1 #ifndef BSWAP_H
2 #define BSWAP_H
3
4 #undef bswap16
5 #define bswap16(_x) __builtin_bswap16(_x)
6 #undef bswap32
7 #define bswap32(_x) __builtin_bswap32(_x)
8 #undef bswap64
9 #define bswap64(_x) __builtin_bswap64(_x)
10
bswap24(uint32_t x)11 static inline uint32_t bswap24(uint32_t x)
12 {
13 return (((x & 0x000000ffU) << 16) |
14 ((x & 0x0000ff00U) << 0) |
15 ((x & 0x00ff0000U) >> 16));
16 }
17
bswap16s(uint16_t * s)18 static inline void bswap16s(uint16_t *s)
19 {
20 *s = __builtin_bswap16(*s);
21 }
22
bswap24s(uint32_t * s)23 static inline void bswap24s(uint32_t *s)
24 {
25 *s = bswap24(*s & 0x00ffffffU);
26 }
27
bswap32s(uint32_t * s)28 static inline void bswap32s(uint32_t *s)
29 {
30 *s = __builtin_bswap32(*s);
31 }
32
bswap64s(uint64_t * s)33 static inline void bswap64s(uint64_t *s)
34 {
35 *s = __builtin_bswap64(*s);
36 }
37
38 #if HOST_BIG_ENDIAN
39 #define be_bswap(v, size) (v)
40 #define le_bswap(v, size) glue(__builtin_bswap, size)(v)
41 #define be_bswap24(v) (v)
42 #define le_bswap24(v) bswap24(v)
43 #define be_bswaps(v, size)
44 #define le_bswaps(p, size) \
45 do { *p = glue(__builtin_bswap, size)(*p); } while (0)
46 #else
47 #define le_bswap(v, size) (v)
48 #define be_bswap24(v) bswap24(v)
49 #define le_bswap24(v) (v)
50 #define be_bswap(v, size) glue(__builtin_bswap, size)(v)
51 #define le_bswaps(v, size)
52 #define be_bswaps(p, size) \
53 do { *p = glue(__builtin_bswap, size)(*p); } while (0)
54 #endif
55
56 /**
57 * Endianness conversion functions between host cpu and specified endianness.
58 * (We list the complete set of prototypes produced by the macros below
59 * to assist people who search the headers to find their definitions.)
60 *
61 * uint16_t le16_to_cpu(uint16_t v);
62 * uint32_t le32_to_cpu(uint32_t v);
63 * uint64_t le64_to_cpu(uint64_t v);
64 * uint16_t be16_to_cpu(uint16_t v);
65 * uint32_t be32_to_cpu(uint32_t v);
66 * uint64_t be64_to_cpu(uint64_t v);
67 *
68 * Convert the value @v from the specified format to the native
69 * endianness of the host CPU by byteswapping if necessary, and
70 * return the converted value.
71 *
72 * uint16_t cpu_to_le16(uint16_t v);
73 * uint32_t cpu_to_le32(uint32_t v);
74 * uint64_t cpu_to_le64(uint64_t v);
75 * uint16_t cpu_to_be16(uint16_t v);
76 * uint32_t cpu_to_be32(uint32_t v);
77 * uint64_t cpu_to_be64(uint64_t v);
78 *
79 * Convert the value @v from the native endianness of the host CPU to
80 * the specified format by byteswapping if necessary, and return
81 * the converted value.
82 *
83 * void le16_to_cpus(uint16_t *v);
84 * void le32_to_cpus(uint32_t *v);
85 * void le64_to_cpus(uint64_t *v);
86 * void be16_to_cpus(uint16_t *v);
87 * void be32_to_cpus(uint32_t *v);
88 * void be64_to_cpus(uint64_t *v);
89 *
90 * Do an in-place conversion of the value pointed to by @v from the
91 * specified format to the native endianness of the host CPU.
92 *
93 * void cpu_to_le16s(uint16_t *v);
94 * void cpu_to_le32s(uint32_t *v);
95 * void cpu_to_le64s(uint64_t *v);
96 * void cpu_to_be16s(uint16_t *v);
97 * void cpu_to_be32s(uint32_t *v);
98 * void cpu_to_be64s(uint64_t *v);
99 *
100 * Do an in-place conversion of the value pointed to by @v from the
101 * native endianness of the host CPU to the specified format.
