Re: Re: [PATCH 3/6] mfd: palmas: Fix potential IRQ action leaks on removal
From: Lee Jones
Date: Thu Nov 19 2026 - 09:25:55 EST
On Thu, 19 Nov 2026, 李广硕 wrote:
>
> Hi Lee,
>
> Thanks for the review.
>
> From: Lee Jones <lee@xxxxxxxxxx>
> Date: 2026-11-13 21:09:18
> To: Guangshuo Li <liguangshuo@xxxxxxxxxxxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx,Aaro Koskinen <aaro.koskinen@xxxxxx>,Andreas Kemnade <andreas@xxxxxxxxxxxx>,Kevin Hilman <khilman@xxxxxxxxxxxx>,Roger Quadros <rogerq@xxxxxxxxxx>,Tony Lindgren <tony@xxxxxxxxxxx>,linux-omap@xxxxxxxxxxxxxxx,mfd@xxxxxxxxxxxxxxx,linux-kernel@xxxxxxxxxxxxxxx
> Subject: Re: [PATCH 3/6] mfd: palmas: Fix potential IRQ action leaks on removal>On Sat, 31 Oct 2026, Guangshuo Li wrote:
Please reconfigure your mailer. These headers shouldn't be in the body.
> >> The remove callback deletes the IRQ chip before managed OF children
> >> are removed. Their later free_irq() calls cannot find the disposed IRQ
> >> descriptors, potentially leaking interrupt actions. Depopulate the
> >> children before deleting the IRQ chip and auxiliary I2C clients.
> >>
> >> This issue was found by our static analysis tool under development.
> >>
> >> Fixes: 124e9deb1c40 ("mfd: palmas: Use devm_of_platform_populate()")
> >> Cc: stable@xxxxxxxxxxxxxxx
> >> Assisted-by: OpenAI:gpt-6-astra
> >> Signed-off-by: Guangshuo Li <liguangshuo@xxxxxxxxxxxxxxx>
> >> ---
> >> Build-tested the changed object with x86_64 defconfig plus COMPILE_TEST
> >> and the driver dependencies (W=1).
> >> No runtime or fault-injection testing was performed.
> >>
> >> drivers/mfd/palmas.c | 3 +++
> >> 1 file changed, 3 insertions(+)
> >>
> >> diff --git a/drivers/mfd/palmas.c b/drivers/mfd/palmas.c
> >> index 7fc886f4f80e..8a0b7b1e89d7 100644
> >> --- a/drivers/mfd/palmas.c
> >> +++ b/drivers/mfd/palmas.c
> >> @@ -677,6 +677,9 @@ static void palmas_i2c_remove(struct i2c_client *i2c)
> >> struct palmas *palmas = i2c_get_clientdata(i2c);
> >> int i;
> >>
> >> + if (i2c->dev.of_node)
> >> + devm_of_platform_depopulate(&i2c->dev);
> >
> >I think it would be nice to devm_* regmap_add_irq_chip() instead.
> >
> >If you also convert the auxiliary I2C clients and 'pm_power_off'
> >teardown to devm, you could drop palmas_i2c_remove() altogether.
> >
> >> +
> >> regmap_del_irq_chip(palmas->irq, palmas->irq_data);
> >>
> >> for (i = 1; i < PALMAS_NUM_CLIENTS; i++) {
> >
> >--
> >Lee Jones
> >
>
> The IRQ chip is registered before devm_of_platform_populate(). With
> devm_regmap_add_irq_chip(), the children would be removed while their
> IRQ mappings are still valid.
>
> If the auxiliary I2C clients and the power-off handler use managed
> cleanup too, I think palmas_i2c_remove() can go away. Is that what you
> had in mind? If so, I'll send a v2 for palmas along those lines.
Yes, that sounds about right.
--
Lee Jones