On 11/02/06, Jan Kiszka <[EMAIL PROTECTED]> wrote:
Dmitry Adamushko wrote:
> And as an additional option,
> it could be interesting to print out to the log if not all "counter" values
> then min,max,average (the same like for the latency :) per second or per
> 1000 interrupts; so to see whether tweaking the MAX_EDGEIRQ_COUNTER may make
> the things better.

Yes, maybe it's too small. But this also depends on the absolute time
required for so many loops, something we should see in the traces then.
I'm afraid that we will finally have to move the UART's read-from-fifo
to task context to reduce the time spent in IRQ context.

Good. Keep me informed as before.

> Maybe we'd better go for 1, just in case somebody would like to know
> the actual number of virq occuried. Although, I can imagine that only
> for debugging reasons and of no avail for the normal users, so maybe 2 :)

I have no final opinion yet. Having some names also behind the virtual
IRQs would be really nice for debugging, but it would also break another
bulk of API calls. Just marking those IRQs virtual would render the
related output of /proc/xenomai/irq not very useful in my eye - as long
as you do not know what's behind it. Just dropping this information is
then likely better.

Yep. Let's just avoid printing it.

>> Then I stumbled over the xnintr structure. Why do you keep a copy of the
>> device name?
>> A "const char *" should be enough, we just have to demand
>> that it will remain valid as long as the xnintr structure itself (i.e.
>> during the IRQ being attached). Saves a few bytes. :)
> Could be if all the users of it are initially kernel-mode entities.
> But e.g. there is a version of rt_intr_create() that may be called by
> user-space
> apps and I extended it to support the "name" parameter too :
> int rt_intr_create(RT_INTR *intr, const char *name, unsigned irq, int mode);
> In such a case, the kernel-side rt_intr_create() is called with the "name"
> allocated on the stack in skins/native/syscall.c : __rt_intr_create())
> so the initial copy of the name can not remain valid.

But you could create buffer space for a copy of the name in the same
block already allocated for RT_INTR. Should be straightforward.

RT_INTR does not have the "name" field any more. I have removed it after adding the same field to the xnintr_t as it became redundant.

To access the "name", any code in native skin does "intr->intr_base.name" (intr_base is of "xnintr_t" type).


Best regards,
Dmitry Adamushko
Xenomai-core mailing list

Reply via email to