[U-Boot] [PATCH 10/16] usb: xhci: Program 'route string' in the input slot context

Wed Jun 28 12:27:19 UTC 2017

Hi Stefan,

On Wed, Jun 28, 2017 at 7:28 PM, Stefan Roese <sr at denx.de> wrote:
> Hi Bin,
>
>
> On 27.06.2017 10:27, Bin Meng wrote:
>>
>> Hi Stefan,
>>
>> On Tue, Jun 27, 2017 at 1:23 PM, Stefan Roese <sr at denx.de> wrote:
>>>
>>> Hi Bin,
>>>
>>>
>>> On 27.06.2017 02:01, Bin Meng wrote:
>>>>
>>>>
>>>> On Tue, Jun 27, 2017 at 2:07 AM, Marek Vasut <marex at denx.de> wrote:
>>>>>
>>>>>
>>>>> On 06/24/2017 03:57 AM, Bin Meng wrote:
>>>>>>
>>>>>>
>>>>>> Hi Marek,
>>>>>>
>>>>>> On Sat, Jun 24, 2017 at 2:02 AM, Marek Vasut <marex at denx.de> wrote:
>>>>>>>
>>>>>>>
>>>>>>> On 06/23/2017 11:54 AM, Bin Meng wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>> xHCI spec says: the values of the 'route string' field shall be
>>>>>>>> initialized by the first 'Address Device' command issued to a
>>>>>>>> device slot, and shall not be modified by any other command.
>>>>>>>>
>>>>>>>> So far U-Boot does not program this field, and it does not prevent
>>>>>>>> SS device directly attached to root port, or HS device behind an HS
>>>>>>>> hub, from working, due to the fact that 'route string' is used by
>>>>>>>> the xHC to target SS packets. But in order to enumerate devices
>>>>>>>> behind an SS hub, this field must be programmed.
>>>>>>>>
>>>>>>>> With this commit and along with previous commits, now SS & HS
>>>>>>>> devices
>>>>>>>> attached to a USB 3.0 hub can be enumerated by U-Boot.
>>>>>>>>
>>>>>>>> As usual, this new feature is only available when DM is on.
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Great, but I really dislike the ifdef pollution, so this needs to be
>>>>>>> sorted out.
>>>>>>
>>>>>>
>>>>>>
>>>>>> The ifdef was needed due to it calls DM APIs or access DM udevice. I
>>>>>> have no intention to add a new feature to the non-DM driver.
>>>>>
>>>>>
>>>>>
>>>>> But then this creates a massive disparity, it's like we're growing two
>>>>> USB stacks.
>>>>>
>>>>
>>>> Yep, unfortunately. But if we continue adding new features/fixes to
>>>> the old non-DM stuff, I am not sure how this can encourage people to
>>>> switch to DM.
>>>
>>>
>>>
>>> Correct. We definitely don't want to add new features to non-DM
>>> drivers / IF, if this is non-trivial.
>>>
>>>> Maybe I can check all boards that use xHCI to see if
>>>> they are switched to DM?
>>>
>>>
>>>
>>> xHCI is still quite new in U-Boot, so I would expect that all
>>> platforms using it, are using DM or at least not far away from using
>>> it. Yes, please check all xHCI "users", if they use DM. Then we
>>> know if and which users need some "persuasion" to switch over to
>>> DM soon. ;)
>>
>>
>> I checked all boards that have CONFIG_USB_XHCI_HCD defined but without
>> CONFIG_DM_USB. Here is the list.
>>
>> configs/uniphier_v8_defconfig:36:CONFIG_USB_XHCI_HCD=y
>> configs/xilinx_zynqmp_zc1751_xm015_dc1_defconfig:62:CONFIG_USB_XHCI_HCD=y
>> configs/am57xx_evm_nodt_defconfig:53:CONFIG_USB_XHCI_HCD=y
>> configs/evb-rk3328_defconfig:34:CONFIG_USB_XHCI_HCD=y
>> configs/ls1021atwr_nor_lpuart_defconfig:43:CONFIG_USB_XHCI_HCD=y
>> configs/uniphier_pxs2_ld6b_defconfig:44:CONFIG_USB_XHCI_HCD=y
>> configs/ls1012ardb_qspi_SECURE_BOOT_defconfig:43:CONFIG_USB_XHCI_HCD=y
>>
>> configs/ls1021atwr_sdcard_ifc_SECURE_BOOT_defconfig:57:CONFIG_USB_XHCI_HCD=y
>> configs/k2e_hs_evm_defconfig:34:CONFIG_USB_XHCI_HCD=y
>> configs/ls1021aqds_sdcard_qspi_defconfig:61:CONFIG_USB_XHCI_HCD=y
>> configs/am43xx_evm_ethboot_defconfig:48:CONFIG_USB_XHCI_HCD=y
>> configs/xilinx_zynqmp_ep_defconfig:70:CONFIG_USB_XHCI_HCD=y
>> configs/ls1021aqds_nand_defconfig:57:CONFIG_USB_XHCI_HCD=y
>> configs/ls1021atwr_qspi_defconfig:50:CONFIG_USB_XHCI_HCD=y
>> configs/k2g_evm_defconfig:45:CONFIG_USB_XHCI_HCD=y
>> configs/am57xx_evm_defconfig:63:CONFIG_USB_XHCI_HCD=y
>> configs/am43xx_hs_evm_defconfig:49:CONFIG_USB_XHCI_HCD=y
>> configs/am43xx_evm_defconfig:39:CONFIG_USB_XHCI_HCD=y
>> configs/ls1021atwr_nor_defconfig:42:CONFIG_USB_XHCI_HCD=y
>> configs/firefly-rk3399_defconfig:59:CONFIG_USB_XHCI_HCD=y
>> configs/puma-rk3399_defconfig:78:CONFIG_USB_XHCI_HCD=y
>> configs/cl-som-am57x_defconfig:55:CONFIG_USB_XHCI_HCD=y
>> configs/ls1021aqds_nor_SECURE_BOOT_defconfig:43:CONFIG_USB_XHCI_HCD=y
>> configs/uniphier_pro4_defconfig:43:CONFIG_USB_XHCI_HCD=y
>> configs/xilinx_zynqmp_zc1751_xm016_dc2_defconfig:61:CONFIG_USB_XHCI_HCD=y
>> configs/xilinx_zynqmp_zcu102_defconfig:63:CONFIG_USB_XHCI_HCD=y
>> configs/ls1021atwr_nor_SECURE_BOOT_defconfig:42:CONFIG_USB_XHCI_HCD=y
>> configs/cm_t43_defconfig:67:CONFIG_USB_XHCI_HCD=y
>> configs/k2g_hs_evm_defconfig:36:CONFIG_USB_XHCI_HCD=y
>> configs/am43xx_evm_qspiboot_defconfig:45:CONFIG_USB_XHCI_HCD=y
>> configs/ls1021aqds_qspi_defconfig:50:CONFIG_USB_XHCI_HCD=y
>> configs/am57xx_hs_evm_defconfig:67:CONFIG_USB_XHCI_HCD=y
>> configs/xilinx_zynqmp_zcu102_revB_defconfig:63:CONFIG_USB_XHCI_HCD=y
>> configs/ls1021aqds_sdcard_ifc_defconfig:55:CONFIG_USB_XHCI_HCD=y
>> configs/uniphier_ld20_defconfig:39:CONFIG_USB_XHCI_HCD=y
>> configs/am43xx_evm_usbhost_boot_defconfig:61:CONFIG_USB_XHCI_HCD=y
>> configs/ls1021atwr_sdcard_qspi_defconfig:61:CONFIG_USB_XHCI_HCD=y
>> configs/evb-rk3399_defconfig:60:CONFIG_USB_XHCI_HCD=y
>> configs/k2hk_evm_defconfig:43:CONFIG_USB_XHCI_HCD=y
>> configs/k2hk_hs_evm_defconfig:34:CONFIG_USB_XHCI_HCD=y
>> configs/k2e_evm_defconfig:43:CONFIG_USB_XHCI_HCD=y
>> configs/ls1021atwr_sdcard_ifc_defconfig:54:CONFIG_USB_XHCI_HCD=y
>> configs/k2l_evm_defconfig:43:CONFIG_USB_XHCI_HCD=y
>>
>> So it looks we have lots of conversion work to be done by many board
>> maintainers. I am not sure how to proceed on this.
>
>
> Marek reminded me, that he thinks that most of these platforms
> above will most likely select DM_USB implicitly via Kconfig.
>
> I did a quick check and it seems that at least these platforms
> have DM_USB enabled:
>
> ARCH_ZYNQ
> ARCH_ZYNQMP
> ARCH_UNIPHIER
> ARCH_ROCKCHIP
>
> and other from above very likely as well.
>
> Before you invest more time and effort into implementing the xHCI
> additions in a "non-DM cleaner way", I suggest to find out which
> targets really use xHCI without USB_DM. An easy check would be to
> add some #error to the non-DM part and run this commit through
> buildman / travis.
>
> What do you think?

Ah, that's really a good idea. Thanks for the hints! I will launch a
buildman testing soon.

Regards,
Bin