Updates of the timekeeper can be done by operating on the shadow timekeeper
and afterwards copying the result into the real timekeeper. This has the
advantage, that the sequence count write protected region is kept as small
as possible.
Convert change_clocksource() to use this scheme.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-18-554456a44a15@linutronix.de
Updates of the timekeeper can be done by operating on the shadow timekeeper
and afterwards copying the result into the real timekeeper. This has the
advantage, that the sequence count write protected region is kept as small
as possible.
Convert timekeeping_inject_offset() to use this scheme.
That allows to use a scoped_guard() for locking the timekeeper lock as the
usage of the shadow timekeeper allows a rollback in the error case instead
of the full timekeeper update of the original code.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-17-554456a44a15@linutronix.de
Updates of the timekeeper can be done by operating on the shadow timekeeper
and afterwards copying the result into the real timekeeper. This has the
advantage, that the sequence count write protected region is kept as small
as possible.
Convert do_settimeofday64() to use this scheme.
That allows to use a scoped_guard() for locking the timekeeper lock as the
usage of the shadow timekeeper allows a rollback in the error case instead
of the full timekeeper update of the original code.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-16-554456a44a15@linutronix.de
Functions which operate on the real timekeeper, e.g. do_settimeofday(),
have error conditions. If they are hit a full timekeeping update is still
required because the already committed operations modified the timekeeper.
When switching these functions to operate on the shadow timekeeper then the
full update can be avoided in the error case, but the modified shadow
timekeeper has to be restored.
Provide a helper function for that.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-15-554456a44a15@linutronix.de
timekeeping_advance() is the only optimized function which uses
shadow_timekeeper for updating the real timekeeper to keep the sequence
counter protected region as small as possible.
To be able to transform timekeeper updates in other functions to use the
same logic, split out functionality into a separate function
timekeeper_update_staged().
While at it, document the reason why the sequence counter must be write
held over the call to timekeeping_update() and the copying to the real
timekeeper and why using a pointer based update is suboptimal.
No functional change.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-13-554456a44a15@linutronix.de
Updates of the timekeeper are done in two ways:
1. Updating timekeeper and afterwards memcpy()'ing the result into
shadow_timekeeper using timekeeping_update(). Used everywhere for
updates except in timekeeping_advance(); the sequence counter protected
region starts before the first change to the timekeeper is done.
2. Updating shadow_timekeeper and then memcpy()'ing the result into
timekeeper. Used only by in timekeeping_advance(); The seqence counter
protected region is only around timekeeping_update() and the memcpy for
copy from shadow to timekeeper.
The second option is fast path optimized. The sequence counter protected
region is as short as possible.
As this behaviour is mainly documented by commit messages, but not in code,
it makes the not easy timekeeping code more complicated to read.
There is no reason why updates to the timekeeper can't use the optimized
version everywhere. With this, the code will be cleaner, as code is reused
instead of duplicated.
To be able to access tk_data which contains all required information, add a
pointer to tk_data as an argument to timekeeping_update(). With that
convert the comment about holding the lock into a lockdep assert.
No functional change.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-12-554456a44a15@linutronix.de
The struct tk_core uses is not reusable. As long as there is only a single
timekeeper, this is not a problem. But when the timekeeper infrastructure
will be reused for per ptp clock timekeepers, an explicit struct type is
required.
Define struct tk_data as explicit struct type for tk_core.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-10-554456a44a15@linutronix.de
timekeeper_lock protects updates to struct tk_core but is not part of
struct tk_core. As long as there is only a single timekeeper, this is not a
problem. But when the timekeeper infrastructure will be reused for per ptp
clock timekeepers, timekeeper_lock needs to be part of tk_core.
Move the lock into tk_core, move initialisation of the lock and sequence
counter into timekeeping_init() and update all users of timekeeper_lock.
As this is touching all lock sites, convert them to use:
guard(raw_spinlock_irqsave)(&tk_core.lock);
instead of lock/unlock functions whenever possible.
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-9-554456a44a15@linutronix.de
timekeeper_lock protects updates of timekeeper (tk_core). It is also used
by vdso_update_begin/end() and not only internally by the timekeeper code.
As long as there is only a single timekeeper, this works fine. But when
the timekeeper infrastructure will be reused for per ptp clock timekeepers,
timekeeper_lock needs to be part of tk_core..
Therefore encapuslate locking/unlocking of timekeeper_lock and make the
lock static.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-8-554456a44a15@linutronix.de
tk_core requires shadow_timekeeper to allow timekeeping_advance() updating
without holding the timekeeper sequence count write locked. This allows the
readers to make progress up to the actual update where the shadow
timekeeper is copied over to the real timekeeper.
