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

[Thomas Bogendoerfer]

On Wed, Mar 25, 2020 at 11:09:10PM +0800, Jiaxun Yang wrote:
> 
> 
> ä 2020å3æ25æ GMT+08:00 äå11:04:37, Thomas Bogendoerfer <tsbogend@xxxxxxxxxxxxxxxx> åå:
> >On Wed, Mar 25, 2020 at 10:31:21PM +0800, Jiaxun Yang wrote:
> >> 
> >> 
> >> ä 2020å3æ25æ GMT+08:00 äå10:15:16, Marc Zyngier <maz@xxxxxxxxxx> åå:
> >> >On 2020-03-25 13:59, Jiaxun Yang wrote:
> >> >
> >> >[...]
> >> >
> >> >>>> So probably we can use legacy domain when  MIPS IRQ BASE is in
> >the
> >> >>>> range of legacy IRQ
> >> >>>> and switch to simple domain when it's not in that range?
> >> >>> 
> >> >>> No, see below.
> >> >>> 
> >> >>>> Here in Loongson systems IRQ 0-15 is occupied by I8259 so I did
> >> >this
> >> >>>> hack.
> >> >>> 
> >> >>> Well, if you have to consider which Linux IRQ gets assigned,
> >> >>> then your platform is definitely not ready for non-legacy
> >> >>> irqdomains. Just stick to legacy for now until you have removed
> >> >>> all the code that knows the hwirq mapping.
> >> >> 
> >> >> Thanks.
> >> >> 
> >> >> So I have to allocate irq_desc here in driver manually?
> >> >
> >> >No, you are probably better off just dropping this patch, as MIPS
> >> >doesn't seem to be ready for a wholesale switch to virtual
> >interrupts.
> >> 
> >> It can't work without this patch.
> >> 
> >> Legacy domain require IRQ number within 0-15 
> >> however it's already occupied by i8259 or "HTPIC" driver.
> >
> >what's the problem here ? AFAIK there could be more than one
> >legacy domain, at least that's what at least IP22/SNI in MIPS world 
> >are doing.
> 
> MIPS_IRQ_BASE must be higher than 15, otherwise it will conflict with i8259.

I still don't get it.

We have following in arch/mips/include/asm/mach-generic/irq.h:

#ifndef MIPS_CPU_IRQ_BASE
#ifdef CONFIG_I8259
#define MIPS_CPU_IRQ_BASE 16
#else
#define MIPS_CPU_IRQ_BASE 0
#endif /* CONFIG_I8259 */
#endif

So every legacy platform with i8259 has MIPS_CPU_IRQ_BASE = 16.

> However we have only preallocated irq_desc for 0-15.
> And legacy domain require irq_desc being preallocated.

maybe I'm too fast by judging the irq code, but without CONFIG_SPARSE_IRQ
the whole irq_desc is pre-allocated.

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]