Subversion Repositories HelenOS-doc

<?xml version="1.0" encoding="UTF-8"?>

<chapter id="mm">

  <?dbhtml filename="mm.html"?>

  <title>Memory management</title>

  <section>

    <title>Virtual memory management</title>

    <section>

      <title>Introduction</title>

      <para>Virtual memory is a special memory management technique, used by

      kernel to achieve a bunch of mission critical goals. <itemizedlist>

          <listitem>

             Isolate each task from other tasks that are running on the system at the same time.

          </listitem>

          <listitem>

             Allow to allocate more memory, than is actual physical memory size of the machine.

          </listitem>

          <listitem>

             Allowing, in general, to load and execute two programs that are linked on the same address without complicated relocations.

          </listitem>

        </itemizedlist></para>

      <para><!--

                TLB shootdown ASID/ASID:PAGE/ALL.

                TLB shootdown requests can come in asynchroniously

                so there is a cache of TLB shootdown requests. Upon cache overflow TLB shootdown ALL is executed

                <para>

                        Address spaces. Address space area (B+ tree). Only for uspace. Set of syscalls (shrink/extend etc).

                        Special address space area type - device - prohibits shrink/extend syscalls to call on it.

                        Address space has link to mapping tables (hierarchical - per Address space, hash - global tables).

                </para>

--></para>

    </section>

    <section>

      <title>Paging</title>

      <para>Virtual memory is usually using paged memory model, where virtual

      memory address space is divided into the <emphasis>pages</emphasis>

      (usually having size 4096 bytes) and physical memory is divided into the

      frames (same sized as a page, of course). Each page may be mapped to

      some frame and then, upon memory access to the virtual address, CPU

      performs <emphasis>address translation</emphasis> during the instruction

      execution. Non-existing mapping generates page fault exception, calling

      kernel exception handler, thus allowing kernel to manipulate rules of

      memory access. Information for pages mapping is stored by kernel in the

      <link linkend="page_tables">page tables</link></para>

      <para>The majority of the architectures use multi-level page tables,

      which means need to access physical memory several times before getting

      physical address. This fact would make serios performance overhead in

      virtual memory management. To avoid this <link linkend="tlb">Traslation

      Lookaside Buffer (TLB)</link> is used.</para>

      <para>At the moment HelenOS does not support swapping.</para>

      <para>- pouzivame vypadky stranky k alokaci ramcu on-demand v ramci

      as_area - na architekturach, ktere to podporuji, podporujeme non-exec

      stranky</para>

    </section>

    <section>

      <title>Address spaces</title>

      <section>

        <title>Address spaces and areas</title>

        <para>- adresovy prostor se sklada z tzv. address space areas

        usporadanych v B+stromu; tyto areas popisuji vyuzivane casti

        adresoveho prostoru patrici do user address space. Kazda cast je dana

        svoji bazovou adresou, velikosti a flagy (rwx/dd).</para>

        <para>- uzivatelske thready maji moznost manipulovat se svym adresovym

        prostorem (vytvaret/resizovat/sdilet) as_areas pomoci syscallu</para>

      </section>

      <section>

        <title>Address Space ID (ASID)</title>

        <para>- nektery hardware umoznuje rozlisit ruzne adresove prostory od

        sebe (cilem je maximalizovat vyuziti TLB); dela to tak, ze s kazdou

        polozkou TLB/strankovacich tabulek sdruzi identifikator adresoveho

        prostoru (ASID, RID, ppc32 ???). Tyto id mivaji ruznou sirku: 8-bitu

        az 24-bitu (kolik ma ppc32?)</para>

        <para>- kernel tomu rozumi a sam pouziva abstrakci ASIDu (na ia64 to

        je napr. cislo odvozene od RIDu, na mips32 to je ASID samotny);

