Re: [PATCH v3 2/6] usb:common Separated decoding functions from dwc3 driver.

From: Greg KH
Date: Mon Feb 04 2019 - 09:47:35 EST


On Mon, Feb 04, 2019 at 04:17:00PM +0200, Felipe Balbi wrote:
>
> Hi,
>
> Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> writes:
> > On Thu, Jan 31, 2019 at 11:52:29AM +0000, Pawel Laszczak wrote:
> >> Patch moves some decoding functions from driver/usb/dwc3/debug.h driver
> >> to driver/usb/common/debug.c file. These moved functions include:
> >> dwc3_decode_get_status
> >> dwc3_decode_set_clear_feature
> >> dwc3_decode_set_address
> >> dwc3_decode_get_set_descriptor
> >> dwc3_decode_get_configuration
> >> dwc3_decode_set_configuration
> >> dwc3_decode_get_intf
> >> dwc3_decode_set_intf
> >> dwc3_decode_synch_frame
> >> dwc3_decode_set_sel
> >> dwc3_decode_set_isoch_delay
> >> dwc3_decode_ctrl
> >>
> >> These functions are used also in inroduced cdns3 driver.
> >>
> >> All functions prefixes were changed from dwc3 to usb.
> >
> > Ick, why?
>
> moving dwc3-specific code into generic code.

That says what it does, but not why :)

And if this really was just moving things around, why was only one
symbol needed to be exported and not all of them?

> >> + * @bRequestType: matches the USB bmRequestType field
> >> + * @bRequest: matches the USB bRequest field
> >> + * @wValue: matches the USB wValue field (CPU byte order)
> >> + * @wIndex: matches the USB wIndex field (CPU byte order)
> >> + * @wLength: matches the USB wLength field (CPU byte order)
> >> + *
> >> + * Function returns decoded, formatted and human-readable description of
> >> + * control request packet.
> >> + *
> >> + * Important: wValue, wIndex, wLength parameters before invoking this function
> >> + * should be processed by le16_to_cpu macro.
> >> + */
> >> +const char *usb_decode_ctrl(char *str, __u8 bRequestType, __u8 bRequest,
> >> + __u16 wValue, __u16 wIndex, __u16 wLength);
> >
> > Why are you returning a value, isn't the data stored in str? Why not
> > just return an int saying if the call succeeded or not?
>
> There is one detail here. The usage scenario for this is for
> tracepoints. When dealing with tracepoints we want to delay decoding of
> the data into strings until print-time. I guess it's best to illustrate
> with an example:
>
> DECLARE_EVENT_CLASS(dwc3_log_ctrl,
> TP_PROTO(struct usb_ctrlrequest *ctrl),
> TP_ARGS(ctrl),
> TP_STRUCT__entry(
> __field(__u8, bRequestType)
> __field(__u8, bRequest)
> __field(__u16, wValue)
> __field(__u16, wIndex)
> __field(__u16, wLength)
> __dynamic_array(char, str, DWC3_MSG_MAX)
> ),
> TP_fast_assign(
> __entry->bRequestType = ctrl->bRequestType;
> __entry->bRequest = ctrl->bRequest;
> __entry->wValue = le16_to_cpu(ctrl->wValue);
> __entry->wIndex = le16_to_cpu(ctrl->wIndex);
> __entry->wLength = le16_to_cpu(ctrl->wLength);
> ),
> TP_printk("%s", dwc3_decode_ctrl(__get_str(str), DWC3_MSG_MAX,
> __entry->bRequestType,
> __entry->bRequest, __entry->wValue,
> __entry->wIndex, __entry->wLength)
> )
> );
>
> The above is the code is today (well, I've added buffer size as an
> argument). If I make dwc3_decode_ctrl() return an integer, I can't call
> it from TP_printk() time. I'd have to move it to TP_fast_assign() time
> which is supposed to be, simply, a copy of the necessary data. IOW, I
> would have this:
>
> DECLARE_EVENT_CLASS(dwc3_log_ctrl,
> TP_PROTO(struct usb_ctrlrequest *ctrl),
> TP_ARGS(ctrl),
> TP_STRUCT__entry(
> __dynamic_array(char, str, DWC3_MSG_MAX)
> ),
> TP_fast_assign(
> dwc3_decode_ctrl(__get_str(str), DWC3_MSG_MAX,
> ctrl->bRequestType,
> ctrl->bRequest,
> le16_to_cpu(ctrl->wValue),
> le16_to_cpu(ctrl->wIndex),
> le16_to_cpu(ctrl->wLength));
> ),
> TP_printk("%s", __get_str(str)
> )
> );
>
> Essentially, we end up moving decoding of the tracepoint to the time it
> is captured; IOW, we reintroduce regular latency of string formatting.
>
> What we could do, is move all functions called by dwc3_decode_ctrl() to
> return int, but leave dwc3_decode_ctrl() returning a pointer to str just
> so we continue decoding the data at printing time.

Ok, it wasn't obvious that this was used in a tracepoint like this, that
makes more sense.

So, it should be documented as well :)

thanks,

greg k-h