Message-ID: <20160218091432.6968b836@arm.com>
Date:	Thu, 18 Feb 2016 09:14:32 +0000
From:	Marc Zyngier <marc.zyngier@....com>
To:	Stefan Agner <stefan@...er.ch>
Cc:	jiang.liu@...ux.intel.com,
	Sanchayan Maity <maitysanchayan@...il.com>, robh@...nel.org,
	linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org
Subject: Re: no irq domain found for ... a GPIO interrupt controller

On Wed, 17 Feb 2016 15:29:22 -0800
Stefan Agner <stefan@...er.ch> wrote:

Hi Stefan,

> Hi Marc,
> 
> I get the following warning on our Vybrid based Colibri VF50 module:
> [    0.147674] irq: no irq domain found for
> /soc/aips-bus@...00000/gpio@...4a000 !
> 
> The device tree arch/arm/boot/dts/vf500-colibri.dtsi uses the following
> interrupt specification in a touchscreen node which seems to trigger the
> issue:
> interrupt-parent = <&gpio0>;
> interrupts = <8 IRQ_TYPE_LEVEL_LOW>;
> 
> We use other GPIO interrupts, but I guess it is a matter of
> initialization order, and the above specification seems to be parsed
> before the GPIO interrupt-controller gets created...?

That's very likely. GPIO interrupt controllers tend to be available
quite late in the boot sequence.

> 
> It seems that Rob Herring addressed a similar issue a while ago in 
> 9ec36ca ("of/irq: do irq resolution in platform_get_irq")
> 
> Is this problem known?
> 
> Currently, the device does not get any interrupts. The interrupt seems
> to be requested successfully at probe time later, so I am not sure if
> the not working device is really related to the message above....

That is what Rob's patch fixes: it ensures that the interrupt specifier
is resolved (turned into a Linux IRQ) when you call platform_get_irq.

> 
> This is a stack trace when I add a WARN_ON(1) in
> irq_create_fwspec_mapping:
> [    0.147674] irq: no irq domain found for
> /soc/aips-bus@...00000/gpio@...4a000 !
> [    0.147823] ------------[ cut here ]------------
> [    0.147941] WARNING: CPU: 0 PID: 1 at kernel/irq/irqdomain.c:584
> irq_create_fwspec_mapping+0xf0/0x1e4()
> [    0.148056] Modules linked in:
> [    0.148142] CPU: 0 PID: 1 Comm: swapper Not tainted
> 4.4.0-00070-g3690255-dirty #764
> [    0.148246] Hardware name: Freescale Vybrid VF5xx/VF6xx (Device Tree)
> [    0.148311] Backtrace:
> [    0.148416] [<80013498>] (dump_backtrace) from [<80013690>]
> (show_stack+0x18/0x1c)
> [    0.148521]  r7:800539f4 r6:00000248 r5:00000009 r4:00000000
> [    0.148646] [<80013678>] (show_stack) from [<8029d070>]
> (dump_stack+0x24/0x28)
> [    0.148782] [<8029d04c>] (dump_stack) from [<800219a4>]
> (warn_slowpath_common+0x88/0xb4)
> [    0.148916] [<8002191c>] (warn_slowpath_common) from [<80021a74>]
> (warn_slowpath_null+0x24/0x2c)
> [    0.149021]  r8:00000000 r7:87cf0f88 r6:87cf0f88 r5:00000000
> r4:87843ca8
> [    0.149158] [<80021a50>] (warn_slowpath_null) from [<800539f4>]
> (irq_create_fwspec_mapping+0xf0/0x1e4)
> [    0.149294] [<80053904>] (irq_create_fwspec_mapping) from
> [<80053b44>] (irq_create_of_mapping+0x5c/0x64)
> [    0.149400]  r6:87cf0f88 r5:878bb7c0 r4:00000000
> [    0.149508] [<80053ae8>] (irq_create_of_mapping) from [<8045e934>]
> (irq_of_parse_and_map+0x2c/0x34)
> [    0.149633] [<8045e908>] (irq_of_parse_and_map) from [<8045e95c>]
> (of_irq_to_resource+0x20/0xc4)
> [    0.149755] [<8045e93c>] (of_irq_to_resource) from [<8045ea44>]
> (of_irq_to_resource_table+0x44/0x54)
> [    0.149859]  r8:00000001 r7:00000001 r6:87cf0f88 r5:878bb7dc
> r4:00000000
> [    0.149990] [<8045ea00>] (of_irq_to_resource_table) from [<8045c5b0>]
> (of_device_alloc+0x114/0x184)
> [    0.150097]  r7:00000000 r6:878be800 r5:87cf0f88 r4:00000000
> [    0.150339] [<8045c49c>] (of_device_alloc) from [<8045c678>]
> (of_platform_device_create_pdata+0x58/0xd4)
> [    0.150457]  r10:00000000 r9:00000000 r8:80650628 r7:00000000
> r6:00000000 r5:87cf0f88
> [    0.150600]  r4:00000000
> [    0.150688] [<8045c620>] (of_platform_device_create_pdata) from
> [<8045c800>] (of_platform_bus_create+0xf0/0x194)
> [    0.150796]  r7:00000001 r6:00000000 r5:87cf0f88 r4:00000000
> [    0.150912] [<8045c710>] (of_platform_bus_create) from [<8045c9e0>]
> (of_platform_populate+0x64/0xc0)
> [    0.151016]  r10:00000000 r9:00000001 r8:00000000 r7:00000000
> r6:80650628 r5:87ce34f4
> [    0.151152]  r4:87cf0f88
> [    0.151244] [<8045c97c>] (of_platform_populate) from [<80806fa8>]
> (vf610_init_machine+0x24/0x2c)
> [    0.151352]  r9:807ff5ec r8:000000b0 r7:87896600 r6:80840420
> r5:80801a48 r4:80840420
> [    0.151515] [<80806f84>] (vf610_init_machine) from [<80801a70>]
> (customize_machine+0x28/0x48)
> [    0.151640] [<80801a48>] (customize_machine) from [<800095fc>]
> (do_one_initcall+0x98/0x1e0)
> [    0.151763] [<80009564>] (do_one_initcall) from [<807ffe34>]
> (kernel_init_freeable+0x13c/0x1e0)
> [    0.151867]  r10:8082b824 r9:807ff5ec r8:000000b0 r7:8086e440
> r6:8086e440 r5:00000003
> [    0.152004]  r4:80838704
> [    0.152091] [<807ffcf8>] (kernel_init_freeable) from [<805ee93c>]
> (kernel_init+0x18/0xf0)
> [    0.152197]  r10:00000000 r9:00000000 r8:00000000 r7:00000000
> r6:00000000 r5:805ee924
> [    0.152333]  r4:8086e440
> [    0.152424] [<805ee924>] (kernel_init) from [<8000f878>]
> (ret_from_fork+0x14/0x3c)
> [    0.152526]  r5:805ee924 r4:00000000
> [    0.152670] ---[ end trace c2d84d5af0139987 ]---

What your stack trace shows is that the failure occurs at boot time,
when of_platform_populate() parses the DT and creates the platform
devices. As part of this process, it tries to resolves the interrupt
specifiers, but fails to do so for this particular device, since the
corresponding irqchip and domain have not been created yet (the GPIO
controller is itself a device, while other irqchips are not).

But as long as your device driver uses platform_get_irq(), you will
force the interrupt specifier to be evaluated again, this time giving
you a valid interrupt (and provided that your GPIO driver has
initialized in the meantime - check for -EPROBE_DEFER). At some point,
we'll be able to kill the interrupt resolution from
of_platform_populate().

To summarize, your driver not working may not be related to this issue.

Hope this helps,

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

Powered by blists - more mailing lists