The scheduler imposes a requirement to sched_clock()
which is to stop the clock during suspend, if we don't
do that any RT thread will be rescheduled in the future
which might cause any sort of problems.
This became an issue on OMAP when we converted omap-i2c.c
to use threaded IRQs, it turned out that depending on how
much time we spent on suspend, the I2C IRQ thread would
end up being rescheduled so far in the future that I2C
transfers would timeout and, because omap_hsmmc depends
on an I2C-connected device to detect if an MMC card is
inserted in the slot, our rootfs would just vanish.
arch/arm/kernel/sched_clock.c already had an optional
implementation (sched_clock_needs_suspend()) which would
handle scheduler's requirement properly, what this patch
does is simply to make that implementation non-optional.
Note that this has the side-effect that printk timings
won't reflect the actual time spent on suspend so other
methods to measure that will have to be used.
This has been tested with beagleboard XM (OMAP3630) and
pandaboard rev A3 (OMAP4430). Suspend to RAM is now working
after this patch.
Thanks to Kevin Hilman for helping out with debugging.
Acked-by: Kevin Hilman <khilman@ti.com>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Add support for irq time accounting. This commit prepares ARM by adding
the call to enable_sched_clock_irqtime() in sched_clock(). We introduce
a new kernel parameter - irqtime - which takes an integer. -1 for auto,
0 for disabled, and 1 for enabled. Auto mode selects IRQ accounting if
we have a sched_clock() tick rate greater than 1MHz.
Frederic Weisbecker is working on a patch set which moves the
IRQ_TIME_ACCOUNTING into arch/, so that part is not incorporated into
this patch; this facility becomes available on ARM only when both this
patch and Frederic's patches are merged.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Many clocks that are used to provide sched_clock will reset during
suspend. If read_sched_clock returns 0 after suspend, sched_clock will
appear to jump forward. This patch resets cd.epoch_cyc to the current
value of read_sched_clock during resume, which causes sched_clock() just
after suspend to return the same value as sched_clock() just before
suspend.
In addition, during the window where epoch_ns has been updated before
suspend, but epoch_cyc has not been updated after suspend, it is unknown
whether the clock has reset or not, and sched_clock() could return a
bogus value. Add a suspended flag, and return the pre-suspend epoch_ns
value during this period.
The new behavior is triggered by calling setup_sched_clock_needs_suspend
instead of setup_sched_clock.
Signed-off-by: Colin Cross <ccross@android.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Ensure that the software state for sched_clock() is updated at the
point of suspend so that we avoid losing ticks since the last update.
This prevents the platform dependent possibility that sched_clock()
may appear to go backwards across a suspend/resume cycle.
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
sched_clock() is yet another blocker on the road to the single
image. This patch implements an idea by Russell King:
http://www.spinics.net/lists/linux-omap/msg49561.html
Instead of asking the platform to implement both sched_clock()
itself and the rollover callback, simply register a read()
function, and let the ARM code care about sched_clock() itself,
the conversion to ns and the rollover. sched_clock() uses
this read() function as an indirection to the platform code.
If the platform doesn't provide a read(), the code falls back
to the jiffy counter (just like the default sched_clock).
This allow some simplifications and possibly some footprint gain
when multiple platforms are compiled in. Among the drawbacks,
the removal of the *_fixed_sched_clock optimization which could
negatively impact some platforms (sa1100, tegra, versatile
and omap).
Tested on 11MPCore, OMAP4 and Tegra.
Cc: Imre Kaloz <kaloz@openwrt.org>
Cc: Eric Miao <eric.y.miao@gmail.com>
Cc: Colin Cross <ccross@android.com>
Cc: Erik Gilling <konkers@android.com>
Cc: Olof Johansson <olof@lixom.net>
Cc: Sascha Hauer <kernel@pengutronix.de>
Cc: Alessandro Rubini <rubini@unipv.it>
Cc: STEricsson <STEricsson_nomadik_linux@list.st.com>
Cc: Lennert Buytenhek <kernel@wantstofly.org>
Cc: Ben Dooks <ben-linux@fluff.org>
Tested-by: Jamie Iles <jamie@jamieiles.com>
Tested-by: Tony Lindgren <tony@atomide.com>
Tested-by: Kyungmin Park <kyungmin.park@samsung.com>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Acked-by: Nicolas Pitre <nico@linaro.org>
Acked-by: Krzysztof Halasa <khc@pm.waw.pl>
Acked-by: Kukjin Kim <kgene.kim@samsung.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
The purpose of the minsec argument is to prevent 64-bit math overflow
when the number of cycles is multiplied up. However, the multipler
is 32-bit, and in the sched_clock() case, the cycle counter is up to
32-bit as well. So the math can never overflow.
With a value of 60, and clock rates greater than 71MHz, the calculated
multiplier is unnecessarily reduced in value, which reduces accuracy by
maybe 70ppt. It's almost not worth bothering with as the oscillator
driving the counter won't be any more than 1ppm - unless you're using
a rubidium lamp or caesium fountain frequency standard.
So, set the minsec argument to zero.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
sched_clock is supposed to be initialized early - in the recently added
init_early platform hook. However, in doing so we end up calling
mod_timer() before the timer lists are initialized, resulting in an
oops.
Split the initialization in two - the part which the platform calls
early which starts things off. The addition of the timer can be
delayed until after we have more of the kernel initialized - when the
normal time sources are initialized.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Provide common sched_clock() infrastructure for platforms to use to
create a 64-bit ns based sched_clock() implementation from a counter
running at a non-variable clock rate.
This implementation is based upon maintaining an epoch for the counter
and an epoch for the nanosecond time. When we desire a sched_clock()
time, we calculate the number of counter ticks since the last epoch
update, convert this to nanoseconds and add to the epoch nanoseconds.
We regularly refresh these epochs within the counter wrap interval.
We perform a similar calculation as above, and store the new epochs.
We read and write the epochs in such a way that sched_clock() can easily
(and locklessly) detect when an update is in progress, and repeat the
loading of these constants when they're known not to be stable. The
one caveat is that sched_clock() is not called in the middle of an
update. We achieve that by disabling IRQs.
Finally, if the clock rate is known at compile time, the counter to ns
conversion factors can be specified, allowing sched_clock() to be tightly
optimized. We ensure that these factors are correct by providing an
initialization function which performs a run-time check.
Acked-by: Peter Zijlstra <peterz@infradead.org>
Tested-by: Santosh Shilimkar <santosh.shilimkar@ti.com>
Tested-by: Will Deacon <will.deacon@arm.com>
Tested-by: Mikael Pettersson <mikpe@it.uu.se>
Tested-by: Eric Miao <eric.y.miao@gmail.com>
Tested-by: Olof Johansson <olof@lixom.net>
Tested-by: Jamie Iles <jamie@jamieiles.com>
Reviewed-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