As long as there is only a single timekeeper, having them separate is
fine. But when the timekeeper infrastructure will be reused for per ptp
clock timekeepers, shadow_timekeeper needs to be part of tk_core.
No functional change.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-7-554456a44a15@linutronix.de
struct timekeeper is ordered suboptimal vs. cachelines. The layout,
including the preceding seqcount (see struct tk_core in timekeeper.c) is:
cacheline 0: seqcount, tkr_mono
cacheline 1: tkr_raw, xtime_sec
cacheline 2: ktime_sec ... tai_offset, internal variables
cacheline 3: next_leap_ktime, raw_sec, internal variables
cacheline 4: internal variables
So any access to via ktime_get*() except for access to CLOCK_MONOTONIC_RAW
will use either cachelines 0 + 1 or cachelines 0 + 2. Access to
CLOCK_MONOTONIC_RAW uses cachelines 0 + 1 + 3.
Reorder the members so that the result is more efficient:
cacheline 0: seqcount, tkr_mono
cacheline 1: xtime_sec, ktime_sec ... tai_offset
cacheline 2: tkr_raw, raw_sec
cacheline 3: internal variables
cacheline 4: internal variables
That means ktime_get*() will access cacheline 0 + 1 and CLOCK_MONOTONIC_RAW
access will use cachelines 0 + 2.
Update kernel-doc and fix formatting issues while at it. Also fix a typo
in struct tk_read_base kernel-doc.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241015100839.12702-1-anna-maria@linutronix.de
timekeeping_advance() takes the timekeeper_lock and releases it before
returning. When an early return is required, goto statements are used to
make sure the lock is realeased properly. When the code was written the
locking guard() was not yet available.
Use the guard() to simplify the code and while at it cleanup ordering of
function variables. No functional change.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-5-554456a44a15@linutronix.de
There is no point to go through a full timekeeping update when acquiring a
module reference or enabling the new clocksource fails.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-4-554456a44a15@linutronix.de
do_adjtimex() invokes tk_update_leap_state() unconditionally even when a
previous invocation of timekeeping_update() already did that update.
Put it into the else path which is invoked when timekeeping_update() is not
called.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-3-554456a44a15@linutronix.de
hard_pps() update does not modify anything which might be required by time
readers so forcing readers out of the way during the update is a pointless
exercise.
The interaction with adjtimex() and timekeeper updates which call into the
NTP code is properly serialized by timekeeper_lock.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241009-devel-anna-maria-b4-timers-ptp-timekeeping-v2-2-554456a44a15@linutronix.de
Since SLOB was removed and since commit 6c6c47b063 ("mm, slab: call
kvfree_rcu_barrier() from kmem_cache_destroy()"), it is not longer
necessary to use call_rcu() when the callback only performs
kmem_cache_free(). Use kfree_rcu() directly.
The changes were made using Coccinelle.
Signed-off-by: Julia Lawall <Julia.Lawall@inria.fr>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Link: https://lore.kernel.org/all/20241013201704.49576-12-Julia.Lawall@inria.fr
The documentation which tries to give advices how to properly inserting
delays or sleeps is outdated. The file name is 'timers-howto.rst' which
might be misleading as it is only about delay and sleep mechanisms and not
how to use timers.
Update the documentation by integrating the important parts from the
related function descriptions and move it all into a self explaining file
with the name "delay_sleep_functions.rst".
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-15-dc8b907cb62f@linutronix.de
anysee driver was transformed to use usbv2 years ago. The comments in
anysee_ctrl_msg() still are referencing the old interfaces where msleep()
was used. The v2 interfaces also changed over the years and with commit
1162c7b383 ("[media] dvb_usb_v2: refactor dvb_usbv2_generic_rw()") the
usage of msleep() was gone anyway.
Remove FIXME comment and update also comment before call to
dvb_usbv2_generic_rw_locked().
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-14-dc8b907cb62f@linutronix.de
When commit 38f7b7067d ("powerpc/rtas: rtas_busy_delay() improvements")
was introduced, documentation about proper usage of sleep related functions
was outdated.
The commit message references the usage of a HZ=100 system. When using a
20ms sleep duration on such a system and therefore using msleep(), the
possible additional slack will be +10ms.
When the system is configured with HZ=100 the granularity of a jiffy and of
a bucket of the lowest timer wheel level is 10ms. To make sure a timer will
not expire early (when queueing of the timer races with an concurrent
update of jiffies), timers are always queued into the next bucket. This is
the reason for the maximal possible slack of 10ms.
fsleep() limits the maximal possible slack to 25% by making threshold
between usleep_range() and msleep() HZ dependent. As soon as the accuracy
of msleep() is sufficient, the less expensive timer list timer based
sleeping function is used instead of the more expensive hrtimer based
usleep_range() function. The udelay() will not be used in this specific
usecase as the lowest sleep length is larger than 1 millisecond.
Use fsleep() directly instead of using an own heuristic for the best
sleeping mechanism to use.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc)
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-13-dc8b907cb62f@linutronix.de
Function descriptions in iopoll.h, regmap.h, phy.h and sound/soc/sof/ops.h
copied all the same outdated documentation about sleep/delay function
limitations. In those comments, the generic (and still outdated) timer
documentation file is referenced.
As proper function descriptions for used delay and sleep functions are in
place, simply update the descriptions to reference to them. While at it fix
missing colon after "Returns" in function description and move return value
description to the end of the function description.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Andrew Lunn <andrew@lunn.ch> # for phy.h
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-12-dc8b907cb62f@linutronix.de
checkpatch.pl checks for several things related to sleep and delay
functions. In all warnings the outdated documentation is referenced. Also
in checkpatch kernel documentation the outdated documentation is
referenced.
Replace the links to the outdated documentation with links to the function
description.
Note: Update of the outdated checkpatch checks is done in a second step.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-10-dc8b907cb62f@linutronix.de
There is a warning in checkpatch script that triggers, when min and max
arguments of usleep_range_state() are in reverse order. This check does
only cover callsites which uses constants. Add this check into the code as
a WARN_ON_ONCE() to also cover callsites not using constants and fix the
mis-usage by resetting the delta to 0.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-9-dc8b907cb62f@linutronix.de
The upper bound for usleep_range_idle() was taken from the outdated
documentation. As a recommondation for the upper bound of usleep_range()
depends on HZ configuration it is not possible to hard code it.
Use the define "USLEEP_RANGE_UPPER_BOUND" instead.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: SeongJae Park <sj@kernel.org>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-8-dc8b907cb62f@linutronix.de
fsleep() simply implements the recommendations of the outdated
documentation in "Documentation/timers/timers-howto.rst". This should be a
user friendly interface to choose always the best timeout function
approach:
- udelay() for very short sleep durations shorter than 10 microseconds
- usleep_range() for sleep durations until 20 milliseconds
- msleep() for the others
The actual implementation has several problems:
- It does not take into account that HZ resolution also has an impact on
granularity of jiffies and has also an impact on the granularity of the
buckets of timer wheel levels. This means that accuracy for the timeout
does not have an upper limit. When executing fsleep(20000) on a HZ=100
system, the possible additional slack will be 50% as the granularity of
the buckets in the lowest level is 10 milliseconds.
- The upper limit of usleep_range() is twice the requested timeout. When no
other interrupts occur in this range, the maximum value is used. This
means that the requested sleep length has then an additional delay of
100%.
Change the thresholds for the decisions in fsleep() to make sure the
maximum slack which is added to the sleep duration is 25%.
Note: Outdated documentation will be updated in a followup patch.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-7-dc8b907cb62f@linutronix.de
udelay() as well as ndelay() are defines and no functions and are using
constants to be able to transform a sleep time into loops and to prevent
too long udelays/ndelays. There was a compiler error with non-const 8 bit
arguments which was fixed by commit a87e553fab ("asm-generic: delay.h fix
udelay and ndelay for 8 bit args"). When using a function, the non-const 8
bit argument is type casted and the problem would be gone.
Transform udelay() and ndelay() into proper functions, remove the no longer
and confusing division, add defines for the magic values and add some
explanations as well.
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-6-dc8b907cb62f@linutronix.de
A lot of commonly used functions for inserting a sleep or delay lack a
proper function description. Add function descriptions to all of them to
have important information in a central place close to the code.
No functional change.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-5-dc8b907cb62f@linutronix.de
usleep_idle_range() is a variant of usleep_range(). Both are using
usleep_range_state() as a base. To be able to find all the related
functions in one go, rename it usleep_idle_range() to usleep_range_idle().
No functional change.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Reviewed-by: SeongJae Park <sj@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-4-dc8b907cb62f@linutronix.de
schedule_timeout*() functions do not have proper kernel-doc formatted
function descriptions. schedule_hrtimeout() and schedule_hrtimeout_range()
have a almost identical description.
