xref: /dpdk/doc/guides/prog_guide/env_abstraction_layer.rst (revision 2a1e2da1bb34c7cc0145fdd8940047ae325277bc)
1..  SPDX-License-Identifier: BSD-3-Clause
2    Copyright(c) 2010-2014 Intel Corporation.
3
4.. _Environment_Abstraction_Layer:
5
6Environment Abstraction Layer
7=============================
8
9The Environment Abstraction Layer (EAL) is responsible for gaining access to low-level resources such as hardware and memory space.
10It provides a generic interface that hides the environment specifics from the applications and libraries.
11It is the responsibility of the initialization routine to decide how to allocate these resources
12(that is, memory space, PCI devices, timers, consoles, and so on).
13
14Typical services expected from the EAL are:
15
16*   DPDK Loading and Launching:
17    The DPDK and its application are linked as a single application and must be loaded by some means.
18
19*   Core Affinity/Assignment Procedures:
20    The EAL provides mechanisms for assigning execution units to specific cores as well as creating execution instances.
21
22*   System Memory Reservation:
23    The EAL facilitates the reservation of different memory zones, for example, physical memory areas for device interactions.
24
25*   PCI Address Abstraction: The EAL provides an interface to access PCI address space.
26
27*   Trace and Debug Functions: Logs, dump_stack, panic and so on.
28
29*   Utility Functions: Spinlocks and atomic counters that are not provided in libc.
30
31*   CPU Feature Identification: Determine at runtime if a particular feature, for example, Intel® AVX is supported.
32    Determine if the current CPU supports the feature set that the binary was compiled for.
33
34*   Interrupt Handling: Interfaces to register/unregister callbacks to specific interrupt sources.
35
36*   Alarm Functions: Interfaces to set/remove callbacks to be run at a specific time.
37
38EAL in a Linux-userland Execution Environment
39---------------------------------------------
40
41In a Linux user space environment, the DPDK application runs as a user-space application using the pthread library.
42PCI information about devices and address space is discovered through the /sys kernel interface and through kernel modules such as uio_pci_generic, or igb_uio.
43Refer to the UIO: User-space drivers documentation in the Linux kernel. This memory is mmap'd in the application.
44
45The EAL performs physical memory allocation using mmap() in hugetlbfs (using huge page sizes to increase performance).
46This memory is exposed to DPDK service layers such as the :ref:`Mempool Library <Mempool_Library>`.
47
48At this point, the DPDK services layer will be initialized, then through pthread setaffinity calls,
49each execution unit will be assigned to a specific logical core to run as a user-level thread.
50
51The time reference is provided by the CPU Time-Stamp Counter (TSC) or by the HPET kernel API through a mmap() call.
52
53Initialization and Core Launching
54~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
55
56Part of the initialization is done by the start function of glibc.
57A check is also performed at initialization time to ensure that the micro architecture type chosen in the config file is supported by the CPU.
58Then, the main() function is called. The core initialization and launch is done in rte_eal_init() (see the API documentation).
59It consist of calls to the pthread library (more specifically, pthread_self(), pthread_create(), and pthread_setaffinity_np()).
60
61.. _figure_linuxapp_launch:
62
63.. figure:: img/linuxapp_launch.*
64
65   EAL Initialization in a Linux Application Environment
66
67
68.. note::
69
70    Initialization of objects, such as memory zones, rings, memory pools, lpm tables and hash tables,
71    should be done as part of the overall application initialization on the master lcore.
72    The creation and initialization functions for these objects are not multi-thread safe.
73    However, once initialized, the objects themselves can safely be used in multiple threads simultaneously.
74
75Shutdown and Cleanup
76~~~~~~~~~~~~~~~~~~~~
77
78During the initialization of EAL resources such as hugepage backed memory can be
79allocated by core components.  The memory allocated during ``rte_eal_init()``
80can be released by calling the ``rte_eal_cleanup()`` function. Refer to the
81API documentation for details.
82
83Multi-process Support
84~~~~~~~~~~~~~~~~~~~~~
85
86The Linuxapp EAL allows a multi-process as well as a multi-threaded (pthread) deployment model.
87See chapter
88:ref:`Multi-process Support <Multi-process_Support>` for more details.
89
90Memory Mapping Discovery and Memory Reservation
91~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
92
93The allocation of large contiguous physical memory is done using the hugetlbfs kernel filesystem.
94The EAL provides an API to reserve named memory zones in this contiguous memory.
95The physical address of the reserved memory for that memory zone is also returned to the user by the memory zone reservation API.
96
97.. note::
98
99    Memory reservations done using the APIs provided by rte_malloc are also backed by pages from the hugetlbfs filesystem.
100
101PCI Access
102~~~~~~~~~~
103
104The EAL uses the /sys/bus/pci utilities provided by the kernel to scan the content on the PCI bus.
105To access PCI memory, a kernel module called uio_pci_generic provides a /dev/uioX device file
106and resource files in /sys
107that can be mmap'd to obtain access to PCI address space from the application.
108The DPDK-specific igb_uio module can also be used for this. Both drivers use the uio kernel feature (userland driver).
109
110Per-lcore and Shared Variables
111~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
112
113.. note::
114
115    lcore refers to a logical execution unit of the processor, sometimes called a hardware *thread*.
116
117Shared variables are the default behavior.
118Per-lcore variables are implemented using *Thread Local Storage* (TLS) to provide per-thread local storage.
119
120Logs
121~~~~
122
123A logging API is provided by EAL.
124By default, in a Linux application, logs are sent to syslog and also to the console.
125However, the log function can be overridden by the user to use a different logging mechanism.
126
127Trace and Debug Functions
128^^^^^^^^^^^^^^^^^^^^^^^^^
129
130There are some debug functions to dump the stack in glibc.
131The rte_panic() function can voluntarily provoke a SIG_ABORT,
132which can trigger the generation of a core file, readable by gdb.
133
134CPU Feature Identification
135~~~~~~~~~~~~~~~~~~~~~~~~~~
136
137The EAL can query the CPU at runtime (using the rte_cpu_get_features() function) to determine which CPU features are available.
138
139User Space Interrupt Event
140~~~~~~~~~~~~~~~~~~~~~~~~~~
141
142+ User Space Interrupt and Alarm Handling in Host Thread
143
144The EAL creates a host thread to poll the UIO device file descriptors to detect the interrupts.
145Callbacks can be registered or unregistered by the EAL functions for a specific interrupt event
146and are called in the host thread asynchronously.
147The EAL also allows timed callbacks to be used in the same way as for NIC interrupts.
148
149.. note::
150
151    In DPDK PMD, the only interrupts handled by the dedicated host thread are those for link status change
152    (link up and link down notification) and for sudden device removal.
153
154
155+ RX Interrupt Event
156
157The receive and transmit routines provided by each PMD don't limit themselves to execute in polling thread mode.
158To ease the idle polling with tiny throughput, it's useful to pause the polling and wait until the wake-up event happens.
159The RX interrupt is the first choice to be such kind of wake-up event, but probably won't be the only one.
160
161EAL provides the event APIs for this event-driven thread mode.
162Taking linuxapp as an example, the implementation relies on epoll. Each thread can monitor an epoll instance
163in which all the wake-up events' file descriptors are added. The event file descriptors are created and mapped to
164the interrupt vectors according to the UIO/VFIO spec.
165From bsdapp's perspective, kqueue is the alternative way, but not implemented yet.
166
167EAL initializes the mapping between event file descriptors and interrupt vectors, while each device initializes the mapping
168between interrupt vectors and queues. In this way, EAL actually is unaware of the interrupt cause on the specific vector.
169The eth_dev driver takes responsibility to program the latter mapping.
170
171.. note::
172
173    Per queue RX interrupt event is only allowed in VFIO which supports multiple MSI-X vector. In UIO, the RX interrupt
174    together with other interrupt causes shares the same vector. In this case, when RX interrupt and LSC(link status change)
175    interrupt are both enabled(intr_conf.lsc == 1 && intr_conf.rxq == 1), only the former is capable.
176
177The RX interrupt are controlled/enabled/disabled by ethdev APIs - 'rte_eth_dev_rx_intr_*'. They return failure if the PMD
178hasn't support them yet. The intr_conf.rxq flag is used to turn on the capability of RX interrupt per device.
179
180+ Device Removal Event
181
182This event is triggered by a device being removed at a bus level. Its
183underlying resources may have been made unavailable (i.e. PCI mappings
184unmapped). The PMD must make sure that on such occurrence, the application can
185still safely use its callbacks.
186
187This event can be subscribed to in the same way one would subscribe to a link
188status change event. The execution context is thus the same, i.e. it is the
189dedicated interrupt host thread.
190
191Considering this, it is likely that an application would want to close a
192device having emitted a Device Removal Event. In such case, calling
193``rte_eth_dev_close()`` can trigger it to unregister its own Device Removal Event
194callback. Care must be taken not to close the device from the interrupt handler
195context. It is necessary to reschedule such closing operation.
196
197Blacklisting
198~~~~~~~~~~~~
199
200The EAL PCI device blacklist functionality can be used to mark certain NIC ports as blacklisted,
201so they are ignored by the DPDK.
202The ports to be blacklisted are identified using the PCIe* description (Domain:Bus:Device.Function).
203
204Misc Functions
205~~~~~~~~~~~~~~
206
207Locks and atomic operations are per-architecture (i686 and x86_64).
208
209Memory Segments and Memory Zones (memzone)
210------------------------------------------
211
212The mapping of physical memory is provided by this feature in the EAL.
213As physical memory can have gaps, the memory is described in a table of descriptors,
214and each descriptor (called rte_memseg ) describes a contiguous portion of memory.
215
216On top of this, the memzone allocator's role is to reserve contiguous portions of physical memory.
217These zones are identified by a unique name when the memory is reserved.
218
219The rte_memzone descriptors are also located in the configuration structure.
220This structure is accessed using rte_eal_get_configuration().
221The lookup (by name) of a memory zone returns a descriptor containing the physical address of the memory zone.
222
223Memory zones can be reserved with specific start address alignment by supplying the align parameter
224(by default, they are aligned to cache line size).
225The alignment value should be a power of two and not less than the cache line size (64 bytes).
226Memory zones can also be reserved from either 2 MB or 1 GB hugepages, provided that both are available on the system.
227
228
229Multiple pthread
230----------------
231
232DPDK usually pins one pthread per core to avoid the overhead of task switching.
233This allows for significant performance gains, but lacks flexibility and is not always efficient.
234
235Power management helps to improve the CPU efficiency by limiting the CPU runtime frequency.
236However, alternately it is possible to utilize the idle cycles available to take advantage of
237the full capability of the CPU.
238
239By taking advantage of cgroup, the CPU utilization quota can be simply assigned.
240This gives another way to improve the CPU efficiency, however, there is a prerequisite;
241DPDK must handle the context switching between multiple pthreads per core.
242
243For further flexibility, it is useful to set pthread affinity not only to a CPU but to a CPU set.
244
245EAL pthread and lcore Affinity
246~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
247
248The term "lcore" refers to an EAL thread, which is really a Linux/FreeBSD pthread.
249"EAL pthreads"  are created and managed by EAL and execute the tasks issued by *remote_launch*.
250In each EAL pthread, there is a TLS (Thread Local Storage) called *_lcore_id* for unique identification.
251As EAL pthreads usually bind 1:1 to the physical CPU, the *_lcore_id* is typically equal to the CPU ID.
252
253When using multiple pthreads, however, the binding is no longer always 1:1 between an EAL pthread and a specified physical CPU.
254The EAL pthread may have affinity to a CPU set, and as such the *_lcore_id* will not be the same as the CPU ID.
255For this reason, there is an EAL long option '--lcores' defined to assign the CPU affinity of lcores.
256For a specified lcore ID or ID group, the option allows setting the CPU set for that EAL pthread.
257
258The format pattern:
259	--lcores='<lcore_set>[@cpu_set][,<lcore_set>[@cpu_set],...]'
260
261'lcore_set' and 'cpu_set' can be a single number, range or a group.
262
263A number is a "digit([0-9]+)"; a range is "<number>-<number>"; a group is "(<number|range>[,<number|range>,...])".
264
265If a '\@cpu_set' value is not supplied, the value of 'cpu_set' will default to the value of 'lcore_set'.
266
267    ::
268
269    	For example, "--lcores='1,2@(5-7),(3-5)@(0,2),(0,6),7-8'" which means start 9 EAL thread;
270    	    lcore 0 runs on cpuset 0x41 (cpu 0,6);
271    	    lcore 1 runs on cpuset 0x2 (cpu 1);
272    	    lcore 2 runs on cpuset 0xe0 (cpu 5,6,7);
273    	    lcore 3,4,5 runs on cpuset 0x5 (cpu 0,2);
274    	    lcore 6 runs on cpuset 0x41 (cpu 0,6);
275    	    lcore 7 runs on cpuset 0x80 (cpu 7);
276    	    lcore 8 runs on cpuset 0x100 (cpu 8).
277
278Using this option, for each given lcore ID, the associated CPUs can be assigned.
279It's also compatible with the pattern of corelist('-l') option.
280
281non-EAL pthread support
282~~~~~~~~~~~~~~~~~~~~~~~
283
284It is possible to use the DPDK execution context with any user pthread (aka. Non-EAL pthreads).
285In a non-EAL pthread, the *_lcore_id* is always LCORE_ID_ANY which identifies that it is not an EAL thread with a valid, unique, *_lcore_id*.
286Some libraries will use an alternative unique ID (e.g. TID), some will not be impacted at all, and some will work but with limitations (e.g. timer and mempool libraries).
287
288All these impacts are mentioned in :ref:`known_issue_label` section.
289
290Public Thread API
291~~~~~~~~~~~~~~~~~
292
293There are two public APIs ``rte_thread_set_affinity()`` and ``rte_thread_get_affinity()`` introduced for threads.
294When they're used in any pthread context, the Thread Local Storage(TLS) will be set/get.
295
296Those TLS include *_cpuset* and *_socket_id*:
297
298*	*_cpuset* stores the CPUs bitmap to which the pthread is affinitized.
299
300*	*_socket_id* stores the NUMA node of the CPU set. If the CPUs in CPU set belong to different NUMA node, the *_socket_id* will be set to SOCKET_ID_ANY.
301
302
303.. _known_issue_label:
304
305Known Issues
306~~~~~~~~~~~~
307
308+ rte_mempool
309
310  The rte_mempool uses a per-lcore cache inside the mempool.
311  For non-EAL pthreads, ``rte_lcore_id()`` will not return a valid number.
312  So for now, when rte_mempool is used with non-EAL pthreads, the put/get operations will bypass the default mempool cache and there is a performance penalty because of this bypass.
313  Only user-owned external caches can be used in a non-EAL context in conjunction with ``rte_mempool_generic_put()`` and ``rte_mempool_generic_get()`` that accept an explicit cache parameter.
314
315+ rte_ring
316
317  rte_ring supports multi-producer enqueue and multi-consumer dequeue.
318  However, it is non-preemptive, this has a knock on effect of making rte_mempool non-preemptable.
319
320  .. note::
321
322    The "non-preemptive" constraint means:
323
324    - a pthread doing multi-producers enqueues on a given ring must not
325      be preempted by another pthread doing a multi-producer enqueue on
326      the same ring.
327    - a pthread doing multi-consumers dequeues on a given ring must not
328      be preempted by another pthread doing a multi-consumer dequeue on
329      the same ring.
330
331    Bypassing this constraint may cause the 2nd pthread to spin until the 1st one is scheduled again.
332    Moreover, if the 1st pthread is preempted by a context that has an higher priority, it may even cause a dead lock.
333
334  This does not mean it cannot be used, simply, there is a need to narrow down the situation when it is used by multi-pthread on the same core.
335
336  1. It CAN be used for any single-producer or single-consumer situation.
337
338  2. It MAY be used by multi-producer/consumer pthread whose scheduling policy are all SCHED_OTHER(cfs). User SHOULD be aware of the performance penalty before using it.
339
340  3. It MUST not be used by multi-producer/consumer pthreads, whose scheduling policies are SCHED_FIFO or SCHED_RR.
341
342+ rte_timer
343
344  Running  ``rte_timer_manage()`` on a non-EAL pthread is not allowed. However, resetting/stopping the timer from a non-EAL pthread is allowed.
345
346+ rte_log
347
348  In non-EAL pthreads, there is no per thread loglevel and logtype, global loglevels are used.
349
350+ misc
351
352  The debug statistics of rte_ring, rte_mempool and rte_timer are not supported in a non-EAL pthread.
353
354cgroup control
355~~~~~~~~~~~~~~
356
357The following is a simple example of cgroup control usage, there are two pthreads(t0 and t1) doing packet I/O on the same core ($CPU).
358We expect only 50% of CPU spend on packet IO.
