Subversion Repositories HelenOS-doc

Rev

Rev 105 | Rev 133 | Go to most recent revision | Blame | Compare with Previous | Last modification | View Log | Download | RSS feed

  1. <?xml version="1.0" encoding="UTF-8"?>
  2. <chapter id="scheduling">
  3.   <?dbhtml filename="scheduling.html"?>
  4.  
  5.   <title>Scheduling</title>
  6.  
  7.   <para>One of the key aims of the operating system is to create and support
  8.   the impression that several activities are executing contemporarily. This is
  9.   true for both uniprocessor as well as multiprocessor systems. In the case of
  10.   multiprocessor systems, the activities are trully happening in parallel. The
  11.   scheduler helps to materialize this impression by planning threads on as
  12.   many processors as possible and, where this means reaches its limits, by
  13.   quickly switching among threads executing on a single processor.</para>
  14.  
  15.   <section>
  16.     <title>Contexts</title>
  17.  
  18.     <para>The term context refers to the set of processor resources that
  19.     define the current state of the computation or the environment and the
  20.     kernel understands it in several more or less narrow sences:</para>
  21.  
  22.     <itemizedlist>
  23.       <listitem>
  24.         <para>synchronous register context,</para>
  25.       </listitem>
  26.  
  27.       <listitem>
  28.         <para>asynchronous register context,</para>
  29.       </listitem>
  30.  
  31.       <listitem>
  32.         <para>FPU context and</para>
  33.       </listitem>
  34.  
  35.       <listitem>
  36.         <para>memory management context.</para>
  37.       </listitem>
  38.     </itemizedlist>
  39.  
  40.     <para>The most narrow sense refers to the the synchronous register
  41.     context. It includes all the preserved registers as defined by the
  42.     architecture. To highlight some, the program counter and stack pointer
  43.     take part in the synchronous register context. These registers must be
  44.     preserved across a procedure call and during synchronous context
  45.     switches.</para>
  46.  
  47.     <para>The next type of the context understood by the kernel is the
  48.     asynchronous register context. On an interrupt, the interrupted execution
  49.     flow's state must be guaranteed to be eventually completely restored.
  50.    Therefore the interrupt context includes, among other things, the scratch
  51.    registers as defined by the architecture. As a special optimization and if
  52.    certain conditions are met, it need not include the architecture's
  53.     preserved registers. The condition mentioned in the previous sentence is
  54.     that the low-level assembly language interrupt routines don't modify the
  55.    preserved registers. The handlers usually call a higher-level C routine.
  56.    The preserved registers are then saved on the stack by the compiler
  57.    generated code of the higher-level function. In HelenOS, several
  58.    architectures can be compiled with this optimization.</para>
  59.  
  60.    <para>Although the kernel does not do any floating point
  61.    arithmetics<footnote>
  62.        <para>Some architectures (e.g. ia64) inevitably use a fixed set of
  63.        floating point registers to carry out their normal operations.</para>
  64.      </footnote>, it must protect FPU context of userspace threads against
  65.    destruction by other threads. Moreover, only a fraction of userspace
  66.    programs use the floating point unit. HelenOS contains a generic framework
  67.    for switching FPU context only when the switch is forced (i.e. a thread
  68.    uses a floating point instruction and its FPU context is not loaded in the
  69.    processor).</para>
  70.  
  71.    <para>The last member of the context family is the memory management
  72.    context. It includes memory management registers that identify address
  73.    spaces on hardware level (i.e. ASIDs and page tables pointers).</para>
  74.  
  75.    <section>
  76.      <title>Synchronous Context Switches</title>
  77.  
  78.      <para>The scheduler, but also other pieces of the kernel, make heavy use
  79.      of synchronous context switches, because it is a natural vehicle not
  80.      only for changes in control flow, but also for switching between two
  81.      kernel stacks. Two functions figure in a synchronous context switch
  82.      implementation: <code>context_save</code> and
  83.      <code>context_restore</code>. Note that these two functions break the
  84.      natural perception of the linear C code execution flow starting at
  85.      function's entry point and ending on one of the function's exit
  86.      points.</para>
  87.  
