blob: d524d733179b96126774ea341e5fcc3dce77bc35 [file] [log] [blame]
Paul Beesleyfc9ee362019-03-07 15:47:15 +00001Coding Style & Guidelines
2=========================
Paul Beesley4e2e1b22019-01-17 15:44:37 +00003
Paul Beesley4e2e1b22019-01-17 15:44:37 +00004The following sections contain TF coding guidelines. They are continually
5evolving and should not be considered "set in stone". Feel free to question them
6and provide feedback.
7
8Some of the guidelines may also apply to other codebases.
9
Paul Beesleyba3ed402019-03-13 16:20:44 +000010.. note::
11 The existing TF codebase does not necessarily comply with all the
12 below guidelines but the intent is for it to do so eventually.
Paul Beesley4e2e1b22019-01-17 15:44:37 +000013
Paul Beesley8b4bdeb2019-01-21 12:06:24 +000014Checkpatch overrides
15--------------------
Paul Beesley4e2e1b22019-01-17 15:44:37 +000016
17Some checkpatch warnings in the TF codebase are deliberately ignored. These
18include:
19
20- ``**WARNING: line over 80 characters**``: Although the codebase should
21 generally conform to the 80 character limit this is overly restrictive in some
22 cases.
23
24- ``**WARNING: Use of volatile is usually wrong``: see
25 `Why the “volatile” type class should not be used`_ . Although this document
26 contains some very useful information, there are several legimate uses of the
27 volatile keyword within the TF codebase.
28
Paul Beesleyfd688342019-01-21 16:11:28 +000029Headers and inclusion
30---------------------
31
32Header guards
33^^^^^^^^^^^^^
34
35For a header file called "some_driver.h" the style used by the Trusted Firmware
36is:
37
38.. code:: c
39
40 #ifndef SOME_DRIVER_H
41 #define SOME_DRIVER_H
42
43 <header content>
44
45 #endif /* SOME_DRIVER_H */
46
Paul Beesley5981d112019-01-22 11:36:41 +000047Include statement ordering
48^^^^^^^^^^^^^^^^^^^^^^^^^^
Paul Beesleyfd688342019-01-21 16:11:28 +000049
Paul Beesley5981d112019-01-22 11:36:41 +000050All header files that are included by a source file must use the following,
51grouped ordering. This is to improve readability (by making it easier to quickly
52read through the list of headers) and maintainability.
Paul Beesleyfd688342019-01-21 16:11:28 +000053
Paul Beesley5981d112019-01-22 11:36:41 +000054#. *System* includes: Header files from the standard *C* library, such as
55 ``stddef.h`` and ``string.h``.
Paul Beesleyfd688342019-01-21 16:11:28 +000056
Paul Beesley5981d112019-01-22 11:36:41 +000057#. *Project* includes: Header files under the ``include/`` directory within TF
58 are *project* includes.
59
60#. *Platform* includes: Header files relating to a single, specific platform,
61 and which are located under the ``plat/<platform_name>`` directory within TF,
62 are *platform* includes.
Paul Beesleyfd688342019-01-21 16:11:28 +000063
Paul Beesley5981d112019-01-22 11:36:41 +000064Within each group, ``#include`` statements must be in alphabetical order,
65taking both the file and directory names into account.
66
67Groups must be separated by a single blank line for clarity.
68
69The example below illustrates the ordering rules using some contrived header
70file names; this type of name reuse should be otherwise avoided.
Paul Beesleye5f0f0a2019-01-31 11:39:01 +000071
Paul Beesleyfd688342019-01-21 16:11:28 +000072.. code:: c
73
Paul Beesley5981d112019-01-22 11:36:41 +000074 #include <string.h>
75
76 #include <a_dir/example/a_header.h>
77 #include <a_dir/example/b_header.h>
78 #include <a_dir/test/a_header.h>
79 #include <b_dir/example/a_header.h>
80
81 #include "./a_header.h"
82
83Include statement variants
Sandrine Bailleuxf5a91002019-02-08 10:50:28 +010084^^^^^^^^^^^^^^^^^^^^^^^^^^
Paul Beesley5981d112019-01-22 11:36:41 +000085
86Two variants of the ``#include`` directive are acceptable in the TF codebase.
