Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "This has a bunch of nouveau fixes, as Ben has been hibernating and has
  lots of small fixes for lots of bugs across nouveau.

  Radeon has one major fix for hdmi/dp audio regression that is larger
  than Alex would like, but seems to fix up a fair few bugs, along with
  some misc fixes.

  And a few msm fixes, one of which is also a bit large.

  But nothing in here seems insane or crazy for this stage, just more
  than I'd like"

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux: (33 commits)
  drm/msm/mdp5: release SMB (shared memory blocks) in various cases
  drm/msm: change to uninterruptible wait in atomic commit
  drm/msm: mdp4: Fix drm_framebuffer dereference crash
  drm/msm: fix msm_gem_prime_get_sg_table()
  drm/amdgpu: add new parameter to seperate map and unmap
  drm/amdgpu: hdp_flush is not needed for inside IB
  drm/amdgpu: different emit_ib for gfx and compute
  drm/amdgpu: information leak in amdgpu_info_ioctl()
  drm/amdgpu: clean up init sequence for failures
  drm/radeon/combios: add some validation of lvds values
  drm/radeon: rework audio modeset to handle non-audio hdmi features
  drm/radeon: rework audio detect (v4)
  drm/amdgpu: Drop drm/ prefix for including drm.h in amdgpu_drm.h
  drm/radeon: Drop drm/ prefix for including drm.h in radeon_drm.h
  drm/nouveau/nouveau/ttm: fix tiled system memory with Maxwell
  drm/nouveau/kms/nv50-: guard against enabling cursor on disabled heads
  drm/nouveau/fbcon/g80: reduce PUSH_SPACE alloc, fire ring on accel init
  drm/nouveau/fbcon/gf100-: reduce RING_SPACE allocation
  drm/nouveau/fbcon/nv11-: correctly account for ring space usage
  drm/nouveau/bios: add proper support for opcode 0x59
  ...

1 files changed, 36 insertions, 11 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index 1c7c992dea37..9e1d4ddbf475 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -3753,7 +3753,7 @@ static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 	amdgpu_ring_write(ring, 0x20); /* poll interval */
 }
 
-static void gfx_v8_0_ring_emit_ib(struct amdgpu_ring *ring,
+static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
 				  struct amdgpu_ib *ib)
 {
 	bool need_ctx_switch = ring->current_ctx != ib->ctx;
@@ -3761,15 +3761,10 @@ static void gfx_v8_0_ring_emit_ib(struct amdgpu_ring *ring,
 	u32 next_rptr = ring->wptr + 5;
 
 	/* drop the CE preamble IB for the same context */
-	if ((ring->type == AMDGPU_RING_TYPE_GFX) &&
-	    (ib->flags & AMDGPU_IB_FLAG_PREAMBLE) &&
-	    !need_ctx_switch)
+	if ((ib->flags & AMDGPU_IB_FLAG_PREAMBLE) && !need_ctx_switch)
 		return;
 
-	if (ring->type == AMDGPU_RING_TYPE_COMPUTE)
-		control |= INDIRECT_BUFFER_VALID;
-
-	if (need_ctx_switch && ring->type == AMDGPU_RING_TYPE_GFX)
+	if (need_ctx_switch)
 		next_rptr += 2;
 
 	next_rptr += 4;
@@ -3780,7 +3775,7 @@ static void gfx_v8_0_ring_emit_ib(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, next_rptr);
 
 	/* insert SWITCH_BUFFER packet before first IB in the ring frame */
-	if (need_ctx_switch && ring->type == AMDGPU_RING_TYPE_GFX) {
+	if (need_ctx_switch) {
 		amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
 		amdgpu_ring_write(ring, 0);
 	}
@@ -3803,6 +3798,36 @@ static void gfx_v8_0_ring_emit_ib(struct amdgpu_ring *ring,
 	amdgpu_ring_write(ring, control);
 }
 
+static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
+				  struct amdgpu_ib *ib)
+{
+	u32 header, control = 0;
+	u32 next_rptr = ring->wptr + 5;
+
+	control |= INDIRECT_BUFFER_VALID;
+
+	next_rptr += 4;
+	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
+	amdgpu_ring_write(ring, WRITE_DATA_DST_SEL(5) | WR_CONFIRM);
+	amdgpu_ring_write(ring, ring->next_rptr_gpu_addr & 0xfffffffc);
+	amdgpu_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
+	amdgpu_ring_write(ring, next_rptr);
+
+	header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
+
+	control |= ib->length_dw |
+			   (ib->vm ? (ib->vm->ids[ring->idx].id << 24) : 0);
+
+	amdgpu_ring_write(ring, header);
+	amdgpu_ring_write(ring,
+#ifdef __BIG_ENDIAN
+					  (2 << 0) |
+#endif
+					  (ib->gpu_addr & 0xFFFFFFFC));
+	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
+	amdgpu_ring_write(ring, control);
+}
+
 static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
 					 u64 seq, unsigned flags)
 {
@@ -4224,7 +4249,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
 	.get_wptr = gfx_v8_0_ring_get_wptr_gfx,
 	.set_wptr = gfx_v8_0_ring_set_wptr_gfx,
 	.parse_cs = NULL,
-	.emit_ib = gfx_v8_0_ring_emit_ib,
+	.emit_ib = gfx_v8_0_ring_emit_ib_gfx,
 	.emit_fence = gfx_v8_0_ring_emit_fence_gfx,
 	.emit_semaphore = gfx_v8_0_ring_emit_semaphore,
 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
@@ -4240,7 +4265,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
 	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
 	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
 	.parse_cs = NULL,
-	.emit_ib = gfx_v8_0_ring_emit_ib,
+	.emit_ib = gfx_v8_0_ring_emit_ib_compute,
 	.emit_fence = gfx_v8_0_ring_emit_fence_compute,
 	.emit_semaphore = gfx_v8_0_ring_emit_semaphore,
 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,