winsys/radeon: consolidate tracing into winsys v2

This move the tracing timeout and printing into winsys and add
an debug environement variable for it (R600_DEBUG=trace_cs).

Lot of file touched because of winsys API changes.

v2: Do not write lockup file if ib uniq id does not match last one

Signed-off-by: Jerome Glisse <[email protected]>
Reviewed-by: Marek Olšák <[email protected]>

5 files changed, 46 insertions, 29 deletions

diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
index 720e0864d19..ea0c99df3fa 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
@@ -167,7 +167,9 @@ static void radeon_destroy_cs_context(struct radeon_cs_context *csc)
 }
 
 
-static struct radeon_winsys_cs *radeon_drm_cs_create(struct radeon_winsys *rws, enum ring_type ring_type)
+static struct radeon_winsys_cs *radeon_drm_cs_create(struct radeon_winsys *rws,
+                                                     enum ring_type ring_type,
+                                                     struct radeon_winsys_cs_handle *trace_buf)
 {
     struct radeon_drm_winsys *ws = radeon_drm_winsys(rws);
     struct radeon_drm_cs *cs;
@@ -179,6 +181,7 @@ static struct radeon_winsys_cs *radeon_drm_cs_create(struct radeon_winsys *rws,
     pipe_semaphore_init(&cs->flush_completed, 0);
 
     cs->ws = ws;
+    cs->trace_buf = (struct radeon_bo*)trace_buf;
 
     if (!radeon_init_cs_context(&cs->csc1, cs->ws)) {
         FREE(cs);
@@ -413,7 +416,7 @@ static void radeon_drm_cs_write_reloc(struct radeon_winsys_cs *rcs,
     OUT_CS(&cs->base, index * RELOC_DWORDS);
 }
 
-void radeon_drm_cs_emit_ioctl_oneshot(struct radeon_cs_context *csc)
+void radeon_drm_cs_emit_ioctl_oneshot(struct radeon_drm_cs *cs, struct radeon_cs_context *csc)
 {
     unsigned i;
 
@@ -432,9 +435,9 @@ void radeon_drm_cs_emit_ioctl_oneshot(struct radeon_cs_context *csc)
         }
     }
 
-#if RADEON_CS_DUMP_ON_LOCKUP
-    radeon_dump_cs_on_lockup(csc);
-#endif
+    if (cs->trace_buf) {
+        radeon_dump_cs_on_lockup(cs, csc);
+    }
 
     for (i = 0; i < csc->crelocs; i++)
         p_atomic_dec(&csc->relocs_bo[i]->num_active_ioctls);
@@ -458,7 +461,7 @@ void radeon_drm_cs_sync_flush(struct radeon_winsys_cs *rcs)
 
 DEBUG_GET_ONCE_BOOL_OPTION(noop, "RADEON_NOOP", FALSE)
 
-static void radeon_drm_cs_flush(struct radeon_winsys_cs *rcs, unsigned flags)
+static void radeon_drm_cs_flush(struct radeon_winsys_cs *rcs, unsigned flags, uint32_t cs_trace_id)
 {
     struct radeon_drm_cs *cs = radeon_drm_cs(rcs);
     struct radeon_cs_context *tmp;
@@ -474,6 +477,8 @@ static void radeon_drm_cs_flush(struct radeon_winsys_cs *rcs, unsigned flags)
     cs->csc = cs->cst;
     cs->cst = tmp;
 
+    cs->cst->cs_trace_id = cs_trace_id;
+
     /* If the CS is not empty or overflowed, emit it in a separate thread. */
     if (cs->base.cdw && cs->base.cdw <= RADEON_MAX_CMDBUF_DWORDS && !debug_get_option_noop()) {
         unsigned i, crelocs = cs->cst->crelocs;
@@ -536,7 +541,7 @@ static void radeon_drm_cs_flush(struct radeon_winsys_cs *rcs, unsigned flags)
                 }
             }
             pipe_mutex_unlock(cs->ws->cs_stack_lock);
-            radeon_drm_cs_emit_ioctl_oneshot(cs->cst);
+            radeon_drm_cs_emit_ioctl_oneshot(cs, cs->cst);
         }
     } else {
         radeon_cs_context_cleanup(cs->cst);
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.h b/src/gallium/winsys/radeon/drm/radeon_drm_cs.h
index 66aee55dae0..7683e96f525 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.h
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.h
@@ -30,8 +30,6 @@
 #include "radeon_drm_bo.h"
 #include <radeon_drm.h>
 
