drm/amdgpu: cleanup HDP flush handling

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Monk Liu <monk.liu@amd.com>

7 files changed, 18 insertions, 54 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index aaa62f33870f..72d9d9ec5c9f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -382,6 +382,7 @@ struct amdgpu_ring_funcs {
                                bool emit_wait);
         void (*emit_vm_flush)(struct amdgpu_ring *ring, unsigned vm_id,
                               uint64_t pd_addr);
+        void (*emit_hdp_flush)(struct amdgpu_ring *ring);
         void (*emit_gds_switch)(struct amdgpu_ring *ring, uint32_t vmid,
                                 uint32_t gds_base, uint32_t gds_size,
                                 uint32_t gws_base, uint32_t gws_size,
@@ -892,7 +893,6 @@ struct amdgpu_ib {
         struct amdgpu_fence             *fence;
         struct amdgpu_user_fence        *user;
         struct amdgpu_vm                *vm;
-        bool                            flush_hdp_writefifo;
         struct amdgpu_sync              sync;
         uint32_t                        gds_base, gds_size;
         uint32_t                        gws_base, gws_size;
@@ -2203,6 +2203,7 @@ static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v)
 #define amdgpu_ring_emit_fence(r, addr, seq, write64bit) (r)->funcs->emit_fence((r), (addr), (seq), (write64bit))
 #define amdgpu_ring_emit_semaphore(r, semaphore, emit_wait) (r)->funcs->emit_semaphore((r), (semaphore), (emit_wait))
 #define amdgpu_ring_emit_gds_switch(r, v, db, ds, wb, ws, ab, as) (r)->funcs->emit_gds_switch((r), (v), (db), (ds), (wb), (ws), (ab), (as))
+#define amdgpu_ring_emit_hdp_flush(r) (r)->funcs->emit_hdp_flush((r))
 #define amdgpu_ih_get_wptr(adev) (adev)->irq.ih_funcs->get_wptr((adev))
 #define amdgpu_ih_decode_iv(adev, iv) (adev)->irq.ih_funcs->decode_iv((adev), (iv))
 #define amdgpu_ih_set_rptr(adev) (adev)->irq.ih_funcs->set_rptr((adev))
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index f2442b2cec00..66c66f83b407 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -143,7 +143,6 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
         struct amdgpu_ib *ib = &ibs[0];
         unsigned i;
         int r = 0;
-        bool flush_hdp = true;
 
         if (num_ibs == 0)
                 return -EINVAL;
@@ -185,6 +184,9 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
                                             ib->gws_base, ib->gws_size,
                                             ib->oa_base, ib->oa_size);
 
+        if (ring->funcs->emit_hdp_flush)
+                amdgpu_ring_emit_hdp_flush(ring);
+
         for (i = 0; i < num_ibs; ++i) {
                 ib = &ibs[i];
 
@@ -192,8 +194,6 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
                         amdgpu_ring_unlock_undo(ring);
                         return -EINVAL;
                 }
-                ib->flush_hdp_writefifo = flush_hdp;
-                flush_hdp = false;
                 amdgpu_ring_emit_ib(ring, ib);
         }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
index 037e3db69547..d5055ea4d112 100644
--- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
+++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
@@ -188,8 +188,6 @@ static void cik_sdma_ring_set_wptr(struct amdgpu_ring *ring)
         WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], (ring->wptr << 2) & 0x3fffc);
 }
 
-static void cik_sdma_hdp_flush_ring_emit(struct amdgpu_ring *);
-
 /**
  * cik_sdma_ring_emit_ib - Schedule an IB on the DMA engine
  *
@@ -204,9 +202,6 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
         u32 extra_bits = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf;
         u32 next_rptr = ring->wptr + 5;
 
-        if (ib->flush_hdp_writefifo)
-                next_rptr += 6;
-
         while ((next_rptr & 7) != 4)
                 next_rptr++;
 
@@ -217,11 +212,6 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
         amdgpu_ring_write(ring, 1); /* number of DWs to follow */
         amdgpu_ring_write(ring, next_rptr);
 
-        if (ib->flush_hdp_writefifo) {
-                /* flush HDP */
-                cik_sdma_hdp_flush_ring_emit(ring);
-        }
-
         /* IB packet must end on a 8 DW boundary */
         while ((ring->wptr & 7) != 4)
                 amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0));
@@ -233,13 +223,13 @@ static void cik_sdma_ring_emit_ib(struct amdgpu_ring *ring,
 }
 
 /**
- * cik_sdma_hdp_flush_ring_emit - emit an hdp flush on the DMA ring
+ * cik_sdma_ring_emit_hdp_flush - emit an hdp flush on the DMA ring
  *
  * @ring: amdgpu ring pointer
  *
  * Emit an hdp flush packet on the requested DMA ring.
  */
-static void cik_sdma_hdp_flush_ring_emit(struct amdgpu_ring *ring)
+static void cik_sdma_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 {
         u32 extra_bits = (SDMA_POLL_REG_MEM_EXTRA_OP(1) |
                           SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */
@@ -1317,6 +1307,7 @@ static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = {
         .emit_fence = cik_sdma_ring_emit_fence,
         .emit_semaphore = cik_sdma_ring_emit_semaphore,
         .emit_vm_flush = cik_sdma_ring_emit_vm_flush,
+        .emit_hdp_flush = cik_sdma_ring_emit_hdp_flush,
         .test_ring = cik_sdma_ring_test_ring,
         .test_ib = cik_sdma_ring_test_ib,
         .is_lockup = cik_sdma_ring_is_lockup,
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
index cec46ebae5f7..855b5272f4b5 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
@@ -2366,14 +2366,14 @@ static int gfx_v7_0_ring_test_ring(struct amdgpu_ring *ring)
 }
 
 /**
- * gfx_v7_0_hdp_flush_cp_ring_emit - emit an hdp flush on the cp
+ * gfx_v7_0_ring_emit_hdp - emit an hdp flush on the cp
  *
  * @adev: amdgpu_device pointer
  * @ridx: amdgpu ring index
  *
  * Emits an hdp flush on the cp.
  */
-static void gfx_v7_0_hdp_flush_cp_ring_emit(struct amdgpu_ring *ring)
+static void gfx_v7_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 {
         u32 ref_and_mask;
 
@@ -2528,9 +2528,6 @@ static void gfx_v7_0_ring_emit_ib(struct amdgpu_ring *ring,
         if (ring->type == AMDGPU_RING_TYPE_COMPUTE)
                 control |= INDIRECT_BUFFER_VALID;
 
-        if (ib->flush_hdp_writefifo)
-                next_rptr += 7;
-
         if (ring->need_ctx_switch && ring->type == AMDGPU_RING_TYPE_GFX)
                 next_rptr += 2;
 
@@ -2541,9 +2538,6 @@ static void gfx_v7_0_ring_emit_ib(struct amdgpu_ring *ring,
         amdgpu_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
         amdgpu_ring_write(ring, next_rptr);
 
-        if (ib->flush_hdp_writefifo)
-                gfx_v7_0_hdp_flush_cp_ring_emit(ring);
-
         /* insert SWITCH_BUFFER packet before first IB in the ring frame */
         if (ring->need_ctx_switch && ring->type == AMDGPU_RING_TYPE_GFX) {
                 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
@@ -5522,6 +5516,7 @@ static const struct amdgpu_ring_funcs gfx_v7_0_ring_funcs_gfx = {
         .emit_semaphore = gfx_v7_0_ring_emit_semaphore,
         .emit_vm_flush = gfx_v7_0_ring_emit_vm_flush,
         .emit_gds_switch = gfx_v7_0_ring_emit_gds_switch,
+        .emit_hdp_flush = gfx_v7_0_ring_emit_hdp_flush,
         .test_ring = gfx_v7_0_ring_test_ring,
         .test_ib = gfx_v7_0_ring_test_ib,
         .is_lockup = gfx_v7_0_ring_is_lockup,
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
index fc8c46209db4..63ed3b01cea1 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
@@ -3610,7 +3610,7 @@ static void gfx_v8_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
         }
 }
 
-static void gfx_v8_0_hdp_flush_cp_ring_emit(struct amdgpu_ring *ring)
+static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 {
         u32 ref_and_mask, reg_mem_engine;
 
@@ -3657,9 +3657,6 @@ static void gfx_v8_0_ring_emit_ib(struct amdgpu_ring *ring,
         if (ring->type == AMDGPU_RING_TYPE_COMPUTE)
                 control |= INDIRECT_BUFFER_VALID;
 
-        if (ib->flush_hdp_writefifo)
-                next_rptr += 7;
-
         if (ring->need_ctx_switch && ring->type == AMDGPU_RING_TYPE_GFX)
                 next_rptr += 2;
 
@@ -3670,9 +3667,6 @@ static void gfx_v8_0_ring_emit_ib(struct amdgpu_ring *ring,
         amdgpu_ring_write(ring, upper_32_bits(ring->next_rptr_gpu_addr) & 0xffffffff);
         amdgpu_ring_write(ring, next_rptr);
 
-        if (ib->flush_hdp_writefifo)
-                gfx_v8_0_hdp_flush_cp_ring_emit(ring);
-
         /* insert SWITCH_BUFFER packet before first IB in the ring frame */
         if (ring->need_ctx_switch && ring->type == AMDGPU_RING_TYPE_GFX) {
                 amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
@@ -4149,6 +4143,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
         .emit_semaphore = gfx_v8_0_ring_emit_semaphore,
         .emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
         .emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
+        .emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
         .test_ring = gfx_v8_0_ring_test_ring,
         .test_ib = gfx_v8_0_ring_test_ib,
         .is_lockup = gfx_v8_0_ring_is_lockup,
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
index 389509aeddf8..f1423a412867 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
@@ -214,8 +214,6 @@ static void sdma_v2_4_ring_set_wptr(struct amdgpu_ring *ring)
         WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2);
 }
 