        existence ASIDu je nutnou podminkou pouziti _global_ page hash table

        mechanismu.</para>

        <para>- na vsech arch. plati, ze asidu je mnohem mene, nez teoreticky

        pocet soucasne bezicich tasku ~ adresovych prostoru, takze je

        implementovan mechanismus, ktery umoznuje jednomu adresovemu prostoru

        ASID odebrat a pridelit ho jinemu</para>

        <para>- vztah task ~ adresovy prostor: teoreticky existuje moznost, ze

        je adresovy prostor sdilen vice tasky, avsak tuto moznost nepouzivame

        a neni ani nijak osetrena. Tim padem plati, ze kazdy task ma vlastni

        adresovy prostor</para>

      </section>

    </section>

    <section>

      <title>Virtual address translation</title>

      <section id="page_tables">

        <title>Page tables</title>

        <para>HelenOS kernel has two different approaches to the paging

        implementation: <emphasis>4 level page tables</emphasis> and

        <emphasis>global hash tables</emphasis>, which are accessible via

        generic paging abstraction layer. This division was caused by the

        major architectural differences between different platforms.</para>

        <formalpara>

          <title>4-level page tables</title>

          <para>4-level page tables are the generalization of the hardware

          capabilities of the certain platforms. <itemizedlist>

              <listitem>

                 ia32 uses 2-level page tables, with full hardware support.

              </listitem>

              <listitem>

                 amd64 uses 4-level page tables, also coming with full hardware support.

              </listitem>

              <listitem>

                 mips and ppc32 have 2-level tables, software simulated support.

              </listitem>

            </itemizedlist></para>

        </formalpara>

        <formalpara>

          <title>Global hash tables</title>

          <para>- global page hash table: existuje jen jedna v celem systemu

          (vyuziva ji ia64), pozn. ia64 ma zatim vypnuty VHPT. Pouziva se

          genericke hash table s oddelenymi collision chains</para>

        </formalpara>

        <para>Thanks to the abstract paging interface, there is possibility

        left have more paging implementations, for example B-Tree page

        tables.</para>

      </section>

      <section id="tlb">

        <title>Translation Lookaside buffer</title>

        <para>- TLB cachuji informace ve strankovacich tabulkach; alternativne

        se lze na strankovaci tabulky (ci ruzne hw rozsireni [e.g. VHPT, ppc32

        hw hash table]) divat jako na velke TLB</para>

        <para>- pri modifikaci mapovani nebo odstraneni mapovani ze

        strankovacich tabulek je potreba zajistit konsistenci TLB a techto

        tabulek; nutne delat na vsech CPU; na to mame zjednodusenou verzi TLB

        shootdown mechanismu; je to variace na algoritmus popsany zde: D.

        Black et al., "Translation Lookaside Buffer Consistency: A Software

        Approach," Proc. Third Int'l Conf. Architectural Support for

        Programming Languages and Operating Systems, 1989, pp. 113-122.</para>

        <para>- nutno poznamenat, ze existuji odlehcenejsi verze TLB shootdown

        algoritmu</para>

      </section>

    </section>

  </section>

  <!-- End of VM -->

  <section>

    <!-- Phys mem -->

    <title>Physical memory management</title>

    <section id="zones_and_frames">

      <title>Zones and frames</title>

      <para><!--graphic fileref="images/mm2.png" /--><!--graphic fileref="images/buddy_alloc.svg" format="SVG" /--></para>

      <para>On some architectures not whole physical memory is available for

      conventional usage. This limitations require from kernel to maintain a

      table of available and unavailable ranges of physical memory addresses.

