clocksource/drivers/arm_arch_timer: Use of_property_present() for non-boolean properties

The use of of_property_read_bool() for non-boolean properties is
deprecated in favor of of_property_present() when testing for property
presence.

Signed-off-by: Rob Herring (Arm) <robh@kernel.org>
Link: https://lore.kernel.org/r/20241104190505.272805-2-robh@kernel.org
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>

clocksource/drivers/gpx: Remove redundant casts

In the function gxp_timer_init, the 'int' type cast in front of the
PTR_ERR() macro is redundant, thus remove it.

Signed-off-by: Tang Bin <tangbin@cmss.chinamobile.com>
Link: https://lore.kernel.org/r/20241107074619.2714-1-tangbin@cmss.chinamobile.com
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>

clocksource/drivers/timer-ti-dm: Fix child node refcount handling

of_find_compatible_node() increments the node's refcount, and it must be
decremented again with a call to of_node_put() when the pointer is no
longer required to avoid leaking the resource.

Instead of adding the missing calls to of_node_put() in all execution
paths, use the cleanup attribute for 'arm_timer' by means of the
__free() macro, which automatically calls of_node_put() when the
variable goes out of scope.

Fixes: 25de4ce5ed02 ("clocksource/drivers/timer-ti-dm: Handle dra7 timer wrap errata i940")
Signed-off-by: Javier Carrasco <javier.carrasco.cruz@gmail.com>
Link: https://lore.kernel.org/r/20241031-timer-ti-dm-systimer-of_node_put-v3-1-063ee822b73a@gmail.com
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>

dt-bindings: timer: actions,owl-timer: convert to YAML

Convert the Actions Semi Owl timer bindings to DT schema.

Changes during conversion:
 - Add a description
 - Add "clocks" as a required property, since the driver searches for it
 - Correct the given example according to owl-s500.dtsi

Signed-off-by: Ivaylo Ivanov <ivo.ivanov.ivanov1@gmail.com>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Link: https://lore.kernel.org/r/20241103123513.2890107-1-ivo.ivanov.ivanov1@gmail.com
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>

clocksource/drivers/ralink: Add Ralink System Tick Counter driver

System Tick Counter is present on Ralink SoCs RT3352 and MT7620. This
driver has been in 'arch/mips/ralink' directory since the beggining of
Ralink architecture support. However, it can be moved into a more proper
place in 'drivers/clocksource'. Hence add it here adding also support for
compile test targets and reducing LOC in architecture code folder.

Signed-off-by: Sergio Paracuellos <sergio.paracuellos@gmail.com>
Link: https://lore.kernel.org/r/20241028203643.191268-2-sergio.paracuellos@gmail.com
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>

clocksource/drivers/mips-gic-timer: Always use cluster 0 counter as clocksource

In a multi-cluster MIPS system, there are multiple GICs - one in each
cluster - each of which has its independent counter. The counters in
each GIC are not synchronized in any way, so they can drift relative
to one another through the lifetime of the system. This is problematic
for a clock source which ought to be global.

Avoid problems by always accessing cluster 0's counter, using
cross-cluster register access. This adds overhead so it is applied only
on multi-cluster systems.

Signed-off-by: Paul Burton <paulburton@kernel.org>
Signed-off-by: Chao-ying Fu <cfu@wavecomp.com>
Signed-off-by: Dragan Mladjenovic <dragan.mladjenovic@syrmia.com>
Signed-off-by: Aleksandar Rikalo <arikalo@gmail.com>
Tested-by: Serge Semin <fancer.lancer@gmail.com>
Acked-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Tested-by: Gregory CLEMENT <gregory.clement@bootlin.com>
Link: https://lore.kernel.org/r/20241019071037.145314-6-arikalo@gmail.com
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>

clocksource/drivers/timer-ti-dm: Don't fail probe if int not found

Some timers may not have an interrupt routed to the A53 GIC, but the
timer PWM functionality can still be used by Linux Kernel. Therefore,
do not fail probe if interrupt is not found and ti,timer-pwm exists.

Signed-off-by: Judith Mendez <jm@ti.com>
Link: https://lore.kernel.org/r/20241011175203.1040568-1-jm@ti.com
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>

clocksource/drivers:sp804: Make user selectable

The sp804 is currently only user selectable if COMPILE_TEST, this was
done by commit dfc82faad725 ("clocksource/drivers/sp804: Add
COMPILE_TEST to CONFIG_ARM_TIMER_SP804") in order to avoid it being
spuriously offered on platforms that won't have the hardware since it's
generally only seen on Arm based platforms.  This config is overly
restrictive, while platforms that rely on the SP804 do select it in
their Kconfig there are others such as the Arm fast models which have a
SP804 available but currently unused by Linux.  Relax the dependency to
allow it to be user selectable on arm and arm64 to avoid surprises and
in case someone comes up with a use for extra timer hardware.

