RE: [Xen-devel] [PATCH] "lock cmpxch8b" and split locks

["Puthiyaparambil, Aravindh" <aravindh.puthiyaparambil@xxxxxxxxxx>]

Which looks something like my attachment? :-)

Signed off by: Aravindh Puthiyaparambil
<aravindh.puthiyaparambil@xxxxxxxxxx>

> -----Original Message-----
> From: Keir Fraser [mailto:Keir.Fraser@xxxxxxxxxxxx]
> Sent: Wednesday, October 05, 2005 6:03 PM
> To: Keir Fraser
> Cc: Koren, Bradley J; xen-devel@xxxxxxxxxxxxxxxxxxx; Puthiyaparambil,
> Aravindh; Vessey, Bruce A; Subrahmanian, Raj
> Subject: Re: [Xen-devel] [PATCH] "lock cmpxch8b" and split locks
> 
> 
> On 5 Oct 2005, at 22:46, Keir Fraser wrote:
> 
> >> Does anyone know if there are other places where the "lock" prefix
is
> >> used with a cache misaligned address?
> >
> > x86 systems are supposed to guarantee that LOCKed instructions
access
> > their memory operand atomically, regardless of alignment (Vol 3 of
the
> > Intel reference manual). Your systems break this application-visible
> > guarantee?
> 
> Also, the patch is way bigger and more invasive than it needs to be.
> There should be no need to make pfn_info bigger than it is. It's
> currently a multiple of 8 bytes (e.g., 24 bytes on 32-bit) which is
> sufficient to avoid cache-line crossing of aligned 8-byte quantities.
> 
> What if we just move 'tlbflush_timestamp' to the end of the structure?
> A one-line fix? :-)
> 
>   -- Keir

Attachment: mm.patch
Description: mm.patch

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