Re: xhci patch 20150623

[Nick Hudson <skrll%netbsd.org@localhost>]

On 06/23/15 07:43, takahiro hayashi wrote:
> Hello,
>
> Here is xhci patch for nick-nhusb branch.
>
>
> nh-xhci-check.diff
>     + Add port range check in xhci_rhpsc().
>     + Add sanity check if xfer->ux_pipe != NULL in xhci_handle_event().
> nh-xhci-prc.diff
>     + Remove SET_FEATURE C_PORT_RESET(PRC) -- what would happen?
> nh-xhci-methods1.diff
>     + Sync return type of xhci_close_pipe() with return type of
>       upm_close method.
> nh-xhci-abort.diff
>     + Add routines for aborting xfer and command.
> nh-xhci-comments.diff
>     + Update comments.
> nh-usb_subr.diff
>     + Don't give up doing SET_CONFIG in usbd_set_config_index()
>       even if it fails to get BOS descriptor.
>       In this case ud_bdesc will be NULL.
Thanks, I'll look them over.

> Known problems:
>
> + HS hub in 3.0 hub under 3.0 port is disconnected and reconnected every
>    several minutes repeatedly.
>    I don't know what is culprit yet.
> + Detaching hubs or devices might cause panic.
>    Especially when the hub that hangs many devices is disconnected.
This is a general problem that has been around forever. I have a plan.

> + KASSERT(sc->sc_command_addr == 0) in xhci_do_command() may fail.
>    If two or more threads run xhci_do_command(), all of them except one
>    should be blocked at mutex_enter. But one of blocked thread can get
>    sc_lock with sc_command_addr != 0 when cv_timedwait() unlocks sc_lock.
> + xhci.c cannot handle cross-64k transfer yet.
>    (It works anyway though.)
> + Power management is not implemented.
> + USB keyboard interaction is laggy and annoying.

Any idea why keyboard is laggy?
> + ASM1042 and some Intel PCH do not work at all.
>    No interrupts.
> + Fresco1100 does not report CSC if the device is connected at boot.
>    Only PortResetChange is set in change bits.
>    uhub ignores ports without CSC bit, so cannot detect devices.
> + SS part of asm107x hub under hudson2 xhci roothub is not recognized.
>    It drops Port Enable Disable (PED) bit after port reset.
>    rhpsc: 0x21203<CSC,PIC=0x0,XSPEED=0x4=SS,PP,PLS=0x0=U0,PED,CCS>
>    after reset: 
> 0x6202c0<PLC,PRC,CSC,PIC=0x0,XSPEED=0x0=NONE,PP,PLS=0x6=SS_INA>

Not sure what you mean by "hudson2 xhci roothub" here.


> + axen does not work after interface up -> down -> up.
>    When the interface goes down, axen driver closes both of RX and TX 
> pipes.
>    That causes transtion slot state of this device to ADDRESSED from 
> CONFIGURED.
> + xhci.c does not snoop SET_CONFIG request and does not transtion the
>    slot state to CONFIGURED from ADDRESSED even if SET_CONFIG is issued.
> + Isochronous transfer is not implemented.
> + Stream protocol is not supported.
> + Conexant CX93010 umodem is not recognized (XACT in ADDRESS_DEVICE).
>    It can be recognized by inserting 150ms delay before port reset
>    while enumeration.
> + usbd_clear_endpoint_stall{,_async}() does not work on xhci to clear 
> stall
>    condition because this function does not issue RESET_ENDPOINT command.
>    However, xhci.c detects whether xfer is stalled and issues 
> RESET_ENDPOINT
>    automatically.
> + Address of root hub is 0 (but it works anyway).
> + Not sure it work on big endian machines.
>
> + usbdevs(8) does not report correctly if num of ports > 16.
>    Size of udi_ports in struct usb_device_info should be
>    expanded from 16 to 256. Needs compat treatment.
>    (currently usbdevs is not part of nick-nhusb branch.)
I can change this :)

Nick