87Correct use of the two styles improves readability by suggesting the location
88of the included header and reducing ambiguity in cases where generic and
89platform-specific headers share a name.
90
91For header files that are in the same directory as the source file that is
92including them, use the ``"..."`` variant.
Paul Beesleyfd688342019-01-21 16:11:28 +000093
Paul Beesley5981d112019-01-22 11:36:41 +000094For header files that are **not** in the same directory as the source file that
95is including them, use the ``<...>`` variant.
Paul Beesleyfd688342019-01-21 16:11:28 +000096
Paul Beesley5981d112019-01-22 11:36:41 +000097Example (bl1_fwu.c):
Paul Beesleye5f0f0a2019-01-31 11:39:01 +000098
Paul Beesley5981d112019-01-22 11:36:41 +000099.. code:: c
100
101 #include <assert.h>
102 #include <errno.h>
103 #include <string.h>
104
105 #include "bl1_private.h"
106
107Platform include paths
108^^^^^^^^^^^^^^^^^^^^^^
109
110Platforms are allowed to add more include paths to be passed to the compiler.
111The ``PLAT_INCLUDES`` variable is used for this purpose. This is needed in
112particular for the file ``platform_def.h``.
113
114Example:
Paul Beesleye5f0f0a2019-01-31 11:39:01 +0000115
Paul Beesley5981d112019-01-22 11:36:41 +0000116.. code:: c
117
118 PLAT_INCLUDES += -Iinclude/plat/myplat/include
Paul Beesleyfd688342019-01-21 16:11:28 +0000119
120Types and typedefs
121------------------
122
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000123Use of built-in *C* and *libc* data types
Sandrine Bailleuxf5a91002019-02-08 10:50:28 +0100124^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000125
126The TF codebase should be kept as portable as possible, especially since both
12764-bit and 32-bit platforms are supported. To help with this, the following data
128type usage guidelines should be followed:
129
130- Where possible, use the built-in *C* data types for variable storage (for
131 example, ``char``, ``int``, ``long long``, etc) instead of the standard *C99*
132 types. Most code is typically only concerned with the minimum size of the
133 data stored, which the built-in *C* types guarantee.
134
135- Avoid using the exact-size standard *C99* types in general (for example,
136 ``uint16_t``, ``uint32_t``, ``uint64_t``, etc) since they can prevent the
137 compiler from making optimizations. There are legitimate uses for them,
138 for example to represent data of a known structure. When using them in struct
139 definitions, consider how padding in the struct will work across architectures.
140 For example, extra padding may be introduced in AArch32 systems if a struct
141 member crosses a 32-bit boundary.
142
143- Use ``int`` as the default integer type - it's likely to be the fastest on all
144 systems. Also this can be assumed to be 32-bit as a consequence of the
Paul Beesley54c83cf2019-01-21 11:57:42 +0000145 `Procedure Call Standard for the Arm Architecture`_ and the `Procedure Call
146 Standard for the Arm 64-bit Architecture`_ .
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000147
148- Avoid use of ``short`` as this may end up being slower than ``int`` in some
149 systems. If a variable must be exactly 16-bit, use ``int16_t`` or
150 ``uint16_t``.
151
152- Avoid use of ``long``. This is guaranteed to be at least 32-bit but, given
153 that `int` is 32-bit on Arm platforms, there is no use for it. For integers of
154 at least 64-bit, use ``long long``.
155
156- Use ``char`` for storing text. Use ``uint8_t`` for storing other 8-bit data.
