Messages in this thread | | | Date | Fri, 3 May 2002 21:36:23 +0100 (BST) | From | Neil Conway <> | Subject | PATCH, IDE corruption, 2.4.18 |
| |
Well, I say patch, but it's really a band-aid.
[Caveats: before reading further, understand that I'm not by any means an IDE expert; I have just figured out how to prevent disk corruption on my PC. I may have completely missed the point.]
While debugging some "0x58" status errors (which I believed to have lead to disk corruption) I discovered that ide_config_drive_speed() uses SELECT_DRIVE without checking to see if there are disk transfers in progress (by definition these have to be DMA transfers). Unless I've misread the ATA specs, this is a Very Bad Thing.
This means that when modules which call ide_register_subdriver() are loaded (so far I think it's just ide-cd and ide-scsi) any disk transfers in progress on the other half of the cable are stuffed.
This is very easy to reproduce: just do a "dd if=/dev/hda of=/dev/null" and a few of "rmmod ide-cd;modprobe ide-cd" (where there must be a CD on hdb, or hdc and hdd but you get the idea). On my box, this is about 80% successful each module-load at causing the "0x58" status error. Repeated usage trashes inodes (don't know why) and even causes lockups (don't know why either). This suggests that the error-recovery doesn't work very well...
Anyway, here's the band aid as promised. A timeout, or a queued request would be better, but I'm not sure of how best to do either, so at least now the real experts can fix it properly...
cheers Neil PS: don't know (but doubt it) if this is what caused the 0x58 problems in 2.2 etc. PPS: I'm off-list at present, so please CC me if you want me to notice responses.
__________________________________________________ Do You Yahoo!? Everything you'll ever need on one web page from News and Sport to Email and Music Charts http://uk.my.yahoo.com[unhandled content-type:application/x-unknown] | |