dwc3: add support for hardware with multiple ports on USB2 hub enabled

[Felipe Balbi]

Hi,

Martin Blumenstingl <martin.blumenstingl at googlemail.com> writes:
> On Mon, Jan 9, 2017 at 12:18 PM, Felipe Balbi
> <felipe.balbi at linux.intel.com> wrote:
>>
>> Hi,
>>
>> Martin Blumenstingl <martin.blumenstingl at googlemail.com> writes:
>>>> Martin Blumenstingl <martin.blumenstingl at googlemail.com> writes:
>>>>> while adding USB support on the Amlogic Meson GXL / GXM SoCs I have
>>>>> come across something I did not know yet:
>>>>> dwc3 has an internal USB2 hub (from what I can read in the code there
>>>>> seem to be multiple USB3 ports supported as well).
>>>>
>>>> no, that's not true. It has a roothub when working as host. But that's
>>>> it. When working as peripheral, it's always single-port AFAIR.
>>> OK, I should have been more clear here: I am only talking about
>>> host-mode since DWC3_GHWPARAMS0 on the GXL/GXM SoCs has a value of
>>> 0x20208009 which translates to "DWC3_GHWPARAMS0_MODE_HOST".
>>>
>>> are you sure about the fact that it does not have an internal hub?
>>> What I see in both, the vendor kernel's and my own patched mainline
>>> kernel log is:
>>> [   19.130331 at 3] xhci-hcd xhci-hcd.0.auto: xHCI Host Controller
>>> [   19.130385 at 3] xhci-hcd xhci-hcd.0.auto: new USB bus registered,
>>> assigned bus number 1
>>> [   19.139666 at 3] xhci-hcd xhci-hcd.0.auto: irq 62, io mem 0xc9000000
>>> [   19.145295 at 3] hub 1-0:1.0: USB hub found
>>
>> this is the roothub :-)
> agreed :)
>
>>> [   19.148098 at 3] hub 1-0:1.0: 3 ports detected
> this is the one I wanted to point out: in this case there are 3 ports
> on the (USB2) roothub.
> the SoC has one USB2 for each of these ports, and all 3 PHYs have to
> be initialized to get USB working (this seems to be my main issue).
>
>>> [   19.152396 at 3] xhci-hcd xhci-hcd.0.auto: xHCI Host Controller
>>> [   19.157813 at 3] xhci-hcd xhci-hcd.0.auto: new USB bus registered,
>>> assigned bus number 2
>>> [   19.166598 at 3] hub 2-0:1.0: USB hub found
>>> [   19.169452 at 3] hub 2-0:1.0: config failed, hub doesn't have any
>>> ports! (err -19)
>>> This is from a GXM SoC which also comes with 3x USB2 PHYs (these are
>>> not Synopsys DesignWare PHYs but custom ones from Amlogic).
>>> I see similar messages but with "2 ports detect" on a GXL SoC which
>>> comes with 2x USB2 PHYs.
>>
>> The fact that is claims that it has "no ports" hints at quirk caused,
>> likely, by poor RTL edits. coreConsultant should make sure that
>> MAX_PORTS (in xHCI HCC_PARAMS* register set) was set to correct value,
>> but somehow that field is set to 0 in one of your Amlogic SoC.
> yes, it's a USB2-only configuration (so there are no USB3 ports on the
> internal roothub)

aha, makes sense.

>>>>> When searching the web I did not come across any SoC that uses a
>>>>> configuration with more than one port enabled.
>>>>>
>>>>> On my Amlogic Meson GXM device (consumer device, no development board)
>>>>> I see the following USB2 PHY register configuration (full register
>>>>> dump from the kernel that was shipped with the device is attached):
>>>>> GUSB2PHYCFG(0) = 0x40102500
>>>>> GUSB2PHYCFG(1) = 0x40102540
>>>>> GUSB2PHYCFG(2) = 0x40102540
>>>>
>>>> multiple PHYs are only used by the host block (xHCI). Don't touch these
>>>> and just let xHCI core handle the ports.
>>> could you be more specific with "xHCI core" - do you mean the core in
>>> the dwc3 IP or drivers/usb/host/xhci-*?
>>> However, we still have a "problem" here: the USB PHYs for each
>>> "enabled" port have to be turned on (if I leave only one USB PHY
>>> disabled then none of the ports is working). unfortunately the current
>>> code (both, in dwc3 and drivers/usb/host/*) assumes that there's
>>> either 0 or 1 PHY for each HCD.

true. But even so, that PHY handle is only needed for devices which
weren't properly configured at coreConsultant time.

>>>>> Then vendor kernel sources (a 3.14 kernel) are adding the resets for
>>>>> GUSB2PHYCFG([1-3]) in dwc3_core_soft_reset().
>>>>
>>>> That shouldn't be necessary, actually. If it is, it means the HW was
>>>> poorly integrated. In that case, we _can_ add the other resets, but I
>>>> need confirmation that they are needed by means of a public errata
>>>> document.
>>>>
>>>>> A mainline 4.9+(Meson GXL USB PHY patches + dwc3/xhci-plat DMA patches
>>>>> from linux-usb) kernel works fine even with just applying the reset to
>>>>> GUSB2PHYCFG(0).
>>>>
>>>> there you go
>>> does that mean that the reset of GUSB2PHYCFG(0) (which is part of the
>>> current dwc3 code in dwc3_phy_setup) is done only because of the
>>> quirks/erratas? in other words: do you mean that one would not have to
>>> reset GUSB2PHYCFG(0) if there were no erratas?
>>
>> no, it's done for peripheral configuration of dwc3. Look at the code
>> more carefully:
> sorry for the confusion - the word "reset" should be "configuration".
> The function I am looking at is dwc3_phy_setup(): apart from toggling
> some "errata bits" it also configures the PHY modes. I am wondering if
> I need to do this setup (DWC3_GUSB2PHYCFG_ULPI_UTMI and
> DWC3_GUSB2PHYCFG_PHYIF_MASK related bits) not only for the *first*
> port ("DWC3_GUSB2PHYCFG(0)") but also for the "other" ports (index 1
> and 2 in above case, where the roothub has 3 ports)

Ideally, that should've been set at coreConsultant (RTL instantiation)
time. If it wasn't, then we need to add a quirk for these SoCs
accordingly. We _do_ need documentation that this quirk is, indeed,
needed.


-- 
balbi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 832 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-amlogic/attachments/20170109/66c864d6/attachment.sig>