On Thu, Feb 28, 2019 at 11:13 AM Paul Thomas <pthomas8...@gmail.com> wrote: > > On Thu, Feb 28, 2019 at 12:35 AM Harini Katakam <hari...@xilinx.com> wrote: > > > > Hi Paul, > > > > > -----Original Message----- > > > From: Paul Thomas [mailto:pthomas8...@gmail.com] > > > Sent: Thursday, February 28, 2019 3:36 AM > > > To: linuxptp-devel@lists.sourceforge.net > > > Subject: strangeness > > > > > <snip> > > > HW/SW setup: > > > This is on an arm64 system (xilinx zynqmp) using the macb ethernet driver: > > > https://xilinx-wiki.atlassian.net/wiki/spaces/A/pages/18841740/Macb+Driver > > > > Some issues were reported when ptp4l was used in conjunction with SSH and > > while they were not easily reproduced, the errors stopped completely after > > this > > patch in the Xilinx tree: > > https://github.com/Xilinx/linux-xlnx/commit/835b6170df72dc73d9386b5838e7f16c09953b9a#diff-41909d180431659ccc1229aa30fd4e5a > > > > It timestamps only PTP event packets as opposed to all packets (which is > > excessive). > > There are also some BD word related stability fixes but I see that you've > > already > > used mainline 5.0 kernel - so they should already be in there. > > > > Regards, > > Harini > > Hi Harini, > > This is very reproducible now, and with further nc testing it almost > seems like the tx packets aren't getting sent until there is rx > traffic. Let me describe the setup, on my host I run the nc server and > then on the zynqmp board I run the nc client. In the "After" case in > the example below you can see that the c4 line doesn't come to the > server until s7 is sent. In this test sN lines are typed from the > server and cN lines are typed from the client where N is the overall > order. In the "Before" case the overall order is correct. > Before running ptp4l: > zynqmp client side host nc server > $ nc host 4567 $ nc -l 4567 > s1 s1 > s2 s2 > c3 c3 > c4 c4 > s5 s5 > s6 s6 > s7 s7 > > After running ptp4l: > zynqmp client side host nc server > $ nc host 4567 $ nc -l 4567 > s1 s1 > s2 s2 > c3 s5 > c4 s6 > s5 c3 > s6 s7 > s7 c4 > > At first I was thinking it was possibly the tx interrupt, but that > maybe doesn't make sense because in a separate (already established) > ssh session I can generate lot's of traffic in both directions and it > never causes the c3/c4 lines to show up on the server nc. It is not > until the traffic on that actual port. > > thanks, > Paul
Hi Harini, Yes changing it to TSTAMP_ALL_PTP_FRAMES instead of TSTAMP_ALL_FRAMES does seem to fix the ssh issue. My worry is that there is still a bug somewhere in the network stack that this is just masking. It seems like we should be able to instrument the TX path to see why the nc send never makes it to the wire? I guess if the contents were some special marker that might help. thanks, Paul _______________________________________________ Linuxptp-devel mailing list Linuxptp-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/linuxptp-devel