Fixes: dfc82faad725 ("clocksource/drivers/sp804: Add COMPILE_TEST to CONFIG_ARM_TIMER_SP804")
Reported-by: Ross Burton <ross.burton@arm.com>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Link: https://lore.kernel.org/r/20241001-arm64-vexpress-sp804-v3-1-0a2d3f7883e4@kernel.org
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>

clocksource/drivers/dw_apb: Remove unused dw_apb_clockevent functions

dw_apb_clockevent_pause(), dw_apb_clockevent_resume() and
dw_apb_clockevent_stop() have been unused since 2021's
commit 1b79fc4f2bfd ("x86/apb_timer: Remove driver for deprecated
platform")

Remove them.

(Some of the other clockevent functions are still called by
dw_apb_timer_of.c  so I guess it is still in use?)

Signed-off-by: Dr. David Alan Gilbert <linux@treblig.org>
Link: https://lore.kernel.org/r/20241025203101.241709-1-linux@treblig.org
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>

hrtimers: Delete hrtimer_init_on_stack()

hrtimer_init_on_stack() is now unused. Delete it.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/510ce0d2944c4a382ea51e51d03dcfb73ba0f4f7.1730386209.git.namcao@linutronix.de

alarmtimer: Switch to use hrtimer_setup() and hrtimer_setup_on_stack()

hrtimer_setup() and hrtimer_setup_on_stack() take the callback function
pointer as argument and initialize the timer completely.

Replace the hrtimer_init*() variants and the open coded initialization of
hrtimer::function with the new setup mechanism.

Switch to use the new functions.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/2bae912336103405adcdab96b88d3ea0353b4228.1730386209.git.namcao@linutronix.de

io_uring: Switch to use hrtimer_setup_on_stack()

hrtimer_setup_on_stack() takes the callback function pointer as argument
and initializes the timer completely.

Replace hrtimer_init_on_stack() and the open coded initialization of
hrtimer::function with the new setup mechanism.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/f0d4ac32ec4050710a656cee8385fa4427be33aa.1730386209.git.namcao@linutronix.de

sched/idle: Switch to use hrtimer_setup_on_stack()

hrtimer_setup_on_stack() takes the callback function pointer as argument
and initializes the timer completely.

Replace hrtimer_init_on_stack() and the open coded initialization of
hrtimer::function with the new setup mechanism.

The conversion was done with Coccinelle.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/17f9421fed6061df4ad26a4cc91873d2c078cb0f.1730386209.git.namcao@linutronix.de

hrtimers: Delete hrtimer_init_sleeper_on_stack()

hrtimer_init_sleeper_on_stack() is now unused. Delete it.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/52549846635c0b3a2abf82101f539efdabcd9778.1730386209.git.namcao@linutronix.de

wait: Switch to use hrtimer_setup_sleeper_on_stack()

hrtimer_setup_sleeper_on_stack() replaces hrtimer_init_sleeper_on_stack()
to keep the naming convention consistent.

Convert the usage site over to it.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/fc91182375df81120a88dbe0263267e24d1bf19e.1730386209.git.namcao@linutronix.de

timers: Switch to use hrtimer_setup_sleeper_on_stack()

hrtimer_setup_sleeper_on_stack() replaces hrtimer_init_sleeper_on_stack()
to keep the naming convention consistent.

Convert the usage sites over to it. The conversion was done with
Coccinelle.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/299c07f0f96af8ab3a7631b47b6ca22b06b20577.1730386209.git.namcao@linutronix.de

net: pktgen: Switch to use hrtimer_setup_sleeper_on_stack()

hrtimer_setup_sleeper_on_stack() replaces hrtimer_init_sleeper_on_stack()
to keep the naming convention consistent.

Convert the usage site over to it. The conversion was done with Coccinelle.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/c4b40b8fef250b6a325e1b8bd6057005fb3cb660.1730386209.git.namcao@linutronix.de

futex: Switch to use hrtimer_setup_sleeper_on_stack()

hrtimer_setup_sleeper_on_stack() replaces hrtimer_init_sleeper_on_stack()
to keep the naming convention consistent.

Convert the usage site over to it. The conversion was done with Coccinelle.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/d92116a17313dee283ebc959869bea80fbf94cdb.1730386209.git.namcao@linutronix.de

fs/aio: Switch to use hrtimer_setup_sleeper_on_stack()

hrtimer_setup_sleeper_on_stack() replaces hrtimer_init_sleeper_on_stack()
to keep the naming convention consistent.

Convert the usage site over to it. The conversion was done with Coccinelle.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/5f10c259fa43ba2fe774de5b2cedc22f5e9cfd2d.1730386209.git.namcao@linutronix.de

hrtimers: Introduce hrtimer_update_function()

Some users of hrtimer need to change the callback function after the
initial setup. They write to hrtimer::function directly.

That's not safe under all circumstances as the write is lockless and a
concurrent timer expiry might end up using the wrong function pointer.

Introduce hrtimer_update_function(), which also performs runtime checks
whether it is safe to modify the callback.

