The context tracking subsystem has the ability to selectively
enable the tracking on any defined subset of CPU. This means that
we can define a CPU range that doesn't run the context tracking
and another range that does.
Now what we want in practice is to enable the tracking on full
dynticks CPUs only. In order to perform this, we just need to pass
our full dynticks CPU range selection from the full dynticks
subsystem to the context tracking.
This way we can spare the overhead of RCU user extended quiescent
state and vtime maintainance on the CPUs that are outside the
full dynticks range. Just keep in mind the raw context tracking
itself is still necessary everywhere.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kevin Hilman <khilman@linaro.org>
As long as the context tracking is enabled on any CPU, even
a single one, all other CPUs need to keep track of their
user <-> kernel boundaries cross as well.
This is because a task can sleep while servicing an exception
that happened in the kernel or in userspace. Then when the task
eventually wakes up and return from the exception, the CPU needs
to know if we resume in userspace or in the kernel. exception_exit()
get this information from exception_enter() that saved the previous
state.
If the CPU where the exception happened didn't keep track of
these informations, exception_exit() doesn't know which state
tracking to restore on the CPU where the task got migrated
and we may return to userspace with the context tracking
subsystem thinking that we are in kernel mode.
This can be fixed in the long term if we move our context tracking
probes on very low level arch fast path user <-> kernel boundary,
although even that is worrisome as an exception can still happen
in the few instructions between the probe and the actual iret.
Also we are not yet ready to set these probes in the fast path given
the potential overhead problem it induces.
So let's fix this by always enable context tracking even on CPUs
that are not in the full dynticks range. OTOH we can spare the
rcu_user_*() and vtime_user_*() calls there because the tick runs
on these CPUs and we can handle RCU state machine and cputime
accounting through it.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kevin Hilman <khilman@linaro.org>
1) If context tracking is enabled with native vtime accounting (which
combo is useless except for dev testing), we call vtime_guest_enter()
and vtime_guest_exit() on host <-> guest switches. But those are stubs
in this configurations. As a result, cputime is not correctly flushed
on kvm context switches.
2) If context tracking runs but is disabled on some CPUs, those
CPUs end up calling __guest_enter/__guest_exit which in turn
call vtime_account_system(). We don't want to call this because we
run in tick based accounting for these CPUs.
Refactor the guest_enter/guest_exit code such that all combinations
finally work.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kevin Hilman <khilman@linaro.org>
If the user enables CONFIG_NO_HZ_FULL and runs the kernel on a machine
with an unstable TSC, it will produce a WARN_ON dump as well as taint
the kernel. This is a bit extreme for a kernel that just enables a
feature but doesn't use it.
The warning should only happen if the user tries to use the feature by
either adding nohz_full to the kernel command line, or by enabling
CONFIG_NO_HZ_FULL_ALL that makes nohz used on all CPUs at boot up. Note,
this second feature should not (yet) be used by distros or anyone that
doesn't care if NO_HZ is used or not.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Li Zhong <zhong@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kevin Hilman <khilman@linaro.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Pull tracing fixes and cleanups from Steven Rostedt:
"This contains fixes, optimizations and some clean ups
Some of the fixes need to go back to 3.10. They are minor, and deal
mostly with incorrect ref counting in accessing event files.
There was a couple of optimizations that should have perf perform a
bit better when accessing trace events.
And some various clean ups. Some of the clean ups are necessary to
help in a fix to a theoretical race between opening a event file and
deleting that event"
* tag 'trace-3.11-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
tracing: Kill the unbalanced tr->ref++ in tracing_buffers_open()
tracing: Kill trace_array->waiter
tracing: Do not (ab)use trace_seq in event_id_read()
tracing: Simplify the iteration logic in f_start/f_next
tracing: Add ref_data to function and fgraph tracer structs
tracing: Miscellaneous fixes for trace_array ref counting
tracing: Fix error handling to ensure instances can always be removed
tracing/kprobe: Wait for disabling all running kprobe handlers
tracing/perf: Move the PERF_MAX_TRACE_SIZE check into perf_trace_buf_prepare()
tracing/syscall: Avoid perf_trace_buf_*() if sys_data->perf_events is empty
tracing/function: Avoid perf_trace_buf_*() if event_function.perf_events is empty
tracing: Typo fix on ring buffer comments
tracing: Use trace_seq_puts()/trace_seq_putc() where possible
tracing: Use correct config guard CONFIG_STACK_TRACER
Pull thermal management fixes from Zhang Rui:
"These are fixes collected over the last week, they fixes several
problems caused by the x86_pkg_temp_thermal introduced in 3.11-rc1.
Specifics:
- the x86_pkg_temp_thermal driver causes crash on systems with no
package MSR support as there is a bug in the logic to check
presence of DTHERM and PTS feature together. Added a change so
that when there is no PTS support, module doesn't get loaded.
- fix krealloc() misuse in pkg_temp_thermal_device_add().
If krealloc() returns NULL, it doesn't free the original. Thus if
we want to exit because of the krealloc() failure, we must make
sure the original one is freed.
