The HPET default is to deliver events on the same INTIn as the legacy
PIT IRQ, and in fact our code requires that because it uses the
"legacy routing" option. So this isn't really a configurable and has
to be set correctly. Do it right in the kconfig default instead of
forcing boards to set it.
(No, I have no idea where "20" came from either.)
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
This patch fixes a bug in System timer driver where
the sys_clock_disable() function was enabling the
timer instead of disabling it.
Change-Id: I4a667d30d43d1f84094d074241ee18d7bb2b2565
Signed-off-by: David Vincze <david.vincze@arm.com>
Two subtractions failed to account for the possibility that a calculated
time exceeded the counter resolution, allowing a comparison to
improperly indicate that a minimum delay was satisfied.
Use the subtraction helper to avoid the problem.
(The subtraction in z_clock_set_timeout was the cause of issue #11694;
the one in rtc1_nrf5_isr was replaced based on inspection rather than
testing.)
Closes#11694
Signed-off-by: Peter A. Bigot <pab@pabigot.com>
If we just had the kernel's implementation, we could
just move this to lib/, but possible arch-specific
implementations dictate that we just make this a
syscall.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
We still have one platform using (for now) the pre-asm2 integration
where the timer interrupt was handled via custom assembly. It calls a
function named "_timer_int_handler" always, not the one we register
with IRQ_CONNECT.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Rewritten Xtensa CCOUNT driver along the lines of all the other new
drivers. The new API permits much smaller code.
Notably: The Xtensa counter is a 32 bit up-counter with a comparator
register. It's in some sense the archetype of this kind of timer as
it's the simplest of the bunch (everything else has quirks: NRF is
very slow and 24 bit, HPET has a runtime frequency detection, RISC-V
is 64 bit...). I should have written this one first.
Note also that this includes a blacklist of the xtensa architecture on
the tests/driver/ipm test. I'm getting spurious failures there where
a k_sem_take() call with a non-zero timeout is being made out of the
console output code in interrupt context. This seems to have nothing
to do with the timer; I suspect it's because the old timer drivers
would (incorrectly!) call z_clock_announce() in non-interrupt context
in some contexts (e.g. "expiring really soon"). Apparently this test
(or something in the IPM or Xtensa console code) was somehow relying
on that on Xtensa. But IPM is a Quark thing and there's no particular
reason to run this test there.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Rewritten driver along the lines of all the other new drivers,
implementing the new timer API. Structurally, the machine timer is an
up-counter with comparator, so it works broadly the same way HPET and
NRF do. The quirk here is that it's a 64 bit counter, which needs a
little more care.
Unlike the other timer reworks, this driver has grown by a few lines
as it used to be very simple. But in exchange, we get full tickless
support on the platform.
Fixes#10609 in the process (the 64 bit timer registers are unlatched
for sub-word transfers, so you have to use careful ordering).
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Reworked using the older hardware interface code, but with an
implementation of the new API only. Much smaller & simpler.
As yet, tested (manually) only on a nrf52_pca10056 board.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Rewritten along the lines of ARM SysTick. Implements only the new,
simplified API. MUCH smaller. Works with tickless pervasively. No
loss of functionality.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Many drivers won't need to implement z_clock_idle_exit() or
sys_clock_disable(). Make those weak stubs too.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Add a TICKLESS_CAPABLE kconfig variable which is used by the kernel to
select tickless mode's default automatically on drivers that support
it (rather than having to set the default per-board). Select it from
the ARM SysTick and Intel HPET drivers.
Also remove the old qemu_cortex_m3 default settings which this
replaces.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Qemu doesn't like tickless. By default[1] it tries to be realtime as
vied by the host CPU -- presenting read values from hardware cycle
counters and interrupt timings at the appropriate real world clock
times according to whatever the simulated counter frequency is. But
when the host system is loaded, there is always the problem that the
qemu process might not see physical CPU time for large chunks of time
(i.e. a host OS scheduling quantum -- generally about the same size as
guest ticks!) leading to lost cycles.
When those timer interrupts are delivered by the emulated hardware at
fixed frequencies without software intervention, that's not so bad:
the work the guest has to do after the interrupt generally happens
synchronously (because the qemu process has just started running) and
nothing notices the dropout.
But with tickless, the interrupts need to be explicitly programmed by
guest software! That means the driver needs to be sure it's going to
get some real CPU time within some small fraction of a Zephyr tick of
the right time, otherwise the computations get wonky.
The end result is that qemu tends to work with tickless well on an
unloaded/idle run, but not in situations (like sanitycheck) where it
needs to content with other processes for host CPU.
So, add a flag that drivers can use to "fake" tickless behavior when
run under qemu (only), and enable it (only!) for the small handful of
tests that are having trouble.
[1] There is an -icount feature to implement proper cycle counting at
the expense of real-world-time correspondence. Maybe someday we might
get it to work for us.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Newer, and much smaller driver written to the new timer API. Supports
all the features the old one did (including shutting off the clock
when clock_always_on is disabled), should be faster in practice, and
should be significantly more accurate due to the "lost cycle" trick
applied in z_clock_set_timeout().
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Since CCOMPARE* registers have undefined values after reset,
set compare value first before enabling timer interrupt.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
I was pretty careful, but these snuck in. Most of them are due to
overbroad string replacements in comments. The pull request is very
large, and I'm too lazy to find exactly where to back-merge all of
these.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Now that the API has been fixed up, replace the existing timeout queue
with a much smaller version. The basic algorithm is unchanged:
timeouts are stored in a sorted dlist with each node nolding a delta
time from the previous node in the list; the announce call just walks
this list pulling off the heads as needed. Advantages:
* Properly spinlocked and SMP-aware. The earlier timer implementation
relied on only CPU 0 doing timeout work, and on an irq_lock() being
taken before entry (something that was violated in a few spots).
Now any CPU can wake up for an event (or all of them) and everything
works correctly.
* The *_thread_timeout() API is now expressible as a clean wrapping
(just one liners) around the lower-level interface based on function
pointer callbacks. As a result the timeout objects no longer need
to store backpointers to the thread and wait_q and have shrunk by
33%.
* MUCH smaller, to the tune of hundreds of lines of code removed.
* Future proof, in that all operations on the queue are now fronted by
just two entry points (_add_timeout() and z_clock_announce()) which
can easily be augmented with fancier data structures.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
The current z_clock_uptime() call (recently renamed from
_get_elapsed_program_time) requires the driver to track a full 64 bit
uptime value in ticks, which is entirely separate from the one the
kernel is already keeping.
Don't do that. Just ask the drivers to track uptime since the last
call to z_clock_announce(), since that is going to map better to
built-in hardware capability.
Obviously existing drivers already have this feature, so they're
actually getting slightly larger in order to implement the new API in
terms of the old one. But future drivers will thank us.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Useful for tick-only drivers like Pulpino that don't support this.
Ideally we'd have a header-level interface definition for individual
timer drivers to eliminate the noop function call, but this is clean
for now (even the Pulpino hardware looks like it should support
timeouts just fine, so effort would be better spent there than on a
clean "ticked" interface).
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
The tickless driver had a bunch of "hairy" APIs which forced the timer
drivers to do needless low-level accounting for the benefit of the
kernel, all of which then proceeded to implement them via cut and
paste. Specifically the "program_time" calls forced the driver to
expose to the kernel exactly when the next interrupt was due and how
much time had elapsed, in a parallel API to the existing "what time is
it" and "announce a tick" interrupts that carry the same information.
Remove these from the kernel, replacing them with synthesized logic
written in terms of the simpler APIs.
In some cases there will be a performance impact due to the use of the
64 bit uptime call, but that will go away soon.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Rename timer driver API functions to be consistent. ADD DOCS TO THE
HEADER so implementations understand what the requirements are.
Remove some unused functions that don't need declarations here.
Also removes the per-platform #if's around the power control callback
in favor of a weak-linked noop function in the driver initialization
(adds a few bytes of code to default platforms -- we'll live, I
think).
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
The existing API had two almost identical functions: _set_time() and
_timer_idle_enter(). Both simply instruct the timer driver to set the
next timer interrupt expiration appropriately so that the call to
z_clock_announce() will be made at the requested number of ticks. On
most/all hardware, these should be implementable identically.
Unfortunately because they are specified differently, existing drivers
have implemented them in parallel.
Specify a new, unified, z_clock_set_timeout(). Document it clearly
for implementors. And provide a shim layer for legacy drivers that
will continue to use the old functions.
Note that this patch fixes an existing bug found by inspection: the
old call to _set_time() out of z_clock_announce() failed to test for
the "wait forever" case in the situation where clock_always_on is
true, meaning that a system that reached this point and then never set
another timeout would freeze its uptime clock incorrectly.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
There were three separate "announce ticks" entry points exposed for
use by drivers. Unify them to just a single z_clock_announce()
function, making the "final" tick announcement the business of the
driver only, not the kernel.
Note the oddness with "_sys_idle_elapsed_ticks": this was a global
variable exposed by the kernel. But it was never actually used by the
kernel. It was updated and inspected only within the timer drivers,
and only so that it could be passed back to the kernel as the default
(actually hidden) argument to the announce function. Break this false
dependency by putting this variable into each timer driver
individually.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
The system tick count is a 64 bit quantity that gets updated from
interrupt context, meaning that it's dangerously non-atomic and has to
be locked. The core kernel clock code did this right.
But the value was also exposed to the rest of the universe as a global
variable, and virtually nothing else was doing this correctly. Even
in the timer ISRs themselves, the interrupts may be themselves
preempted (most of our architectures support nested interrupts) by
code that wants to set timeouts and inspect system uptime.
