drm/nouveau/fifo: allow direct access to channel control registers where possible

The indirect method has been left in-place here as a fallback path, as
it may not be possible to map the non-PAGE_SIZE aligned control areas
across some chipset+interface combinations.

This isn't a problem for the primary use-case where the core and drm
are linked together in kernel-land, but across a VM or (in the case
where it applies now) between the core in the kernel and a userspace
test tool.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

drivers/gpu/drm/nouveau/core/engine/fifo/base.c

@@ -103,15 +103,10 @@ nouveau_fifo_channel_create_(struct nouveau_object *parent,
 		return -ENOSPC;
 	}
 
-	/* map fifo control registers */
-	chan->user = ioremap(nv_device_resource_start(device, bar) + addr +
-			     (chan->chid * size), size);
-	if (!chan->user)
-		return -EFAULT;
-
-	nvkm_event_send(&priv->cevent, 1, 0, NULL, 0);
-
+	chan->addr = nv_device_resource_start(device, bar) +
+		     addr + size * chan->chid;
 	chan->size = size;
+	nvkm_event_send(&priv->cevent, 1, 0, NULL, 0);
 	return 0;
 }
 
@@ -121,6 +116,7 @@ nouveau_fifo_channel_destroy(struct nouveau_fifo_chan *chan)
 	struct nouveau_fifo *priv = (void *)nv_object(chan)->engine;
 	unsigned long flags;
 
+	if (chan->user)
 		iounmap(chan->user);
 
 	spin_lock_irqsave(&priv->lock, flags);
@@ -139,10 +135,24 @@ _nouveau_fifo_channel_dtor(struct nouveau_object *object)
 	nouveau_fifo_channel_destroy(chan);
 }
 
+int
+_nouveau_fifo_channel_map(struct nouveau_object *object, u64 *addr, u32 *size)
+{
+	struct nouveau_fifo_chan *chan = (void *)object;
+	*addr = chan->addr;
+	*size = chan->size;
+	return 0;
+}
+
 u32
 _nouveau_fifo_channel_rd32(struct nouveau_object *object, u64 addr)
 {
 	struct nouveau_fifo_chan *chan = (void *)object;
+	if (unlikely(!chan->user)) {
+		chan->user = ioremap(chan->addr, chan->size);
+		if (WARN_ON_ONCE(chan->user == NULL))
+			return 0;
+	}
 	return ioread32_native(chan->user + addr);
 }
 
@@ -150,6 +160,11 @@ void
 _nouveau_fifo_channel_wr32(struct nouveau_object *object, u64 addr, u32 data)
 {
 	struct nouveau_fifo_chan *chan = (void *)object;
+	if (unlikely(!chan->user)) {
+		chan->user = ioremap(chan->addr, chan->size);
+		if (WARN_ON_ONCE(chan->user == NULL))
+			return;
+	}
 	iowrite32_native(data, chan->user + addr);
 }
 

drivers/gpu/drm/nouveau/core/engine/fifo/nv04.c

@@ -252,6 +252,7 @@ nv04_fifo_ofuncs = {
 	.dtor = nv04_fifo_chan_dtor,
 	.init = nv04_fifo_chan_init,
 	.fini = nv04_fifo_chan_fini,
+	.map  = _nouveau_fifo_channel_map,
 	.rd32 = _nouveau_fifo_channel_rd32,
 	.wr32 = _nouveau_fifo_channel_wr32,
 };

drivers/gpu/drm/nouveau/core/engine/fifo/nv10.c

@@ -110,6 +110,7 @@ nv10_fifo_ofuncs = {
 	.dtor = nv04_fifo_chan_dtor,
 	.init = nv04_fifo_chan_init,
 	.fini = nv04_fifo_chan_fini,
+	.map  = _nouveau_fifo_channel_map,
 	.rd32 = _nouveau_fifo_channel_rd32,
 	.wr32 = _nouveau_fifo_channel_wr32,
 };

drivers/gpu/drm/nouveau/core/engine/fifo/nv17.c

@@ -117,6 +117,7 @@ nv17_fifo_ofuncs = {
 	.dtor = nv04_fifo_chan_dtor,
 	.init = nv04_fifo_chan_init,
 	.fini = nv04_fifo_chan_fini,
+	.map  = _nouveau_fifo_channel_map,
 	.rd32 = _nouveau_fifo_channel_rd32,
 	.wr32 = _nouveau_fifo_channel_wr32,
 };

drivers/gpu/drm/nouveau/core/engine/fifo/nv40.c

@@ -236,6 +236,7 @@ nv40_fifo_ofuncs = {
 	.dtor = nv04_fifo_chan_dtor,
 	.init = nv04_fifo_chan_init,
 	.fini = nv04_fifo_chan_fini,
+	.map  = _nouveau_fifo_channel_map,
 	.rd32 = _nouveau_fifo_channel_rd32,
 	.wr32 = _nouveau_fifo_channel_wr32,
 };

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

@@ -363,6 +363,7 @@ nv50_fifo_ofuncs_dma = {
 	.dtor = nv50_fifo_chan_dtor,
 	.init = nv50_fifo_chan_init,
 	.fini = nv50_fifo_chan_fini,
+	.map  = _nouveau_fifo_channel_map,
 	.rd32 = _nouveau_fifo_channel_rd32,
 	.wr32 = _nouveau_fifo_channel_wr32,
 };
@@ -373,6 +374,7 @@ nv50_fifo_ofuncs_ind = {
 	.dtor = nv50_fifo_chan_dtor,
 	.init = nv50_fifo_chan_init,
 	.fini = nv50_fifo_chan_fini,
+	.map  = _nouveau_fifo_channel_map,
 	.rd32 = _nouveau_fifo_channel_rd32,
 	.wr32 = _nouveau_fifo_channel_wr32,
 };

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

@@ -324,6 +324,7 @@ nv84_fifo_ofuncs_dma = {
 	.dtor = nv50_fifo_chan_dtor,
 	.init = nv84_fifo_chan_init,
 	.fini = nv50_fifo_chan_fini,
+	.map  = _nouveau_fifo_channel_map,
 	.rd32 = _nouveau_fifo_channel_rd32,
 	.wr32 = _nouveau_fifo_channel_wr32,
 };
@@ -334,6 +335,7 @@ nv84_fifo_ofuncs_ind = {
 	.dtor = nv50_fifo_chan_dtor,
 	.init = nv84_fifo_chan_init,
 	.fini = nv50_fifo_chan_fini,
+	.map  = _nouveau_fifo_channel_map,
 	.rd32 = _nouveau_fifo_channel_rd32,
 	.wr32 = _nouveau_fifo_channel_wr32,
 };

drivers/gpu/drm/nouveau/core/engine/fifo/nvc0.c

@@ -302,6 +302,7 @@ nvc0_fifo_ofuncs = {
 	.dtor = _nouveau_fifo_channel_dtor,
 	.init = nvc0_fifo_chan_init,
 	.fini = nvc0_fifo_chan_fini,
+	.map  = _nouveau_fifo_channel_map,
 	.rd32 = _nouveau_fifo_channel_rd32,
 	.wr32 = _nouveau_fifo_channel_wr32,
 };

drivers/gpu/drm/nouveau/core/engine/fifo/nve0.c

@@ -336,6 +336,7 @@ nve0_fifo_ofuncs = {
 	.dtor = _nouveau_fifo_channel_dtor,
 	.init = nve0_fifo_chan_init,
 	.fini = nve0_fifo_chan_fini,
+	.map  = _nouveau_fifo_channel_map,
 	.rd32 = _nouveau_fifo_channel_rd32,
 	.wr32 = _nouveau_fifo_channel_wr32,
 };

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

@@ -10,6 +10,7 @@ struct nouveau_fifo_chan {
 	struct nouveau_dmaobj *pushdma;
 	struct nouveau_gpuobj *pushgpu;
 	void __iomem *user;
+	u64 addr;
 	u32 size;
 	u16 chid;
 	atomic_t refcnt; /* NV04_NVSW_SET_REF */
@@ -40,6 +41,7 @@ void nouveau_fifo_channel_destroy(struct nouveau_fifo_chan *);
 #define _nouveau_fifo_channel_fini _nouveau_namedb_fini
 
 void _nouveau_fifo_channel_dtor(struct nouveau_object *);
+int  _nouveau_fifo_channel_map(struct nouveau_object *, u64 *, u32 *);
 u32  _nouveau_fifo_channel_rd32(struct nouveau_object *, u64);
 void _nouveau_fifo_channel_wr32(struct nouveau_object *, u64, u32);
 

drivers/gpu/drm/nouveau/nouveau_chan.c

@@ -291,6 +291,8 @@ nouveau_channel_init(struct nouveau_channel *chan, u32 vram, u32 gart)
 	struct nv_dma_v0 args = {};
 	int ret, i;
 
+	nvif_object_map(chan->object);
+
 	/* allocate dma objects to cover all allowed vram, and gart */
 	if (device->info.family < NV_DEVICE_INFO_V0_FERMI) {
 		if (device->info.family >= NV_DEVICE_INFO_V0_TESLA) {