[U-Boot] [PATCH v6 2/4] usb: ehci: add weak-aliased functions to portsc & tdi

[Marek Vasut]

Dear Kuo-Jung Su,

> 2013/5/15 Kuo-Jung Su <dantesu at gmail.com>:
> > 2013/5/14 Marek Vasut <marex at denx.de>:
> >> Dear Kuo-Jung Su,
> >> 
> >>> 2013/5/13 Marek Vasut <marex at denx.de>:
> >>> > Dear Kuo-Jung Su,
> >>> > 
> >>> >> 2013/5/13 Marek Vasut <marex at denx.de>:
> >>> >> > Dear Kuo-Jung Su,
> >>> >> > 
> >>> >> >> From: Kuo-Jung Su <dantesu at faraday-tech.com>
> >>> >> >> 
> >>> >> >> There is at least one non-EHCI compliant controller (i.e. Faraday
> >>> >> >> EHCI) known to implement a non-standard TDI stuff.
> >>> >> >> Futhermore, it not only leave reserved and CONFIGFLAG registers
> >>> >> >> un-implemented but also has their address spaces removed.
> >>> >> >> 
> >>> >> >> And thus, we need weak-aliased functions to both TDI stuff
> >>> >> >> and PORTSC registers for interface abstraction.
> >>> >> >> 
> >>> >> >> Signed-off-by: Kuo-Jung Su <dantesu at faraday-tech.com>
> >>> >> >> CC: Marek Vasut <marex at denx.de>
> >>> >> >> ---
> >>> >> >> 
> >>> >> >> Changes for v6:
> >>> >> >>    - Simplify weak aliased function declaration
> >>> >> >>    - Drop redundant line feed
> >>> >> >> 
> >>> >> >> Changes for v5:
> >>> >> >>    - Split up from Faraday EHCI patch
> >>> >> >> 
> >>> >> >> Changes for v2 - v4:
> >>> >> >>    - See 'usb: ehci: add Faraday USB 2.0 EHCI support'
> >>> >> >>  
> >>> >> >>  drivers/usb/host/ehci-hcd.c |   91
> >>> >> >> 
> >>> >> >> ++++++++++++++++++++++++++----------------- 1 file changed, 55
> >>> >> >> insertions(+), 36 deletions(-)
> >>> >> >> 
> >>> >> >> diff --git a/drivers/usb/host/ehci-hcd.c
> >>> >> >> b/drivers/usb/host/ehci-hcd.c index c816878..ae3f2a4 100644
> >>> >> >> --- a/drivers/usb/host/ehci-hcd.c
> >>> >> >> +++ b/drivers/usb/host/ehci-hcd.c
> >>> >> >> @@ -117,10 +117,44 @@ static struct descriptor {
> >>> >> >> 
> >>> >> >>  };
> >>> >> >>  
> >>> >> >>  #if defined(CONFIG_EHCI_IS_TDI)
> >>> >> >> 
> >>> >> >> -#define ehci_is_TDI()        (1)
> >>> >> >> -#else
> >>> >> >> -#define ehci_is_TDI()        (0)
> >>> >> >> +# define ehci_is_TDI()       (1)
> >>> >> > 
> >>> >> > btw you can remove those braces around (1) and (0) below. But I
> >>> >> > have one more question ...
> >>> >> 
> >>> >> Got it, thanks
> >>> >> 
> >>> >> > [...]
> >>> >> > 
> >>> >> >> @@ -609,13 +644,10 @@ ehci_submit_root(struct usb_device *dev,
> >>> >> >> unsigned long pipe, void *buffer, uint32_t *status_reg;
> >>> >> >> 
> >>> >> >>       struct ehci_ctrl *ctrl = dev->controller;
> >>> >> >> 
> >>> >> >> -     if (le16_to_cpu(req->index) >
> >>> >> >> CONFIG_SYS_USB_EHCI_MAX_ROOT_PORTS) { -             printf("The
> >>> >> >> request port(%d) is not configured\n", -
> >>> >> >> 
> >>> >> >>                  le16_to_cpu(req->index) - 1);
> >>> >> >> 
> >>> >> >> +     status_reg = ehci_get_portsc_register(ctrl->hcor,
> >>> >> >> +             le16_to_cpu(req->index) - 1);
> >>> >> >> +     if (!status_reg)
> >>> >> > 
> >>> >> > What happens here if req->index is zero ?
> >>> >> > 
> >>> >> > Hint: the above code always does unsigned comparison ...
> >>> >> > 
> >>> >> > I think you should make the second argument of
> >>> >> > ehci_get_portsc_register() unsigned short too (as is req->index in
> >>> >> > struct devrequest).
> >>> >> 