      Main idea of zones is in creating memory zone entity, that is a

      continuous chunk of memory available for allocation. If some chunk is

      not available, we simply do not put it in any zone.</para>

      <para>Zone is also serves for informational purposes, containing

      information about number of free and busy frames. Physical memory

      allocation is also done inside the certain zone. Allocation of zone

      frame must be organized by the <link linkend="frame_allocator">frame

      allocator</link> associated with the zone.</para>

      <para>Some of the architectures (mips32, ppc32) have only one zone, that

      covers whole physical memory, and the others (like ia32) may have

      multiple zones. Information about zones on current machine is stored in

      BIOS hardware tables or can be hardcoded into kernel during compile

      time.</para>

    </section>

    <section id="frame_allocator">

      <title>Frame allocator</title>

      <para><mediaobject id="frame_alloc">

          <imageobject role="html">

            <imagedata fileref="images/frame_alloc.png" format="PNG" />

          </imageobject>

          <imageobject role="fop">

            <imagedata fileref="images.vector/frame_alloc.svg" format="SVG" />

          </imageobject>

        </mediaobject></para>

      <formalpara>

        <title>Overview</title>

        <para>Frame allocator provides physical memory allocation for the

        kernel. Because of zonal organization of physical memory, frame

        allocator is always working in context of some zone, thus making

        impossible to allocate a piece of memory, which lays in different

        zone, which cannot happen, because two adjacent zones can be merged

        into one. Frame allocator is also being responsible to update

        information on the number of free/busy frames in zone. Physical memory

        allocation inside one <link linkend="zones_and_frames">memory

        zone</link> is being handled by an instance of <link

        linkend="buddy_allocator">buddy allocator</link> tailored to allocate

        blocks of physical memory frames.</para>

      </formalpara>

      <formalpara>

        <title>Allocation / deallocation</title>

        <para>Upon allocation request, frame allocator tries to find first

        zone, that can satisfy the incoming request (has required amount of

        free frames to allocate). During deallocation, frame allocator needs

        to find zone, that contain deallocated frame. This approach could

        bring up two potential problems: <itemizedlist>

            <listitem>

               Linear search of zones does not any good to performance, but number of zones is not expected to be high. And if yes, list of zones can be replaced with more time-efficient B-tree.

            </listitem>

            <listitem>

               Quickly find out if zone contains required number of frames to allocate and if this chunk of memory is properly aligned. This issue is perfectly solved bu the buddy allocator.

            </listitem>

          </itemizedlist></para>

      </formalpara>

    </section>

    <section id="buddy_allocator">

      <title>Buddy allocator</title>

      <section>

        <title>Overview</title>

        <para><mediaobject id="buddy_alloc">

            <imageobject role="html">

              <imagedata fileref="images/buddy_alloc.png" format="PNG" />

            </imageobject>

            <imageobject role="fop">

              <imagedata fileref="images.vector/buddy_alloc.svg" format="SVG" />

            </imageobject>

          </mediaobject></para>

        <para>In buddy allocator, memory is broken down into power-of-two

        sized naturally aligned blocks. These blocks are organized in an array

        of lists in which list with index i contains all unallocated blocks of

        the size <mathphrase>2<superscript>i</superscript></mathphrase>. The

        index i is called the order of block. Should there be two adjacent

        equally sized blocks in list <mathphrase>i</mathphrase> (i.e.

        buddies), the buddy allocator would coalesce them and put the

        resulting block in list <mathphrase>i + 1</mathphrase>, provided that

        the resulting block would be naturally aligned. Similarily, when the

        allocator is asked to allocate a block of size

        <mathphrase>2<superscript>i</superscript></mathphrase>, it first tries

        to satisfy the request from list with index i. If the request cannot

        be satisfied (i.e. the list i is empty), the buddy allocator will try

        to allocate and split larger block from list with index i + 1. Both of

        these algorithms are recursive. The recursion ends either when there

        are no blocks to coalesce in the former case or when there are no

        blocks that can be split in the latter case.</para>

        <!--graphic fileref="images/mm1.png" format="EPS" /-->

        <para>This approach greatly reduces external fragmentation of memory

        and helps in allocating bigger continuous blocks of memory aligned to

        their size. On the other hand, the buddy allocator suffers increased

        internal fragmentation of memory and is not suitable for general

        kernel allocations. This purpose is better addressed by the <link

        linkend="slab">slab allocator</link>.</para>

      </section>

      <section>

        <title>Implementation</title>

        <para>The buddy allocator is, in fact, an abstract framework wich can

        be easily specialized to serve one particular task. It knows nothing

        about the nature of memory it helps to allocate. In order to beat the

        lack of this knowledge, the buddy allocator exports an interface that

        each of its clients is required to implement. When supplied an

        implementation of this interface, the buddy allocator can use

        specialized external functions to find buddy for a block, split and

        coalesce blocks, manipulate block order and mark blocks busy or

        available. For precize documentation of this interface, refer to

        <emphasis>"HelenOS Generic Kernel Reference Manual"</emphasis>.</para>

        <formalpara>

          <title>Data organization</title>

          <para>Each entity allocable by the buddy allocator is required to

          contain space for storing block order number and a link variable

          used to interconnect blocks within the same order.</para>

          <para>Whatever entities are allocated by the buddy allocator, the

          first entity within a block is used to represent the entire block.