157
158- Use ``unsigned`` for integers that can never be negative (counts,
159 indices, sizes, etc). TF intends to comply with MISRA "essential type" coding
160 rules (10.X), where signed and unsigned types are considered different
161 essential types. Choosing the correct type will aid this. MISRA static
162 analysers will pick up any implicit signed/unsigned conversions that may lead
163 to unexpected behaviour.
164
165- For pointer types:
166
167 - If an argument in a function declaration is pointing to a known type then
168 simply use a pointer to that type (for example: ``struct my_struct *``).
169
170 - If a variable (including an argument in a function declaration) is pointing
171 to a general, memory-mapped address, an array of pointers or another
172 structure that is likely to require pointer arithmetic then use
173 ``uintptr_t``. This will reduce the amount of casting required in the code.
174 Avoid using ``unsigned long`` or ``unsigned long long`` for this purpose; it
175 may work but is less portable.
176
177 - For other pointer arguments in a function declaration, use ``void *``. This
178 includes pointers to types that are abstracted away from the known API and
179 pointers to arbitrary data. This allows the calling function to pass a
180 pointer argument to the function without any explicit casting (the cast to
181 ``void *`` is implicit). The function implementation can then do the
182 appropriate casting to a specific type.
183
184 - Use ``ptrdiff_t`` to compare the difference between 2 pointers.
185
186- Use ``size_t`` when storing the ``sizeof()`` something.
187
Paul Beesley244c7832019-01-21 12:02:09 +0000188- Use ``ssize_t`` when returning the ``sizeof()`` something from a function that
189 can also return an error code; the signed type allows for a negative return
190 code in case of error. This practice should be used sparingly.
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000191
192- Use ``u_register_t`` when it's important to store the contents of a register
193 in its native size (32-bit in AArch32 and 64-bit in AArch64). This is not a
194 standard *C99* type but is widely available in libc implementations,
195 including the FreeBSD version included with the TF codebase. Where possible,
196 cast the variable to a more appropriate type before interpreting the data. For
197 example, the following struct in ``ep_info.h`` could use this type to minimize
198 the storage required for the set of registers:
199
200.. code:: c
201
202 typedef struct aapcs64_params {
203 u_register_t arg0;
204 u_register_t arg1;
205 u_register_t arg2;
206 u_register_t arg3;
207 u_register_t arg4;
208 u_register_t arg5;
209 u_register_t arg6;
210 u_register_t arg7;
211 } aapcs64_params_t;
212
Sandrine Bailleuxf5a91002019-02-08 10:50:28 +0100213If some code wants to operate on ``arg0`` and knows that it represents a 32-bit
214unsigned integer on all systems, cast it to ``unsigned int``.
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000215
216These guidelines should be updated if additional types are needed.
217
Paul Beesleyfd688342019-01-21 16:11:28 +0000218Avoid anonymous typedefs of structs/enums in headers
219^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000220
Paul Beesleyfd688342019-01-21 16:11:28 +0000221For example, the following definition:
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000222
Paul Beesleyfd688342019-01-21 16:11:28 +0000223.. code:: c
224
225 typedef struct {
226 int arg1;
227 int arg2;
228 } my_struct_t;
229
230
231is better written as:
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000232
233.. code:: c
234
Paul Beesleyfd688342019-01-21 16:11:28 +0000235 struct my_struct {
236 int arg1;
237 int arg2;
238 };
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000239
Paul Beesleyfd688342019-01-21 16:11:28 +0000240This allows function declarations in other header files that depend on the
241struct/enum to forward declare the struct/enum instead of including the
242entire header:
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000243
Paul Beesleyfd688342019-01-21 16:11:28 +0000244.. code:: c
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000245
Paul Beesleyfd688342019-01-21 16:11:28 +0000246 #include <my_struct.h>
247 void my_func(my_struct_t *arg);
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000248
Paul Beesleyfd688342019-01-21 16:11:28 +0000249instead of:
250
251.. code:: c
252
253 struct my_struct;
254 void my_func(struct my_struct *arg);
255
256Some TF definitions use both a struct/enum name **and** a typedef name. This
257is discouraged for new definitions as it makes it difficult for TF to comply
258with MISRA rule 8.3, which states that "All declarations of an object or
259function shall use the same names and type qualifiers".
