The fbcon imposes unpredictable latencies on tests - each drmIoctl has been observed to trigger two 650us calls to console_unlock() as it flushes printk buffer for the DRM_DEBUG around the ioctl. This makes tests such as gem_wait fail as they expect the ioctl to be spent on the operation under test not clogged up by the console.
References: https://bugs.freedesktop.org/show_bug.cgi?id=99130 Signed-off-by: Chris Wilson <ch...@chris-wilson.co.uk> --- lib/igt_core.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/lib/igt_core.c b/lib/igt_core.c index a18a728c..403b9423 100644 --- a/lib/igt_core.c +++ b/lib/igt_core.c @@ -67,6 +67,7 @@ #include "igt_core.h" #include "igt_aux.h" +#include "igt_sysfs.h" #ifdef HAVE_LIBGEN_H #include <libgen.h> /* for basename() on Solaris */ @@ -512,6 +513,7 @@ static void common_exit_handler(int sig) { if (!igt_only_list_subtests()) { low_mem_killer_disable(false); + kick_fbcon(true); } /* When not killed by a signal check that igt_exit() has been properly @@ -731,6 +733,7 @@ out: exit(ret == -1 ? 0 : IGT_EXIT_INVALID); if (!list_subtests) { + kick_fbcon(false); kmsg(KERN_INFO "[IGT] %s: executing\n", command_str); print_version(); -- 2.11.0 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
