Re: [Xen-devel] [PATCH] x86: compact supposedly unused entry point code

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 20.06.16 at 14:54, <andrew.cooper3@xxxxxxxxxx> wrote:
> On 20/06/16 13:48, Jan Beulich wrote:
>>>>> On 20.06.16 at 14:15, <andrew.cooper3@xxxxxxxxxx> wrote:
>>> On 20/06/16 12:04, Jan Beulich wrote:
>>>> No point in aligning entry points which aren't supposed to be used
>>>> anyway.
>>>>
>>>> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
>>> Reviewed-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
>> Thanks, but - any thoughts on this part:
>>
>> TBD: Might consider simply using "andq $-15,%rsp", delivering an
>> uninitialized error code (which shouldn't matter).
> 
> I was still considering that part.
> 
> These are entries we never expect to actually take.  At that point, the
> small overhead of setting up the error code to 0 is probably better than
> leaving it uninitialised.

I understand - it's really a matter of balancing the overhead on
these paths (which will never have an effect if these entries indeed
are unused, and which is of no interest if they are used by due some
other flaw) with the (likely negligible, but non-zero) overhead they
introduce on _other_ paths (due to cache and TLB consumption). I.e.
my goal was to make these unused entries as small as possible. And

        andq    $-15,%rsp
        movl    $vector,4(%rsp)

(obviously we can't use movb here) is smaller than the current

        testb   $8,%spl
        jz      1f
        pushq   $0
        movb    $vector,4(%rsp)

afaict.

Jan


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