Re: [Qemu-devel] [PATCH V2] [PowerPC][RFC] booke timers

[Fabien Chouteau]

On 05/07/2011 18:02, Scott Wood wrote:
> On Mon, 4 Jul 2011 17:06:54 +0200
> Fabien Chouteau <address@hidden> wrote:
> 
>> On 01/07/2011 22:22, Scott Wood wrote:
>>> On Fri, 1 Jul 2011 16:13:41 +0200
>>> Fabien Chouteau <address@hidden> wrote:
>>>> +static void booke_update_fixed_timer(CPUState         *env,
>>>> +                                     uint8_t           target_bit,
>>>> +                                     uint64_t          *next,
>>>> +                                     struct QEMUTimer *timer)
>>>> +{
>>>> +    ppc_tb_t *tb_env = env->tb_env;
>>>> +    uint64_t lapse;
>>>> +    uint64_t tb;
>>>> +    uint64_t period = 1 << (target_bit + 1);
>>>> +    uint64_t now;
>>>> +
>>>> +    now = qemu_get_clock_ns(vm_clock);
>>>> +    tb  = cpu_ppc_get_tb(tb_env, now, tb_env->tb_offset);
>>>> +
>>>> +    if (tb <= (1 << target_bit)) {
>>>> +        lapse = (1 << target_bit) - tb;
>>>> +    } else {
>>>> +        lapse = period - ((tb - (1 << target_bit)) % period);
>>>
>>> We know period is a power of two, so just do "& (period - 1)".
>>>
>>> That should let you get rid of the special case for
>>> "tb <= (1 << target_bit)" as well.
>>>
>>
>> Do you mean "lapse = period - ((tb - (1 << target_bit)) & (period - 1));" ?
> 
> Yes.
> 
> Or more simply:
> 
> lapse = period - ((tb - period) & (period - 1));
> 

Are you sure? Note that period != (1 << target_bit).

>> I don't see how this solves the "tb <= (1 << target_bit)" case.
> 
> Actually, since everything is unsigned the special case shouldn't be needed
> regardless.

You're right about this one, it's tricky though :)

-- 
Fabien Chouteau