Xen devel - Mar 2010 - [PATCH] x86: fix improper return value from relinquish_memory()

While apparently only a theoretical possibility (domain_kill() has a
BUG_ON() that wasn''t reported to trigger so far), I still think it is
better to have the code cleaned up.

Signed-off-by: Jan Beulich <jbeulich@novell.com>

--- 2010-03-22.orig/xen/arch/x86/domain.c	2010-03-22 00:00:00.000000000 +0100
+++ 2010-03-22/xen/arch/x86/domain.c	2010-03-24 11:57:49.000000000 +0100
@@ -1821,8 +1821,10 @@ static int relinquish_memory(
         {
         case 0:
             break;
-        case -EAGAIN:
         case -EINTR:
+            ret = -EAGAIN;
+            /* fallthrough */
+        case -EAGAIN:
             page_list_add(page, list);
             set_bit(_PGT_pinned, &page->u.inuse.type_info);
             put_page(page);





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

>>> "Jan Beulich" <JBeulich@novell.com> 24.03.10 12:07
>>>
>While apparently only a theoretical possibility (domain_kill() has a
>BUG_ON() that wasn''t reported to trigger so far), I still think it
is
>better to have the code cleaned up.
Btw., the reason I was looking at that code was that we observe
zombie domains - ones in DOMDYING_dead state, perhaps having
almost none of their memory freed (shadowed guests appear to be
particularly bad). In one of the reports, an interesting extra fact
was that this happened only for the first 100 guests - any
subsequent ones got destroyed properly (obviously to get there
this requires quite a bit of memory in the host). Has anyone else
observed this? Does this ring any bells?

Jan


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

On Wed, Mar 24, 2010 at 11:23:32AM +0000, Jan Beulich
wrote:
> >>> "Jan Beulich" <JBeulich@novell.com> 24.03.10
12:07 >>>
> >While apparently only a theoretical possibility (domain_kill() has a
> >BUG_ON() that wasn''t reported to trigger so far), I still
think it is
> >better to have the code cleaned up.
> 
> Btw., the reason I was looking at that code was that we observe
> zombie domains - ones in DOMDYING_dead state, perhaps having
> almost none of their memory freed (shadowed guests appear to be
> particularly bad). In one of the reports, an interesting extra fact
> was that this happened only for the first 100 guests - any
> subsequent ones got destroyed properly (obviously to get there
> this requires quite a bit of memory in the host). Has anyone else
> observed this? Does this ring any bells?
Yes.
http://lists.xensource.com/archives/html/xen-devel/2008-12/msg00222.html

B/c of the page count we had guests that would never have their mmap
count removed causing them to be zombie guests. Our fix, which wasn''t
nice, was to have the guest domain id re-number and shove it and its
remaining page ownership (at that point it only has some pages in Dom0
and DomU) in a corner. 

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

On 24/03/2010 14:43, "Konrad Rzeszutek Wilk"
<konrad.wilk@oracle.com> wrote:
>> Btw., the reason I was looking at that code was that we observe
>> zombie domains - ones in DOMDYING_dead state, perhaps having
>> almost none of their memory freed (shadowed guests appear to be
>> particularly bad). In one of the reports, an interesting extra fact
>> was that this happened only for the first 100 guests - any
>> subsequent ones got destroyed properly (obviously to get there
>> this requires quite a bit of memory in the host). Has anyone else
>> observed this? Does this ring any bells?
> 
> Yes.
> http://lists.xensource.com/archives/html/xen-devel/2008-12/msg00222.html
> 
> B/c of the page count we had guests that would never have their mmap
> count removed causing them to be zombie guests. Our fix, which
wasn''t
> nice, was to have the guest domain id re-number and shove it and its
> remaining page ownership (at that point it only has some pages in Dom0
> and DomU) in a corner.
There''s a big difference between a zombie domain owning a few pages
versus a
zombie domain still having most of its memory, though. One could be a
ref-count leak, the other sounds potentially like something wrong with the
domain-killing routines (not that there''s enough data to definitely say
either way for sure yet -- but at least they do sound like different bugs).

 -- Keir



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