mjb@acd4.UUCP ( Mike Bryan ) (10/25/89)
Just found a pretty nasty bug in Ultrix 3.1. Under periods of heavy paging/swapping, a user of shared memory may become permanently blocked at a priority of -24. Our specific configuration is a DECSystem 5400 with 16MB of memory, running Ultrix 3.1, Rev 26. I don't know if it's specific to the MIPS CPU, or if it affects the VAX line as well. I have seen it happen several times, usually in conjunction with ORACLE V6.0 (it is a heavy user of shared memory). I was finally able to duplicate it this evening in a simplified situation. The program basically calls shmat(), does lots of reads and some writes to the entire shared memory region (785000 bytes), interspersed with some sleeps, and then exits. I kept three copies of the program running at a time. Then I started up a program which is a major CPU/memory hog, and got the system to go into heavy paging and swapping. After about 10 minutes of this, the kernel broke, and one of the processes became locked. The other two were able to continue working, even though the one process was locked. It appears that the process is waiting on address smem[n].sm_flag. (This based on info from a "ps -axl", "nm -gn /vmunix", and the "smem" structure in shm.h.) The priority is -24, so of course it can't be killed or recovered in any way, except via a reboot. I don't particularly like doing this for our ORACLE development system, though. (Actually, I don't mind *that* much, but the development managers don't seem to like us loosing productivity! :-) ) Has anyone else seen this problem? Is it specific to the MIPS CPU, or does it happen on VAXen as well? Does DEC know of the problem, and more importantly, a solution? They'll be getting a call tomorrow morning from us, that's for sure... hopefully they can fix this one a lot quicker than the raw mode / PASSALL problem in LAT. Later... -- Mike Bryan, Applied Computing Devices, 100 N Campus Dr, Terre Haute IN 47802 Phone: 812/232-6051 FAX: 812/231-5280 Home: 812/232-0815 UUCP: uunet!acd4!mjb INTERNET: mjb%acd4@uunet.uu.net "Agony is born of desire; that's what you get for wanting." --- Moev