Re: [PATCH v12 3/6] usb: dwc3: Resize TX FIFOs to meet EP bursting requirements

[Greg KH]

On Tue, Jul 06, 2021 at 01:19:38PM -0700, Wesley Cheng wrote:
> 
> 
> On 7/6/2021 11:13 AM, Greg KH wrote:
> > On Fri, Jul 02, 2021 at 02:37:32AM -0700, Wesley Cheng wrote:
> >> Some devices have USB compositions which may require multiple endpoints
> >> that support EP bursting.  HW defined TX FIFO sizes may not always be
> >> sufficient for these compositions.  By utilizing flexible TX FIFO
> >> allocation, this allows for endpoints to request the required FIFO depth to
> >> achieve higher bandwidth.  With some higher bMaxBurst configurations, using
> >> a larger TX FIFO size results in better TX throughput.
> >>
> >> By introducing the check_config() callback, the resizing logic can fetch
> >> the maximum number of endpoints used in the USB composition (can contain
> >> multiple configurations), which helps ensure that the resizing logic can
> >> fulfill the configuration(s), or return an error to the gadget layer
> >> otherwise during bind time.
> >>
> >> Signed-off-by: Wesley Cheng <wcheng@xxxxxxxxxxxxxx>
> >> ---
> >>  drivers/usb/dwc3/core.c   |   9 ++
> >>  drivers/usb/dwc3/core.h   |  15 ++++
> >>  drivers/usb/dwc3/ep0.c    |   2 +
> >>  drivers/usb/dwc3/gadget.c | 221 ++++++++++++++++++++++++++++++++++++++++++++++
> >>  4 files changed, 247 insertions(+)
> >>
> >> diff --git a/drivers/usb/dwc3/core.c b/drivers/usb/dwc3/core.c
> >> index e0a8e79..a7bcdb9d 100644
> >> --- a/drivers/usb/dwc3/core.c
> >> +++ b/drivers/usb/dwc3/core.c
> >> @@ -1267,6 +1267,7 @@ static void dwc3_get_properties(struct dwc3 *dwc)
> >>  	u8			rx_max_burst_prd;
> >>  	u8			tx_thr_num_pkt_prd;
> >>  	u8			tx_max_burst_prd;
> >> +	u8			tx_fifo_resize_max_num;
> >>  	const char		*usb_psy_name;
> >>  	int			ret;
> >>  
> >> @@ -1282,6 +1283,8 @@ static void dwc3_get_properties(struct dwc3 *dwc)
> >>  	 */
> >>  	hird_threshold = 12;
> >>  
> >> +	tx_fifo_resize_max_num = 6;
> >> +
> 
> Hi Greg,
> > 
> > No comment as to why 6 was picked, like the other defaults in this
> > function?
> > 
> > Why was 6 picked?
> > 
> > 
> Talked with Thinh about this sometime back about why 6 was picked.  It
> was just an arbitrary setting we decided on throughout our testing, as
> that was what provided the best tput numbers for our system.  Hence why
> it was suggested to have a separate property, so other vendors can set
> this to accommodate their difference in HW latencies.

My point is, this needs to be documented!!!

Right now it just looks like you made up a magic number here.  Look at
the other defaults in this function right above this line.  Those
comments explain what is happening, unlike your change.

Do you want to look at this code in 3 years and wonder why this number
was picked?

> >>  	dwc->maximum_speed = usb_get_maximum_speed(dev);
> >>  	dwc->max_ssp_rate = usb_get_maximum_ssp_rate(dev);
> >>  	dwc->dr_mode = usb_get_dr_mode(dev);
> >> @@ -1325,6 +1328,10 @@ static void dwc3_get_properties(struct dwc3 *dwc)
> >>  				&tx_thr_num_pkt_prd);
> >>  	device_property_read_u8(dev, "snps,tx-max-burst-prd",
> >>  				&tx_max_burst_prd);
> >> +	dwc->do_fifo_resize = device_property_read_bool(dev,
> >> +							"tx-fifo-resize");
> >> +	device_property_read_u8(dev, "tx-fifo-max-num",
> >> +				&tx_fifo_resize_max_num);
> > 
> > So you overwrite the "max" with whatever is given to you?  What if
> > tx-fifo-resize is not enabled?
> >
> If tx-fifo-resize is not enabled, then there shouldn't be anything that
> will reference this property.  As mentioned in the previous comment, HW
> vendors may not need a FIFO size of 6 max packets for their particular
> system, so they should be able to program this to their needs.


That's fine, but is that what is really happening here?  You are not
looking at the "do_fifo_resize" value before you try to read the "max"
value, why not?

> If someone programs to this a large number, the logic works where it
> will allocate based off the space left after ensuring enough space for 1
> FIFO per ep.

Where is that documented or happening?

thanks,

greg k-h