[PATCH v7 09/13] PCI: Give pcim_set_mwi() its own devres callback

[Philipp Stanner]

On Thu, 2024-06-13 at 20:19 +0300, Ilpo Järvinen wrote:
> On Wed, 5 Jun 2024, Philipp Stanner wrote:
> 
> > Managing pci_set_mwi() with devres can easily be done with its own
> > callback, without the necessity to store any state about it in a
> > device-related struct.
> > 
> > Remove the MWI state from struct pci_devres.
> > Give pcim_set_mwi() a separate devres-callback.
> > 
> > Signed-off-by: Philipp Stanner <pstanner at redhat.com>
> > ---
> >  drivers/pci/devres.c | 29 ++++++++++++++++++-----------
> >  drivers/pci/pci.h    |  1 -
> >  2 files changed, 18 insertions(+), 12 deletions(-)
> > 
> > diff --git a/drivers/pci/devres.c b/drivers/pci/devres.c
> > index 936369face4b..0bafb67e1886 100644
> > --- a/drivers/pci/devres.c
> > +++ b/drivers/pci/devres.c
> > @@ -361,24 +361,34 @@ void __iomem
> > *devm_pci_remap_cfg_resource(struct device *dev,
> >  }
> >  EXPORT_SYMBOL(devm_pci_remap_cfg_resource);
> >  
> > +static void __pcim_clear_mwi(void *pdev_raw)
> > +{
> > +       struct pci_dev *pdev = pdev_raw;
> > +
> > +       pci_clear_mwi(pdev);
> > +}
> > +
> >  /**
> >   * pcim_set_mwi - a device-managed pci_set_mwi()
> > - * @dev: the PCI device for which MWI is enabled
> > + * @pdev: the PCI device for which MWI is enabled
> >   *
> >   * Managed pci_set_mwi().
> >   *
> >   * RETURNS: An appropriate -ERRNO error value on error, or zero
> > for success.
> >   */
> > -int pcim_set_mwi(struct pci_dev *dev)
> > +int pcim_set_mwi(struct pci_dev *pdev)
> >  {
> > -       struct pci_devres *dr;
> > +       int ret;
> >  
> > -       dr = find_pci_dr(dev);
> > -       if (!dr)
> > -               return -ENOMEM;
> > +       ret = devm_add_action(&pdev->dev, __pcim_clear_mwi, pdev);
> > +       if (ret != 0)
> > +               return ret;
> > +
> > +       ret = pci_set_mwi(pdev);
> > +       if (ret != 0)
> > +               devm_remove_action(&pdev->dev, __pcim_clear_mwi,
> > pdev);
> 
> I'm sorry if this is a stupid question but why this cannot use 
> devm_add_action_or_reset()?

For MWI that could be done.

This is basically just consistent with the new pcim_enable_device() in
patch No.11 where devm_add_action_or_reset() could collide with
pcim_pin_device().

We could squash usage of devm_add_action_or_reset() in here. I don't
care.

P.


> 
> > -       dr->mwi = 1;
> > -       return pci_set_mwi(dev);
> > +       return ret;
> >  }
> >  EXPORT_SYMBOL(pcim_set_mwi);
>