Merge tag 'block-5.11-2021-01-16' of git://git.kernel.dk/linux-block
[linux/fpc-iii.git] / lib / lzo / lzo1x_compress.c
blob76758e9296ba65221fee14ec2884e404a3bf06da
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * LZO1X Compressor from LZO
5 * Copyright (C) 1996-2012 Markus F.X.J. Oberhumer <markus@oberhumer.com>
7 * The full LZO package can be found at:
8 * http://www.oberhumer.com/opensource/lzo/
10 * Changed for Linux kernel use by:
11 * Nitin Gupta <nitingupta910@gmail.com>
12 * Richard Purdie <rpurdie@openedhand.com>
15 #include <linux/module.h>
16 #include <linux/kernel.h>
17 #include <asm/unaligned.h>
18 #include <linux/lzo.h>
19 #include "lzodefs.h"
21 static noinline size_t
22 lzo1x_1_do_compress(const unsigned char *in, size_t in_len,
23 unsigned char *out, size_t *out_len,
24 size_t ti, void *wrkmem, signed char *state_offset,
25 const unsigned char bitstream_version)
27 const unsigned char *ip;
28 unsigned char *op;
29 const unsigned char * const in_end = in + in_len;
30 const unsigned char * const ip_end = in + in_len - 20;
31 const unsigned char *ii;
32 lzo_dict_t * const dict = (lzo_dict_t *) wrkmem;
34 op = out;
35 ip = in;
36 ii = ip;
37 ip += ti < 4 ? 4 - ti : 0;
39 for (;;) {
40 const unsigned char *m_pos = NULL;
41 size_t t, m_len, m_off;
42 u32 dv;
43 u32 run_length = 0;
44 literal:
45 ip += 1 + ((ip - ii) >> 5);
46 next:
47 if (unlikely(ip >= ip_end))
48 break;
49 dv = get_unaligned_le32(ip);
51 if (dv == 0 && bitstream_version) {
52 const unsigned char *ir = ip + 4;
53 const unsigned char *limit = ip_end
54 < (ip + MAX_ZERO_RUN_LENGTH + 1)
55 ? ip_end : ip + MAX_ZERO_RUN_LENGTH + 1;
56 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && \
57 defined(LZO_FAST_64BIT_MEMORY_ACCESS)
58 u64 dv64;
60 for (; (ir + 32) <= limit; ir += 32) {
61 dv64 = get_unaligned((u64 *)ir);
62 dv64 |= get_unaligned((u64 *)ir + 1);
63 dv64 |= get_unaligned((u64 *)ir + 2);
64 dv64 |= get_unaligned((u64 *)ir + 3);
65 if (dv64)
66 break;
68 for (; (ir + 8) <= limit; ir += 8) {
69 dv64 = get_unaligned((u64 *)ir);
70 if (dv64) {
71 # if defined(__LITTLE_ENDIAN)
72 ir += __builtin_ctzll(dv64) >> 3;
73 # elif defined(__BIG_ENDIAN)
74 ir += __builtin_clzll(dv64) >> 3;
75 # else
76 # error "missing endian definition"
77 # endif
78 break;
81 #else
82 while ((ir < (const unsigned char *)
83 ALIGN((uintptr_t)ir, 4)) &&
84 (ir < limit) && (*ir == 0))
85 ir++;
86 if (IS_ALIGNED((uintptr_t)ir, 4)) {
87 for (; (ir + 4) <= limit; ir += 4) {
88 dv = *((u32 *)ir);
89 if (dv) {
90 # if defined(__LITTLE_ENDIAN)
91 ir += __builtin_ctz(dv) >> 3;
92 # elif defined(__BIG_ENDIAN)
93 ir += __builtin_clz(dv) >> 3;
94 # else
95 # error "missing endian definition"
96 # endif
97 break;
101 #endif
102 while (likely(ir < limit) && unlikely(*ir == 0))
103 ir++;
104 run_length = ir - ip;
105 if (run_length > MAX_ZERO_RUN_LENGTH)
106 run_length = MAX_ZERO_RUN_LENGTH;
107 } else {
108 t = ((dv * 0x1824429d) >> (32 - D_BITS)) & D_MASK;
109 m_pos = in + dict[t];
110 dict[t] = (lzo_dict_t) (ip - in);
111 if (unlikely(dv != get_unaligned_le32(m_pos)))
112 goto literal;
115 ii -= ti;
116 ti = 0;
117 t = ip - ii;
118 if (t != 0) {
119 if (t <= 3) {
120 op[*state_offset] |= t;
121 COPY4(op, ii);
122 op += t;
123 } else if (t <= 16) {
124 *op++ = (t - 3);
125 COPY8(op, ii);
126 COPY8(op + 8, ii + 8);
127 op += t;
128 } else {
129 if (t <= 18) {
130 *op++ = (t - 3);
131 } else {
132 size_t tt = t - 18;
133 *op++ = 0;
134 while (unlikely(tt > 255)) {
135 tt -= 255;
136 *op++ = 0;
138 *op++ = tt;
140 do {
141 COPY8(op, ii);
