linux-kernel - Re: [PATCH v1] driver core: Add device links from fwnode only for the primary device

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 23 Mar 2020 17:28:03 -0500
From:   Bjorn Helgaas <helgaas@...nel.org>
To:     "Rafael J. Wysocki" <rafael@...nel.org>
Cc:     Saravana Kannan <saravanak@...gle.com>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Linux PCI <linux-pci@...r.kernel.org>,
        "Cc: Android Kernel" <kernel-team@...roid.com>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v1] driver core: Add device links from fwnode only for
 the primary device

On Sat, Mar 21, 2020 at 11:20:07AM +0100, Rafael J. Wysocki wrote:
> On Sat, Mar 21, 2020 at 5:55 AM Saravana Kannan <saravanak@...gle.com> wrote:
> >
> > Sometimes, more than one (generally two) device can point to the same
> > fwnode.  However, only one device is set as the fwnode's device
> > (fwnode->dev) and can be looked up from the fwnode.
> >
> > Typically, only one of these devices actually have a driver and actually
> > probe. If we create device links for all these devices, then the
> > suppliers' of these devices (with the same fwnode) will never get a
> > sync_state() call because one of their consumer devices will never probe
> > (because they don't have a driver).
> >
> > So, create device links only for the device that is considered as the
> > fwnode's device.
> >
> > One such example of this is the PCI bridge platform_device and the
> > corresponding pci_bus device. Both these devices will have the same
> > fwnode. It's the platform_device that is registered first and is set as
> > the fwnode's device. Also the platform_device is the one that actually
> > probes. Without this patch none of the suppliers of a PCI bridge
> > platform_device would get a sync_state() callback.
> 
> For the record, I think that this is a PCI subsystem problem, but I
> agree with the patch here.

I don't understand the issue here.  Can somebody educate me?  I'm
guessing this is related to pci_set_bus_of_node(), which does (for
PCI-to-PCI bridges):

  bus->dev.of_node = of_node_get(bus->self->dev.of_node);
  bus->dev.fwnode = &bus->dev.of_node->fwnode;

where "bus" points to a struct pci_bus and "bus->self" points to the
struct pci_dev for the bridge leading to the bus?

Is this related to the fact that we have a struct device for both a
PCI-to-PCI bridge and for its downstream bus?

Any suggestions for how could we fix this problem in the PCI
subsystem?