-#define RADEON_CS_DUMP_ON_LOCKUP    0
-
 struct radeon_cs_context {
     uint32_t                    buf[RADEON_MAX_CMDBUF_DWORDS];
 
@@ -41,6 +39,8 @@ struct radeon_cs_context {
     uint64_t                    chunk_array[3];
     uint32_t                    flags[2];
 
+    uint32_t                    cs_trace_id;
+
     /* Relocs. */
     unsigned                    nrelocs;
     unsigned                    crelocs;
@@ -78,6 +78,7 @@ struct radeon_drm_cs {
 
     int flush_started;
     pipe_semaphore flush_completed;
+    struct radeon_bo                    *trace_buf;
 };
 
 int radeon_get_reloc(struct radeon_cs_context *csc, struct radeon_bo *bo);
@@ -121,10 +122,8 @@ radeon_bo_is_referenced_by_any_cs(struct radeon_bo *bo)
 
 void radeon_drm_cs_sync_flush(struct radeon_winsys_cs *rcs);
 void radeon_drm_cs_init_functions(struct radeon_drm_winsys *ws);
-void radeon_drm_cs_emit_ioctl_oneshot(struct radeon_cs_context *csc);
+void radeon_drm_cs_emit_ioctl_oneshot(struct radeon_drm_cs *cs, struct radeon_cs_context *csc);
 
-#if RADEON_CS_DUMP_ON_LOCKUP
-void radeon_dump_cs_on_lockup(struct radeon_cs_context *csc);
-#endif
+void radeon_dump_cs_on_lockup(struct radeon_drm_cs *cs, struct radeon_cs_context *csc);
 
 #endif
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs_dump.c b/src/gallium/winsys/radeon/drm/radeon_drm_cs_dump.c
index a3634bf2566..e738a988679 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs_dump.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs_dump.c
@@ -34,18 +34,18 @@
 #include "radeon_drm_cs.h"
 #include "radeon_drm_bo.h"
 
-#if RADEON_CS_DUMP_ON_LOCKUP
-static bool dumped = false;
+#define RADEON_CS_DUMP_AFTER_MS_TIMEOUT         500
 
-void radeon_dump_cs_on_lockup(struct radeon_cs_context *csc)
+void radeon_dump_cs_on_lockup(struct radeon_drm_cs *cs, struct radeon_cs_context *csc)
 {
     struct drm_radeon_gem_busy args;
     FILE *dump;
     unsigned i, lockup;
     uint32_t *ptr;
+    char fname[32];
 
     /* only dump the first cs to cause a lockup */
-    if (!csc->crelocs || dumped) {
+    if (!csc->crelocs) {
         /* can not determine if there was a lockup if no bo were use by
          * the cs and most likely in such case no lockup occurs
          */
@@ -54,19 +54,27 @@ void radeon_dump_cs_on_lockup(struct radeon_cs_context *csc)
 
     memset(&args, 0, sizeof(args));
     args.handle = csc->relocs_bo[0]->handle;
-    for (i = 0; i < 10; i++) {
-        usleep(5);
+    for (i = 0; i < RADEON_CS_DUMP_AFTER_MS_TIMEOUT; i++) {
+        usleep(1);
         lockup = drmCommandWriteRead(csc->fd, DRM_RADEON_GEM_BUSY, &args, sizeof(args));
         if (!lockup) {
             break;
         }
     }
-    if (!lockup || i < 10) {
+    if (!lockup || i < RADEON_CS_DUMP_AFTER_MS_TIMEOUT) {
+        return;
+    }
+
+    ptr = radeon_bo_do_map(cs->trace_buf);
+    fprintf(stderr, "timeout on cs lockup likely happen at cs 0x%08x dw 0x%08x\n", ptr[1], ptr[0]);
+
+    if (csc->cs_trace_id != ptr[1]) {
         return;
     }
 