  88.      <para>When the <code>context_save</code> function is called, the
  89.      synchronous context is saved in a memory structure passed to it. After
  90.      executing <code>context_save</code>, the caller is returned 1 as a
  91.      return value. The execution of instructions continues as normally until
  92.      <code>context_restore</code> is called. For the caller, it seems like
  93.      the call never returns<footnote>
  94.          <para>Which might be a source of problems with variable liveliness
  95.          after <code>context_restore</code>.</para>
  96.        </footnote>. Nevertheless, a synchronous register context, which is
  97.      saved in a memory structure passed to <code>context_restore,</code> is
  98.      restored, thus transfering the control flow to the place of occurrence
  99.      of the corresponding call to <code>context_save</code>. From the
  100.      perspective of the caller of the corresponding
  101.      <code>context_save</code>, it looks as though a return from
  102.      <code>context_save</code>. However, this time a return value of 0 is
  103.      returned.</para>
  104.    </section>
  105.  </section>
  106.  
  107.  <section>
  108.    <title>Threads</title>
  109.  
  110.    <para>A thread is the basic executable entity with some code and stack.
  111.    While the code, implemented by a C language function, can be shared by
  112.    several threads, the stack is always private to each instance of the
  113.    thread. Each thread belongs to exactly one task through which it shares
  114.    address space with its sibling threads. Threads that execute purely in the
  115.    kernel don't have any userspace memory allocated. However, when a thread
  116.     has ambitions to run in userspace, it must be allocated a userspace stack.
  117.     The distinction between the purely kernel threads and threads running also
  118.     in userspace is made by refering to the former group as to kernel threads
  119.     and to the latter group as to userspace threads. Both kernel and userspace
  120.     threads are visible to the scheduler and can become a subject of kernel
  121.     preemption and thread migration during times when preemption is
  122.     possible.</para>
  123.  
  124.     <formalpara>
  125.       <title>Thread States</title>
  126.  
  127.       <para>In each moment, a thread exists in one of six possible thread
  128.       states. When the thread is created and first readied into the
  129.       scheduler's run queues or when a thread is migrated to a new processor,
  130.      it is put into the <constant>Entering</constant> state. After some time,
  131.      the scheduler picks up the thread and starts executing it. A thread
  132.      being currently executed on a processor is in the
  133.      <constant>Running</constant> state. From there, the thread has three
  134.      possibilities. It either runs until it is preemtped, in which case the
  135.      state changes to <constant>Ready</constant>, goes to the
  136.      <constant>Sleeping</constant> state by going to sleep or enters the
  137.      <constant>Exiting</constant> state when it reaches termination. When the
  138.      thread exits, its kernel structure usually stays in memory, until the
  139.      thread is detached by another thread using <code>thread_detach</code>
  140.      function. Terminated but undetached threads are in the
  141.      <constant>Undead</constant> state. When the thread is detached or
  142.      detaches itself during its life, it is destroyed in the
  143.      <constant>Exiting</constant> state and the <constant>Undead</constant>
  144.      state is not reached.<figure float="1">
  145.          <title>Transitions among thread states.</title>
  146.  
  147.          <mediaobject id="thread_states" xreflabel="">
  148.            <imageobject role="pdf">
  149.              <imagedata fileref="images/thread_states.pdf" format="PDF" />
  150.            </imageobject>
  151.  
  152.            <imageobject role="html">
  153.              <imagedata fileref="images/thread_states.png" format="PNG" />
  154.            </imageobject>
  155.  
  156.            <imageobject role="fop">
  157.              <imagedata fileref="images/thread_states.svg" format="SVG" />
  158.            </imageobject>
  159.          </mediaobject>
  160.        </figure></para>
  161.    </formalpara>
  162.  
  163.    <formalpara>
  164.      <title>Pseudo Threads</title>
  165.  
  166.      <para>HelenOS userspace layer knows even smaller units of execution.
  167.      Each userspace thread can make use of an arbitrary number of pseudo
  168.      threads. These pseudo threads have their own synchronous register
  169.      context, userspace code and stack. They live their own life within the
  170.      userspace thread and the scheduler does not have any idea about them
  171.      because they are completely implemented by the userspace library. This
  172.      implies several things:<itemizedlist>
  173.          <listitem>
  174.            <para>pseudothreads schedule themselves cooperatively within the
  175.            time slice given to their userspace thread,</para>
  176.          </listitem>
  177.  
  178.          <listitem>
  179.            <para>pseudothreads share FPU context of their containing thread
  180.            and</para>
  181.          </listitem>
  182.  
  183.          <listitem>
  184.            <para>all pseudothreads of one userspace thread block when one of
  185.            them goes to sleep.</para>
  186.          </listitem>
  187.        </itemizedlist></para>
  188.    </formalpara>
  189.  </section>
  190.  
