drm/nouveau: use alternate memory type for system-memory buffers with kind != 0

Fixes bug on Tegra where we'd strip kind information from system memory
(ie. all) buffers, resulting in misrendering.

Behaviour on dGPU should be unchanged.

Reported-by: Thierry Reding <treding@nvidia.com>
Fixes: d7722134b8 ("drm/nouveau: switch over to new memory and vmm interfaces")
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Tested-by: Thierry Reding <treding@nvidia.com>

drivers/gpu/drm/nouveau/nouveau_bo.c

@@ -224,7 +224,7 @@ nouveau_bo_new(struct nouveau_cli *cli, u64 size, int align,
 		/* Determine if we can get a cache-coherent map, forcing
 		 * uncached mapping if we can't.
 		 */
-		if (mmu->type[drm->ttm.type_host].type & NVIF_MEM_UNCACHED)
+		if (!nouveau_drm_use_coherent_gpu_mapping(drm))
 			nvbo->force_coherent = true;
 	}
 

drivers/gpu/drm/nouveau/nouveau_drv.h

@@ -156,8 +156,8 @@ struct nouveau_drm {
 		struct nvif_object copy;
 		int mtrr;
 		int type_vram;
-		int type_host;
-		int type_ncoh;
+		int type_host[2];
+		int type_ncoh[2];
 	} ttm;
 
 	/* GEM interface support */
@@ -216,6 +216,13 @@ nouveau_drm(struct drm_device *dev)
 	return dev->dev_private;
 }
 
+static inline bool
+nouveau_drm_use_coherent_gpu_mapping(struct nouveau_drm *drm)
+{
+	struct nvif_mmu *mmu = &drm->client.mmu;
+	return !(mmu->type[drm->ttm.type_host[0]].type & NVIF_MEM_UNCACHED);
+}
+
 int nouveau_pmops_suspend(struct device *);
 int nouveau_pmops_resume(struct device *);
 bool nouveau_pmops_runtime(void);

drivers/gpu/drm/nouveau/nouveau_mem.c

@@ -103,10 +103,10 @@ nouveau_mem_host(struct ttm_mem_reg *reg, struct ttm_dma_tt *tt)
 	u8 type;
 	int ret;
 
-	if (mmu->type[drm->ttm.type_host].type & NVIF_MEM_UNCACHED)
-		type = drm->ttm.type_ncoh;
+	if (!nouveau_drm_use_coherent_gpu_mapping(drm))
+		type = drm->ttm.type_ncoh[!!mem->kind];
 	else
-		type = drm->ttm.type_host;
+		type = drm->ttm.type_host[0];
 
 	if (mem->kind && !(mmu->type[type].type & NVIF_MEM_KIND))
 		mem->comp = mem->kind = 0;

drivers/gpu/drm/nouveau/nouveau_ttm.c

@@ -235,27 +235,46 @@ nouveau_ttm_global_release(struct nouveau_drm *drm)
 	drm->ttm.mem_global_ref.release = NULL;
 }
 
-int
-nouveau_ttm_init(struct nouveau_drm *drm)
+static int
+nouveau_ttm_init_host(struct nouveau_drm *drm, u8 kind)
 {
-	struct nvkm_device *device = nvxx_device(&drm->client.device);
-	struct nvkm_pci *pci = device->pci;
 	struct nvif_mmu *mmu = &drm->client.mmu;
-	struct drm_device *dev = drm->dev;
-	int typei, ret;
+	int typei;
 
 	typei = nvif_mmu_type(mmu, NVIF_MEM_HOST | NVIF_MEM_MAPPABLE |
-						   NVIF_MEM_COHERENT);
+					    kind | NVIF_MEM_COHERENT);
 	if (typei < 0)
 		return -ENOSYS;
 
-	drm->ttm.type_host = typei;
+	drm->ttm.type_host[!!kind] = typei;
 
-	typei = nvif_mmu_type(mmu, NVIF_MEM_HOST | NVIF_MEM_MAPPABLE);
+	typei = nvif_mmu_type(mmu, NVIF_MEM_HOST | NVIF_MEM_MAPPABLE | kind);
 	if (typei < 0)
 		return -ENOSYS;
 
-	drm->ttm.type_ncoh = typei;
+	drm->ttm.type_ncoh[!!kind] = typei;
+	return 0;
+}
+
+int
+nouveau_ttm_init(struct nouveau_drm *drm)
+{
+	struct nvkm_device *device = nvxx_device(&drm->client.device);
+	struct nvkm_pci *pci = device->pci;
+	struct nvif_mmu *mmu = &drm->client.mmu;
+	struct drm_device *dev = drm->dev;
+	int typei, ret;
+
+	ret = nouveau_ttm_init_host(drm, 0);
+	if (ret)
+		return ret;
+
+	if (drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA &&
+	    drm->client.device.info.chipset != 0x50) {
+		ret = nouveau_ttm_init_host(drm, NVIF_MEM_KIND);
+		if (ret)
+			return ret;
+	}
 
 	if (drm->client.device.info.platform != NV_DEVICE_INFO_V0_SOC &&
 	    drm->client.device.info.family >= NV_DEVICE_INFO_V0_TESLA) {