Bug#345067: My understanding of the IDE mess, and why it does not make sense to apply the proposed patch
On Fri, 10 Mar 2006, Jonas Smedegaard wrote:
If modular-ide is the sole source of trouble here, then what worked in
2.6.14-4 and earlier?
The bugreports seem to indicate that things broke in 2.6.14-5 that
worked in 2.6.14-4. And it seems nothing related else than linux-2.6
changed then - not yaird and not kernel-package.
I have compared the 2.6.14-4 and 2.6.14-5 linux-2.6 packages. The
configs used to build the kernel are almost identical, and the 3
patches added in 2.6.14-5 do not touch anything in the IDE subsystem
at all. Both have the modular IDE patch applied. File lists of
/lib/modules/[version]/kernel/drivers/ide are exactly the same. Thus,
it is pretty unlikely that kernel is the cause of problems with
ide-generic.
Yaird, on the other hand, contains some relevant changes between the
last known working version (0.0.11-12) and the broken one (0.0.12-1).
Out of the complete diff between the source the following three hunks
are relevant for ide-generic handling:
--- yaird-0.0.11-12/perl/Hardware.pm 2006-03-10 17:28:03.852469888 -0800
+++ yaird-0.0.12-1/perl/Hardware.pm 2005-12-08 14:42:33.000000000 -0800
[..]
@@ -88,9 +113,42 @@
# Other mac-io devices are not storage related.
$modules = [ "mesh" ];
}
-
+ elsif ($abspath =~ m!/ide\d+$!) {
+ # IDE bus. On this end of the
+ # cable we normally have on the PCI bus a chipset
+ # controlling the IDE bus, so that the whole looks
+ # like this:
+ # /sys/devices/pci0000:00/0000:00:11.1/ide0/0.0
+ # Here the driver for the PCI device will know how
+ # to manage the IDE bus, and there's no need to load
+ # a driver for the bus as such.
+ #
+ # Except that there also is this:
+ # /sys/devices/ide0/0.0
+ # This happens with the ide-generic driver. This
+ # is a driver that manages any PCI/IDE controller,
+ # regardless of device or vendor ID.
+ # The ide-generic driver does not do DMA, which
+ # means it can be an order of magnitude slower than
+ # drivers specific for a chipset. Avoid if possible.
+ # The ide-generic driver will not show the PCI device
+ # in /sys/devices; this is an indicator for yaird to
+ # add ide-generic to the image.
+ #
+ # Note that there is discussion over whether
+ # ide-generic # should grab otherwise unhandled
+ # IDE devices.
+ # - http://thread.gmane.org/gmane.linux.hotplug.devel/6003
+ # - http://lists.debian.org/debian-kernel/2004/11/msg00218.h
tml
+ # - http://www.ussg.iu.edu/hypermail/linux/kernel/0410.1/145
2.html
+ #
+ if ($i == 0) {
+ $modules = [ "ide-generic" ];
+ }
+ }
elsif ($abspath =~ m!/ide\d+/\d+\.\d+$!) {
- # IDE device
+ # IDE device - this is the device at the other end
+ # of the cable; normally a disk.
my $dev = IdeDev->new (path => $abspath);
$modules = IdeDev::findModuleByIdeDev ($dev);
}
[..]
@@ -145,6 +205,19 @@
push @{$result}, @{$modules};
}
}
+
+ #
+ # Hmm, via82cxxx (2.6.8) also needs ide-generic
+ # to load it seems. That could be because ide-generic
+ # contains a call to ide_probe_init() which is in
+ # the ide-core module. After loading ide-generic
+ # it's still the via82cxxx module that manages the device;
+ # see /sys/bus/pci/drivers/VIA_IDE/.
+ # The above error persists in 2.6.12, and is solved
+ # in 2.6.14.
+ #
+ $result = [ map { $_ eq "via82cxxx" ? ($_, "ide-generic") : $_ } @{$result}
];
+
return $result;
}
[..]
--- yaird-0.0.11-12/perl/IdeDev.pm 2005-08-07 13:57:40.000000000 -0700
+++ yaird-0.0.12-1/perl/IdeDev.pm 2005-12-08 14:42:33.000000000 -0800
[..]
@@ -108,16 +86,14 @@
my ($ideDev) = @_;
my $media = $ideDev->media();
my $result = [];
- if (! KConfig::isOmitted ("ide-generic")) {
- # Supply ide-generic as default chipseet driver only if
- # it was compiled into the kernel.
- push @{$result}, "ide-generic";
- }
my $driver;
$driver = "ide-disk" if ($media eq "disk");
$driver = "ide-tape" if ($media eq "tape");
+
+ # The CD may also need ide-generic.
$driver = "ide-cd" if ($media eq "cdrom");
+
$driver = "ide-floppy" if ($media eq "floppy");
if (defined ($driver)) {
push @{$result}, $driver;
Judging by the comments, the first one is supposed to take care of
situations when one of the IDE devices is handled by ide-generic. In
that case the sysfs entry looks like /sys/devices/ide0/0.0, and
ide-generic is added to initrd in that case. The second one is the
via82cxxx kludge. And I believe that the third one is responsible for
breakage. It appears, that 0.0.11-12 had some logic in perl/IdeDev.pm
to add ide-generic to the initrd in the case when CONFIG_IDE_GENERIC
is present in the kernel config, but for some reason it got removed in
0.0.12-1. Testing on my box (with ICH4 controller) shows that both
2.6.14-4 and 2.6.14-5 work with 0.0.11-12 (ide-generic is included in
initrd), fail with 0.0.12-1 (ide-generic is not included in initrd),
and work again with 0.0.12-1 if the third hunk mentioned above is
reversed.
Best regards,
Jurij Smakov jurij@wooyd.org
Key: http://www.wooyd.org/pgpkey/ KeyID: C99E03CC
Reply to: