Hi, Now this patch should be even better.
On Sat, Jun 19, 2004 at 10:15:14AM +1000, Vincent McIntyre wrote: > thanks for that careful analysis, Osamu. > I have very similar hardware and it appears the problem may also be > avoidable by configuring the kernel to not use DMA on cdroms. > See bugs #250323 (and referenced therein #222939.b) > Your solution is probably the least intrusive however. > If this patch is applied I would be happy to try it out as well. This prompted me to look in what really happening in stage 1 phase when CDROM is mounted. After couple learning experience, I now can build CD with modified initrd :-) (See http://people.debian.org/~osamu/hackdi/ ) I changed cdrom-detect.postinst as: --- cdrom-detect.postinst.org 1970-01-01 01:00:00.000000000 +0100 +++ cdrom-detect.postinst 2004-06-22 23:11:21.000000000 +0200 @@ -40,11 +40,28 @@ fi for device in $devices do - if mount -t iso9660 -o ro,exec $device /cdrom 2>/dev/null + if mount -v -t iso9660 -o ro,exec $device /cdrom >>error.log 2>&1 then mounted=1 db_set cdrom-detect/cdrom_device $device break + else + echo "ERROR_1=$?" >>/error.log + umount -v /cdrom >>/error.log 2>&1 || true + echo "=== UMOUNT 1 finished ===" >>/error.log + echo "=== SLEEP 1 ===" >>/error.log + sleep 1 >>/error.log || true + echo "=== SLEEP OVER ===" >>/error.log + if mount -v -t iso9660 -o ro,exec $device /cdrom >>/error.log 2>&1 + then + mounted=1 + db_set cdrom-detect/cdrom_device $device + break + else + echo "ERROR_2=$?" >>/error.log + umount -v /cdrom >>/error.log 2>&1 || true + echo "=== UMOUNT 2 finished ===" >>/error.log + fi fi done ------------- Idea was to read error output as much and any DMA error which trigger kernel to reset CDROM Disk IO as PIO mode should automatically fix issues. I added 1 second just be sure that kernel reset is complete. Alas, this worked without flaw. Here is the error.log file: -------------- mount: Mounting /dev/cdroms/cdrom0 on /cdrom failed: Not a directory ERROR_1=1 umount: /cdrom: Invalid argument === UMOUNT 1 finished === === SLEEP 1 === === SLEEP OVER === ---------------- I guess first read action produce error in Kernel and mount complain. Second try succeed since it is in PIO mode. ERROR=1 means incorrect invocation or permissions Applying above patch should give releaf to many old DELL users. Of course we should do some code clean up such as redirecting output to /dev/null instead. Osamu
signature.asc
Description: Digital signature
