smp.c and irq_work.c implement the same inline helper. Move it to
apic.h and use it everywhere.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Reschedule vector tracepoints may be called in cpu idle state.
This causes lockdep check warning below.
The tracepoint requires rcu but for accuracy it also
requires irq_enter() (tracepoints record the irq context), thus,
the tracepoint interrupt handler should be calling irq_enter()
and not rcu_irq_enter() (irq_enter() calls rcu_irq_enter()).
So, add irq_enter/exit() to smp_trace_reschedule_interrupt()
with common pre/post processing functions, smp_entering_irq()
and exiting_irq() (exiting_irq() calls just irq_exit()
in arch/x86/include/asm/apic.h),
because these can be shared among reschedule, call_function,
and call_function_single vectors.
[ 50.720557] Testing event reschedule_exit:
[ 50.721349]
[ 50.721502] ===============================
[ 50.721835] [ INFO: suspicious RCU usage. ]
[ 50.722169] 3.10.0-rc6-00004-gcf910e8 #190 Not tainted
[ 50.722582] -------------------------------
[ 50.722915] /c/kernel-tests/src/linux/arch/x86/include/asm/trace/irq_vectors.h:50 suspicious rcu_dereference_check() usage!
[ 50.723770]
[ 50.723770] other info that might help us debug this:
[ 50.723770]
[ 50.724385]
[ 50.724385] RCU used illegally from idle CPU!
[ 50.724385] rcu_scheduler_active = 1, debug_locks = 0
[ 50.725232] RCU used illegally from extended quiescent state!
[ 50.725690] no locks held by swapper/0/0.
[ 50.726010]
[ 50.726010] stack backtrace:
[...]
Signed-off-by: Seiji Aguchi <seiji.aguchi@hds.com>
Reviewed-by: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/51CDCFA3.9080101@hds.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
[Purpose of this patch]
As Vaibhav explained in the thread below, tracepoints for irq vectors
are useful.
http://www.spinics.net/lists/mm-commits/msg85707.html
<snip>
The current interrupt traces from irq_handler_entry and irq_handler_exit
provide when an interrupt is handled. They provide good data about when
the system has switched to kernel space and how it affects the currently
running processes.
There are some IRQ vectors which trigger the system into kernel space,
which are not handled in generic IRQ handlers. Tracing such events gives
us the information about IRQ interaction with other system events.
The trace also tells where the system is spending its time. We want to
know which cores are handling interrupts and how they are affecting other
processes in the system. Also, the trace provides information about when
the cores are idle and which interrupts are changing that state.
<snip>
On the other hand, my usecase is tracing just local timer event and
getting a value of instruction pointer.
I suggested to add an argument local timer event to get instruction pointer before.
But there is another way to get it with external module like systemtap.
So, I don't need to add any argument to irq vector tracepoints now.
[Patch Description]
Vaibhav's patch shared a trace point ,irq_vector_entry/irq_vector_exit, in all events.
But there is an above use case to trace specific irq_vector rather than tracing all events.
In this case, we are concerned about overhead due to unwanted events.
So, add following tracepoints instead of introducing irq_vector_entry/exit.
so that we can enable them independently.
- local_timer_vector
- reschedule_vector
- call_function_vector
- call_function_single_vector
- irq_work_entry_vector
- error_apic_vector
- thermal_apic_vector
- threshold_apic_vector
- spurious_apic_vector
- x86_platform_ipi_vector
Also, introduce a logic switching IDT at enabling/disabling time so that a time penalty
makes a zero when tracepoints are disabled. Detailed explanations are as follows.
- Create trace irq handlers with entering_irq()/exiting_irq().
- Create a new IDT, trace_idt_table, at boot time by adding a logic to
_set_gate(). It is just a copy of original idt table.
- Register the new handlers for tracpoints to the new IDT by introducing
macros to alloc_intr_gate() called at registering time of irq_vector handlers.
- Add checking, whether irq vector tracing is on/off, into load_current_idt().
This has to be done below debug checking for these reasons.