     /* ok we are most likely facing a lockup write the standalone replay file */
-    dump = fopen("radeon_lockup.c", "w");
+    snprintf(fname, sizeof(fname), "rlockup_0x%08x.c", csc->cs_trace_id);
+    dump = fopen(fname, "w");
     if (dump == NULL) {
         return;
     }
@@ -74,8 +82,9 @@ void radeon_dump_cs_on_lockup(struct radeon_cs_context *csc)
     fprintf(dump, " * in same directory. You can find radeon_ctx.h in mesa tree :\n");
     fprintf(dump, " * mesa/src/gallium/winsys/radeon/tools/radeon_ctx.h\n");
     fprintf(dump, " * Build with :\n");
-    fprintf(dump, " * gcc -O0 -g radeon_lockup.c -ldrm -o radeon_lockup -I/usr/include/libdrm\n");
+    fprintf(dump, " * gcc -O0 -g %s -ldrm -o rlockup_0x%08x -I/usr/include/libdrm\n", fname, csc->cs_trace_id);
     fprintf(dump, " */\n");
+    fprintf(dump, " /* timeout on cs lockup likely happen at cs 0x%08x dw 0x%08x*/\n", ptr[1], ptr[0]);
     fprintf(dump, "#include <stdio.h>\n");
     fprintf(dump, "#include <stdint.h>\n");
     fprintf(dump, "#include \"radeon_ctx.h\"\n");
@@ -107,6 +116,7 @@ void radeon_dump_cs_on_lockup(struct radeon_cs_context *csc)
     }
     fprintf(dump, "};\n\n");
 
+    fprintf(dump, "/* cs %d dw */\n", csc->chunks[0].length_dw);
     fprintf(dump, "static uint32_t cs[] = {\n");
     ptr = csc->buf;
     for (i = 0; i < csc->chunks[0].length_dw; i++) {
@@ -147,6 +157,4 @@ void radeon_dump_cs_on_lockup(struct radeon_cs_context *csc)
     fprintf(dump, "    bo_wait(&ctx, bo[0]);\n");
     fprintf(dump, "}\n");
     fclose(dump);
-    dumped = true;
 }
-#endif
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
index f6877d5e95d..368902039c4 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
@@ -551,7 +551,7 @@ next:
         pipe_mutex_unlock(ws->cs_stack_lock);
 
         if (cs) {
-            radeon_drm_cs_emit_ioctl_oneshot(cs->cst);
+            radeon_drm_cs_emit_ioctl_oneshot(cs, cs->cst);
 
             pipe_mutex_lock(ws->cs_stack_lock);
             for (i = 1; i < p_atomic_read(&ws->ncs); i++) {
diff --git a/src/gallium/winsys/radeon/drm/radeon_winsys.h b/src/gallium/winsys/radeon/drm/radeon_winsys.h
index a37bd120cb8..1c2fb698981 100644
--- a/src/gallium/winsys/radeon/drm/radeon_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_winsys.h
@@ -368,8 +368,12 @@ struct radeon_winsys {
      * Create a command stream.
      *
      * \param ws        The winsys this function is called from.
+     * \param ring_type The ring type (GFX, DMA, UVD)
+     * \param trace_buf Trace buffer when tracing is enabled
      */
-    struct radeon_winsys_cs *(*cs_create)(struct radeon_winsys *ws, enum ring_type ring_type);
+    struct radeon_winsys_cs *(*cs_create)(struct radeon_winsys *ws,
+                                          enum ring_type ring_type,
+                                          struct radeon_winsys_cs_handle *trace_buf);
 
     /**
      * Destroy a command stream.
@@ -425,10 +429,11 @@ struct radeon_winsys {
     /**
      * Flush a command stream.
      *
-     * \param cs        A command stream to flush.
-     * \param flags,    RADEON_FLUSH_ASYNC or 0.
+     * \param cs          A command stream to flush.
+     * \param flags,      RADEON_FLUSH_ASYNC or 0.
+     * \param cs_trace_id A unique identifiant for the cs
      */
-    void (*cs_flush)(struct radeon_winsys_cs *cs, unsigned flags);
+    void (*cs_flush)(struct radeon_winsys_cs *cs, unsigned flags, uint32_t cs_trace_id);
 
     /**
      * Set a flush callback which is called from winsys when flush is