[Intel-gfx] [PATCH 02/51] drm/i915: Don't clear drvdata in ->release

[Daniel Vetter]

On Sat, Feb 22, 2020 at 10:48 AM Daniel Vetter <daniel.vetter at ffwll.ch> wrote:
>
> On Fri, Feb 21, 2020 at 10:36 PM Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > Quoting Daniel Vetter (2020-02-21 21:02:30)
> > > For two reasons:
> > >
> > > - The driver core clears this already for us after we're unloaded in
> > >   __device_release_driver().
> >
> > Even if we abort before loading?
> >
> > History notes that i915_pci_remove was called with a stale pointer on
> > error.
>
> So even if there's a bug we still have the problem that clearing the
> pci_drvdata in our drm_driver->release hook is way too late. You could
> already have bound a new driver to the underlying device. So if driver
> core doesn't clear drvdata on bind failure and we need to clear this
> ourselves, then this line here could actually clear the drvdata of the
> next driver instance bound to the pci device. Not that that's ever
> going to happen outside of very contrived testing.
>
> But looking at really_probe() in base/dd.c we do clear drvdata on
> failure. So no idea how/why that stale drvdata came to be. Anyway
> that's kinda why I cc'ed Greg, so he could confirm that this is
> correct.

Looking at git history, this was fixed in driver core in

commit 0998d0631001288a5974afc0b2a5f568bcdecb4d
Author: Hans de Goede <hdegoede at redhat.com>
Date:   Wed May 23 00:09:34 2012 +0200

    device-core: Ensure drvdata = NULL when no driver is bound

I'll add that to the commit message.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch