drm/nouveau: Fix pushbufs over the 4GB mark.

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

drivers/gpu/drm/nouveau/nouveau_channel.c

@@ -187,6 +187,8 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 	nouveau_dma_pre_init(chan);
 	chan->user_put = 0x40;
 	chan->user_get = 0x44;
+	if (dev_priv->card_type >= NV_50)
+                chan->user_get_hi = 0x60;
 
 	/* disable the fifo caches */
 	pfifo->reassign(dev, false);

drivers/gpu/drm/nouveau/nouveau_debugfs.c

@@ -44,7 +44,7 @@ nouveau_debugfs_channel_info(struct seq_file *m, void *data)
 	seq_printf(m, "channel id    : %d\n", chan->id);
 
 	seq_printf(m, "cpu fifo state:\n");
-	seq_printf(m, "          base: 0x%08x\n", chan->pushbuf_base);
+	seq_printf(m, "          base: 0x%10llx\n", chan->pushbuf_base);
 	seq_printf(m, "           max: 0x%08x\n", chan->dma.max << 2);
 	seq_printf(m, "           cur: 0x%08x\n", chan->dma.cur << 2);
 	seq_printf(m, "           put: 0x%08x\n", chan->dma.put << 2);

drivers/gpu/drm/nouveau/nouveau_dma.c

@@ -134,11 +134,13 @@ OUT_RINGp(struct nouveau_channel *chan, const void *data, unsigned nr_dwords)
  *  -EBUSY if timeout exceeded
  */
 static inline int
-READ_GET(struct nouveau_channel *chan, uint32_t *prev_get, uint32_t *timeout)
+READ_GET(struct nouveau_channel *chan, uint64_t *prev_get, int *timeout)
 {
-	uint32_t val;
+	uint64_t val;
 
 	val = nvchan_rd32(chan, chan->user_get);
+        if (chan->user_get_hi)
+                val |= (uint64_t)nvchan_rd32(chan, chan->user_get_hi) << 32;
 
 	/* reset counter as long as GET is still advancing, this is
 	 * to avoid misdetecting a GPU lockup if the GPU happens to
@@ -218,8 +220,8 @@ nv50_dma_push_wait(struct nouveau_channel *chan, int count)
 static int
 nv50_dma_wait(struct nouveau_channel *chan, int slots, int count)
 {
-	uint32_t cnt = 0, prev_get = 0;
-	int ret;
+	uint64_t prev_get = 0;
+	int ret, cnt = 0;
 
 	ret = nv50_dma_push_wait(chan, slots + 1);
 	if (unlikely(ret))
@@ -261,8 +263,8 @@ nv50_dma_wait(struct nouveau_channel *chan, int slots, int count)
 int
 nouveau_dma_wait(struct nouveau_channel *chan, int slots, int size)
 {
-	uint32_t prev_get = 0, cnt = 0;
-	int get;
+	uint64_t prev_get = 0;
+	int cnt = 0, get;
 
 	if (chan->dma.ib_max)
 		return nv50_dma_wait(chan, slots, size);

drivers/gpu/drm/nouveau/nouveau_drv.h

@@ -232,6 +232,7 @@ struct nouveau_channel {
 	/* mapping of the regs controlling the fifo */
 	void __iomem *user;
 	uint32_t user_get;
+	uint32_t user_get_hi;
 	uint32_t user_put;
 
 	/* Fencing */
@@ -249,7 +250,7 @@ struct nouveau_channel {
 	struct nouveau_gpuobj *pushbuf;
 	struct nouveau_bo     *pushbuf_bo;
 	struct nouveau_vma     pushbuf_vma;
-	uint32_t               pushbuf_base;
+	uint64_t               pushbuf_base;
 
 	/* Notifier memory */
 	struct nouveau_bo *notifier_bo;

drivers/gpu/drm/nouveau/nv50_fifo.c

@@ -230,6 +230,7 @@ nv50_fifo_create_context(struct nouveau_channel *chan)
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_gpuobj *ramfc = NULL;
+        uint64_t ib_offset = chan->pushbuf_base + chan->dma.ib_base * 4;
 	unsigned long flags;
 	int ret;
 
@@ -280,8 +281,9 @@ nv50_fifo_create_context(struct nouveau_channel *chan)
 	nv_wo32(ramfc, 0x7c, 0x30000001);
 	nv_wo32(ramfc, 0x78, 0x00000000);
 	nv_wo32(ramfc, 0x3c, 0x403f6078);
-	nv_wo32(ramfc, 0x50, chan->pushbuf_base + chan->dma.ib_base * 4);
-	nv_wo32(ramfc, 0x54, drm_order(chan->dma.ib_max + 1) << 16);
+	nv_wo32(ramfc, 0x50, lower_32_bits(ib_offset));
+	nv_wo32(ramfc, 0x54, upper_32_bits(ib_offset) |
+                drm_order(chan->dma.ib_max + 1) << 16);
 
 	if (dev_priv->chipset != 0x50) {
 		nv_wo32(chan->ramin, 0, chan->id);