drm/ttm: move memory accounting into vmwgfx v4

This is just another feature which is only used by VMWGFX, so move
it into the driver instead.

I've tried to add the accounting sysfs file to the kobject of the drm
minor, but I'm not 100% sure if this works as expected.

v2: fix typo in KFD and avoid 64bit divide
v3: fix init order in VMWGFX
v4: use pdev sysfs reference instead of drm

Signed-off-by: Christian König <[email protected]>
Reviewed-by: Zack Rusin <[email protected]> (v3)
Tested-by: Nirmoy Das <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

Author: ChristianKoenigAMD

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c

@@ -118,6 +118,16 @@ void amdgpu_amdkfd_gpuvm_init_mem_limits(void)
  */
 #define ESTIMATE_PT_SIZE(mem_size) ((mem_size) >> 14)
 
+static size_t amdgpu_amdkfd_acc_size(uint64_t size)
+{
+	size >>= PAGE_SHIFT;
+	size *= sizeof(dma_addr_t) + sizeof(void *);
+
+	return __roundup_pow_of_two(sizeof(struct amdgpu_bo)) +
+		__roundup_pow_of_two(sizeof(struct ttm_tt)) +
+		PAGE_ALIGN(size);
+}
+
 static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
 		uint64_t size, u32 domain, bool sg)
 {
@@ -126,8 +136,7 @@ static int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
 	size_t acc_size, system_mem_needed, ttm_mem_needed, vram_needed;
 	int ret = 0;
 
-	acc_size = ttm_bo_dma_acc_size(&adev->mman.bdev, size,
-				       sizeof(struct amdgpu_bo));
+	acc_size = amdgpu_amdkfd_acc_size(size);
 
 	vram_needed = 0;
 	if (domain == AMDGPU_GEM_DOMAIN_GTT) {
@@ -174,8 +183,7 @@ static void unreserve_mem_limit(struct amdgpu_device *adev,
 {
 	size_t acc_size;
 
-	acc_size = ttm_bo_dma_acc_size(&adev->mman.bdev, size,
-				       sizeof(struct amdgpu_bo));
+	acc_size = amdgpu_amdkfd_acc_size(size);
 
 	spin_lock(&kfd_mem_limit.mem_limit_lock);
 	if (domain == AMDGPU_GEM_DOMAIN_GTT) {

drivers/gpu/drm/amd/amdgpu/amdgpu_object.c

@@ -523,7 +523,6 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev,
 	};
 	struct amdgpu_bo *bo;
 	unsigned long page_align, size = bp->size;
-	size_t acc_size;
 	int r;
 
 	/* Note that GDS/GWS/OA allocates 1 page per byte/resource. */
@@ -546,9 +545,6 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev,
 
 	*bo_ptr = NULL;
 
-	acc_size = ttm_bo_dma_acc_size(&adev->mman.bdev, size,
-				       sizeof(struct amdgpu_bo));
-
 	bo = kzalloc(sizeof(struct amdgpu_bo), GFP_KERNEL);
 	if (bo == NULL)
 		return -ENOMEM;
@@ -577,8 +573,8 @@ static int amdgpu_bo_do_create(struct amdgpu_device *adev,
 		bo->tbo.priority = 1;
 
 	r = ttm_bo_init_reserved(&adev->mman.bdev, &bo->tbo, size, bp->type,
-				 &bo->placement, page_align, &ctx, acc_size,
-				 NULL, bp->resv, &amdgpu_bo_destroy);
+				 &bo->placement, page_align, &ctx,  NULL,
+				 bp->resv, &amdgpu_bo_destroy);
 	if (unlikely(r != 0))
 		return r;
 

drivers/gpu/drm/drm_gem_vram_helper.c

@@ -189,7 +189,6 @@ struct drm_gem_vram_object *drm_gem_vram_create(struct drm_device *dev,
 	struct drm_vram_mm *vmm = dev->vram_mm;
 	struct ttm_device *bdev;
 	int ret;
-	size_t acc_size;
 
 	if (WARN_ONCE(!vmm, "VRAM MM not initialized"))
 		return ERR_PTR(-EINVAL);
@@ -216,7 +215,6 @@ struct drm_gem_vram_object *drm_gem_vram_create(struct drm_device *dev,
 	}
 
 	bdev = &vmm->bdev;
-	acc_size = ttm_bo_dma_acc_size(bdev, size, sizeof(*gbo));
 
