Hi all, On 08/17/2018 10:00 PM, Frank Scheiner wrote:
On 08/17/2018 09:35 PM, John Paul Adrian Glaubitz wrote:On 08/17/2018 09:23 PM, Frank Scheiner wrote:My question now is, to which `linux-image[...]` package version should I actually file the bug report?You should rather test an upstream kernel and report the bug to bugzilla.kernel.org.Ok, thanks for the info, will do that instead.
An update on this issue:As mentioned earlier `4.17.0-rc3` from Debian works. The other still available `4.17-rc[...]` from Debian was `4.17.0-rc7` and this version doesn't work.
I tried to further narrow down when the problem was introduced and did some bisecting using the vanilla kernel sources with `4.17-rc3` as known good and `4.17-rc7` as known bad. Unfortunately the results are somewhat inconclusive.
I could determine, that the kernel config plays into this issue because when doing `make localmodconfig` and compiling kernels, the resulting kernels always worked, even with `ipr` module loaded and even for the initially assumed known bad `4.17-rc7`. But with a kernel config derived (i.e. maintainer key configuration deactivated) from the original `4.17.0-rc3` Debian config the resulting kernels show the same issue as the `4.17.0-rc7`. But to my astonishment also for the assumed `4.17-rc3`, so the vanilla `4.17-rc3` shows the issue but the Debian `4.17.0-rc3` does not. ?-/
In addition compilation with the Debian config is rather time consuming and takes about 85 minutes for kernel and modules, then add 10 to 15 minutes for testing startup three times, as I sometimes saw the kernels working correctly exactly one time but not later.
To rule out a hardware problem I also tested on two other POWER5(+) based machines, an Intellistation POWER 285 - which is more or less identical to the p5 520Q except for the CPU - and an older revision p5 520 with POWER5 and DDR memory instead of POWER5+ and DDR2. So far I could reproduce the exact same behaviour there, too. So it could of course still be that all machines have the exact same defect, but I consider that rather unlikely.
The question now is, how to proceed? Maybe the best approach would be to continue with the bisecting of the vanilla kernel sources with the Debian config and start with `4.17-rc3` as known bad and some older revision as known good. I hope I won't loose too much time determining a good version that's reasonably close to the bad version.
Any other ideas or suggestions? Cheers, Frank