<?xml version="1.0" encoding="UTF-8"?>
<chapter id="time">
<?dbhtml filename="time.html"?>
<title>Time management</title>
<para>Time is one of the dimensions in which kernel, as well as the whole
system, operates. It is of special importance to many kernel subsytems.
Knowledge of time makes it possible for the scheduler to preemptively plan
threads for execution. Different parts of the kernel can request execution
of their callback function with some specified delay. A good example of such
kernel code is the synchronization subsystem which uses this functionality
to implement timeouting versions of synchronization primitives.</para>
<section>
<title>System clock</title>
<para>Every hardware architecture supported by HelenOS must support some
kind of a device that can be programmed to yield periodic time signals
(i.e. clock interrupts). Some architectures have external clock that is
merely programmed by the kernel to interrupt the processor multiple times
in a second. This is the case of ia32 and amd64 architectures<footnote>
<para>When running in uniprocessor mode.</para>
</footnote>, which use i8254 or a compatible chip to achieve the
goal.</para>
<para>Other architectures' processors typically contain two registers. The
first register is usually called a compare or a match register and can be
set to an arbitrary value by the operating system. The contents of the
compare register then stays unaltered until it is written by the kernel
again. The second register, often called a counter register, can be also
written by the kernel, but the processor automatically increments it after
every executed instruction or in some fixed relation to processor speed.
The point is that a clock interrupt is generated whenever the values of
the counter and the compare registers match. Sometimes, the scheme of two
registers is modified so that only one register is needed. Such a
register, called a decrementer, then counts towards zero and an interrupt
is generated when zero is reached.</para>
<para>In any case, the initial value of the decrementer or the initial
difference between the counter and the compare registers, respectively,
must be set accordingly to a known relation between the real time and the
speed of the decrementer or the counter register, respectively.</para>
<para>The rest of this section will, for the sake of clarity, focus on the
two-register scheme. The decrementer scheme is very similar.</para>
<para>The kernel must reinitialize the counter registers after each clock
interrupt in order to schedule next interrupt. However this step is tricky
and must be done with caution. Imagine that the clock interrupt is masked
either because the kernel is servicing another interrupt or because the
processor locally disabled interrupts for a while. If the clock interrupt
occurs during this period, it will be pending until interrupts are enabled
again. In theory, that could happen arbitrary counter register ticks
later. Which is worse, the ideal time period between two non-delayed clock
interrupts can also elapse arbitrary number of times before the delayed
interrupt gets serviced. The architecture-specific part of the clock
interrupt driver must avoid time drifts caused by this by taking proactive
counter-measures.</para>
<para>Let us assume that the kernel wants each clock interrupt be
generated every <constant>TICKCONST</constant> ticks. This value
represents the ideal number of ticks between two non-delayed clock
interrupts and has some known relation to real time. On each clock
interrupt, the kernel computes and writes down the expected value of the
counter register as it hopes to read it on the next clock interrupt. When
that interrupt comes, the kernel reads the counter register again and
compares it with the written down value. If the difference is smaller than
or equal to <constant>TICKCONST</constant>, then the time drift is none or
small and the next interrupt is scheduled earlier with a penalty of so
many ticks as is the value of the difference. However, if the difference
is bigger, then at least one clock signal was missed. In that case, the
missed clock signal is remembered in the special counter. If there are
more missed signals, each of them is recorded there. The next interrupt is
scheduled with respect to the difference similarily to the former case.
This time, the penalty is taken modulo <constant>TICKCONST</constant>. The
effect of missed clock signals is remedied in the generic clock interrupt
handler.</para>
</section>
<section>
<title>Timeouts</title>
<para>Kernel subsystems can register a callback function to be executed
with a specified delay. Such a registration is represented by a kernel
structure called <classname>timeout</classname>. Timeouts are registered
via <code>timeout_register</code> function. This function takes a pointer
to a timeout structure, a callback function, a parameter of the callback
function and a delay in microseconds as parameters. After the structure is
initialized with all these values, it is sorted into the processor's list
of active timeouts, according to the number of clock interrupts remaining
to their expiration and relatively to already listed timeouts.</para>
<para>Timeouts can be unregistered via <code>timeout_unregister</code>.
This function can, as opposed to <code>timeout_register</code>, fail when
it is too late to remove the timeout from the list of active
timeouts.</para>
<para>Timeouts are nearing their expiration in the list of active timeouts
which exists on every processor in the system. The expiration counters are
decremented on each clock interrupt by the generic clock interrupt
handler. Due to the relative ordering of timeouts in the list, it is
sufficient to decrement expiration counter only of the first timeout in
the list. Timeouts with expiration counter equal to zero are removed from
the list and their callback function is called with respective
parameter.</para>
</section>
<section>
<title>Generic clock interrupt handler</title>
<para>On each clock interrupt, the architecture specific part of the clock
interrupt handler makes a call to the generic clock interrupt handler
implemented by the <code>clock</code> function. The generic handler takes
care of several mission critical goals:</para>
<itemizedlist>
<listitem>
<para>expiration of timeouts,</para>
</listitem>
<listitem>
<para>updating time of the day counters for userspace and</para>
</listitem>
<listitem>
<para>preemption of threads.</para>
</listitem>
</itemizedlist>
<para>The <code>clock</code> function checks for expired timeouts and
decrements unexpired timeout expiration counters exactly one more times
than is the number of missed clock signals (i.e. at least once and
possibly more times, depending on the missed clock signals counter). The
time of the day counters are also updated one more times than is the
number of missed clock signals. And finally, the remaining timeslice of
the running thread is decremented with respect to this counter as well. By
considering its value, the kernel performs actions that would otherwise be
lost due to an occasional excessive time drift described in previous
paragraphs.</para>
</section>
<section>
<title>Time source for userspace</title>
<para>In HelenOS, userspace tasks don't communicate with the kernel in
order to read system time. Instead, a mechanism that shares kernel time of
the day counters with userspace address spaces is deployed. On the kernel
side, during system initialization, HelenOS allocates a frame of physical
memory and stores the time of the day counters there. The counters have
the following structure:</para>
<itemizedlist>
<listitem>
<para>first 32-bit counter for seconds,</para>
</listitem>
<listitem>
<para>32-bit counter for microseconds and</para>
</listitem>
<listitem>
<para>second 32-bit counter for seconds.</para>
</listitem>
</itemizedlist>
<para>One of the userspace tasks with capabilities of memory manager (e.g.
ns) asks the kernel to map this frame into its address space. Other,
non-privileged, tasks then use IPC to communicate read-only sharing of
this memory. Reading time in a userspace task is therefore just a matter
of reading memory.</para>
<para>There are two interesting points about this. First, the counters are
32-bit even on 64-bit machines. The goal is to provide subsecond precision
with the possibility to span roughly 136 years. Note that a single 64-bit
microsecond counter could not be usually read atomically on 32-bit
platforms. Now the second point is that 32-bit platforms cannot atomically
read two 32-bit counters either. However, a generic protocol is used to
guarantee that sequentially read times will create a non-decreasing
sequence.</para>
<para>The problematic part is updating both seconds and microseconds once
in a second. Seconds must be incremented and microseconds must be reset.
However, without any synchronization, the two kernel stores and the two
userspace reads can arbitrarily interleave. Furthemore, the reader has no
chance to detect that the counters were updated only from half. Therefore
three counters are used in HelenOS.</para>
<para>If seconds need to be updated, the kernel increments the first
second counter, issues a write memory barrier operation, updates the
microsecond counter, issues another write memory barrier operation and
increments the second second counter. When only microseconds need to be
updated, no special action is taken by the kernel. On the other hand, the
userspace task must always read all three counters and in reversed order.
A read memory barrier operation must be issued between each two reads. A
non-atomic read is detected when the two second counters differ. The
userspace library solves this situation by returning zero instead of the
value read from the microsecond counter.</para>
</section>
</chapter>