<?xml version="1.0" encoding="UTF-8"?>
<chapter id="mm">
<?dbhtml filename="mm.html"?>
<title>Memory management</title>
<section>
<!-- VM -->
<title>Virtual memory management</title>
<section>
<title>Address spaces</title>
<para></para>
</section>
<section>
<title>Virtual address translation</title>
<para></para>
</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>
<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>
<section id="buddy_allocator">
<title>Buddy allocator</title>
<section>
<title>Overview</title>
<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 <link linkend="???">HelenOS
Generic Kernel Reference Manual</link>.</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>
</formalpara>
<formalpara>
<title>Data organization</title>
<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 id="slab">
<title>Slab allocator</title>
<section>
<title>Introduction</title>
<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 a 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 as inodes, threads, virtual memory
structures etc.</para>
<para>Original slab allocator locking mechanism has become a
significant preformance bottleneck on SMP architectures. <termdef>Slab
SMP perfromance bottleneck was resolved by introducing a per-CPU
caching scheme called as <glossterm>magazine
layer</glossterm></termdef>.</para>
</section>
<section>
<title>Implementation details (needs revision)</title>
<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 growing (different magazine sizes are already supported, but we would need to adjust allocation strategy)
</listitem>
</itemizedlist></para>
<para>The SLAB allocator supports per-CPU caches ('magazines') to
facilitate good SMP scaling.</para>
<para>When a new object is being allocated, it is first checked, if it
is available in CPU-bound magazine. If it is not found there, it is
allocated from CPU-shared SLAB - if partial full is found, it is used,
otherwise a new one is allocated.</para>
<para>When an object is being deallocated, it is put to CPU-bound
magazine. If there is no such magazine, new one is allocated (if it
fails, the object is deallocated into SLAB). If the magazine is full,
it is put into cpu-shared list of magazines and new one is
allocated.</para>
<para>The CPU-bound magazine is actually a pair of magazines to avoid
thrashing when somebody is allocating/deallocating 1 item at the
magazine size boundary. LIFO order is enforced, which should avoid
fragmentation as much as possible.</para>
<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 information structure is kept inside the data area, if
possible. The cache can be marked that it should not use magazines.
This is used only for SLAB related caches to avoid deadlocks and
infinite recursion (the SLAB allocator uses itself for allocating all
it's control structures).</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>
<para>TODO: <itemizedlist>
<listitem>
For better CPU-scaling the magazine allocation strategy should be extended. Currently, if the cache does not have magazine, it asks for non-cpu cached magazine cache to provide one. It might be feasible to add cpu-cached magazine cache (which would allocate it's magazines from non-cpu-cached mag. cache). This would provide a nice per-cpu buffer. The other possibility is to use the per-cache 'empty-magazine-list', which decreases competing for 1 per-system magazine cache.
</listitem>
<listitem>
- it might be good to add granularity of locks even to slab level, we could then try_spinlock over all partial slabs and thus improve scalability even on slab level
</listitem>
</itemizedlist></para>
</section>
</section>
<!-- End of Physmem -->
</section>
<section>
<title>Memory sharing</title>
<para>Not implemented yet(?)</para>
</section>
</chapter>