Re: [git patches] xfs and block fixes for virtually indexed arches

[tytso@mit.edu]

On Thu, Dec 17, 2009 at 12:10:25PM -0500, Christoph Hellwig wrote:
> On Thu, Dec 17, 2009 at 11:30:36AM -0500, tytso@mit.edu wrote:
> > That's because apparently the iSCSI and DMA blocks assume that they
> > have Real Pages (tm) passed to block I/O requests, and apparently XFS
> > ran into problems when sending vmalloc'ed pages.  I don't know if this
> > is a problem if we pass the bio layer addresses coming from the SLAB
> > allocator, but oral tradition seems to indicate this is problematic,
> > although no one has given me the full chapter and verse explanation
> > about why this is so.
> 
> Actually at least iscsi now has a workaround for that by checking for
> PageSlab.  Back when we deal with the XFS issue that check was only
> available with debug options enabled.  I tried to sort it out by
> agreeing with the block and iscsi folks that either
> 
>  a) we need to send down refcountable pages
>  b) block drivers need to accept kmalloced pages
> 
> I could not get any afreement, and thus we stopped using the kmalloced
> pages in XFS which was easy enough.  A bit later people fixed iscsi,
> but we still don't have formal rules about what is acceptable to the
> block layer.

It would be good to get some formal rules articulated.  Someone has
asserted that the AoE (ATA over Ethernet) driver will barf on
SLAB/kmalloc allocated memory.   True, false?

					- Ted