142 COPY8(op + 8, ii + 8);
143 op += 16;
144 ii += 16;
145 t -= 16;
146 } while (t >= 16);
147 if (t > 0) do {
148 *op++ = *ii++;
149 } while (--t > 0);
153 if (unlikely(run_length)) {
154 ip += run_length;
155 run_length -= MIN_ZERO_RUN_LENGTH;
156 put_unaligned_le32((run_length << 21) | 0xfffc18
157 | (run_length & 0x7), op);
158 op += 4;
159 run_length = 0;
160 *state_offset = -3;
161 goto finished_writing_instruction;
164 m_len = 4;
166 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && defined(LZO_USE_CTZ64)
167 u64 v;
168 v = get_unaligned((const u64 *) (ip + m_len)) ^
169 get_unaligned((const u64 *) (m_pos + m_len));
170 if (unlikely(v == 0)) {
171 do {
172 m_len += 8;
173 v = get_unaligned((const u64 *) (ip + m_len)) ^
174 get_unaligned((const u64 *) (m_pos + m_len));
175 if (unlikely(ip + m_len >= ip_end))
176 goto m_len_done;
177 } while (v == 0);
179 # if defined(__LITTLE_ENDIAN)
180 m_len += (unsigned) __builtin_ctzll(v) / 8;
181 # elif defined(__BIG_ENDIAN)
182 m_len += (unsigned) __builtin_clzll(v) / 8;
183 # else
184 # error "missing endian definition"
185 # endif
186 #elif defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && defined(LZO_USE_CTZ32)
187 u32 v;
188 v = get_unaligned((const u32 *) (ip + m_len)) ^
189 get_unaligned((const u32 *) (m_pos + m_len));
190 if (unlikely(v == 0)) {
191 do {
192 m_len += 4;
193 v = get_unaligned((const u32 *) (ip + m_len)) ^
194 get_unaligned((const u32 *) (m_pos + m_len));
195 if (v != 0)
196 break;
197 m_len += 4;
198 v = get_unaligned((const u32 *) (ip + m_len)) ^
199 get_unaligned((const u32 *) (m_pos + m_len));
200 if (unlikely(ip + m_len >= ip_end))
201 goto m_len_done;
202 } while (v == 0);
204 # if defined(__LITTLE_ENDIAN)
205 m_len += (unsigned) __builtin_ctz(v) / 8;
206 # elif defined(__BIG_ENDIAN)
207 m_len += (unsigned) __builtin_clz(v) / 8;
208 # else
209 # error "missing endian definition"
210 # endif
211 #else
212 if (unlikely(ip[m_len] == m_pos[m_len])) {
213 do {
214 m_len += 1;
215 if (ip[m_len] != m_pos[m_len])
216 break;
217 m_len += 1;
218 if (ip[m_len] != m_pos[m_len])
219 break;
220 m_len += 1;
221 if (ip[m_len] != m_pos[m_len])
222 break;
223 m_len += 1;
224 if (ip[m_len] != m_pos[m_len])
225 break;
226 m_len += 1;
227 if (ip[m_len] != m_pos[m_len])
228 break;
229 m_len += 1;
230 if (ip[m_len] != m_pos[m_len])
231 break;
232 m_len += 1;
233 if (ip[m_len] != m_pos[m_len])
234 break;
235 m_len += 1;
236 if (unlikely(ip + m_len >= ip_end))
237 goto m_len_done;
238 } while (ip[m_len] == m_pos[m_len]);
240 #endif
242 m_len_done:
244 m_off = ip - m_pos;
245 ip += m_len;
246 if (m_len <= M2_MAX_LEN && m_off <= M2_MAX_OFFSET) {
247 m_off -= 1;
248 *op++ = (((m_len - 1) << 5) | ((m_off & 7) << 2));
249 *op++ = (m_off >> 3);
250 } else if (m_off <= M3_MAX_OFFSET) {
251 m_off -= 1;
252 if (m_len <= M3_MAX_LEN)
253 *op++ = (M3_MARKER | (m_len - 2));
254 else {
255 m_len -= M3_MAX_LEN;
256 *op++ = M3_MARKER | 0;
257 while (unlikely(m_len > 255)) {
258 m_len -= 255;
259 *op++ = 0;
261 *op++ = (m_len);
263 *op++ = (m_off << 2);
264 *op++ = (m_off >> 6);
265 } else {
266 m_off -= 0x4000;
267 if (m_len <= M4_MAX_LEN)
268 *op++ = (M4_MARKER | ((m_off >> 11) & 8)
269 | (m_len - 2));
270 else {
271 if (unlikely(((m_off & 0x403f) == 0x403f)
272 && (m_len >= 261)
273 && (m_len <= 264))
274 && likely(bitstream_version)) {
275 // Under lzo-rle, block copies
276 // for 261 <= length <= 264 and
277 // (distance & 0x80f3) == 0x80f3
278 // can result in ambiguous
279 // output. Adjust length
280 // to 260 to prevent ambiguity.