- Switching to debug IDT may be kicked while tracing is enabled.
- On the other hands, switching to trace IDT is kicked only when debugging
is disabled.
In addition, the new IDT is created only when CONFIG_TRACING is enabled to avoid being
used for other purposes.
Signed-off-by: Seiji Aguchi <seiji.aguchi@hds.com>
Link: http://lkml.kernel.org/r/51C323ED.5050708@hds.com
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
When implementing tracepoints in interrupt handers, if the tracepoints are
simply added in the performance sensitive path of interrupt handers,
it may cause potential performance problem due to the time penalty.
To solve the problem, an idea is to prepare non-trace/trace irq handers and
switch their IDTs at the enabling/disabling time.
So, let's introduce entering_irq()/exiting_irq() for pre/post-
processing of each irq handler.
A way to use them is as follows.
Non-trace irq handler:
smp_irq_handler()
{
entering_irq(); /* pre-processing of this handler */
__smp_irq_handler(); /*
* common logic between non-trace and trace handlers
* in a vector.
*/
exiting_irq(); /* post-processing of this handler */
}
Trace irq_handler:
smp_trace_irq_handler()
{
entering_irq(); /* pre-processing of this handler */
trace_irq_entry(); /* tracepoint for irq entry */
__smp_irq_handler(); /*
* common logic between non-trace and trace handlers
* in a vector.
*/
trace_irq_exit(); /* tracepoint for irq exit */
exiting_irq(); /* post-processing of this handler */
}
If tracepoints can place outside entering_irq()/exiting_irq() as follows,
it looks cleaner.
smp_trace_irq_handler()
{
trace_irq_entry();
smp_irq_handler();
trace_irq_exit();
}
But it doesn't work.
The problem is with irq_enter/exit() being called. They must be called before
trace_irq_enter/exit(), because of the rcu_irq_enter() must be called before
any tracepoints are used, as tracepoints use rcu to synchronize.
As a possible alternative, we may be able to call irq_enter() first as follows
if irq_enter() can nest.
smp_trace_irq_hander()
{
irq_entry();
trace_irq_entry();
smp_irq_handler();
trace_irq_exit();
irq_exit();
}
But it doesn't work, either.
If irq_enter() is nested, it may have a time penalty because it has to check if it
was already called or not. The time penalty is not desired in performance sensitive
paths even if it is tiny.
Signed-off-by: Seiji Aguchi <seiji.aguchi@hds.com>
Link: http://lkml.kernel.org/r/51C3238D.9040706@hds.com
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
For v3.3, I added code to use the NMI to stop other cpus in the
panic case. The idea was to make sure all cpus on the system
were definitely halted to help serialize the panic path to
execute the rest of the code on a single cpu.
The main problem it was trying to solve was how to stop a cpu
that was spinning with its irqs disabled. A IPI irq would be
stuck and couldn't get in there, but an NMI could.
Things were great until we had another conversation about some
pstore changes. Because some of the backend pstore still uses
spinlocks to protect the device access, things could get ugly if
a panic happened and we were stuck spinning on a lock.
Now with the NMI shutting down cpus, we could assume no other
cpus were running and just bust the spin lock and proceed.
The counter argument was, well if you do that the backend could
be in a screwed up state and you might not be able to save
anything as a result. If we could have just given the cpu a
little more time to finish things, we could have grabbed the
spin lock cleanly and everything would have been fine.
Well, how do give a cpu a 'little more time' in the panic case?
For the most part you can't without spinning on the lock and
even in that case, how long do you spin for?
So instead of making it ugly in the pstore code, just mimic the
idea that stop_machine had, which is block on an IRQ IPI until
the remote cpu has re-enabled interrupts and left the critical
region. Which is what happens now using REBOOT_IRQ.
Then leave the NMI case for those cpus that are truly stuck
after a short time. This leaves the current behaviour alone and
just handle a corner case. Most systems should never have to
enter the NMI code and if they do, print out a message in case
the NMI itself causes another issue.
Signed-off-by: Don Zickus <dzickus@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1336761675-24296-3-git-send-email-dzickus@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This reverts commit 3603a2512f.
Originally I wanted a better hammer to shutdown cpus during
panic. However, this really steps on the toes of various
spinlocks in the panic path. Sometimes it is easier to wait for
the IRQ to become re-enabled to indictate the cpu left the
critical region and then shutdown the cpu.
The next patch moves the NMI addition after the IRQ part. To
make it easier to see the logic of everything, revert this patch
and apply the next simpler patch.
Signed-off-by: Don Zickus <dzickus@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1336761675-24296-2-git-send-email-dzickus@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
A recent discussion started talking about the locking on the
pstore fs and how it relates to the kmsg infrastructure. We
noticed it was possible for userspace to r/w to the pstore fs
(grabbing the locks in the process) and block the panic path
from r/w to the same fs.
The reason was the cpu with the lock could be doing work while
the crashing cpu is panic'ing. Busting those spinlocks might
cause those cpus to step on each other's data. Fine, fair
enough.
It was suggested it would be nice to serialize the panic path
(ie stop the other cpus) and have only one cpu running. This
would allow us to bust the spinlocks and not worry about another
cpu stepping on the data.
Of course, smp_send_stop() does this in the panic case.
kmsg_dump() would have to be moved to be called after it. Easy
enough.
The only problem is on x86 the smp_send_stop() function calls
the REBOOT_VECTOR. Any cpu with irqs disabled (which pstore and
its backend ERST would do), block this IPI and thus do not stop.
This makes it difficult to reliably log data to the pstore fs.
The patch below switches from the REBOOT_VECTOR to NMI (and
mimics what kdump does). Switching to NMI allows us to deliver
the IPI when irqs are disabled, increasing the reliability of
this function.
However, Andi carefully noted that on some machines this
approach does not work because of broken BIOSes or whatever.
To help accomodate this, the next couple of patches will run a
selftest and provide a knob to disable.
V2:
uses atomic ops to serialize the cpu that shuts everyone down
V3:
comment cleanup
Signed-off-by: Don Zickus <dzickus@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Robert Richter <robert.richter@amd.com>
Cc: seiji.aguchi@hds.com
Cc: vgoyal@redhat.com
Cc: mjg@redhat.com
Cc: tony.luck@intel.com
Cc: gong.chen@intel.com
Cc: satoru.moriya@hds.com
Cc: avi@redhat.com
Cc: Andi Kleen <andi@firstfloor.org>
Link: http://lkml.kernel.org/r/1318533267-18880-2-git-send-email-dzickus@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>
These files were implicitly getting EXPORT_SYMBOL via device.h
which was including module.h, but that will be fixed up shortly.
By fixing these now, we can avoid seeing things like:
arch/x86/kernel/rtc.c:29: warning: type defaults to ‘int’ in declaration of ‘EXPORT_SYMBOL’
arch/x86/kernel/pci-dma.c:20: warning: type defaults to ‘int’ in declaration of ‘EXPORT_SYMBOL’
arch/x86/kernel/e820.c:69: warning: type defaults to ‘int’ in declaration of ‘EXPORT_SYMBOL_GPL’
[ with input from Randy Dunlap <rdunlap@xenotime.net> and also
from Stephen Rothwell <sfr@canb.auug.org.au> ]
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
x86 smp_ops now has a new op, stop_other_cpus which takes a parameter
"wait" this allows the caller to specify if it wants to stop until all
the cpus have processed the stop IPI. This is required specifically
for the kexec case where we should wait for all the cpus to be stopped
before starting the new kernel. We now wait for the cpus to stop in
all cases except for panic/kdump where we expect things to be broken
and we are doing our best to make things work anyway.
This patch fixes a legitimate regression, which was introduced during
2.6.30, by commit id 4ef702c10b.
