On 11/29/24 5:52 PM, John David Anglin wrote:
Hi Helge,
Hi Dave,
I have done a lot of work on gcc in the past month or so trying to resolve issues with LRA.
Thanks!
Sometimes cc1plus will hang during bootstrap. This usually occurs fairly soon after the stage1 compiler is built. Sometimes it hangs doing conf tests or building libstdc++ for stage2. top shows hung process running at 100%. Once a hang occurs, it seems system is more likely to hang again (i.e., the gdb trick only lasts for a few compiles before another hang occurs). Maybe hang is more likely after system has been running for sometime. There is no output from strace if it is attached to processor. If I attach gdb to hung process, I see cc1plus is usually in parser. If I continue process or just quit, process runs successfully to completion. I suspect this problem is somehow related to the size of cc1plus. If I build stage1 with -O1, the problem is reduced or goes away. I've never seen a hang building stage 1 or with the stage2 compiler. Could this be a swap problem? Or maybe trace bit gets turned on so process runs really slow?
I doubt that's caused by swap/trace bits.
Any other thoughts?
Which kernel is affected? I suspect some issue with kernel timer (and thus scheduler). You could try reverting this patch: "parisc: Convert to generic clockevents" https://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux.git/commit/?h=parisc-6.12-1&id=b5ff52be891347f8847872c49d7a5c2fa29400a7 Do you still need to press enter so that systemd continues boot? Might be related. I see some 100% hangs with apt-cacher-ng (again) on one of the machines at SAP, although apt-cacher-ng has my latest patches applied. Helge
