ac/sqtt: add ac_sqtt_update_bo_size

buffer_size is uint32_t so we must be careful to not overflow it.
radeonsi had code for this but radv doesn't, which means it will
hang if RADV_THREAD_TRACE_BUFFER_SIZE is too big or if buffer_size
is being doubled up to the point it overflows.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/41383>

src/amd/common/ac_sqtt.c

@@ -700,3 +700,47 @@ ac_sqtt_emit_wait(const struct radeon_info *info, struct ac_pm4_state *pm4,
    ac_pm4_set_reg(pm4, R_030800_GRBM_GFX_INDEX, S_030800_SE_BROADCAST_WRITES(1) |
                   S_030800_SH_BROADCAST_WRITES(1) | S_030800_INSTANCE_BROADCAST_WRITES(1));
 }
+
+bool ac_sqtt_update_bo_size(struct ac_sqtt *sqtt, const char *env_var_prefix)
+{
+   if (strlen(env_var_prefix) > 4)
+      return false;
+
+   if (sqtt->buffer_size == 0) {
+      char envvar[sizeof("xxxx_THREAD_TRACE_BUFFER_SIZE")];
+
+      sprintf(envvar, "%s_THREAD_TRACE_BUFFER_SIZE", env_var_prefix);
+
+      /* Default buffer size set to 32MB per SE. */
+      uint64_t s = debug_get_num_option(envvar, 32 * 1024 * 1024);
+      sqtt->buffer_size = s;
+      if ((uint64_t)sqtt->buffer_size != s) {
+         fprintf(stderr,
+                 "Invalid %s value (must be <= %u).\n",
+                 envvar, UINT32_MAX);
+         return false;
+      }
+
+      return true;
+   }
+   if (sqtt->buffer_size < UINT32_MAX / 2) {
+      /* Double the size of the thread trace buffer per SE. */
+      sqtt->buffer_size *= 2;
+      fprintf(stderr,
+              "Failed to get the thread trace because the buffer "
+              "was too small, resizing to %u kB per se\n",
+              sqtt->buffer_size / 1024);
+   } else {
+      fprintf(stderr,
+              "Failed to get the thread trace because the buffer "
+              "was too small (%u kB per se). Cancelling trace capture.\n",
+               sqtt->buffer_size / 1024);
+      if (sqtt->instruction_timing_enabled)
+         fprintf(stderr,
+                 "Try again with %s_THREAD_TRACE_INSTRUCTION_TIMING=false"
+                 " to reduce the size of the captured data.\n",
+                 env_var_prefix);
+      return false;
+   }
+   return true;
+}

src/amd/common/ac_sqtt.h

@@ -577,4 +577,6 @@ void ac_sqtt_emit_stop(const struct radeon_info *info, struct ac_pm4_state *pm4,
 void ac_sqtt_emit_wait(const struct radeon_info *info, struct ac_pm4_state *pm4,
                        const struct ac_sqtt *sqtt, bool is_compute_queue);
 
+bool ac_sqtt_update_bo_size(struct ac_sqtt *sqtt, const char *env_var_prefix);
+
 #endif

src/amd/vulkan/radv_sqtt.c

@@ -506,8 +506,9 @@ radv_sqtt_init(struct radv_device *device)
    const struct radv_physical_device *pdev = radv_device_physical(device);
    struct ac_sqtt *sqtt = &device->sqtt;
 
-   /* Default buffer size set to 32MB per SE. */
-   device->sqtt.buffer_size = (uint32_t)debug_get_num_option("RADV_THREAD_TRACE_BUFFER_SIZE", 32 * 1024 * 1024);
+   if (!ac_sqtt_update_bo_size(&device->sqtt, "RADV"))
+      return false;
+
    device->sqtt.instruction_timing_enabled = radv_is_instruction_timing_enabled();
 
    /* Whether to use a staging buffer for faster reads on dGPUs. */
@@ -556,13 +557,8 @@ radv_sqtt_resize_bo(struct radv_device *device)
    /* Destroy the previous thread trace BO. */
    radv_sqtt_finish_bo(device);
 
-   /* Double the size of the thread trace buffer per SE. */
-   device->sqtt.buffer_size *= 2;
-
-   fprintf(stderr,
-           "Failed to get the thread trace because the buffer "
-           "was too small, resizing to %d KB\n",
-           device->sqtt.buffer_size / 1024);
+   if (!ac_sqtt_update_bo_size(&device->sqtt, "RADV"))
+      return false;
 
    /* Re-create the thread trace BO. */
    return radv_sqtt_init_bo(device);

src/gallium/drivers/radeonsi/gfx/si_sqtt.c

@@ -260,21 +260,8 @@ si_sqtt_resize_bo(struct si_context *sctx)
    pipe_resource_reference(&bo, NULL);
    sctx->sqtt->bo = NULL;
 
-   if (sctx->sqtt->buffer_size < UINT32_MAX / 2) {
-      /* Double the size of the thread trace buffer per SE. */
-      sctx->sqtt->buffer_size *= 2;
-      mesa_loge("Failed to get the thread trace because the buffer "
-                "was too small, resizing to %d kB per se",
-                sctx->sqtt->buffer_size / 1024);
-   } else {
-      mesa_loge("Failed to get the thread trace because the buffer "
-                "was too small (%d kB per se). Cancelling trace capture.",
-                 sctx->sqtt->buffer_size / 1024);
-      if (sctx->sqtt->instruction_timing_enabled)
-         mesa_loge("Try again with AMD_THREAD_TRACE_INSTRUCTION_TIMING=false"
-                   " to reduce the size of the captured data.");
+   if (!ac_sqtt_update_bo_size(sctx->sqtt, "AMD"))
       return false;
-   }
 
    /* Re-create the thread trace BO. */
    return si_sqtt_init_bo(sctx);
@@ -335,10 +322,9 @@ bool si_init_sqtt(struct si_context *sctx)
       return false;
    }
 
-   /* Default buffer size set to 32MB per SE. */
-   sctx->sqtt->buffer_size =
-      debug_get_num_option("AMD_THREAD_TRACE_BUFFER_SIZE", 32 * 1024) * 1024;
-   assert(sctx->sqtt->buffer_size);
+   if (!ac_sqtt_update_bo_size(sctx->sqtt, "AMD"))
+      return false;
+
    sctx->sqtt->instruction_timing_enabled =
       debug_get_bool_option("AMD_THREAD_TRACE_INSTRUCTION_TIMING", true);
    sctx->sqtt->start_frame = 10;