crypto: 842 - change 842 alg to use software
[deliverable/linux.git] / drivers / crypto / nx / nx-842.c
1 /*
2 * Driver for IBM Power 842 compression accelerator
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
17 *
18 * Copyright (C) IBM Corporation, 2012
19 *
20 * Authors: Robert Jennings <rcj@linux.vnet.ibm.com>
21 * Seth Jennings <sjenning@linux.vnet.ibm.com>
22 */
23
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/nx842.h>
27 #include <linux/of.h>
28 #include <linux/slab.h>
29
30 #include <asm/page.h>
31 #include <asm/vio.h>
32
33 #include "nx_csbcpb.h" /* struct nx_csbcpb */
34
35 #define MODULE_NAME "nx-compress"
36 MODULE_LICENSE("GPL");
37 MODULE_AUTHOR("Robert Jennings <rcj@linux.vnet.ibm.com>");
38 MODULE_DESCRIPTION("842 H/W Compression driver for IBM Power processors");
39
40 #define SHIFT_4K 12
41 #define SHIFT_64K 16
42 #define SIZE_4K (1UL << SHIFT_4K)
43 #define SIZE_64K (1UL << SHIFT_64K)
44
45 /* IO buffer must be 128 byte aligned */
46 #define IO_BUFFER_ALIGN 128
47
48 struct nx842_header {
49 int blocks_nr; /* number of compressed blocks */
50 int offset; /* offset of the first block (from beginning of header) */
51 int sizes[0]; /* size of compressed blocks */
52 };
53
54 static inline int nx842_header_size(const struct nx842_header *hdr)
55 {
56 return sizeof(struct nx842_header) +
57 hdr->blocks_nr * sizeof(hdr->sizes[0]);
58 }
59
60 /* Macros for fields within nx_csbcpb */
61 /* Check the valid bit within the csbcpb valid field */
62 #define NX842_CSBCBP_VALID_CHK(x) (x & BIT_MASK(7))
63
64 /* CE macros operate on the completion_extension field bits in the csbcpb.
65 * CE0 0=full completion, 1=partial completion
66 * CE1 0=CE0 indicates completion, 1=termination (output may be modified)
67 * CE2 0=processed_bytes is source bytes, 1=processed_bytes is target bytes */
68 #define NX842_CSBCPB_CE0(x) (x & BIT_MASK(7))
69 #define NX842_CSBCPB_CE1(x) (x & BIT_MASK(6))
70 #define NX842_CSBCPB_CE2(x) (x & BIT_MASK(5))
71
72 /* The NX unit accepts data only on 4K page boundaries */
73 #define NX842_HW_PAGE_SHIFT SHIFT_4K
74 #define NX842_HW_PAGE_SIZE (ASM_CONST(1) << NX842_HW_PAGE_SHIFT)
75 #define NX842_HW_PAGE_MASK (~(NX842_HW_PAGE_SIZE-1))
76
77 enum nx842_status {
78 UNAVAILABLE,
79 AVAILABLE
80 };
81
82 struct ibm_nx842_counters {
83 atomic64_t comp_complete;
84 atomic64_t comp_failed;
85 atomic64_t decomp_complete;
86 atomic64_t decomp_failed;
87 atomic64_t swdecomp;
88 atomic64_t comp_times[32];
89 atomic64_t decomp_times[32];
90 };
91
92 static struct nx842_devdata {
93 struct vio_dev *vdev;
94 struct device *dev;
95 struct ibm_nx842_counters *counters;
96 unsigned int max_sg_len;
97 unsigned int max_sync_size;
98 unsigned int max_sync_sg;
99 enum nx842_status status;
100 } __rcu *devdata;
101 static DEFINE_SPINLOCK(devdata_mutex);
102
103 #define NX842_COUNTER_INC(_x) \
104 static inline void nx842_inc_##_x( \
105 const struct nx842_devdata *dev) { \
106 if (dev) \
107 atomic64_inc(&dev->counters->_x); \
108 }
109 NX842_COUNTER_INC(comp_complete);
110 NX842_COUNTER_INC(comp_failed);
111 NX842_COUNTER_INC(decomp_complete);
112 NX842_COUNTER_INC(decomp_failed);
113 NX842_COUNTER_INC(swdecomp);
114
115 #define NX842_HIST_SLOTS 16
116
117 static void ibm_nx842_incr_hist(atomic64_t *times, unsigned int time)
118 {
119 int bucket = fls(time);
120
121 if (bucket)
122 bucket = min((NX842_HIST_SLOTS - 1), bucket - 1);
123
124 atomic64_inc(&times[bucket]);
125 }
126
127 /* NX unit operation flags */
128 #define NX842_OP_COMPRESS 0x0
129 #define NX842_OP_CRC 0x1
130 #define NX842_OP_DECOMPRESS 0x2
131 #define NX842_OP_COMPRESS_CRC (NX842_OP_COMPRESS | NX842_OP_CRC)
132 #define NX842_OP_DECOMPRESS_CRC (NX842_OP_DECOMPRESS | NX842_OP_CRC)
133 #define NX842_OP_ASYNC (1<<23)
134 #define NX842_OP_NOTIFY (1<<22)
135 #define NX842_OP_NOTIFY_INT(x) ((x & 0xff)<<8)
136
137 static unsigned long nx842_get_desired_dma(struct vio_dev *viodev)
138 {
139 /* No use of DMA mappings within the driver. */
140 return 0;
141 }
142
143 struct nx842_slentry {
144 unsigned long ptr; /* Real address (use __pa()) */
145 unsigned long len;
146 };
147
148 /* pHyp scatterlist entry */
149 struct nx842_scatterlist {
150 int entry_nr; /* number of slentries */
151 struct nx842_slentry *entries; /* ptr to array of slentries */
152 };
153
154 /* Does not include sizeof(entry_nr) in the size */
155 static inline unsigned long nx842_get_scatterlist_size(
156 struct nx842_scatterlist *sl)
157 {
158 return sl->entry_nr * sizeof(struct nx842_slentry);
159 }
160
161 static inline unsigned long nx842_get_pa(void *addr)
162 {
163 if (is_vmalloc_addr(addr))
164 return page_to_phys(vmalloc_to_page(addr))
165 + offset_in_page(addr);
166 else
167 return __pa(addr);
168 }
169
170 static int nx842_build_scatterlist(unsigned long buf, int len,
171 struct nx842_scatterlist *sl)
172 {
173 unsigned long nextpage;
174 struct nx842_slentry *entry;
175
176 sl->entry_nr = 0;
177
178 entry = sl->entries;
179 while (len) {
180 entry->ptr = nx842_get_pa((void *)buf);
181 nextpage = ALIGN(buf + 1, NX842_HW_PAGE_SIZE);
182 if (nextpage < buf + len) {
183 /* we aren't at the end yet */
184 if (IS_ALIGNED(buf, NX842_HW_PAGE_SIZE))
185 /* we are in the middle (or beginning) */
186 entry->len = NX842_HW_PAGE_SIZE;
187 else
188 /* we are at the beginning */
189 entry->len = nextpage - buf;
190 } else {
191 /* at the end */
192 entry->len = len;
193 }
194
195 len -= entry->len;
196 buf += entry->len;
197 sl->entry_nr++;
198 entry++;
199 }
200
201 return 0;
202 }
203
204 /*
205 * Working memory for software decompression
206 */
207 struct sw842_fifo {
208 union {
209 char f8[256][8];
210 char f4[512][4];
211 };
212 char f2[256][2];
213 unsigned char f84_full;
214 unsigned char f2_full;
215 unsigned char f8_count;
216 unsigned char f2_count;
217 unsigned int f4_count;
218 };
219
220 /*
221 * Working memory for crypto API
222 */
223 struct nx842_workmem {
224 char bounce[PAGE_SIZE]; /* bounce buffer for decompression input */
225 union {
226 /* hardware working memory */
227 struct {
228 /* scatterlist */
229 char slin[SIZE_4K];
230 char slout[SIZE_4K];
231 /* coprocessor status/parameter block */
232 struct nx_csbcpb csbcpb;
233 };
234 /* software working memory */
235 struct sw842_fifo swfifo; /* software decompression fifo */
236 };
237 };
238
239 int nx842_get_workmem_size(void)
240 {
241 return sizeof(struct nx842_workmem) + NX842_HW_PAGE_SIZE;
242 }
243 EXPORT_SYMBOL_GPL(nx842_get_workmem_size);
244
245 int nx842_get_workmem_size_aligned(void)
246 {
247 return sizeof(struct nx842_workmem);
248 }
249 EXPORT_SYMBOL_GPL(nx842_get_workmem_size_aligned);
250
251 static int nx842_validate_result(struct device *dev,
252 struct cop_status_block *csb)
253 {
254 /* The csb must be valid after returning from vio_h_cop_sync */
255 if (!NX842_CSBCBP_VALID_CHK(csb->valid)) {
256 dev_err(dev, "%s: cspcbp not valid upon completion.\n",
257 __func__);
258 dev_dbg(dev, "valid:0x%02x cs:0x%02x cc:0x%02x ce:0x%02x\n",
259 csb->valid,
260 csb->crb_seq_number,
261 csb->completion_code,
262 csb->completion_extension);
263 dev_dbg(dev, "processed_bytes:%d address:0x%016lx\n",
264 csb->processed_byte_count,
265 (unsigned long)csb->address);
266 return -EIO;
267 }
268
269 /* Check return values from the hardware in the CSB */
270 switch (csb->completion_code) {
271 case 0: /* Completed without error */
272 break;
273 case 64: /* Target bytes > Source bytes during compression */
274 case 13: /* Output buffer too small */
275 dev_dbg(dev, "%s: Compression output larger than input\n",
276 __func__);
277 return -ENOSPC;
278 case 66: /* Input data contains an illegal template field */
279 case 67: /* Template indicates data past the end of the input stream */
280 dev_dbg(dev, "%s: Bad data for decompression (code:%d)\n",
281 __func__, csb->completion_code);
282 return -EINVAL;
283 default:
284 dev_dbg(dev, "%s: Unspecified error (code:%d)\n",
285 __func__, csb->completion_code);
286 return -EIO;
287 }
288
289 /* Hardware sanity check */
290 if (!NX842_CSBCPB_CE2(csb->completion_extension)) {
291 dev_err(dev, "%s: No error returned by hardware, but "
292 "data returned is unusable, contact support.\n"
293 "(Additional info: csbcbp->processed bytes "
294 "does not specify processed bytes for the "
295 "target buffer.)\n", __func__);
296 return -EIO;
297 }
298
299 return 0;
300 }
301
302 /**
303 * nx842_compress - Compress data using the 842 algorithm
304 *
305 * Compression provide by the NX842 coprocessor on IBM Power systems.
306 * The input buffer is compressed and the result is stored in the
307 * provided output buffer.
308 *
309 * Upon return from this function @outlen contains the length of the
310 * compressed data. If there is an error then @outlen will be 0 and an
311 * error will be specified by the return code from this function.
312 *
313 * @in: Pointer to input buffer, must be page aligned
314 * @inlen: Length of input buffer, must be PAGE_SIZE
315 * @out: Pointer to output buffer
316 * @outlen: Length of output buffer
317 * @wrkmem: ptr to buffer for working memory, size determined by
318 * nx842_get_workmem_size()
319 *
320 * Returns:
321 * 0 Success, output of length @outlen stored in the buffer at @out
322 * -ENOMEM Unable to allocate internal buffers
323 * -ENOSPC Output buffer is to small
324 * -EMSGSIZE XXX Difficult to describe this limitation
325 * -EIO Internal error
326 * -ENODEV Hardware unavailable
327 */
328 int nx842_compress(const unsigned char *in, unsigned int inlen,
329 unsigned char *out, unsigned int *outlen, void *wmem)
330 {
331 struct nx842_header *hdr;
332 struct nx842_devdata *local_devdata;
333 struct device *dev = NULL;
334 struct nx842_workmem *workmem;
335 struct nx842_scatterlist slin, slout;
336 struct nx_csbcpb *csbcpb;
337 int ret = 0, max_sync_size, i, bytesleft, size, hdrsize;
338 unsigned long inbuf, outbuf, padding;
339 struct vio_pfo_op op = {
340 .done = NULL,
341 .handle = 0,
342 .timeout = 0,
343 };
344 unsigned long start_time = get_tb();
345
346 /*
347 * Make sure input buffer is 64k page aligned. This is assumed since
348 * this driver is designed for page compression only (for now). This
349 * is very nice since we can now use direct DDE(s) for the input and
350 * the alignment is guaranteed.
351 */
352 inbuf = (unsigned long)in;
353 if (!IS_ALIGNED(inbuf, PAGE_SIZE) || inlen != PAGE_SIZE)
354 return -EINVAL;
355
356 rcu_read_lock();
357 local_devdata = rcu_dereference(devdata);
358 if (!local_devdata || !local_devdata->dev) {
359 rcu_read_unlock();
360 return -ENODEV;
361 }
362 max_sync_size = local_devdata->max_sync_size;
363 dev = local_devdata->dev;
364
365 /* Create the header */
366 hdr = (struct nx842_header *)out;
367 hdr->blocks_nr = PAGE_SIZE / max_sync_size;
368 hdrsize = nx842_header_size(hdr);
369 outbuf = (unsigned long)out + hdrsize;
370 bytesleft = *outlen - hdrsize;
371
372 /* Init scatterlist */
373 workmem = (struct nx842_workmem *)ALIGN((unsigned long)wmem,
374 NX842_HW_PAGE_SIZE);
375 slin.entries = (struct nx842_slentry *)workmem->slin;
376 slout.entries = (struct nx842_slentry *)workmem->slout;
377
378 /* Init operation */
379 op.flags = NX842_OP_COMPRESS;
380 csbcpb = &workmem->csbcpb;
381 memset(csbcpb, 0, sizeof(*csbcpb));
382 op.csbcpb = nx842_get_pa(csbcpb);
383 op.out = nx842_get_pa(slout.entries);
384
385 for (i = 0; i < hdr->blocks_nr; i++) {
386 /*
387 * Aligning the output blocks to 128 bytes does waste space,
388 * but it prevents the need for bounce buffers and memory
389 * copies. It also simplifies the code a lot. In the worst
390 * case (64k page, 4k max_sync_size), you lose up to
391 * (128*16)/64k = ~3% the compression factor. For 64k
392 * max_sync_size, the loss would be at most 128/64k = ~0.2%.
393 */
394 padding = ALIGN(outbuf, IO_BUFFER_ALIGN) - outbuf;
395 outbuf += padding;
396 bytesleft -= padding;
397 if (i == 0)
398 /* save offset into first block in header */
399 hdr->offset = padding + hdrsize;
400
401 if (bytesleft <= 0) {
402 ret = -ENOSPC;
403 goto unlock;
404 }
405
406 /*
407 * NOTE: If the default max_sync_size is changed from 4k
408 * to 64k, remove the "likely" case below, since a
409 * scatterlist will always be needed.
410 */
411 if (likely(max_sync_size == NX842_HW_PAGE_SIZE)) {
412 /* Create direct DDE */
413 op.in = nx842_get_pa((void *)inbuf);
414 op.inlen = max_sync_size;
415
416 } else {
417 /* Create indirect DDE (scatterlist) */
418 nx842_build_scatterlist(inbuf, max_sync_size, &slin);
419 op.in = nx842_get_pa(slin.entries);
420 op.inlen = -nx842_get_scatterlist_size(&slin);
421 }
422
423 /*
424 * If max_sync_size != NX842_HW_PAGE_SIZE, an indirect
425 * DDE is required for the outbuf.
426 * If max_sync_size == NX842_HW_PAGE_SIZE, outbuf must
427 * also be page aligned (1 in 128/4k=32 chance) in order
428 * to use a direct DDE.
429 * This is unlikely, just use an indirect DDE always.
430 */
431 nx842_build_scatterlist(outbuf,
432 min(bytesleft, max_sync_size), &slout);
433 /* op.out set before loop */
434 op.outlen = -nx842_get_scatterlist_size(&slout);
435
436 /* Send request to pHyp */
437 ret = vio_h_cop_sync(local_devdata->vdev, &op);
438
439 /* Check for pHyp error */
440 if (ret) {
441 dev_dbg(dev, "%s: vio_h_cop_sync error (ret=%d, hret=%ld)\n",
442 __func__, ret, op.hcall_err);
443 ret = -EIO;
444 goto unlock;
445 }
446
447 /* Check for hardware error */
448 ret = nx842_validate_result(dev, &csbcpb->csb);
449 if (ret && ret != -ENOSPC)
450 goto unlock;
451
452 /* Handle incompressible data */
453 if (unlikely(ret == -ENOSPC)) {
454 if (bytesleft < max_sync_size) {
455 /*
456 * Not enough space left in the output buffer
457 * to store uncompressed block
458 */
459 goto unlock;
460 } else {
461 /* Store incompressible block */
462 memcpy((void *)outbuf, (void *)inbuf,
463 max_sync_size);
464 hdr->sizes[i] = -max_sync_size;
465 outbuf += max_sync_size;
466 bytesleft -= max_sync_size;
467 /* Reset ret, incompressible data handled */
468 ret = 0;
469 }
470 } else {
471 /* Normal case, compression was successful */
472 size = csbcpb->csb.processed_byte_count;
473 dev_dbg(dev, "%s: processed_bytes=%d\n",
474 __func__, size);
475 hdr->sizes[i] = size;
476 outbuf += size;
477 bytesleft -= size;
478 }
479
480 inbuf += max_sync_size;
481 }
482
483 *outlen = (unsigned int)(outbuf - (unsigned long)out);
484
485 unlock:
486 if (ret)
487 nx842_inc_comp_failed(local_devdata);
488 else {
489 nx842_inc_comp_complete(local_devdata);
490 ibm_nx842_incr_hist(local_devdata->counters->comp_times,
491 (get_tb() - start_time) / tb_ticks_per_usec);
492 }
493 rcu_read_unlock();
494 return ret;
495 }
496 EXPORT_SYMBOL_GPL(nx842_compress);
497
498 static int sw842_decompress(const unsigned char *, int, unsigned char *, int *,
499 const void *);
500
501 /**
502 * nx842_decompress - Decompress data using the 842 algorithm
503 *
504 * Decompression provide by the NX842 coprocessor on IBM Power systems.
505 * The input buffer is decompressed and the result is stored in the
506 * provided output buffer. The size allocated to the output buffer is
507 * provided by the caller of this function in @outlen. Upon return from
508 * this function @outlen contains the length of the decompressed data.
509 * If there is an error then @outlen will be 0 and an error will be
510 * specified by the return code from this function.
511 *
512 * @in: Pointer to input buffer, will use bounce buffer if not 128 byte
513 * aligned
514 * @inlen: Length of input buffer
515 * @out: Pointer to output buffer, must be page aligned
516 * @outlen: Length of output buffer, must be PAGE_SIZE
517 * @wrkmem: ptr to buffer for working memory, size determined by
518 * nx842_get_workmem_size()
519 *
520 * Returns:
521 * 0 Success, output of length @outlen stored in the buffer at @out
522 * -ENODEV Hardware decompression device is unavailable
523 * -ENOMEM Unable to allocate internal buffers
524 * -ENOSPC Output buffer is to small
525 * -EINVAL Bad input data encountered when attempting decompress
526 * -EIO Internal error
527 */
528 int nx842_decompress(const unsigned char *in, unsigned int inlen,
529 unsigned char *out, unsigned int *outlen, void *wmem)
530 {
531 struct nx842_header *hdr;
532 struct nx842_devdata *local_devdata;
533 struct device *dev = NULL;
534 struct nx842_workmem *workmem;
535 struct nx842_scatterlist slin, slout;
536 struct nx_csbcpb *csbcpb;
537 int ret = 0, i, size, max_sync_size;
538 unsigned long inbuf, outbuf;
539 struct vio_pfo_op op = {
540 .done = NULL,
541 .handle = 0,
542 .timeout = 0,
543 };
544 unsigned long start_time = get_tb();
545
546 /* Ensure page alignment and size */
547 outbuf = (unsigned long)out;
548 if (!IS_ALIGNED(outbuf, PAGE_SIZE) || *outlen != PAGE_SIZE)
549 return -EINVAL;
550
551 rcu_read_lock();
552 local_devdata = rcu_dereference(devdata);
553 if (local_devdata)
554 dev = local_devdata->dev;
555
556 /* Get header */
557 hdr = (struct nx842_header *)in;
558
559 workmem = (struct nx842_workmem *)ALIGN((unsigned long)wmem,
560 NX842_HW_PAGE_SIZE);
561
562 inbuf = (unsigned long)in + hdr->offset;
563 if (likely(!IS_ALIGNED(inbuf, IO_BUFFER_ALIGN))) {
564 /* Copy block(s) into bounce buffer for alignment */
565 memcpy(workmem->bounce, in + hdr->offset, inlen - hdr->offset);
566 inbuf = (unsigned long)workmem->bounce;
567 }
568
569 /* Init scatterlist */
570 slin.entries = (struct nx842_slentry *)workmem->slin;
571 slout.entries = (struct nx842_slentry *)workmem->slout;
572
573 /* Init operation */
574 op.flags = NX842_OP_DECOMPRESS;
575 csbcpb = &workmem->csbcpb;
576 memset(csbcpb, 0, sizeof(*csbcpb));
577 op.csbcpb = nx842_get_pa(csbcpb);
578
579 /*
580 * max_sync_size may have changed since compression,
581 * so we can't read it from the device info. We need
582 * to derive it from hdr->blocks_nr.
583 */
584 max_sync_size = PAGE_SIZE / hdr->blocks_nr;
585
586 for (i = 0; i < hdr->blocks_nr; i++) {
587 /* Skip padding */
588 inbuf = ALIGN(inbuf, IO_BUFFER_ALIGN);
589
590 if (hdr->sizes[i] < 0) {
591 /* Negative sizes indicate uncompressed data blocks */
592 size = abs(hdr->sizes[i]);
593 memcpy((void *)outbuf, (void *)inbuf, size);
594 outbuf += size;
595 inbuf += size;
596 continue;
597 }
598
599 if (!dev)
600 goto sw;
601
602 /*
603 * The better the compression, the more likely the "likely"
604 * case becomes.
605 */
606 if (likely((inbuf & NX842_HW_PAGE_MASK) ==
607 ((inbuf + hdr->sizes[i] - 1) & NX842_HW_PAGE_MASK))) {
608 /* Create direct DDE */
609 op.in = nx842_get_pa((void *)inbuf);
610 op.inlen = hdr->sizes[i];
611 } else {
612 /* Create indirect DDE (scatterlist) */
613 nx842_build_scatterlist(inbuf, hdr->sizes[i] , &slin);
614 op.in = nx842_get_pa(slin.entries);
615 op.inlen = -nx842_get_scatterlist_size(&slin);
616 }
617
618 /*
619 * NOTE: If the default max_sync_size is changed from 4k
620 * to 64k, remove the "likely" case below, since a
621 * scatterlist will always be needed.
622 */
623 if (likely(max_sync_size == NX842_HW_PAGE_SIZE)) {
624 /* Create direct DDE */
625 op.out = nx842_get_pa((void *)outbuf);
626 op.outlen = max_sync_size;
627 } else {
628 /* Create indirect DDE (scatterlist) */
629 nx842_build_scatterlist(outbuf, max_sync_size, &slout);
630 op.out = nx842_get_pa(slout.entries);
631 op.outlen = -nx842_get_scatterlist_size(&slout);
632 }
633
634 /* Send request to pHyp */
635 ret = vio_h_cop_sync(local_devdata->vdev, &op);
636
637 /* Check for pHyp error */
638 if (ret) {
639 dev_dbg(dev, "%s: vio_h_cop_sync error (ret=%d, hret=%ld)\n",
640 __func__, ret, op.hcall_err);
641 dev = NULL;
642 goto sw;
643 }
644
645 /* Check for hardware error */
646 ret = nx842_validate_result(dev, &csbcpb->csb);
647 if (ret) {
648 dev = NULL;
649 goto sw;
650 }
651
652 /* HW decompression success */
653 inbuf += hdr->sizes[i];
654 outbuf += csbcpb->csb.processed_byte_count;
655 continue;
656
657 sw:
658 /* software decompression */
659 size = max_sync_size;
660 ret = sw842_decompress(
661 (unsigned char *)inbuf, hdr->sizes[i],
662 (unsigned char *)outbuf, &size, wmem);
663 if (ret)
664 pr_debug("%s: sw842_decompress failed with %d\n",
665 __func__, ret);
666
667 if (ret) {
668 if (ret != -ENOSPC && ret != -EINVAL &&
669 ret != -EMSGSIZE)
670 ret = -EIO;
671 goto unlock;
672 }
673
674 /* SW decompression success */
675 inbuf += hdr->sizes[i];
676 outbuf += size;
677 }
678
679 *outlen = (unsigned int)(outbuf - (unsigned long)out);
680
681 unlock:
682 if (ret)
683 /* decompress fail */
684 nx842_inc_decomp_failed(local_devdata);
685 else {
686 if (!dev)
687 /* software decompress */
688 nx842_inc_swdecomp(local_devdata);
689 nx842_inc_decomp_complete(local_devdata);
690 ibm_nx842_incr_hist(local_devdata->counters->decomp_times,
691 (get_tb() - start_time) / tb_ticks_per_usec);
692 }
693
694 rcu_read_unlock();
695 return ret;
696 }
697 EXPORT_SYMBOL_GPL(nx842_decompress);
698
699 /**
700 * nx842_OF_set_defaults -- Set default (disabled) values for devdata
701 *
702 * @devdata - struct nx842_devdata to update
703 *
704 * Returns:
705 * 0 on success
706 * -ENOENT if @devdata ptr is NULL
707 */
708 static int nx842_OF_set_defaults(struct nx842_devdata *devdata)
709 {
710 if (devdata) {
711 devdata->max_sync_size = 0;
712 devdata->max_sync_sg = 0;
713 devdata->max_sg_len = 0;
714 devdata->status = UNAVAILABLE;
715 return 0;
716 } else
717 return -ENOENT;
718 }
719
720 /**
721 * nx842_OF_upd_status -- Update the device info from OF status prop
722 *
723 * The status property indicates if the accelerator is enabled. If the
724 * device is in the OF tree it indicates that the hardware is present.
725 * The status field indicates if the device is enabled when the status
726 * is 'okay'. Otherwise the device driver will be disabled.
727 *
728 * @devdata - struct nx842_devdata to update
729 * @prop - struct property point containing the maxsyncop for the update
730 *
731 * Returns:
732 * 0 - Device is available
733 * -EINVAL - Device is not available
734 */
735 static int nx842_OF_upd_status(struct nx842_devdata *devdata,
736 struct property *prop) {
737 int ret = 0;
738 const char *status = (const char *)prop->value;
739
740 if (!strncmp(status, "okay", (size_t)prop->length)) {
741 devdata->status = AVAILABLE;
742 } else {
743 dev_info(devdata->dev, "%s: status '%s' is not 'okay'\n",
744 __func__, status);
745 devdata->status = UNAVAILABLE;
746 }
747
748 return ret;
749 }
750
751 /**
752 * nx842_OF_upd_maxsglen -- Update the device info from OF maxsglen prop
753 *
754 * Definition of the 'ibm,max-sg-len' OF property:
755 * This field indicates the maximum byte length of a scatter list
756 * for the platform facility. It is a single cell encoded as with encode-int.
757 *
758 * Example:
759 * # od -x ibm,max-sg-len
760 * 0000000 0000 0ff0
761 *
762 * In this example, the maximum byte length of a scatter list is
763 * 0x0ff0 (4,080).
764 *
765 * @devdata - struct nx842_devdata to update
766 * @prop - struct property point containing the maxsyncop for the update
767 *
768 * Returns:
769 * 0 on success
770 * -EINVAL on failure
771 */
772 static int nx842_OF_upd_maxsglen(struct nx842_devdata *devdata,
773 struct property *prop) {
774 int ret = 0;
775 const int *maxsglen = prop->value;
776
777 if (prop->length != sizeof(*maxsglen)) {
778 dev_err(devdata->dev, "%s: unexpected format for ibm,max-sg-len property\n", __func__);
779 dev_dbg(devdata->dev, "%s: ibm,max-sg-len is %d bytes long, expected %lu bytes\n", __func__,
780 prop->length, sizeof(*maxsglen));
781 ret = -EINVAL;
782 } else {
783 devdata->max_sg_len = (unsigned int)min(*maxsglen,
784 (int)NX842_HW_PAGE_SIZE);
785 }
786
787 return ret;
788 }
789
790 /**
791 * nx842_OF_upd_maxsyncop -- Update the device info from OF maxsyncop prop
792 *
793 * Definition of the 'ibm,max-sync-cop' OF property:
794 * Two series of cells. The first series of cells represents the maximums
795 * that can be synchronously compressed. The second series of cells
796 * represents the maximums that can be synchronously decompressed.
797 * 1. The first cell in each series contains the count of the number of
798 * data length, scatter list elements pairs that follow – each being
799 * of the form
800 * a. One cell data byte length
801 * b. One cell total number of scatter list elements
802 *
803 * Example:
804 * # od -x ibm,max-sync-cop
805 * 0000000 0000 0001 0000 1000 0000 01fe 0000 0001
806 * 0000020 0000 1000 0000 01fe
807 *
808 * In this example, compression supports 0x1000 (4,096) data byte length
809 * and 0x1fe (510) total scatter list elements. Decompression supports
810 * 0x1000 (4,096) data byte length and 0x1f3 (510) total scatter list
811 * elements.
812 *
813 * @devdata - struct nx842_devdata to update
814 * @prop - struct property point containing the maxsyncop for the update
815 *
816 * Returns:
817 * 0 on success
818 * -EINVAL on failure
819 */
820 static int nx842_OF_upd_maxsyncop(struct nx842_devdata *devdata,
821 struct property *prop) {
822 int ret = 0;
823 const struct maxsynccop_t {
824 int comp_elements;
825 int comp_data_limit;
826 int comp_sg_limit;
827 int decomp_elements;
828 int decomp_data_limit;
829 int decomp_sg_limit;
830 } *maxsynccop;
831
832 if (prop->length != sizeof(*maxsynccop)) {
833 dev_err(devdata->dev, "%s: unexpected format for ibm,max-sync-cop property\n", __func__);
834 dev_dbg(devdata->dev, "%s: ibm,max-sync-cop is %d bytes long, expected %lu bytes\n", __func__, prop->length,
835 sizeof(*maxsynccop));
836 ret = -EINVAL;
837 goto out;
838 }
839
840 maxsynccop = (const struct maxsynccop_t *)prop->value;
841
842 /* Use one limit rather than separate limits for compression and
843 * decompression. Set a maximum for this so as not to exceed the
844 * size that the header can support and round the value down to
845 * the hardware page size (4K) */
846 devdata->max_sync_size =
847 (unsigned int)min(maxsynccop->comp_data_limit,
848 maxsynccop->decomp_data_limit);
849
850 devdata->max_sync_size = min_t(unsigned int, devdata->max_sync_size,
851 SIZE_64K);
852
853 if (devdata->max_sync_size < SIZE_4K) {
854 dev_err(devdata->dev, "%s: hardware max data size (%u) is "
855 "less than the driver minimum, unable to use "
856 "the hardware device\n",
857 __func__, devdata->max_sync_size);
858 ret = -EINVAL;
859 goto out;
860 }
861
862 devdata->max_sync_sg = (unsigned int)min(maxsynccop->comp_sg_limit,
863 maxsynccop->decomp_sg_limit);
864 if (devdata->max_sync_sg < 1) {
865 dev_err(devdata->dev, "%s: hardware max sg size (%u) is "
866 "less than the driver minimum, unable to use "
867 "the hardware device\n",
868 __func__, devdata->max_sync_sg);
869 ret = -EINVAL;
870 goto out;
871 }
872
873 out:
874 return ret;
875 }
876
877 /**
878 *
879 * nx842_OF_upd -- Handle OF properties updates for the device.
880 *
881 * Set all properties from the OF tree. Optionally, a new property
882 * can be provided by the @new_prop pointer to overwrite an existing value.
883 * The device will remain disabled until all values are valid, this function
884 * will return an error for updates unless all values are valid.
885 *
886 * @new_prop: If not NULL, this property is being updated. If NULL, update
887 * all properties from the current values in the OF tree.
888 *
889 * Returns:
890 * 0 - Success
891 * -ENOMEM - Could not allocate memory for new devdata structure
892 * -EINVAL - property value not found, new_prop is not a recognized
893 * property for the device or property value is not valid.
894 * -ENODEV - Device is not available
895 */
896 static int nx842_OF_upd(struct property *new_prop)
897 {
898 struct nx842_devdata *old_devdata = NULL;
899 struct nx842_devdata *new_devdata = NULL;
900 struct device_node *of_node = NULL;
901 struct property *status = NULL;
902 struct property *maxsglen = NULL;
903 struct property *maxsyncop = NULL;
904 int ret = 0;
905 unsigned long flags;
906
907 spin_lock_irqsave(&devdata_mutex, flags);
908 old_devdata = rcu_dereference_check(devdata,
909 lockdep_is_held(&devdata_mutex));
910 if (old_devdata)
911 of_node = old_devdata->dev->of_node;
912
913 if (!old_devdata || !of_node) {
914 pr_err("%s: device is not available\n", __func__);
915 spin_unlock_irqrestore(&devdata_mutex, flags);
916 return -ENODEV;
917 }
918
919 new_devdata = kzalloc(sizeof(*new_devdata), GFP_NOFS);
920 if (!new_devdata) {
921 dev_err(old_devdata->dev, "%s: Could not allocate memory for device data\n", __func__);
922 ret = -ENOMEM;
923 goto error_out;
924 }
925
926 memcpy(new_devdata, old_devdata, sizeof(*old_devdata));
927 new_devdata->counters = old_devdata->counters;
928
929 /* Set ptrs for existing properties */
930 status = of_find_property(of_node, "status", NULL);
931 maxsglen = of_find_property(of_node, "ibm,max-sg-len", NULL);
932 maxsyncop = of_find_property(of_node, "ibm,max-sync-cop", NULL);
933 if (!status || !maxsglen || !maxsyncop) {
934 dev_err(old_devdata->dev, "%s: Could not locate device properties\n", __func__);
935 ret = -EINVAL;
936 goto error_out;
937 }
938
939 /*
940 * If this is a property update, there are only certain properties that
941 * we care about. Bail if it isn't in the below list
942 */
943 if (new_prop && (strncmp(new_prop->name, "status", new_prop->length) ||
944 strncmp(new_prop->name, "ibm,max-sg-len", new_prop->length) ||
945 strncmp(new_prop->name, "ibm,max-sync-cop", new_prop->length)))
946 goto out;
947
948 /* Perform property updates */
949 ret = nx842_OF_upd_status(new_devdata, status);
950 if (ret)
951 goto error_out;
952
953 ret = nx842_OF_upd_maxsglen(new_devdata, maxsglen);
954 if (ret)
955 goto error_out;
956
957 ret = nx842_OF_upd_maxsyncop(new_devdata, maxsyncop);
958 if (ret)
959 goto error_out;
960
961 out:
962 dev_info(old_devdata->dev, "%s: max_sync_size new:%u old:%u\n",
963 __func__, new_devdata->max_sync_size,
964 old_devdata->max_sync_size);
965 dev_info(old_devdata->dev, "%s: max_sync_sg new:%u old:%u\n",
966 __func__, new_devdata->max_sync_sg,
967 old_devdata->max_sync_sg);
968 dev_info(old_devdata->dev, "%s: max_sg_len new:%u old:%u\n",
969 __func__, new_devdata->max_sg_len,
970 old_devdata->max_sg_len);
971
972 rcu_assign_pointer(devdata, new_devdata);
973 spin_unlock_irqrestore(&devdata_mutex, flags);
974 synchronize_rcu();
975 dev_set_drvdata(new_devdata->dev, new_devdata);
976 kfree(old_devdata);
977 return 0;
978
979 error_out:
980 if (new_devdata) {
981 dev_info(old_devdata->dev, "%s: device disabled\n", __func__);
982 nx842_OF_set_defaults(new_devdata);
983 rcu_assign_pointer(devdata, new_devdata);
984 spin_unlock_irqrestore(&devdata_mutex, flags);
985 synchronize_rcu();
986 dev_set_drvdata(new_devdata->dev, new_devdata);
987 kfree(old_devdata);
988 } else {
989 dev_err(old_devdata->dev, "%s: could not update driver from hardware\n", __func__);
990 spin_unlock_irqrestore(&devdata_mutex, flags);
991 }
992
993 if (!ret)
994 ret = -EINVAL;
995 return ret;
996 }
997
998 /**
999 * nx842_OF_notifier - Process updates to OF properties for the device
1000 *
1001 * @np: notifier block
1002 * @action: notifier action
1003 * @update: struct pSeries_reconfig_prop_update pointer if action is
1004 * PSERIES_UPDATE_PROPERTY
1005 *
1006 * Returns:
1007 * NOTIFY_OK on success
1008 * NOTIFY_BAD encoded with error number on failure, use
1009 * notifier_to_errno() to decode this value
1010 */
1011 static int nx842_OF_notifier(struct notifier_block *np, unsigned long action,
1012 void *data)
1013 {
1014 struct of_reconfig_data *upd = data;
1015 struct nx842_devdata *local_devdata;
1016 struct device_node *node = NULL;
1017
1018 rcu_read_lock();
1019 local_devdata = rcu_dereference(devdata);
1020 if (local_devdata)
1021 node = local_devdata->dev->of_node;
1022
1023 if (local_devdata &&
1024 action == OF_RECONFIG_UPDATE_PROPERTY &&
1025 !strcmp(upd->dn->name, node->name)) {
1026 rcu_read_unlock();
1027 nx842_OF_upd(upd->prop);
1028 } else
1029 rcu_read_unlock();
1030
1031 return NOTIFY_OK;
1032 }
1033
1034 static struct notifier_block nx842_of_nb = {
1035 .notifier_call = nx842_OF_notifier,
1036 };
1037
1038 #define nx842_counter_read(_name) \
1039 static ssize_t nx842_##_name##_show(struct device *dev, \
1040 struct device_attribute *attr, \
1041 char *buf) { \
1042 struct nx842_devdata *local_devdata; \
1043 int p = 0; \
1044 rcu_read_lock(); \
1045 local_devdata = rcu_dereference(devdata); \
1046 if (local_devdata) \
1047 p = snprintf(buf, PAGE_SIZE, "%ld\n", \
1048 atomic64_read(&local_devdata->counters->_name)); \
1049 rcu_read_unlock(); \
1050 return p; \
1051 }
1052
1053 #define NX842DEV_COUNTER_ATTR_RO(_name) \
1054 nx842_counter_read(_name); \
1055 static struct device_attribute dev_attr_##_name = __ATTR(_name, \
1056 0444, \
1057 nx842_##_name##_show,\
1058 NULL);
1059
1060 NX842DEV_COUNTER_ATTR_RO(comp_complete);
1061 NX842DEV_COUNTER_ATTR_RO(comp_failed);
1062 NX842DEV_COUNTER_ATTR_RO(decomp_complete);
1063 NX842DEV_COUNTER_ATTR_RO(decomp_failed);
1064 NX842DEV_COUNTER_ATTR_RO(swdecomp);
1065
1066 static ssize_t nx842_timehist_show(struct device *,
1067 struct device_attribute *, char *);
1068
1069 static struct device_attribute dev_attr_comp_times = __ATTR(comp_times, 0444,
1070 nx842_timehist_show, NULL);
1071 static struct device_attribute dev_attr_decomp_times = __ATTR(decomp_times,
1072 0444, nx842_timehist_show, NULL);
1073
1074 static ssize_t nx842_timehist_show(struct device *dev,
1075 struct device_attribute *attr, char *buf) {
1076 char *p = buf;
1077 struct nx842_devdata *local_devdata;
1078 atomic64_t *times;
1079 int bytes_remain = PAGE_SIZE;
1080 int bytes;
1081 int i;
1082
1083 rcu_read_lock();
1084 local_devdata = rcu_dereference(devdata);
1085 if (!local_devdata) {
1086 rcu_read_unlock();
1087 return 0;
1088 }
1089
1090 if (attr == &dev_attr_comp_times)
1091 times = local_devdata->counters->comp_times;
1092 else if (attr == &dev_attr_decomp_times)
1093 times = local_devdata->counters->decomp_times;
1094 else {
1095 rcu_read_unlock();
1096 return 0;
1097 }
1098
1099 for (i = 0; i < (NX842_HIST_SLOTS - 2); i++) {
1100 bytes = snprintf(p, bytes_remain, "%u-%uus:\t%ld\n",
1101 i ? (2<<(i-1)) : 0, (2<<i)-1,
1102 atomic64_read(&times[i]));
1103 bytes_remain -= bytes;
1104 p += bytes;
1105 }
1106 /* The last bucket holds everything over
1107 * 2<<(NX842_HIST_SLOTS - 2) us */
1108 bytes = snprintf(p, bytes_remain, "%uus - :\t%ld\n",
1109 2<<(NX842_HIST_SLOTS - 2),
1110 atomic64_read(&times[(NX842_HIST_SLOTS - 1)]));
1111 p += bytes;
1112
1113 rcu_read_unlock();
1114 return p - buf;
1115 }
1116
1117 static struct attribute *nx842_sysfs_entries[] = {
1118 &dev_attr_comp_complete.attr,
1119 &dev_attr_comp_failed.attr,
1120 &dev_attr_decomp_complete.attr,
1121 &dev_attr_decomp_failed.attr,
1122 &dev_attr_swdecomp.attr,
1123 &dev_attr_comp_times.attr,
1124 &dev_attr_decomp_times.attr,
1125 NULL,
1126 };
1127
1128 static struct attribute_group nx842_attribute_group = {
1129 .name = NULL, /* put in device directory */
1130 .attrs = nx842_sysfs_entries,
1131 };
1132
1133 static int __init nx842_probe(struct vio_dev *viodev,
1134 const struct vio_device_id *id)
1135 {
1136 struct nx842_devdata *old_devdata, *new_devdata = NULL;
1137 unsigned long flags;
1138 int ret = 0;
1139
1140 spin_lock_irqsave(&devdata_mutex, flags);
1141 old_devdata = rcu_dereference_check(devdata,
1142 lockdep_is_held(&devdata_mutex));
1143
1144 if (old_devdata && old_devdata->vdev != NULL) {
1145 dev_err(&viodev->dev, "%s: Attempt to register more than one instance of the hardware\n", __func__);
1146 ret = -1;
1147 goto error_unlock;
1148 }
1149
1150 dev_set_drvdata(&viodev->dev, NULL);
1151
1152 new_devdata = kzalloc(sizeof(*new_devdata), GFP_NOFS);
1153 if (!new_devdata) {
1154 dev_err(&viodev->dev, "%s: Could not allocate memory for device data\n", __func__);
1155 ret = -ENOMEM;
1156 goto error_unlock;
1157 }
1158
1159 new_devdata->counters = kzalloc(sizeof(*new_devdata->counters),
1160 GFP_NOFS);
1161 if (!new_devdata->counters) {
1162 dev_err(&viodev->dev, "%s: Could not allocate memory for performance counters\n", __func__);
1163 ret = -ENOMEM;
1164 goto error_unlock;
1165 }
1166
1167 new_devdata->vdev = viodev;
1168 new_devdata->dev = &viodev->dev;
1169 nx842_OF_set_defaults(new_devdata);
1170
1171 rcu_assign_pointer(devdata, new_devdata);
1172 spin_unlock_irqrestore(&devdata_mutex, flags);
1173 synchronize_rcu();
1174 kfree(old_devdata);
1175
1176 of_reconfig_notifier_register(&nx842_of_nb);
1177
1178 ret = nx842_OF_upd(NULL);
1179 if (ret && ret != -ENODEV) {
1180 dev_err(&viodev->dev, "could not parse device tree. %d\n", ret);
1181 ret = -1;
1182 goto error;
1183 }
1184
1185 rcu_read_lock();
1186 dev_set_drvdata(&viodev->dev, rcu_dereference(devdata));
1187 rcu_read_unlock();
1188
1189 if (sysfs_create_group(&viodev->dev.kobj, &nx842_attribute_group)) {
1190 dev_err(&viodev->dev, "could not create sysfs device attributes\n");
1191 ret = -1;
1192 goto error;
1193 }
1194
1195 return 0;
1196
1197 error_unlock:
1198 spin_unlock_irqrestore(&devdata_mutex, flags);
1199 if (new_devdata)
1200 kfree(new_devdata->counters);
1201 kfree(new_devdata);
1202 error:
1203 return ret;
1204 }
1205
1206 static int __exit nx842_remove(struct vio_dev *viodev)
1207 {
1208 struct nx842_devdata *old_devdata;
1209 unsigned long flags;
1210
1211 pr_info("Removing IBM Power 842 compression device\n");
1212 sysfs_remove_group(&viodev->dev.kobj, &nx842_attribute_group);
1213
1214 spin_lock_irqsave(&devdata_mutex, flags);
1215 old_devdata = rcu_dereference_check(devdata,
1216 lockdep_is_held(&devdata_mutex));
1217 of_reconfig_notifier_unregister(&nx842_of_nb);
1218 RCU_INIT_POINTER(devdata, NULL);
1219 spin_unlock_irqrestore(&devdata_mutex, flags);
1220 synchronize_rcu();
1221 dev_set_drvdata(&viodev->dev, NULL);
1222 if (old_devdata)
1223 kfree(old_devdata->counters);
1224 kfree(old_devdata);
1225 return 0;
1226 }
1227
1228 static struct vio_device_id nx842_driver_ids[] = {
1229 {"ibm,compression-v1", "ibm,compression"},
1230 {"", ""},
1231 };
1232
1233 static struct vio_driver nx842_driver = {
1234 .name = MODULE_NAME,
1235 .probe = nx842_probe,
1236 .remove = __exit_p(nx842_remove),
1237 .get_desired_dma = nx842_get_desired_dma,
1238 .id_table = nx842_driver_ids,
1239 };
1240
1241 static int __init nx842_init(void)
1242 {
1243 struct nx842_devdata *new_devdata;
1244 pr_info("Registering IBM Power 842 compression driver\n");
1245
1246 RCU_INIT_POINTER(devdata, NULL);
1247 new_devdata = kzalloc(sizeof(*new_devdata), GFP_KERNEL);
1248 if (!new_devdata) {
1249 pr_err("Could not allocate memory for device data\n");
1250 return -ENOMEM;
1251 }
1252 new_devdata->status = UNAVAILABLE;
1253 RCU_INIT_POINTER(devdata, new_devdata);
1254
1255 return vio_register_driver(&nx842_driver);
1256 }
1257
1258 module_init(nx842_init);
1259
1260 static void __exit nx842_exit(void)
1261 {
1262 struct nx842_devdata *old_devdata;
1263 unsigned long flags;
1264
1265 pr_info("Exiting IBM Power 842 compression driver\n");
1266 spin_lock_irqsave(&devdata_mutex, flags);
1267 old_devdata = rcu_dereference_check(devdata,
1268 lockdep_is_held(&devdata_mutex));
1269 RCU_INIT_POINTER(devdata, NULL);
1270 spin_unlock_irqrestore(&devdata_mutex, flags);
1271 synchronize_rcu();
1272 if (old_devdata)
1273 dev_set_drvdata(old_devdata->dev, NULL);
1274 kfree(old_devdata);
1275 vio_unregister_driver(&nx842_driver);
1276 }
1277
1278 module_exit(nx842_exit);
1279
1280 /*********************************
1281 * 842 software decompressor
1282 *********************************/
1283 typedef int (*sw842_template_op)(const char **, int *, unsigned char **,
1284 struct sw842_fifo *);
1285
1286 static int sw842_data8(const char **, int *, unsigned char **,
1287 struct sw842_fifo *);
1288 static int sw842_data4(const char **, int *, unsigned char **,
1289 struct sw842_fifo *);
1290 static int sw842_data2(const char **, int *, unsigned char **,
1291 struct sw842_fifo *);
1292 static int sw842_ptr8(const char **, int *, unsigned char **,
1293 struct sw842_fifo *);
1294 static int sw842_ptr4(const char **, int *, unsigned char **,
1295 struct sw842_fifo *);
1296 static int sw842_ptr2(const char **, int *, unsigned char **,
1297 struct sw842_fifo *);
1298
1299 /* special templates */
1300 #define SW842_TMPL_REPEAT 0x1B
1301 #define SW842_TMPL_ZEROS 0x1C
1302 #define SW842_TMPL_EOF 0x1E
1303
1304 static sw842_template_op sw842_tmpl_ops[26][4] = {
1305 { sw842_data8, NULL}, /* 0 (00000) */
1306 { sw842_data4, sw842_data2, sw842_ptr2, NULL},
1307 { sw842_data4, sw842_ptr2, sw842_data2, NULL},
1308 { sw842_data4, sw842_ptr2, sw842_ptr2, NULL},
1309 { sw842_data4, sw842_ptr4, NULL},
1310 { sw842_data2, sw842_ptr2, sw842_data4, NULL},
1311 { sw842_data2, sw842_ptr2, sw842_data2, sw842_ptr2},
1312 { sw842_data2, sw842_ptr2, sw842_ptr2, sw842_data2},
1313 { sw842_data2, sw842_ptr2, sw842_ptr2, sw842_ptr2,},
1314 { sw842_data2, sw842_ptr2, sw842_ptr4, NULL},
1315 { sw842_ptr2, sw842_data2, sw842_data4, NULL}, /* 10 (01010) */
1316 { sw842_ptr2, sw842_data4, sw842_ptr2, NULL},
1317 { sw842_ptr2, sw842_data2, sw842_ptr2, sw842_data2},
