lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
Open Source and information security mailing list archives
| ||
|
Date: Fri, 19 Dec 2014 15:55:23 -0800 From: Linus Torvalds <torvalds@...ux-foundation.org> To: Thomas Gleixner <tglx@...utronix.de> Cc: Dave Jones <davej@...hat.com>, Chris Mason <clm@...com>, Mike Galbraith <umgwanakikbuti@...il.com>, Ingo Molnar <mingo@...nel.org>, Peter Zijlstra <peterz@...radead.org>, Dâniel Fraga <fragabr@...il.com>, Sasha Levin <sasha.levin@...cle.com>, "Paul E. McKenney" <paulmck@...ux.vnet.ibm.com>, Linux Kernel Mailing List <linux-kernel@...r.kernel.org>, Suresh Siddha <sbsiddha@...il.com>, Oleg Nesterov <oleg@...hat.com>, Peter Anvin <hpa@...ux.intel.com> Subject: Re: frequent lockups in 3.18rc4 On Fri, Dec 19, 2014 at 3:14 PM, Thomas Gleixner <tglx@...utronix.de> wrote: > > Now that all looks correct. So there is something else going on. After > staring some more at it, I think we are looking at it from the wrong > angle. > > The watchdog always detects CPU1 as stuck and we got completely > fixated on the csd_wait() in the stack trace on CPU1. Now we have > stack traces which show a different picture, i.e. CPU1 makes progress > after a gazillion of seconds. .. but that doesn't explain why CPU0 ends up always being at that *exact* same instruction in the NMI backtrace. While a fairly tight loop, together with "mmio read is very expensive and synchronizing" would explain it. An MMIO read can easily be as expensive as several thousand instructions. > I think we really need to look at CPU1 itself. Not so fast. Take another look at CPU0. [24998.083577] [<ffffffff810e0d3e>] ktime_get+0x3e/0xa0 [24998.084450] [<ffffffff810e9cd3>] tick_sched_timer+0x23/0x160 [24998.085315] [<ffffffff810daf96>] __run_hrtimer+0x76/0x1f0 [24998.086173] [<ffffffff810e9cb0>] ? tick_init_highres+0x20/0x20 [24998.087025] [<ffffffff810db2e7>] hrtimer_interrupt+0x107/0x260 [24998.087877] [<ffffffff81031a4b>] local_apic_timer_interrupt+0x3b/0x70 [24998.088732] [<ffffffff8179bca5>] smp_apic_timer_interrupt+0x45/0x60 [24998.089583] [<ffffffff8179a0df>] apic_timer_interrupt+0x6f/0x80 [24998.090435] <EOI> [24998.091279] [<ffffffff810da66e>] ? __remove_hrtimer+0x4e/0xa0 [24998.092118] [<ffffffff812c7c7a>] ? ipcget+0x8a/0x1e0 [24998.092951] [<ffffffff812c7c6c>] ? ipcget+0x7c/0x1e0 [24998.093779] [<ffffffff812c8d6d>] SyS_msgget+0x4d/0x70 Really. None of that changed. NONE. The likelihood that we hit the exact same instruction every time? Over a timefraem of more than a minute? The only way I see that happening is (a) NMI is completely buggered, and the backtrace is just random crap that is always the same. Or (b) it's really a fairly tight loop. The fact that you had a hrtimer interrupt happen in the *middle* of __remove_hrtimer() is really another fairly strong hint. That smells like "__remove_hrtimer() has a race with hrtimer interrupts". And that race results in a basically endless loop (which perhaps ends when the HRtimer overflows, in what, a few minutes?) I really don't think you should look at CPU1. Not when CPU0 has such an interesting pattern that you dismissed just because the HPET is making progress. Linus -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@...r.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists