Re: [Xen-devel] [PATCH 1/2] fix locking in offline_page()

[Tim Deegan <tim@xxxxxxx>]

At 11:48 +0100 on 27 Nov (1385549319), Tim Deegan wrote:
> At 10:43 +0000 on 27 Nov (1385545399), Jan Beulich wrote:
> > >>> On 27.11.13 at 11:34, Tim Deegan <tim@xxxxxxx> wrote:
> > > At 10:05 +0000 on 27 Nov (1385543154), Jan Beulich wrote:
> > >> >>> On 27.11.13 at 11:01, Andrew Cooper <andrew.cooper3@xxxxxxxxxx> 
> > >> >>> wrote:
> > >> > On 27/11/13 08:25, Jan Beulich wrote:
> > >> >>          else
> > >> >>          {
> > >> >>              *status = PG_OFFLINE_OWNED | PG_OFFLINE_PENDING |
> > >> >> -              (owner->domain_id << PG_OFFLINE_OWNER_SHIFT);
> > >> >> +                      (owner->domain_id << PG_OFFLINE_OWNER_SHIFT);
> > >> > 
> > >> > domain_id will be promoted from a uint16_t to an int32_t, then shifted
> > >> > left by 16 bits which is undefined if the top of domain_id bit was set.
> > >> 
> > >> That promotion is done by zero extension, so I don't figure what
> > >> you think is undefined here.
> > > 
> > > If the domid has bit 15 set, it will be shifted into the sign bit of
> > > the promoted integer; it should be explicitly cast to an unsigned type
> > > before the shift.
> > 
> > Again - I don't see this. Promotion happens before the shift,
> > i.e. 0x8000 -> 0x00008000 -> 0x80000000.
> 
> AIUI the default promotion is to a signed integer if the value will
> fit, i.e.:
>              (unsigned short) 0x8000 
> promoted     (signed int) 0x00008000 
> shifted left (signed int) 0x80000000 (undefined behaviour)
> 
> (but I don't have my copy of the standard here to check).

(and AFAICT there's no way for a real domain's domain_id to hva the
top bit set so it's a moot point).

Tim.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel