[PATCH 132/159] drm/amdgpu: Add function to allocate and fill PDB0

[Alex Deucher]

From: Oak Zeng <Oak.Zeng at amd.com>

Add functions to allocate PDB0, map it for CPU access,
and fill it.

Those functions are only used for 2-level vmid0 page
table construction

Signed-off-by: Oak Zeng <Oak.Zeng at amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling at amd.com>
Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c | 103 ++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h |   5 ++
 2 files changed, 108 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
index f492ff8facd3..2be31ad8e31c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c
@@ -31,6 +31,57 @@
 #include "amdgpu_ras.h"
 #include "amdgpu_xgmi.h"
 
+/**
+ * amdgpu_gmc_pdb0_alloc - allocate vram for pdb0
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * Allocate video memory for pdb0 and map it for CPU access
+ * Returns 0 for success, error for failure.
+ */
+int amdgpu_gmc_pdb0_alloc(struct amdgpu_device *adev)
+{
+	int r;
+	struct amdgpu_bo_param bp;
+	u64 vram_size = adev->gmc.xgmi.node_segment_size * adev->gmc.xgmi.num_physical_nodes;
+	uint32_t pde0_page_shift = adev->gmc.vmid0_page_table_block_size + 21;
+	uint32_t npdes = (vram_size + (1ULL << pde0_page_shift) -1) >> pde0_page_shift;
+
+	memset(&bp, 0, sizeof(bp));
+	bp.size = PAGE_ALIGN((npdes + 1) * 8);
+	bp.byte_align = PAGE_SIZE;
+	bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+	bp.flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+		AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+	bp.type = ttm_bo_type_kernel;
+	bp.resv = NULL;
+	r = amdgpu_bo_create(adev, &bp, &adev->gmc.pdb0_bo);
+	if (r)
+		return r;
+
+	r = amdgpu_bo_reserve(adev->gmc.pdb0_bo, false);
+	if (unlikely(r != 0))
+		goto bo_reserve_failure;
+
+	r = amdgpu_bo_pin(adev->gmc.pdb0_bo, AMDGPU_GEM_DOMAIN_VRAM);
+	if (r)
+		goto bo_pin_failure;
+	r = amdgpu_bo_kmap(adev->gmc.pdb0_bo, &adev->gmc.ptr_pdb0);
+	if (r)
+		goto bo_kmap_failure;
+
+	amdgpu_bo_unreserve(adev->gmc.pdb0_bo);
+	return 0;
+
+bo_kmap_failure:
+	amdgpu_bo_unpin(adev->gmc.pdb0_bo);
+bo_pin_failure:
+	amdgpu_bo_unreserve(adev->gmc.pdb0_bo);
+bo_reserve_failure:
+	amdgpu_bo_unref(&adev->gmc.pdb0_bo);
+	return r;
+}
+
 /**
  * amdgpu_gmc_get_pde_for_bo - get the PDE for a BO
  *
@@ -552,3 +603,55 @@ void amdgpu_gmc_get_vbios_allocations(struct amdgpu_device *adev)
 		adev->mman.stolen_extended_size = 0;
 	}
 }
+
+/**
+ * amdgpu_gmc_init_pdb0 - initialize PDB0
+ *
+ * @adev: amdgpu_device pointer
+ *
+ * This function is only used when GART page table is used
+ * for FB address translatioin. In such a case, we construct
+ * a 2-level system VM page table: PDB0->PTB, to cover both
+ * VRAM of the hive and system memory.
+ *
+ * PDB0 is static, initialized once on driver initialization.
+ * The first n entries of PDB0 are used as PTE by setting
+ * P bit to 1, pointing to VRAM. The n+1'th entry points
+ * to a big PTB covering system memory.
+ *
+ */
+void amdgpu_gmc_init_pdb0(struct amdgpu_device *adev)
+{
+	int i;
+	uint64_t flags = adev->gart.gart_pte_flags; //TODO it is UC. explore NC/RW?
+	/* Each PDE0 (used as PTE) covers (2^vmid0_page_table_block_size)*2M
+	 */
+	u64 vram_size = adev->gmc.xgmi.node_segment_size * adev->gmc.xgmi.num_physical_nodes;
+	u64 pde0_page_size = (1ULL<<adev->gmc.vmid0_page_table_block_size)<<21;
+	u64 vram_addr = adev->vm_manager.vram_base_offset -
+		adev->gmc.xgmi.physical_node_id * adev->gmc.xgmi.node_segment_size;
+	u64 vram_end = vram_addr + vram_size;
+	u64 gart_ptb_gpu_pa = amdgpu_bo_gpu_offset(adev->gart.bo) +
+		adev->vm_manager.vram_base_offset - adev->gmc.vram_start;
+
+	flags |= AMDGPU_PTE_VALID | AMDGPU_PTE_READABLE;
+	flags |= AMDGPU_PTE_WRITEABLE;
+	flags |= AMDGPU_PTE_SNOOPED;
+	flags |= AMDGPU_PTE_FRAG((adev->gmc.vmid0_page_table_block_size + 9*1));
+	flags |= AMDGPU_PDE_PTE;
+
+	/* The first n PDE0 entries are used as PTE,
+	 * pointing to vram
+	 */
+	for (i = 0; vram_addr < vram_end; i++, vram_addr += pde0_page_size)
+		amdgpu_gmc_set_pte_pde(adev, adev->gmc.ptr_pdb0, i, vram_addr, flags);
+
+	/* The n+1'th PDE0 entry points to a huge
+	 * PTB who has more than 512 entries each
+	 * pointing to a 4K system page
+	 */
+	flags = AMDGPU_PTE_VALID | AMDGPU_PTE_SYSTEM;
+	flags |= AMDGPU_PDE_BFS(0) | AMDGPU_PTE_SNOOPED;
+	/* Requires gart_ptb_gpu_pa to be 4K aligned */
+	amdgpu_gmc_set_pte_pde(adev, adev->gmc.ptr_pdb0, i, gart_ptb_gpu_pa, flags);
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
index d5312b8e6abf..d5f3825cd479 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
@@ -247,6 +247,9 @@ struct amdgpu_gmc {
 
 	uint32_t	vmid0_page_table_block_size;
 	uint32_t	vmid0_page_table_depth;
+	struct amdgpu_bo		*pdb0_bo;
+	/* CPU kmapped address of pdb0*/
+	void				*ptr_pdb0;
 };
 
 #define amdgpu_gmc_flush_gpu_tlb(adev, vmid, vmhub, type) ((adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid), (vmhub), (type)))
@@ -288,6 +291,7 @@ static inline uint64_t amdgpu_gmc_sign_extend(uint64_t addr)
 	return addr;
 }
 
+int amdgpu_gmc_pdb0_alloc(struct amdgpu_device *adev);
 void amdgpu_gmc_get_pde_for_bo(struct amdgpu_bo *bo, int level,
 			       uint64_t *addr, uint64_t *flags);
 int amdgpu_gmc_set_pte_pde(struct amdgpu_device *adev, void *cpu_pt_addr,
@@ -317,4 +321,5 @@ amdgpu_gmc_set_vm_fault_masks(struct amdgpu_device *adev, int hub_type,
 
 void amdgpu_gmc_get_vbios_allocations(struct amdgpu_device *adev);
 
+void amdgpu_gmc_init_pdb0(struct amdgpu_device *adev);
 #endif
-- 
2.29.2