Re: 2.4.19pre5aa1 and splitted vm-33

[Christoph Hellwig]

> Only in 2.4.19pre5aa1: 20_kiobuf-slab-2
> 
> 	Other patch from Chuck Lever to allocate the whole kiobuf from slab,
> 	using pointer to arrays, to have small structures not larger
> 	than one page (that would otherwise become sensitive to
> 	physical ram page granular fragmentation). While auditing and
> 	merging the patch I also did various small fixes, so please rely
> 	on this patch as a final version, because the original one had a few
> 	bugs/leftovers.

What is the reason for adding kio_ prefixes to the 'blocks' and 'bh' members
of struct kiobuf? - These fields haven't any new semantics at all, but that
change increases the delta dramatically.

> 	Merged various fixes, in particular delayed writes should now be
> 	really flushed by the VM, previously the fact the vm holds
> 	the page lock was forbidding that. New logic is much cleaner too.

Do you plan to submit the XFS delalloc changes back to SGI?

	Christoph