Received-SPF: pass (google.com: domain of linux-wireless-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <c2b6599182b2f543da2d5fee17dce4b3a094535a.camel@sipsolutions.net>
Subject: Re: [PATCH v2] PCI/PM: enable runtime PM later during device
 scanning
From:   Johannes Berg <johannes@sipsolutions.net>
To:     Lukas Wunner <lukas@wunner.de>
Cc:     linux-wireless@vger.kernel.org, linux-pci@vger.kernel.org,
        linux-pm@vger.kernel.org, rafael@kernel.org,
        Bjorn Helgaas <bhelgaas@google.com>
Date:   Wed, 07 Jun 2023 09:58:21 +0200
In-Reply-To: <20230607074939.GA17700@wunner.de>
References: <20230605121621.4259f1be6cd2.Idbaa55b93f780838af44ebccb84c36f60716df04@changeid>
         <20230605203519.bc4232207449.Idbaa55b93f780838af44ebccb84c36f60716df04@changeid>
         <20230607074939.GA17700@wunner.de>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.48.2 (3.48.2-1.fc38) 
MIME-Version: 1.0
Precedence: bulk

On Wed, 2023-06-07 at 09:49 +0200, Lukas Wunner wrote:
> On Mon, Jun 05, 2023 at 08:35:45PM +0200, Johannes Berg wrote:
> > @@ -3139,6 +3139,7 @@ void pci_pm_init(struct pci_dev *dev)
> >  	u16 pmc;
> > =20
> >  	pm_runtime_forbid(&dev->dev);
> > +	pm_runtime_get_noresume(&dev->dev);
> >  	pm_runtime_set_active(&dev->dev);
> >  	pm_runtime_enable(&dev->dev);
> >  	device_enable_async_suspend(&dev->dev);
> > @@ -335,9 +336,12 @@ void pci_bus_add_device(struct pci_dev *dev)
> >  	int retval;
> > =20
> >  	/*
> > -	 * Can not put in pci_device_add yet because resources
> > -	 * are not assigned yet for some devices.
> > +	 * Allow runtime PM only here, since otherwise we may
> > +	 * try to suspend a device that isn't fully configured
> > +	 * yet, which causes problems.
> >  	 */
> > +	pm_runtime_put_noidle(&dev->dev);
> > +
> >  	pcibios_bus_add_device(dev);
> >  	pci_fixup_device(pci_fixup_final, dev);
> >  	pci_create_sysfs_dev_files(dev);
>=20
> There seem to be many different callers that end up in pci_pm_init()
> and pci_bus_add_device().
>=20
> Is it guaranteed that the two functions are always called in order?
> Do callers exist which only invoke the former but not the latter or
> vice-versa?  Can it happen that a caller of the former errors out,
> so the latter is never called, leading to a runtime PM ref imbalance?

I did ask myself that too, and honestly, I'm not entirely sure - need
somebody more familiar to really understand that, I think.

Most places elsewhere _do_ call both, and it feels like you have to call
both if you want to do something with the device.

However there are a few places that seem to call the first part and then
remove the device again immediately after. That also seems harmless
though.

> It would be easier to ascertain correctness if you could find a
> function at a higher level which (indirectly) calls both pci_pm_init()
> and pci_bus_add_device() so that you can acquire and release the
> runtimw PM ref in that single function.
>=20

Unfortunately, there isn't such a place, since the scanning is done by
various bus walks.

johannes