> >>> >> Sorry, but I'll prefer 'int' over 'unsigned short', since it looks
> >>> >> to me that the u-boot would set 'req->index' to 0 at startup, which
> >>> >> results in a 'port = -1' to be passed to
> >>> >> ehci_get_portsc_register().
> >>> >> 
> >>> >> And I think '-1' is a better self-explain value, so I'd like to
> >>> >> stick with 'int'
> >>> > 
> >>> > Sure, but then the comparison is signed, not unsigned. Besides, it's
> >>> > unnecessary change to the logic of the code. Or did I miss something
> >>> > ?
> >>> 
> >>> 1. There is a bug in ehci_submit_root() of usb ehci:
> >>>     int ehci_submit_root()
> >>>     {
> >>>     
> >>>          ......
> >>>          if (port > CONFIG_SYS_USB_EHCI_MAX_ROOT_PORTS) {
> >>>          
> >>>             printf("The request port(%d) is not configured\n", port -
> >>>             1); return -1;
> >>>          
> >>>          }
> >>>          status_reg = (uint32_t *)&ctrl->hcor->or_portsc[port - 1];
> >>>          ......
> >>>     
> >>>     }
> >>>     
> >>>     The 'port' is actually a '0' at start-up, so we actually accessed
> >>> 
> >>> a wrong register.
> >>> 
> >>>     But fortunately the wrong register actually points to
> >>>     CONFIGFLAG(0x40)
> >>> 
> >>> with a safe value for the following codes.
> >>> 
> >>> 2. One of Vivek Gautam's usb patches has altered the logic of usb host
> >>> 
> >>>     upon launching 'usb start', if we report a error upon (port - 1 <
> >>>     0), the current u-boot usb would failed to scan ports. (At least
> >>>     it
> >>> 
> >>> failed at Faraday platforms.)
> >>> 
> >>>     However it looks to me that it's o.k to report a error upon (port
> >>> 
> >>> - 1 < 0) at old usb ehci stack.
> >>> 
> >>>     (i.e. 10 days ago, in master branch of u-boot)
> >>> 
> >>> And thus I add a quick check to PATCH v7.
> >>> 
> >>> __weak uint32_t *ehci_get_portsc_register(struct ehci_hcor *hcor, int
> >>> port) {
> >>> 
> >>>  /*
> >>>  
> >>>   * The u-boot would somehow set port=-1 at usb start-up,
> >>>   * so this quick fix is necessary.
> >>>   */
> >>>  
> >>>  if (port < 0)
> >>>  
> >>>   port = 0;
> >> 
> >> Maybe we should return fail, no ?
> > 
> > No, it would make the 'usb start' to terminate immediately,
> > and results in a port scan failure to at least Faraday EHCI.
> > 
> >> Can you pinpoint where does the req->index
> >> (resp. port) get set to -1 ?
> > 
> > Later I'll try to find out where we have 'req->index' set as a '0' in
> > 'usb start'.
> 
> It's from usb_new_device() --> usb_get_descriptor(), and thus
> it's definitely correct to set index = 0 right here.
> The only problem is 'We shall not always report an portsc error for
> all request!'
> Please refer to the patch attached at the tail of this mail.
> 
> >> And which commit introduced this breakage ?
> > 
> > I believe it's there long ago, we just fortunately bypass the error at
> > old day, and now one of Vivek Gautam's USB patch make us face up to this
> > issue.
> 
> Shame on me....
> It's nothing to do with Vivek Gautam's USB patch or the old USB ehci stack,
> it's only mattered while reporting an portsc error. :P
> 
> >> Best regards,
> >> Marek Vasut

Ok, I think I almost see it there. Can you just make a patchset out of this 
stuff so I can digest it a little bit easier? Maybe stick the fixes first so I 
can pick them ASAP and the FHCI stuff after that.