System freezes only when reboot. Userspace seems to finish OK, disks are unmounted cleanly. Another user reports kernel panic, not in my case. Simply rolling back to previous versions (3.0.45 or 3.5.7) the problem dissapears. Tested on two machines, one with CPU AMD Phenom II X4, the other a single-core processor. Only occurs on the multicore. Reviewing the changelog for both kernels I could see a commit related to behavior on reboot (f96972f2dc6365421cf2366ebd61ee4cf060c8d5 upstream) kernel/sys.c: call disable_nonboot_cpus() in kernel_restart()
Recompiling kernel/sys.c in 3.0.46 without the call to disable_nonboot_cpus() at the function kernel_restart() makes my system reboots normally. Otherwise, kernel_restart() freezes my machine.
Same behavior on 3.0.48, kernel_restart() freezes system if call disable_nonboot_cpus() Only occurs on 4-core CPU. Tested on single-core, no freeze.
Same machine (4-core AMD), now running Ubuntu 12.04, until kernel version 3.2.33, normal behavior, after update to kernel 3.2.34 the system freezes on reboot. As we can see in the 3.2.34 patch, the call to disable_nonboot_cpus() was just added to kernel_restart() in sys.c: http://www.kernel.org/diff/diffview.cgi?file=%2Fpub%2Flinux%2Fkernel%2Fv3.0%2Fpatch-3.2.34.bz2;z=1402
References: http://bugs.debian.org/697635 First-bad-commit: f96972f2dc6365421cf2366ebd61ee4cf060c8d5
I was starting kernel with "acpi_skip_timer_override" parameter, otherwise I could not enable "AMD C1E" state in BIOS (system freezes when booting). This option saves about 10W of power at idle for me. If I disable AMD C1E, system works perfectly (but consumming more power at idle). "AMD C1E" + "acpi_skip_timer_override" = system works ok except the commented freeze when reboot. "AMD C1E" + "lapic" + "nolapic_timer" = testing, but system seems to work OK (and kernel_restart works) Is "AMD C1E state" supported by current kernels? The debian bug... an Intel CPU.
Said to be fixed by cf7df378aa4ff7da3a44769b7ff6e9eef1a9f3db (reboot: rigrate shutdown/reboot to boot cpu, 2013-06-12). Thanks to Nicolas Le Cam for the update. Looks like the bugzilla configuration has changed so I can't claim the bug and close it any more?