Re: [PATCH] irqchip/jcore-aic: Fix missing allocation of IRQ descriptors

On Tue, 2023-05-16 at 11:06 +0100, Marc Zyngier wrote:
> > It did still work which is why the previously suggested change was to make a
> > failing call to irq_alloc_descs() non-fatal. The boards still booted
> > up.
> 
> I don't get it. Either the descriptors are already allocated, and you
> don't need this call, or they were never allocated and this never
> worked. Which one is it?

I haven't tried it myself yet. The original report is that the kernel prints
a lot of backtraces for each IRQ descriptor not allocated but still works:

> SH generic board support: scanning for interrupt controllers
> Initializing J-Core AIC
> ------------[ cut here ]------------
> error: virq16 is not allocated
> WARNING: CPU: 0 PID: 0 at kernel/irq/irqdomain.c:571
> irq_domain_associate+0x120/0x178
> 
> CPU: 0 PID: 0 Comm: swapper/0 Not tainted 5.19.0-rc2 #1
> PC is at irq_domain_associate+0x120/0x178
> PR is at irq_domain_associate+0x120/0x178
> PC  : 10049b90 SP  : 103bdec0 SR  : 400001f1
> R0  : 0000001e R1  : 1042d024 R2  : 1042d024 R3  : 00000028
> R4  : 00000001 R5  : 0006f1ff R6  : 00000008 R7  : 103bde04
> R8  : 1200c000 R9  : 00000010 R10 : 00000000 R11 : 00000010
> R12 : 10049a70 R13 : 103bfcac R14 : 1030a398
> MACH: 00000000 MACL: 00057fa8 GBR : 00000000 PR  : 10049b90
> 
> Call trace:
>  [<100496f0>] __irq_domain_add+0x80/0x1dc
>  [<10049cd2>] irq_domain_create_legacy+0x46/0x68
>  [<10049a70>] irq_domain_associate+0x0/0x178
>  [<104517da>] aic_irq_of_init+0x82/0xd8
>  [<1020ab90>] of_iomap+0x0/0x30
>  [<1031df1c>] _printk+0x0/0x24
>  [<1045630c>] of_irq_init+0xe4/0x228
>  [<100a5a10>] kfree+0x0/0x250
>  [<10042376>] vprintk_emit+0xde/0x1fc
>  [<1004239c>] vprintk_emit+0x104/0x1fc
>  [<10309940>] strlen+0x0/0x60
>  [<100424a6>] vprintk_default+0x12/0x20
>  [<10309940>] strlen+0x0/0x60
>  [<10002a2c>] arch_local_save_flags+0x0/0x8
>  [<1031df1c>] _printk+0x0/0x24
>  [<104456f8>] init_IRQ+0x14/0x28
>  [<10309940>] strlen+0x0/0x60
>  [<10002a2c>] arch_local_save_flags+0x0/0x8
>  [<1031df1c>] _printk+0x0/0x24
>  [<1044394c>] start_kernel+0x3b8/0x73c
>  [<1044320c>] unknown_bootoption+0x0/0x170
>  [<1000202a>] _stext+0x2a/0x34
> 
> Code:
>   10049b8a:  mov.l     10049bd8 <irq_domain_associate+0x168/0x178>, r4  !
> 10393da0 <0x10393da0>
>   10049b8c:  jsr       @r1
>   10049b8e:  mov       r11, r5
> ->10049b90:  trapa     #62
>   10049b92:  bra       10049b0e
>   10049b94:  mov       #-22, r12
>   10049b96:  mov.l     10049bd0 <irq_domain_associate+0x160/0x178>, r1  !
> 1031da2c <__warn_printk+0x0/0x38>
>   10049b98:  mov.l     10049bdc <irq_domain_associate+0x16c/0x178>, r4  !
> 10393dc0 <0x10393dc0>
>   10049b9a:  jsr       @r1
> 
> ---[ end trace 0000000000000000 ]---

Adding the call to irq_alloc_descs() addresses the problem.

> > > > > Is there any plan to modernise the port and get it to allocate
> > > > > irq_descs on demand, as we do on most architectures?
> > > > 
> > > > Yes, there are plans to modernize the port. We're first working on
> > > > upstreaming all kinds of patches that have been queuing up over the
> > > > time.
> > > 
> > > I'd rather you skip that step and focus on making it work as a modern
> > > architecture. This really looks like ARM circa 2007... :-/
> > 
> > We have a patch-set for switching it to device tree in the pipeline.
> 
> Again: why aren't we reviewing that instead of beating a long dead
> horse?

Because we're not there yet. I just picked up maintenance of the SuperH port and we
want to fix various bugs first before tackling the big changes to the architecture.

Thanks,
Adrian

-- 
 .''`.  John Paul Adrian Glaubitz
: :' :  Debian Developer
`. `'   Physicist
  `-    GPG: 62FF 8A75 84E0 2956 9546  0006 7426 3B37 F5B5 F913