- The error code path of the x86 package temperature thermal driver's
initialization routine makes an unbalanced call to
get_online_cpus(), which causes subsequent CPU offline operations,
and consequently system suspend, to permanently block in
cpu_hotplug_begin() on systems where get_core_online() returns an
error code.
Remove the extra get_online_cpus() to fix the problem"
* 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/rzhang/linux:
Thermal: Fix lockup of cpu_down()
Thermal: x86_pkg_temp: Limit number of pkg temp zones
Thermal: x86_pkg_temp: fix krealloc() misuse in in pkg_temp_thermal_device_add()
Thermal: x86 package temp thermal crash
Pull gpio fixes from Linus Walleij:
"A first round of GPIO fixes for the v3.11 series:
- OMAP device tree boot fix
- Handle an error condition in the MSM driver
The OMAP patches have been around since around the merge window, but
since they first caused more breakage I let them boil in -next for a
while. These should be fine now"
* tag 'gpio-for-v3.11-2' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio:
drivers: gpio: msm: Fix the error condition for reading ngpio
gpio/omap: fix build error when OF_GPIO is not defined.
gpio/omap: auto request GPIO as input if used as IRQ via DT
gpio/omap: don't create an IRQ mapping for every GPIO on DT
Pull block IO driver bits from Jens Axboe:
"As I mentioned in the core block pull request, due to real life
circumstances the driver pull request would be late. Now it looks
like -rc2 late... On the plus side, apart form the rsxx update, these
are all things that I could argue could go in later in the cycle as
they are fixes and not features. So even though things are late, it's
not ALL bad.
The pull request contains:
- Updates to bcache, all bug fixes, from Kent.
- A pile of drbd bug fixes (no big features this time!).
- xen blk front/back fixes.
- rsxx driver updates, some of them deferred form 3.10. So should be
well cooked by now"
* 'for-3.11/drivers' of git://git.kernel.dk/linux-block: (63 commits)
bcache: Allocation kthread fixes
bcache: Fix GC_SECTORS_USED() calculation
bcache: Journal replay fix
bcache: Shutdown fix
bcache: Fix a sysfs splat on shutdown
bcache: Advertise that flushes are supported
bcache: check for allocation failures
bcache: Fix a dumb race
bcache: Use standard utility code
bcache: Update email address
bcache: Delete fuzz tester
bcache: Document shrinker reserve better
bcache: FUA fixes
drbd: Allow online change of al-stripes and al-stripe-size
drbd: Constants should be UPPERCASE
drbd: Ignore the exit code of a fence-peer handler if it returns too late
drbd: Fix rcu_read_lock balance on error path
drbd: fix error return code in drbd_init()
drbd: Do not sleep inside rcu
bcache: Refresh usage docs
...
The expression '(1 << 32)' happens to evaluate as 0 on ARM, but
it evaluates as 1 on xtensa and x86_64. This zeros sched_clock_mask,
and breaks sched_clock().
Set the type of 1 to 'unsigned long long' to get the value we need.
Reported-by: Max Filippov <jcmvbkbc@gmail.com>
Tested-by: Max Filippov <jcmvbkbc@gmail.com>
Acked-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: John Stultz <john.stultz@linaro.org>
Commit f1a18a105 "Thermal: CPU Package temperature thermal" had code
that did a get_online_cpus(), run a loop and then do a
put_online_cpus(). The problem is that the loop had an error exit that
would skip the put_online_cpus() part.
In the error exit part of the function, it also did a get_online_cpus(),
run a loop and then put_online_cpus(). The only way to get to the error
exit part is with get_online_cpus() already performed. If this error
condition is hit, the system will be prevented from taking CPUs offline.
The process taking the CPU offline will lock up hard.
Removing the get_online_cpus() removes the lockup as the hotplug CPU
refcount is back to zero.
This was bisected with ktest.
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Pull ACPI video support fixes from Rafael Wysocki:
"I'm sending a separate pull request for this as it may be somewhat
controversial. The breakage addressed here is not really new and the
fixes may not satisfy all users of the affected systems, but we've had
so much back and forth dance in this area over the last several weeks
that I think it's time to actually make some progress.
The source of the problem is that about a year ago we started to tell
BIOSes that we're compatible with Windows 8, which we really need to
do, because some systems shipping with Windows 8 are tested with it
and nothing else, so if we tell their BIOSes that we aren't compatible
with Windows 8, we expose our users to untested BIOS/AML code paths.
However, as it turns out, some Windows 8-specific AML code paths are
not tested either, because Windows 8 actually doesn't use the ACPI
methods containing them, so if we declare Windows 8 compatibility and
attempt to use those ACPI methods, things break. That occurs mostly
in the backlight support area where in particular the _BCM and _BQC
methods are plain unusable on some systems if the OS declares Windows
8 compatibility.
[ The additional twist is that they actually become usable if the OS
says it is not compatible with Windows 8, but that may cause
problems to show up elsewhere ]
Investigation carried out by Matthew Garrett indicates that what
Windows 8 does about backlight is to leave backlight control up to
individual graphics drivers. At least there's evidence that it does
that if the Intel graphics driver is used, so we've decided to follow
Windows 8 in that respect and allow i915 to control backlight (Daniel
likes that part).
The first commit from Aaron Lu makes ACPICA export the variable from
which we can infer whether or not the BIOS believes that we are
compatible with Windows 8.
The second commit from Matthew Garrett prepares the ACPI video driver
by making it initialize the ACPI backlight even if it is not going to
be used afterward (that is needed for backlight control to work on
Thinkpads).
The third commit implements the actual workaround making i915 take
over backlight control if the firmware thinks it's dealing with
Windows 8 and is based on the work of multiple developers, including
Matthew Garrett, Chun-Yi Lee, Seth Forshee, and Aaron Lu.
The final commit from Aaron Lu makes us follow Windows 8 by informing
the firmware through the _DOS method that it should not carry out
automatic brightness changes, so that brightness can be controlled by
GUI.
Hopefully, this approach will allow us to avoid using blacklists of
systems that should not declare Windows 8 compatibility just to avoid
backlight control problems in the future.
- Change from Aaron Lu makes ACPICA export a variable which can be
used by driver code to determine whether or not the BIOS believes
that we are compatible with Windows 8.
- Change from Matthew Garrett makes the ACPI video driver initialize
the ACPI backlight even if it is not going to be used afterward
(that is needed for backlight control to work on Thinkpads).
- Fix from Rafael J Wysocki implements Windows 8 backlight support
workaround making i915 take over bakclight control if the firmware
thinks it's dealing with Windows 8. Based on the work of multiple
developers including Matthew Garrett, Chun-Yi Lee, Seth Forshee,
and Aaron Lu.
- Fix from Aaron Lu makes the kernel follow Windows 8 by informing
the firmware through the _DOS method that it should not carry out
automatic brightness changes, so that brightness can be controlled
by GUI"
* tag 'acpi-video-3.11' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
ACPI / video: no automatic brightness changes by win8-compatible firmware
ACPI / video / i915: No ACPI backlight if firmware expects Windows 8
ACPI / video: Always call acpi_video_init_brightness() on init
ACPICA: expose OSI version
Pull ext[34] tmpfile bugfix from Ted Ts'o:
"Fix regression caused by commit af51a2ac36 which added ->tmpfile()
support (along with a similar fix for ext3)"
* tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
ext3: fix a BUG when opening a file with O_TMPFILE flag
ext4: fix a BUG when opening a file with O_TMPFILE flag
Pull staging tree fixes from Greg KH:
"Here are a few iio driver fixes for 3.11-rc2. They are still spread
across drivers/iio and drivers/staging/iio so they are coming in
through this tree.
I've also removed the drivers/staging/csr/ driver as the developers
who originally sent it to me have moved on to other companies, and CSR
still will not send us the specs for the device, making the driver
pretty much obsolete and impossible to fix up. Deleting it now
prevents people from sending in lots of tiny codingsyle fixes that
will never go anywhere.
It also helps to offset the large lustre filesystem merge that
happened in 3.11-rc1 in the overall 3.11.0 diffstat. :)"
* tag 'staging-3.11-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
staging: csr: remove driver
iio: lps331ap: Fix wrong in_pressure_scale output value
iio staging: fix lis3l02dq, read error handling
staging:iio:ad7291: add missing .driver_module to struct iio_info
iio: ti_am335x_adc: add missing .driver_module to struct iio_info
iio: mxs-lradc: Remove useless check in read_raw
iio: mxs-lradc: Fix misuse of iio->trig
iio: inkern: fix iio_convert_raw_to_processed_unlocked
iio: Fix iio_channel_has_info
iio:trigger: device_unregister->device_del to avoid double free
iio: dac: ad7303: fix error return code in ad7303_probe()
Pull vfs fixes from Al Viro:
"The sget() one is a long-standing bug and will need to go into -stable
(in fact, it had been originally caught in RHEL6), the other two are
3.11-only"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
vfs: constify dentry parameter in d_count()
livelock avoidance in sget()
allow O_TMPFILE to work with O_WRONLY
Pull ext4 bugfixes from Ted Ts'o:
"Fixes for 3.11-rc2, sent at 5pm, in the professoinal style. :-)"
I'm not sure I like this new level of "professionalism".
9-5, people, 9-5.
* tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
ext4: call ext4_es_lru_add() after handling cache miss
ext4: yield during large unlinks
ext4: make the extent_status code more robust against ENOMEM failures
ext4: simplify calculation of blocks to free on error
ext4: fix error handling in ext4_ext_truncate()
Pull NFS client bugfixes from Trond Myklebust:
- Fix a regression against NFSv4 FreeBSD servers when creating a new
file
- Fix another regression in rpc_client_register()
* tag 'nfs-for-3.11-3' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
NFSv4: Fix a regression against the FreeBSD server
SUNRPC: Fix another issue with rpc_client_register()
Frederic Weisbecker
Ingo Molnar
Li Zhong
Steven Rostedt
Linus Torvalds
Baruch Siach
Zheng Liu
Rohit Vaswani