2.5.30 breaks cciss driver?

2.5.30 breaks cciss driver?

[Stephen Cameron]

I just saw this problem with 2.5.30.

I can't mount my 2nd volume on a cciss controller (SmartArray 5i)

< /dev/cciss/c0d1p1 /u1
No such device or address

The first volume, /dev/cciss/c0d0p1, works fine
(I'm booted from it.)

Reboot 2.5.29, both volumes work fine.

I don't have time to look into this right now,
but I thought I'd mention it in case someone else
does have time.  Looks like there was some partition 
code and/or devfs changes...

-- steve

Re: 2.5.30 breaks cciss driver?

[Greg KH]

On Fri, Aug 02, 2002 at 03:47:51PM -0500, Stephen Cameron wrote:
> 
> I just saw this problem with 2.5.30.
> 
> I can't mount my 2nd volume on a cciss controller (SmartArray 5i)
> 
> < /dev/cciss/c0d1p1 /u1
> No such device or address
> 
> The first volume, /dev/cciss/c0d0p1, works fine
> (I'm booted from it.)
> 
> Reboot 2.5.29, both volumes work fine.
> 
> I don't have time to look into this right now,
> but I thought I'd mention it in case someone else
> does have time.  Looks like there was some partition 
> code and/or devfs changes...

Are you running in "devfs=only" mode?  If so, the changes I made
probably are the cause of this.

thanks,

greg k-h

Re: 2.5.30 breaks cciss driver?

[Cameron, Steve]

Message 2 in thread 
Greg KH (greg@kroah.com) wrote:
> On Fri, Aug 02, 2002 at 03:47:51PM -0500, Stephen Cameron wrote:
> > 
> > I just saw this problem with 2.5.30.
> > 
> > I can't mount my 2nd volume on a cciss controller (SmartArray 5i)
> > 
> > < /dev/cciss/c0d1p1 
> > No such device or address
> > 
> > The first volume, /dev/cciss/c0d0p1, works fine
> > (I'm booted from it.)
> > 
> > Reboot 2.5.29, both volumes work fine.
> > 
> > I don't have time to look into this right now,
> > but I thought I'd mention it in case someone else
> > does have time.  Looks like there was some partition 
> > code and/or devfs changes...
>
> Are you running in "devfs=only" mode?  If so, the changes I made
> probably are the cause of this.
> 

No.  The changes that broke it are to do with the sizes[]
being gone, I think.  That yours?

/dev/cciss/c0d1 shows up in /proc/partitions, but not
/dev/cciss/c0d1p1 or /dev/cciss/c0d1p7

All the partitions of /dev/cciss/c0d0 show up.

(still broken in 2.5.31, of course)

I'm having some trouble with instability of 2.5.3[01],
don't know if it's just life with 2.5, or maybe my userland
is too old for my kernel...(redhat 7.1) Anyway, it's 
slowing me down, running fsck constantly after puking. :-(

-- steve



-

Re: 2.5.30 breaks cciss driver?

[Greg KH]

On Tue, Aug 13, 2002 at 10:09:32AM -0500, Cameron, Steve wrote:
> Message 2 in thread 
> Greg KH (greg@kroah.com) wrote:
> >
> > Are you running in "devfs=only" mode?  If so, the changes I made
> > probably are the cause of this.
> > 
> 
> No.  The changes that broke it are to do with the sizes[]
> being gone, I think.  That yours?

Nope, thankfully :)

Sorry, but I have no idea of what could be wrong.

greg k-h

Re: 2.5.30 breaks cciss driver?

[Stephen Cameron]

Hmmm, It appears that now we need a gendisk per disk, rather 
than per controller, as this snippet from patch-2.5.30 implies?
And we need to set the gendisk's first_minor to be the minor 
number of the first partition for each disk, is that about right?

-       hd_gendisk.nr_real = NR_HD;
-
-       for(drive=0; drive < NR_HD; drive++)
-               register_disk(&hd_gendisk, mk_kdev(MAJOR_NR,drive<<6), 1<<6,
+       for(drive=0; drive < NR_HD; drive++) {
+               hd_gendisk[i].nr_real = 1;
+               add_gendisk(hd_gendisk + drive);
+               register_disk(hd_gendisk + drive,
+                       mk_kdev(MAJOR_NR,drive<<6), 1<<6,
                        &hd_fops, hd_info[drive].head * hd_info[drive].sect *
                        hd_info[drive].cyl)

I think this is why cciss is broken, we have just one gendisk
per controller.  (please pardon my ignorance, I haven't ever 
had to look at this partition related code before.)

-- steve