Re: [PATCH 07/10] iio: adc: qcom-spmi-adc5-gen3: Free TM auxiliary device wrapper in release callback

From: Jonathan Cameron

Date: Sat Nov 21 2026 - 18:08:42 EST


On Thu, 19 Nov 2026 19:30:35 +0200
Andy Shevchenko <andriy.shevchenko@xxxxxxxxx> wrote:

> On Thu, Nov 19, 2026 at 03:30:40PM +0000, Lee Jones wrote:
> > struct tm5_aux_dev_wrapper embeds a reference-counted struct
> > auxiliary_device, whose memory must be released from its dev.release
> > callback once the final device reference is dropped.
> >
> > Allocating the wrapper with devm_kzalloc_obj() on the parent ADC device
> > and providing an empty adc5_gen3_aux_device_release() causes parent
> > devres teardown to free the wrapper immediately after
> > auxiliary_device_uninit(), even if external references to the embedded
> > struct device are still held.
> >
> > Allocate the wrapper with kzalloc_obj(), release it from
> > adc5_gen3_aux_device_release(), and unwind cleanly on early error paths
> > in adc5_gen3_add_aux_tm_device().
>
> ...
>
> > #include <linux/platform_device.h>
> > #include <linux/property.h>
> > #include <linux/regmap.h>
> > +#include <linux/slab.h>
> > #include <linux/types.h>
> > #include <linux/unaligned.h>
>
> > +static void adc5_gen3_aux_device_release(struct device *dev)
> > +{
> > + struct tm5_aux_dev_wrapper *aux_device =
> > + container_of(dev, struct tm5_aux_dev_wrapper, aux_dev.dev);
>
> container_of() is also in a separate header, id est container_of.h.
>
> > + kfree(aux_device);
> > +}
>
> ...
>
> > ret = auxiliary_device_add(&aux_device->aux_dev);
> > - if (ret)
> > + if (ret) {
> > + devm_release_action(adc->dev, adc5_gen3_uninit_aux,
> > + &aux_device->aux_dev);
>
> Oh, this is interesting... Usually the release devm_*() calls point
> to a misdesign somewhere. But I haven't checked the twisted flow of
> this code.

It's one of those complex / annoying cases where a driver uses devm but
carries on with a subset of features if a particular bit doesn't work.

Right answer here is probably a devres group.

That is independent of the handoff dance done here.

I'd move the auxiliary_device_init() earlier, stop messing
with the memory being devm controlled at all except by adding
a put_device() in a dev_add_action_or_reset().
Actually that's already done by the auxiliary_device_uniit.

Anyhow, whilst I haven't looked at this that closely looks
like right clean answer is a dev_res group for the whole
function (can be done at the caller if neater). That allows
either handling the group over to normal devres cleanup on
success, or cleaning it up on erorr.

That plus the move of the kfree into release as Lee has done.
Also smells like tm_props lifetime is probably wrong as well
and should be part of the release rather than devm_kzalloc.

Jonathan

>
> > return ret;
> > + }
> > +
> > ret = devm_add_action_or_reset(adc->dev, adc5_gen3_delete_aux,
> > &aux_device->aux_dev);
> > - if (ret)
> > + if (ret) {
> > + devm_release_action(adc->dev, adc5_gen3_uninit_aux,
> > + &aux_device->aux_dev);
> > return ret;
> > + }
>