Lockdep reports the warning below when Li tries to offline one cpu:
[ 110.835487] =================================
[ 110.835616] [ INFO: inconsistent lock state ]
[ 110.835688] 2.6.30-rc4-00336-g8c9ed89 #52
[ 110.835757] ---------------------------------
[ 110.835828] inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage.
[ 110.835908] swapper/0 [HC1[1]:SC0[0]:HE0:SE1] takes:
[ 110.835982] (cmci_discover_lock){?.+...}, at: [<ffffffff80236dc0>] cmci_clear+0x30/0x9b
cmci_clear() can be called via smp_call_function_single().
It is better to disable interrupt while holding cmci_discover_lock,
to turn it into an irq-safe lock - we can deadlock otherwise.
[ Impact: fix possible deadlock in the MCE code ]
Reported-by: Shaohua Li <shaohua.li@intel.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
LKML-Reference: <4A03ED38.8000700@jp.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Reported-by: Shaohua Li<shaohua.li@intel.com>
The earlier patch to change the poller to a separate function subtly
broke the boot logging logic. This could lead to machine checks
getting logged at boot even when disabled or defaulting to off
on some systems. Fix that.
[ Impact: bug fix - avoid spurious MCE in log ]
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Reviewed-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
The polling timer while running per CPU still uses a global next_interval
variable, which lead to some CPUs either polling too fast or too slow.
This was not a serious problem because all errors get picked up eventually,
but it's still better to avoid it. Turn next_interval into a per cpu variable.
v2: Fix check_interval == 0 case (Hidetoshi Seto)
[ Impact: minor bug fix ]
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Reviewed-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Attempting to rid us of the problematic work_on_cpu(). Just use
smp_call_function_single() here.
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
LKML-Reference: <20090318042217.EF3F1DDF39@ozlabs.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: Bug fix on UP
Referring commit cc3ca22063,
Peter removed __cpuinit annotations for mce_cpu_features()
and its successor functions, which caused troubles on UP
configurations.
However the intel_init_cmci() was introduced after that and
it also has __cpuinit annotation even though it is called from
mce_cpu_features(). Remove the annotation from that function
too.
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: reduce kernel memory usage when CONFIG_CPUMASK_OFFSTACK=y
Simple conversion of mce_device_initialized to cpumask_var_t. We don't
check the alloc_cpumask_var() return since it's boot-time only, and
the misc_register() in that same function isn't checked.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Impact: reduce per-cpu size for CONFIG_CPUMASK_OFFSTACK=y
In most places it's cleaner to use the accessors cpu_sibling_mask()
and cpu_core_mask() wrappers which already exist.
I couldn't avoid cleaning up the access in oprofile, either.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Impact: saving power _very_ little
round_jiffies() round up absolute jiffies to full second.
round_jiffies_relative() round up relative jiffies to full second.
The "t->expires" is absolute jiffies. Then, round_jiffies() should be
used instead round_jiffies_relative().
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: H. Peter Anvin <hpa@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: build fix
The APIC code rewrite in the x86 tree broke the x86/mce branch:
arch/x86/kernel/cpu/mcheck/threshold.c: In function ‘mce_threshold_interrupt’:
arch/x86/kernel/cpu/mcheck/threshold.c:24: error: implicit declaration of function ‘ack_APIC_irq’
Also tidy up the file a bit while at it.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: Bug fix on UP
The MCE code is reinitialized from resume, so we can't use
__cpuinit/__cpuexit for most of the code. Remove those annotations
for anything downstream of mce_init().
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: Major new feature
Intel CMCI (Corrected Machine Check Interrupt) is a new
feature on Nehalem CPUs. It allows the CPU to trigger
interrupts on corrected events, which allows faster
reaction to them instead of with the traditional
polling timer.
Also use CMCI to discover shared banks. Machine check banks
can be shared by CPU threads or even cores. Using the CMCI enable
bit it is possible to detect the fact that another CPU already
saw a specific bank. Use this to assign shared banks only
to one CPU to avoid reporting duplicated events.
On CPU hot unplug bank sharing is re discovered. This is done
using a thread that cycles through all the CPUs.
To avoid races between the poller and CMCI we only poll
for banks that are not CMCI capable and only check CMCI
owned banks on a interrupt.
