drm/nouveau: Simplify tile region handling.

Instead of emptying the caches to avoid a race with the PFIFO puller,
go straight ahead and try to recover from it when it happens. Also,
kill pfifo->cache_flush and tile->lock, we don't need them anymore.

Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

6 files changed, 36 insertions, 60 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 6313ba4..912b04b 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -296,7 +296,6 @@ struct nouveau_fifo_engine {
 	void (*disable)(struct drm_device *);
 	void (*enable)(struct drm_device *);
 	bool (*reassign)(struct drm_device *, bool enable);
-	bool (*cache_flush)(struct drm_device *dev);
 	bool (*cache_pull)(struct drm_device *dev, bool enable);
 
 	int  (*channel_id)(struct drm_device *);
@@ -569,10 +568,7 @@ struct drm_nouveau_private {
 	} gart_info;
 
 	/* nv10-nv40 tiling regions */
-	struct {
-		struct nouveau_tile_reg reg[NOUVEAU_MAX_TILE_NR];
-		spinlock_t lock;
-	} tile;
+	struct nouveau_tile_reg tile[NOUVEAU_MAX_TILE_NR];
 
 	/* VRAM/fb configuration */
 	uint64_t vram_size;
@@ -917,7 +913,6 @@ extern int  nv04_fifo_init(struct drm_device *);
 extern void nv04_fifo_disable(struct drm_device *);
 extern void nv04_fifo_enable(struct drm_device *);
 extern bool nv04_fifo_reassign(struct drm_device *, bool);
-extern bool nv04_fifo_cache_flush(struct drm_device *);
 extern bool nv04_fifo_cache_pull(struct drm_device *, bool);
 extern int  nv04_fifo_channel_id(struct drm_device *);
 extern int  nv04_fifo_create_context(struct nouveau_channel *);
@@ -955,7 +950,6 @@ extern void nvc0_fifo_takedown(struct drm_device *);
 extern void nvc0_fifo_disable(struct drm_device *);
 extern void nvc0_fifo_enable(struct drm_device *);
 extern bool nvc0_fifo_reassign(struct drm_device *, bool);
-extern bool nvc0_fifo_cache_flush(struct drm_device *);
 extern bool nvc0_fifo_cache_pull(struct drm_device *, bool);
 extern int  nvc0_fifo_channel_id(struct drm_device *);
 extern int  nvc0_fifo_create_context(struct nouveau_channel *);
diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c
index 02aa9d2..2db01f8 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -47,18 +47,14 @@ nv10_mem_set_region_tiling(struct drm_device *dev, int i, uint32_t addr,
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
 	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
 	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
+	struct nouveau_tile_reg *tile = &dev_priv->tile[i];
 
 	tile->addr = addr;
 	tile->size = size;
 	tile->used = !!pitch;
 	nouveau_fence_unref((void **)&tile->fence);
 
-	if (!pfifo->cache_flush(dev))
-		return;
-
 	pfifo->reassign(dev, false);
-	pfifo->cache_flush(dev);
 	pfifo->cache_pull(dev, false);
 
 	nouveau_wait_for_idle(dev);
@@ -76,34 +72,36 @@ nv10_mem_set_tiling(struct drm_device *dev, uint32_t addr, uint32_t size,
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
-	struct nouveau_tile_reg *tile = dev_priv->tile.reg, *found = NULL;
-	int i;
+	struct nouveau_tile_reg *found = NULL;
+	unsigned long i, flags;
 
-	spin_lock(&dev_priv->tile.lock);
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
 
 	for (i = 0; i < pfb->num_tiles; i++) {
-		if (tile[i].used)
+		struct nouveau_tile_reg *tile = &dev_priv->tile[i];
+
+		if (tile->used)
 			/* Tile region in use. */
 			continue;
 
-		if (tile[i].fence &&
-		    !nouveau_fence_signalled(tile[i].fence, NULL))
+		if (tile->fence &&
+		    !nouveau_fence_signalled(tile->fence, NULL))
 			/* Pending tile region. */
 			continue;
 
