Messages in this thread |  | | From | (Frank Pavageau) | Subject | Crashes under load with 2.0.27 | Date | 5 Jan 1997 18:27:46 GMT |
| |
I've been having crashes for the past 3 weeks, 3 of them led to reinstallation of the system. It might be a hardware problem, but there are a few strange things that happen. So here's the story :
The first crash happened while I was connected by PPP. I was under X, writing a mail with pine. As I sent the mail and pine forked into sendmail, X scrolled up and the computer froze completely. Reset, reboot, e2fsck which finds almost nothing (a few inodes with zero dtime), and back under X with PPP up. Then I launch suck to get my news, and after innxmit has locally posted the articles, I've lost /usr/bin and /usr/lib. Unfortunately, it was kind of late and I forgot to save the kernel log.
I reinstall the system, and 2 days later, I get a new crash with the HD making a spooky noise, but this time X is not locked. I switch to tty0 and get a lot of error messages from the IDE driver saying that it gets timeouts and is trying to reset. This time I had Crack running for a couple of hours before using my computer, and it looks like the crash had been waiting for me, since there had a been a lot of VFS: Wrong blocksize on device 16:01 in the kernel log while it was running alone. But since Crack doesn't make a lot of disk access, I find it quite strange. Anyway the crash happened just after I started PPP with oopses from sh and chat, and Unable to handle kernel paging request at virtual address c813fe7f.
Next day, another crash : Aiee: scheduling in interrupt 0011ad3d (repeated 59 times) general protection: 0000 (from sendmail) and more astonishing : attempt to access beyond end of device 16:01: rw=0, want=1735289205, limit=487336 attempt to access beyond end of device 16:01: rw=0, want=135652385, limit=487336 attempt to access beyond end of device 16:01: rw=0, want=1952541795, limit=487336 ... EXT2-fs warning (device 16:01): ext2_free_inode: bit already cleared for inode 49217 (twice) and then Oops from suck.
Next day, yet another crash, but I can't find the log anymore. Anyway, the message was from the scheduler this time : wait_queue is bad (eip = xxxx) q = xxxxx *q = yyyyy
Then I tried with another HD to put the system on. It ran without any problem for 10 days, and then, yesterday, I got a crash after trying Crack5 for a couple of hours. This time, I had no / left (but I still had /sbin, /usr, /whatever, just no /). I managed to shutdown, but then it wouldn't boot (kernel panic trying to mount its root readonly), and of course I didn't find any boot disk with e2fsck on it (it would be cool if debian had it). So I reinstalled once again. I've not yet crashed again, but it could be on its way cause I got that : EXT2-fs error (device 03:41): ext2_read_inode: bad inode number: 1920169263 after compiling the kernel twice.
Since I was almost every time running crack when it happened (either 4.1 or 5), it looks like it happens under CPU load. What bugs me is the memory corruption that seems to result (abnormally large inode or block numbers). I ran memtest-86 for 6 hours yesterday night and it found no error at all, so the RAM seems fine. It happened with 2 different HD (both Quantum ones, but the first one was a 540 Mb 18 months old, and the second one is a brand new 2.5 Gb). This leaves the EIDE controller, which is integrated on my Gigabyte GA-586 MB. The only things I find about it in the MB documentation is "Onboard CMD IDE port" (it's not a CMD640, I tried enabling the kernel support and it never said anything), and the basic electronic diagram which link the IDE ports with 2 chips : S82371FB (Triton I guess), and 74F245. But then I don't understand why a suddenly buggy hardware would corrupt inode tables (returning strange block numbers, I could understand that).
I never had any problem before, and I've had this MB for 18 months. I can just add a few things about my configuration : MB: Gigabyte GA-586 with P100, 256 kb cache, integrated EIDE controller RAM: 32 Mb HDs (first config) : hda: WDC AC2200F, 202MB w/64kB Cache, CHS=989/12/35 (DOS) hdb: ST3491A-XR, 408MB w/120kB Cache, CHS=899/15/62 (hdb1 DOS, hdb2 /home) hdc: QUANTUM MAVERICK 540A, 516MB w/98kB Cache, LBA, CHS=1049/16/63 (hdc1 / hdc2 swap, 40 Mb) hdd: FX001DE, ATAPI CDROM drive HDs (new config) : hda: same as before hdb: QUANTUM BIGFOOT2550A, 2457MB w/87kB Cache, LBA, CHS=624/128/63, DMA (hdb1 / hdb2 swap, 64 Mb hdb3 /usr hdb4 /home) hdc: old hdb, but hdc2 (old /home) unused right now hdd: same as before Kernel 2.0.27, hdparm 3.1.
After yesterday crash, I used hdc2 as /var, so should the computer crash again, I will hopefully be able to get the kernel log... AND, I won't lose my news and mail spools once again.
So, does anyone have any idea about what could be happening ? Should I get a new EIDE controller, or is it related to another hardware problem, or could it be the kernel ? Help ! :)
Frank -- X is a single letter denoting the unknown. This is X too. Motif is what everyone uses to annoy people that dont have it. Openwindoze is related to X only harder to spell and slower to use. [Peter Evans (peter@gol1.gol.com) in comp.unix.admin]
|  |