Re: [PATCH v3] iio: trigger: use put_device() in viio_trigger_alloc() error path

[Nuno Sá]

On Tue, 2026-02-17 at 17:08 +0200, Andy Shevchenko wrote:
> On Tue, Feb 17, 2026 at 01:51:08PM +0000, Nuno Sá wrote:
> > On Mon, 2026-02-16 at 09:45 +0100, Salah Triki wrote:
> > > 
> > > You are absolutely right. My previous version (v3) was logically flawed as
> > > it could trigger the release callback before the necessary fields were
> > > initialized, leading to an unsafe irq_free_descs() call.
> > > 
> > > Since I don't have the physical hardware to perform runtime injection
> > > tests,I relied on manual code path analysis and clearly failed to account
> > > for the side effects of put_device().
> > > 
> > > I'm sending a v4 which takes the safer approach: moving
> > > device_initialize() after all potential failure points. This way, we can
> > > safely use kfree() and irq_free_descs() in the error path without
> > > involving the device lifecycle prematurely.
> > > 
> > > Thank you for the catch.
> > 
> > Please just use the same approach Andy did for iio_device_alloc(). I posted
> > links (in one of your versions) to that and for the discussion we already had
> > on this same function some time ago.
> 
> I believe this is the approach mentioned above, id est move device_initialize()
> further in the code. But I might misread something...

Me too. But if that was the case, I would not expect any put_device() call. Ah, but I just read
Salah's reply. It will be done in v4.

- Nuno Sá