Re: [Qemu-devel] [kvm-unit-tests PATCH v5 10/11] arm/arm64: gicv3: add an IPI test

[Andrew Jones]

On Fri, Nov 11, 2016 at 10:02:59AM +0000, Alex Bennée wrote:
> 
> Andrew Jones <address@hidden> writes:
> 
> > On Thu, Nov 10, 2016 at 07:53:58PM +0000, Alex Bennée wrote:
> > [...]
> >> > +struct gic gicv2 = {
> >> > +        .ipi = {
> >> > +                .enable = gicv2_enable_defaults,
> >> > +                .send_self = gicv2_ipi_send_self,
> >> > +                .send_tlist = gicv2_ipi_send_tlist,
> >> > +                .send_broadcast = gicv2_ipi_send_broadcast,
> >> > +        },
> >> > +        .read_iar = gicv2_read_iar,
> >> > +        .irqnr = gicv2_irqnr,
> >> > +        .write_eoi = gicv2_write_eoi,
> >> > +};
> >> > +
> >> > +struct gic gicv3 = {
> >> > +        .ipi = {
> >> > +                .enable = gicv3_enable_defaults,
> >> > +                .send_self = gicv3_ipi_send_self,
> >> > +                .send_tlist = gicv3_ipi_send_tlist,
> >> > +                .send_broadcast = gicv3_ipi_send_broadcast,
> >> > +        },
> >> > +        .read_iar = gicv3_read_iar,
> >> > +        .irqnr = gicv3_irqnr,
> >> > +        .write_eoi = gicv3_write_eoir,
> >> > +};
> >> > +
> >>
> >> So I was re-basing my kvm-unit-tests against your GIC rework and found
> >> myself copy and pasting a bunch of this into my tests that fire IRQs.
> >> That makes me think the abstraction should be in the library code so
> >> other tests can fiddle with sending IRQs.
> >>
> >> What do you think?
> >>
> >
> > I guess you mean moving the above two structs and their corresponding
> > functions (all which aren't already common) to lib/arm/ ? Or do you
> > just mean the one non-trivial function gicv3_ipi_send_tlist? I think
> > agree with gicv3_ipi_send_tlist getting shared, but the others are
> > mostly one-liners, so I'm not sure. I guess I'd have to see how you're
> > using them first.
> 
> So it looked like there were some functions in the common code for one
> GIC which had local test defined functions for the other. They should at
> least be consistent.

gicv3_read_iar and gicv3_write_eoir being common already is a product of
being sysreg wrappers, allowing for both arm32 and arm64 to use functions
of the same names, not because I wanted gicv3 to be inconsistent with
gicv2 (which uses MMIO and thus doesn't need wrappers)

> 
> For my use case I could do with a common:
> 
>   gic_enable

OK, I can extend gic_init() to initialize a 'struct gic_common_ops' that
includes an enable -> *_enable_defaults(void), ipi_send(int cpu),
read_iar(void), iar_irqnr(u32 iar), and write_eoi(u32 irqstat). And also
provide the wrappers gic_enable, gic_ipi_send(cpu), ...

>   gic_send_spi(cpu, irq)

I'll let you add this one to the new common ops struct :-)

>   gic_irq_ack() which returns the iar.

This one will be called read_iar.

Would that work for you, Alex?

Andre,

Would this also satisfy your needs for more common code?

Thanks,
drew

> 
> See:
> 
>   
> https://github.com/stsquad/kvm-unit-tests/blob/mttcg/current-tests-v6/arm/tcg-test.c#L113
> 
> >
> > Thanks,
> > drew
> 
> 
> --
> Alex Bennée
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to address@hidden
> More majordomo info at  http://vger.kernel.org/majordomo-info.html