Re: [PATCH v2 1/2] PCI: Convert to device_create_managed_software_node()

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Fri, Oct 1, 2021 at 12:36 PM Heikki Krogerus
<heikki.krogerus@xxxxxxxxxxxxxxx> wrote:
>
> On Thu, Sep 30, 2021 at 10:04:02AM -0500, Bjorn Helgaas wrote:
> > On Thu, Sep 30, 2021 at 03:12:45PM +0300, Heikki Krogerus wrote:
> > > In quirk_huawei_pcie_sva(), use device_create_managed_software_node()
> > > instead of device_add_properties() to set the "dma-can-stall"
> > > property.
> > >
> > > Reviewed-by: Andy Shevchenko <andy.shevchenko@xxxxxxxxx>
> > > Acked-by: Zhangfei Gao <zhangfei.gao@xxxxxxxxxx>
> > > Signed-off-by: Heikki Krogerus <heikki.krogerus@xxxxxxxxxxxxxxx>
> > > ---
> > > Hi,
> > >
> > > The commit message now says what Bjorn requested, except I left out
> > > the claim that the patch fixes a lifetime issue.
> >
> > Thanks.
> >
> > The commit log should help reviewers determine whether the change is
> > safe and necessary.  So far it doesn't have any hints along that line.
> >
> > Comparing device_add_properties() [1] and
> > device_create_managed_software_node() [2], the only difference in this
> > case is that the latter sets "swnode->managed = true".  The function
> > comment says "managed" means the lifetime of the swnode is tied to the
> > lifetime of dev, hence my question about a lifetime issue.
> >
> > I can see that one reason for this change is to remove the last caller
> > of device_add_properties(), so device_add_properties() itself can be
> > removed.  That's a good reason for wanting to do it, and the commit
> > log could mention it.
>
> Fair enough. I need to explain the why as well as the what.
>
> I'll improve the commit message, but just to be clear, the goal is
> actually not to get rid of device_add_properties(). It is removed in
> the second patch together with device_remove_properties() because
> there are simply no more users for that API.
>
> > But it doesn't help me figure out whether it's safe.  For that,
> > I need to know the effect of setting "managed = true".  Obviously
> > it means *something*, but I don't know what.  It looks like the only
> > test is in software_node_notify():
> >
> >   device_del
> >     device_platform_notify_remove
> >       software_node_notify_remove
> >         sysfs_remove_link(dev_name)
> >         sysfs_remove_link("software_node")
> >         if (swnode->managed)                 <--
> >           set_secondary_fwnode(dev, NULL)
> >           kobject_put(&swnode->kobj)
> >     device_remove_properties
> >       if (is_software_node())
> >         fwnode_remove_software_node
> >           kobject_put(&swnode->kobj)
> >         set_secondary_fwnode(dev, NULL)
> >
> > I'm not sure what's going on here; it looks like some redundancy with
> > multiple calls of kobject_put() and set_secondary_fwnode().  Maybe you
> > are in the process of removing device_remove_properties() as well as
> > device_add_properties()?
>
> It'll get removed, but that's not the goal. The goal is to get rid of
> the call to it in device_del(), so not the function itself. That call
> is the problem here as explained in commit 151f6ff78cdf ("software
> node: Provide replacement for device_add_properties()").
>
> I'll split the second patch, and first only remove that
> device_remove_properties() call from device_del(), and only after
> that remove the functions themselves.

So I'm expecting a v3 of this.



[Index of Archives]     [DMA Engine]     [Linux Coverity]     [Linux USB]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [Greybus]

  Powered by Linux