On 30 September 2015 at 17:44, Daniel Vetter daniel@ffwll.ch wrote:
On Sun, Sep 06, 2015 at 06:53:00PM +1000, Dave Airlie wrote:
There is a race where the reply could get processed by another work queue before we've updated the state.
Update the state before sending the msg to close it.
v2: reset value if return indicates we haven't send the msg.
Pointed out by Adam J Richter on
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91481 Signed-off-by: Dave Airlie airlied@redhat.com
Ok we see to have two callers of drm_dp_send_link_address:
- drm_dp_add_port, but the only caller of that in turn is again drm_dp_send_link_address. This is just self-recursion.
- drm_dp_check_and_send_link_address, which in turn is called only by itself and by drm_dp_mst_link_probe_work.
drm_dp_mst_link_probe_work is only called from the mgr->work function, and a single work item is never run concurrently.
I don't see any race here at all. A good cleanup though would be what Adam's original patch has done and move the link_address_sent = true into drm_dp_send_link_address.
I think we were being paranoid about the work queue vs the send_link_address if you can't have multiple then it matters not.
But I'll repost the cleaned up version as it makes things easier to grok.
Dave.