drm/nv50/vm: remove explicit vm knowledge from engines

This reverses the lock ordering between VM and gr/nv84:nvc0.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/core/engine/copy/nva3.c

@@ -116,13 +116,6 @@ nva3_copy_intr(struct nouveau_subdev *subdev)
 	nouveau_engctx_put(engctx);
 }
 
-static int
-nva3_copy_tlb_flush(struct nouveau_engine *engine)
-{
-	nv50_vm_flush_engine(&engine->base, 0x0d);
-	return 0;
-}
-
 static int
 nva3_copy_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
 	       struct nouveau_oclass *oclass, void *data, u32 size,
@@ -142,7 +135,6 @@ nva3_copy_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
 	nv_subdev(priv)->intr = nva3_copy_intr;
 	nv_engine(priv)->cclass = &nva3_copy_cclass;
 	nv_engine(priv)->sclass = nva3_copy_sclass;
-	nv_engine(priv)->tlb_flush = nva3_copy_tlb_flush;
 	nv_falcon(priv)->code.data = nva3_pcopy_code;
 	nv_falcon(priv)->code.size = sizeof(nva3_pcopy_code);
 	nv_falcon(priv)->data.data = nva3_pcopy_data;

drivers/gpu/drm/nouveau/core/engine/crypt/nv84.c

@@ -140,13 +140,6 @@ nv84_crypt_intr(struct nouveau_subdev *subdev)
 	nouveau_engctx_put(engctx);
 }
 
-static int
-nv84_crypt_tlb_flush(struct nouveau_engine *engine)
-{
-	nv50_vm_flush_engine(&engine->base, 0x0a);
-	return 0;
-}
-
 static int
 nv84_crypt_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
 	       struct nouveau_oclass *oclass, void *data, u32 size,
@@ -165,7 +158,6 @@ nv84_crypt_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
 	nv_subdev(priv)->intr = nv84_crypt_intr;
 	nv_engine(priv)->cclass = &nv84_crypt_cclass;
 	nv_engine(priv)->sclass = nv84_crypt_sclass;
-	nv_engine(priv)->tlb_flush = nv84_crypt_tlb_flush;
 	return 0;
 }
 

drivers/gpu/drm/nouveau/core/engine/crypt/nv98.c

@@ -118,13 +118,6 @@ nv98_crypt_intr(struct nouveau_subdev *subdev)
 	nouveau_engctx_put(engctx);
 }
 
-static int
-nv98_crypt_tlb_flush(struct nouveau_engine *engine)
-{
-	nv50_vm_flush_engine(&engine->base, 0x0a);
-	return 0;
-}
-
 static int
 nv98_crypt_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
 	       struct nouveau_oclass *oclass, void *data, u32 size,
@@ -143,7 +136,6 @@ nv98_crypt_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
 	nv_subdev(priv)->intr = nv98_crypt_intr;
 	nv_engine(priv)->cclass = &nv98_crypt_cclass;
 	nv_engine(priv)->sclass = nv98_crypt_sclass;
-	nv_engine(priv)->tlb_flush = nv98_crypt_tlb_flush;
 	nv_falcon(priv)->code.data = nv98_pcrypt_code;
 	nv_falcon(priv)->code.size = sizeof(nv98_pcrypt_code);
 	nv_falcon(priv)->data.data = nv98_pcrypt_data;

drivers/gpu/drm/nouveau/core/engine/graph/nv50.c

@@ -186,13 +186,6 @@ nv50_graph_cclass = {
  * PGRAPH engine/subdev functions
  ******************************************************************************/
 
-static int
-nv50_graph_tlb_flush(struct nouveau_engine *engine)
-{
-	nv50_vm_flush_engine(&engine->base, 0x00);
-	return 0;
-}
-
 static const struct nouveau_bitfield nv50_pgraph_status[] = {
 	{ 0x00000001, "BUSY" }, /* set when any bit is set */
 	{ 0x00000002, "DISPATCH" },
@@ -302,8 +295,10 @@ nv84_graph_tlb_flush(struct nouveau_engine *engine)
 				nv_rd32(priv, 0x400388));
 	}
 
-	nv50_vm_flush_engine(&engine->base, 0x00);
 
+	nv_wr32(priv, 0x100c80, 0x00000001);
+	if (!nv_wait(priv, 0x100c80, 0x00000001, 0x00000000))
+		nv_error(priv, "vm flush timeout\n");
 	nv_mask(priv, 0x400500, 0x00000001, 0x00000001);
 	spin_unlock_irqrestore(&priv->lock, flags);
 	return timeout ? -EBUSY : 0;
@@ -857,10 +852,9 @@ nv50_graph_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
 
 	};
 
-	if (nv_device(priv)->chipset == 0x50 ||
-	    nv_device(priv)->chipset == 0xac)
-		nv_engine(priv)->tlb_flush = nv50_graph_tlb_flush;
-	else
+	/* unfortunate hw bug workaround... */
+	if (nv_device(priv)->chipset != 0x50 &&
+	    nv_device(priv)->chipset != 0xac)
 		nv_engine(priv)->tlb_flush = nv84_graph_tlb_flush;
 
 	spin_lock_init(&priv->lock);

drivers/gpu/drm/nouveau/core/engine/mpeg/nv50.c

@@ -125,13 +125,6 @@ nv50_mpeg_cclass = {
  * PMPEG engine/subdev functions
  ******************************************************************************/
 