          The first entity keeps the order of the whole block. Other entities

          within the block are assigned the magic value

          <constant>BUDDY_INNER_BLOCK</constant>. This is especially important

          for effective identification of buddies in one-dimensional array

          because the entity that represents a potential buddy cannot be

          associated with <constant>BUDDY_INNER_BLOCK</constant> (i.e. if it

          is associated with <constant>BUDDY_INNER_BLOCK</constant> then it is

          not a buddy).</para>

          <para>Buddy allocator always uses first frame to represent frame

          block. This frame contains <varname>buddy_order</varname> variable

          to provide information about the block size it actually represents (

          <mathphrase>2<superscript>buddy_order</superscript></mathphrase>

          frames block). Other frames in block have this value set to magic

          <constant>BUDDY_INNER_BLOCK</constant> that is much greater than

          buddy <varname>max_order</varname> value.</para>

          <para>Each <varname>frame_t</varname> also contains pointer member

          to hold frame structure in the linked list inside one order.</para>

        </formalpara>

        <formalpara>

          <title>Allocation algorithm</title>

          <para>Upon <mathphrase>2<superscript>i</superscript></mathphrase>

          frames block allocation request, allocator checks if there are any

          blocks available at the order list <varname>i</varname>. If yes,

          removes block from order list and returns its address. If no,

          recursively allocates

          <mathphrase>2<superscript>i+1</superscript></mathphrase> frame

          block, splits it into two

          <mathphrase>2<superscript>i</superscript></mathphrase> frame blocks.

          Then adds one of the blocks to the <varname>i</varname> order list

          and returns address of another.</para>

        </formalpara>

        <formalpara>

          <title>Deallocation algorithm</title>

          <para>Check if block has so called buddy (another free

          <mathphrase>2<superscript>i</superscript></mathphrase> frame block

          that can be linked with freed block into the

          <mathphrase>2<superscript>i+1</superscript></mathphrase> block).

          Technically, buddy is a odd/even block for even/odd block

          respectively. Plus we can put an extra requirement, that resulting

          block must be aligned to its size. This requirement guarantees

          natural block alignment for the blocks coming out the allocation

          system.</para>

          <para>Using direct pointer arithmetics,

          <varname>frame_t::ref_count</varname> and

          <varname>frame_t::buddy_order</varname> variables, finding buddy is

          done at constant time.</para>

        </formalpara>

      </section>

    </section>

    <section id="slab">

      <title>Slab allocator</title>

      <section>

        <title>Overview</title>

        <para><termdef><glossterm>Slab</glossterm> represents a contiguous

        piece of memory, usually made of several physically contiguous

        pages.</termdef> <termdef><glossterm>Slab cache</glossterm> consists

        of one or more slabs.</termdef></para>

        <para>The majority of memory allocation requests in the kernel are for

        small, frequently used data structures. For this purpose the slab

        allocator is a perfect solution. The basic idea behind the slab

        allocator is to have lists of commonly used objects available packed

        into pages. This avoids the overhead of allocating and destroying

        commonly used types of objects such threads, virtual memory structures

        etc. Also due to the exact allocated size matching, slab allocation

        completely eliminates internal fragmentation issue.</para>

      </section>

      <section>

        <title>Implementation</title>

        <para><mediaobject id="slab_alloc">

            <imageobject role="html">

              <imagedata fileref="images/slab_alloc.png" format="PNG" />

            </imageobject>

            <imageobject role="fop">

              <imagedata fileref="images.vector/slab_alloc.svg" format="SVG" />

            </imageobject>

          </mediaobject></para>

        <para>The SLAB allocator is closely modelled after <ulink

        url="http://www.usenix.org/events/usenix01/full_papers/bonwick/bonwick_html/">

        OpenSolaris SLAB allocator by Jeff Bonwick and Jonathan Adams </ulink>

        with the following exceptions: <itemizedlist>

            <listitem>

               empty SLABS are deallocated immediately (in Linux they are kept in linked list, in Solaris ???)

            </listitem>

            <listitem>

               empty magazines are deallocated when not needed (in Solaris they are held in linked list in slab cache)

            </listitem>

          </itemizedlist> Following features are not currently supported but

        would be easy to do: <itemizedlist>

            <listitem>

               - cache coloring

            </listitem>

            <listitem>

               - dynamic magazine grow (different magazine sizes are already supported, but we would need to adjust allocation strategy)

            </listitem>

          </itemizedlist></para>

        <section>

          <title>Magazine layer</title>

          <para>Due to the extensive bottleneck on SMP architures, caused by

          global SLAB locking mechanism, making processing of all slab

          allocation requests serialized, a new layer was introduced to the

          classic slab allocator design. Slab allocator was extended to

          support per-CPU caches 'magazines' to achieve good SMP scaling.

