Re: 2.6.14 status summary, and upcoming 2.6.15 ...
On Mon, Nov 28, 2005 at 04:12:34AM +0100, Jonas Smedegaard wrote:
> On Mon, 28 Nov 2005 02:48:01 +0100 Sven Luther <sven.luther@wanadoo.fr> wrote:
> > On Mon, Nov 28, 2005 at 02:23:32AM +0100, Jonas Smedegaard wrote:
> > > On Mon, 28 Nov 2005 00:13:50 +0100 Sven Luther <sven.luther@wanadoo.fr> wrote:
> Yaird only by default supports drivers using sysfs. I suspect that the
> issue of bug#328534 is a driver module not properly using sysfs. And I
> guess the workaround is simply to add it manually
> to /etc/yaird/Default.cfg. But it is pure speculation - I'd really want
> Erik's opinion.
Hmm, long and wandering history in that bug. If I read it correctly,
user can boot 2.6.8, then run yaird for 2.6.14. This results in a boot
image, booting it ends with complaints about 'sda1 fails to show up'.
The SCSI driver for relevant controller has caused problems in recent kernels.
This suggests:
1. Yaird believes it can build an image. This happens only if /sys/block/sda1
contains a symlink to something that looks credible as a disk driver in /sys/devices/
2. Loading the drivers that yaird selects is not enough to get the hardware
running to a level where it shows up in /sys/block/sda.
Possible explanations:
a. Driver has misleading sysfs information, that leads to a module not being loaded.
b. Driver only becomes active after an unrelated module is loaded,
like old versions of via82cxxx that only work after ide-generic is loaded.
c. Driver does something clever that is not recognised by yaird,
like /sbin/modprobe to start raid6 support.
d. Driver does not work at all.
To find out what is the case, start by looking at the /init script in the
boot image. Does this contain the modules you want it to contain?
The last module before mkbdev sda is suspect: it should be the one that
creates the missing /sys/block/sda/dev. What's that last module?
So looks like a driver problem, at this point not enough information to be
sure the problem is in the sysfs part of the driver, or to trust that
it can be made to work by loading extra modules.
> > BTW, also, about ia64 situation, is it possible to use yaird in
> > initrd case still ? I know Eric was disabling this for 0.12, but we
> > are still using 0.11 ?
>
> Yaird 0.0.11 is newest release - so it _will_ be removed but _was_ not
> yet.
>
> It is deprecated due to lack of testing. So if someone starts testing
> it then Erik might get convinced to keep it:
> Replace /etc/yaird/Templates.cfg
> with /usr/share/doc/yaird/examples/Debian-initrd.cfg.gz gunzipped
> should do the trick.
Seems to be fixed in lilo in the mean time, so I'll not look into this.
Regards,
Erik
Reply to: