nouveau: send back a debug message when waiting for a fence to complete

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

src/gallium/drivers/nouveau/nouveau_buffer.c

@@ -225,21 +225,22 @@ nouveau_transfer_write(struct nouveau_context *nv, struct nouveau_transfer *tx,
  * for write/read by waiting on the buffer's relevant fences.
  */
 static inline bool
-nouveau_buffer_sync(struct nv04_resource *buf, unsigned rw)
+nouveau_buffer_sync(struct nouveau_context *nv,
+                    struct nv04_resource *buf, unsigned rw)
 {
    if (rw == PIPE_TRANSFER_READ) {
       if (!buf->fence_wr)
          return true;
       NOUVEAU_DRV_STAT_RES(buf, buf_non_kernel_fence_sync_count,
                            !nouveau_fence_signalled(buf->fence_wr));
-      if (!nouveau_fence_wait(buf->fence_wr))
+      if (!nouveau_fence_wait(buf->fence_wr, &nv->debug))
          return false;
    } else {
       if (!buf->fence)
          return true;
       NOUVEAU_DRV_STAT_RES(buf, buf_non_kernel_fence_sync_count,
                            !nouveau_fence_signalled(buf->fence));
-      if (!nouveau_fence_wait(buf->fence))
+      if (!nouveau_fence_wait(buf->fence, &nv->debug))
          return false;
 
       nouveau_fence_ref(NULL, &buf->fence);
@@ -478,7 +479,7 @@ nouveau_buffer_transfer_map(struct pipe_context *pipe,
       if (unlikely(usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE)) {
          /* Discarding was not possible, must sync because
           * subsequent transfers might use UNSYNCHRONIZED. */
-         nouveau_buffer_sync(buf, usage & PIPE_TRANSFER_READ_WRITE);
+         nouveau_buffer_sync(nv, buf, usage & PIPE_TRANSFER_READ_WRITE);
       } else
       if (usage & PIPE_TRANSFER_DISCARD_RANGE) {
          /* The whole range is being discarded, so it doesn't matter what was
@@ -490,7 +491,7 @@ nouveau_buffer_transfer_map(struct pipe_context *pipe,
          if (usage & PIPE_TRANSFER_DONTBLOCK)
             map = NULL;
          else
-            nouveau_buffer_sync(buf, usage & PIPE_TRANSFER_READ_WRITE);
+            nouveau_buffer_sync(nv, buf, usage & PIPE_TRANSFER_READ_WRITE);
       } else {
          /* It is expected that the returned buffer be a representation of the
           * data in question, so we must copy it over from the buffer. */
@@ -615,7 +616,7 @@ nouveau_resource_map_offset(struct nouveau_context *nv,
    if (res->mm) {
       unsigned rw;
       rw = (flags & NOUVEAU_BO_WR) ? PIPE_TRANSFER_WRITE : PIPE_TRANSFER_READ;
-      nouveau_buffer_sync(res, rw);
+      nouveau_buffer_sync(nv, res, rw);
       if (nouveau_bo_map(res->bo, 0, NULL))
          return NULL;
    } else {

src/gallium/drivers/nouveau/nouveau_context.h

@@ -2,6 +2,7 @@
 #define __NOUVEAU_CONTEXT_H__
 
 #include "pipe/p_context.h"
+#include "pipe/p_state.h"
 #include <nouveau.h>
 
 #define NOUVEAU_MAX_SCRATCH_BUFS 4

src/gallium/drivers/nouveau/nouveau_fence.c

@@ -23,6 +23,7 @@
 #include "nouveau_screen.h"
 #include "nouveau_winsys.h"
 #include "nouveau_fence.h"
+#include "os/os_time.h"
 
 #ifdef PIPE_OS_UNIX
 #include <sched.h>
@@ -182,10 +183,11 @@ nouveau_fence_signalled(struct nouveau_fence *fence)
 }
 
 bool
-nouveau_fence_wait(struct nouveau_fence *fence)
+nouveau_fence_wait(struct nouveau_fence *fence, struct pipe_debug_callback *debug)
 {
    struct nouveau_screen *screen = fence->screen;
    uint32_t spins = 0;
+   int64_t start = 0;
 
    /* wtf, someone is waiting on a fence in flush_notify handler? */
    assert(fence->state != NOUVEAU_FENCE_STATE_EMITTING);
@@ -206,11 +208,19 @@ nouveau_fence_wait(struct nouveau_fence *fence)
    if (fence == screen->fence.current)
       nouveau_fence_next(screen);
 
+   if (debug && debug->debug_message)
+      start = os_time_get_nano();
+
    do {
       nouveau_fence_update(screen, false);
 