 	gbo->bo.bdev = bdev;
 	drm_gem_vram_placement(gbo, DRM_GEM_VRAM_PL_FLAG_SYSTEM);
@@ -226,8 +224,8 @@ struct drm_gem_vram_object *drm_gem_vram_create(struct drm_device *dev,
 	 * to release gbo->bo.base and kfree gbo.
 	 */
 	ret = ttm_bo_init(bdev, &gbo->bo, size, ttm_bo_type_device,
-			  &gbo->placement, pg_align, false, acc_size,
-			  NULL, NULL, ttm_buffer_object_destroy);
+			  &gbo->placement, pg_align, false, NULL, NULL,
+			  ttm_buffer_object_destroy);
 	if (ret)
 		return ERR_PTR(ret);
 

drivers/gpu/drm/nouveau/nouveau_bo.c

@@ -300,18 +300,15 @@ nouveau_bo_init(struct nouveau_bo *nvbo, u64 size, int align, u32 domain,
 		struct sg_table *sg, struct dma_resv *robj)
 {
 	int type = sg ? ttm_bo_type_sg : ttm_bo_type_device;
-	size_t acc_size;
 	int ret;
 
-	acc_size = ttm_bo_dma_acc_size(nvbo->bo.bdev, size, sizeof(*nvbo));
-
 	nvbo->bo.mem.num_pages = size >> PAGE_SHIFT;
 	nouveau_bo_placement_set(nvbo, domain, 0);
 	INIT_LIST_HEAD(&nvbo->io_reserve_lru);
 
 	ret = ttm_bo_init(nvbo->bo.bdev, &nvbo->bo, size, type,
-			  &nvbo->placement, align >> PAGE_SHIFT, false,
-			  acc_size, sg, robj, nouveau_bo_del_ttm);
+			  &nvbo->placement, align >> PAGE_SHIFT, false, sg,
+			  robj, nouveau_bo_del_ttm);
 	if (ret) {
 		/* ttm will call nouveau_bo_del_ttm if it fails.. */
 		return ret;

drivers/gpu/drm/nouveau/nouveau_drv.h

@@ -54,7 +54,6 @@
 #include <drm/ttm/ttm_bo_api.h>
 #include <drm/ttm/ttm_bo_driver.h>
 #include <drm/ttm/ttm_placement.h>
-#include <drm/ttm/ttm_memory.h>
 
 #include <drm/drm_audio_component.h>
 

drivers/gpu/drm/qxl/qxl_object.c

@@ -138,8 +138,8 @@ int qxl_bo_create(struct qxl_device *qdev,
 	qxl_ttm_placement_from_domain(bo, domain);
 
 	r = ttm_bo_init_reserved(&qdev->mman.bdev, &bo->tbo, size, type,
-				 &bo->placement, 0, &ctx, size,
-				 NULL, NULL, &qxl_ttm_bo_destroy);
+				 &bo->placement, 0, &ctx, NULL, NULL,
+				 &qxl_ttm_bo_destroy);
 	if (unlikely(r != 0)) {
 		if (r != -ERESTARTSYS)
 			dev_err(qdev->ddev.dev,

drivers/gpu/drm/radeon/radeon_object.c

@@ -159,7 +159,6 @@ int radeon_bo_create(struct radeon_device *rdev,
 	struct radeon_bo *bo;
 	enum ttm_bo_type type;
 	unsigned long page_align = roundup(byte_align, PAGE_SIZE) >> PAGE_SHIFT;
-	size_t acc_size;
 	int r;
 
 	size = ALIGN(size, PAGE_SIZE);
@@ -173,9 +172,6 @@ int radeon_bo_create(struct radeon_device *rdev,
 	}
 	*bo_ptr = NULL;
 
-	acc_size = ttm_bo_dma_acc_size(&rdev->mman.bdev, size,
-				       sizeof(struct radeon_bo));
-
 	bo = kzalloc(sizeof(struct radeon_bo), GFP_KERNEL);
 	if (bo == NULL)
 		return -ENOMEM;
@@ -230,8 +226,8 @@ int radeon_bo_create(struct radeon_device *rdev,
 	/* Kernel allocation are uninterruptible */
 	down_read(&rdev->pm.mclk_lock);
 	r = ttm_bo_init(&rdev->mman.bdev, &bo->tbo, size, type,
-			&bo->placement, page_align, !kernel, acc_size,
-			sg, resv, &radeon_ttm_bo_destroy);
+			&bo->placement, page_align, !kernel, sg, resv,
+			&radeon_ttm_bo_destroy);
 	up_read(&rdev->pm.mclk_lock);
 	if (unlikely(r != 0)) {
 		return r;

drivers/gpu/drm/ttm/Makefile

@@ -2,10 +2,9 @@
 #
 # Makefile for the drm device driver.  This driver provides support for the
 