This allows to make hrtimer::function private once all users are converted.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20a937b0ae09ad54b5b6d86eabead7c570f1b72e.1730386209.git.namcao@linutronix.de

hrtimers: Introduce hrtimer_setup_sleeper_on_stack()

The hrtimer_init*() API is replaced by hrtimer_setup*() variants to
initialize the timer including the callback function at once.

hrtimer_init_sleeper_on_stack() does not need user to setup the callback
function separately, so a new variant would not be strictly necessary.

Nonetheless, to keep the naming convention consistent, introduce
hrtimer_setup_sleeper_on_stack(). hrtimer_init_on_stack() will be removed
once all users are converted.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/7b5e18e6dd0ace9eaa211201528cb9dc23752454.1730386209.git.namcao@linutronix.de

hrtimers: Introduce hrtimer_setup_on_stack()

To initialize hrtimer on stack, hrtimer_init_on_stack() needs to be called
and also hrtimer::function must be set. This is error-prone and awkward to
use.

Introduce hrtimer_setup_on_stack() which does both of these things, so that
users of hrtimer can be simplified.

The new setup function also has a sanity check for the provided function
pointer. If NULL, a warning is emitted and a dummy callback installed.

hrtimer_init_on_stack() will be removed as soon as all of its users have
been converted to the new function.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/4b05e2ab3a82c517adf67fabc0f0cd8fe118b97c.1730386209.git.namcao@linutronix.de

hrtimers: Introduce hrtimer_setup() to replace hrtimer_init()

To initialize hrtimer, hrtimer_init() needs to be called and also
hrtimer::function must be set. This is error-prone and awkward to use.

Introduce hrtimer_setup() which does both of these things, so that users of
hrtimer can be simplified.

The new setup function also has a sanity check for the provided function
pointer. If NULL, a warning is emitted and a dummy callback installed.

hrtimer_init() will be removed as soon as all of its users have been
converted to the new function.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/5057c1ddbfd4b92033cd93d37fe38e6b069d5ba6.1730386209.git.namcao@linutronix.de

io_uring: Remove redundant hrtimer's callback function setup

The IORING_OP_TIMEOUT command uses hrtimer underneath. The timer's callback
function is setup in io_timeout(), and then the callback function is setup
again when the timer is rearmed.

Since the callback function is the same for both cases, the latter setup is
redundant, therefore remove it.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Jens Axboe <axboe@kernel.dk:
Link: https://lore.kernel.org/all/07b28dfd5691478a2d250f379c8b90dd37f9bb9a.1730386209.git.namcao@linutronix.de

_RESEND_PATCH_v2_04_19_wifi_rt2x00_Remove_redundant_hrtimer_init_

rt2x00usb_probe() executes a hrtimer_init() for txstatus_timer. Afterwards,
rt2x00lib_probe_dev() is called which also initializes this txstatus_timer
with the same settings.

Remove the redundant hrtimer_init() call in rt2x00usb_probe().

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Kalle Valo <kvalo@kernel.org>
Link: https://lore.kernel.org/all/66116057f788e18a6603d50a554417eee459e02c.1730386209.git.namcao@linutronix.de

KVM: x86/xen: Initialize hrtimer in kvm_xen_init_vcpu()

The hrtimer is initialized in the KVM_XEN_VCPU_SET_ATTR ioctl. That caused
problem in the past, because the hrtimer can be initialized multiple times,
which was fixed by commit af735db31285 ("KVM: x86/xen: Initialize Xen timer
only once"). This commit avoids initializing the timer multiple times by
checking the field 'function' of struct hrtimer to determine if it has
already been initialized.

This is not required and in the way to make the function field private.

Move the hrtimer initialization into kvm_xen_init_vcpu() so that it will
only be initialized once.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Sean Christopherson <seanjc@google.com>
Link: https://lore.kernel.org/all/9c33c7224d97d08f4fa30d3cc8687981c1d3e953.1730386209.git.namcao@linutronix.de

drm/i915/request: Remove unnecessary modification of hrtimer:: Function

When a request is created, the hrtimer is not initialized and only its
'function' field is set to NULL. The hrtimer is only initialized when the
request is enqueued. The point of setting 'function' to NULL is that, it
can be used to check whether hrtimer_try_to_cancel() should be called while
retiring the request.

This "trick" is unnecessary, because hrtimer_try_to_cancel() already does
its own check whether the timer is armed. If the timer is not armed,
hrtimer_try_to_cancel() returns 0.

Fully initialize the timer when the request is created, which allows to
make the hrtimer::function field private once all users of hrtimer_init()
are converted to hrtimer_setup(), which requires a valid callback function
to be set.

Because hrtimer_try_to_cancel() returns 0 if the timer is not armed, the
logic to check whether to call i915_request_put() remains equivalent.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/50f865045aa672a9730343ad131543da332b1d8d.1730386209.git.namcao@linutronix.de

hrtimers: Add missing hrtimer_init() trace points

hrtimer_init*_on_stack() is not covered by tracing when
CONFIG_DEBUG_OBJECTS_TIMERS=y.

Rework the functions similar to hrtimer_init() and hrtimer_init_sleeper()
so that the hrtimer_init() tracepoint is unconditionally available.

The rework makes hrtimer_init_sleeper() unused. Delete it.

Signed-off-by: Nam Cao <namcao@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/74528e8abf2bb96e8bee85ffacbf14e15cf89f0d.1730386209.git.namcao@linutronix.de

alarmtimers: Remove return value from alarm functions

Now that the SIG_IGN problem is solved in the core code, the alarmtimer
callbacks do not require a return value anymore.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241105064214.318837272@linutronix.de

alarmtimers: Remove the throttle mechanism from alarm_forward_now()

Now that ignored posix timer signals are requeued and the timers are
rearmed on signal delivery the workaround to keep such timers alive and
self rearm them is not longer required.

Remove the unused alarm timer parts.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064214.252443020@linutronix.de

posix-timers: Cleanup SIG_IGN workaround leftovers

Now that ignored posix timer signals are requeued and the timers are
rearmed on signal delivery the workaround to keep such timers alive and
self rearm them is not longer required.

Remove the relevant hacks and the not longer required return values from
the related functions. The alarm timer workarounds will be cleaned up in a
separate step.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064214.187239060@linutronix.de

signal: Queue ignored posixtimers on ignore list

Queue posixtimers which have their signal ignored on the ignored list:

   1) When the timer fires and the signal has SIG_IGN set

   2) When SIG_IGN is installed via sigaction() and a timer signal
      is already queued

This only happens when the signal is for a valid timer, which delivered the
signal in periodic mode. One-shot timer signals are correctly dropped.

Due to the lock order constraints (sighand::siglock nests inside
timer::lock) the signal code cannot access any of the timer fields which
are relevant to make this decision, e.g. timer::it_status.

This is addressed by establishing a protection scheme which requires to
lock both locks on the timer side for modifying decision fields in the
timer struct and therefore makes it possible for the signal delivery to
evaluate with only sighand:siglock being held:

  1) Move the NULLification of timer->it_signal into the sighand::siglock
     protected section of timer_delete() and check timer::it_signal in the
     code path which determines whether the signal is dropped or queued on
     the ignore list.

     This ensures that a deleted timer cannot be moved onto the ignore
     list, which would prevent it from being freed on exit() as it is not
     longer in the process' posix timer list.

     If the timer got moved to the ignored list before deletion then it is
     removed from the ignored list under sighand lock in timer_delete().

  2) Provide a new timer::it_sig_periodic flag, which gets set in the
     signal queue path with both timer and sighand locks held if the timer
     is actually in periodic mode at expiry time.

     The ignore list code checks this flag under sighand::siglock and drops
     the signal when it is not set.

     If it is set, then the signal is moved to the ignored list independent
     of the actual state of the timer.

     When the signal is un-ignored later then the signal is moved back to
     the signal queue. On signal delivery the posix timer side decides
     about dropping the signal if the timer was re-armed, dis-armed or
     deleted based on the signal sequence counter check.

     If the thread/process exits then not yet delivered signals are
     discarded which means the reference of the timer containing the
     sigqueue is dropped and frees the timer.

     This is way cheaper than requiring all code paths to lock
     sighand::siglock of the target thread/process on any modification of
     timer::it_status or going all the way and removing pending signals
     from the signal queues on every rearm, disarm or delete operation.

So the protection scheme here is that on the timer side both timer::lock
and sighand::siglock have to be held for modifying

   timer::it_signal
   timer::it_sig_periodic

which means that on the signal side holding sighand::siglock is enough to
evaluate these fields.

In posixtimer_deliver_signal() holding timer::lock is sufficient to do the
sequence validation against timer::it_signal_seq because a concurrent
expiry is waiting on timer::lock to be released.

This completes the SIG_IGN handling and such timers are not longer self
rearmed which avoids pointless wakeups.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064214.120756416@linutronix.de

signal: Handle ignored signals in do_sigaction(action != SIG_IGN)

When a real handler (including SIG_DFL) is installed for a signal, which
had previously SIG_IGN set, then the list of ignored posix timers has to be
checked for timers which are affected by this change.

Add a list walk function which checks for the matching signal number and if
found requeues the timers signal, so the timer is rearmed on signal
delivery.

Rearming the timer right away is not possible because that requires to drop
sighand lock.

No functional change as the counter part which queues the timers on the
ignored list is still missing.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064214.054091076@linutronix.de

posix-timers: Handle ignored list on delete and exit

To handle posix timer signals on sigaction(SIG_IGN) properly, the timers
will be queued on a separate ignored list.

Add the necessary cleanup code for timer_delete() and exit_itimers().

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064213.987530588@linutronix.de

signal: Provide ignored_posix_timers list

To prepare for handling posix timer signals on sigaction(SIG_IGN) properly,
add a list to task::signal.

This list will be used to queue posix timers so their signal can be
requeued when SIG_IGN is lifted later.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064213.920101900@linutronix.de

posix-timers: Move sequence logic into struct k_itimer

The posix timer signal handling uses siginfo::si_sys_private for handling
the sequence counter check. That indirection is not longer required and the
sequence count value at signal queueing time can be stored in struct
k_itimer itself.

This removes the requirement of treating siginfo::si_sys_private special as
it's now always zero as the kernel does not touch it anymore.

Suggested-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Link: https://lore.kernel.org/all/20241105064213.852619866@linutronix.de

signal: Cleanup unused posix-timer leftovers

Remove the leftovers of sigqueue preallocation as it's not longer used.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064213.786506636@linutronix.de

posix-timers: Embed sigqueue in struct k_itimer

To cure the SIG_IGN handling for posix interval timers, the preallocated
sigqueue needs to be embedded into struct k_itimer to prevent life time
races of all sorts.

Now that the prerequisites are in place, embed the sigqueue into struct
k_itimer and fixup the relevant usage sites.

Aside of preparing for proper SIG_IGN handling, this spares an extra
allocation.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064213.719695194@linutronix.de

signal: Replace resched_timer logic

In preparation for handling ignored posix timer signals correctly and
embedding the sigqueue struct into struct k_itimer, hand down a pointer to
the sigqueue struct into posix_timer_deliver_signal() instead of just
having a boolean flag.

No functional change.

Suggested-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Link: https://lore.kernel.org/all/20241105064213.652658158@linutronix.de

signal: Refactor send_sigqueue()

To handle posix timers which have their signal ignored via SIG_IGN properly
it is required to requeue a ignored signal for delivery when SIG_IGN is
lifted so the timer gets rearmed.

Split the required code out of send_sigqueue() so it can be reused in
context of sigaction().

While at it rename send_sigqueue() to posixtimer_send_sigqueue() so its
clear what this is about.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064213.586453412@linutronix.de

posix-timers: Store PID type in the timer

instead of re-evaluating the signal delivery mode everywhere.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064213.519086500@linutronix.de

signal: Provide posixtimer_sigqueue_init()

To cure the SIG_IGN handling for posix interval timers, the preallocated
sigqueue needs to be embedded into struct k_itimer to prevent life time
races of all sorts.

Provide a new function to initialize the embedded sigqueue to prepare for
that.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064213.450427515@linutronix.de

signal: Split up __sigqueue_alloc()

To cure the SIG_IGN handling for posix interval timers, the preallocated
sigqueue needs to be embedded into struct k_itimer to prevent life time
races of all sorts.

Reorganize __sigqueue_alloc() so the ucounts retrieval and the
initialization can be used independently.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064213.371410037@linutronix.de

posix-timers: Add a refcount to struct k_itimer

To cure the SIG_IGN handling for posix interval timers, the preallocated
sigqueue needs to be embedded into struct k_itimer to prevent life time
races of all sorts.

To make that work correctly it needs reference counting so that timer
deletion does not free the timer prematuraly when there is a signal queued
or delivered concurrently.

Add a rcuref to the posix timer part.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064213.304756440@linutronix.de

posix-cpu-timers: Use dedicated flag for CPU timer nanosleep

POSIX CPU timer nanosleep creates a k_itimer on stack and uses the sigq
pointer to detect the nanosleep case in the expiry function.

Prepare for embedding sigqueue into struct k_itimer by using a dedicated
flag for nanosleep.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064213.238550394@linutronix.de

posix-cpu-timers: Cleanup the firing logic

The firing flag of a posix CPU timer is tristate:

  0: when the timer is not about to deliver a signal

  1: when the timer has expired, but the signal has not been delivered yet

 -1: when the timer was queued for signal delivery and a rearm operation
     raced against it and supressed the signal delivery.

This is a pointless exercise as this can be simply expressed with a
boolean. Only if set, the signal is delivered. This makes delete and rearm
consistent with the rest of the posix timers.

Convert firing to bool and fixup the usage sites accordingly and add
comments why the timer cannot be dequeued right away.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241105064213.172848618@linutronix.de

posix-timers: Make signal overrun accounting sensible

The handling of the timer overrun in the signal code is inconsistent as it
takes previous overruns into account. This is just wrong as after the
reprogramming of a timer the overrun count starts over from a clean state,
i.e. 0.

Don't touch info::si_overrun in send_sigqueue() and only store the overrun
value at signal delivery time, which is computed from the timer itself
relative to the expiry time.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241105064213.106738193@linutronix.de

posix-timers: Make signal delivery consistent

Signals of timers which are reprogammed, disarmed or deleted can deliver
signals related to the past. The POSIX spec is blury about this:

 - "The effect of disarming or resetting a timer with pending expiration
    notifications is unspecified."

 - "The disposition of pending signals for the deleted timer is
    unspecified."

In both cases it is reasonable to expect that pending signals are
discarded. Especially in the reprogramming case it does not make sense to
account for previous overruns or to deliver a signal for a timer which has
been disarmed. This makes the behaviour consistent and understandable.

Remove the si_sys_private check from the signal delivery code and invoke
posix_timer_deliver_signal() unconditionally for posix timer related
signals.

Change posix_timer_deliver_signal() so it controls the actual signal
delivery via the return value. It now instructs the signal code to drop the
signal when:

  1) The timer does not longer exist in the hash table

  2) The timer signal_seq value is not the same as the si_sys_private value
     which was set when the signal was queued.

