From ed7673afd2d1a31bfcec0520743acd2681b462e6 Mon Sep 17 00:00:00 2001 From: Bruce Cherniak Date: Fri, 4 Jan 2019 14:53:59 -0600 Subject: [PATCH] gallium/swr: Fix multi-context sync fence deadlock. Various recreation scenarios lead to API thread getting stuck in swr_fence_finish(). This is a multi-context issue, whereby one context overwrites the fence read-value with a previous sync's lesser value. The fence sync value is supposed to be always increasing. In swr_fence_cb(), only update the "read" value if the new value is greater. (This may seem like we're not waiting on the other context to finish, but had we needed for it to finish there would have been a wait prior to submitting a new sync.) cc: mesa-stable@lists.freedesktop.org --- src/gallium/drivers/swr/swr_fence.cpp | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/src/gallium/drivers/swr/swr_fence.cpp b/src/gallium/drivers/swr/swr_fence.cpp index b05ac8cec02..074d82a3b47 100644 --- a/src/gallium/drivers/swr/swr_fence.cpp +++ b/src/gallium/drivers/swr/swr_fence.cpp @@ -50,7 +50,9 @@ swr_fence_cb(uint64_t userData, uint64_t userData2, uint64_t userData3) swr_fence_do_work(fence); /* Correct value is in SwrSync data, and not the fence write field. */ - fence->read = userData2; + /* Contexts may not finish in order, but fence value always increases */ + if (fence->read < userData2) + fence->read = userData2; } /* -- 2.30.2