260
261The Linux coding standards also discourage new typedefs and checkpatch emits
262a warning for this.
263
264Existing typedefs will be retained for compatibility.
265
266Error handling and robustness
267-----------------------------
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000268
269Using CASSERT to check for compile time data errors
270^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
271
272Where possible, use the ``CASSERT`` macro to check the validity of data known at
273compile time instead of checking validity at runtime, to avoid unnecessary
274runtime code.
275
276For example, this can be used to check that the assembler's and compiler's views
277of the size of an array is the same.
278
279.. code:: c
280
281 #include <cassert.h>
282
283 define MY_STRUCT_SIZE 8 /* Used by assembler source files */
284
285 struct my_struct {
286 uint32_t arg1;
287 uint32_t arg2;
288 };
289
290 CASSERT(MY_STRUCT_SIZE == sizeof(struct my_struct), assert_my_struct_size_mismatch);
291
292
293If ``MY_STRUCT_SIZE`` in the above example were wrong then the compiler would
294emit an error like this:
295
Paul Beesley493e3492019-03-13 15:11:04 +0000296::
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000297
298 my_struct.h:10:1: error: size of array ‘assert_my_struct_size_mismatch’ is negative
299
300
301Using assert() to check for programming errors
302^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
303
304In general, each secure world TF image (BL1, BL2, BL31 and BL32) should be
305treated as a tightly integrated package; the image builder should be aware of
306and responsible for all functionality within the image, even if code within that
307image is provided by multiple entities. This allows us to be more aggressive in
308interpreting invalid state or bad function arguments as programming errors using
309``assert()``, including arguments passed across platform porting interfaces.
310This is in contrast to code in a Linux environment, which is less tightly
311integrated and may attempt to be more defensive by passing the error back up the
312call stack.
313
314Where possible, badly written TF code should fail early using ``assert()``. This
315helps reduce the amount of untested conditional code. By default these
316statements are not compiled into release builds, although this can be overridden
317using the ``ENABLE_ASSERTIONS`` build flag.
318
319Examples:
320
321- Bad argument supplied to library function
322- Bad argument provided by platform porting function
323- Internal secure world image state is inconsistent
324
325
326Handling integration errors
327^^^^^^^^^^^^^^^^^^^^^^^^^^^
328
329Each secure world image may be provided by a different entity (for example, a
330Trusted Boot vendor may provide the BL2 image, a TEE vendor may provide the BL32
331image and the OEM/SoC vendor may provide the other images).
332
333An image may contain bugs that are only visible when the images are integrated.
334The system integrator may not even have access to the debug variants of all the
335images in order to check if asserts are firing. For example, the release variant
336of BL1 may have already been burnt into the SoC. Therefore, TF code that detects
337an integration error should _not_ consider this a programming error, and should
338always take action, even in release builds.
339
340If an integration error is considered non-critical it should be treated as a
341recoverable error. If the error is considered critical it should be treated as
342an unexpected unrecoverable error.
343
344Handling recoverable errors
345^^^^^^^^^^^^^^^^^^^^^^^^^^^
346
347The secure world **must not** crash when supplied with bad data from an external
348source. For example, data from the normal world or a hardware device. Similarly,
349the secure world **must not** crash if it detects a non-critical problem within
350itself or the system. It must make every effort to recover from the problem by
351emitting a ``WARN`` message, performing any necessary error handling and
352continuing.
353
354Examples:
355
356- Secure world receives SMC from normal world with bad arguments.
357- Secure world receives SMC from normal world at an unexpected time.
358- BL31 receives SMC from BL32 with bad arguments.
359- BL31 receives SMC from BL32 at unexpected time.
360- Secure world receives recoverable error from hardware device. Retrying the
361 operation may help here.
