Change in add_gendisk() in 2.4.11-preXXX

Change in add_gendisk() in 2.4.11-preXXX

[John Byrne]

Cristoph,

Looking in the Changelog for 2.4.11-pre, I find that you are given
credit for the change to add_gendisk() which should prevent the
/proc/paritions loop; which is good. However, I was tracing the bug
myself and come to the conclusion the culprit in my case was the "sd"
driver. One of our systems has two different SCSI HBAs and this resulted
in two calls to sd_finish() which results in the sd_gendisk structures
being added twice and, hence, the loop. So, I am a little concerned that
your change is covering up the problem so well, that the actual issue
may not be addressed. Unfortunately, I don't understand the ins and outs
of the SCSI and blkdev layers to suggest a fix for "sd".

So, my question to you or anyone is, is anyone looking further at the
causes of the problem right now?

John Byrne

Re: Change in add_gendisk() in 2.4.11-preXXX

[Christoph Hellwig]

On Fri, Oct 05, 2001 at 11:02:32AM -0700, John Byrne wrote:
> 
> Cristoph,
> 
> Looking in the Changelog for 2.4.11-pre, I find that you are given
> credit for the change to add_gendisk() which should prevent the
> /proc/paritions loop; which is good.

In fact it was Alan's fix, I just sent it to Linus 8)

> However, I was tracing the bug
> myself and come to the conclusion the culprit in my case was the "sd"
> driver. One of our systems has two different SCSI HBAs and this resulted
> in two calls to sd_finish() which results in the sd_gendisk structures
> being added twice and, hence, the loop. So, I am a little concerned that
> your change is covering up the problem so well, that the actual issue
> may not be addressed. Unfortunately, I don't understand the ins and outs
> of the SCSI and blkdev layers to suggest a fix for "sd".

If you look at the new add_gendisk structure it is clearly marked that
the new bahaviour is a workaround.

It will be fixed in 2.5, but the scsi layer is too fragile to do such
changes in 2.4.

	Christoph

-- 
Of course it doesn't work. We've performed a software upgrade.