XPost: linux.debian.maint.boot
From:
[email protected]
Hi,
Executive summary:
1. Patch by Vincent McIntyre (Bug#265636) has good intent but buggy
as is. (I wish this type of syntax exists for "fail" function.)
2. My previous patch was not general enough.
3. My new patch here should work but intrusive at this moment.
(This can be better. "fail" and "exit 1" need to be cleaned.)
4. I lack time to test throrouly for the release.
5. Realistically, ERRATA is the good aproach now considering release
schedule.
Proposed ERRATA:
----------
In some CDROM drives such as the ones on the older Dell laptop PCs,
second detection may be successful. If you fail in mounting CD, you
simply run detect CDROM drive again from the menu. If reading CD fails
with installation log in vt4 indicating:
.... anna[4147]: /cdrom/dists//Release
.... anna[4147]: : No such file or directory
then, you need to go to console and run "umount /cdrom" before running
to detect CDROM drive from the menu.
------------
My thoughts...
After some thought, I think situation of funny CDROM firmware/Linux
kernel interaction as followings.
In old tc1, buggy DELL CDROM drive DMA related initialization during
mount caused mount error but mounted CD halfway (no symlink).
In new rc1, buggy DELL CDROM drive the same DMA related initialization
during mount causes no mount error while still mounting CD halfway (no symlink). agh.
My initial attempt to work around in RC1 relied on failed action of
mount.
My new thought is if HW is buggy, we should tell user about it and offer
way to proceed. Currently, you need to umount cd from the console
unless you are expert. This is not desirable. Centrarized automatic
umount should be implemented and easier to manage.
Also dummy mount/umount code in my previous mail was commented out now
so no one will use it unless you change source but as reminder. I also
removed section which is based on tc1 fix scenario.
Also I fixed funny confused use of fail/log functions in Vincent's
oroginal patch.
* Any opinion?
* Should dummy mount/umount be commented out?
* Is it worth patching now?
* I think reorganizing whole code makes it much more readable.
* Further code fix should be etch thing.
Osamu
--- cdrom-detect.postinst.old 2004-07-02 02:23:52.000000000 +0200
+++ cdrom-detect.postinst 2004-08-17 00:21:06.000000000 +0200
@@ -9,7 +9,9 @@
}
fail () {
- log "CDROM-detect failed."
+ log "CDROM-detect failed. Unmounting cdrom just to be sure."
+ umount /cdrom 2>/dev/null || true
+ log "For buggy CDROM drive, you may try CDROM-detect again."
db_input critical cdrom-detect/failure || [ $? -eq 30 ]
db_go
exit 1
@@ -40,6 +42,10 @@
fi
for device in $devices
do
+# For some buggy CDROM drive, blindly mount and umount
+# CDROM will avoid DMA issues with kernel.
+ #mount -t iso9660 -o ro,exec $device /cdrom || true
+ #umount /cdrom 2>/dev/null || true
if mount -t iso9660 -o ro,exec $device /cdrom; then
log "CDROM-mount succeeded: device=$device"
mounted=1
@@ -49,18 +55,6 @@
log "CDROM-mount failed (error=$?): device=$device"
log "Unmounting CD just to be sure."