362- Non-critical secure world service is not functioning correctly.
363- BL31 SPD discovers minor configuration problem with corresponding SP.
364
365Handling unrecoverable errors
366^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
367
368In some cases it may not be possible for the secure world to recover from an
369error. This situation should be handled in one of the following ways:
370
3711. If the unrecoverable error is unexpected then emit an ``ERROR`` message and
372 call ``panic()``. This will end up calling the platform-specific function
373 ``plat_panic_handler()``.
3742. If the unrecoverable error is expected to occur in certain circumstances,
375 then emit an ``ERROR`` message and call the platform-specific function
376 ``plat_error_handler()``.
377
378Cases 1 and 2 are subtly different. A platform may implement ``plat_panic_handler``
379and ``plat_error_handler`` in the same way (for example, by waiting for a secure
380watchdog to time-out or by invoking an interface on the platform's power
381controller to reset the platform). However, ``plat_error_handler`` may take
382additional action for some errors (for example, it may set a flag so the
383platform resets into a different mode). Also, ``plat_panic_handler()`` may
384implement additional debug functionality (for example, invoking a hardware
385breakpoint).
386
387Examples of unexpected unrecoverable errors:
388
389- BL32 receives an unexpected SMC response from BL31 that it is unable to
390 recover from.
391- BL31 Trusted OS SPD code discovers that BL2 has not loaded the corresponding
392 Trusted OS, which is critical for platform operation.
393- Secure world discovers that a critical hardware device is an unexpected and
394 unrecoverable state.
395- Secure world receives an unexpected and unrecoverable error from a critical
396 hardware device.
397- Secure world discovers that it is running on unsupported hardware.
398
399Examples of expected unrecoverable errors:
400
401- BL1/BL2 fails to load the next image due to missing/corrupt firmware on disk.
402- BL1/BL2 fails to authenticate the next image due to an invalid certificate.
403- Secure world continuously receives recoverable errors from a hardware device
404 but is unable to proceed without a valid response.
405
406Handling critical unresponsiveness
407^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
408
409If the secure world is waiting for a response from an external source (for
410example, the normal world or a hardware device) which is critical for continued
411operation, it must not wait indefinitely. It must have a mechanism (for example,
412a secure watchdog) for resetting itself and/or the external source to prevent
413the system from executing in this state indefinitely.
414
415Examples:
416
417- BL1 is waiting for the normal world to raise an SMC to proceed to the next
418 stage of the secure firmware update process.
419- A Trusted OS is waiting for a response from a proxy in the normal world that
420 is critical for continued operation.
421- Secure world is waiting for a hardware response that is critical for continued
422 operation.
423
424Security considerations
425-----------------------
426
427Part of the security of a platform is handling errors correctly, as described in
428the previous section. There are several other security considerations covered in
429this section.
430
431Do not leak secrets to the normal world
432^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
433
434The secure world **must not** leak secrets to the normal world, for example in
435response to an SMC.
436
437Handling Denial of Service attacks
438^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
439
440The secure world **should never** crash or become unusable due to receiving too
441many normal world requests (a *Denial of Service* or *DoS* attack). It should
442have a mechanism for throttling or ignoring normal world requests.
443
Paul Beesleyfd688342019-01-21 16:11:28 +0000444Performance considerations
445--------------------------
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000446
Paul Beesleyfd688342019-01-21 16:11:28 +0000447Avoid printf and use logging macros
448^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000449
Paul Beesleyfd688342019-01-21 16:11:28 +0000450``debug.h`` provides logging macros (for example, ``WARN`` and ``ERROR``)
451which wrap ``tf_log`` and which allow the logging call to be compiled-out
452depending on the ``make`` command. Use these macros to avoid print statements
453being compiled unconditionally into the binary.