The shared banks ownership information is currently only used for
CMCI interrupts, not polled banks.
The sharing discovery code follows the algorithm recommended in the
IA32 SDM Vol3a 14.5.2.1
The CMCI interrupt handler just calls the machine check poller to
pick up the machine check event that caused the interrupt.
I decided not to implement a separate threshold event like
the AMD version has, because the threshold is always one currently
and adding another event didn't seem to add any value.
Some code inspired by Yunhong Jiang's Xen implementation,
which was in term inspired by a earlier CMCI implementation
by me.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Define a per cpu bitmap that contains the banks polled by the machine
check poller. This is needed for the CMCI code in the next patches
to be able to disable polling on specific banks.
The bank by default contains all banks, so there is no behaviour
change. Only future code will remove some banks from the polling
set.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: behavior change, use common code
Use a standard leaky bucket ratelimit for the machine check
warning print interval instead of waiting every check_interval.
Also decrease the limit to twice per minute.
This interacts better with threshold interrupts because
they can happen more often than check_interval.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: minor bugfix
The threshold handler on AMD (and soon on Intel) could be theoretically
reentered by the hardware. This could lead to corrupted events
because the machine check poll code assumes it is not reentered.
Move the APIC ACK to the end of the interrupt handler to let
the hardware avoid that.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: cleanup; preparation for feature
The mce_amd_64 code has an own private MC threshold vector with an own
interrupt handler. Since Intel needs a similar handler
it makes sense to share the vector because both can not
be active at the same time.
I factored the common APIC handler code into a separate file which can
be used by both the Intel or AMD MC code.
This is needed for the next patch which adds an Intel specific
CMCI handler.
This patch should be a nop for AMD, it just moves some code
around.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: Cleanup (code movement)
Move MAX_NR_BANKS into mce.h because it's needed there
for followup patches.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: Bug fix when CPU hotplug is disabled
Correct the following broken __cpuinit/__cpuexit annotations:
- mce_cpu_features() is called from mce_resume(), and so cannot be
__cpuinit.
- mce_disable_cpu() and mce_reenable_cpu() are called from
mce_cpu_callback(), and so cannot be __cpuexit().
Cc: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Impact: Bug fix on UP
Checkin 6ec68bff3c:
x86, mce: reinitialize per cpu features on resume
introduced a call to mce_cpu_features() in the resume path, in order
for the MCE machinery to get properly reinitialized after a resume.
However, this function (and its successors) was flagged __cpuinit,
which becomes __init on UP configurations (on SMP suspend/resume
requires CPU hotplug and so this would not be seen.)
Remove the offending __cpuinit annotations for mce_cpu_features() and
its successor functions.
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: Cleanup
The standard spelling of a printf pattern for long long is "ll", not
"L", which is for long double.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Impact: cleanup, performance enhancement
The machine check poller is diverging more and more from the fatal
exception handler. Instead of adding more special cases separate the code
paths completely. The corrected poll path is actually quite simple,
and this doesn't result in much code duplication.
This makes both handlers much easier to read and results in
cleaner code flow. The exception handler now only needs to care
about uncorrected errors, which also simplifies the handling of multiple
errors. The corrected poller also now always runs in standard interrupt
context and does not need to do anything special to handle NMI context.
Minor behaviour changes:
- MCG status is now not cleared on polling.
- Only the banks which had corrected errors get cleared on polling
- The exception handler only clears banks with errors now
v2: Forward port to new patch order. Add "uc" argument.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Impact: cleanup
This merely factors out duplicated code to set up
the initial struct mce state into a single function.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Impact: cleanup; making code future proof; memory saving on small systems
This patch replaces the hardcoded max number of machine check banks with
dynamic allocation depending on what the CPU reports. The sysfs
data structures and the banks array are dynamically allocated.
There is still a hard bank limit (128) because the mcelog protocol uses
banks >= 128 as pseudo banks to escape other events. But we expect
that 128 banks is beyond any reasonable CPU for now.
This supersedes an earlier patch by Venki, but it solves the problem
more completely by making the limit fully dynamic (up to the 128
boundary).
This saves some memory on machines with less than 6 banks because
they won't need sysdevs for unused ones and also allows to
use sysfs to control these banks on possible future CPUs with
more than 6 banks.
