1 =============================
2 No-MMU memory mapping support
3 =============================
4
5 The kernel has limited support for memory mapp
6 as are used in uClinux environments. From the
7 mapping is made use of in conjunction with the
8 call and the execve() system call. From the ke
9 mapping is actually performed by the binfmt dr
10 mmap() routines to do the actual work.
11
12 Memory mapping behaviour also involves the way
13 ptrace() work. Under uClinux there is no fork(
14 the CLONE_VM flag.
15
16 The behaviour is similar between the MMU and n
17 and it's also much more restricted in the latt
18
19 (#) Anonymous mapping, MAP_PRIVATE
20
21 In the MMU case: VM regions backed by
22 across fork.
23
24 In the no-MMU case: VM regions backed
25 pages.
26
27 (#) Anonymous mapping, MAP_SHARED
28
29 These behave very much like private ma
30 shared across fork() or clone() withou
31 the no-MMU case doesn't support these,
32 MAP_PRIVATE there.
33
34 (#) File, MAP_PRIVATE, PROT_READ / PROT_EXEC,
35
36 In the MMU case: VM regions backed by
37 the underlying file are reflected in t
38
39 In the no-MMU case:
40
41 - If one exists, the kernel will re-u
42 same segment of the same file if th
43 even if this was created by another
44
45 - If possible, the file mapping will
46 if the backing device has the NOMMU
47 appropriate mapping protection capa
48 and mtd might all permit this.
49
50 - If the backing device can't or won'
51 but does have the NOMMU_MAP_COPY ca
52 appropriate bit of the file will be
53 memory and any extraneous space bey
54
55 - Writes to the file do not affect th
56 are visible in other processes (no
57 happen.
58
59 (#) File, MAP_PRIVATE, PROT_READ / PROT_EXEC,
60
61 In the MMU case: like the non-PROT_WRI
62 question get copied before the write a
63 on writes to the file underneath that
64 the mapping's backing pages. The page
65
66 In the no-MMU case: works much like th
67 that a copy is always taken and never
68
69 (#) Regular file / blockdev, MAP_SHARED, PROT
70
71 In the MMU case: VM regions backed by
72 pages written back to file; writes to
73 mapping; shared across fork.
74
75 In the no-MMU case: not supported.
76
77 (#) Memory backed regular file, MAP_SHARED, P
78
79 In the MMU case: As for ordinary regul
80
81 In the no-MMU case: The filesystem pro
82 (such as ramfs or tmpfs) may choose to
83 sequence by providing a contiguous seq
84 case, a shared-writable memory mapping
85 as for the MMU case. If the filesystem
86 support, then the mapping request will
87
88 (#) Memory backed blockdev, MAP_SHARED, PROT_
89
90 In the MMU case: As for ordinary regul
91
92 In the no-MMU case: As for memory back
93 blockdev must be able to provide a con
94 truncate being called. The ramdisk dri
95 all its memory as a contiguous array u
96
97 (#) Memory backed chardev, MAP_SHARED, PROT_R
98
99 In the MMU case: As for ordinary regul
100
101 In the no-MMU case: The character devi
102 the mmap() by providing direct access
103 provides memory or quasi-memory that c
104 of such are frame buffers and flash de
105 provide any such support, then the map
106
107
108 Further notes on no-MMU MMAP
109 ============================
110
111 (#) A request for a private mapping of a file
112 page-aligned. This is because XIP may ta
113 paged aligned in the backing store.
114
115 (#) A request for an anonymous mapping will a
116 possible the size of the request should b
117 of the space may be wasted as the kernel
118 granule but will only discard the excess
119 this has an effect on fragmentation.
120
121 (#) The memory allocated by a request for an
122 be cleared by the kernel before being ret
123 Linux man pages (ver 2.22 or later).
124
125 In the MMU case this can be achieved with
126 regions are backed by virtual pages, with
127 to cleared physical pages when a write ha
128 (prior to which, the pages are effectivel
129 from which reads can take place). This s
130 initialize the contents of a page - depen
131 mapping.
132
133 In the no-MMU case, however, anonymous ma
134 pages, and the entire map is cleared at a
135 significant delays during a userspace mal
136 anonymous mapping and the kernel then doe
137
138 However, for memory that isn't required t
139 returned by malloc() - mmap() can take a
140 indicate to the kernel that it shouldn't
141 returning it. Note that CONFIG_MMAP_ALLO
142 to permit this, otherwise the flag will b
143
144 uClibc uses this to speed up malloc(), an
145 to allocate the brk and stack region.
