[PATCH] irqchip/irq-imx-irqsteer: Get/put PM runtime in ->irq_unmask()/irq_mask()

[Liu Ying]

On Wed, 2022-06-08 at 14:54 +0100, Marc Zyngier wrote:
> On Wed, 08 Jun 2022 13:02:46 +0100,
> Lucas Stach <l.stach at pengutronix.de> wrote:
> > 
> > Am Mittwoch, dem 08.06.2022 um 19:29 +0800 schrieb Liu Ying:
> > > On Wed, 2022-06-08 at 12:56 +0200, Lucas Stach wrote:
> > > > Am Mittwoch, dem 08.06.2022 um 18:50 +0800 schrieb Liu Ying:
> > > > > Now that runtime PM support was added in this driver, we have
> > > > > to enable power before accessing irqchip registers.  And,
> > > > > after
> > > > > the access is done, we should disable power.  This patch
> > > > > calls
> > > > > pm_runtime_get_sync() in ->irq_unmask() and pm_runtime_put()
> > > > > in
> > > > > ->irq_mask() to make sure power is managed for the register
> > > > > access.
> > > > > 
> > > > 
> > > > Can you tell me in which case this is necessary? IIRC the IRQ
> > > > core
> > > 
> > > With the i.MX8qxp DPU driver[1], I see below synchronous external
> > > abort:
> > > 
> > > [    1.207270] Internal error: synchronous external abort:
> > > 96000210
> > > [#1] PREEMPT SMP
> > > [    1.207287] Modules linked in:
> > > [    1.207299] CPU: 1 PID: 64 Comm: kworker/u8:2 Not tainted
> > > 5.18.0-
> > > rc6-next-20220509-00053-gf01f74ee1c18 #272
> > > [    1.207311] Hardware name: Freescale i.MX8QXP MEK (DT)
> > > [    1.207319] Workqueue: events_unbound deferred_probe_work_func
> > > [    1.207339] pstate: 400000c5 (nZcv daIF -PAN -UAO -TCO -DIT
> > > -SSBS
> > > BTYPE=--)
> > > [    1.207349] pc : imx_irqsteer_irq_unmask+0x48/0x80
> > > [    1.207360] lr : imx_irqsteer_irq_unmask+0x38/0x80
> > > [    1.207368] sp : ffff80000a88b900
> > > [    1.207372] x29: ffff80000a88b900 x28: ffff8000080fed90 x27:
> > > ffff8000080fefe0
> > > [    1.207388] x26: ffff8000080fef40 x25: ffff0008012538d4 x24:
> > > ffff8000092fe388
> > > [    1.207407] x23: 0000000000000001 x22: ffff0008013295b4 x21:
> > > ffff000801329580
> > > [    1.207425] x20: ffff0008003faa60 x19: 000000000000000e x18:
> > > 0000000000000000
> > > [    1.207443] x17: 0000000000000003 x16: 0000000000000162 x15:
> > > 0000000000000001
> > > [    1.207459] x14: 0000000000000002 x13: 0000000000000018 x12:
> > > 0000000000000040
> > > [    1.207477] x11: ffff000800682480 x10: ffff000800682482 x9 :
> > > ffff80000a072678
> > > [    1.207495] x8 : ffff0008006a64a8 x7 : 0000000000000000 x6 :
> > > ffff0008006a6608
> > > [    1.207513] x5 : ffff800009070a18 x4 : 0000000000000000 x3 :
> > > ffff80000b240000
> > > [    1.207529] x2 : ffff80000b240038 x1 : 00000000000000c0 x0 :
> > > 00000000000000c0
> > > [    1.207549] Call trace:
> > > [    1.207553]  imx_irqsteer_irq_unmask+0x48/0x80
> > > [    1.207562]  irq_enable+0x40/0x8c
> > > [    1.207575]  __irq_startup+0x78/0xa4
> > > [    1.207588]  irq_startup+0x78/0x16c
> > > [    1.207601]  irq_activate_and_startup+0x38/0x70
> > > [    1.207612]  __irq_do_set_handler+0xcc/0x1e0
> > > [    1.207626]  irq_set_chained_handler_and_data+0x58/0xa0
> > 
> > Ooh, I think this is the problem. The IRQ is not requested in the
> > usual
> > way when a chained handler is added, so this might bypass the
> > runtime
> > PM handling normally done in the IRQ core. In that case this is a
> > core
> > issue and should not be worked around in the driver, but the core
> > should take the RPM reference for the chained handler, just like it
> > does for normal IRQs.
> 
> Well spotted. Could you please give the hack below (compile-tested
> only) a go?

I don't see the splat after your patch is applied.

Regards,
Liu Ying

> 
> Thanks,
> 
> 	M.
> 
> From 1426cadd87717f1d876c7563f2a29b00283a847e Mon Sep 17 00:00:00
> 2001
> From: Marc Zyngier <maz at kernel.org>
> Date: Wed, 8 Jun 2022 14:45:35 +0100
> Subject: [PATCH] genirq: PM: Use runtime PM for chained interrupts
> 
> When requesting an interrupt, we correctly call into the runtime
> PM framework to guarantee that the underlying interrupt controller
> is up and running.
> 
> However, we fail to do so for chained interrupt controllers, as
> the mux interrupt is not requested along the same path.
> 
> Augment __irq_do_set_handler() to call into the runtime PM code
> in this case, making sure the PM flow is the same for all interrupts.
> 
> Reported-by: Lucas Stach <l.stach at pengutronix.de>
> Signed-off-by: Marc Zyngier <maz at kernel.org>
> Link: 
> https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Flore.kernel.org%2Fr%2F26973cddee5f527ea17184c0f3fccb70bc8969a0.camel%40pengutronix.de&data=05%7C01%7Cvictor.liu%40nxp.com%7C2ce8f74775494461d51808da49566940%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C637902932736234542%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=jqe8nsRFWQpVjRl7C6Bgz%2BlwtkaMv%2BKq%2Fui5F1jwf4c%3D&reserved=0
> ---
>  kernel/irq/chip.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
> index e6b8e564b37f..886789dcee43 100644
> --- a/kernel/irq/chip.c
> +++ b/kernel/irq/chip.c
> @@ -1006,8 +1006,10 @@ __irq_do_set_handler(struct irq_desc *desc,
> irq_flow_handler_t handle,
>  		if (desc->irq_data.chip != &no_irq_chip)
>  			mask_ack_irq(desc);
>  		irq_state_set_disabled(desc);
> -		if (is_chained)
> +		if (is_chained) {
>  			desc->action = NULL;
> +			WARN_ON(irq_chip_pm_put(irq_desc_get_irq_data(d
> esc)));
> +		}
>  		desc->depth = 1;
>  	}
>  	desc->handle_irq = handle;
> @@ -1033,6 +1035,7 @@ __irq_do_set_handler(struct irq_desc *desc,
> irq_flow_handler_t handle,
>  		irq_settings_set_norequest(desc);
>  		irq_settings_set_nothread(desc);
>  		desc->action = &chained_action;
> +		WARN_ON(irq_chip_pm_get(irq_desc_get_irq_data(desc)));
>  		irq_activate_and_startup(desc, IRQ_RESEND);
>  	}
>  }
> -- 
> 2.34.1
> 
>