drm/nouveau/mmu: serialise mmu invalidations with private mutex

nvkm_subdev.mutex is going away.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>

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

@@ -117,6 +117,8 @@ struct nvkm_mmu {
 		struct list_head list;
 	} ptc, ptp;
 
+	struct mutex mutex; /* serialises mmu invalidations */
+
 	struct nvkm_device_oclass user;
 };
 

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

@@ -402,6 +402,7 @@ nvkm_mmu_dtor(struct nvkm_subdev *subdev)
 	nvkm_vmm_unref(&mmu->vmm);
 
 	nvkm_mmu_ptc_fini(mmu);
+	mutex_destroy(&mmu->mutex);
 	return mmu;
 }
 
@@ -420,6 +421,7 @@ nvkm_mmu_ctor(const struct nvkm_mmu_func *func, struct nvkm_device *device,
 	mmu->func = func;
 	mmu->dma_bits = func->dma_bits;
 	nvkm_mmu_ptc_init(mmu);
+	mutex_init(&mmu->mutex);
 	mmu->user.ctor = nvkm_ummu_new;
 	mmu->user.base = func->mmu.user;
 }

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

@@ -187,12 +187,11 @@ gf100_vmm_invalidate_pdb(struct nvkm_vmm *vmm, u64 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_device *device = vmm->mmu->subdev.device;
 	struct nvkm_mmu_pt *pd = vmm->pd->pt[0];
 	u64 addr = 0;
 
-	mutex_lock(&subdev->mutex);
+	mutex_lock(&vmm->mmu->mutex);
 	/* Looks like maybe a "free flush slots" counter, the
 	 * faster you write to 0x100cbc to more it decreases.
 	 */
@@ -222,7 +221,7 @@ gf100_vmm_invalidate(struct nvkm_vmm *vmm, u32 type)
 		if (nvkm_rd32(device, 0x100c80) & 0x00008000)
 			break;
 	);
-	mutex_unlock(&subdev->mutex);
+	mutex_unlock(&vmm->mmu->mutex);
 }
 
 void

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmnv41.c

@@ -80,17 +80,16 @@ 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;
+	struct nvkm_device *device = vmm->mmu->subdev.device;
 
-	mutex_lock(&subdev->mutex);
+	mutex_lock(&vmm->mmu->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);
+	mutex_unlock(&vmm->mmu->mutex);
 }
 
 static const struct nvkm_vmm_func

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmnv50.c

@@ -184,7 +184,7 @@ nv50_vmm_flush(struct nvkm_vmm *vmm, int level)
 	struct nvkm_device *device = subdev->device;
 	int i, id;
 
-	mutex_lock(&subdev->mutex);
+	mutex_lock(&vmm->mmu->mutex);
 	for (i = 0; i < NVKM_SUBDEV_NR; i++) {
 		if (!atomic_read(&vmm->engref[i]))
 			continue;
@@ -220,7 +220,7 @@ nv50_vmm_flush(struct nvkm_vmm *vmm, int level)
 			nvkm_error(subdev, "%s mmu invalidate timeout\n",
 				   nvkm_subdev_name[i]);
 	}
-	mutex_unlock(&subdev->mutex);
+	mutex_unlock(&vmm->mmu->mutex);
 }
 
 int

drivers/gpu/drm/nouveau/nvkm/subdev/mmu/vmmtu102.c

@@ -26,15 +26,14 @@
 static void
 tu102_vmm_flush(struct nvkm_vmm *vmm, int depth)
 {
-	struct nvkm_subdev *subdev = &vmm->mmu->subdev;
-	struct nvkm_device *device = subdev->device;
+	struct nvkm_device *device = vmm->mmu->subdev.device;
 	u32 type = (5 /* CACHE_LEVEL_UP_TO_PDE3 */ - depth) << 24;
 
 	type |= 0x00000001; /* PAGE_ALL */
 	if (atomic_read(&vmm->engref[NVKM_SUBDEV_BAR]))
 		type |= 0x00000004; /* HUB_ONLY */
 
-	mutex_lock(&subdev->mutex);
+	mutex_lock(&vmm->mmu->mutex);
 
 	nvkm_wr32(device, 0xb830a0, vmm->pd->pt[0]->addr >> 8);
 	nvkm_wr32(device, 0xb830a4, 0x00000000);
@@ -46,7 +45,7 @@ tu102_vmm_flush(struct nvkm_vmm *vmm, int depth)
 			break;
 	);
 
-	mutex_unlock(&subdev->mutex);
+	mutex_unlock(&vmm->mmu->mutex);
 }
 
 static const struct nvkm_vmm_func