Define a z_tick_{get,set}() API, eliminate the old variable, and make
sure everyone uses the right mechanism.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
This was another "global variable" API. Give it function syntax too.
Also add a warning, because on nRF devices (at least) the cycle clock
runs in kHz and is too slow to give a precise answer here.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
This just got turned into a function from a "variable" API, but
post-the-most-recent-patch it turns out to be degenerate anyway.
Everyone everywhere should always have been using the kconfig variable
directly, and it was only a weirdness in the tickless API that made it
confusing. Fix.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
The existing API defined sys_clock_{hw_cycles,ticks}_per_sec as simple
"variables" to be shared, except that they were only real storage in
certain modes (the HPET driver, basically) and everywhere else they
were a build constant.
Properly, these should be an API defined by the timer driver (who
controls those rates) and consumed by the clock subsystem. So give
them function syntax as a stepping stone to get there.
Note that this also removes the deprecated variable
_sys_clock_us_per_tick rather than give it the same treatment.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Simplify the Kconfig dependency for the nrf timer driver.
CLOCK_CONTROL_NRF5 depends on the SOC_FAMILY_NRF already.
Signed-off-by: Alberto Escolar Piedras <alpi@oticon.com>
Add support for CONFIG_SYSTEM_CLOCK_DISABLE so applications
may be compiled with CONFIG_REBOOT.
Signed-off-by: Alberto Escolar Piedras <alpi@oticon.com>
The code assumes that when the systick counter hits zero,
the timer interrupt will be taken before the loop can
read the LOAD/VAL registers, but this is not architecturally
guaranteed, and so the code can see a post-reload SysTick->VAL
and a pre-reload clock_accumulated_count, which causes it to
return an incorrectly small cycle count. By adding a ISB we
overcome this issue.
Signed-off-by: Adithya Baglody <adithya.nagaraj.baglody@intel.com>
Move to more generic tracing hooks that can be implemented in different
ways and do not interfere with the kernel.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Define generic interface and hooks for tracing to replace
kernel_event_logger and existing tracing facilities with something more
common.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
This patch provides support needed to get timing related
information from riscv32 based SOC.
Signed-off-by: Adithya Baglody <adithya.nagaraj.baglody@intel.com>
This patch provides support needed to get timing related
information from nios2 based SOC.
Signed-off-by: Adithya Baglody <adithya.nagaraj.baglody@intel.com>
This patch provides support needed to get timing related
information from xtensa based SOC.
Signed-off-by: Adithya Baglody <adithya.nagaraj.baglody@intel.com>
The benchmark application timing_info needs certain hooks to be
present in the kernel to get the accurate measurements. This
patch adds these hook at all the required locations.
Signed-off-by: Adithya Baglody <adithya.nagaraj.baglody@intel.com>
irq_lock returns an unsigned int, though, several places was using
signed int. This commit fix this behaviour.
In order to avoid this error happens again, a coccinelle script was
added and can be used to check violations.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Consistently use
config FOO
bool/int/hex/string "Prompt text"
instead of
config FOO
bool/int/hex/string
prompt "Prompt text"
(...and a bunch of other variations that e.g. swapped the order of the
type and the 'prompt', or put other properties between them).
The shorthand is fully equivalent to using 'prompt'. It saves lines and
avoids tricking people into thinking there is some semantic difference.
Most of the grunt work was done by a modified version of
https://unix.stackexchange.com/questions/26284/how-can-i-use-sed-to-replace-a-multi-line-string/26290#26290, but some
of the rarer variations had to be converted manually.
Signed-off-by: Ulf Magnusson <Ulf.Magnusson@nordicsemi.no>
This commit removes redundant reads of RTC_COUNTER register
propagating optimizations made in the k_cycle_get_32() function.
Signed-off-by: Piotr Zięcik <piotr.ziecik@nordicsemi.no>
The previous implementation of _timer_cycle_get_32() (which is directly
mapped to k_cycle_get_32()) taken from 87 to 132 cycles. As result it
was too heavy for using it as source of time for logger.
This commit makes this function faster by removing redundant access
to the RTC register (each access consumed 28 CPU cycles) as well as
loop, which made this call non-deterministic.
After these changes the k_cycle_get_32() needs only 50-52 cycles
in to calculate 32-bit timestamp.
Signed-off-by: Piotr Zięcik <piotr.ziecik@nordicsemi.no>
A few function prototypes were missing in the native_posix
board and its drivers.
Let's add them.
Signed-off-by: Alberto Escolar Piedras <alpi@oticon.com>
Intel SDM Vol3 10.5.4.1 states that "A write to LVT Timer Register that
changes the timer mode disarms the local APIC timer".
This implies that LVT Timer register needs to be programmed before
Initial Count register, otherwise the LOAPIC timer could not be armed.
Signed-off-by: Zide Chen <zide.chen@intel.com>
