On Fri, 24 May 2019 at 15:54, Eduardo Valentin <edubezval@xxxxxxxxx> wrote: > > Hello, > > On Fri, May 24, 2019 at 10:23:09AM +0200, Tomeu Vizoso wrote: > > On Fri, 24 May 2019 at 04:40, Eduardo Valentin <edubezval@xxxxxxxxx> wrote: > > > > > > On Thu, May 23, 2019 at 11:46:47AM +0200, Tomeu Vizoso wrote: > > > > Hi Eduardo, > > > > > > > > I saw that for 5.1 [0] you included a kernelci boot report for your > > > > tree, but not for 5.2. Have you found anything that should be improved > > > > in KernelCI for it to be more useful to maintainers like you? > > > > > > Honestly, I take a couple of automated testing as input before sending > > > my pulls to Linux: (a) my local test, (b) kernel-ci, and (c) 0-day. > > > > > > There was really no reason specifically for me to not add the report > > > from kernelci, except.. > > > > > > > > [0] https://lore.kernel.org/lkml/20190306161207.GA7365@localhost.localdomain/ > > > > > > > > I found about this when trying to understand why the boot on the > > > > veyron-jaq board has been broken in 5.2-rc1. > > > > > > > > > > I remember a report saying this failed, but from what I could tell from > > > the boot log, the board booted and hit terminal. But apparently, after > > > all reports from developers, the veyron-jaq boards were in a hang state. > > > > > > That was hard for me to tell from your logs, as they looked like > > > a regular boot that hits terminal. > > > > > > Maybe I should have looked for a specific output of a command you guys > > > run, saying "successful boot" somewhere? > > > > I think what is easiest and clearest is to consider the bisection > > reports as a very strong indication that something is quite wrong in > > the branch. > > OK. I hear you. > > > > > Because if a board stopped booting and the bisection found a > > suspicious patch, and reverting it gets the board booting again, then > > chances are very high that the patch in question broke that boot. > > > > > Yeah, for sure If I had understood the report properly I could have > nacked the patch. > > > Do you think the wording could be improved to make it clearer? Or > > maybe some other changes to make all this more useful to maintainers > > like you? > > > > Well, from my perspective, I need to judge if the failure on your report > is really related to my changes. Many times, specially on build errors, > we get failures that are unrelated. Build errors are more straight > forward do judge. Similarly, we need to find out if a boot issue is > caused by a change on the branch or something existing. On boot issues > from kernelci reports, I think the false negatives I have been seeing > is lab/boards failing to boot. Those can also be easy to spot as the > in most cases the kernel wont even load. > > For this particular case, as I described before, the kernel would > load and hit the shell command line, but in fact it was in hang state > IIRC. That is probably why it has not straight forward to understand > from the log. Maybe a successful boot message somewhere would have > helped to spot the problem (or the opposite of it, something > saying, I was expecting to execute a command and board was > unresponsive). Ah, I think I see now what you meant. In the boot log linked below, near the end we have: 22:14:41.981401 ShellCommand command timed out.: Sending # in case of corruption. Connection timeout 00:04:25, retry in 00:02:12 22:14:42.083594 # The # character is sent by the LAVA machine that the DUT is connected to, in the hope that a userspace shell would reply with something. But the kernel failed to boot to userspace, so we have this at the end: 22:16:54.558087 depthcharge-retry failed: 1 of 1 attempts. 'auto-login-action timed out after 285 seconds' 22:16:54.560479 depthcharge-action failed: 1 of 1 attempts. 'auto-login-action timed out after 285 seconds' 22:16:54.855023 JobError: Your job cannot terminate cleanly. https://storage.kernelci.org/evalenti/for-kernelci/v5.1-rc6-58-gbe827ffd38ea/arm/multi_v7_defconfig/gcc-8/lab-collabora/boot-rk3288-veyron-jaq.html Do you think that's the source of the confusion? Thanks, Tomeu
