| From: | Matthias Rampke <matthias.rampke@xxxxxxxxxxxxxx> |
| Date: | Tue, 15 Feb 2011 18:16:21 +0100 |
Hello, I had two (separate, as far as I can tell) problems with my Western Digital Elements Desktop USB hard disk on x86_64 2.8-RELEASE. (a) the disk would be recognised, but not usable after a reboot, but worked fine after either power cycling the machine (i.e. turning it off and on again) or after re-attaching the disk. In both cases the disk powers itself down, during a reboot it does not. I have not cross-tested this on master, but I think I had the same problem there. (b) when writing large amounts of data to the same disk (formatted with HAMMER) the system would panic (see the screenshot at [1]). This does NOT happen on master. I have found (through guessing and trial&error) a fix for both problems: adding an entry to the quirks table in sys/dev/usbmisc/umass/umass.c - see the attached patch or the gist at [2]. (a) is fixed by adding the NO_TEST_UNIT_READY quirk. From the description of said flag this is plausible. (b) is fixed through IGNORE_RESIDUE. This sounds more-or-less plausible as well, but I do not understand why this panic did not occur on master even without this patch (umass.c has not changed between 2.8-RELEASE and master). I can not find where this particular quirk is ever observed, so I am at loss here concerning the underlying reason for this. My guess is (from what I understand about IGNORE_RESIDUE) that this flag causes some part of the driver stack to disregard some count from the disk and relies solely on it's own knowledge of the state of a transfer, thus not choking on this count being off. Has this perhaps been made the default? Regards, Matthias PS: Thanks swildner for the advice and patience! [1] http://twitpic.com/405ogp [2] https://gist.github.com/825244
Attachment:
wd-extern-hdd-1021.patch
Description: Binary data