Xen devel - Aug 2013 - question about patch that allow guest to set/clear MSIx mask bit

Hi , Jan,

I saw you reverts the patch that allow guest to set/clear MSI-x bit. And you
only said it is not fully cooked. What''s the wrong with this patch?
Actually, we found a bug which introduced by this patch: MSI-x has several pirqs
that each vector mappings one. So the following assert is wrong:

ASSERT(entry->pirq == pirq);

And I found this patch was reverted before I send out the fixing patch.

commit 8a7769b4453168e23e8935a85e9a875ef5117253
Author: Jan Beulich <jbeulich@suse.com>
Date:   Mon Aug 26 12:40:44 2013 +0200

    Revert "interrupts: allow guest to set/clear MSI-X mask bit"
    
    This reverts commit 54a46bce768033b1c36e25eace15f7abde972389.
    It''s not fully cooked yet.



best regards
yang

>>> On 27.08.13 at 07:21, "Zhang, Yang Z"
<yang.z.zhang@intel.com> wrote:
> Hi , Jan,
> 
> I saw you reverts the patch that allow guest to set/clear MSI-x bit. And
you
> only said it is not fully cooked. What''s the wrong with this
patch?
> Actually, we found a bug which introduced by this patch: MSI-x has several 
> pirqs that each vector mappings one. So the following assert is wrong:
> 
> ASSERT(entry->pirq == pirq);
Did you not see the respective discussion
(http://lists.xenproject.org/archives/html/xen-devel/2013-08/msg02346.html)?

Bottom line - the assertion is _not_ wrong, what''s wrong is the
code. And the trivial fix suggested by Sander has other downsides.
Hence the author and I agreed to revert the patch until he can
provide a full, properly done one. Of course you''re welcome to
step in.

Jan