Pixel wrote:
> Borsenkow Andrej <[EMAIL PROTECTED]> writes:
>
>
>>Pixel wrote:
>>
>>>in fact, the best solution is to use
>>>probeall scsi_hostadapter scsi_hostadapter1 scsi_hostadapter2 ...
>>>and to perl -pi -e 's/scsi-hosts/scsi_hostadapter/g' /etc/modules.devfs
>>>that way, both devfs and non-devfs will work nicely.
>>>
>>How should it work with non-devfs? And I do not see any point in chaging the
>>name - scsi-hosts is already used in modules.devfs and is good mnemonic name.
>>
>
> because of:
>
> pixel@no:/home/chmou/rpm/BUILD/linux>grep -2 -r scsi_hostadapter .
> ./drivers/scsi/scsi.c-#ifdef CONFIG_KMOD
> ./drivers/scsi/scsi.c- if (scsi_hosts == NULL)
> ./drivers/scsi/scsi.c: request_module("scsi_hostadapter");
> ./drivers/scsi/scsi.c-#endif
> ./drivers/scsi/scsi.c- return scsi_register_device_module((struct
>Scsi_Device_Template *) ptr);
>
> The kernel explicitly modprobe's scsi_hostadapter. So unless we patch the
> kernel to make it modprobe scsi-hosts, we should patch devfsd to modprobe
> scsi_hostadapter instead of scsi-hosts.
>
> # If they asked for ide-scsi, load it
Nope, it does not work for non-devfs case (tested). Kernel will probe
scsi_hostadapter *only* if no hostadapters are registered. It means, if
you have two (me as example :-) and one of them is loaded and other not
- kernel will not try to modporbe host_adapter at all (which is awfull
kludge anyway). Some more random comments:
- loading such modules as ide-scsi and ppa results in nasty console
output. If they are going to be loaded on demand, something needs to be
done about it
- ide-scsi is unconditionally loaded in rc.sysinit:
if grep -q "ide-scsi" /proc/cmdline ; then
modprobe ide-cd >/dev/null 2>&1
modprobe ide-scsi >/dev/null 2>&1
It may be leftover, now ide-scsi is preloaded in /etc/modules; so this
probably should be removed? In any case, it prevents autocleaning module.
- there is someting strange with ide-scsi CDs; instead of creating link
to /dev/scsi/... plain block device is created:
filr-xr-xr-x 1 root root 33 ??? 1 1970 cdrom0 ->
../ide/host0/bus1/target0/lun0/cd
brw-rw-rw- 1 root root 11, 0 ??? 1 1970 cdrom1
I do not know if it may have some side-effect; in any case, permisions
from pam_console are not applied here so you get world-readable device.
It may be considered a problem. OTOH canonical path has correct permissions:
{pts/1}% ll /dev/scsi/host0/bus0/target0/lun0/cd
brw-rw---- 1 bor cdwriter 11, 0 ??? 9 23:15
/dev/scsi/host0/bus0/target0/lun0/cd
as tey should be. I have no idea who does it.
-andrej
