Re: 2.6.25-rc1 ahci regression

[Anton Vorontsov <cbouatmailru@gmail.com>]

On Thu, Jun 03, 2010 at 11:45:33PM +0800, Américo Wang wrote:
> On Thu, Jun 03, 2010 at 07:04:22PM +0400, Anton Vorontsov wrote:
> >On Thu, Jun 03, 2010 at 10:13:57PM +0800, Luming Yu wrote:
> >> On Thu, Jun 3, 2010 at 9:50 PM, Anton Vorontsov <cbouatmailru@gmail.com> wrote:
> >> > On Thu, Jun 03, 2010 at 05:23:19PM +0400, Anton Vorontsov wrote:
> >> > [...]
> >> >> Oh. It seems I found it. It must be the module order loading
> >> >> thing.
> >> >>
> >> >> Can you try the patch down below?
> >> >
> >> > OTOH... libahci is a link-time dependency. And the order doesn't
> >> > matter here. So you should just have the libahci module in
> >> > the initrd (actually, mkinitrd or whatever you use should
> >> > have included this automatically).
> >> >
> >> 
> >> Neither ahci nor libachi is automatically included in initrd..
> >
> >What commands you use to build the kernel and initrd?
> >
> >> Manually added ahci and libachi into initrd seems to make the problem go away..
> >> Any idea why that could happen?
> >
> >No idea off hand.
> >
> >Cc'ing Kbuild folks. Question is: is there something wrong with
> >writing stuff like this:
> >obj-$(CONFIG_SATA_AHCI)         += ahci.o libahci.o
> >obj-$(CONFIG_SATA_AHCI_PLATFORM) += ahci_platform.o libahci.o
> >as in drivers/ata/Makefile?
> >
> >It seem to link fine, .ko's are created. But for some reason
> >initrd doesn't include these modules for Luming...
> >
> 
> Odd.
> Are these modules exported to /lib/modules/`uname -r`/ directory?

At least for me, yes (debian sid). It all looks sane.

$ find /tmp/abc/
/tmp/abc/
/tmp/abc/lib
/tmp/abc/lib/modules
/tmp/abc/lib/modules/2.6.35-rc1+
/tmp/abc/lib/modules/2.6.35-rc1+/modules.dep
/tmp/abc/lib/modules/2.6.35-rc1+/modules.dep.bin
/tmp/abc/lib/modules/2.6.35-rc1+/modules.alias
/tmp/abc/lib/modules/2.6.35-rc1+/modules.softdep
/tmp/abc/lib/modules/2.6.35-rc1+/source
/tmp/abc/lib/modules/2.6.35-rc1+/modules.symbols
/tmp/abc/lib/modules/2.6.35-rc1+/kernel
/tmp/abc/lib/modules/2.6.35-rc1+/kernel/drivers
/tmp/abc/lib/modules/2.6.35-rc1+/kernel/drivers/ata
/tmp/abc/lib/modules/2.6.35-rc1+/kernel/drivers/ata/ahci_platform.ko
/tmp/abc/lib/modules/2.6.35-rc1+/kernel/drivers/ata/libahci.ko
/tmp/abc/lib/modules/2.6.35-rc1+/kernel/drivers/ata/ahci.ko
/tmp/abc/lib/modules/2.6.35-rc1+/kernel/drivers/scsi
/tmp/abc/lib/modules/2.6.35-rc1+/kernel/drivers/scsi/scsi_wait_scan.ko
/tmp/abc/lib/modules/2.6.35-rc1+/modules.builtin.bin
/tmp/abc/lib/modules/2.6.35-rc1+/modules.alias.bin
/tmp/abc/lib/modules/2.6.35-rc1+/build
/tmp/abc/lib/modules/2.6.35-rc1+/modules.order
/tmp/abc/lib/modules/2.6.35-rc1+/modules.symbols.bin
/tmp/abc/lib/modules/2.6.35-rc1+/modules.builtin
$ cat /tmp/abc/lib/modules/2.6.35-rc1+/modules.dep
kernel/drivers/scsi/scsi_wait_scan.ko:
kernel/drivers/ata/libahci.ko:
kernel/drivers/ata/ahci.ko: kernel/drivers/ata/libahci.ko
kernel/drivers/ata/ahci_platform.ko: kernel/drivers/ata/libahci.ko

I just found Fedora's mkinitrd, and I see this:

~/mkinitrd$ git grep -A 5 -B 5 ahci
functions-    for m in $items ; do
functions-      char=$(echo $m | cut -c1)
functions-      if [ $char = '=' ]; then
functions-          NAME=$(echo $m | cut -c2-)
functions-          if [ "$NAME" = "ata" ]; then
functions:              MODS="$MODS $(cat /lib/modules/$kernel/modules.block |egrep '(ata|ahci)' |sed -e 's/.ko//')"
functions-          else
functions-                # Ignore if group list does not exist
functions-                if [ -e /lib/modules/$kernel/modules.$NAME ]; then
functions-                  MODS="$MODS $(cat /lib/modules/$kernel/modules.$NAME |sed -e 's/.ko//')"
functions-                fi

Hm? What is this? Maybe it breaks when it sees libahci?

-- 
Anton Vorontsov
email: cbouatmailru@gmail.com
irc://irc.freenode.net/bd2