Hi, Jean-Jacques Hiblot <jjhiblot@xxxxxx> writes: >>> Jean-Jacques Hiblot <jjhiblot@xxxxxx> writes: >>>> Hi Felipe, >>>> >>>> While trying out the v4.9 kernel on a DRA7x, I stumbled upon this >>>> crash. I haven't tried to bisect yet and probably won't be do it >>>> before next year. >>>> >>>> This crash doesn't occur with a 4.4 kernel. I suspect it has to do >>>> with the changes in the TRB preparation and interrupt handling. >>>> >>>> I don't see the issue if qmult isn't increased. I suspect it's still >>>> there, just harder to reproduce. >>> can't help you without tracepoints: >>> >>> # mkdir -p /t >>> # mount -t tracefs none /t >>> # cd /t >>> # echo 81920 > buffer_size_kb >>> # echo 1 > events/dwc3/enable >>> >>> (reproduce) >>> >>> # cp trace ~/trace.txt >>> >>> send me trace.txt, you probably need to compress it. >> nevermind, found it. I just pushed a branch called >> dwc3-fix-v4.9-NULL-pointer-deref to my k.org tree. Tested here and I can >> still see my 850 MBits/sec with MTU 1500. When setting MTU to 10000, >> then I got 1.15 Gbit/sec on iperf. >> >> Let me know your results. I'll send patches out after Xmas. > The results are quite good. > > default qmult + MTU 1500 : 430 MBits/s > qmult=20 + MTU 1500 : 500-660 MBits/s. There is a lot of variability > there but it seems to be due to the migration of the iperf process > (using taskset to attach iperf to a specific core helps a lot) > default qmult + MTU 10000 : 2.3 GBits/s. the qmult parameter here has > little effect (+10-20 MBits/s). > > Thank you. cool, thanks for testing. I'll try messing around with taskset on my side too. Interesting finding ;-) -- balbi -- 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
