Re: Checkin 7404ad3b6d04efbd918e9e2e776bf560fbedf47d breaks boot on KVM

["H. Peter Anvin" <hpa@zytor.com>]

James Bottomley wrote:
>>
>> :040000 040000 98fc7ae95211b8d16e2e8ca46997be660ad9ba93 
>> 2d35d0a2b6232411b840a9ccf6a52b946172566e M      drivers
>>
>> However, simply reverting this commit caused a panic on boot (not 
>> entirely surprising.)
> 
> Actually, it is surprising.  That patch takes the default arithmetic for
> calculating the disk size out of sd and uses a routine to do it more
> efficiently in lib/string_helpers.c
> 
> So there are two problems:  Why does it panic on revert (could you post
> the oops) and what does kvm object to in string_get_size ... it's a
> fairly innocuous routine as I read it ... your symptoms sound like the
> for loop isn't terminating.
> 

Looks like it's trying to print a zero.  I made a diff to print out the 
raw numbers, and did indeed get zero.

Now, *why* it is printing a zero is another matter (as is why using 
ffz() instead of ilog2() there...).  Some quick investigation showed 
that sdkp->capacity is a 32-bit quantity in this configuration, and 
shifting it left by 9 of course ends up with zero; being a virtual disk, 
it's an exact power of two.

Still, it is bad that string_get_size() hangs on passing zero.

The spinlock lockup is still happening, so I'm assuming it is an 
unrelated bug that was masked by the string_get_size issue.

Patch series will follow shortly.

	-hpa