359
360  .. code-block:: console
361
362    mkdir /sys/fs/cgroup/cpu/pkt_io
363    mkdir /sys/fs/cgroup/cpuset/pkt_io
364
365    echo $cpu > /sys/fs/cgroup/cpuset/cpuset.cpus
366
367    echo $t0 > /sys/fs/cgroup/cpu/pkt_io/tasks
368    echo $t0 > /sys/fs/cgroup/cpuset/pkt_io/tasks
369
370    echo $t1 > /sys/fs/cgroup/cpu/pkt_io/tasks
371    echo $t1 > /sys/fs/cgroup/cpuset/pkt_io/tasks
372
373    cd /sys/fs/cgroup/cpu/pkt_io
374    echo 100000 > pkt_io/cpu.cfs_period_us
375    echo  50000 > pkt_io/cpu.cfs_quota_us
376
377
378Malloc
379------
380
381The EAL provides a malloc API to allocate any-sized memory.
382
383The objective of this API is to provide malloc-like functions to allow
384allocation from hugepage memory and to facilitate application porting.
385The *DPDK API Reference* manual describes the available functions.
386
387Typically, these kinds of allocations should not be done in data plane
388processing because they are slower than pool-based allocation and make
389use of locks within the allocation and free paths.
390However, they can be used in configuration code.
391
392Refer to the rte_malloc() function description in the *DPDK API Reference*
393manual for more information.
394
395Cookies
396~~~~~~~
397
398When CONFIG_RTE_MALLOC_DEBUG is enabled, the allocated memory contains
399overwrite protection fields to help identify buffer overflows.
400
401Alignment and NUMA Constraints
402~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
403
404The rte_malloc() takes an align argument that can be used to request a memory
405area that is aligned on a multiple of this value (which must be a power of two).
406
407On systems with NUMA support, a call to the rte_malloc() function will return
408memory that has been allocated on the NUMA socket of the core which made the call.
409A set of APIs is also provided, to allow memory to be explicitly allocated on a
410NUMA socket directly, or by allocated on the NUMA socket where another core is
411located, in the case where the memory is to be used by a logical core other than
412on the one doing the memory allocation.
413
414Use Cases
415~~~~~~~~~
416
417This API is meant to be used by an application that requires malloc-like
418functions at initialization time.
419
420For allocating/freeing data at runtime, in the fast-path of an application,
421the memory pool library should be used instead.
422
423Internal Implementation
424~~~~~~~~~~~~~~~~~~~~~~~
425
426Data Structures
427^^^^^^^^^^^^^^^
428
429There are two data structure types used internally in the malloc library:
430
431*   struct malloc_heap - used to track free space on a per-socket basis
432
433*   struct malloc_elem - the basic element of allocation and free-space
434    tracking inside the library.
435
436Structure: malloc_heap
437""""""""""""""""""""""
438
439The malloc_heap structure is used to manage free space on a per-socket basis.
440Internally, there is one heap structure per NUMA node, which allows us to
441allocate memory to a thread based on the NUMA node on which this thread runs.
442While this does not guarantee that the memory will be used on that NUMA node,
443it is no worse than a scheme where the memory is always allocated on a fixed
444or random node.
445
446The key fields of the heap structure and their function are described below
447(see also diagram above):
448
449*   lock - the lock field is needed to synchronize access to the heap.
450    Given that the free space in the heap is tracked using a linked list,
451    we need a lock to prevent two threads manipulating the list at the same time.
452
453*   free_head - this points to the first element in the list of free nodes for
454    this malloc heap.
455
456.. note::
457
458    The malloc_heap structure does not keep track of in-use blocks of memory,
459    since these are never touched except when they are to be freed again -
460    at which point the pointer to the block is an input to the free() function.
461
462.. _figure_malloc_heap:
463
464.. figure:: img/malloc_heap.*
465
466   Example of a malloc heap and malloc elements within the malloc library
467
468
469.. _malloc_elem:
470
471Structure: malloc_elem
472""""""""""""""""""""""
473
474The malloc_elem structure is used as a generic header structure for various
475blocks of memory.
476It is used in three different ways - all shown in the diagram above:
477
478#.  As a header on a block of free or allocated memory - normal case
479
480#.  As a padding header inside a block of memory
481
482#.  As an end-of-memseg marker
483
484The most important fields in the structure and how they are used are described below.
485
486.. note::
487
488    If the usage of a particular field in one of the above three usages is not
489    described, the field can be assumed to have an undefined value in that
490    situation, for example, for padding headers only the "state" and "pad"
491    fields have valid values.
492
493*   heap - this pointer is a reference back to the heap structure from which
494    this block was allocated.
495    It is used for normal memory blocks when they are being freed, to add the
496    newly-freed block to the heap's free-list.
497
498*   prev - this pointer points to the header element/block in the memseg
499    immediately behind the current one. When freeing a block, this pointer is
500    used to reference the previous block to check if that block is also free.