          <termdef>Slab SMP perfromance bottleneck was resolved by introducing

          a per-CPU caching scheme called as <glossterm>magazine

          layer</glossterm></termdef>.</para>

          <para>Magazine is a N-element cache of objects, so each magazine can

          satisfy N allocations. Magazine behaves like a automatic weapon

          magazine (LIFO, stack), so the allocation/deallocation become simple

          push/pop pointer operation. Trick is that CPU does not access global

          slab allocator data during the allocation from its magazine, thus

          making possible parallel allocations between CPUs.</para>

          <para>Implementation also requires adding another feature as the

          CPU-bound magazine is actually a pair of magazines to avoid

          thrashing when during allocation/deallocatiion of 1 item at the

          magazine size boundary. LIFO order is enforced, which should avoid

          fragmentation as much as possible.</para>

          <para>Another important entity of magazine layer is a full magazine

          depot, that stores full magazines which are used by any of the CPU

          magazine caches to reload active CPU magazine. Magazine depot can be

          pre-filled with full magazines during initialization, but in current

          implementation it is filled during object deallocation, when CPU

          magazine becomes full.</para>

          <para>Slab allocator control structures are allocated from special

          slabs, that are marked by special flag, indicating that it should

          not be used for slab magazine layer. This is done to avoid possible

          infinite recursions and deadlock during conventional slab allocaiton

          requests.</para>

        </section>

        <section>

          <title>Allocation/deallocation</title>

          <para>Every cache contains list of full slabs and list of partialy

          full slabs. Empty slabs are immediately freed (thrashing will be

          avoided because of magazines).</para>

          <para>The SLAB allocator allocates lots of space and does not free

          it. When frame allocator fails to allocate the frame, it calls

          slab_reclaim(). It tries 'light reclaim' first, then brutal reclaim.

          The light reclaim releases slabs from cpu-shared magazine-list,

          until at least 1 slab is deallocated in each cache (this algorithm

          should probably change). The brutal reclaim removes all cached

          objects, even from CPU-bound magazines.</para>

          <formalpara>

            <title>Allocation</title>

            <para><emphasis>Step 1.</emphasis> When it comes to the allocation

            request, slab allocator first of all checks availability of memory

            in local CPU-bound magazine. If it is there, we would just "pop"

            the CPU magazine and return the pointer to object.</para>

            <para><emphasis>Step 2.</emphasis> If the CPU-bound magazine is

            empty, allocator will attempt to reload magazin, swapping it with

            second CPU magazine and returns to the first step.</para>

            <para><emphasis>Step 3.</emphasis> Now we are in the situation

            when both CPU-bound magazines are empty, which makes allocator to

            access shared full-magazines depot to reload CPU-bound magazines.

            If reload is succesful (meaning there are full magazines in depot)

            algoritm continues at Step 1.</para>

            <para><emphasis>Step 4.</emphasis> Final step of the allocation.

            In this step object is allocated from the conventional slab layer

            and pointer is returned.</para>

          </formalpara>

          <formalpara>

            <title>Deallocation</title>

            <para><emphasis>Step 1.</emphasis> During deallocation request,

            slab allocator will check if the local CPU-bound magazine is not

            full. In this case we will just push the pointer to this

            magazine.</para>

            <para><emphasis>Step 2.</emphasis> If the CPU-bound magazine is

            full, allocator will attempt to reload magazin, swapping it with

            second CPU magazine and returns to the first step.</para>

            <para><emphasis>Step 3.</emphasis> Now we are in the situation

            when both CPU-bound magazines are full, which makes allocator to

            access shared full-magazines depot to put one of the magazines to

            the depot and creating new empty magazine. Algoritm continues at

            Step 1.</para>

          </formalpara>

        </section>

      </section>

    </section>

    <!-- End of Physmem -->

  </section>

  <section>

    <title>Memory sharing</title>

    <para>Not implemented yet(?)</para>

  </section>

</chapter>