102 *
103 * Both X_to_cpu() and cpu_to_X() perform the same operation; you
104 * should use whichever one is better documenting of the function your
105 * code is performing.
106 *
107 * Do not use these functions for conversion of values which are in guest
108 * memory, since the data may not be sufficiently aligned for the host CPU's
109 * load and store instructions. Instead you should use the ld*_p() and
110 * st*_p() functions, which perform loads and stores of data of any
111 * required size and endianness and handle possible misalignment.
112 */
113
114 #define CPU_CONVERT(endian, size, type)\
115 static inline type endian ## size ## _to_cpu(type v)\
116 {\
117 return glue(endian, _bswap)(v, size);\
118 }\
119 \
120 static inline type cpu_to_ ## endian ## size(type v)\
121 {\
122 return glue(endian, _bswap)(v, size);\
123 }\
124 \
125 static inline void endian ## size ## _to_cpus(type *p)\
126 {\
127 glue(endian, _bswaps)(p, size);\
128 }\
129 \
130 static inline void cpu_to_ ## endian ## size ## s(type *p)\
131 {\
132 glue(endian, _bswaps)(p, size);\
133 }
134
135 CPU_CONVERT(be, 16, uint16_t)
136 CPU_CONVERT(be, 32, uint32_t)
137 CPU_CONVERT(be, 64, uint64_t)
138
139 CPU_CONVERT(le, 16, uint16_t)
140 CPU_CONVERT(le, 32, uint32_t)
141 CPU_CONVERT(le, 64, uint64_t)
142
143 #undef CPU_CONVERT
144
145 /*
146 * Same as cpu_to_le{16,32,64}, except that gcc will figure the result is
147 * a compile-time constant if you pass in a constant. So this can be
148 * used to initialize static variables.
149 */
150 #if HOST_BIG_ENDIAN
151 # define const_le64(_x) \
152 ((((_x) & 0x00000000000000ffULL) << 56) | \
153 (((_x) & 0x000000000000ff00ULL) << 40) | \
154 (((_x) & 0x0000000000ff0000ULL) << 24) | \
155 (((_x) & 0x00000000ff000000ULL) << 8) | \
156 (((_x) & 0x000000ff00000000ULL) >> 8) | \
157 (((_x) & 0x0000ff0000000000ULL) >> 24) | \
158 (((_x) & 0x00ff000000000000ULL) >> 40) | \
159 (((_x) & 0xff00000000000000ULL) >> 56))
160 # define const_le32(_x) \
161 ((((_x) & 0x000000ffU) << 24) | \
162 (((_x) & 0x0000ff00U) << 8) | \
163 (((_x) & 0x00ff0000U) >> 8) | \
164 (((_x) & 0xff000000U) >> 24))
165 # define const_le16(_x) \
166 ((((_x) & 0x00ff) << 8) | \
167 (((_x) & 0xff00) >> 8))
168 #else
169 # define const_le64(_x) (_x)
170 # define const_le32(_x) (_x)
171 # define const_le16(_x) (_x)
172 #endif
173
174 /* unaligned/endian-independent pointer access */
175
176 /*
177 * the generic syntax is:
178 *
179 * load: ld{type}{sign}{size}_{endian}_p(ptr)
180 *
181 * store: st{type}{size}_{endian}_p(ptr, val)
182 *
183 * Note there are small differences with the softmmu access API!
184 *
185 * type is:
186 * (empty): integer access
187 * f : float access
188 *
189 * sign is:
190 * (empty): for 32 or 64 bit sizes (including floats and doubles)
191 * u : unsigned
192 * s : signed
193 *
194 * size is:
195 * b: 8 bits
196 * w: 16 bits
197 * 24: 24 bits
198 * l: 32 bits
199 * q: 64 bits
200 *
201 * endian is:
202 * he : host endian
203 * be : big endian
204 * le : little endian
205 * te : target endian
206 * (except for byte accesses, which have no endian infix).
207 *
208 * In all cases these functions take a host pointer.
