drm/nouveau/mmu/gf100-: virtualise setting pdb base address for invalidation

It appears that Pascal and newer need something different.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmm.h

@@ -141,6 +141,8 @@ struct nvkm_vmm_func {
 		     struct nvkm_vmm_map *);
 	void (*flush)(struct nvkm_vmm *, int depth);
 
+	void (*invalidate_pdb)(struct nvkm_vmm *, u64 addr);
+
 	u64 page_block;
 	const struct nvkm_vmm_page page[];
 };
@@ -186,8 +188,9 @@ int gf100_vmm_join(struct nvkm_vmm *, struct nvkm_memory *);
 void gf100_vmm_part(struct nvkm_vmm *, struct nvkm_memory *);
 int gf100_vmm_aper(enum nvkm_memory_target);
 int gf100_vmm_valid(struct nvkm_vmm *, void *, u32, struct nvkm_vmm_map *);
-void gf100_vmm_invalidate(struct nvkm_vmm *, u32 type);
 void gf100_vmm_flush(struct nvkm_vmm *, int);
+void gf100_vmm_invalidate(struct nvkm_vmm *, u32 type);
+void gf100_vmm_invalidate_pdb(struct nvkm_vmm *, u64 addr);
 
 int gk20a_vmm_aper(enum nvkm_memory_target);
 
@@ -200,6 +203,7 @@ int gm200_vmm_join(struct nvkm_vmm *, struct nvkm_memory *);
 int gp100_vmm_join(struct nvkm_vmm *, struct nvkm_memory *);
 int gp100_vmm_valid(struct nvkm_vmm *, void *, u32, struct nvkm_vmm_map *);
 void gp100_vmm_flush(struct nvkm_vmm *, int);
+void gp100_vmm_invalidate_pdb(struct nvkm_vmm *, u64 addr);
 
 int gv100_vmm_join(struct nvkm_vmm *, struct nvkm_memory *);
 

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgf100.c

@@ -177,11 +177,20 @@ gf100_vmm_desc_16_16[] = {
 	{}
 };
 
+void
+gf100_vmm_invalidate_pdb(struct nvkm_vmm *vmm, u64 addr)
+{
+	struct nvkm_device *device = vmm->mmu->subdev.device;
+	nvkm_wr32(device, 0x100cb8, addr);
+}
+
 void
 gf100_vmm_invalidate(struct nvkm_vmm *vmm, u32 type)
 {
 	struct nvkm_subdev *subdev = &vmm->mmu->subdev;
 	struct nvkm_device *device = subdev->device;
+	struct nvkm_mmu_pt *pd = vmm->pd->pt[0];
+	u64 addr = 0;
 
 	mutex_lock(&subdev->mutex);
 	/* Looks like maybe a "free flush slots" counter, the
@@ -192,7 +201,20 @@ gf100_vmm_invalidate(struct nvkm_vmm *vmm, u32 type)
 			break;
 	);
 
-	nvkm_wr32(device, 0x100cb8, vmm->pd->pt[0]->addr >> 8);
+	if (!(type & 0x00000002) /* ALL_PDB. */) {
+		switch (nvkm_memory_target(pd->memory)) {
+		case NVKM_MEM_TARGET_VRAM: addr |= 0x00000000; break;
+		case NVKM_MEM_TARGET_HOST: addr |= 0x00000002; break;
+		case NVKM_MEM_TARGET_NCOH: addr |= 0x00000003; break;
+		default:
+			WARN_ON(1);
+			break;
+		}
+		addr |= (vmm->pd->pt[0]->addr >> 12) << 4;
+
+		vmm->func->invalidate_pdb(vmm, addr);
+	}
+
 	nvkm_wr32(device, 0x100cbc, 0x80000000 | type);
 
 	/* Wait for flush to be queued? */
@@ -352,6 +374,7 @@ gf100_vmm_17 = {
 	.aper = gf100_vmm_aper,
 	.valid = gf100_vmm_valid,
 	.flush = gf100_vmm_flush,
+	.invalidate_pdb = gf100_vmm_invalidate_pdb,
 	.page = {
 		{ 17, &gf100_vmm_desc_17_17[0], NVKM_VMM_PAGE_xVxC },
 		{ 12, &gf100_vmm_desc_17_12[0], NVKM_VMM_PAGE_xVHx },
@@ -366,6 +389,7 @@ gf100_vmm_16 = {
 	.aper = gf100_vmm_aper,
 	.valid = gf100_vmm_valid,
 	.flush = gf100_vmm_flush,
+	.invalidate_pdb = gf100_vmm_invalidate_pdb,
 	.page = {
 		{ 16, &gf100_vmm_desc_16_16[0], NVKM_VMM_PAGE_xVxC },
 		{ 12, &gf100_vmm_desc_16_12[0], NVKM_VMM_PAGE_xVHx },

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgk104.c

@@ -71,6 +71,7 @@ gk104_vmm_17 = {
 	.aper = gf100_vmm_aper,
 	.valid = gf100_vmm_valid,
 	.flush = gf100_vmm_flush,
+	.invalidate_pdb = gf100_vmm_invalidate_pdb,
 	.page = {
 		{ 17, &gk104_vmm_desc_17_17[0], NVKM_VMM_PAGE_xVxC },
 		{ 12, &gk104_vmm_desc_17_12[0], NVKM_VMM_PAGE_xVHx },
@@ -85,6 +86,7 @@ gk104_vmm_16 = {
 	.aper = gf100_vmm_aper,
 	.valid = gf100_vmm_valid,
 	.flush = gf100_vmm_flush,
+	.invalidate_pdb = gf100_vmm_invalidate_pdb,
 	.page = {
 		{ 16, &gk104_vmm_desc_16_16[0], NVKM_VMM_PAGE_xVxC },
 		{ 12, &gk104_vmm_desc_16_12[0], NVKM_VMM_PAGE_xVHx },