281 ip -= m_len - 260;
282 m_len = 260;
284 m_len -= M4_MAX_LEN;
285 *op++ = (M4_MARKER | ((m_off >> 11) & 8));
286 while (unlikely(m_len > 255)) {
287 m_len -= 255;
288 *op++ = 0;
290 *op++ = (m_len);
292 *op++ = (m_off << 2);
293 *op++ = (m_off >> 6);
295 *state_offset = -2;
296 finished_writing_instruction:
297 ii = ip;
298 goto next;
300 *out_len = op - out;
301 return in_end - (ii - ti);
304 static int lzogeneric1x_1_compress(const unsigned char *in, size_t in_len,
305 unsigned char *out, size_t *out_len,
306 void *wrkmem, const unsigned char bitstream_version)
308 const unsigned char *ip = in;
309 unsigned char *op = out;
310 unsigned char *data_start;
311 size_t l = in_len;
312 size_t t = 0;
313 signed char state_offset = -2;
314 unsigned int m4_max_offset;
316 // LZO v0 will never write 17 as first byte (except for zero-length
317 // input), so this is used to version the bitstream
318 if (bitstream_version > 0) {
319 *op++ = 17;
320 *op++ = bitstream_version;
321 m4_max_offset = M4_MAX_OFFSET_V1;
322 } else {
323 m4_max_offset = M4_MAX_OFFSET_V0;
326 data_start = op;
328 while (l > 20) {
329 size_t ll = l <= (m4_max_offset + 1) ? l : (m4_max_offset + 1);
330 uintptr_t ll_end = (uintptr_t) ip + ll;
331 if ((ll_end + ((t + ll) >> 5)) <= ll_end)
332 break;
333 BUILD_BUG_ON(D_SIZE * sizeof(lzo_dict_t) > LZO1X_1_MEM_COMPRESS);
334 memset(wrkmem, 0, D_SIZE * sizeof(lzo_dict_t));
335 t = lzo1x_1_do_compress(ip, ll, op, out_len, t, wrkmem,
336 &state_offset, bitstream_version);
337 ip += ll;
338 op += *out_len;
339 l -= ll;
341 t += l;
343 if (t > 0) {
344 const unsigned char *ii = in + in_len - t;
346 if (op == data_start && t <= 238) {
347 *op++ = (17 + t);
348 } else if (t <= 3) {
349 op[state_offset] |= t;
350 } else if (t <= 18) {
351 *op++ = (t - 3);
352 } else {
353 size_t tt = t - 18;
354 *op++ = 0;
355 while (tt > 255) {
356 tt -= 255;
357 *op++ = 0;
359 *op++ = tt;
361 if (t >= 16) do {
362 COPY8(op, ii);
363 COPY8(op + 8, ii + 8);
364 op += 16;
365 ii += 16;
366 t -= 16;
367 } while (t >= 16);
368 if (t > 0) do {
369 *op++ = *ii++;
370 } while (--t > 0);
373 *op++ = M4_MARKER | 1;
374 *op++ = 0;
375 *op++ = 0;
377 *out_len = op - out;
378 return LZO_E_OK;
381 int lzo1x_1_compress(const unsigned char *in, size_t in_len,
382 unsigned char *out, size_t *out_len,
383 void *wrkmem)
385 return lzogeneric1x_1_compress(in, in_len, out, out_len, wrkmem, 0);
388 int lzorle1x_1_compress(const unsigned char *in, size_t in_len,
389 unsigned char *out, size_t *out_len,
390 void *wrkmem)
392 return lzogeneric1x_1_compress(in, in_len, out, out_len,
393 wrkmem, LZO_VERSION);
396 EXPORT_SYMBOL_GPL(lzo1x_1_compress);
397 EXPORT_SYMBOL_GPL(lzorle1x_1_compress);
399 MODULE_LICENSE("GPL");
400 MODULE_DESCRIPTION("LZO1X-1 Compressor");