On 6/3/2016 3:55 AM, Felipe Balbi wrote: > > Hi, > > John Youn <John.Youn@xxxxxxxxxxxx> writes: >>>>> This reverts back to the original buggy behavior. This will fail when >>>>> a SET_INTERFACE occurs multiple times. >>>>> >>>>> You can run testusb to see the failure: >>>>> testusb -t 9 -c 5000 -s 2048 -a >>>> >>>> I came up with something else for this. It's still unstable and I need >>>> to debug further to figure out where we're wrong. But it seems to me >>>> that we're following databook down to the last comma with patch >>>> below. Note that we're partially reverting your original changes and >>>> adding some extra knowledge about current configuration and interface, >>>> then we only reassign transfer resources if those change. Care to have a >>>> look? >>> >>> here's a version that passes testusb and normal enumeration with g_zero >>> and g_mass_storage. After some experimenting, it seems like we should >>> always MODIFY resource allocation, unless we're doing a SetConfiguration >>> to the same configuration that is already chosen or a SetInterface to >>> the same interface/alt-setting pair that's already being used. This is >>> working for me. Can you test on your end too? >> >> Hi Felipe, >> >> You might be treading down a path we've already visited here :) > > :) > >> It looks like this patch will still fail the case with multiple >> interfaces, with one or more having alt-settings. You will end up in a >> situation where multiple endpoints are assigned the same transfer >> resource which will cause failures. >> >> Unfortunately I don't have an easy test outside of our test >> environment that exposes this. > > oh okay > >> You could try creating a composite device. Interface #1 having a >> single alt setting, interface #2 with multiple alt-settings. Start >> traffic to both interfaces, then start setting different alt-settings >> on interface #2. >> >> The problem is the databook does not cover this condition. The >> documentation will be fixed in 3.20a. >> >> Ravi Babu originally reported this and you can see discussion in this >> thread: >> >> http://marc.info/?l=linux-usb&m=145396682025403&w=2 > > in that case, I'll just wait for your fix. But we really this fixed > during this -rc cycle, so I can get my latest improvements (see > testing/next) merged for v4.8. The previous fix does have a different issue, but not anything to do with End Transfer. I will try to reproduce with your testing/next. Unfortunately I need to work on a different project for the next 2-3 weeks so I don't know if I'll have enough time to look at this for rc. Regards, John -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
