Re: [PATCH v8 5/8] PCI/portdrv: add mechanism to turn on subdev regulators

["Pali Rohár" <pali@kernel.org>]

On Wednesday 17 November 2021 10:14:19 Jim Quinlan wrote:
> On Tue, Nov 16, 2021 at 3:53 PM Pali Rohár <pali@kernel.org> wrote:
> >
> > On Tuesday 16 November 2021 11:41:22 Rob Herring wrote:
> > > +Pali
> > >
> > > On Mon, Nov 15, 2021 at 2:44 PM Jim Quinlan <james.quinlan@broadcom.com> wrote:
> > > >
> > > > On Thu, Nov 11, 2021 at 5:57 PM Rob Herring <robh@kernel.org> wrote:
> > > > >
> > > > > On Wed, Nov 10, 2021 at 4:15 PM Jim Quinlan <jim2101024@gmail.com> wrote:
> > > > > >
> > > > > > Adds a mechanism inside the root port device to identify standard PCIe
> > > > > > regulators in the DT, allocate them, and turn them on before the rest of
> > > > > > the bus is scanned during pci_host_probe().  A root complex driver can
> > > > > > leverage this mechanism by setting the pci_ops methods add_bus and
> > > > > > remove_bus to pci_subdev_regulators_{add,remove}_bus.
> > > > > >
> > > > > > The allocated structure that contains the regulators is stored in
> > > > > > dev.driver_data.
> > > > > >
> > > > > > The unabridged reason for doing this is as follows.  We would like the
> > > > > > Broadcom STB PCIe root complex driver (and others) to be able to turn
> > > > > > off/on regulators[1] that provide power to endpoint[2] devices.  Typically,
> > > > > > the drivers of these endpoint devices are stock Linux drivers that are not
> > > > > > aware that these regulator(s) exist and must be turned on for the driver to
> > > > > > be probed.  The simple solution of course is to turn these regulators on at
> > > > > > boot and keep them on.  However, this solution does not satisfy at least
> > > > > > three of our usage modes:
> > > > > >
> > > > > > 1. For example, one customer uses multiple PCIe controllers, but wants the
> > > > > > ability to, by script invoking and unbind, turn any or all of them by and
> > > > > > their subdevices off to save power, e.g. when in battery mode.
> > > > > >
> > > > > > 2. Another example is when a watchdog script discovers that an endpoint
> > > > > > device is in an unresponsive state and would like to unbind, power toggle,
> > > > > > and re-bind just the PCIe endpoint and controller.
> > > > > >
> > > > > > 3. Of course we also want power turned off during suspend mode.  However,
> > > > > > some endpoint devices may be able to "wake" during suspend and we need to
> > > > > > recognise this case and veto the nominal act of turning off its regulator.
> > > > > > Such is the case with Wake-on-LAN and Wake-on-WLAN support where PCIe
> > > > > > end-point device needs to be kept powered on in order to receive network
> > > > > > packets and wake-up the system.
> > > > > >
> > > > > > In all of these cases it is advantageous for the PCIe controller to govern
> > > > > > the turning off/on the regulators needed by the endpoint device.  The first
> > > > > > two cases can be done by simply unbinding and binding the PCIe controller,
> > > > > > if the controller has control of these regulators.
> > > > > >
> > > > > > [1] These regulators typically govern the actual power supply to the
> > > > > >     endpoint chip.  Sometimes they may be a the official PCIe socket
> > > > > >     power -- such as 3.3v or aux-3.3v.  Sometimes they are truly
> > > > > >     the regulator(s) that supply power to the EP chip.
> > > > > >
> > > > > > [2] The 99% configuration of our boards is a single endpoint device
> > > > > >     attached to the PCIe controller.  I use the term endpoint but it could
> > > > > >     possible mean a switch as well.
> > > > > >
> > > > > > Signed-off-by: Jim Quinlan <jim2101024@gmail.com>
> > > > > > ---
> > > > > >  drivers/pci/bus.c              | 72 ++++++++++++++++++++++++++++++++++
> > > > > >  drivers/pci/pci.h              |  8 ++++
> > > > > >  drivers/pci/pcie/portdrv_pci.c | 32 +++++++++++++++
> > > > > >  3 files changed, 112 insertions(+)
> > > > > >
> > > > > > diff --git a/drivers/pci/bus.c b/drivers/pci/bus.c
> > > > > > index 3cef835b375f..c39fdf36b0ad 100644
> > > > > > --- a/drivers/pci/bus.c
> > > > > > +++ b/drivers/pci/bus.c
> > > > > > @@ -419,3 +419,75 @@ void pci_bus_put(struct pci_bus *bus)
> > > > > >         if (bus)
> > > > > >                 put_device(&bus->dev);
> > > > > >  }
> > > > > > +
> > > > > > +static void *alloc_subdev_regulators(struct device *dev)
> > > > > > +{
> > > > > > +       static const char * const supplies[] = {
> > > > > > +               "vpcie3v3",
> > > > > > +               "vpcie3v3aux",
> > > > > > +               "vpcie12v",
> > > > > > +       };
> > > > > > +       const size_t size = sizeof(struct subdev_regulators)
> > > > > > +               + sizeof(struct regulator_bulk_data) * ARRAY_SIZE(supplies);
> > > > > > +       struct subdev_regulators *sr;
> > > > > > +       int i;
> > > > > > +
> > > > > > +       sr = devm_kzalloc(dev, size, GFP_KERNEL);
> > > > > > +
> > > > > > +       if (sr) {
> > > > > > +               sr->num_supplies = ARRAY_SIZE(supplies);
> > > > > > +               for (i = 0; i < ARRAY_SIZE(supplies); i++)
> > > > > > +                       sr->supplies[i].supply = supplies[i];
> > > > > > +       }
> > > > > > +
> > > > > > +       return sr;
> > > > > > +}
> > > > > > +
> > > > > > +
> > > > > > +int pci_subdev_regulators_add_bus(struct pci_bus *bus)
> > > > > > +{
> > > > > > +       struct device *dev = &bus->dev;
> > > > > > +       struct subdev_regulators *sr;
> > > > > > +       int ret;
> > > > > > +
> > > > > > +       if (!pcie_is_port_dev(bus->self))
> > > > > > +               return 0;
> > > > > > +
> > > > > > +       if (WARN_ON(bus->dev.driver_data))
> > > > > > +               dev_err(dev, "multiple clients using dev.driver_data\n");
> > > > > > +
> > > > > > +       sr = alloc_subdev_regulators(&bus->dev);
> > > > > > +       if (!sr)
> > > > > > +               return -ENOMEM;
> > > > > > +
> > > > > > +       bus->dev.driver_data = sr;
> > > > > > +       ret = regulator_bulk_get(dev, sr->num_supplies, sr->supplies);
> > > > > > +       if (ret)
> > > > > > +               return ret;
> > > > > > +
> > > > > > +       ret = regulator_bulk_enable(sr->num_supplies, sr->supplies);
> > > > > > +       if (ret) {
> > > > > > +               dev_err(dev, "failed to enable regulators for downstream device\n");
> > > > > > +               return ret;
> > > > > > +       }
> > > > > > +
> > > > > > +       return 0;
> > > > > > +}
> > > > > > +EXPORT_SYMBOL_GPL(pci_subdev_regulators_add_bus);
> > > > >
> > > > > Can't these just go in the portdrv probe and remove functions now?
> > > > >
> > > > > Rob
> > > >
> > > > Not really.  The idea is that  only when a host controller driver does this
> > > >
> > > > static struct pci_ops my_pcie_ops = {
> > > >     .add_bus = pci_subdev_regulators_add_bus , /* see  note below */
> > > >     .remove_bus = pci_subdev_regulators_remove_bus,
> > > >     ...
> > > > }
> > > >
> > > > does it explicitly want this feature.  Without doing this, every PCI
> > > > port in the world will execute a devm_kzalloc() and
> > > > devm_regulator_bulk_get() to (likely) grab nothing, and then there
> > > > will be three superfluous lines in the boot log:
> > >
> > > You can opt-in based on there being a DT node.
> > >
> > > > pci_bus 0001:01: 0001:01 supply vpcie12v not found, using dummy regulator
> > > > pci_bus 0001:01: 0001:01 supply vpcie3v3 not found, using dummy regulator
> > > > pci_bus 0001:01: 0001:01 supply vpcie3v3aux not found, using dummy regulator
> > >
> > > This would be annoying, but not really a reason for how to design this.
> > >
> > > > Secondly, our  HW needs to know when the  alloc/get/enable of
> > > > regulators is done so that the PCIe link can then be attempted.   This
> > > > is pretty much the cornerstone of this patchset.   To do this the brcm
> > > > RC driver's call to pci_subdev_regulators_add_bus() is wrapped by
> > > > brcm_pcie_add_bus() so that we can do this:
> > > >
> > > > static struct pci_ops my_pcie_ops = {
> > > >     .add_bus = brcm_pcie_add_bus ,   /* calls pci_subdev_regulators_add_bus() */
> > > >     .remove_bus = pci_subdev_regulators_remove_bus,
> > >
> > > Do add_bus/remove_bus get called during resume/suspend? If not, how do
> > > you handle the link during resume?
> > >
> > > Maybe there needs to be explicit hooks for link handling. Pali has
> > > been looking into this some.
> > >
> > > Rob
> >
> > Yes, I was looking at it... main power (12V/3.3V) and AUX power (3.3V)
> > needs to be supplied at the "correct" time during establishing link
> > procedure. I wrote it in my RFC email:
> > https://lore.kernel.org/linux-pci/20211022183808.jdeo7vntnagqkg7g@pali/
> Hello Pali,
> 
> I really like your proposal although I would like to get my patchset
> first :-) :-)
> 
> Suppose you came up with a patchset for your ideas-- would that include
> changes to existing RC drivers to use the proposed framework?  If so,
> I am wary that it would
> break at least a few of them.  Or would you just present the framework
> and allow the
> RC drivers' authors to opt-in, one by one?

My idea is to add new "framework" to allow drivers implement new
callbacks for this "framework". There would be no change in drivers
which do not provide these callbacks to ensure that nothing is going to
be broken. I'm planning to implement these callbacks only for RC drivers
for which I have hardware and can properly test to not introduce any
regression. For other existing RC drivers it is up to other authors +
testers. But to decrease future maintenance cost of all RC drivers I
expect that new drivers would not implement any ad-hoc solution in their
"probe" function and instead implement these new callbacks. That is my
idea.

> At any rate, if you want someone to test some of your ideas I can work
> with you.

Perfect! If you have any concerns or you see any issues, please reply
that my RFC email. So I can collect feedback.

Also I sent draft for updating DTS schema for PCIe devices:
https://github.com/devicetree-org/dt-schema/pull/64

> Regards,
> Jim Quinlan
> Broadcom STB
> 
> 
> >
> > I'm not sure if regulator API is the most suitable for this task in PCI
> > core code as there are planty ways how it can be controllers. My idea
> > presented in that email was that driver provides power callback and core
> > pci code would use it.
> >
> > Because power needs to be enabled at the "correct" time during link up,
> > I think that add/remove bus callbacks are unsuitable for this task. This
> > would just cause adding another msleep() calls on different places to
> > make correct timing of link up...
> >
> > I think it is needed to implement generic function for establishing link
> > in pci core code with all required steps.