Signed-off-by: Alok N Kataria <akataria@vmware.com>
LKML-Reference: <1286833028.1372.20.camel@ank32.eng.vmware.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: <stable@kernel.org> v2.6.30-36
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files. percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.
percpu.h -> slab.h dependency is about to be removed. Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability. As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.
http://userweb.kernel.org/~tj/misc/slabh-sweep.py
The script does the followings.
* Scan files for gfp and slab usages and update includes such that
only the necessary includes are there. ie. if only gfp is used,
gfp.h, if slab is used, slab.h.
* When the script inserts a new include, it looks at the include
blocks and try to put the new include such that its order conforms
to its surrounding. It's put in the include block which contains
core kernel includes, in the same order that the rest are ordered -
alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
doesn't seem to be any matching order.
* If the script can't find a place to put a new include (mostly
because the file doesn't have fitting include block), it prints out
an error message indicating which .h file needs to be added to the
file.
The conversion was done in the following steps.
1. The initial automatic conversion of all .c files updated slightly
over 4000 files, deleting around 700 includes and adding ~480 gfp.h
and ~3000 slab.h inclusions. The script emitted errors for ~400
files.
2. Each error was manually checked. Some didn't need the inclusion,
some needed manual addition while adding it to implementation .h or
embedding .c file was more appropriate for others. This step added
inclusions to around 150 files.
3. The script was run again and the output was compared to the edits
from #2 to make sure no file was left behind.
4. Several build tests were done and a couple of problems were fixed.
e.g. lib/decompress_*.c used malloc/free() wrappers around slab
APIs requiring slab.h to be added manually.
5. The script was run on all .h files but without automatically
editing them as sprinkling gfp.h and slab.h inclusions around .h
files could easily lead to inclusion dependency hell. Most gfp.h
inclusion directives were ignored as stuff from gfp.h was usually
wildly available and often used in preprocessor macros. Each
slab.h inclusion directive was examined and added manually as
necessary.
6. percpu.h was updated not to include slab.h.
7. Build test were done on the following configurations and failures
were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my
distributed build env didn't work with gcov compiles) and a few
more options had to be turned off depending on archs to make things
build (like ipr on powerpc/64 which failed due to missing writeq).
* x86 and x86_64 UP and SMP allmodconfig and a custom test config.
* powerpc and powerpc64 SMP allmodconfig
* sparc and sparc64 SMP allmodconfig
* ia64 SMP allmodconfig
* s390 SMP allmodconfig
* alpha SMP allmodconfig
* um on x86_64 SMP allmodconfig
8. percpu.h modifications were reverted so that it could be applied as
a separate patch and serve as bisection point.
Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
This reverts commit 9bcbdd9c58.
The real bug producing LatencyTop latencies has been fixed in:
f5dc375: sched: Update the clock of runqueue select_task_rq() selected
And the commit being reverted here triggers local timer processing
from every device IRQ. If device IRQs come in at a high frequency,
this could cause a performance regression.
The commit being reverted here purely 'fixed' the reported latency
as a side effect, because CPUs were being moved out of idle more
often.
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Cc: Frans Pop <elendil@planet.nl>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <20091008064041.67219b13@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Now that range timers and deferred timers are common, I found a
problem with these using the "perf timechart" tool. Frans Pop also
reported high scheduler latencies via LatencyTop, when using
iwlagn.
It turns out that on x86, these two 'opportunistic' timers only get
checked when another "real" timer happens. These opportunistic
timers have the objective to save power by hitchhiking on other
wakeups, as to avoid CPU wakeups by themselves as much as possible.
The change in this patch runs this check not only at timer
interrupts, but at all (device) interrupts. The effect is that:
1) the deferred timers/range timers get delayed less
2) the range timers cause less wakeups by themselves because
the percentage of hitchhiking on existing wakeup events goes up.
I've verified the working of the patch using "perf timechart", the
original exposed bug is gone with this patch. Frans also reported
success - the latencies are now down in the expected ~10 msec
range.
