From mboxrd@z Thu Jan 1 00:00:00 1970 From: sergei.shtylyov@cogentembedded.com (Sergei Shtylyov) Date: Fri, 04 Jul 2014 02:39:53 +0400 Subject: [PATCH 2/4] usb: host: xhci-plat: Add support to get PHYs In-Reply-To: References: <1402056736-12674-1-git-send-email-gautam.vivek@samsung.com> <1402056736-12674-3-git-send-email-gautam.vivek@samsung.com> <53A9FCDB.6060805@cogentembedded.com> Message-ID: <53B5DBB9.3080202@cogentembedded.com> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Hello. On 06/25/2014 12:44 PM, Vivek Gautam wrote: >>>> diff --git a/drivers/usb/host/xhci.h b/drivers/usb/host/xhci.h >>>> index 9ffecd5..453d89e 100644 >>>> --- a/drivers/usb/host/xhci.h >>>> +++ b/drivers/usb/host/xhci.h >>>> @@ -1582,6 +1582,9 @@ struct xhci_hcd { >>>> u32 port_status_u0; >>>> /* Compliance Mode Timer Triggered every 2 seconds */ >>>> #define COMP_MODE_RCVRY_MSECS 2000 >>>> + /* phys for the controller */ >>>> + struct phy *phy2_gen; >>>> + struct phy *phy3_gen; >>>> }; >>> I don't think adding new variables here and restricting most of this >>> logic to xhci-plat.c (in the next patch) is the best way to do it. >> Indeed. Actually, I haven't given enough thought to this... >>> There's no conceptual reason why other host controllers (e.g. xhci-pci >>> or even EHCI) could not have a similar need to tune their PHY after >>> reset. PHYs are universal to all host controllers. >>> There is already a 'phy' member in struct usb_hcd which I think is >>> mostly unused right now. I think it would be much less >>> confusing/redundant to reuse that member for this purpose (you could >>> still set it up from xhci_plat_probe(), and then call it from >>> hcd_bus_resume() or something like that). >> That member has type 'struct usb_phy *' while here we have 'struct phy *' >> -- feel the difference. >> I have already tried adding 'struct phy *gen_phy' to 'struct usb_hcd', > So the 'struct phy *' available in the usb_hcd is requested in usb_add_hcd(). > This is requested with the constant string 'usb' : > struct phy *phy = phy_get(hcd->self.controller, "usb"); > This can get the phy with string 'usb' only if, either the host > controller has a device node wherein the phys are given. Yes, that was the plan. Currently, the PHY driver for which this was written can be probed from the device tree only. > Even in this case one can't give same constant string for two > different phys, UTMI+ and PIPE3 phy, isn't it ? Yes, you're right. I didn't really consider the case of some host needing 2 distinct PHY. > Or, the other way can be when host gets a lookup table to look into to > find the relevant phys, something like > Heikki has suggested: > usb: dwc3: host: convey the PHYs to xhci > (https://lkml.org/lkml/2014/6/5/585) and related patch series. Well, AFAIK the look-up table method is already provided by the current PHY core for non-DT use cases; this doesn't seem to have changed with that patch set, does it? > So if we use this second approach, we would need to override the > 'phy_get()' that has been done in usb_add_hcd() > in xhci_plat_probe(), and then use them in later operations. I guess it'd probably be simpler to ignore the 'gen_phy' member of 'struct usb_hcd' in your case (if it gets eventually added). WBR, Sergei