-		if (max(tile[i].addr, addr) <
-		    min(tile[i].addr + tile[i].size, addr + size))
+		if (max(tile->addr, addr) <
+		    min(tile->addr + tile->size, addr + size))
 			/* Kill an intersecting tile region. */
 			nv10_mem_set_region_tiling(dev, i, 0, 0, 0);
 
 		if (pitch && !found) {
 			/* Free tile region. */
 			nv10_mem_set_region_tiling(dev, i, addr, size, pitch);
-			found = &tile[i];
+			found = tile;
 		}
 	}
 
-	spin_unlock(&dev_priv->tile.lock);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 
 	return found;
 }
@@ -568,8 +566,6 @@ nouveau_mem_vram_init(struct drm_device *dev)
 		return ret;
 	}
 
-	spin_lock_init(&dev_priv->tile.lock);
-
 	dev_priv->fb_available_size = dev_priv->vram_size;
 	dev_priv->fb_mappable_pages = dev_priv->fb_available_size;
 	if (dev_priv->fb_mappable_pages > pci_resource_len(dev->pdev, 1))
diff --git a/drivers/gpu/drm/nouveau/nouveau_reg.h b/drivers/gpu/drm/nouveau/nouveau_reg.h
index 4b81328..1b42541 100644
--- a/drivers/gpu/drm/nouveau/nouveau_reg.h
+++ b/drivers/gpu/drm/nouveau/nouveau_reg.h
@@ -551,6 +551,8 @@
 #define NV10_PFIFO_CACHE1_DMA_SUBROUTINE                   0x0000324C
 #define NV03_PFIFO_CACHE1_PULL0                            0x00003240
 #define NV04_PFIFO_CACHE1_PULL0                            0x00003250
+#    define NV04_PFIFO_CACHE1_PULL0_HASH_FAILED            0x00000010
+#    define NV04_PFIFO_CACHE1_PULL0_HASH_BUSY              0x00001000
 #define NV03_PFIFO_CACHE1_PULL1                            0x00003250
 #define NV04_PFIFO_CACHE1_PULL1                            0x00003254
 #define NV04_PFIFO_CACHE1_HASH                             0x00003258
diff --git a/drivers/gpu/drm/nouveau/nouveau_state.c b/drivers/gpu/drm/nouveau/nouveau_state.c
index 9a4bf44..be85960 100644
--- a/drivers/gpu/drm/nouveau/nouveau_state.c
+++ b/drivers/gpu/drm/nouveau/nouveau_state.c
@@ -79,7 +79,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->fifo.disable		= nv04_fifo_disable;
 		engine->fifo.enable		= nv04_fifo_enable;
 		engine->fifo.reassign		= nv04_fifo_reassign;
-		engine->fifo.cache_flush	= nv04_fifo_cache_flush;
 		engine->fifo.cache_pull		= nv04_fifo_cache_pull;
 		engine->fifo.channel_id		= nv04_fifo_channel_id;
 		engine->fifo.create_context	= nv04_fifo_create_context;
@@ -131,7 +130,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->fifo.disable		= nv04_fifo_disable;
 		engine->fifo.enable		= nv04_fifo_enable;
 		engine->fifo.reassign		= nv04_fifo_reassign;
-		engine->fifo.cache_flush	= nv04_fifo_cache_flush;
 		engine->fifo.cache_pull		= nv04_fifo_cache_pull;
 		engine->fifo.channel_id		= nv10_fifo_channel_id;
 		engine->fifo.create_context	= nv10_fifo_create_context;
@@ -183,7 +181,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->fifo.disable		= nv04_fifo_disable;
 		engine->fifo.enable		= nv04_fifo_enable;
 		engine->fifo.reassign		= nv04_fifo_reassign;
-		engine->fifo.cache_flush	= nv04_fifo_cache_flush;
 		engine->fifo.cache_pull		= nv04_fifo_cache_pull;
 		engine->fifo.channel_id		= nv10_fifo_channel_id;
 		engine->fifo.create_context	= nv10_fifo_create_context;
@@ -235,7 +232,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->fifo.disable		= nv04_fifo_disable;
 		engine->fifo.enable		= nv04_fifo_enable;
 		engine->fifo.reassign		= nv04_fifo_reassign;
-		engine->fifo.cache_flush	= nv04_fifo_cache_flush;
 		engine->fifo.cache_pull		= nv04_fifo_cache_pull;
 		engine->fifo.channel_id		= nv10_fifo_channel_id;
 		engine->fifo.create_context	= nv10_fifo_create_context;
@@ -288,7 +284,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->fifo.disable		= nv04_fifo_disable;
 		engine->fifo.enable		= nv04_fifo_enable;
 		engine->fifo.reassign		= nv04_fifo_reassign;
-		engine->fifo.cache_flush	= nv04_fifo_cache_flush;
 		engine->fifo.cache_pull		= nv04_fifo_cache_pull;
 		engine->fifo.channel_id		= nv10_fifo_channel_id;
 		engine->fifo.create_context	= nv40_fifo_create_context;
diff --git a/drivers/gpu/drm/nouveau/nv04_fifo.c b/drivers/gpu/drm/nouveau/nv04_fifo.c
index 64dc0e2..708293b 100644
--- a/drivers/gpu/drm/nouveau/nv04_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv04_fifo.c
@@ -71,37 +71,32 @@ nv04_fifo_reassign(struct drm_device *dev, bool enable)
 }
 
