Re: [RFC PATCH 4/5] arm: omap2: support port power on lan95xx devices

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

 



On Mon, Dec 3, 2012 at 12:52 PM, Andy Green <andy.green@xxxxxxxxxx> wrote:
>>>> +static void ehci_hub_power_off(struct power_controller *pc, struct
>>>> device
>>>> *dev)
>>>> +{
>>>> +       gpio_set_value(GPIO_HUB_NRESET, 0);
>>>> +       gpio_set_value(GPIO_HUB_POWER, 0);
>>>> +}
>>>> +
>>>> +static struct usb_port_power_switch_data root_hub_port_data = {
>>>> +       .hub_tier       = 0,
>>>> +       .port_number = 1,
>>>> +       .type = USB_PORT_CONNECT_TYPE_HARD_WIRED,
>>>> +};
>>>> +
>>>> +static struct usb_port_power_switch_data smsc_hub_port_data = {
>>>> +       .hub_tier       = 1,
>>>> +       .port_number = 1,
>>>> +       .type = USB_PORT_CONNECT_TYPE_HARD_WIRED,
>>>> +};
>>>> +
>>>> +static struct power_controller pc = {
>>>> +       .name = "omap_hub_eth_pc",
>>>> +       .count = ATOMIC_INIT(0),
>>>> +       .power_on = ehci_hub_power_on,
>>>> +       .power_off = ehci_hub_power_off,
>>>> +};
>>>> +
>>>> +static inline int omap_ehci_hub_port(struct device *dev)
>>>> +{
>>>> +       /* we expect dev->parent points to ehcd controller */
>>>> +       if (dev->parent && !strcmp(dev_name(dev->parent),
>>>> "ehci-omap.0"))
>>>> +               return 1;
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +static inline int dev_pc_match(struct device *dev)
>>>> +{
>>>> +       struct device *anc;
>>>> +       int ret = 0;
>>>> +
>>>> +       if (likely(strcmp(dev_name(dev), "port1")))
>>>> +               goto exit;
>>>> +
>>>> +       if (dev->parent && (anc = dev->parent->parent)) {
>>>> +               if (omap_ehci_hub_port(anc)) {
>>>> +                       ret = 1;
>>>> +                       goto exit;
>>>> +               }
>>>> +
>>>> +               /* is it port of lan95xx hub? */
>>>> +               if ((anc = anc->parent) && omap_ehci_hub_port(anc)) {
>>>> +                       ret = 2;
>>>> +                       goto exit;
>>>> +               }
>>>> +       }
>>>> +exit:
>>>> +       return ret;
>>>> +}
>>>> +
>>>> +/*
>>>> + * Notifications of device registration
>>>> + */
>>>> +static int device_notify(struct notifier_block *nb, unsigned long
>>>> action,
>>>> void *data)
>>>> +{
>>>> +       struct device *dev = data;
>>>> +       int ret;
>>>> +
>>>> +       switch (action) {
>>>> +       case DEV_NOTIFY_ADD_DEVICE:
>>>> +               ret = dev_pc_match(dev);
>>>> +               if (likely(!ret))
>>>> +                       goto exit;
>>>> +               if (ret == 1)
>>>> +                       dev_pc_bind(&pc, dev, &root_hub_port_data,
>>>> sizeof(root_hub_port_data));
>>>> +               else
>>>> +                       dev_pc_bind(&pc, dev, &smsc_hub_port_data,
>>>> sizeof(smsc_hub_port_data));
>>>> +               break;
>>>> +
>>>> +       case DEV_NOTIFY_DEL_DEVICE:
>>>> +               break;
>>>> +       }
>>>> +exit:
>>>> +       return 0;
>>>> +}
>>>> +
>>>> +static struct notifier_block usb_port_nb = {
>>>> +       .notifier_call = device_notify,
>>>> +};
>>>> +
>>>
>>>
>>>
>>> Some thoughts on trying to make this functionality specific to power only
>>> and ehci hub port only:
>>>
>>>   - Quite a few boards have smsc95xx... they're all going to carry these
>>> additions in the board file?  Surely you'll have to generalize the pieces
>>
>>
>> All things are board dependent because we are discussing peripheral
>> device(not builtin SoC devices), so it is proper to put it in the board
>> file.
>> If some boards want to share it, we can put it in a single .c file and
>> let board file include it.
>
>
> Where would the .c file go... SMSC is not platform, or even arch specific
> chip (eg, iMX or MIPS or even x86 boards can have it), so not
> arch/arm/mach-xxxx or arch/arm/plat-xxx or arch/arm.  I guess it would go in
> drivers/usb or drivers/net.

