Heiko,
On Mon, Nov 9, 2015 at 1:08 PM, Heiko Stuebner <heiko at sntech.de> wrote:
> Hi Doug,
>
> Am Montag, 9. November 2015, 12:59:58 schrieb Doug Anderson:
>> On Sun, Nov 8, 2015 at 8:04 AM, Heiko Stuebner <heiko at sntech.de> wrote:
>> > +static const struct rockchip_usb_phy_pdata rk3066a_pdata = {
>> > + .phys = (struct rockchip_usb_phys[]){
>> > + { .reg = 0x17c, .pll_name = "sclk_otgphy0_480m" },
>> > + { .reg = 0x188, .pll_name = "sclk_otgphy1_480m" },
>> > + { /* sentinel */ }
>> > + },
>> > +};
>> > +
>> > +static const struct rockchip_usb_phy_pdata rk3188_pdata = {
>> > + .phys = (struct rockchip_usb_phys[]){
>> > + { .reg = 0x10c, .pll_name = "sclk_otgphy0_480m" },
>> > + { .reg = 0x11c, .pll_name = "sclk_otgphy1_480m" },
>> > + { /* sentinel */ }
>> > + },
>> > +};
>> > +
>> > +static const struct rockchip_usb_phy_pdata rk3288_pdata = {
>> > + .phys = (struct rockchip_usb_phys[]){
>> > + { .reg = 0x320, .pll_name = "sclk_otgphy0_480m" },
>> > + { .reg = 0x334, .pll_name = "sclk_otgphy1_480m" },
>> > + { .reg = 0x348, .pll_name = "sclk_otgphy2_480m" },
>> > + { /* sentinel */ }
>> > + },
>>
>> Slighly annoying to have to add a table for each SoC. I'd imagine
>> this growing quite large.
>>
>> Would it be possible to query our parent clock name and then append a
>> "_480m" suffix? Then you can magically get the name you want, right?
>> That doesn't preclude you from later overriding this or adding a "dts"
>> property for it, but it means no tables for now and (hopefully) for
>> the forseeable future. ...and you get to kill a bunch of code in this
>> patch...
>>
>> You could fall back to some generic name based on the PHY device tree
>> node name if you wanted, or use the ID allocator...
>>
>>
>> Otherwise, this all looks good to me. If you have a good reason why
>> using the parent name isn't a good idea, let me know and I can add my
>> Reviewed-by on this patch.
>
> Adding a _480m suffix was my initial solution. But the phy-supply clock
> is actually optional in the binding, so it may not be specified at all.
> Still the code needs this pll to work at all after the change, and
> inventing some completely random clock name seemed very strange
> (like pll_usbphy_REG or something) - especially as this would never
> have worked with the downstream clocks. As it is now, the downstream
> clock-tree will work indepent of the supply clock being specified.
>
> Also Rockchip seems to have moved to a different phy-ip, with one
> supplying clock, one pll and everything more tightly coupled,
> so we're probably getting a new driver for it, instead of additional
> entries for the picophy.
OK, seems quite reasonable to me.
Reviewed-by: Douglas Anderson <dianders at chromium.org>
