Re: [PATCH 2.6.21-rt2] PowerPC: decrementer clockevent driver

[Sergei Shtylyov <sshtylyov@ru.mvista.com>]

Segher Boessenkool wrote:
>>>> +     * We must write a positive value to the decrementer to clear
>>>> +     * the interrupt on the IBM 970 CPU series.  In periodic mode,
>>>> +     * this happens when the decrementer gets reloaded later, but
>>>> +     * in one-shot mode, we have to do it here since an event handler
>>>> +     * may skip loading the new value...
>>
>>
>>> Nothing special about 970 here -- on *every* PowerPC,
>>> a decrementer exception exists as long as the high
>>> bit of the decrementer equals 1.  BookE is different
>>> of course.  Some other CPUs might deviate from the
>>> architecture as well.
>>
>>
>>    Quoting "PowerPC Operating Environment Architecture":

> <snip>

>> Otherwise, when the contents of DEC0 change, the exception effects of 
>> the Decrementer become consistent with the new contents of the 
>> Decrementer reasonably soon after the change.

> And that is guaranteed on all PowerPC as far as I can see.
> The main thing is that a decrementer exception won't go
> away until the high bit becomes 0.

On both POWER4 and POWER4+, the Decrementer must be implemented such that 
requirements 1 to 3 below are satisfied. On POWER4, requirements 4 and 5 must 
also be satisfied.

1. The operation of the Time Base and the Decrementer is coherent, i.e., the 
counters are driven by the same fundamental time base.
2. Loading a GPR from the Decrementer shall have no effect on the accuracy of 
the Decrementer.

3. Storing a GPR to the Decrementer shall replace the value in the Decrementer 
with the value in the GPR.

4. Whenever bit 0 of the Decrementer changes from 0 to 1, an interrupt request 
is signaled. If multiple Decrementer interrupt requests are received before 
the first can be reported, only one interrupt is reported. The occurrence of a 
Decrementer interrupt cancels the request.

5. If the Decrementer is altered by software and the contents of bit 0 are 
changed from 0 to 1, an interrupt request is signaled.

    (4) clearly contradicts your point. I don't mind changing #ifdef though 
(so it'll cover all non Book E cases)

> Segher

WBR, Sergei