Bump event read size to 4 KiBstaging/fixes

drmHandleEvent() currently reads events in 1 KiB chunks. Within the
kernel each DRM device is allowed to carry up to 4 KiB of event data,
and the drm_read() implementation will only hand out complete events.
All of this combined could cause applications to go into a busy loop
if drivers queue events larger than 1 KiB, because drm_read() will
requeue the event and drm_poll() would never see the list of queued
events become empty and therefore always return immediately.

This patch increases the buffer for event reads to 4 KiB, so that all
events will be read at once. This ensures that at least one event can
always be dequeued, therefore avoiding the busy loop condition.

A complementary patch will be applied to the kernel to limit the size
of events to 1 KiB to ensure that versions of the libdrm library prior
to this change won't break. Fortunately, there currently aren't any
event producers that exceed the 1 KiB limit.

Signed-off-by: Thierry Reding <treding@nvidia.com>

1 files changed, 1 insertions, 1 deletions

diff --git a/xf86drmMode.c b/xf86drmMode.c
index d3bc20ea..c4bccd6f 100644
--- a/xf86drmMode.c
+++ b/xf86drmMode.c
@@ -885,7 +885,7 @@ int drmModeCrtcSetGamma(int fd, uint32_t crtc_id, uint32_t size,
 
 int drmHandleEvent(int fd, drmEventContextPtr evctx)
 {
-	char buffer[1024];
+	char buffer[4096];
 	int len, i;
 	struct drm_event *e;
 	struct drm_event_vblank *vblank;