Re: [Xen-devel] gnttab_setup_table attempting copy_from_guest in atomic context

[Keir Fraser <keir@xxxxxxx>]

On 01/12/2010 22:48, "Daniel De Graaf" <dgdegra@xxxxxxxxxxxxx> wrote:

> do_grant_table_op takes domain_lock(d), and then calls gnttab_setup_table
> with the lock held. Since changeset 22402:7d2fdc083c9c, this will cause the
> copy to fail with a rather misleading message:
> 
> (XEN) grant_table.c:1315:d1 Fault while reading gnttab_query_size_t.
> 
> This causes grants between HVM guests to fail.
> 
>> From looking at the code, it appears that at least some of the functions
> within do_grant_table_op do not need to hold domain_lock. Is there a reason
> to hold domain_lock when manipulating the grant table? Most of the critical
> operations already appear to be protected by the grant table lock, so it may
> be safe to completely remove the lock_domain calls in this function.

Yes, the grant locking will need reworking. Xenpaging is going to require
the ability to sleep in __hvm_copy(), and this means that all callers --
direct or indirect -- must hold no locks. Will have to take a look.

 -- Keir



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