-      if (fence->state == NOUVEAU_FENCE_STATE_SIGNALLED)
+      if (fence->state == NOUVEAU_FENCE_STATE_SIGNALLED) {
+         if (debug && debug->debug_message)
+            pipe_debug_message(debug, PERF_INFO,
+                               "stalled %.3f ms waiting for fence",
+                               (os_time_get_nano() - start) / 1000000.f);
          return true;
+      }
       if (!spins)
          NOUVEAU_DRV_STAT(screen, any_non_kernel_fence_sync_count, 1);
       spins++;

src/gallium/drivers/nouveau/nouveau_fence.h

@@ -11,6 +11,8 @@
 #define NOUVEAU_FENCE_STATE_FLUSHED   3
 #define NOUVEAU_FENCE_STATE_SIGNALLED 4
 
+struct pipe_debug_callback;
+
 struct nouveau_fence_work {
    struct list_head list;
    void (*func)(void *);
@@ -34,7 +36,7 @@ bool nouveau_fence_new(struct nouveau_screen *, struct nouveau_fence **,
 bool nouveau_fence_work(struct nouveau_fence *, void (*)(void *), void *);
 void nouveau_fence_update(struct nouveau_screen *, bool flushed);
 void nouveau_fence_next(struct nouveau_screen *);
-bool nouveau_fence_wait(struct nouveau_fence *);
+bool nouveau_fence_wait(struct nouveau_fence *, struct pipe_debug_callback *);
 bool nouveau_fence_signalled(struct nouveau_fence *);
 
 void nouveau_fence_unref_bo(void *data); /* generic unref bo callback */

src/gallium/drivers/nouveau/nouveau_screen.c

@@ -76,7 +76,7 @@ nouveau_screen_fence_finish(struct pipe_screen *screen,
    if (!timeout)
       return nouveau_fence_signalled(nouveau_fence(pfence));
 
-   return nouveau_fence_wait(nouveau_fence(pfence));
+   return nouveau_fence_wait(nouveau_fence(pfence), NULL);
 }
 
 

src/gallium/drivers/nouveau/nv30/nv30_screen.c

@@ -383,7 +383,7 @@ nv30_screen_destroy(struct pipe_screen *pscreen)
        * _current_ one, and remove both.
        */
       nouveau_fence_ref(screen->base.fence.current, &current);
-      nouveau_fence_wait(current);
+      nouveau_fence_wait(current, NULL);
       nouveau_fence_ref(NULL, &current);
       nouveau_fence_ref(NULL, &screen->base.fence.current);
    }

src/gallium/drivers/nouveau/nv50/nv50_screen.c

@@ -350,7 +350,7 @@ nv50_screen_destroy(struct pipe_screen *pscreen)
        * _current_ one, and remove both.
        */
       nouveau_fence_ref(screen->base.fence.current, &current);
-      nouveau_fence_wait(current);
+      nouveau_fence_wait(current, NULL);
       nouveau_fence_ref(NULL, &current);
       nouveau_fence_ref(NULL, &screen->base.fence.current);
    }

src/gallium/drivers/nouveau/nv50/nv50_vbo.c

@@ -636,7 +636,7 @@ nv50_draw_elements(struct nv50_context *nv50, bool shorten,
        * pushbuf submit, but it's probably not a big performance difference.
        */
       if (buf->fence_wr && !nouveau_fence_signalled(buf->fence_wr))
-         nouveau_fence_wait(buf->fence_wr);
+         nouveau_fence_wait(buf->fence_wr, &nv50->base.debug);
 
       while (instance_count--) {
          BEGIN_NV04(push, NV50_3D(VERTEX_BEGIN_GL), 1);

src/gallium/drivers/nouveau/nvc0/nvc0_screen.c

@@ -415,7 +415,7 @@ nvc0_screen_destroy(struct pipe_screen *pscreen)
        * _current_ one, and remove both.
        */
       nouveau_fence_ref(screen->base.fence.current, &current);
-      nouveau_fence_wait(current);
+      nouveau_fence_wait(current, NULL);
       nouveau_fence_ref(NULL, &current);
       nouveau_fence_ref(NULL, &screen->base.fence.current);
    }

src/gallium/drivers/nouveau/nvc0/nvc0_transfer.c

@@ -340,8 +340,8 @@ nvc0_mt_sync(struct nvc0_context *nvc0, struct nv50_miptree *mt, unsigned usage)
       return !nouveau_bo_wait(mt->base.bo, access, nvc0->base.client);
    }
    if (usage & PIPE_TRANSFER_WRITE)
-      return !mt->base.fence || nouveau_fence_wait(mt->base.fence);
-   return !mt->base.fence_wr || nouveau_fence_wait(mt->base.fence_wr);
+      return !mt->base.fence || nouveau_fence_wait(mt->base.fence, &nvc0->base.debug);
+   return !mt->base.fence_wr || nouveau_fence_wait(mt->base.fence_wr, &nvc0->base.debug);
 }
 
 void *