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000454
Paul Beesleyfd688342019-01-21 16:11:28 +0000455Each logging macro has a numerical log level:
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000456
457.. code:: c
458
Paul Beesleyfd688342019-01-21 16:11:28 +0000459 #define LOG_LEVEL_NONE 0
460 #define LOG_LEVEL_ERROR 10
461 #define LOG_LEVEL_NOTICE 20
462 #define LOG_LEVEL_WARNING 30
463 #define LOG_LEVEL_INFO 40
464 #define LOG_LEVEL_VERBOSE 50
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000465
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000466
Paul Beesleyfd688342019-01-21 16:11:28 +0000467By default, all logging statements with a log level ``<= LOG_LEVEL_INFO`` will
468be compiled into debug builds and all statements with a log level
469``<= LOG_LEVEL_NOTICE`` will be compiled into release builds. This can be
470overridden from the command line or by the platform makefile (although it may be
471necessary to clean the build directory first). For example, to enable
472``VERBOSE`` logging on FVP:
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000473
Paul Beesleyfd688342019-01-21 16:11:28 +0000474``make PLAT=fvp LOG_LEVEL=50 all``
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000475
476Use const data where possible
Paul Beesleyfd688342019-01-21 16:11:28 +0000477^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000478
479For example, the following code:
480
481.. code:: c
482
483 struct my_struct {
484 int arg1;
485 int arg2;
486 };
487
488 void init(struct my_struct *ptr);
489
490 void main(void)
491 {
492 struct my_struct x;
493 x.arg1 = 1;
494 x.arg2 = 2;
495 init(&x);
496 }
497
498is better written as:
499
500.. code:: c
501
502 struct my_struct {
503 int arg1;
504 int arg2;
505 };
506
507 void init(const struct my_struct *ptr);
508
509 void main(void)
510 {
511 const struct my_struct x = { 1, 2 };
512 init(&x);
513 }
514
515This allows the linker to put the data in a read-only data section instead of a
516writeable data section, which may result in a smaller and faster binary. Note
517that this may require dependent functions (``init()`` in the above example) to
518have ``const`` arguments, assuming they don't need to modify the data.
519
Paul Beesleyfd688342019-01-21 16:11:28 +0000520Library and driver code
521-----------------------
522
523TF library code (under ``lib/`` and ``include/lib``) is any code that provides a
524reusable interface to other code, potentially even to code outside of TF.
525
526In some systems drivers must conform to a specific driver framework to provide
527services to the rest of the system. TF has no driver framework and the
528distinction between a driver and library is somewhat subjective.
529
530A driver (under ``drivers/`` and ``include/drivers/``) is defined as code that
531interfaces with hardware via a memory mapped interface.
532
533Some drivers (for example, the Arm CCI driver in ``include/drivers/arm/cci.h``)
534provide a general purpose API to that specific hardware. Other drivers (for
535example, the Arm PL011 console driver in ``drivers/arm/pl011/pl011_console.S``)
536provide a specific hardware implementation of a more abstract library API. In
537the latter case there may potentially be multiple drivers for the same hardware
538device.
539
540Neither libraries nor drivers should depend on platform-specific code. If they
541require platform-specific data (for example, a base address) to operate then
542they should provide an initialization function that takes the platform-specific
543data as arguments.
544
545TF common code (under ``common/`` and ``include/common/``) is code that is re-used
546by other generic (non-platform-specific) TF code. It is effectively internal
547library code.
548
Paul Beesley4e2e1b22019-01-17 15:44:37 +0000549.. _`Why the “volatile” type class should not be used`: https://www.kernel.org/doc/html/latest/process/volatile-considered-harmful.html
Paul Beesley54c83cf2019-01-21 11:57:42 +0000550.. _`Procedure Call Standard for the Arm Architecture`: http://infocenter.arm.com/help/topic/com.arm.doc.ihi0042f/IHI0042F_aapcs.pdf
551.. _`Procedure Call Standard for the Arm 64-bit Architecture`: http://infocenter.arm.com/help/topic/com.arm.doc.ihi0055b/IHI0055B_aapcs64.pdf