209 * For accessors that take a guest address rather than a
210 * host address, see the cpu_{ld,st}_* accessors defined in
211 * cpu_ldst.h.
212 *
213 * For cases where the size to be used is not fixed at compile time,
214 * there are
215 * stn_{endian}_p(ptr, sz, val)
216 * which stores @val to @ptr as an @endian-order number @sz bytes in size
217 * and
218 * ldn_{endian}_p(ptr, sz)
219 * which loads @sz bytes from @ptr as an unsigned @endian-order number
220 * and returns it in a uint64_t.
221 */
222
ldub_p(const void * ptr)223 static inline int ldub_p(const void *ptr)
224 {
225 return *(uint8_t *)ptr;
226 }
227
ldsb_p(const void * ptr)228 static inline int ldsb_p(const void *ptr)
229 {
230 return *(int8_t *)ptr;
231 }
232
stb_p(void * ptr,uint8_t v)233 static inline void stb_p(void *ptr, uint8_t v)
234 {
235 *(uint8_t *)ptr = v;
236 }
237
238 /*
239 * Any compiler worth its salt will turn these memcpy into native unaligned
240 * operations. Thus we don't need to play games with packed attributes, or
241 * inline byte-by-byte stores.
242 * Some compilation environments (eg some fortify-source implementations)
243 * may intercept memcpy() in a way that defeats the compiler optimization,
244 * though, so we use __builtin_memcpy() to give ourselves the best chance
245 * of good performance.
246 */
247
lduw_he_p(const void * ptr)248 static inline int lduw_he_p(const void *ptr)
249 {
250 uint16_t r;
251 __builtin_memcpy(&r, ptr, sizeof(r));
252 return r;
253 }
254
ldsw_he_p(const void * ptr)255 static inline int ldsw_he_p(const void *ptr)
256 {
257 int16_t r;
258 __builtin_memcpy(&r, ptr, sizeof(r));
259 return r;
260 }
261
stw_he_p(void * ptr,uint16_t v)262 static inline void stw_he_p(void *ptr, uint16_t v)
263 {
264 __builtin_memcpy(ptr, &v, sizeof(v));
265 }
266
st24_he_p(void * ptr,uint32_t v)267 static inline void st24_he_p(void *ptr, uint32_t v)
268 {
269 __builtin_memcpy(ptr, &v, 3);
270 }
271
ldl_he_p(const void * ptr)272 static inline int ldl_he_p(const void *ptr)
273 {
274 int32_t r;
275 __builtin_memcpy(&r, ptr, sizeof(r));
276 return r;
277 }
278
stl_he_p(void * ptr,uint32_t v)279 static inline void stl_he_p(void *ptr, uint32_t v)
280 {
281 __builtin_memcpy(ptr, &v, sizeof(v));
282 }
283
ldq_he_p(const void * ptr)284 static inline uint64_t ldq_he_p(const void *ptr)
285 {
286 uint64_t r;
287 __builtin_memcpy(&r, ptr, sizeof(r));
288 return r;
289 }
290
stq_he_p(void * ptr,uint64_t v)291 static inline void stq_he_p(void *ptr, uint64_t v)
292 {
293 __builtin_memcpy(ptr, &v, sizeof(v));
294 }
295
lduw_le_p(const void * ptr)296 static inline int lduw_le_p(const void *ptr)
297 {
298 return (uint16_t)le_bswap(lduw_he_p(ptr), 16);
299 }
300
ldsw_le_p(const void * ptr)301 static inline int ldsw_le_p(const void *ptr)
302 {
303 return (int16_t)le_bswap(lduw_he_p(ptr), 16);
304 }
305
ldl_le_p(const void * ptr)306 static inline int ldl_le_p(const void *ptr)
307 {
308 return le_bswap(ldl_he_p(ptr), 32);
309 }
310
ldq_le_p(const void * ptr)311 static inline uint64_t ldq_le_p(const void *ptr)