How does drivers/usb or drivers/net know the SMSC is used on beagle or
panda? Different power control approach is used in the two boards even
same SMSC chip is shipped in the two boards.

>
> Push in ARM-Land is towards single kernels which support many platforms, a
> good case in point is omap2plus_defconfg which wants to allow to support all
> OMAP2/3/4/5 platforms in one kernel.  If you "include" that C file over and
> over in board files, it's not very nice for that.  They anyway want to
> eliminate board files for the single kernel binary reason, and just have one
> load of config come in by dtb.

I only mean it is reasonable to put the power control code into board
file because
each board may take different power control approach even same SMSC chip
is used. I understand DT only describes the difference of the board via device
node, and I am not sure if the current DT is enough to convert the board file
into data as far as the problem is concerned.

>
> So it guides you towards having static helper code once in drivers/ for the
> scenarios you support... if that's where you end up smsc is less good a
> target for a helper than to have helpers for classes of object like
> regulator and clk, that you can combine and reuse on all sorts of target
> devices, which is device_asset approach.
>
> It also guides you to having the special platform sauce be something that
> can go into the dtb: per-board code can't.  That's why device_asset stuff
> only places asset structs in the board file and is removing code from there.
>
>
>>> that perform device_path business out of the omap4panda board file at
>>> least.
>>> At that point the path matching code becomes generic
>>> end-of-the-device-path
>>> matching code.
>>
>>
>> Looks Alan has mentioned, there might be no generic way, and any device's
>> name change in the path may make the way fragile.
>
>
> What you have provided is no less fragile if you allow "port1" and the
> ehci-omap.0 to be set from the outside.
>
> Unless someone NAKs it for sure already (if you're already sure you're going
> to, please do so to avoid wasting time), I'll issue a try#2 of my code later
> which demonstrates what I mean.  At least I guess it's useful for
> comparative purposes.
>
>
>>>   - How could these literals like "port1" etc be nicely provided by
>>> Device
>>> Tree?  In ARM-land there's pressure to eventually eliminate board files
>>> completely and pass in everything from dtb.  device_asset can neatly grow
>>> DT
>>> bindings in a generic way, since the footprint in the board file is some
>>
>>
>> IMO, it isn't necessary to expose these assets to device or users, from
>> the
>> view of device or user, which only cares two actions(poweron and poweroff)
>> about the discussed problem. Also it should be better to put these assets
>> into device resource list, instead of introducing them in 'struct device'.
>
>
> From the point of view of allowing it to be reused on different boards /
> platforms / arches, you are going to have to do something better than have
> literals in the code.
>
>
>>> regulators that already have dt bindings and some device_asset structs.
>>> Similarly there's pressure for magic code to service a board (rather than
>>> SoC) to go elsewhere than the board file.
>>
>>
>> Looks you associate these assets with ehci-omap device, which mightn't be
>> enough, because we need to control port's power for supporting port
>> power off mechanism. Do you have generic way to associate these assets
>> with usb port device and let port use it generally?
>
>
> Yes, you need a parent device pointer (ehci host controller in this case)
> and the path rhs, but only stuff that is defined by usb stack code.  Needing
> a parent pointer is OK because this stuff only has meaning for hardwired
> assets on the platform, so the parent device will always be known as a
> platform_device at boot time anyway.

parent device pointer may work on the panda problem, but may not work
on other case if one hardwired device is powered by another power domain.

So it is not a general solution on usb port power off.

> The code I'll provide will work the same in sdio or other bus case, just use
> mmc host controller pointer and the sdio device name the same way.
>
>
>>>   - Shouldn't this take care of enabling and disabling the ULPI PHY clock
>>> on
>>> Panda too?  There's no purpose leaving it running if the one thing the
>>> ULPI
>>> PHY is connected to is depowered, and when you do power it, on Panda you
>>> will reset the ULPI PHY at the same time anyway (smsc reset and ULPI PHY
>>> reset are connected together on Panda).  Then you can eliminate
>>> omap4_ehci_init() in the board file.
>>
>>
>> OK, we can include the ULPI PHY clock things easily in ->power_on() and
>> ->power_off() of 'power controller'
>
>
> Yes if the ARM people will accept establishing more code in board files that
> doesn't have a DT story.

As I explained above, it is reasonable to put the power control code in board
file, but current DT could convert these board file into device node?

Thanks,
-- 
Ming Lei
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


[Index of Archives]     [Linux Arm (vger)]     [ARM Kernel]     [ARM MSM]     [Linux Tegra]     [Linux WPAN Networking]     [Linux Wireless Networking]     [Maemo Users]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Trails]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux