Xen devel - Oct 2009 - [PATCH] Bug in xmalloc_bytes()

Keir,

Dan Magenheimer said I should forward this on to you, last week, as he was out.
I meant to do some further testing against the top of xen-unstable, but
something came up and I can''t deal with this, further, right now.
I''m sending my patch, as-is, so people can be aware of the bug.

John Byrne
 

-----Original Message-----
From: Byrne, John (HP Labs) 
Sent: Tuesday, October 13, 2009 6:32 PM
To: ''dan.magenheimer@oracle.com''
Subject: Bug in xmalloc_bytes()

Dan,

Since you ported the xmalloc_tlsf code, I thought I''d talk to you
first.

With the various roundings, xmalloc_bytes() fails consistently with requests of
3921 to 3952 bytes.

3921 + 128 (SMP_CACHE_BYTES)- 16 (MEM_ALIGN) = 4033 in _xmalloc(); then
MAPPING_SEARCH() rounds the size up to 4096, at which point you fail the
"if ( size > (pool->grow_size - 2 * BHDR_OVERHEAD) )" test in
xmem_pool_alloc().

Attached is the patch I came up with. I''ve tested it against:

changeset:   19105:696351cde9a4
user:        Keir Fraser <keir.fraser@citrix.com>
date:        Wed Jan 28 16:58:41 2009 +0000
summary:     Allow memflags to be specified to alloc_xenheap_pages().

Not the latest stuff with the tmem hooks, but I don''t think they will
affect anything.

Feel free to test, fold, spindle, or mutilate it and/or send it on to Keir.

John







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

Thanks, I''ve applied a different fix as xen-unstable:20349

 -- Keir

On 21/10/2009 02:49, "Byrne, John (HP Labs)"
<john.l.byrne@hp.com> wrote:
> Keir,
> 
> Dan Magenheimer said I should forward this on to you, last week, as he was
> out. I meant to do some further testing against the top of xen-unstable,
but
> something came up and I can''t deal with this, further, right now.
I''m sending
> my patch, as-is, so people can be aware of the bug.
> 
> John Byrne
>  
> 
> -----Original Message-----
> From: Byrne, John (HP Labs)
> Sent: Tuesday, October 13, 2009 6:32 PM
> To: ''dan.magenheimer@oracle.com''
> Subject: Bug in xmalloc_bytes()
> 
> Dan,
> 
> Since you ported the xmalloc_tlsf code, I thought I''d talk to you
first.
> 
> With the various roundings, xmalloc_bytes() fails consistently with
requests
> of 3921 to 3952 bytes.
> 
> 3921 + 128 (SMP_CACHE_BYTES)- 16 (MEM_ALIGN) = 4033 in _xmalloc(); then
> MAPPING_SEARCH() rounds the size up to 4096, at which point you fail the
"if (
> size > (pool->grow_size - 2 * BHDR_OVERHEAD) )" test in
xmem_pool_alloc().
> 
> Attached is the patch I came up with. I''ve tested it against:
> 
> changeset:   19105:696351cde9a4
> user:        Keir Fraser <keir.fraser@citrix.com>
> date:        Wed Jan 28 16:58:41 2009 +0000
> summary:     Allow memflags to be specified to alloc_xenheap_pages().
> 
> Not the latest stuff with the tmem hooks, but I don''t think they
will affect
> anything.
> 
> Feel free to test, fold, spindle, or mutilate it and/or send it on to Keir.
> 
> John
> 
> 
> 
> 
> 


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