-int
-nv50_mpeg_tlb_flush(struct nouveau_engine *engine)
-{
-	nv50_vm_flush_engine(&engine->base, 0x08);
-	return 0;
-}
-
 void
 nv50_mpeg_intr(struct nouveau_subdev *subdev)
 {
@@ -191,7 +184,6 @@ nv50_mpeg_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
 	nv_subdev(priv)->intr = nv50_vpe_intr;
 	nv_engine(priv)->cclass = &nv50_mpeg_cclass;
 	nv_engine(priv)->sclass = nv50_mpeg_sclass;
-	nv_engine(priv)->tlb_flush = nv50_mpeg_tlb_flush;
 	return 0;
 }
 

drivers/gpu/drm/nouveau/core/engine/mpeg/nv84.c

@@ -88,7 +88,6 @@ nv84_mpeg_ctor(struct nouveau_object *parent, struct nouveau_object *engine,
 	nv_subdev(priv)->intr = nv50_mpeg_intr;
 	nv_engine(priv)->cclass = &nv84_mpeg_cclass;
 	nv_engine(priv)->sclass = nv84_mpeg_sclass;
-	nv_engine(priv)->tlb_flush = nv50_mpeg_tlb_flush;
 	return 0;
 }
 

drivers/gpu/drm/nouveau/core/include/engine/mpeg.h

@@ -54,7 +54,6 @@ extern struct nouveau_ofuncs nv50_mpeg_ofuncs;
 int  nv50_mpeg_context_ctor(struct nouveau_object *, struct nouveau_object *,
 			    struct nouveau_oclass *, void *, u32,
 			    struct nouveau_object **);
-int  nv50_mpeg_tlb_flush(struct nouveau_engine *);
 void nv50_mpeg_intr(struct nouveau_subdev *);
 int  nv50_mpeg_init(struct nouveau_object *);
 

drivers/gpu/drm/nouveau/core/include/subdev/vm.h

@@ -117,7 +117,6 @@ int  nv04_vm_create(struct nouveau_vmmgr *, u64, u64, u64,
 		    struct nouveau_vm **);
 void nv04_vmmgr_dtor(struct nouveau_object *);
 
-void nv50_vm_flush_engine(struct nouveau_subdev *, int engine);
 void nvc0_vm_flush_engine(struct nouveau_subdev *, u64 addr, int type);
 
 /* nouveau_vm.c */

drivers/gpu/drm/nouveau/core/subdev/vm/nv50.c

@@ -151,31 +151,37 @@ nv50_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
 static void
 nv50_vm_flush(struct nouveau_vm *vm)
 {
+	struct nv50_vmmgr_priv *priv = (void *)vm->vmm;
 	struct nouveau_engine *engine;
-	int i;
+	unsigned long flags;
+	int i, vme;
 
+	spin_lock_irqsave(&priv->lock, flags);
 	for (i = 0; i < NVDEV_SUBDEV_NR; i++) {
-		if (atomic_read(&vm->engref[i]) && i == NVDEV_SUBDEV_BAR) {
-			nv50_vm_flush_engine(nv_subdev(vm->vmm), 6);
-		} else
-		if (atomic_read(&vm->engref[i])) {
-			engine = nouveau_engine(vm->vmm, i);
-			if (engine && engine->tlb_flush)
-				engine->tlb_flush(engine);
+		if (!atomic_read(&vm->engref[i]))
+			continue;
+
+		/* unfortunate hw bug workaround... */
+		engine = nouveau_engine(priv, i);
+		if (engine && engine->tlb_flush) {
+			engine->tlb_flush(engine);
+			continue;
 		}
-	}
-}
 
-void
-nv50_vm_flush_engine(struct nouveau_subdev *subdev, int engine)
-{
-	struct nv50_vmmgr_priv *priv = (void *)nouveau_vmmgr(subdev);
-	unsigned long flags;
+		switch (i) {
+		case NVDEV_ENGINE_GR   : vme = 0x00; break;
+		case NVDEV_SUBDEV_BAR  : vme = 0x06; break;
+		case NVDEV_ENGINE_MPEG : vme = 0x08; break;
+		case NVDEV_ENGINE_CRYPT: vme = 0x0a; break;
+		case NVDEV_ENGINE_COPY0: vme = 0x0d; break;
+		default:
+			continue;
+		}
 
-	spin_lock_irqsave(&priv->lock, flags);
-	nv_wr32(subdev, 0x100c80, (engine << 16) | 1);
-	if (!nv_wait(subdev, 0x100c80, 0x00000001, 0x00000000))
-		nv_error(subdev, "vm flush timeout: engine %d\n", engine);
+		nv_wr32(priv, 0x100c80, (vme << 16) | 1);
+		if (!nv_wait(priv, 0x100c80, 0x00000001, 0x00000000))
+			nv_error(priv, "vm flush timeout: engine %d\n", vme);
+	}
 	spin_unlock_irqrestore(&priv->lock, flags);
 }