This is also a preparatory change to embed the sigqueue into the k_itimer
structure, which in turn allows to remove the si_sys_private magic.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241105064213.040348644@linutronix.de

posix-cpu-timers: Correctly update timer status in posix_cpu_timer_del()

If posix_cpu_timer_del() exits early due to task not found or sighand
invalid, it fails to clear the state of the timer. That's harmless but
inconsistent.

These early exits are accounted as successful delete. Move the update of
the timer state into the success return path, so all "successful" deletions
are handled.

Reported-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241105064212.974053438@linutronix.de

timekeeping: Always check for negative motion

clocksource_delta() has two variants. One with a check for negative motion,
which is only selected by x86. This is a historic leftover as this function
was previously used in the time getter hot paths.

Since 135225a363ae timekeeping_cycles_to_ns() has unconditional protection
against this as a by-product of the protection against 64bit math overflow.

clocksource_delta() is only used in the clocksource watchdog and in
timekeeping_advance(). The extra conditional there is not hurting anyone.

Remove the config option and unconditionally prevent negative motion of the
readout.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241031120328.599430157@linutronix.de

timekeeping: Remove CONFIG_DEBUG_TIMEKEEPING

Since 135225a363ae timekeeping_cycles_to_ns() handles large offsets which
would lead to 64bit multiplication overflows correctly. It's also protected
against negative motion of the clocksource unconditionally, which was
exclusive to x86 before.

timekeeping_advance() handles large offsets already correctly.

That means the value of CONFIG_DEBUG_TIMEKEEPING which analyzed these cases
is very close to zero. Remove all of it.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Stultz <jstultz@google.com>
Link: https://lore.kernel.org/all/20241031120328.536010148@linutronix.de

timers: Add missing READ_ONCE() in __run_timer_base()

__run_timer_base() checks base::next_expiry without holding
base::lock. That can race with a remote CPU updating next_expiry under the
lock. This is an intentional and harmless data race, but lacks a
READ_ONCE(), so KCSAN complains about this.

Add the missing READ_ONCE(). All other places are covered already.

Fixes: 79f8b28e85f8 ("timers: Annotate possible non critical data race of next_expiry")
Reported-by: kernel test robot <oliver.sang@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/87a5emyqk0.ffs@tglx
Closes: https://lore.kernel.org/oe-lkp/202410301205.ef8e9743-lkp@intel.com

clocksource/drivers/timer-tegra: Remove clockevents shutdown call on offlining

The clockevents core already detached and unregistered it at this stage.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241029125451.54574-11-frederic@kernel.org

clocksource/drivers/qcom: Remove clockevents shutdown call on offlining

The clockevents core already detached and unregistered it at this stage.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241029125451.54574-10-frederic@kernel.org

clocksource/drivers/armada-370-xp: Remove clockevents shutdown call on offlining

The clockevents core already detached and unregistered it at this stage.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241029125451.54574-9-frederic@kernel.org

clocksource/drivers/exynos_mct: Remove clockevents shutdown call on offlining

The clockevents core already detached and unregistered it at this stage.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241029125451.54574-8-frederic@kernel.org

clocksource/drivers/arm_global_timer: Remove clockevents shutdown call on offlining

The clockevents core already detached and unregistered it at this stage.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241029125451.54574-7-frederic@kernel.org

clocksource/drivers/arm_arch_timer: Remove clockevents shutdown call on offlining

The clockevents core already detached and unregistered it at this stage.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241029125451.54574-6-frederic@kernel.org

ARM: smp_twd: Remove clockevents shutdown call on offlining

The clockevents core already detached and unregistered it at this stage.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241029125451.54574-5-frederic@kernel.org

tick: Remove now unneeded low-res tick stop on CPUHP_AP_TICK_DYING

The generic clockevent layer now detaches and stops the underlying
clockevent from the dying CPU, unifying the tick behaviour for both
periodic and oneshot mode on offline CPUs. There is no more need for
the tick layer to care about that.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241029125451.54574-4-frederic@kernel.org

clockevents: Shutdown and unregister current clockevents at CPUHP_AP_TICK_DYING

The way the clockevent devices are finally stopped while a CPU is
offlining is currently chaotic. The layout being by order:

1) tick_sched_timer_dying() stops the tick and the underlying clockevent
  but only for oneshot case. The periodic tick and its related
  clockevent still runs.

2) tick_broadcast_offline() detaches and stops the per-cpu oneshot
  broadcast and append it to the released list.

3) Some individual clockevent drivers stop the clockevents (a second time if
  the tick is oneshot)

4) Once the CPU is dead, a control CPU remotely detaches and stops
  (a 3rd time if oneshot mode) the CPU clockevent and adds it to the
  released list.

5) The released list containing the broadcast device released on step 2)
   and the remotely detached clockevent from step 4) are unregistered.

These random events can be factorized if the current clockevent is
detached and stopped by the dying CPU at the generic layer, that is
from the dying CPU:

a) Stop the tick
b) Stop/detach the underlying per-cpu oneshot broadcast clockevent
c) Stop/detach the underlying clockevent
d) Release / unregister the clockevents from b) and c)
e) Release / unregister the remaining clockevents from the dying CPU.
   This part could be performed by the dying CPU

This way the drivers and the tick layer don't need to care about
clockevent operations during cpuhotplug down. This also unifies the tick
behaviour on offline CPUs between oneshot and periodic modes, avoiding
offline ticks altogether for sanity.

Adopt the simplification.

[ tglx: Remove the WARN_ON() in clockevents_register_device() as that
   is called from an upcoming CPU before the CPU is marked online ]

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241029125451.54574-3-frederic@kernel.org

clockevents: Improve clockevents_notify_released() comment

When a new clockevent device is added and replaces a previous device,
the latter is put into the released list. Then the released list is
added back.

This may look counter-intuitive but the reason is that released device
might be suitable for other uses. For example a released CPU regular
clockevent can be a better replacement for the current broadcast event.
Similarly a released broadcast clockevent can be a better replacement
for the current regular clockevent of a given CPU.

Improve comments stating about these subtleties.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241029125451.54574-2-frederic@kernel.org

jiffies: Define secs_to_jiffies()

secs_to_jiffies() is defined in hci_event.c and cannot be reused by
other call sites. Hoist it into the core code to allow conversion of the
~1150 usages of msecs_to_jiffies() that either:

 - use a multiplier value of 1000 or equivalently MSEC_PER_SEC, or
 - have timeouts that are denominated in seconds (i.e. end in 000)

It's implemented as a macro to allow usage in static initializers.

This will also allow conversion of yet more sites that use (sec * HZ)
directly, and improve their readability.

Suggested-by: Michael Kelley <mhklinux@outlook.com>
Signed-off-by: Easwar Hariharan <eahariha@linux.microsoft.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Link: https://lore.kernel.org/all/20241030-open-coded-timeouts-v3-1-9ba123facf88@linux.microsoft.com

posix-timers: Add proper state tracking

Right now the state tracking is done by two struct members:

 - it_active:
     A boolean which tracks armed/disarmed state

 - it_signal_seq:
     A sequence counter which is used to invalidate settings
     and prevent rearming

Replace it_active with it_status and keep properly track about the states
in one place.

This allows to reuse it_signal_seq to track reprogramming, disarm and
delete operations in order to drop signals which are related to the state
previous of those operations.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241001083835.670337048@linutronix.de

posix-timers: Rename k_itimer:: It_requeue_pending

Prepare for using this struct member to do a proper reprogramming and
deletion accounting so that stale signals can be dropped.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241001083835.611997737@linutronix.de

posix-timers: Drop signal if timer has been deleted or reprogrammed

No point in delivering a signal from the past. POSIX does not specify the
behaviour here:

 - "The effect of disarming or resetting a timer with pending expiration
    notifications is unspecified."

 - "The disposition of pending signals for the deleted timer is unspecified."

In both cases it is reasonable to expect that pending signals are
discarded. Especially in the reprogramming case it does not make sense to
account for previous overruns or to deliver a signal for a timer which has
been disarmed.

Drop the signal as that is conistent and understandable behaviour.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241001083835.553646280@linutronix.de

signal: Allow POSIX timer signals to be dropped

In case that a timer was reprogrammed or deleted an already pending signal
is obsolete. Right now such signals are kept around and eventually
delivered. While POSIX is blury about this:

 - "The effect of disarming or resetting a timer with pending expiration
    notifications is unspecified."

 - "The disposition of pending signals for the deleted timer is
    unspecified."

it is reasonable in both cases to expect that pending signals are discarded
as they have no meaning anymore.

Prepare the signal code to allow dropping posix timer signals.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241001083835.494416923@linutronix.de

posix-timers: Cure si_sys_private race

The si_sys_private member of the siginfo which is embedded in the
preallocated sigqueue is used by the posix timer code to decide whether a
timer must be reprogrammed on signal delivery.

The handling of this is racy as a long standing comment in that code
documents. It is modified with the timer lock held, but without sighand
lock being held. The actual signal delivery code checks for it under
sighand lock without holding the timer lock.

Hand the new value to send_sigqueue() as argument and store it with sighand
lock held. This is an intermediate change to address this issue.

The arguments to this function will be cleanup in subsequent changes.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241001083835.434338954@linutronix.de

signal: Cleanup flush_sigqueue_mask()

Mop up the stale return value comment and add a lockdep check instead of
commenting on the locking requirement.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/20241001083835.374933959@linutronix.de

signal: Confine POSIX_TIMERS properly

Move the itimer rearming out of the signal code and consolidate all posix
timer related functions in the signal code under one ifdef.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/all/20241001083835.314100569@linutronix.de

time: Fix references to _msecs_to_jiffies() handling of values

The details about the handling of the "normal" values were moved
to the _msecs_to_jiffies() helpers in commit ca42aaf0c861 ("time:
Refactor msecs_to_jiffies"). However, the same commit still mentioned
__msecs_to_jiffies() in the added documentation.

Thus point to _msecs_to_jiffies() instead.

