Re: [PATCH] pwm: Don't dereference the chip after put_device() in __pwm_put()

From: Wong Steve

Date: Thu Nov 19 2026 - 14:27:47 EST


Hi Uwe,

On Thu, Nov 19, 2026, Uwe Kleine-König wrote:
> I wonder what ensures that the compiler doesn't reorder put_device() and
> the assignment to owner. Maybe it would be more sensible to call
> module_put() before put_device()?

put_device() is an out-of-line call that may free *chip, so the compiler
has to do the load of chip->owner before the call. But your way is
simpler and doesn't need that argument: pwmchip_release() lives in the
PWM core, so dropping the module reference first is fine. I'll send a
v2 that does that.

Thanks,
Suyuan

Uwe Kleine-König <ukleinek@xxxxxxxxxx> 于2026年11月19日周四 22:27写道:
>
> On Thu, Nov 19, 2026 at 10:56:17AM +0000, Suyuan Wang via B4 Relay wrote:
> > From: Suyuan Wang <stevewong@xxxxxxxxxxx>
> >
> > __pwm_put() drops the chip reference the consumer took and then reads the
> > chip to drop the module reference:
> >
> > put_device(&chip->dev);
> > module_put(chip->owner);
> >
> > A consumer takes both references in pwm_device_request()
> > (try_module_get(chip->owner) and get_device(&chip->dev)). The chip's base
> > reference is released by pwmchip_put(), not by pwmchip_remove(), so a
> > consumer can still hold a pwm after the provider has removed and put the
> > chip. When that consumer finally calls pwm_put(), the put_device() here
> > drops the last reference, pwmchip_release() frees the chip, and
> > module_put(chip->owner) then reads the freed chip.
> >
> > Read chip->owner into a local before put_device() and use that.
> >
> > Found with a static analysis tool and checked by hand.
> >
> > Fixes: 4c56b1434b81 ("pwm: Add a struct device to struct pwm_chip")
> > Cc: stable@xxxxxxxxxxxxxxx
> > Co-developed-by: Kaixuan Li <kaixuan.li@xxxxxxxxxx>
> > Signed-off-by: Kaixuan Li <kaixuan.li@xxxxxxxxxx>
> > Signed-off-by: Suyuan Wang <stevewong@xxxxxxxxxxx>
> > ---
> > Compile tested on x86_64 (W=1 and sparse, no new warnings). I don't have
> > the hardware.
> > ---
> > drivers/pwm/core.c | 3 ++-
> > 1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/pwm/core.c b/drivers/pwm/core.c
> > index 4c6d716e5fde..8988a7dbe5b1 100644
> > --- a/drivers/pwm/core.c
> > +++ b/drivers/pwm/core.c
> > @@ -2020,6 +2020,7 @@ EXPORT_SYMBOL_GPL(pwm_get);
> > static void __pwm_put(struct pwm_device *pwm)
> > {
> > struct pwm_chip *chip = pwm->chip;
> > + struct module *owner = chip->owner;
> >
> > /*
> > * Trigger a warning if a consumer called pwm_put() twice.
> > @@ -2038,7 +2039,7 @@ static void __pwm_put(struct pwm_device *pwm)
> >
> > put_device(&chip->dev);
> >
> > - module_put(chip->owner);
> > + module_put(owner);
> > }
>
> I wonder what ensures that the compiler doesn't reorder put_device() and
> the assignment to owner. Maybe it would be more sensible to call
> module_put() before put_device()?
>
> Best regards
> Uwe