On Tue, 27 Jul 2004, Marc Haber wrote: > > I think that's not the USB subsystem getting into a bad state; more likely > > it's the SCSI subsystem taking the disk off-line because of unrecoverable > > errors. Unplugging the drive and then plugging it back in ought to help. > > Yes, that usually helps, but doesn't help when the drive is at a > remote site.
Very true. There are some commands you can send to the SCSI subsystem using procfs that will have much the same effect. See section 4.3.1 of the SCSI-Programming-HOWTO. > I must admit that I have become quite confused between two computers, > a notebook, one PCI USB card, two enclosures and three disks. I should > have been more thorough with my documentation. > > If I need to repeat the tests, is it enough to have the sector number > reported from the SCSI subsystem, or do I really need to run with the > _very_ verbose USB logging enabled? I think the sector number reported by SCSI will be enough. I was interested in the verbose USB logging only because I wanted to see exactly how the failure occurred and what was failing. > > "... after different time spans..." Does that mean sometimes you are able > > to read the sector without getting an error? > > Obviously, I need to note the wallclock time from test start to test > failure as well. Maybe you can come up with a version of the test that only tries to read sectors in the vicinity of the bad ones. > > Again, that's probably because the disk is off-line. Once that happens > > all further I/O requests fail, so you see a series of errors with > > incrementing sector numbers. > > Yes, of course. So it's really only the first error that matters. Right. > > Can you try using this device on a system running Windows? If you still > > get similar errors (and especially if they occur at the same sector > > locations) then you should try to exchange the enclosure or get it > > repaired. > > Windows is not so easily scriptable, and since I use ext3 as file > system on the USB disks, I would need to build a new file system which > will probably destroy the "bad" bit pattern, influencing the test > result. Yes, using Windows for this sort of thing can be difficult. And you shouldn't erase the disks. Can you run your test under Cygwin? Even if you're unable to mount the partitions in Windows, perhaps you can read the raw disk device. > I have two desktop boxes, but both only have USB 1.1, so I need to use > a PCI USB card which I only have one. So, all tests involving the > desktop boxes have been done with the same PCI card with its PCI data > listed above. The enclosure works fine when connected to my Notebook > which has a USB 2.0 interface built into the Centrino chipset. Oho! > Tomorrow, I will get an different USB PCI card which hopefully has a > different chipset. Good idea. If replacing the USB host controller card will solve your problem, it's an easy answer. Then there wouldn't be any need to run more tests or mess around with Windows. Alan Stern ------------------------------------------------------- This SF.Net email is sponsored by BEA Weblogic Workshop FREE Java Enterprise J2EE developer tools! Get your free copy of BEA WebLogic Workshop 8.1 today. http://ads.osdn.com/?ad_id=4721&alloc_id=10040&op=click _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-users
