On Wed, Aug 22, 2018 at 11:14:19AM -0700, Guenter Roeck wrote:
Various mips64 and ppc64 qemu tests crash as follows in v4.14.y and v4.17.y (the log is from ppc64).
------------[ cut here ]------------ kernel BUG at kernel/time/hrtimer.c:1673! Oops: Exception in kernel mode, sig: 5 [#1] BE NUMA CoreNet Generic Modules linked in: CPU: 0 PID: 1 Comm: init Not tainted 4.17.19-rc1-00309-g8fe1830 #1 NIP: c000000000085d6c LR: c00000000089d840 CTR: c00000000000cd00 REGS: c00000003e1e7990 TRAP: 0700 Not tainted (4.17.19-rc1-00309-g8fe1830) MSR: 000000008002b000 <CE,EE,FP,ME> CR: 48000284 XER: 00000000 SOFTE: 0 GPR00: c00000000089d7ec c00000003e1e7c10 c000000000cb9c00 c00000003e1e8238 GPR04: c00000003e1e7c80 ffffffffffffffff 000000003b9aca00 0000000000000000 GPR08: 0000000031012c01 0000000031012c01 0000000000000002 0000000031012c01 GPR12: 0000000028000482 c000000000d35000 0000000000000000 0000000000000000 GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR24: 0000000000000000 0000000000000000 0000000000000016 00000000ffff9008 GPR28: c00000003e1e7e10 c00000003e1e8000 0000000000000000 000000009336eabb NIP [c000000000085d6c] .nanosleep_copyout+0x4c/0x50 LR [c00000000089d840] .do_nanosleep+0x160/0x190 Call Trace: [c00000003e1e7c10] [c00000000089d7ec] .do_nanosleep+0x10c/0x190 (unreliable) [c00000003e1e7cc0] [c000000000085e78] .hrtimer_nanosleep+0x108/0x1d0 [c00000003e1e7da0] [c000000000086068] .__se_compat_sys_nanosleep+0x78/0xb0 [c00000003e1e7e30] [c000000000000618] system_call+0x58/0x64 Instruction dump: 7c832378 e8890010 4bffbadd 60000000 38210070 e8010010 2fa30000 3940fff2 3860fdfc 7c63579e 7c0803a6 4e800020 <0fe00000> 7c0802a6 fb81ffe0 fbc1fff0 ---[ end trace 15c7fbc119007c42 ]---
I started to bisect, but abandoned it after finding commit 62d7ce7f40a9 ("posix-timers: Fix nanosleep_copyout() for CONFIG_COMPAT_32BIT_TIME") in both branches. Since there is no "config COMPAT_32BIT_TIME" in v4.14.y or v4.17.y, some relevant code is commented out by the commit, which in turn results in the crash.
That patch is already gone from my patch queue, let me push out new -rc git trees to help make it easier for you.
thanks,
greg k-h