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Tested-by: Frans Pop <elendil@planet.nl>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
LKML-Reference: <20091008064041.67219b13@infradead.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'kvm-updates/2.6.31' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (138 commits)
KVM: Prevent overflow in largepages calculation
KVM: Disable large pages on misaligned memory slots
KVM: Add VT-x machine check support
KVM: VMX: Rename rmode.active to rmode.vm86_active
KVM: Move "exit due to NMI" handling into vmx_complete_interrupts()
KVM: Disable CR8 intercept if tpr patching is active
KVM: Do not migrate pending software interrupts.
KVM: inject NMI after IRET from a previous NMI, not before.
KVM: Always request IRQ/NMI window if an interrupt is pending
KVM: Do not re-execute INTn instruction.
KVM: skip_emulated_instruction() decode instruction if size is not known
KVM: Remove irq_pending bitmap
KVM: Do not allow interrupt injection from userspace if there is a pending event.
KVM: Unprotect a page if #PF happens during NMI injection.
KVM: s390: Verify memory in kvm run
KVM: s390: Sanity check on validity intercept
KVM: s390: Unlink vcpu on destroy - v2
KVM: s390: optimize float int lock: spin_lock_bh --> spin_lock
KVM: s390: use hrtimer for clock wakeup from idle - v2
KVM: s390: Fix memory slot versus run - v3
...
KVM uses a function call IPI to cause the exit of a guest running on a
physical cpu. For virtual interrupt notification there is no need to
wait on IPI receival, or to execute any function.
This is exactly what the reschedule IPI does, without the overhead
of function IPI. So use it instead of smp_call_function_single in
kvm_vcpu_kick.
Also change the "guest_mode" variable to a bit in vcpu->requests, and
use that to collapse multiple IPI's that would be issued between the
first one and zeroing of guest mode.
This allows kvm_vcpu_kick to called with interrupts disabled.
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
For some time each panic() called with interrupts disabled
triggered the !irqs_disabled() WARN_ON in smp_call_function(),
producing ugly backtraces and confusing users.
This is a common situation with machine checks for example which
tend to call panic with interrupts disabled, but will also hit
in other situations e.g. panic during early boot. In fact it
means that panic cannot be called in many circumstances, which
would be bad.
This all started with the new fancy queued smp_call_function,
which is then used by the shutdown path to shut down the other
CPUs.
On closer examination it turned out that the fancy RCU
smp_call_function() does lots of things not suitable in a panic
situation anyways, like allocating memory and relying on complex
system state.
I originally tried to patch this over by checking for panic
there, but it was quite complicated and the original patch
was also not very popular. This also didn't fix some of the
underlying complexity problems.
The new code in post 2.6.29 tries to patch around this by
checking for oops_in_progress, but that is not enough to make
this fully safe and I don't think that's a real solution
because panic has to be reliable.
So instead use an own vector to reboot. This makes the reboot
code extremly straight forward, which is definitely a big plus
in a panic situation where it is important to avoid relying on
too much kernel state. The new simple code is also safe to be
called from interupts off region because it is very very simple.
There can be situations where it is important that panic
is reliable. For example on a fatal machine check the panic
is needed to get the system up again and running as quickly
as possible. So it's important that panic is reliable and
all function it calls simple.
This is why I came up with this simple vector scheme.
It's very hard to beat in simplicity. Vectors are not
particularly precious anymore since all big systems are
using per CPU vectors.
Another possibility would have been to use an NMI similar
to kdump, but there is still the problem that NMIs don't
work reliably on some systems due to BIOS issues. NMIs
would have been able to stop CPUs running with interrupts
off too. In the sake of universal reliability I opted for
using a non NMI vector for now.
I put the reboot vector into the highest priority bucket of
the APIC vectors and moved the 64bit UV_BAU message down
instead into the next lower priority.
[ Impact: bug fix, fixes an old regression ]
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Ashok Raj
Thomas Gleixner
Andi Kleen
Seiji Aguchi
Don Zickus
Paul Gortmaker
Peter Zijlstra
Alok Kataria
Tejun Heo
Ingo Molnar
Arjan van de Ven
Linus Torvalds
Marcelo Tosatti
Andi Kleen
Cyrill Gorcunov