On 10/30/12 16:51, Shally Verma wrote: > Hi > > I am seeing DFB apps (df_andi or df_matrix) deadlock conditions with DFB > 1.4.3 lib on linux 2.6.34 and seems all of the processes go in sleep > state.Sometimes it > comes very soon. > > > > I ran fusion /cat/proc interface stat , property and skirmishes command and > see in an output of /proc/fusion/0/stat lease count> cede count AND prevail> > dismiss . > > Also if I look at cat /proc/fusion/property and /proc/fusion/skirmishes > output its shown locked by current process. > > > > Could anyone please confirm me if its clear indication of deadlock so that I > can be confirm if am debugging in right directionand will help if you could > give > some description of same .How can I debug it where it can get freeze?I > enabled FUSION_DEBUG_SKIRMISH_DEADLOCK but still I don’t see any message on > console once > deadlock is reached.
The skirmish dead lock debugging just detects inversion of lock order which is a potential dead lock. What you sent looks like a dead lock, but to debug further we'd need logs or stack traces. You can also try the latest fusion kernel module that should print the waiters on a skirmish, but it's obvious that pid 592 has taken the lock without releasing it. As the property is also "locked" it seems to be related to the graphics driver, though we did some fixes in that code area recently. Actually, having both the property and skirmish locked at the same time does not make sense, so I am wondering if you have a modified gfxcard.c in use. Please send that one for us to check. Thanks! -- Denis Oliver Kropp DirectFB integrated media GmbH _______________________________________________ directfb-dev mailing list directfb-dev@directfb.org http://mail.directfb.org/cgi-bin/mailman/listinfo/directfb-dev
