Re: [PATCH] usb: devio: update max count of DPs per interval for ISOC

[Greg Kroah-Hartman]

On Sat, Dec 29, 2018 at 10:48:40AM +0800, Chunfeng Yun wrote:
> Hi,
> On Fri, 2018-12-28 at 11:51 +0100, Greg Kroah-Hartman wrote:
> > On Fri, Dec 28, 2018 at 06:27:44PM +0800, Chunfeng Yun wrote:
> > > The host shall be able to accept and send up to 96 DPs for
> > > devices operating at Gen 2 speed.
> > 
> > Why?  What is this keeping from working properly today?
> 
> The failure happened when I tried to send up to 96DPs per an interval
> for SSP ISOC transations by libusb, this is used to verify SSP ISOC
> function of USB3 GEN2 controller. Maybe for normal scenarios they
> needn't transfer greater than 48 DPs in an ISOC interval.
> (refer usb3.1r1.0 section 8.12.6 Isochronous Transactions)

Ok, please add this information in the changelog text please.

> > 
> > I need a much better changelog text here in order to be able to accept
> > this patch.
> I'll make the change in next version
> > 
> > > 
> > > Signed-off-by: Chunfeng Yun <chunfeng.yun@xxxxxxxxxxxx>
> > > ---
> > >  drivers/usb/core/devio.c | 8 +++-----
> > >  1 file changed, 3 insertions(+), 5 deletions(-)
> > > 
> > > diff --git a/drivers/usb/core/devio.c b/drivers/usb/core/devio.c
> > > index a75bc0b8a50f..82c16210e34c 100644
> > > --- a/drivers/usb/core/devio.c
> > > +++ b/drivers/usb/core/devio.c
> > > @@ -1564,12 +1564,10 @@ static int proc_do_submiturb(struct usb_dev_state *ps, struct usbdevfs_urb *uurb
> > >  		}
> > >  		for (totlen = u = 0; u < number_of_packets; u++) {
> > >  			/*
> > > -			 * arbitrary limit need for USB 3.0
> > > -			 * bMaxBurst (0~15 allowed, 1~16 packets)
> > > -			 * bmAttributes (bit 1:0, mult 0~2, 1~3 packets)
> > > -			 * sizemax: 1024 * 16 * 3 = 49152
> > > +			 * arbitrary limit need for USB 3.1 Gen2
> > > +			 * sizemax: 96 DPs at SSP, 96 * 1024 = 98304
> > >  			 */
> > > -			if (isopkt[u].length > 49152) {
> > > +			if (isopkt[u].length > 98304) {
> > 
> > Are we going to have to keep bumping this up as speeds get faster and
> > faster?
> Or remove the check of data length? I'm not sure which way is better

No, we added that check for a good reason, please see the git log for
the details :)

thanks,

greg k-h