ttm/ttm_bo_api.h             |  4 +++-
 4 files changed, 24 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
index 716e724..240c462 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
@@ -224,7 +224,7 @@ static int ttm_bo_vm_fault(struct vm_fault *vmf)
 		cvma.vm_page_prot = ttm_io_prot(bo->mem.placement,
 						cvma.vm_page_prot);
 	} else {
-		struct ttm_operation_ctx ctx = {
+		struct ttm_operation_ctx ttm_opt_ctx = {
 			.interruptible = false,
 			.no_wait_gpu = false
 		};
@@ -233,8 +233,10 @@ static int ttm_bo_vm_fault(struct vm_fault *vmf)
 		cvma.vm_page_prot = ttm_io_prot(bo->mem.placement,
 						cvma.vm_page_prot);
 
+		if (ttm->page_flags & TTM_PAGE_FLAG_NO_RETRY)
+			ttm_opt_ctx.flags |= TTM_OPT_FLAG_ALLOW_ALLOC_ANYWAY;
 		/* Allocate all page at once, most common usage */
-		if (ttm->bdev->driver->ttm_tt_populate(ttm, &ctx)) {
+		if (ttm->bdev->driver->ttm_tt_populate(ttm, &ttm_opt_ctx)) {
 			ret = VM_FAULT_OOM;
 			goto out_io_unlock;
 		}
diff --git a/drivers/gpu/drm/ttm/ttm_memory.c b/drivers/gpu/drm/ttm/ttm_memory.c
index 3f00ed8..5e357b4 100644
--- a/drivers/gpu/drm/ttm/ttm_memory.c
+++ b/drivers/gpu/drm/ttm/ttm_memory.c
@@ -488,9 +488,6 @@ ttm_check_over_swaplimit(struct ttm_mem_global *glob)
 {
 	bool ret = false;
 
-	if (!glob->no_retry)
-		return ret;
-
 	if (get_nr_swap_pages() < FREE_SWAP_SPACE
 	    && si_mem_available() < glob->sys_mem_limit)
 		ret = true;
@@ -500,17 +497,28 @@ ttm_check_over_swaplimit(struct ttm_mem_global *glob)
 
 static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
 				  struct ttm_mem_zone *single_zone,
-				  uint64_t amount, bool reserve)
+				  uint64_t amount, bool reserve,
+				  bool allow_alloc_anyway)
 {
 	uint64_t limit;
 	int ret = -ENOMEM;
 	unsigned int i;
 	struct ttm_mem_zone *zone;
 
-	if (ttm_check_over_swaplimit(glob))
+	if (glob->no_retry && !allow_alloc_anyway
+	    && ttm_check_over_swaplimit(glob))
 		return ret;
 
 	spin_lock(&glob->lock);
+	/*
+	 * to cover two special cases:
+	 * a. if serving page_fault allow reservation anyway since
+	 * it already allocated system pages. Otherwise it will trigger OOM.
+	 * b. if serving suspend, allow reservation anyway as well.
+	 */
+	if (glob->no_retry && allow_alloc_anyway)
+		goto reserve_direct;
+
 	for (i = 0; i < glob->num_zones; ++i) {
 		zone = glob->zones[i];
 		if (single_zone && zone != single_zone)
@@ -523,6 +531,7 @@ static int ttm_mem_global_reserve(struct ttm_mem_global *glob,
 			goto out_unlock;
 	}
 
+reserve_direct:
 	if (reserve) {
 		for (i = 0; i < glob->num_zones; ++i) {
 			zone = glob->zones[i];
@@ -547,10 +556,10 @@ static int ttm_mem_global_alloc_zone(struct ttm_mem_global *glob,
 				     struct ttm_operation_ctx *ctx)
 {
 	int count = TTM_MEMORY_ALLOC_RETRIES;
+	bool alloc_anyway = ctx->flags & TTM_OPT_FLAG_ALLOW_ALLOC_ANYWAY;
 
-	while (unlikely(ttm_mem_global_reserve(glob,
-					       single_zone,
-					       memory, true)
+	while (unlikely(ttm_mem_global_reserve(glob, single_zone, memory,
+					       true, alloc_anyway)
 			!= 0)) {
 		if (ctx->no_wait_gpu)
 			return -ENOMEM;
diff --git a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
index b122f6e..354e0e1 100644
--- a/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
+++ b/drivers/gpu/drm/ttm/ttm_page_alloc_dma.c
@@ -944,7 +944,6 @@ int ttm_dma_populate(struct ttm_dma_tt *ttm_dma, struct device *dev,
 	i = 0;
 
 	type = ttm_to_type(ttm->page_flags, ttm->caching_state);
-
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 	if (ttm->page_flags & TTM_PAGE_FLAG_DMA32)
 		goto skip_huge;
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index 872ff6c..f7304c5 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -278,7 +278,9 @@ struct ttm_operation_ctx {
 };
 
 /* Allow eviction of reserved BOs */
-#define TTM_OPT_FLAG_ALLOW_RES_EVICT	0x1
+#define TTM_OPT_FLAG_ALLOW_RES_EVICT		0x1
+/* when serving page fault, allow alloc anyway */
+#define TTM_OPT_FLAG_ALLOW_ALLOC_ANYWAY		0x2
 
 /**
  * ttm_bo_reference - reference a struct ttm_buffer_object
-- 
2.7.4



More information about the amd-gfx mailing list