Messages in this thread | | | Date | Mon, 12 Oct 1998 02:05:32 +0200 (MET DST) | From | Andries.Brouwer@cwi ... | Subject | Re: [ME TOO] Re: Linux-2.1.125 ... pre-2.2 |
| |
> > IF 2.2 is released without these issues resolved, we'd better mark > > Adaptec 1542 support "EXPERIMENTAL" again.
> It just needs fixing. The AHA1542 driver in 2.0.x is absolutely rock solid
No, it is not, and has never been. The truth is: when no errors occur, the aha1542 driver functions fine. As soon as something goes wrong, garbage code is invoked. The error recovery code in aha1542.c is wrong or nonexistent.
Five years ago a bad block on the disk would kill the system in a loop of resetting scsi bus... Today this is still true for 2.0. For 2.1 the details are a bit different because we now have scsi_error.c, but the end result is more or less the same: either the system is killed entirely, or the offending device is taken off-line.
I have seen systems killed because the cdrom code tried to lock a nonexistent door, and scsi_error decided to bring down everything in an attempt to fix the problem that this cdrom drive kept returning errors. On this list we learn that a bad block size on tape takes the tape unit off-line etc.
There are two conclusions: 1. scsi_error is far too drastic, and it should become milder before 2.2 is released 2. aha1542.c must be fixed.
I made some progress on part 2, but will need (much) more time, especially since I do not have a 1542 myself these days, so testing is slow. Part 1 is trivial: put #if 0 ... #endif around the parts where scsi_error tries to reset the scsi bus and the host adapter. (It is possible that the only very bad driver is the aha1542, in which case a test instead of an #if 0 might be appropriate.)
Andries
P.S. I hope to release util-linux-2.9 this week. mkswap/mount/fdisk/shutdown have changed a lot. Comments, especially from people with peculiar architectures and/or libc's are welcome. See ftp.win.tue.nl:/pub/linux/util/util-linux-pre2.9c.tar.gz .
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |