[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <659f00ed271b3_5cee2942@dwillia2-xfh.jf.intel.com.notmuch>
Date: Wed, 10 Jan 2024 12:41:17 -0800
From: Dan Williams <dan.j.williams@...el.com>
To: Lukas Wunner <lukas@...ner.de>, Bartosz Golaszewski <brgl@...ev.pl>
CC: Kalle Valo <kvalo@...nel.org>, "David S . Miller" <davem@...emloft.net>,
Eric Dumazet <edumazet@...gle.com>, Jakub Kicinski <kuba@...nel.org>, "Paolo
Abeni" <pabeni@...hat.com>, Rob Herring <robh+dt@...nel.org>, "Krzysztof
Kozlowski" <krzysztof.kozlowski+dt@...aro.org>, Conor Dooley
<conor+dt@...nel.org>, Bjorn Andersson <andersson@...nel.org>, Konrad Dybcio
<konrad.dybcio@...aro.org>, Catalin Marinas <catalin.marinas@....com>, "Will
Deacon" <will@...nel.org>, Bjorn Helgaas <bhelgaas@...gle.com>, "Heiko
Stuebner" <heiko@...ech.de>, Jernej Skrabec <jernej.skrabec@...il.com>,
"Chris Morgan" <macromorgan@...mail.com>, Linus Walleij
<linus.walleij@...aro.org>, Geert Uytterhoeven <geert+renesas@...der.be>,
Arnd Bergmann <arnd@...db.de>, Neil Armstrong <neil.armstrong@...aro.org>,
Nícolas F . R . A . Prado <nfraprado@...labora.com>,
Marek Szyprowski <m.szyprowski@...sung.com>, Peng Fan <peng.fan@....com>,
Robert Richter <rrichter@....com>, Dan Williams <dan.j.williams@...el.com>,
Jonathan Cameron <Jonathan.Cameron@...wei.com>, Terry Bowman
<terry.bowman@....com>, Kuppuswamy Sathyanarayanan
<sathyanarayanan.kuppuswamy@...ux.intel.com>, Ilpo
Järvinen <ilpo.jarvinen@...ux.intel.com>, Huacai Chen
<chenhuacai@...nel.org>, Alex Elder <elder@...aro.org>, Srini Kandagatla
<srinivas.kandagatla@...aro.org>, Greg Kroah-Hartman
<gregkh@...uxfoundation.org>, <linux-wireless@...r.kernel.org>,
<netdev@...r.kernel.org>, <devicetree@...r.kernel.org>,
<linux-kernel@...r.kernel.org>, <linux-arm-msm@...r.kernel.org>,
<linux-arm-kernel@...ts.infradead.org>, <linux-pci@...r.kernel.org>, "Bartosz
Golaszewski" <bartosz.golaszewski@...aro.org>
Subject: Re: [RFC 3/9] PCI/portdrv: create platform devices for child OF nodes
[ add Terry ]
Lukas Wunner wrote:
> On Wed, Jan 10, 2024 at 01:55:18PM +0100, Bartosz Golaszewski wrote:
> > On Tue, Jan 9, 2024 at 3:43???PM Lukas Wunner <lukas@...ner.de> wrote:
> > > On Thu, Jan 04, 2024 at 02:01:17PM +0100, Bartosz Golaszewski wrote:
> > > > In order to introduce PCIe power-sequencing, we need to create platform
> > > > devices for child nodes of the port driver node. They will get matched
> > > > against the pwrseq drivers (if one exists) and then the actuak PCIe
> > > > device will reuse the node once it's detected on the bus.
> > > [...]
> > > > --- a/drivers/pci/pcie/portdrv.c
> > > > +++ b/drivers/pci/pcie/portdrv.c
> > > > @@ -715,7 +716,7 @@ static int pcie_portdrv_probe(struct pci_dev *dev,
> > > > pm_runtime_allow(&dev->dev);
> > > > }
> > > >
> > > > - return 0;
> > > > + return devm_of_platform_populate(&dev->dev);
> > > > }
> > >
> > > I think this belongs in of_pci_make_dev_node(), portdrv seems totally
> > > the wrong place. Note that you're currently calling this for RCECs
> > > (Root Complex Event Collectors) as well, which is likely not what
> > > you want.
> > >
> >
> > of_pci_make_dev_node() is only called when the relevant PCI device is
> > instantiated which doesn't happen until it's powered-up and scanned -
> > precisely the problem I'm trying to address.
>
> No, of_pci_make_dev_node() is called *before* device_attach(),
> i.e. before portdrv has even probed. So it seems this should
> work perfectly well for your use case.
>
>
> > > devm functions can't be used in the PCI core, so symmetrically call
> > > of_platform_unpopulate() from of_pci_remove_node().
> >
> > I don't doubt what you're saying is true (I've seen worse things) but
> > this is the probe() callback of a driver using the driver model. Why
> > wouldn't devres work?
>
> The long term plan is to move the functionality in portdrv to
> the PCI core. Because devm functions can't be used in the PCI
> core, adding new ones to portdrv will *add* a new roadblock to
> migrating portdrv to the PCI core. In other words, it makes
> future maintenance more difficult.
>
> Generally, only PCIe port services which share the same interrupt
> (hotplug, PME, bandwith notification, flit error counter, ...)
> need to live in portdrv. Arbitrary other stuff should not be
> shoehorned into portdrv.
I came here to say the same thing. It is already the case that portdrv
is not a good model to build new functionality upon [1], and PCI core
enlightenment should be considered first.
The portdrv model is impeding Terry's CXL Port error handling effort, so
I am on the lookout for portdrv growing new entanglements to unwind
later.
[1]: http://lore.kernel.org/r/20221025232535.GA579167@bhelgaas
Powered by blists - more mailing lists