  191.  <section>
  192.    <title>Scheduler</title>
  193.  
  194.    <section>
  195.      <title>Run Queues</title>
  196.  
  197.      <para>There is an array of several run queues on each processor. The
  198.      current version of HelenOS uses 16 run queues implemented by 16 doubly
  199.      linked lists. Each of the run queues is associated with thread priority.
  200.      The lower the run queue index in the array is, the higher is the
  201.      priority of threads linked in that run queue and the shorter is the time
  202.      in which those threads will execute. When kernel code wants to access
  203.      the run queue, it must first acquire its lock.</para>
  204.    </section>
  205.  
  206.    <section>
  207.      <title>Scheduler Operation</title>
  208.  
  209.      <para>The scheduler is invoked either explicitly when a thread calls the
  210.      <code>scheduler</code> function (e.g. goes to sleep or merely wants to
  211.      relinquish the processor for a while) or implicitly on a periodic basis
  212.      when the generic clock interrupt preempts the current thread. After its
  213.      invocation, the scheduler saves the synchronous register context of the
  214.      current thread and switches to its private stack. Afterwards, a new
  215.      thread is selected according to the scheduling policy. If there is no
  216.      suitable thread, the processor is idle and no thread executes on it.
  217.      Note that the act of switching to the private scheduler stack is
  218.      essential. If the processor kept running using the stack of the
  219.      preempted thread it could damage it because the old thread can be
  220.      migrated to another processor and scheduled there. In the worst case
  221.      scenario, two execution flows would be using the same stack.</para>
  222.  
  223.      <para>The scheduling policy is implemented in function
  224.      <code>find_best_thread</code>. This function walks the processor run
  225.      queues from lower towards higher indices and looks for a thread. If the
  226.      visited run queue is empty, it simply searches the next run queue. If it
  227.      is known in advance that there are no ready threads waiting for
  228.      execution, <code>find_best_thread</code> interruptibly halts the
  229.      processor or busy waits until some threads arrive. This process repeats
  230.      until <code>find_best_thread</code> succeeds.</para>
  231.  
  232.      <para>After the best thread is chosen, the scheduler switches to the
  233.      thread's task and memory management context. Finally, the saved
  234.       synchronous register context is restored and the thread runs. Each
  235.       scheduled thread is given a time slice depending on its priority (i.e.
  236.       run queue). The higher priority, the shorter timeslice. To summarize,
  237.       this policy schedules threads with high priorities more frequently but
  238.       gives them smaller time slices. On the other hand, lower priority
  239.       threads are scheduled less frequently, but run for longer periods of
  240.       time.</para>
  241.  
  242.       <para>When a thread uses its entire time slice, it is preempted and put
  243.       back into the run queue that immediately follows the previous run queue
  244.       from which the thread ran. Threads that are woken up from a sleep are
  245.       put into the biggest priority run queue. Low priority threads are
  246.       therefore those that don't go to sleep so often and just occupy the
  247.      processor.</para>
  248.  
  249.      <para>In order to avoid complete starvation of the low priority threads,
  250.      from time to time, the scheduler will provide them with a bonus of one
  251.      point priority increase. In other words, the scheduler will now and then
  252.      move the entire run queues one level up.</para>
  253.    </section>
  254.  
  255.    <section>
  256.      <title>Processor Load Balancing</title>
  257.  
  258.      <para>Normally, for the sake of cache locality, threads are scheduled on
  259.      one of the processors and don't leave it. Nevertheless, a situation in
  260.       which one processor is heavily overloaded while others sit idle can
  261.       occur. HelenOS deploys special kernel threads to help mitigate this
  262.       problem. Each processor is associated with one load balancing thread
  263.       called <code>kcpulb</code> that wakes up regularly to see whether its
  264.       processor is underbalanced or not. If it is, the thread attempts to
  265.       migrate threads from other overloaded processors to its own processor's
  266.      run queues. When the job is done or there is no need for load balancing,
  267.      the thread goes to sleep.</para>
  268.  
  269.      <para>The balancing threads operate very gently and try to migrate low
  270.      priority threads first; one <code>kcpulb</code> never takes from one
  271.      processor twice in a row. The load balancing threads as well as threads
  272.      that were just stolen cannot be migrated. The <code>kcpulb</code>
  273.      threads are wired to their processors and cannot be migrated whatsoever.
  274.      The ordinary threads are protected only until they are
  275.      rescheduled.</para>
  276.    </section>
  277.  </section>
  278. </chapter>