-ttm-y := ttm_memory.o ttm_tt.o ttm_bo.o \
-	ttm_bo_util.o ttm_bo_vm.o ttm_module.o \
-	ttm_execbuf_util.o ttm_range_manager.o \
-	ttm_resource.o ttm_pool.o ttm_device.o
+ttm-y := ttm_tt.o ttm_bo.o ttm_bo_util.o ttm_bo_vm.o ttm_module.o \
+	ttm_execbuf_util.o ttm_range_manager.o ttm_resource.o ttm_pool.o \
+	ttm_device.o
 ttm-$(CONFIG_AGP) += ttm_agp_backend.o
 
 obj-$(CONFIG_DRM_TTM) += ttm.o

drivers/gpu/drm/ttm/ttm_bo.c

@@ -425,7 +425,6 @@ static void ttm_bo_release(struct kref *kref)
 	struct ttm_buffer_object *bo =
 	    container_of(kref, struct ttm_buffer_object, kref);
 	struct ttm_device *bdev = bo->bdev;
-	size_t acc_size = bo->acc_size;
 	int ret;
 
 	if (!bo->deleted) {
@@ -485,7 +484,6 @@ static void ttm_bo_release(struct kref *kref)
 	if (!ttm_bo_uses_embedded_gem_object(bo))
 		dma_resv_fini(&bo->base._resv);
 	bo->destroy(bo);
-	ttm_mem_global_free(&ttm_mem_glob, acc_size);
 }
 
 void ttm_bo_put(struct ttm_buffer_object *bo)
@@ -1046,25 +1044,13 @@ int ttm_bo_init_reserved(struct ttm_device *bdev,
 			 struct ttm_placement *placement,
 			 uint32_t page_alignment,
 			 struct ttm_operation_ctx *ctx,
-			 size_t acc_size,
 			 struct sg_table *sg,
 			 struct dma_resv *resv,
 			 void (*destroy) (struct ttm_buffer_object *))
 {
-	struct ttm_mem_global *mem_glob = &ttm_mem_glob;
 	bool locked;
 	int ret = 0;
 
-	ret = ttm_mem_global_alloc(mem_glob, acc_size, ctx);
-	if (ret) {
-		pr_err("Out of kernel memory\n");
-		if (destroy)
-			(*destroy)(bo);
-		else
-			kfree(bo);
-		return -ENOMEM;
-	}
-
 	bo->destroy = destroy ? destroy : ttm_bo_default_destroy;
 
 	kref_init(&bo->kref);
@@ -1081,7 +1067,6 @@ int ttm_bo_init_reserved(struct ttm_device *bdev,
 	bo->mem.bus.addr = NULL;
 	bo->moving = NULL;
 	bo->mem.placement = 0;
-	bo->acc_size = acc_size;
 	bo->pin_count = 0;
 	bo->sg = sg;
 	if (resv) {
@@ -1142,7 +1127,6 @@ int ttm_bo_init(struct ttm_device *bdev,
 		struct ttm_placement *placement,
 		uint32_t page_alignment,
 		bool interruptible,
-		size_t acc_size,
 		struct sg_table *sg,
 		struct dma_resv *resv,
 		void (*destroy) (struct ttm_buffer_object *))
@@ -1151,8 +1135,7 @@ int ttm_bo_init(struct ttm_device *bdev,
 	int ret;
 
 	ret = ttm_bo_init_reserved(bdev, bo, size, type, placement,
-				   page_alignment, &ctx, acc_size,
-				   sg, resv, destroy);
+				   page_alignment, &ctx, sg, resv, destroy);
 	if (ret)
 		return ret;
 
@@ -1163,20 +1146,6 @@ int ttm_bo_init(struct ttm_device *bdev,
 }
 EXPORT_SYMBOL(ttm_bo_init);
 
-size_t ttm_bo_dma_acc_size(struct ttm_device *bdev,
-			   unsigned long bo_size,
-			   unsigned struct_size)
-{
-	unsigned npages = (PAGE_ALIGN(bo_size)) >> PAGE_SHIFT;
-	size_t size = 0;
-
-	size += ttm_round_pot(struct_size);
-	size += ttm_round_pot(npages * (2*sizeof(void *) + sizeof(dma_addr_t)));
-	size += ttm_round_pot(sizeof(struct ttm_tt));
-	return size;
-}
-EXPORT_SYMBOL(ttm_bo_dma_acc_size);
-
 /*
  * buffer object vm functions.
  */

drivers/gpu/drm/ttm/ttm_bo_util.c

@@ -309,7 +309,6 @@ static int ttm_buffer_object_transfer(struct ttm_buffer_object *bo,
 
