On Mon, Oct 19, 2009 at 03:29, Hennerich, Michael wrote:
>From: Mike Frysinger
>>On Fri, Oct 16, 2009 at 05:02, Zhang, Sonic wrote:
>>> Write to XMT_DATA8 register should be done after INT_STAT is cleared.
>>> So, as least one sync before XMIT_DATA8 is necessary.
>>> I will do a bit clean up.
>>>
>>> Yes, a SSYNC_ATOMIC() macro is better for interrupt routine usage.
>>
>>Michael pointed out something not so obvious (at least to me) ...
>>higher priority interrupts can still interrupt things, so
>>SSYNC_ATOMIC() would only be usable if the irq handler were registered
>>w/IRQF_DISABLED.  and people really hate when you use that flag.
>
> Besides that - you should also keep in mind that Linux uses DELAYED
> DISABLE.
> This means if you disable an IRQ - it's not disabled until it actually
> triggered.
> They use this concept, because interrupt enable/disable can take quite
> long on some architecture.

thanks, i wasnt aware of that.  guess the only place this has a use is
exception/early context.  and we already use the builtin there.
-mike
_______________________________________________
Linux-kernel-commits mailing list
[email protected]
https://blackfin.uclinux.org/mailman/listinfo/linux-kernel-commits

Reply via email to