Fixes: ca42aaf0c861 ("time: Refactor msecs_to_jiffies")
Signed-off-by: Miguel Ojeda <ojeda@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241025110141.157205-2-ojeda@kernel.org

time: Partially revert cleanup on msecs_to_jiffies() documentation

The documentation's intention is to compare msecs_to_jiffies() (first
sentence) with __msecs_to_jiffies() (second sentence), which is what the
original documentation did. One of the cleanups in commit f3cb80804b82
("time: Fix various kernel-doc problems") may have thought the paragraph
was talking about the latter since that is what it is being documented.

Thus revert that part of the change.

Fixes: f3cb80804b82 ("time: Fix various kernel-doc problems")
Signed-off-by: Miguel Ojeda <ojeda@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241025110141.157205-1-ojeda@kernel.org

timekeeping: Merge timekeeping_update_staged() and timekeeping_update()

timekeeping_update_staged() is the only call site of timekeeping_update().

Merge those functions. 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-v2-25-554456a44a15@linutronix.de

timekeeping: Remove TK_MIRROR timekeeping_update() action

All call sites of using TK_MIRROR flag in timekeeping_update() are
gone. The TK_MIRROR dependent code path is therefore dead code.

Remove it along with the TK_MIRROR define.

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-v2-24-554456a44a15@linutronix.de

timekeeping: Rework do_adjtimex() to use shadow_timekeeper

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_adjtimex() to use this scheme and take the opportunity to use a
scoped_guard() for locking.

That requires to have a separate function for updating the leap state so
that the update is protected by the sequence count. This also brings the
timekeeper and the shadow timekeeper in sync for this state, which was not
the case so far. That's not a correctness problem as the state is only used
at the read sides which use the real timekeeper, but it's inconsistent
nevertheless.

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-v2-23-554456a44a15@linutronix.de

timekeeping: Rework timekeeping_suspend() to use shadow_timekeeper

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.

While the sequence count held time is not relevant for the resume path as
there is no concurrency, there is no reason to have this function
different than all the other update sites.

Convert timekeeping_inject_offset() to use this scheme and cleanup the
variable declarations while at it.

As halt_fast_timekeeper() does not need protection sequence counter, it is
no problem to move it with this change outside of the sequence counter
protected area. But it still needs to be executed while holding the lock.

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-v2-22-554456a44a15@linutronix.de

timekeeping: Rework timekeeping_resume() to use shadow_timekeeper

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.

While the sequence count held time is not relevant for the resume path as
there is no concurrency, there is no reason to have this function
different than all the other update sites.

Convert timekeeping_inject_offset() to use this scheme and cleanup the
variable declaration while at it.

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-v2-21-554456a44a15@linutronix.de

timekeeping: Rework timekeeping_inject_sleeptime64() to use shadow_timekeeper

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_sleeptime64() 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-v2-20-554456a44a15@linutronix.de

timekeeping: Rework timekeeping_init() to use shadow_timekeeper

For timekeeping_init() the sequence count write held time is not relevant
and it could keep working on the real timekeeper, but there is no reason to
make it different from other timekeeper updates.

Convert it to operate on the shadow timekeeper.

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-v2-19-554456a44a15@linutronix.de

timekeeping: Rework change_clocksource() to use shadow_timekeeper

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

timekeeping: Rework timekeeping_inject_offset() to use shadow_timekeeper

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

timekeeping: Rework do_settimeofday64() to use shadow_timekeeper

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

timekeeping: Provide timekeeping_restore_shadow()

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: Introduce combined timekeeping action flag

Instead of explicitly listing all the separate timekeeping actions flags,
introduce a new one which covers all actions except TK_MIRROR action.

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-v2-14-554456a44a15@linutronix.de

timekeeping: Split out timekeeper update of timekeeping_advanced()

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

timekeeping: Add struct tk_data as argument to timekeeping_update()

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

timekeeping: Introduce tkd_basic_setup() to make lock and seqcount init reusable

Initialization of lock and seqcount needs to be done for every instance of
timekeeper struct. To be able to easily reuse it, create a separate
function for it.

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-v2-11-554456a44a15@linutronix.de

timekeeping: Define a struct type for tk_core to make it reusable

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

timekeeping: Move timekeeper_lock into tk_core

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

timekeeping: Encapsulate locking/unlocking of timekeeper_lock

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

timekeeping: Move shadow_timekeeper into tk_core

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

timekeeping: Reorder struct timekeeper

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: Simplify code in timekeeping_advance()

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

timekeeping: Abort clocksource change in case of failure

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

timekeeping: Avoid duplicate leap state update

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

timekeeping: Don't stop time readers across hard_pps() update

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

timekeeping: Read NTP tick length only once

No point in reading it a second time when the comparison 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-v2-1-554456a44a15@linutronix.de

posix-timers: Replace call_rcu() by kfree_rcu() for simple kmem_cache_free() callback

Since SLOB was removed and since commit 6c6c47b063b5 ("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

timers/Documentation: Cleanup delay/sleep documentation

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

media: anysee: Fix and remove outdated comment

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
1162c7b383a6 ("[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