On Wed, May 15, 2013 at 04:30:26PM -0500, Scott Wood wrote: > On 05/13/2013 09:03:17 PM, Kevin Hao wrote: > >On Mon, May 13, 2013 at 10:47:17AM -0500, Scott Wood wrote: > >> On 05/11/2013 06:26:21 PM, Kevin Hao wrote: > >> >In the external proxy facility mode, the interrupt is automatically > >> >acknowledged with the same effect as reading the IACK register. So > >> >this makes external input interrupt more like edge sensitive. That > >> >means we can leave the irq hard enabled when it occurs with irq > >soft > >> >disabled just like the dec and doorbell interrupt. But the External > >> >Proxy Register(EPR) is only considered valid from the time that the > >> >external interrupt occurs until MSR[EE] is set to 1. So we have to > >> >save the EPR before irq hard enabled. > >> > >> Is it really worth it? > > > >Maybe. :-) > >Compare with the current kernel: > > * The overhead is that we need additional load & store the > >contents of > > the EPR from/to PACA. > > There's also mental overhead of the extra complexity.
Yes, I agree. But since we already have the support for the edge sensitive interrupt such as doorbell, decrementer, adding another one doesn't really introduce much code complexity in my opinion. > The lazy EE > stuff is already fiddly enough (e.g. the recent KVM patches). :-) > > > * The bonus is we keep the irq hard enabled when a external > >interrupt occurs > > with irq soft-disabled. As I know we should leave the irq hard > >enabled as > > much as possible. This is also the primary reason that we > >introduce the > > Lazy EE. > > I don't think "as much as possible" is a good way to look at it, so > much as "as much as is practical", balanced by also wanting to keep > the code as simple as is practical. Yes, I also like simple. That is why I make the following patch first. http://patchwork.ozlabs.org/patch/235530/ But it seems that Ben doesn't like it. And it also seem not so difficulty to support the external interrupt as edge sensitive for external proxy, so I scratch these patches. It seems that you and Ben have different view about this issue. Anyway I have no strong preference for these two ways and will leave it to you guys to determine which way we like to adopt. Thanks, Kevin > > -Scott
pgpCNerI0DCNe.pgp
Description: PGP signature
_______________________________________________ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
