--- Jens Owen <[EMAIL PROTECTED]> wrote: > Mike Mestnik wrote: > > > > --- Michael <[EMAIL PROTECTED]> wrote: > > > Can we repair the damage? That doesn't seem to have been that successful > > > in the past in terms of recovery / carrying on running the 3d app - > > > others have more knowledge here. If anything, cleanly exiting from 3d > > > and, if recovery is possible or the 2d side / keyboard and mouse aren't > > > all broken, the 2d side trying to recover might work better? Failing > > > that, finding the bugs that cause those hangs is the only real fix. > > > > > I don't know about getting the current app to still be working, but we > should > > still try to if nothing else sync, remount-ro so that the user can hit the > > power button. I don't think it should ever get that bad, the GPU could > loose > > power temporarily flushing its core memory (Because of a hard drive spin up > and > > a bad power supply) the app could reload the GPU and life goes on. Since > this > > code would only be used in extreme situations it can't hurt having it > around. > > > > Maby I'm just crazy about video games :) > > When I'm 19 to 20 in CTF I don't ever want anything spoiling my game. > > > > > > > > I guess you could make this function use an ioctl() into the drm and get > > > it to put you on a wait queue woken after the driver updates the > > > register it's polling? Maybe others will chip in with their thoughts. > > > > > > -- > > > Michael. > > > > > I was thinking the same thing, but I don't realy know the proformance hit > in an > > ioctl. In my testing it usualy gets out on the first or second try. This > > could be from PageFlip and when we do a CopyBuffer this is when the wait is > > longer (200 - 700 tryes /w hammering the bus). I'l do some more testing to > see > > whitch. This may sound crazy but could we do the ioctl only after trying 3 > or > > 5 times? I like the idea of letting other apps run. Maby the next-gen 3D > > games are multi-threded. > > Are we talking about recovering from driver hangs here? There are > really two classes of hangs: > > 1) System locks up tight. No external login. Usually caused by hanging > the bus. No error recovery is possible without some type of watch dog > timer to reset the bus. > > 2) X subsystem hangs, but rest of system is still running. Can login > externally and > gracefully reboot the system. Error recovery w/o reboot may be > possible, but difficult. Can login remotely and kill 3D client > application and possibly recover. > > Which class of hang are you guys talking about? >
Mostly the Second. However if we know were'r going to do a #1 sync first please. __________________________________________________ Do You Yahoo!? LAUNCH - Your Yahoo! Music Experience http://launch.yahoo.com _______________________________________________________________ Don't miss the 2002 Sprint PCS Application Developer's Conference August 25-28 in Las Vegas -- http://devcon.sprintpcs.com/adp/index.cfm _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel
