home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!usc!zaphod.mps.ohio-state.edu!pacific.mps.ohio-state.edu!linac!mp.cs.niu.edu!fnnews!fncrda.fnal.gov!rohrer
- From: rohrer@fncrda.fnal.gov (Keith Rohrer)
- Newsgroups: comp.os.linux
- Subject: timeouts on hard drive (long)
- Message-ID: <2141@fnnews.fnal.gov>
- Date: 30 Jul 92 19:16:57 GMT
- Sender: news@fnnews.fnal.gov
- Organization: Fermi National Accelerator Laboratory, Batavia IL
- Lines: 40
- Nntp-Posting-Host: fncrda.fnal.gov
-
- I've had Linux up and running for a week or two now, and I added another
- partition to my Linux zone. There is a DOG partition (/dev/hda1), a Linux-
- Minix FS partition (/dev/hda2), 8 megs of swap (/dev/hda3), and the fourth
- partition (/dev/hda4). I know that the fourth partition is physically perfect,
- because mkfs -c and mkefs -c both mark no bad blocks, and I formatted the
- partition under DOG and ran Norton Witch Doctor on it (in continuous-pass
- weekly mode) while I had dinner. So no problem there.
-
- However, when I fill the new partition with data (it's not the number of files,
- as a couple tars of the Nethack source will do), and fsck it, or try to cp
- or cat > /dev/null from it, *sometimes* it hangs up for about 5 seconds at a
- time, then prints out that it timed out on the hard drive, and was resetting
- the controller. Then it goes on as normal, and does the right thing. The
- process always works correctly if the data is already in the buffer cache (of
- course) and only sometimes does this when reading the actual disk. I've hacked
- the kernel to printk the device and sectors being requested, and they are
- always the right device and reasonalble sectors to be reading. Further,
- back when I could get it to do this on demand, they sectors were very
- similar (I fsck'd 3 times in a row, one at a time, on three VC's and wrote
- down the results; most of the sectors timed out on were timed out on twice,
- a fair number all three times, a few only once). I was doing *all* of this
- in a very linear, DOS-like way, one thing at a time, only on one VC until
- I went to test the repeatability of the timeouts on particular sectors, at
- which point I did one, switched to the next VC, logged in, fsck'd, waited
- for it to finish, switched to the next, &c.
-
- I'm using Linux kernel 0.96c pl 2 with the third extfs patch (triple indirection
- patch?), everything else but the datafile was stock MCC Interim 0.96/0.96c.
- The hardware is a 386-40 clone, with 4 megs of RAM (looks like 4.25 as the
- chipset scavenges most of 640K-1M), a Maxtor 7213A IDE hard drive, and floppies.
-
- Any ideas? I posted about this problem, though with less concrete evidence,
- before, and the only vaguely useful suggestion I got is that I should have
- my BIOS updated (it's currently dated 5-91 or so, and the person said it
- should be 6-6-91 or newer "for correct operation with IDE drives"... anyone
- confirm that?), but I don't think that's the case, as DOS uses it just fine,
- and I don't think Linux uses any firmware for the hard drive, just the CMOS...
-
- Keith
- (rohrer@fncrd0.fnal.gov)
-