[Mesa-dev] [PATCH 08/10] radeonsi: prevent a deadlock in util_queue_add_job with too many GL contexts

[Marek Olšák]

From: Marek Olšák <marek.olsak at amd.com>

If the queue is full, util_queue_add_job will wait while bo_fence_lock is
held.

It pb_slab wants to reuse a buffer, it will lock the pb_slab mutex and
try to check BO fence busyness, but it has to wait for bo_fence_lock to get
released. Both bo_fence_lock and pb_slab mutex are locked now.

When the CS thread unreferences and releases a suballocated buffer,
it will try to lock the pb_slab mutex and has to wait. The CS thread
can't finish its job in order to free a queue slot and unblock
util_queue_add_job ==> deadlock.
---
 src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
index 30f4dfb..837c1e2 100644
--- a/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
+++ b/src/gallium/winsys/amdgpu/drm/amdgpu_winsys.c
@@ -309,21 +309,22 @@ amdgpu_winsys_create(int fd, unsigned flags,
    ws->base.get_chip_name = amdgpu_get_chip_name;
 
    amdgpu_bo_init_functions(ws);
    amdgpu_cs_init_functions(ws);
    amdgpu_surface_init_functions(ws);
 
    LIST_INITHEAD(&ws->global_bo_list);
    (void) mtx_init(&ws->global_bo_list_lock, mtx_plain);
    (void) mtx_init(&ws->bo_fence_lock, mtx_plain);
 
-   if (!util_queue_init(&ws->cs_queue, "amdgpu_cs", 8, 1, 0)) {
+   if (!util_queue_init(&ws->cs_queue, "amdgpu_cs", 8, 1,
+                        UTIL_QUEUE_INIT_RESIZE_IF_FULL)) {
       amdgpu_winsys_destroy(&ws->base);
       mtx_unlock(&dev_tab_mutex);
       return NULL;
    }
 
    /* Create the screen at the end. The winsys must be initialized
     * completely.
     *
     * Alternatively, we could create the screen based on "ws->gen"
     * and link all drivers into one binary blob. */
-- 
2.7.4