-static void sdma_v2_4_hdp_flush_ring_emit(struct amdgpu_ring *);
-
 /**
  * sdma_v2_4_ring_emit_ib - Schedule an IB on the DMA engine
  *
@@ -230,9 +228,6 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
         u32 vmid = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf;
         u32 next_rptr = ring->wptr + 5;
 
-        if (ib->flush_hdp_writefifo)
-                next_rptr += 6;
-
         while ((next_rptr & 7) != 2)
                 next_rptr++;
 
@@ -245,11 +240,6 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
         amdgpu_ring_write(ring, SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(1));
         amdgpu_ring_write(ring, next_rptr);
 
-        if (ib->flush_hdp_writefifo) {
-                /* flush HDP */
-                sdma_v2_4_hdp_flush_ring_emit(ring);
-        }
-
         /* IB packet must end on a 8 DW boundary */
         while ((ring->wptr & 7) != 2)
                 amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_NOP));
@@ -271,7 +261,7 @@ static void sdma_v2_4_ring_emit_ib(struct amdgpu_ring *ring,
  *
  * Emit an hdp flush packet on the requested DMA ring.
  */
-static void sdma_v2_4_hdp_flush_ring_emit(struct amdgpu_ring *ring)
+static void sdma_v2_4_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 {
         u32 ref_and_mask = 0;
 
@@ -1340,6 +1330,7 @@ static const struct amdgpu_ring_funcs sdma_v2_4_ring_funcs = {
         .emit_fence = sdma_v2_4_ring_emit_fence,
         .emit_semaphore = sdma_v2_4_ring_emit_semaphore,
         .emit_vm_flush = sdma_v2_4_ring_emit_vm_flush,
+        .emit_hdp_flush = sdma_v2_4_ring_emit_hdp_flush,
         .test_ring = sdma_v2_4_ring_test_ring,
         .test_ib = sdma_v2_4_ring_test_ib,
         .is_lockup = sdma_v2_4_ring_is_lockup,
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
index d3eda315e719..7c4abff5005d 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
@@ -269,8 +269,6 @@ static void sdma_v3_0_ring_set_wptr(struct amdgpu_ring *ring)
         }
 }
 
-static void sdma_v3_0_hdp_flush_ring_emit(struct amdgpu_ring *);
-
 /**
  * sdma_v3_0_ring_emit_ib - Schedule an IB on the DMA engine
  *
@@ -285,9 +283,6 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
         u32 vmid = (ib->vm ? ib->vm->ids[ring->idx].id : 0) & 0xf;
         u32 next_rptr = ring->wptr + 5;
 
-        if (ib->flush_hdp_writefifo)
-                next_rptr += 6;
-
         while ((next_rptr & 7) != 2)
                 next_rptr++;
         next_rptr += 6;
@@ -299,11 +294,6 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
         amdgpu_ring_write(ring, SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(1));
         amdgpu_ring_write(ring, next_rptr);
 
-        /* flush HDP */
-        if (ib->flush_hdp_writefifo) {
-                sdma_v3_0_hdp_flush_ring_emit(ring);
-        }
-
         /* IB packet must end on a 8 DW boundary */
         while ((ring->wptr & 7) != 2)
                 amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_NOP));
@@ -320,13 +310,13 @@ static void sdma_v3_0_ring_emit_ib(struct amdgpu_ring *ring,
 }
 
 /**
- * sdma_v3_0_hdp_flush_ring_emit - emit an hdp flush on the DMA ring
+ * sdma_v3_0_ring_emit_hdp_flush - emit an hdp flush on the DMA ring
  *
  * @ring: amdgpu ring pointer
  *
  * Emit an hdp flush packet on the requested DMA ring.
  */
-static void sdma_v3_0_hdp_flush_ring_emit(struct amdgpu_ring *ring)
+static void sdma_v3_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
 {
         u32 ref_and_mask = 0;
 
@@ -1407,6 +1397,7 @@ static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = {
         .emit_fence = sdma_v3_0_ring_emit_fence,
         .emit_semaphore = sdma_v3_0_ring_emit_semaphore,
         .emit_vm_flush = sdma_v3_0_ring_emit_vm_flush,
+        .emit_hdp_flush = sdma_v3_0_ring_emit_hdp_flush,
         .test_ring = sdma_v3_0_ring_test_ring,
         .test_ib = sdma_v3_0_ring_test_ib,
         .is_lockup = sdma_v3_0_ring_is_lockup,