Add missing function descriptions. Remove copy of function description and
add a pointer to the existing description instead.
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-3-dc8b907cb62f@linutronix.de
All schedule_timeout() and *sleep*() related functions are interfaces on
top of timer list timers and hrtimers to add a sleep to the code. As they
are built on top of the timer list timers and hrtimers, the [hr]timer
interfaces are already used except when queuing the timer in
schedule_timeout(). But there exists the appropriate interface add_timer()
which does the same job with an extra check for an already pending timer.
Split all those functions as they are into a separate file and use
add_timer() instead of __mod_timer() in schedule_timeout().
While at it fix minor formatting issues and a multi line printk function
call in schedule_timeout().
Signed-off-by: Anna-Maria Behnsen <anna-maria@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241014-devel-anna-maria-b4-timers-flseep-v3-2-dc8b907cb62f@linutronix.de
clocksource_change_rating() has been unused since 2017's commit
63ed4e0c67 ("Drivers: hv: vmbus: Consolidate all Hyper-V specific clocksource code")
Remove it.
__clocksource_change_rating now only has one use which is ifdef'd.
Move it into the ifdef'd section.
Signed-off-by: Dr. David Alan Gilbert <linux@treblig.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241010135446.213098-1-linux@treblig.org
The mgtime_floor value is a global variable for tracking the latest
fine-grained timestamp handed out. Because it's a global, track the
number of times that a new floor value is assigned.
Add a new percpu counter to the timekeeping code to track the number of
floor swap events that have occurred. A later patch will add a debugfs
file to display this counter alongside other stats involving multigrain
timestamps.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Randy Dunlap <rdunlap@infradead.org> # documentation bits
Link: https://lore.kernel.org/all/20241002-mgtime-v10-2-d1c4717f5284@kernel.org
Multigrain timestamps allow the kernel to use fine-grained timestamps when
an inode's attributes is being actively observed via ->getattr(). With
this support, it's possible for a file to get a fine-grained timestamp, and
another modified after it to get a coarse-grained stamp that is earlier
than the fine-grained time. If this happens then the files can appear to
have been modified in reverse order, which breaks VFS ordering guarantees
[1].
To prevent this, maintain a floor value for multigrain timestamps.
Whenever a fine-grained timestamp is handed out, record it, and when later
coarse-grained stamps are handed out, ensure they are not earlier than that
value. If the coarse-grained timestamp is earlier than the fine-grained
floor, return the floor value instead.
Add a static singleton atomic64_t into timekeeper.c that is used to keep
track of the latest fine-grained time ever handed out. This is tracked as a
monotonic ktime_t value to ensure that it isn't affected by clock
jumps. Because it is updated at different times than the rest of the
timekeeper object, the floor value is managed independently of the
timekeeper via a cmpxchg() operation, and sits on its own cacheline.
Add two new public interfaces:
- ktime_get_coarse_real_ts64_mg() fills a timespec64 with the later of the
coarse-grained clock and the floor time
- ktime_get_real_ts64_mg() gets the fine-grained clock value, and tries
to swap it into the floor. A timespec64 is filled with the result.
The floor value is global and updated via a single try_cmpxchg(). If
that fails then the operation raced with a concurrent update. Any
concurrent update must be later than the existing floor value, so any
racing tasks can accept any resulting floor value without retrying.
[1]: POSIX requires that files be stamped with realtime clock values, and
makes no provision for dealing with backward clock jumps. If a backward
realtime clock jump occurs, then files can appear to have been modified
in reverse order.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Randy Dunlap <rdunlap@infradead.org> # documentation bits
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241002-mgtime-v10-1-d1c4717f5284@kernel.org
ktime_mono_to_any() only fetches the offset inside the loop. This is a
single word on 64-bit CPUs, and seqcount_read_begin() implies a full SMP
barrier.
Use READ_ONCE() to fetch the offset instead of doing a seqcount loop on
64-bit and add the matching WRITE_ONCE()'s to update the offsets in
tk_set_wall_to_mono() and tk_update_sleep_time().
[ tglx: Get rid of the #ifdeffery ]
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20240910-mgtime-v3-1-84406ed53fad@kernel.org
For tracing purpose, the boot clock is interesting as it doesn't stop on
suspend. Export it as part of the time snapshot. This will later allow
the hypervisor to add boot clock timestamps to its events.
Signed-off-by: Vincent Donnefort <vdonnefort@google.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20240911093029.3279154-5-vdonnefort@google.com