146
147 (#) A list of all the private copy and anonym
148 visible through /proc/maps in no-MMU mode
149
150 (#) A list of all the mappings in use by a pr
151 /proc/<pid>/maps in no-MMU mode.
152
153 (#) Supplying MAP_FIXED or a requesting a par
154 result in an error.
155
156 (#) Files mapped privately usually have to ha
157 driver or filesystem so that the contents
158 allocated if mmap() chooses not to map th
159 error will result if they don't. This is
160 with character device files, pipes, fifos
161
162
163 Interprocess shared memory
164 ==========================
165
166 Both SYSV IPC SHM shared memory and POSIX shar
167 mode. The former through the usual mechanism,
168 on ramfs or tmpfs mounts.
169
170
171 Futexes
172 =======
173
174 Futexes are supported in NOMMU mode if the arc
175 be given if an address passed to the futex sys
176 mappings made by a process or if the mapping i
177 support futexes (such as an I/O chardev mappin
178
179
180 No-MMU mremap
181 =============
182
183 The mremap() function is partially supported.
184 mapping, and may move it [#]_ if MREMAP_MAYMOV
185 of the mapping exceeds the size of the slab ob
186 memory to which the mapping refers, or if a sm
187
188 MREMAP_FIXED is not supported, though it is ig
189 address and the object does not need to be mov
190
191 Shared mappings may not be moved. Shareable m
192 even if they are not currently shared.
193
194 The mremap() function must be given an exact m
195 a previously mapped object. It may not be use
196 mappings, move parts of existing mappings or r
197 act on a complete mapping.
198
199 .. [#] Not currently supported.
200
201
202 Providing shareable character device support
203 ============================================
204
205 To provide shareable character device support,
206 file->f_op->get_unmapped_area() operation. The
207 to get a proposed address for the mapping. Thi
208 doesn't wish to honour the mapping because it'
209 under some unsupported combination of flags or
210
211 The driver should also provide backing device
212 to indicate the permitted types of mapping on
213 assumed to be readable and writable, not execu
214 directly (can't be copied).
215
216 The file->f_op->mmap() operation will be calle
217 mapping. It can be rejected at that point. Ret
218 cause the mapping to be copied instead if NOMM
219
220 The vm_ops->close() routine will be invoked wh
221 is removed. An existing mapping will be shared
222 without notifying the driver.
223
224 It is permitted also for the file->f_op->get_u
225 return -ENOSYS. This will be taken to mean tha
226 want to handle it, despite the fact it's got a
227 might try directing the call to a secondary dr
228 implement it. Such is the case for the framebu
229 direct the call to the device-specific driver.
230 mapping request will be rejected if NOMMU_MAP_
231 copy mapped otherwise.
232
233 .. important::
234
235 Some types of device may present a dif
236 looking at them in certain modes. Flas
237 instance if they're in programming or
238 status reflected in the mapping, inste
239
240 In such a case, care must be taken les
241 private mapping showing such informati
242 controlling the device. Remember espec
243 mappings may still be mapped directly
244 circumstances!
245
246
247 Providing shareable memory-backed file support
248 ==============================================
249
250 Provision of shared mappings on memory backed
251 of support for shared mapped character devices
252 filesystem providing the service will probably
253 of pages and permit mappings to be made on tha
254
255 It is recommended that a truncate operation ap
256 increases the file size, if that file is empty
257 enough pages to honour a mapping. This is requ
258 memory.
259
260 Memory backed devices are indicated by the map
261 the memory_backed flag set.
262
263
264 Providing shareable block device support
265 ========================================
266
267 Provision of shared mappings on block device f
268 character devices. If there isn't a real devic
269 should allocate sufficient contiguous memory t
270
271
272 Adjusting page trimming behaviour
273 =================================
274
275 NOMMU mmap automatically rounds up to the near
276 when performing an allocation. This can have
277 fragmentation, and as such, is left configurab
278 aggressively trim allocations and discard any
279 allocator. In order to retain finer-grained c
280 behaviour can either be disabled completely, o
281 watermark where trimming begins.
282
283 Page trimming behaviour is configurable via th
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.