Re: [Xen-devel] [PATCH v2 1/2] x86: defer processing events on the NMI exit path

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 01.03.13 at 17:01, Andrew Cooper <andrew.cooper3@xxxxxxxxxx> wrote:
> On 01/03/13 15:56, Keir Fraser wrote:
>> On 01/03/2013 11:53, "Jan Beulich" <JBeulich@xxxxxxxx> wrote:
>>
>>>> __softirq_pending is an unsigned long.  Would it not be prudent to use
>>>> cmpq to save obscure bugs if the implementation changes, or are we
>>>> sufficiently sure that this wont happen?
>>> All other existing instances of similar assembly code use testl or
>>> cmpl, and in fact I merely copied some other instance.
>>>
>>> As we're not getting close to 32, I think we might rather want to
>>> adjust the __softirq_pending type. Keir?
>> Yup, I don't see any reason why it couldn't be a uint32_t.
> 
> Further to that, is there any reason that it cant be per-cpu, to save
> having information like this moving around pcpus in a hot cache line?

The structure definition has __cacheline_aligned, there shouldn't
be much moving around (as long as our cache line size definition
is still suitable for modern CPUs' L1 caches at least).

Jan


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