Re: [Bug #11342] Linux 2.6.27-rc3: kernel BUG at mm/vmalloc.c - bisected

[Mike Travis]

Linus Torvalds wrote:
> 
> On Mon, 25 Aug 2008, Linus Torvalds wrote:
>> checkstack.pl 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
>>
>> 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.
> 
> In fact, they contain multiple CPU-masks, each 4k-bits - 512 bytes - in 
> size. And they tend to call each other.
> 
> Quite frankly, I don't think we were really ready for 4k CPU's. I'm going 
> to commit this patch to make sure others don't do that many CPU's by 
> mistake. It marks MAXCPU's as being 'broken' so you cannot select it, and 
> also limits the number of CPU's that you _can_ select to "just" 512.
> 
> Right now, 4k cpu's is known broken because of the stack usage. I'm not 
> willing to debug more of these kinds of stack smashers, they're really 
> nasty to work with. I wonder how many other random failures these have 
> been involved with?
> 
> This patch also makes the ifdef mess in Kconfig much cleaner and avoids 
> duplicate definitions by just conditionally suppressing the question and 
> giving higher defaults. 
> 
> We can enable MAXSMP and raise the CPU limits some time in the future. But 
> that future is not going to be before 2.6.27 - the code simply isn't ready 
> for it.
> 
> The reason I picked 512 CPU's as the limit is that we _used_ to limit 
> things to 255. So it's higher than it used to be, but low enough to still 
> feel safe. Considering that a 4k-bit CPU mask (512 bytes) _almost_ worked, 
> the 512-bit (64 bytes) masks are almost certainly fine.
> 
> Still, sane people should limit their NR_CPUS to 8 or 16 or something like 
> that. Very very few people really need the pain of big NR_CPUS. Not even 
> "just" 512 CPU's.
> 
> Travis, Ingo and Thomas cc'd, since they were involved in the original 
> commit (1184dc2ffe2c8fb9afb766d870850f2c3165ef25) that raised the limit.
> 
> 		Linus

Hi Linus,

[Sorry for the long winded response, but I felt that sufficient background
is needed to address this issue.... YOMV :-)]

The need to allow distros to set NR_CPUS=4096 (and NODES_SHIFT=9) is
critical to our upcoming SGI systems using what we have been calling
"UV".  This system will be capable of supporting 4096 cpu threads in a
single system image (and 16k cpus/2k nodes is right around the corner).
While obviously I cannot divulge too many details, it's sufficient to
say there are customers who not only require this extended capability,
but are extremely excited about it.

But the nature of some of these system environments is that they will
not accept a specially built kernel, but only a kernel that has been
built and certified (both from the application standpoint as well as
the security standpoint) by standard distributions.  And you probably
know how extensively these distributions test and certify for many known
defects and absolutely require that incoming source changes come from
the community supported source bases, primarily yours.

Due to the lead time required to accomplish these certifications,
the version of the distributions that will be available when this
system releases will be based on 2.6.27.  (They will allow patches
"post-2.6.27-rc.final" as long as those are committed in the source base.)
The two distributions that SGI supports for our customers is SLES
(SUSE Linux Enterprise Server) and RHEL (Red Hat Enterprise Linux).
[They, of course, are free to run any OS of their choosing, but SGI only
provides front line support for those two.]

I started last August to begin analyzing how to accomplish the above goals
and where exactly are the hot spots in the kernel that would require
attention.  It quickly became clear that cpumask_t and nodemask_t are
two variables that are very casually used (along with NR_CPUS), because
the assumption was that 64 "was more than sufficient" for an upper limit
and even extending it to 128 or 255 (254 was the maximum IPI broadcast
ID until x2apic), only added a few more bytes here and there.

I chose not to introduce too many dramatic changes and instead analyzed
every instance where cpumask_t and NR_CPUS was being used (along with the
node counterparts.)  An initial proposal was to allow the default stack
size to be increased, but this was met with a lot of objections because
of the extensive work that was done to bring it to it's current size.

So in summary, the goals of the changes that I have been making since
last October are:

	1. Allow a "typical distro" configured kernel with NR_CPUS=4096
	   and NODES_SHIFT=9 to be booted on an x86_64 system with 2GB
	   of memory.  (Some thought was given to use a 512Mb laptop
	   as the base, but because of other memory bloat from using a
	   64-bit kernel, that was considered not very useful.)

	   [Note I frequently use an allyes and allmod config for
	   testing.]

	2. To lessen as much as possible, the impact on memory usage for
	   that same kernel on that same system.

	3. To lessen as much as possible, the impact on system performance
	   for that same kernel on that same system.  [Which mostly
	   depended on #2.]

I booted the first 4096 cpu kernel last February, and since around March
or April, Ingo has been (build and boot) testing the x86 branch using
"MAXSMP" to trigger the increased defaults quite extensively (IIRC, it
was somewhere between 75% and 90% of all kernels built.)  We here at SGI
nightly build 4 trees (linux-2.6, linux-next, linux-mm, linux-x86) to
insure new checkins don't conflict with changes we've made in the past.
Unfortunately, our run testing wasn't sufficient to catch this latest
error (and I will be quickly fixing that.)

I will also revisit all the past areas to analyze if there have been
other abuses of stack and memory space added since the 4k cpu limit
was "certified" as usable and releasable.  (See below for an initial
survey of size increases between a 512cpu/64node configuration and a
4096cpu/512node configuration.)

So perhaps "MAXSMP" is not needed (or perhaps should be more hidden to
reduce accidental uses), but allowing the defaults listed above to be in
the standard x86/Kconfig insures that the distros can at least attempt
certification with the maximally configured kernels for their enterprise
editions of Linux.

There are many more changes that will be proposed for the 2.6.28 window.
Most certainly your concerns, as well as others, about how to change the
current "cpumask paradigm" to be more easily manageable for systems with
huge cpu counts, will be visited.  (And surely be well discussed. :-)

Thanks,
Mike
---

linux-2.6: v2.6.27-rc4-176-gb8e6c91

====== Data (-l 500)
... files 2 vars 1421 all 0 lim 500 unch 0

    1 - 512-64-allmodconfig
    2 - 4096-512-allmodconfig

      .1.       .2.    ..final..
  1671168  +3899392 5570560  +233%  irq_desc(.data.cacheline_aligned)
   591872  +3899392 4491264  +658%  irq_cfg(.data.read_mostly)
    76800   +537600  614400  +700%  early_node_map(.init.data)
    66176   +462336  528512  +698%  init_mem_cgroup(.bss)
    65536   +458752  524288  +700%  boot_pageset(.bss)
    63648   +419328  482976  +658%  kmalloc_caches(.data.cacheline_aligned)
    15328    +61376   76704  +400%  def_root_domain(.bss)
    10240    +43008   53248  +420%  change_point_list(.init.data)
     8760      +504    9264    +5%  init_task(.data)
     8192    +57344   65536  +700%  kgdb_info(.bss)
     6404    +26880   33284  +419%  e820_saved(.bss)
     6404    +26880   33284  +419%  e820(.bss)
     6400    +26880   33280  +420%  new_bios(.init.data)
     5120    +35840   40960  +700%  node_devices(.bss)
     5120    +21504   26624  +420%  change_point(.init.data)
     4160    +29120   33280  +700%  cpu_bit_bitmap(.rodata)
     4096    +28672   32768  +700%  __cpu_pda(.init.data)
     3776    +25088   28864  +664%  hstates(.bss)
     3584    +25088   28672  +700%  bootmem_node_data(.init.data)
     2560    +10752   13312  +420%  overlap_list(.init.data)
     2048    +14336   16384  +700%  x86_cpu_to_node_map_early_map(.init.data)
     2048    +14336   16384  +700%  was_in_debug_nmi(.bss)
     2048    +14336   16384  +700%  rio_devs(.init.data)
     2048    +14336   16384  +700%  passive_cpu_wait(.bss)
     2048    +14336   16384  +700%  node_memblk_range(.init.data)
     2048    +14336   16384  +700%  ints(.init.data)
     2048    +14336   16384  +700%  cpu_in_kgdb(.bss)
     1024     +7168    8192  +700%  x86_cpu_to_apicid_early_map(.init.data)
     1024     +7168    8192  +700%  x86_bios_cpu_apicid_early_map(.init.data)
     1024     +1024    2048  +100%  pxm_to_node_map(.data)
     1024     +7168    8192  +700%  nodes_add(.bss)
     1024     +7168    8192  +700%  nodes(.init.data)
      512     +3584    4096  +700%  zone_movable_pfn(.init.data)
      512     +3584    4096  +700%  tvec_base_done(.data)
      512     +3584    4096  +700%  scal_devs(.init.data)
      512     +3584    4096  +700%  node_data(.data.read_mostly)
      512     +3584    4096  +700%  memblk_nodeid(.init.data)
        0     +2048    2048      .  node_to_pxm_map(.data)
        0     +2048    2048      .  node_order(.bss)
        0     +2048    2048      .  node_load(.bss)
        0     +2048    2048      .  fake_node_to_pxm_map(.init.data)
        0      +768     768      .  rcu_ctrlblk(.data)
        0      +768     768      .  rcu_bh_ctrlblk(.data)
        0      +768     768      .  per_cpu__cpu_info(.data.percpu)
        0      +768     768      .  boot_cpu_data(.data.read_mostly)
        0      +760     760      .  per_cpu__phys_domains(.data.percpu)
        0      +760     760      .  per_cpu__node_domains(.data.percpu)
        0      +760     760      .  per_cpu__cpu_domains(.data.percpu)
        0      +760     760      .  per_cpu__core_domains(.data.percpu)
        0      +760     760      .  per_cpu__allnodes_domains(.data.percpu)
        0      +720     720      .  top_cpuset(.data)
        0      +640     640      .  per_cpu__flush_state(.data.percpu)
        0      +632     632      .  pit_clockevent(.data)
        0      +632     632      .  per_cpu__lapic_events(.data.percpu)
        0      +632     632      .  lapic_clockevent(.data)
        0      +632     632      .  hpet_clockevent(.data)
        0      +616     616      .  net_dma(.data)
        0      +579     579      .  do_migrate_pages(.text)
        0      +568     568      .  irq2(.data)
        0      +568     568      .  irq0(.data)
        0      +528     528      .  per_cpu__sched_group_phys(.data.percpu)
        0      +528     528      .  per_cpu__sched_group_cpus(.data.percpu)
        0      +528     528      .  per_cpu__sched_group_core(.data.percpu)
        0      +528     528      .  per_cpu__sched_group_allnodes(.data.percpu)
        0      +520     520      .  out_of_memory(.text)
        0      +520     520      .  nohz(.data)
        0      +512     512      .  tick_broadcast_oneshot_mask(.bss)
        0      +512     512      .  tick_broadcast_mask(.bss)
        0      +512     512      .  prof_cpu_mask(.data)
        0      +512     512      .  per_cpu__local_cpu_mask(.data.percpu)
        0      +512     512      .  per_cpu__cpu_sibling_map(.data.percpu)
        0      +512     512      .  per_cpu__cpu_core_map(.data.percpu)
        0      +512     512      .  nohz_cpu_mask(.bss)
        0      +512     512      .  mce_device_initialized(.bss)
        0      +512     512      .  mce_cpus(.bss)
        0      +512     512      .  marked_cpus(.bss)
        0      +512     512      .  kmem_cach_cpu_free_init_once(.bss)
        0      +512     512      .  irq_default_affinity(.data)
        0      +512     512      .  frozen_cpus(.bss)
        0      +512     512      .  fallback_doms(.bss)
        0      +512     512      .  cpu_singlethread_map(.data.read_mostly)
        0      +512     512      .  cpu_sibling_setup_map(.bss)
        0      +512     512      .  cpu_present_map(.data.read_mostly)
        0      +512     512      .  cpu_possible_map(.bss)
        0      +512     512      .  cpu_populated_map(.data.read_mostly)
        0      +512     512      .  cpu_online_map(.data.read_mostly)
        0      +512     512      .  cpu_mask_none(.bss)
        0      +512     512      .  cpu_mask_all(.data.read_mostly)
        0      +512     512      .  cpu_isolated_map(.bss)
        0      +512     512      .  cpu_initialized(.data)
        0      +512     512      .  cpu_callout_map(.bss)
        0      +512     512      .  cpu_callin_map(.bss)
        0      +512     512      .  cpu_active_map(.bss)
        0      +512     512      .  cache_dev_map(.bss)
        0      +512     512      .  c1e_mask(.bss)
        0      +512     512      .  backtrace_mask(.bss)
 2647360 +10283499 12930859 +388%  Totals

====== Sections (-l 500)
... files 2 vars 36 all 0 lim 500 unch 0

    1 - 512-64-allmodconfig
    2 - 4096-512-allmodconfig

       .1.        .2.    ..final..
  66688274  +10345296 77033570   +15%  Total
  38237848     +44031 38281879    <1%  .debug_info
   8441752   +1215872  9657624   +14%  .bss
   2551715      +3136  2554851    <1%  .text
   1737600   +4318720  6056320  +248%  .data.cacheline_aligned
   1640096      +6784  1646880    <1%  .data.percpu
   1175061     +29104  1204165    +2%  .rodata
   1073400     +13712  1087112    +1%  .debug_abbrev
    901760      +1392   903152    <1%  .debug_ranges
    608192   +3906016  4514208  +642%  .data.read_mostly
    302704     +13504   316208    +4%  .data
    244896    +792112  1037008  +323%  .init.data
123603298 +20689679 144292977  +16%  Totals

====== Text/Data ()
... files 2 vars 6 all 0 lim 0 unch 0

    1 - 512-64-allmodconfig
    2 - 4096-512-allmodconfig

      .1.       .2.    ..final..
  2551808     +2048  2553856    <1%  TextSize
  1679360    +43008  1722368    +2%  DataSize
  8441856  +1216512  9658368   +14%  BssSize
  2138112   +798720  2936832   +37%  InitSize
  1640448     +6144  1646592    <1%  PerCPU
  2383872  +8228864 10612736  +345%  OtherSize
18835456 +10295296 29130752  +54%  Totals

====== PerCPU ()
... files 2 vars 22 all 0 lim 0 unch 0

    1 - 512-64-allmodconfig
    2 - 4096-512-allmodconfig

   .1.    .2.    ..final..
  2048  -2048    .  -100%  vm_event_states
  2048  -2048    .  -100%  softnet_data
  2048  -2048    .  -100%  init_sched_rt_entity
  2048  -2048    .  -100%  core_domains
     0  +2048 2048      .  sched_group_core
     0  +2048 2048      .  node_domains
     0  +2048 2048      .  lru_add_active_pvecs
     0  +2048 2048      .  init_rt_rq
     0  +2048 2048      .  cpu_domains
     0  +2048 2048      .  cpu_core_map
     0  +2048 2048      .  cpu_buffer
 8192 +6144 14336  +75%  Totals

====== Stack (-l 500)
... files 2 vars 126 all 0 lim 500 unch 0

    1 - 512-64-allmodconfig
    2 - 4096-512-allmodconfig

  .1.    .2.    ..final..
    0  +2712 2712      .  smp_call_function_mask
    0  +1576 1576      .  setup_IO_APIC_irq
    0  +1576 1576      .  move_task_off_dead_cpu
    0  +1560 1560      .  arch_setup_ht_irq
    0  +1560 1560      .  __assign_irq_vector
    0  +1544 1544      .  tick_handle_oneshot_broadcast
    0  +1544 1544      .  msi_compose_msg
    0  +1440 1440      .  cpuset_write_resmask
    0  +1352 1352      .  store_scaling_governor
    0  +1352 1352      .  cpufreq_add_dev
    0  +1320 1320      .  cpufreq_update_policy
    0  +1312 1312      .  store_scaling_min_freq
    0  +1312 1312      .  store_scaling_max_freq
    0  +1176 1176      .  threshold_create_device
    0  +1128 1128      .  setup_IO_APIC
    0  +1096 1096      .  sched_balance_self
    0  +1080 1080      .  sched_rt_period_timer
    0  +1080 1080      .  _cpu_down
    0  +1064 1064      .  set_ioapic_affinity_irq
    0  +1048 1048      .  store_interrupt_enable
    0  +1048 1048      .  setup_timer_IRQ0_pin
    0  +1048 1048      .  setup_ioapic_dest
    0  +1048 1048      .  set_msi_irq_affinity
    0  +1048 1048      .  set_ht_irq_affinity
    0  +1048 1048      .  native_machine_crash_shutdown
    0  +1048 1048      .  native_flush_tlb_others
    0  +1048 1048      .  dmar_msi_set_affinity
    0  +1040 1040      .  store_threshold_limit
    0  +1040 1040      .  show_error_count
    0  +1040 1040      .  acpi_map_lsapic
    0  +1032 1032      .  tick_do_periodic_broadcast
    0  +1032 1032      .  sched_setaffinity
    0  +1032 1032      .  native_send_call_func_ipi
    0  +1032 1032      .  local_cpus_show
    0  +1032 1032      .  local_cpulist_show
    0  +1032 1032      .  irq_select_affinity
    0  +1032 1032      .  irq_complete_move
    0  +1032 1032      .  irq_affinity_proc_write
    0  +1032 1032      .  flush_tlb_mm
    0  +1032 1032      .  flush_tlb_current_task
    0  +1032 1032      .  fixup_irqs
    0  +1032 1032      .  create_irq
    0  +1024 1024      .  uv_vector_allocation_domain
    0  +1024 1024      .  uv_send_IPI_allbutself
    0  +1024 1024      .  store_error_count
    0  +1024 1024      .  physflat_send_IPI_allbutself
    0  +1024 1024      .  pci_bus_show_cpuaffinity
    0  +1024 1024      .  move_masked_irq
    0  +1024 1024      .  flush_tlb_page
    0  +1024 1024      .  flat_send_IPI_allbutself
    0   +784  784      .  sd_init_ALLNODES
    0   +776  776      .  sd_init_SIBLING
    0   +776  776      .  sd_init_NODE
    0   +768  768      .  sd_init_MC
    0   +768  768      .  sd_init_CPU
    0   +728  728      .  update_flag
    0   +696  696      .  init_intel_cacheinfo
    0   +680  680      .  __build_sched_domains
    0   +648  648      .  thread_return
    0   +648  648      .  schedule
    0   +640  640      .  cpuset_attach
    0   +616  616      .  rebalance_domains
    0   +600  600      .  select_task_rq_fair
    0   +600  600      .  cache_add_dev
    0   +584  584      .  shmem_getpage
    0   +568  568      .  pdflush
    0   +552  552      .  tick_notify
    0   +552  552      .  partition_sched_domains
    0   +552  552      .  free_sched_groups
    0   +552  552      .  __percpu_alloc_mask
    0   +544  544      .  taskstats_user_cmd
    0   +536  536      .  sched_init_smp
    0   +536  536      .  pci_device_probe
    0   +536  536      .  cpuset_common_file_read
    0   +536  536      .  cpupri_find
    0   +536  536      .  acpi_processor_ffh_cstate_probe
    0   +536  536      .  __cpu_disable
    0   +520  520      .  uv_send_IPI_all
    0   +520  520      .  tick_do_broadcast
    0   +520  520      .  smp_call_function
    0   +520  520      .  show_related_cpus
    0   +520  520      .  show_affected_cpus
    0   +520  520      .  prof_cpu_mask_write_proc
    0   +520  520      .  physflat_send_IPI_mask
    0   +520  520      .  physflat_send_IPI_all
    0   +520  520      .  native_smp_send_reschedule
    0   +520  520      .  native_send_call_func_single_ipi
    0   +520  520      .  lapic_timer_broadcast
    0   +520  520      .  irq_set_affinity
    0   +520  520      .  flat_vector_allocation_domain
    0   +520  520      .  flat_send_IPI_all
    0   +520  520      .  find_lowest_rq
    0   +520  520      .  cpuset_can_attach
    0   +520  520      .  cpu_callback
    0   +520  520      .  compat_sys_sched_setaffinity
    0   +520  520      .  add_del_listener
    0   +512  512      .  sys_sched_setaffinity
    0   +512  512      .  sys_sched_getaffinity
    0   +512  512      .  run_rebalance_domains
    0   +512  512      .  ioapic_retrigger_irq
    0   +512  512      .  generic_processor_info
    0   +512  512      .  force_quiescent_state
    0   +512  512      .  destroy_irq
    0   +512  512      .  default_affinity_write
    0   +512  512      .  cpu_to_phys_group
    0   +512  512      .  cpu_to_allnodes_group
    0   +512  512      .  compat_sys_sched_getaffinity
    0   +512  512      .  check_preempt_curr_rt
    0   +512  512      .  assign_irq_vector
   0 +92248 92248   +0%  Totals

====== MemInfo ()
... files 0 vars 0 all 0 lim 0 unch 0

(runtime meminfo not collected.)
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/