Re: [Xen-devel] build problem in linux-2.6.18-xen.hg?

[Grzegorz Milos <gm281@xxxxxxxxx>]

> *However*, cc'ing Gregor and Patrick as that raises the question of why this
> wasn't noticed sooner. Why do we have grant-table interface extensions being
> defined in a guest, but apparently no use of them (and hence no
> implementation of them?) in Xen itself? I'm talking about
> GNTMAP_can_fail/GNTCOPY_can_fail/GNTST_eagain -- is linux-2.6.18-xen
> defining using these just for its own amusement?

GNTMAP_* are meant to be used by Xen, however at the moment we've got
some leftovers from the original approach we used to handle OOM
conditions. In there, we maintained a pool of pages to handle certain
'critical' operations which should not fail (such as unsharing grated
pages). However, it quickly turned out that the memory pool would have
to grow unacceptably large. A more invasive approach, which we finally
decide to take, was to fixup all grant op callers to handle EAGAIN.

The last remaining patch (which I'm already working on) is to cleanup
the mem pool code in Xen, and return EAGAIN on failure instead.

For the time being, I have an explicit check (with a BUG()) on failed
page unshares, so that they'll never go unnoticed.

Thanks
Gregor

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel