Hi Francesco,
> -----Original Message-----
> From: Francesco Dolcini <francesco@xxxxxxxxxx>
> Sent: Friday, October 28, 2022 8:32 PM
> To: Peter Chen <peter.chen@xxxxxxxxxx>; linux-usb@xxxxxxxxxxxxxxx; Jun Li
> <jun.li@xxxxxxx>
> Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>; Shawn Guo
> <shawnguo@xxxxxxxxxx>; Sascha Hauer <s.hauer@xxxxxxxxxxxxxx>; Pengutronix
> Kernel Team <kernel@xxxxxxxxxxxxxx>; Fabio Estevam <festevam@xxxxxxxxx>;
> dl-linux-imx <linux-imx@xxxxxxx>; Felipe Balbi <balbi@xxxxxxxxxx>;
> philippe.schenker@xxxxxxxxxxx; Francesco Dolcini
> <francesco.dolcini@xxxxxxxxxxx>
> Subject: USB runtime PM issues on i.MX6ULL
>
> Hello all,
>
> I am debugging some unexpected USB behavior on a i.MX6ULL SOC,
> chipidea controller ("fsl,imx6ul-usb") and a fsl mxs usbphy
> ("fsl,imx6ul-usbphy").
>
> The HW design has 2 USB interface, the first one is dual-role, while the
> second
> one is a host port with NO way to re-read the VBUS (USB_OTG2_VBUS is not
> really connected, there is just a capacitor to GND).
How is your USB_OTG1_VBUS status? Can you try to make your USB_OTG1_VBUS pad
has a valid VBUS voltage, then run your Host only port test with runtime PM
enabled?
Li Jun
>
> Focusing on the second interface, USB Host only, what I do experience is
> the
> following:
>
> - if there is no USB HUB connected and no device connected at boot, any
> USB
> device hotplugged is not working, ci_runtime_suspend is called and it
> never
> resume.
> - if there is a HUB in between it somehow works, however I have a continuos
> runtime powermanagement reset loop every 2 seconds
> [ 1026.146360] ci_hdrc ci_hdrc.1: at ci_runtime_suspend
> [ 1026.164725] ci_hdrc ci_hdrc.1: at ci_controller_resume
> [ 1026.487844] usb 1-1: reset high-speed USB device number 2 using
> ci_hdrc
> [ 1028.326789] ci_hdrc ci_hdrc.1: at ci_runtime_suspend
> [ 1028.335378] ci_hdrc ci_hdrc.1: at ci_controller_resume
> [ 1028.657853] usb 1-1: reset high-speed USB device number 2 using
> ci_hdrc
>
>
> I was able to have it fully working disabling the runtime power management
> in
> the chipidea driver or using sysfs (`echo on > /sys/.../power/control`).
>
> --- a/drivers/usb/chipidea/ci_hdrc_imx.c
> +++ b/drivers/usb/chipidea/ci_hdrc_imx.c
> @@ -56,7 +56,7 @@ static const struct ci_hdrc_imx_platform_flag
> imx6sx_usb_data = {
> };
>
> static const struct ci_hdrc_imx_platform_flag imx6ul_usb_data = {
> - .flags = CI_HDRC_SUPPORTS_RUNTIME_PM |
> + .flags =
>
> I was digging even more into the topic and I found out that what is happening
> is that mxs_phy_disconnect_line() is called. I than tried to remove the
> MXS_PHY_DISCONNECT_LINE_WITHOUT_VBUS flag.
>
> --- a/drivers/usb/phy/phy-mxs-usb.c
> +++ b/drivers/usb/phy/phy-mxs-usb.c
> @@ -178,7 +178,7 @@ static const struct mxs_phy_data imx6sx_phy_data = {
> };
>
> static const struct mxs_phy_data imx6ul_phy_data = {
> - .flags = MXS_PHY_DISCONNECT_LINE_WITHOUT_VBUS,
> + /*.flags = MXS_PHY_DISCONNECT_LINE_WITHOUT_VBUS,*/
> };
>
> This commit from NXP downstream kernel seems somehow related
> https://github.com/nxp-imx/linux-imx/commit/89ec73836a9b1347743e406d62d
> d446dc4365db3
> however it builds on other commits that allow to communicate the actual mode
> to
> the USB PHY driver and prevent mxs_phy_disconnect_line() to be called for
> the
> USB host case.
>
> With that the situation is way better, but while without
> `CI_HDRC_SUPPORTS_RUNTIME_PM` it works perfectly, without
> `MXS_PHY_DISCONNECT_LINE_WITHOUT_VBUS` it does not really work smootly
> in case there is a hub in between (~20% of the time the device is not
> enumerated after plugging it in).
>
> When it does not work I see that after plugging in a device runtime resume
> function is called, but after that the device is not enumerated on the USB
> bus.
> It looks like something else is missing.
>
> It seems like having a pure USB Host interface without having a way to re-read
> the VBUS is not really supported in SW at the moment, am I wrong?
> Any idea?
>
> One last comment, even the USB dual role port is not working smoothly, however
> I have not investigated this specific case that much because disabling runtime
> pm
> solves everything also in that case.
>
> Francesco
