From: Ben Skeggs <bskeggs@redhat.com>
Date: Wed, 1 Nov 2017 03:56:19 +1000
Subject: drm/nouveau/mmu/nv41: implement new vmm backend
Git-commit: 473f9aca6c1063bf77cd61c2683fc496850d63b3
Patch-mainline: v4.15-rc1
References: FATE#326289 FATE#326079 FATE#326049 FATE#322398 FATE#326166

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Acked-by: Petr Tesarik <ptesarik@suse.com>
---
 drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c    |   69 ----------------------
 drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmnv41.c |   64 ++++++++++++++++++++
 2 files changed, 64 insertions(+), 69 deletions(-)

--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/nv41.c
@@ -24,75 +24,10 @@
 #include "vmm.h"
 
 #include <core/option.h>
-#include <subdev/timer.h>
 
 #include <nvif/class.h>
 
 #define NV41_GART_SIZE (512 * 1024 * 1024)
-#define NV41_GART_PAGE (  4 * 1024)
-
-/*******************************************************************************
- * VM map/unmap callbacks
- ******************************************************************************/
-
-static void
-nv41_vm_map_sg(struct nvkm_vma *vma, struct nvkm_memory *pgt,
-	       struct nvkm_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
-{
-	pte = pte * 4;
-	nvkm_kmap(pgt);
-	while (cnt) {
-		u32 page = PAGE_SIZE / NV41_GART_PAGE;
-		u64 phys = (u64)*list++;
-		while (cnt && page--) {
-			nvkm_wo32(pgt, pte, (phys >> 7) | 1);
-			phys += NV41_GART_PAGE;
-			pte += 4;
-			cnt -= 1;
-		}
-	}
-	nvkm_done(pgt);
-}
-
-static void
-nv41_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt)
-{
-	pte = pte * 4;
-	nvkm_kmap(pgt);
-	while (cnt--) {
-		nvkm_wo32(pgt, pte, 0x00000000);
-		pte += 4;
-	}
-	nvkm_done(pgt);
-}
-
-static void
-nv41_vm_flush(struct nvkm_vm *vm)
-{
-	struct nvkm_subdev *subdev = &vm->mmu->subdev;
-	struct nvkm_device *device = subdev->device;
-
-	mutex_lock(&subdev->mutex);
-	nvkm_wr32(device, 0x100810, 0x00000022);
-	nvkm_msec(device, 2000,
-		if (nvkm_rd32(device, 0x100810) & 0x00000020)
-			break;
-	);
-	nvkm_wr32(device, 0x100810, 0x00000000);
-	mutex_unlock(&subdev->mutex);
-}
-
-/*******************************************************************************
- * MMU subdev
- ******************************************************************************/
-
-static int
-nv41_mmu_oneinit(struct nvkm_mmu *mmu)
-{
-	mmu->vmm->pgt[0].mem[0] = mmu->vmm->pd->pt[0]->memory;
-	mmu->vmm->pgt[0].refcount[0] = 1;
-	return 0;
-}
 
 static void
 nv41_mmu_init(struct nvkm_mmu *mmu)
@@ -105,16 +40,12 @@ nv41_mmu_init(struct nvkm_mmu *mmu)
 
 static const struct nvkm_mmu_func
 nv41_mmu = {
-	.oneinit = nv41_mmu_oneinit,
 	.init = nv41_mmu_init,
 	.limit = NV41_GART_SIZE,
 	.dma_bits = 39,
 	.pgt_bits = 32 - 12,
 	.spg_shift = 12,
 	.lpg_shift = 12,
-	.map_sg = nv41_vm_map_sg,
-	.unmap = nv41_vm_unmap,
-	.flush = nv41_vm_flush,
 	.vmm = {{ -1, -1, NVIF_CLASS_VMM_NV04}, nv41_vmm_new, true },
 };
 
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmnv41.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmnv41.c
@@ -21,8 +21,54 @@
  */
 #include "vmm.h"
 
+#include <subdev/timer.h>
+
+static void
+nv41_vmm_pgt_pte(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
+		 u32 ptei, u32 ptes, struct nvkm_vmm_map *map, u64 addr)
+{
+	u32 data = (addr >> 7) | 0x00000001; /* VALID. */
+	while (ptes--) {
+		VMM_WO032(pt, vmm, ptei++ * 4, data);
+		data += 0x00000020;
+	}
+}
+
+static void
+nv41_vmm_pgt_sgl(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
+		 u32 ptei, u32 ptes, struct nvkm_vmm_map *map)
+{
+	VMM_MAP_ITER_SGL(vmm, pt, ptei, ptes, map, nv41_vmm_pgt_pte);
+}
+
+static void
+nv41_vmm_pgt_dma(struct nvkm_vmm *vmm, struct nvkm_mmu_pt *pt,
+		 u32 ptei, u32 ptes, struct nvkm_vmm_map *map)
+{
+#if PAGE_SHIFT == 12
+	nvkm_kmap(pt->memory);
+	while (ptes--) {
+		const u32 data = (*map->dma++ >> 7) | 0x00000001;
+		VMM_WO032(pt, vmm, ptei++ * 4, data);
+	}
+	nvkm_done(pt->memory);
+#else
+	VMM_MAP_ITER_DMA(vmm, pt, ptei, ptes, map, nv41_vmm_pgt_pte);
+#endif
+}
+
+static void
+nv41_vmm_pgt_unmap(struct nvkm_vmm *vmm,
+		   struct nvkm_mmu_pt *pt, u32 ptei, u32 ptes)
+{
+	VMM_FO032(pt, vmm, ptei * 4, 0, ptes);
+}
+
 static const struct nvkm_vmm_desc_func
 nv41_vmm_desc_pgt = {
+	.unmap = nv41_vmm_pgt_unmap,
+	.dma = nv41_vmm_pgt_dma,
+	.sgl = nv41_vmm_pgt_sgl,
 };
 
 static const struct nvkm_vmm_desc
@@ -31,8 +77,26 @@ nv41_vmm_desc_12[] = {
 	{}
 };
 
+static void
+nv41_vmm_flush(struct nvkm_vmm *vmm, int level)
+{
+	struct nvkm_subdev *subdev = &vmm->mmu->subdev;
+	struct nvkm_device *device = subdev->device;
+
+	mutex_lock(&subdev->mutex);
+	nvkm_wr32(device, 0x100810, 0x00000022);
+	nvkm_msec(device, 2000,
+		if (nvkm_rd32(device, 0x100810) & 0x00000020)
+			break;
+	);
+	nvkm_wr32(device, 0x100810, 0x00000000);
+	mutex_unlock(&subdev->mutex);
+}
+
 static const struct nvkm_vmm_func
 nv41_vmm = {
+	.valid = nv04_vmm_valid,
+	.flush = nv41_vmm_flush,
 	.page = {
 		{ 12, &nv41_vmm_desc_12[0], NVKM_VMM_PAGE_HOST },
 		{}