501    If so, then the two free blocks are merged to form a single larger block.
502
503*   next_free - this pointer is used to chain the free-list of unallocated
504    memory blocks together.
505    It is only used in normal memory blocks; on ``malloc()`` to find a suitable
506    free block to allocate and on ``free()`` to add the newly freed element to
507    the free-list.
508
509*   state - This field can have one of three values: ``FREE``, ``BUSY`` or
510    ``PAD``.
511    The former two are to indicate the allocation state of a normal memory block
512    and the latter is to indicate that the element structure is a dummy structure
513    at the end of the start-of-block padding, i.e. where the start of the data
514    within a block is not at the start of the block itself, due to alignment
515    constraints.
516    In that case, the pad header is used to locate the actual malloc element
517    header for the block.
518    For the end-of-memseg structure, this is always a ``BUSY`` value, which
519    ensures that no element, on being freed, searches beyond the end of the
520    memseg for other blocks to merge with into a larger free area.
521
522*   pad - this holds the length of the padding present at the start of the block.
523    In the case of a normal block header, it is added to the address of the end
524    of the header to give the address of the start of the data area, i.e. the
525    value passed back to the application on a malloc.
526    Within a dummy header inside the padding, this same value is stored, and is
527    subtracted from the address of the dummy header to yield the address of the
528    actual block header.
529
530*   size - the size of the data block, including the header itself.
531    For end-of-memseg structures, this size is given as zero, though it is never
532    actually checked.
533    For normal blocks which are being freed, this size value is used in place of
534    a "next" pointer to identify the location of the next block of memory that
535    in the case of being ``FREE``, the two free blocks can be merged into one.
536
537Memory Allocation
538^^^^^^^^^^^^^^^^^
539
540On EAL initialization, all memsegs are setup as part of the malloc heap.
541This setup involves placing a dummy structure at the end with ``BUSY`` state,
542which may contain a sentinel value if ``CONFIG_RTE_MALLOC_DEBUG`` is enabled,
543and a proper :ref:`element header<malloc_elem>` with ``FREE`` at the start
544for each memseg.
545The ``FREE`` element is then added to the ``free_list`` for the malloc heap.
546
547When an application makes a call to a malloc-like function, the malloc function
548will first index the ``lcore_config`` structure for the calling thread, and
549determine the NUMA node of that thread.
550The NUMA node is used to index the array of ``malloc_heap`` structures which is
551passed as a parameter to the ``heap_alloc()`` function, along with the
552requested size, type, alignment and boundary parameters.
553
554The ``heap_alloc()`` function will scan the free_list of the heap, and attempt
555to find a free block suitable for storing data of the requested size, with the
556requested alignment and boundary constraints.
557
558When a suitable free element has been identified, the pointer to be returned
559to the user is calculated.
560The cache-line of memory immediately preceding this pointer is filled with a
561struct malloc_elem header.
562Because of alignment and boundary constraints, there could be free space at
563the start and/or end of the element, resulting in the following behavior:
564
565#. Check for trailing space.
566   If the trailing space is big enough, i.e. > 128 bytes, then the free element
567   is split.
568   If it is not, then we just ignore it (wasted space).
569
570#. Check for space at the start of the element.
571   If the space at the start is small, i.e. <=128 bytes, then a pad header is
572   used, and the remaining space is wasted.
573   If, however, the remaining space is greater, then the free element is split.
574
575The advantage of allocating the memory from the end of the existing element is
576that no adjustment of the free list needs to take place - the existing element
577on the free list just has its size pointer adjusted, and the following element
578has its "prev" pointer redirected to the newly created element.
579
580Freeing Memory
581^^^^^^^^^^^^^^
582
583To free an area of memory, the pointer to the start of the data area is passed
584to the free function.
585The size of the ``malloc_elem`` structure is subtracted from this pointer to get
586the element header for the block.
587If this header is of type ``PAD`` then the pad length is further subtracted from
588the pointer to get the proper element header for the entire block.
589
590From this element header, we get pointers to the heap from which the block was
591allocated and to where it must be freed, as well as the pointer to the previous
592element, and via the size field, we can calculate the pointer to the next element.
593These next and previous elements are then checked to see if they are also
594``FREE``, and if so, they are merged with the current element.
595This means that we can never have two ``FREE`` memory blocks adjacent to one
596another, as they are always merged into a single block.
597