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgk20a.c

@@ -40,6 +40,7 @@ gk20a_vmm_17 = {
 	.aper = gf100_vmm_aper,
 	.valid = gf100_vmm_valid,
 	.flush = gf100_vmm_flush,
+	.invalidate_pdb = gf100_vmm_invalidate_pdb,
 	.page = {
 		{ 17, &gk104_vmm_desc_17_17[0], NVKM_VMM_PAGE_xxHC },
 		{ 12, &gk104_vmm_desc_17_12[0], NVKM_VMM_PAGE_xxHx },
@@ -54,6 +55,7 @@ gk20a_vmm_16 = {
 	.aper = gf100_vmm_aper,
 	.valid = gf100_vmm_valid,
 	.flush = gf100_vmm_flush,
+	.invalidate_pdb = gf100_vmm_invalidate_pdb,
 	.page = {
 		{ 16, &gk104_vmm_desc_16_16[0], NVKM_VMM_PAGE_xxHC },
 		{ 12, &gk104_vmm_desc_16_12[0], NVKM_VMM_PAGE_xxHx },

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgm200.c

@@ -113,6 +113,7 @@ gm200_vmm_17 = {
 	.aper = gf100_vmm_aper,
 	.valid = gf100_vmm_valid,
 	.flush = gf100_vmm_flush,
+	.invalidate_pdb = gf100_vmm_invalidate_pdb,
 	.page = {
 		{ 27, &gm200_vmm_desc_17_17[1], NVKM_VMM_PAGE_Sxxx },
 		{ 17, &gm200_vmm_desc_17_17[0], NVKM_VMM_PAGE_SVxC },
@@ -128,6 +129,7 @@ gm200_vmm_16 = {
 	.aper = gf100_vmm_aper,
 	.valid = gf100_vmm_valid,
 	.flush = gf100_vmm_flush,
+	.invalidate_pdb = gf100_vmm_invalidate_pdb,
 	.page = {
 		{ 27, &gm200_vmm_desc_16_16[1], NVKM_VMM_PAGE_Sxxx },
 		{ 16, &gm200_vmm_desc_16_16[0], NVKM_VMM_PAGE_SVxC },

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgm20b.c

@@ -28,6 +28,7 @@ gm20b_vmm_17 = {
 	.aper = gk20a_vmm_aper,
 	.valid = gf100_vmm_valid,
 	.flush = gf100_vmm_flush,
+	.invalidate_pdb = gf100_vmm_invalidate_pdb,
 	.page = {
 		{ 27, &gm200_vmm_desc_17_17[1], NVKM_VMM_PAGE_Sxxx },
 		{ 17, &gm200_vmm_desc_17_17[0], NVKM_VMM_PAGE_SxHC },
@@ -43,6 +44,7 @@ gm20b_vmm_16 = {
 	.aper = gk20a_vmm_aper,
 	.valid = gf100_vmm_valid,
 	.flush = gf100_vmm_flush,
+	.invalidate_pdb = gf100_vmm_invalidate_pdb,
 	.page = {
 		{ 27, &gm200_vmm_desc_16_16[1], NVKM_VMM_PAGE_Sxxx },
 		{ 16, &gm200_vmm_desc_16_16[0], NVKM_VMM_PAGE_SxHC },

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgp100.c

@@ -306,6 +306,14 @@ gp100_vmm_valid(struct nvkm_vmm *vmm, void *argv, u32 argc,
 	return 0;
 }
 
+void
+gp100_vmm_invalidate_pdb(struct nvkm_vmm *vmm, u64 addr)
+{
+	struct nvkm_device *device = vmm->mmu->subdev.device;
+	nvkm_wr32(device, 0x100cb8, lower_32_bits(addr));
+	nvkm_wr32(device, 0x100cec, upper_32_bits(addr));
+}
+
 void
 gp100_vmm_flush(struct nvkm_vmm *vmm, int depth)
 {
@@ -331,6 +339,7 @@ gp100_vmm = {
 	.aper = gf100_vmm_aper,
 	.valid = gp100_vmm_valid,
 	.flush = gp100_vmm_flush,
+	.invalidate_pdb = gp100_vmm_invalidate_pdb,
 	.page = {
 		{ 47, &gp100_vmm_desc_16[4], NVKM_VMM_PAGE_Sxxx },
 		{ 38, &gp100_vmm_desc_16[3], NVKM_VMM_PAGE_Sxxx },

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgp10b.c

@@ -28,6 +28,7 @@ gp10b_vmm = {
 	.aper = gk20a_vmm_aper,
 	.valid = gp100_vmm_valid,
 	.flush = gp100_vmm_flush,
+	.invalidate_pdb = gp100_vmm_invalidate_pdb,
 	.page = {
 		{ 47, &gp100_vmm_desc_16[4], NVKM_VMM_PAGE_Sxxx },
 		{ 38, &gp100_vmm_desc_16[3], NVKM_VMM_PAGE_Sxxx },

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmgv100.c

@@ -66,6 +66,7 @@ gv100_vmm = {
 	.aper = gf100_vmm_aper,
 	.valid = gp100_vmm_valid,
 	.flush = gp100_vmm_flush,
+	.invalidate_pdb = gp100_vmm_invalidate_pdb,
 	.page = {
 		{ 47, &gp100_vmm_desc_16[4], NVKM_VMM_PAGE_Sxxx },
 		{ 38, &gp100_vmm_desc_16[3], NVKM_VMM_PAGE_Sxxx },