 	kref_init(&fbo->base.kref);
 	fbo->base.destroy = &ttm_transfered_destroy;
-	fbo->base.acc_size = 0;
 	fbo->base.pin_count = 0;
 	if (bo->type != ttm_bo_type_sg)
 		fbo->base.base.resv = &fbo->base.base._resv;

drivers/gpu/drm/ttm/ttm_device.c

@@ -27,9 +27,12 @@
 
 #define pr_fmt(fmt) "[TTM DEVICE] " fmt
 
+#include <linux/mm.h>
+
 #include <drm/ttm/ttm_device.h>
-#include <drm/ttm/ttm_memory.h>
+#include <drm/ttm/ttm_tt.h>
 #include <drm/ttm/ttm_placement.h>
+#include <drm/ttm/ttm_bo_api.h>
 
 #include "ttm_module.h"
 
@@ -49,7 +52,9 @@ static void ttm_global_release(void)
 	if (--ttm_glob_use_count > 0)
 		goto out;
 
-	ttm_mem_global_release(&ttm_mem_glob);
+	ttm_pool_mgr_fini();
+	ttm_tt_mgr_fini();
+
 	__free_page(glob->dummy_read_page);
 	memset(glob, 0, sizeof(*glob));
 out:
@@ -59,16 +64,23 @@ static void ttm_global_release(void)
 static int ttm_global_init(void)
 {
 	struct ttm_global *glob = &ttm_glob;
+	unsigned long num_pages;
+	struct sysinfo si;
 	int ret = 0;
 	unsigned i;
 
 	mutex_lock(&ttm_global_mutex);
 	if (++ttm_glob_use_count > 1)
 		goto out;
 
-	ret = ttm_mem_global_init(&ttm_mem_glob);
-	if (ret)
-		goto out;
+	si_meminfo(&si);
+
+	/* Limit the number of pages in the pool to about 50% of the total
+	 * system memory.
+	 */
+	num_pages = ((u64)si.totalram * si.mem_unit) >> PAGE_SHIFT;
+	ttm_pool_mgr_init(num_pages * 50 / 100);
+	ttm_tt_mgr_init();
 
 	spin_lock_init(&glob->lru_lock);
 	glob->dummy_read_page = alloc_page(__GFP_ZERO | GFP_DMA32);

drivers/gpu/drm/ttm/ttm_pool.c

@@ -404,16 +404,10 @@ int ttm_pool_alloc(struct ttm_pool *pool, struct ttm_tt *tt,
 			caching = pages + (1 << order);
 		}
 
-		r = ttm_mem_global_alloc_page(&ttm_mem_glob, p,
-					      (1 << order) * PAGE_SIZE,
-					      ctx);
-		if (r)
-			goto error_free_page;
-
 		if (dma_addr) {
 			r = ttm_pool_map(pool, order, p, &dma_addr);
 			if (r)
-				goto error_global_free;
+				goto error_free_page;
 		}
 
 		num_pages -= 1 << order;
@@ -427,9 +421,6 @@ int ttm_pool_alloc(struct ttm_pool *pool, struct ttm_tt *tt,
 
 	return 0;
 
-error_global_free:
-	ttm_mem_global_free_page(&ttm_mem_glob, p, (1 << order) * PAGE_SIZE);
-
 error_free_page:
 	ttm_pool_free_page(pool, tt->caching, order, p);
 
@@ -464,8 +455,6 @@ void ttm_pool_free(struct ttm_pool *pool, struct ttm_tt *tt)
 
 		order = ttm_pool_page_order(pool, p);
 		num_pages = 1ULL << order;
-		ttm_mem_global_free_page(&ttm_mem_glob, p,
-					 num_pages * PAGE_SIZE);
 		if (tt->dma_address)
 			ttm_pool_unmap(pool, tt->dma_address[i], num_pages);
 

drivers/gpu/drm/vmwgfx/Makefile

@@ -9,7 +9,7 @@ vmwgfx-y := vmwgfx_execbuf.o vmwgfx_gmr.o vmwgfx_kms.o vmwgfx_drv.o \
 	    vmwgfx_cotable.o vmwgfx_so.o vmwgfx_binding.o vmwgfx_msg.o \
 	    vmwgfx_simple_resource.o vmwgfx_va.o vmwgfx_blit.o \
 	    vmwgfx_validation.o vmwgfx_page_dirty.o vmwgfx_streamoutput.o \
-	    ttm_object.o ttm_lock.o
+	    ttm_object.o ttm_lock.o ttm_memory.o
 
 vmwgfx-$(CONFIG_TRANSPARENT_HUGEPAGE) += vmwgfx_thp.o
 obj-$(CONFIG_DRM_VMWGFX) := vmwgfx.o