Re: [Xen-devel] [PATCH v3 02/11] hvm/hpet: Only call guest_time_hpet(h) one time per action.

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 23.04.14 at 17:42, <dslutz@xxxxxxxxxxx> wrote:
> On 4/23/2014 11:07 AM, Jan Beulich wrote:
>>>>> On 17.04.14 at 19:42, <dslutz@xxxxxxxxxxx> wrote:
>>> v3:
>>>    Did not add Reviewed-by (Jan Beulich) do to amount of change
>>>    Added passing of guest_time to hpet_read64() and
>>>      hpet_stop_timer().
>> But that wasn't as a result of the review of v2, was it?
> 
> Not on this thread.  I was using gdb to get answers for a later patch.
> 
>>   Especially for
>> the hpet_read64() case I fear this is going a little too far, since now
>> you call the supposedly expensive function even when the value isn't
>> needed. I suppose you could resolve this by passing a known-invalid
>> value to the function from hpet_read() (so it knows to call
>> guest_time_hpet() itself) and use the always-preset variant on from
>> hpet_write().
> 
> I also considered a case statement on the call (since the value is 
> needed later when
> hpet_write also uses it).  I do not think that any value is invalid.

With

#define STIME_PER_HPET_TICK 16
#define guest_time_hpet(hpet) \
    (hvm_get_guest_time(vhpet_vcpu(hpet)) / STIME_PER_HPET_TICK)

any value with one or more of the high 4 bits set would be invalid.

Jan


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