Re: [PATCH v8 06/11] irqchip: mips-cpu: Convert to simple domain

[Marc Zyngier]

On 2020-03-25 12:37, Thomas Bogendoerfer wrote:
> On Wed, Mar 25, 2020 at 11:54:59AM +0800, Jiaxun Yang wrote:
> > The old code is using legacy domain to setup irq_domain for CPU 
> > interrupts
> > which requires irq_desc to be preallocated.
> >
> > However, when MIPS_CPU_IRQ_BASE >= 16, irq_desc for CPU IRQs may end 
> > up
> > unallocated and lead to incorrect behavior.
> >
> > Thus we convert the legacy domain to simple domain which can allocate
> > irq_desc during initialization.
> >
> > Signed-off-by: Jiaxun Yang <jiaxun.yang@xxxxxxxxxxx>
> > Co-developed-by: Huacai Chen <chenhc@xxxxxxxxxx>
> > Signed-off-by: Huacai Chen <chenhc@xxxxxxxxxx>
> > Reviewed-by: Marc Zyngier <maz@xxxxxxxxxx>
> > ---
> >  drivers/irqchip/irq-mips-cpu.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/irqchip/irq-mips-cpu.c 
> > b/drivers/irqchip/irq-mips-cpu.c
> > index 95d4fd8f7a96..c3cf7fa76424 100644
> > --- a/drivers/irqchip/irq-mips-cpu.c
> > +++ b/drivers/irqchip/irq-mips-cpu.c
> > @@ -251,7 +251,7 @@ static void __init __mips_cpu_irq_init(struct 
> > device_node *of_node)
> >  	clear_c0_status(ST0_IM);
> >  	clear_c0_cause(CAUSEF_IP);
> >
> > -	irq_domain = irq_domain_add_legacy(of_node, 8, MIPS_CPU_IRQ_BASE, 0,
> > +	irq_domain = irq_domain_add_simple(of_node, 8, MIPS_CPU_IRQ_BASE,
> >  					   &mips_cpu_intc_irq_domain_ops,
> >  					   NULL);
>
> this breaks at least IP30 and guess it will break every platform where
> MIPS_CPU_IRQ_BASE == 0. add_legacy will always do 
> irq_domain_associate_many(),
> while add_simple doesn't do it, if first_irq == 0.
>
> Marc, what is the reason not doing it all the time ? What's the correct
> way here to work with irq_domain_add_simple() in this case ?

On a fully DT-ified platform, using non-legacy irqdomains, virtual 
interrupts
are allocated as a "random" number, depending on the order of 
allocation,
and on demand.

The first_irq hack in irq_domain_add_simple() is just a way to still 
allocate
descriptors upfront (and I wish we could drop it...).

If you have legacy code that "knows" about the relationship between 
Linux's
virtual interrupt and the hwirq (that is only meaningful to the 
interrupt
controller), you're screwed, and need to stick to the legacy irqdomain.

It feels like the MIPS code is squarely in the latter case, so I guess 
this
patch is probably the wrong thing to do for this architecture.

Thanks,

        M.
--
Jazz is not dead. It just smells funny...