[PATCH 6/9] phy: ti: tusb1210: Improve ulpi_read()/_write() error checking

[Hans de Goede]

Hi,

On 6/13/22 17:48, Andy Shevchenko wrote:
> +Cc: Ferry
> 
> On Sat, Feb 05, 2022 at 05:45:32PM +0100, Hans de Goede wrote:
>> ulpi_read() and ulpi_write() calls can fail. Add wrapper functions to log
>> errors when this happens and add error checking to the read + write of
>> the phy parameters from the TUSB1210_VENDOR_SPECIFIC2 register.`
> 
> 
> This patch seems to break Intel Merrifield platform.
> 
> Before:
> 
> [   36.333644] xhci-hcd xhci-hcd.1.auto: xHCI Host Controller
> [   36.339828] xhci-hcd xhci-hcd.1.auto: new USB bus registered, assigned bus number 1
> [   36.352566] xhci-hcd xhci-hcd.1.auto: hcc params 0x0220f06c hci version 0x100 quirks 0x0000000002010010
> [   36.367062] xhci-hcd xhci-hcd.1.auto: irq 16, io mem 0xf9100000
> [   36.378429] xhci-hcd xhci-hcd.1.auto: xHCI Host Controller
> [   36.384705] xhci-hcd xhci-hcd.1.auto: new USB bus registered, assigned bus number 2
> 
> After:
> [   32.986776] tusb1210 dwc3.0.auto.ulpi: error -110 writing val 0x41 to reg 0x80
> [   32.994053] phy phy-dwc3.0.auto.ulpi.0: phy poweron failed --> -110
> [   33.000601] dwc3 dwc3.0.auto: error -ETIMEDOUT: failed to initialize core
> [   33.007486] dwc3: probe of dwc3.0.auto failed with error -110
> 
> 
> Any ideas?

In my experience with using the phy for charger-type detection on some
x86 android tablets which don't have any other way to do charger detection,
these errors indicate a real communication issue for reading/writing
phy registers. At the same time this usually does not seem to be a big
problem since the phy seems to work fine with its power-on defaults.

In case of Bay Trail these errors were related to 2 things:

1. Autosuspend of the phy-interface block in the dwc3, fixed by:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=d7c93a903f33ff35aa0e6b5a8032eb9755b00826

But dwc3_pci_mrfld_properties[] already sets "snps,dis_u2_susphy_quirk",
so I guess it is not this.

2. There being no delay in tusb1210_power_on() between toggling the
reset IO and then trying to communicate with the phy, fixed in:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=df37c99815d9e0775e67276d70c93cbc25f31c70

Maybe the:

#define TUSB1210_RESET_TIME_MS				30

Added by that commit needs to be a bit bigger for the possibly
older phy revision used on the merifield boards?

(note it is fine to just increase it a bit everywhere).

Regards,

Hans




 
> 
> P.S> There is no bisect log, since it's done manually with a good guess by
> Ferry. I have just reverted patches on ULPI from this series and start applying
> them one-by-one.
>