312 {
313 return le_bswap(ldq_he_p(ptr), 64);
314 }
315
stw_le_p(void * ptr,uint16_t v)316 static inline void stw_le_p(void *ptr, uint16_t v)
317 {
318 stw_he_p(ptr, le_bswap(v, 16));
319 }
320
st24_le_p(void * ptr,uint32_t v)321 static inline void st24_le_p(void *ptr, uint32_t v)
322 {
323 st24_he_p(ptr, le_bswap24(v));
324 }
325
stl_le_p(void * ptr,uint32_t v)326 static inline void stl_le_p(void *ptr, uint32_t v)
327 {
328 stl_he_p(ptr, le_bswap(v, 32));
329 }
330
stq_le_p(void * ptr,uint64_t v)331 static inline void stq_le_p(void *ptr, uint64_t v)
332 {
333 stq_he_p(ptr, le_bswap(v, 64));
334 }
335
lduw_be_p(const void * ptr)336 static inline int lduw_be_p(const void *ptr)
337 {
338 return (uint16_t)be_bswap(lduw_he_p(ptr), 16);
339 }
340
ldsw_be_p(const void * ptr)341 static inline int ldsw_be_p(const void *ptr)
342 {
343 return (int16_t)be_bswap(lduw_he_p(ptr), 16);
344 }
345
ldl_be_p(const void * ptr)346 static inline int ldl_be_p(const void *ptr)
347 {
348 return be_bswap(ldl_he_p(ptr), 32);
349 }
350
ldq_be_p(const void * ptr)351 static inline uint64_t ldq_be_p(const void *ptr)
352 {
353 return be_bswap(ldq_he_p(ptr), 64);
354 }
355
stw_be_p(void * ptr,uint16_t v)356 static inline void stw_be_p(void *ptr, uint16_t v)
357 {
358 stw_he_p(ptr, be_bswap(v, 16));
359 }
360
st24_be_p(void * ptr,uint32_t v)361 static inline void st24_be_p(void *ptr, uint32_t v)
362 {
363 st24_he_p(ptr, be_bswap24(v));
364 }
365
stl_be_p(void * ptr,uint32_t v)366 static inline void stl_be_p(void *ptr, uint32_t v)
367 {
368 stl_he_p(ptr, be_bswap(v, 32));
369 }
370
stq_be_p(void * ptr,uint64_t v)371 static inline void stq_be_p(void *ptr, uint64_t v)
372 {
373 stq_he_p(ptr, be_bswap(v, 64));
374 }
375
leul_to_cpu(unsigned long v)376 static inline unsigned long leul_to_cpu(unsigned long v)
377 {
378 #if HOST_LONG_BITS == 32
379 return le_bswap(v, 32);
380 #elif HOST_LONG_BITS == 64
381 return le_bswap(v, 64);
382 #else
383 # error Unknown sizeof long
384 #endif
385 }
386
387 /* Store v to p as a sz byte value in host order */
388 #define DO_STN_LDN_P(END) \
389 static inline void stn_## END ## _p(void *ptr, int sz, uint64_t v) \
390 { \
391 switch (sz) { \
392 case 1: \
393 stb_p(ptr, v); \
394 break; \
395 case 2: \
396 stw_ ## END ## _p(ptr, v); \
397 break; \
398 case 4: \
399 stl_ ## END ## _p(ptr, v); \
400 break; \
401 case 8: \
402 stq_ ## END ## _p(ptr, v); \
403 break; \
404 default: \
405 g_assert_not_reached(); \
406 } \
407 } \
408 static inline uint64_t ldn_## END ## _p(const void *ptr, int sz) \
409 { \
410 switch (sz) { \
411 case 1: \
412 return ldub_p(ptr); \
413 case 2: \
414 return lduw_ ## END ## _p(ptr); \
415 case 4: \
416 return (uint32_t)ldl_ ## END ## _p(ptr); \
417 case 8: \
418 return ldq_ ## END ## _p(ptr); \
419 default: \
420 g_assert_not_reached(); \
421 } \
422 }
423
424 DO_STN_LDN_P(he)
425 DO_STN_LDN_P(le)
426 DO_STN_LDN_P(be)
427
428 #undef DO_STN_LDN_P
429
430 #undef le_bswap
431 #undef be_bswap
432 #undef le_bswaps
433 #undef be_bswaps
434
435 #endif /* BSWAP_H */
436