 bool
-nv04_fifo_cache_flush(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_timer_engine *ptimer = &dev_priv->engine.timer;
-	uint64_t start = ptimer->read(dev);
-
-	do {
-		if (nv_rd32(dev, NV03_PFIFO_CACHE1_GET) ==
-		    nv_rd32(dev, NV03_PFIFO_CACHE1_PUT))
-			return true;
-
-	} while (ptimer->read(dev) - start < 100000000);
-
-	NV_ERROR(dev, "Timeout flushing the PFIFO cache.\n");
-
-	return false;
-}
-
-bool
 nv04_fifo_cache_pull(struct drm_device *dev, bool enable)
 {
-	uint32_t pull = nv_rd32(dev, NV04_PFIFO_CACHE1_PULL0);
+	int pull = nv_mask(dev, NV04_PFIFO_CACHE1_PULL0, 1, enable);
+
+	if (!enable) {
+		/* In some cases the PFIFO puller may be left in an
+		 * inconsistent state if you try to stop it when it's
+		 * busy translating handles. Sometimes you get a
+		 * PFIFO_CACHE_ERROR, sometimes it just fails silently
+		 * sending incorrect instance offsets to PGRAPH after
+		 * it's started up again. To avoid the latter we
+		 * invalidate the most recently calculated instance.
+		 */
+		if (!nv_wait(dev, NV04_PFIFO_CACHE1_PULL0,
+			     NV04_PFIFO_CACHE1_PULL0_HASH_BUSY, 0))
+			NV_ERROR(dev, "Timeout idling the PFIFO puller.\n");
+
+		if (nv_rd32(dev, NV04_PFIFO_CACHE1_PULL0) &
+		    NV04_PFIFO_CACHE1_PULL0_HASH_FAILED)
+			nv_wr32(dev, NV03_PFIFO_INTR_0,
+				NV_PFIFO_INTR_CACHE_ERROR);
 
-	if (enable) {
-		nv_wr32(dev, NV04_PFIFO_CACHE1_PULL0, pull | 1);
-	} else {
-		nv_wr32(dev, NV04_PFIFO_CACHE1_PULL0, pull & ~1);
 		nv_wr32(dev, NV04_PFIFO_CACHE1_HASH, 0);
 	}
 
-	return !!(pull & 1);
+	return pull & 1;
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/nvc0_fifo.c b/drivers/gpu/drm/nouveau/nvc0_fifo.c
index d643758..890c2b9 100644
--- a/drivers/gpu/drm/nouveau/nvc0_fifo.c
+++ b/drivers/gpu/drm/nouveau/nvc0_fifo.c
@@ -43,12 +43,6 @@ nvc0_fifo_reassign(struct drm_device *dev, bool enable)
 }
 
 bool
-nvc0_fifo_cache_flush(struct drm_device *dev)
-{
-	return true;
-}
-
-bool
 nvc0_fifo_cache_pull(struct drm_device *dev, bool enable)
 {
 	return false;