drm/amdgpu: Set TTM pools for memory partitions

For native mode only, create TTM pool for each memory partition to store
the NUMA node id, then the TTM pool will be selected using memory
partition id to allocate memory from the correct partition.

Acked-by: Christian König <christian.koenig@amd.com>
(rajneesh: changed need_swiotlb and need_dma32 to false for pool init)
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-and-tested-by:  Mukul Joshi <mukul.joshi@amd.com>
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index c8a2d030f226c85b7f121c24e1c003c2db4e276f..7674109810b0601d58c05ef5ee8ca9fcb6489c27 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -631,6 +631,7 @@ struct amdgpu_ttm_tt {
        struct task_struct      *usertask;
        uint32_t                userflags;
        bool                    bound;
+       int32_t                 pool_id;
 };
 
 #define ttm_to_amdgpu_ttm_tt(ptr)      container_of(ptr, struct amdgpu_ttm_tt, ttm)
@@ -1059,6 +1060,7 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
                return NULL;
        }
        gtt->gobj = &bo->base;
+       gtt->pool_id = NUMA_NO_NODE;
 
        if (abo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
                caching = ttm_write_combined;
@@ -1085,6 +1087,7 @@ static int amdgpu_ttm_tt_populate(struct ttm_device *bdev,
 {
        struct amdgpu_device *adev = amdgpu_ttm_adev(bdev);
        struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
+       struct ttm_pool *pool;
        pgoff_t i;
        int ret;
 
@@ -1099,7 +1102,11 @@ static int amdgpu_ttm_tt_populate(struct ttm_device *bdev,
        if (ttm->page_flags & TTM_TT_FLAG_EXTERNAL)
                return 0;
 
-       ret = ttm_pool_alloc(&adev->mman.bdev.pool, ttm, ctx);
+       if (adev->mman.ttm_pools && gtt->pool_id >= 0)
+               pool = &adev->mman.ttm_pools[gtt->pool_id];
+       else
+               pool = &adev->mman.bdev.pool;
+       ret = ttm_pool_alloc(pool, ttm, ctx);
        if (ret)
                return ret;
 
@@ -1120,6 +1127,7 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_device *bdev,
 {
        struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
        struct amdgpu_device *adev;
+       struct ttm_pool *pool;
        pgoff_t i;
 
        amdgpu_ttm_backend_unbind(bdev, ttm);
@@ -1138,7 +1146,13 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_device *bdev,
                ttm->pages[i]->mapping = NULL;
 
        adev = amdgpu_ttm_adev(bdev);
-       return ttm_pool_free(&adev->mman.bdev.pool, ttm);
+
+       if (adev->mman.ttm_pools && gtt->pool_id >= 0)
+               pool = &adev->mman.ttm_pools[gtt->pool_id];
+       else
+               pool = &adev->mman.bdev.pool;
+
+       return ttm_pool_free(pool, ttm);
 }
 
 /**
@@ -1728,6 +1742,41 @@ static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev)
        return 0;
 }
 
+static int amdgpu_ttm_pools_init(struct amdgpu_device *adev)
+{
+       int i;
+
+       if (!adev->gmc.is_app_apu || !adev->gmc.num_mem_partitions)
+               return 0;
+
+       adev->mman.ttm_pools = kcalloc(adev->gmc.num_mem_partitions,
+                                      sizeof(*adev->mman.ttm_pools),
+                                      GFP_KERNEL);
+       if (!adev->mman.ttm_pools)
+               return -ENOMEM;
+
+       for (i = 0; i < adev->gmc.num_mem_partitions; i++) {
+               ttm_pool_init(&adev->mman.ttm_pools[i], adev->dev,
+                             adev->gmc.mem_partitions[i].numa.node,
+                             false, false);
+       }
+       return 0;
+}
+
+static void amdgpu_ttm_pools_fini(struct amdgpu_device *adev)
+{
+       int i;
+
+       if (!adev->gmc.is_app_apu || !adev->mman.ttm_pools)
+               return;
+
+       for (i = 0; i < adev->gmc.num_mem_partitions; i++)
+               ttm_pool_fini(&adev->mman.ttm_pools[i]);
+
+       kfree(adev->mman.ttm_pools);
+       adev->mman.ttm_pools = NULL;
+}
+
 /*
  * amdgpu_ttm_init - Init the memory management (ttm) as well as various
  * gtt/vram related fields.
@@ -1754,6 +1803,12 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
                DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
                return r;
        }
+
+       r = amdgpu_ttm_pools_init(adev);
+       if (r) {
+               DRM_ERROR("failed to init ttm pools(%d).\n", r);
+               return r;
+       }
        adev->mman.initialized = true;
 
        /* Initialize VRAM pool with all of VRAM divided into pages */
@@ -1901,6 +1956,8 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev)
        if (!adev->mman.initialized)
                return;
 
+       amdgpu_ttm_pools_fini(adev);
+
        amdgpu_ttm_training_reserve_vram_fini(adev);
        /* return the stolen vga memory back to VRAM */
        if (!adev->gmc.is_app_apu) {

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index da6544fdc8ddd354d2a7fbf8a3f0c803d789a2b4..8ef048a0a33e7c2ed8d82b9b0fb202d0ffd371e1 100644 (file)
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -49,6 +49,7 @@ struct amdgpu_gtt_mgr {
 
 struct amdgpu_mman {
        struct ttm_device               bdev;
+       struct ttm_pool                 *ttm_pools;
        bool                            initialized;
        void __iomem                    *aper_base_kaddr;