[lkml]   [2008]   [Aug]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [Bug #11342] Linux 2.6.27-rc3: kernel BUG at mm/vmalloc.c - bisected
Linus Torvalds wrote:
> On Mon, 25 Aug 2008, Linus Torvalds wrote:
>> But I'll look at your vmlinux, see what stands out.
> Oops. I already see the problem.
> Your .config has soem _huge_ CPU count, doesn't it?
> shows these things as the top problems:
> 0xffffffff80266234 smp_call_function_mask [vmlinux]: 2736
> 0xffffffff80234747 __build_sched_domains [vmlinux]: 2232
> 0xffffffff8023523f __build_sched_domains [vmlinux]: 2232
> 0xffffffff8021e884 setup_IO_APIC_irq [vmlinux]: 1616
> 0xffffffff8021ee24 arch_setup_ht_irq [vmlinux]: 1600
> 0xffffffff8021f144 arch_setup_msi_irq [vmlinux]: 1600
> 0xffffffff8021e3b0 __assign_irq_vector [vmlinux]: 1592
> 0xffffffff8021e626 __assign_irq_vector [vmlinux]: 1592
> 0xffffffff8023257e move_task_off_dead_cpu [vmlinux]: 1592
> 0xffffffff802326e8 move_task_off_dead_cpu [vmlinux]: 1592
> 0xffffffff8025dbc5 tick_handle_oneshot_broadcast [vmlinux]:1544
> 0xffffffff8025dcb4 tick_handle_oneshot_broadcast [vmlinux]:1544
> 0xffffffff803f3dc4 store_scaling_governor [vmlinux]: 1376
> 0xffffffff80279ef4 cpuset_write_resmask [vmlinux]: 1360
> 0xffffffff803f465d cpufreq_add_dev [vmlinux]: 1352
> 0xffffffff803f495b cpufreq_add_dev [vmlinux]: 1352
> 0xffffffff803f3fc4 store_scaling_max_freq [vmlinux]: 1328
> 0xffffffff803f4064 store_scaling_min_freq [vmlinux]: 1328
> 0xffffffff803f44c4 cpufreq_update_policy [vmlinux]: 1328
> ..
> and sys_init_module is actually way way down the list. I bet the only
> reason it showed up at all was because dynamically it was such a deep
> callchain, and part of that callchain probably called some of those really
> nasty things.
> Anyway, the reason smp_call_function_mask and friends have such _huge_
> stack usages for you is that they contain a 'cpumask_t' on the stack.
> For example, for me, usign a sane NR_CPU, the size of the stack frame for
> smp_call_function_mask is under 200 bytes. For you, it's 2736 bytes.
> How about you make CONFIG_NR_CPU's something _sane_? Like 16? Or do you
> really have four thousand CPU's in that system?
> Oh, I guess you have the MAXSMP config enabled? I really think that was a
> bit too aggressive.
> Linus

This probably all started when I was working on a software tool (aiod)
that was failing because somebody ELSE had 4,096 CPUs configured.
[[Seems that gcc had/has? it's MAX CPU value set to 1,024 (bits/sched.h
__CPU_SETSIZE), so when you issue system calls like sched_getaffinity,
it will "fail" for systems configured w/ 4,096 CPUs. I worked around it
by simply forgetting about the gcc values, and kept allocating larger
CPU masks until it worked.]]

I think you're right: the kernel as a whole may not be ready for 4,096
CPUs apparently...

Thanks for taking the time to look into this...


 \ /
  Last update: 2008-08-25 23:33    [W:0.194 / U:14.912 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site