This is an updated patch addressing Venki's comments. I also added in
another patch from Thomas which fixed the error allocation path (that
patch was previously separated)
Cc: Venki Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Impact: bugfix
Considering the situation as follow:
before: mcelog.next == 1, mcelog.entry[0].finished = 1
+--------------------------------------------------------------------------
R W1 W2 W3
read mcelog.next (1)
mcelog.next++ (2)
(working on entry 1,
finished == 0)
mcelog.next = 0
mcelog.next++ (1)
(working on entry 0)
mcelog.next++ (2)
(working on entry 1)
<----------------- race ---------------->
(done on entry 1,
finished = 1)
(done on entry 1,
finished = 1)
To fix the race condition, a cmpxchg loop is added to mce_read() to
ensure no new MCE record can be added between mcelog.next reading and
mcelog.next = 0.
Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: Lower priority bug fix
Offlined CPUs could still get machine checks, but the machine check handler
cannot handle them properly, leading to an unconditional crash. Disable
machine checks on CPUs that are going down.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: bug fix, in this case the resume handler shouldn't run which
avoids incorrectly reenabling machine checks on resume
When MCEs are completely disabled on the command line don't set
up the sysdev devices for them either.
Includes a comment fix from Thomas Gleixner.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: Higher priority bug fix
The machine check poller runs a single timer and then broadcasted an
IPI to all CPUs to check them. This leads to unnecessary
synchronization between CPUs. The original CPU running the timer has
to wait potentially a long time for all other CPUs answering. This is
also real time unfriendly and in general inefficient.
This was especially a problem on systems with a lot of events where
the poller run with a higher frequency after processing some events.
There could be more and more CPU time wasted with this, to
the point of significantly slowing down machines.
The machine check polling is actually fully independent per CPU, so
there's no reason to not just do this all with per CPU timers. This
patch implements that.
Also switch the poller also to use standard timers instead of work
queues. It was using work queues to be able to execute a user program
on a event, but mce_notify_user() handles this case now with a
separate callback. So instead always run the poll code in in a
standard per CPU timer, which means that in the common case of not
having to execute a trigger there will be less overhead.
This allows to clean up the initialization significantly, because
standard timers are already up when machine checks get init'ed. No
multiple initialization functions.
Thanks to Thomas Gleixner for some help.
Cc: thockin@google.com
v2: Use del_timer_sync() on cpu shutdown and don't try to handle
migrated timers.
v3: Add WARN_ON for timer running on unexpected CPU
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: Needed for bug fix in next patch
This relaxes the requirement that mce_notify_user has to run in process
context. Useful for future changes, but also leads to cleaner
behaviour now. Now instead mce_notify_user can be called directly
from interrupt (but not NMI) context.
The work queue only uses a single global work struct, which can be done safely
because it is always free to reuse before the trigger function is executed.
This way no events can be lost.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: low priority bug fix
This removes part of a a patch I added myself some time ago. After some
consideration the patch was a bad idea. In particular it stopped machine check
exceptions during code patching.
To quote the comment:
* MCEs only happen when something got corrupted and in this
* case we must do something about the corruption.
* Ignoring it is worse than a unlikely patching race.
* Also machine checks tend to be broadcast and if one CPU
* goes into machine check the others follow quickly, so we don't
* expect a machine check to cause undue problems during to code
* patching.
So undo the machine check related parts of
8f4e956b31 NMIs are still disabled.
This only removes code, the only additions are a new comment.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: Bug fix
During suspend it is not reliable to process machine check
exceptions, because CPUs disappear but can still get machine check
broadcasts. Also the system is slightly more likely to
machine check them, but the handler is typically not a position
to handle them in a meaningfull way.
So disable them during suspend and enable them during resume.
Also make sure they are always disabled on hot-unplugged CPUs.
This new code assumes that suspend always hotunplugs all
non BP CPUs.
v2: Remove the WARN_ONs Thomas objected to.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: bug fix (with tolerant == 3)
do_exit cannot be called directly from the exception handler because
it can sleep and the exception handler runs on the exception stack.
Use force_sig() instead.
Based on a earlier patch by Ying Huang who debugged the problem.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Impact: Bug fix
This fixes a long standing bug in the machine check code. On resume the
boot CPU wouldn't get its vendor specific state like thermal handling
reinitialized. This means the boot cpu wouldn't ever get any thermal
events reported again.
Call the respective initialization functions on resume
v2: Remove ancient init because they don't have a resume device anyways.
Pointed out by Thomas Gleixner.
v3: Now fix the Subject too to reflect v2 change
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
- make oprofile build
- select X86_X2APIC from X86_UV - it relies on it
- export genapic for oprofile modular build
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
make it simpler, don't need have one extra struct.
v2: fix the sgi_uv build
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Impact: cleanup
APIC definitions aren't needed here. Remove the include and fix
up the fallout.
tj: added include to mce_intel_64.c.
Signed-off-by: Brian Gerst <brgerst@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Impact: Reduce memory usage, use new cpumask API.
Use cpumask_var_t for 'cpus' cpumask in struct threshold_bank and update
remaining old cpumask_t functions to new cpumask API.
Signed-off-by: Mike Travis <travis@sgi.com>
Mark the function local_allocate_threshold_blocks() with __cpuinit,
in order to remove the following section mismatch messages:
WARNING: arch/x86/kernel/cpu/mcheck/built-in.o(.text+0x1363): Section mismatch in reference from the function local_allocate_threshold_blocks() to the function .cpuinit.text:allocate_threshold_blocks()
The function local_allocate_threshold_blocks() references
the function __cpuinit allocate_threshold_blocks().
This is often because local_allocate_threshold_blocks lacks a __cpuinit
annotation or the annotation of allocate_threshold_blocks is wrong.
WARNING: arch/x86/kernel/cpu/built-in.o(.text+0x1def): Section mismatch in reference from the function local_allocate_threshold_blocks() to the function .cpuinit.text:allocate_threshold_blocks()
The function local_allocate_threshold_blocks() references
the function __cpuinit allocate_threshold_blocks().
This is often because local_allocate_threshold_blocks lacks a __cpuinit
annotation or the annotation of allocate_threshold_blocks is wrong.
WARNING: arch/x86/kernel/built-in.o(.text+0xef2b): Section mismatch in reference from the function local_allocate_threshold_blocks() to the function .cpuinit.text:allocate_threshold_blocks()
The function local_allocate_threshold_blocks() references
the function __cpuinit allocate_threshold_blocks().
This is often because local_allocate_threshold_blocks lacks a __cpuinit
annotation or the annotation of allocate_threshold_blocks is wrong.
All the callsites of this function are __cpuinit already, and all the
functions it calls are __cpuinit as well.
Signed-off-by: Leonardo Potenza <lpotenza@inwind.it>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'cpus4096-for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (66 commits)
x86: export vector_used_by_percpu_irq
x86: use logical apicid in x2apic_cluster's x2apic_cpu_mask_to_apicid_and()
sched: nominate preferred wakeup cpu, fix
x86: fix lguest used_vectors breakage, -v2
x86: fix warning in arch/x86/kernel/io_apic.c
sched: fix warning in kernel/sched.c
sched: move test_sd_parent() to an SMP section of sched.h
sched: add SD_BALANCE_NEWIDLE at MC and CPU level for sched_mc>0
sched: activate active load balancing in new idle cpus
sched: bias task wakeups to preferred semi-idle packages
sched: nominate preferred wakeup cpu
sched: favour lower logical cpu number for sched_mc balance
sched: framework for sched_mc/smt_power_savings=N
sched: convert BALANCE_FOR_xx_POWER to inline functions
x86: use possible_cpus=NUM to extend the possible cpus allowed
x86: fix cpu_mask_to_apicid_and to include cpu_online_mask
x86: update io_apic.c to the new cpumask code
x86: Introduce topology_core_cpumask()/topology_thread_cpumask()
x86: xen: use smp_call_function_many()
x86: use work_on_cpu in x86/kernel/cpu/mcheck/mce_amd_64.c
...
Fixed up trivial conflict in kernel/time/tick-sched.c manually
Impact: Remove cpumask_t's from stack.
Simple transition to work_on_cpu(), rather than cpumask games.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Mike Travis <travis@sgi.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Robert Richter <robert.richter@amd.com>
Cc: jacob.shin@amd.com
