Keith Whitwell wrote: >Thomas Hellström wrote: > > >>Keith Whitwell wrote: >> >> >>>Thomas Hellstrom wrote: >>> >>> >>> >>>> linux-core/drmP.h | 13 +++- >>>> linux-core/drm_bo.c | 31 ++++++----- >>>> linux-core/drm_fence.c | 128 >>>> +++++++++++------------------------------------- >>>> linux-core/drm_lock.c | 4 - >>>> >>>> >>>> >>>Thomas, >>> >>>Can you be more specific about the changes in drm_lock.c -- what is the >>>background there - the change doesn't seem related to the commit comment? >>> >>>Keith >>> >>> >>> >>> >>Ah, OK, this one more or less slipped through. >>I might need to revert that. >> >>In a situation where there are many processes waiting on the hardware >>lock, and one process grabs it, the lock will lose the CONTENDED flag, >>until one of the sleeping processes is woken up by a clock tick and >>sends the CONTENDED flag again. The process taking the lock will not >>call the kernel to wake up sleeping processes, and can retake the lock >>again without calling the kernel. >> >>This is not an ideal situation, and the change makes sure that the >>process taking the lock calls the kernel again to wake up any sleeping >>processes even if there are none available. This seems improve >>scheduling at the cost of some extra CPU when there are more than one >>DRI client. >> >> > >Ah, interesting. This sounds like the cause of some pretty >long-standing bad behaviour of dri locking where some contexts end up >being 'locked out' despite the dri lock being repeatedly grabbed and >released by a given process. > >Keith > > Yes. Hmm, Actually the sleeping processes will not wake up by clock ticks since schedule() is called instead of schedule_timeout(), so the starvation can potentially go on forever. That's probably why some window movement sometimes seemed to wake a locked-out context.
Anyway, while the current code seems to fix the problem, the best fix is probably to add a sleeping process counter to the lock. That would avoid unnecessary kernel calls when the lock is released and there are no processes waiting. Thomas >------------------------------------------------------------------------- >Using Tomcat but need to do more? Need to support web services, security? >Get stuff done quickly with pre-integrated technology to make your job easier >Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo >http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 >-- >_______________________________________________ >Dri-devel mailing list >Dri-devel@lists.sourceforge.net >https://lists.sourceforge.net/lists/listinfo/dri-devel > > ------------------------------------------------------------------------- Using Tomcat but need to do more? Need to support web services, security? Get stuff done quickly with pre-integrated technology to make your job easier Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642 -- _______________________________________________ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