1318 { sw842_ptr2, sw842_data2, sw842_ptr2, sw842_ptr2},
1319 { sw842_ptr2, sw842_data2, sw842_ptr4, NULL},
1320 { sw842_ptr2, sw842_ptr2, sw842_data4, NULL},
1321 { sw842_ptr2, sw842_ptr2, sw842_data2, sw842_ptr2},
1322 { sw842_ptr2, sw842_ptr2, sw842_ptr2, sw842_data2},
1323 { sw842_ptr2, sw842_ptr2, sw842_ptr2, sw842_ptr2},
1324 { sw842_ptr2, sw842_ptr2, sw842_ptr4, NULL},
1325 { sw842_ptr4, sw842_data4, NULL}, /* 20 (10100) */
1326 { sw842_ptr4, sw842_data2, sw842_ptr2, NULL},
1327 { sw842_ptr4, sw842_ptr2, sw842_data2, NULL},
1328 { sw842_ptr4, sw842_ptr2, sw842_ptr2, NULL},
1329 { sw842_ptr4, sw842_ptr4, NULL},
1330 { sw842_ptr8, NULL}
1331 };
1332
1333 /* Software decompress helpers */
1334
1335 static uint8_t sw842_get_byte(const char *buf, int bit)
1336 {
1337 uint8_t tmpl;
1338 uint16_t tmp;
1339 tmp = htons(*(uint16_t *)(buf));
1340 tmp = (uint16_t)(tmp << bit);
1341 tmp = ntohs(tmp);
1342 memcpy(&tmpl, &tmp, 1);
1343 return tmpl;
1344 }
1345
1346 static uint8_t sw842_get_template(const char **buf, int *bit)
1347 {
1348 uint8_t byte;
1349 byte = sw842_get_byte(*buf, *bit);
1350 byte = byte >> 3;
1351 byte &= 0x1F;
1352 *buf += (*bit + 5) / 8;
1353 *bit = (*bit + 5) % 8;
1354 return byte;
1355 }
1356
1357 /* repeat_count happens to be 5-bit too (like the template) */
1358 static uint8_t sw842_get_repeat_count(const char **buf, int *bit)
1359 {
1360 uint8_t byte;
1361 byte = sw842_get_byte(*buf, *bit);
1362 byte = byte >> 2;
1363 byte &= 0x3F;
1364 *buf += (*bit + 6) / 8;
1365 *bit = (*bit + 6) % 8;
1366 return byte;
1367 }
1368
1369 static uint8_t sw842_get_ptr2(const char **buf, int *bit)
1370 {
1371 uint8_t ptr;
1372 ptr = sw842_get_byte(*buf, *bit);
1373 (*buf)++;
1374 return ptr;
1375 }
1376
1377 static uint16_t sw842_get_ptr4(const char **buf, int *bit,
1378 struct sw842_fifo *fifo)
1379 {
1380 uint16_t ptr;
1381 ptr = htons(*(uint16_t *)(*buf));
1382 ptr = (uint16_t)(ptr << *bit);
1383 ptr = ptr >> 7;
1384 ptr &= 0x01FF;
1385 *buf += (*bit + 9) / 8;
1386 *bit = (*bit + 9) % 8;
1387 return ptr;
1388 }
1389
1390 static uint8_t sw842_get_ptr8(const char **buf, int *bit,
1391 struct sw842_fifo *fifo)
1392 {
1393 return sw842_get_ptr2(buf, bit);
1394 }
1395
1396 /* Software decompress template ops */
1397
1398 static int sw842_data8(const char **inbuf, int *inbit,
1399 unsigned char **outbuf, struct sw842_fifo *fifo)
1400 {
1401 int ret;
1402
1403 ret = sw842_data4(inbuf, inbit, outbuf, fifo);
1404 if (ret)
1405 return ret;
1406 ret = sw842_data4(inbuf, inbit, outbuf, fifo);
1407 return ret;
1408 }
1409
1410 static int sw842_data4(const char **inbuf, int *inbit,
1411 unsigned char **outbuf, struct sw842_fifo *fifo)
1412 {
1413 int ret;
1414
1415 ret = sw842_data2(inbuf, inbit, outbuf, fifo);
1416 if (ret)
1417 return ret;
1418 ret = sw842_data2(inbuf, inbit, outbuf, fifo);
1419 return ret;
1420 }
1421
1422 static int sw842_data2(const char **inbuf, int *inbit,
1423 unsigned char **outbuf, struct sw842_fifo *fifo)
1424 {
1425 **outbuf = sw842_get_byte(*inbuf, *inbit);
1426 (*inbuf)++;
1427 (*outbuf)++;
1428 **outbuf = sw842_get_byte(*inbuf, *inbit);
1429 (*inbuf)++;
1430 (*outbuf)++;
1431 return 0;
1432 }
1433
1434 static int sw842_ptr8(const char **inbuf, int *inbit,
1435 unsigned char **outbuf, struct sw842_fifo *fifo)
1436 {
1437 uint8_t ptr;
1438 ptr = sw842_get_ptr8(inbuf, inbit, fifo);
1439 if (!fifo->f84_full && (ptr >= fifo->f8_count))
1440 return 1;
1441 memcpy(*outbuf, fifo->f8[ptr], 8);
1442 *outbuf += 8;
1443 return 0;
1444 }
1445
1446 static int sw842_ptr4(const char **inbuf, int *inbit,
1447 unsigned char **outbuf, struct sw842_fifo *fifo)
1448 {
1449 uint16_t ptr;
1450 ptr = sw842_get_ptr4(inbuf, inbit, fifo);
1451 if (!fifo->f84_full && (ptr >= fifo->f4_count))
1452 return 1;
1453 memcpy(*outbuf, fifo->f4[ptr], 4);
1454 *outbuf += 4;
1455 return 0;
1456 }
1457
1458 static int sw842_ptr2(const char **inbuf, int *inbit,
1459 unsigned char **outbuf, struct sw842_fifo *fifo)
1460 {
1461 uint8_t ptr;
1462 ptr = sw842_get_ptr2(inbuf, inbit);
1463 if (!fifo->f2_full && (ptr >= fifo->f2_count))
1464 return 1;
1465 memcpy(*outbuf, fifo->f2[ptr], 2);
1466 *outbuf += 2;
1467 return 0;
1468 }
1469
1470 static void sw842_copy_to_fifo(const char *buf, struct sw842_fifo *fifo)
1471 {
1472 unsigned char initial_f2count = fifo->f2_count;
1473
1474 memcpy(fifo->f8[fifo->f8_count], buf, 8);
1475 fifo->f4_count += 2;
1476 fifo->f8_count += 1;
1477
1478 if (!fifo->f84_full && fifo->f4_count >= 512) {
1479 fifo->f84_full = 1;
1480 fifo->f4_count /= 512;
1481 }
1482
1483 memcpy(fifo->f2[fifo->f2_count++], buf, 2);
1484 memcpy(fifo->f2[fifo->f2_count++], buf + 2, 2);
1485 memcpy(fifo->f2[fifo->f2_count++], buf + 4, 2);
1486 memcpy(fifo->f2[fifo->f2_count++], buf + 6, 2);
1487 if (fifo->f2_count < initial_f2count)
1488 fifo->f2_full = 1;
1489 }
1490
1491 static int sw842_decompress(const unsigned char *src, int srclen,
1492 unsigned char *dst, int *destlen,
1493 const void *wrkmem)
1494 {
1495 uint8_t tmpl;
1496 const char *inbuf;
1497 int inbit = 0;
1498 unsigned char *outbuf, *outbuf_end, *origbuf, *prevbuf;
1499 const char *inbuf_end;
1500 sw842_template_op op;
1501 int opindex;
1502 int i, repeat_count;
1503 struct sw842_fifo *fifo;
1504 int ret = 0;
1505
1506 fifo = &((struct nx842_workmem *)(wrkmem))->swfifo;
1507 memset(fifo, 0, sizeof(*fifo));
1508
1509 origbuf = NULL;
1510 inbuf = src;
1511 inbuf_end = src + srclen;
1512 outbuf = dst;
1513 outbuf_end = dst + *destlen;
1514
1515 while ((tmpl = sw842_get_template(&inbuf, &inbit)) != SW842_TMPL_EOF) {
1516 if (inbuf >= inbuf_end) {
1517 ret = -EINVAL;
1518 goto out;
1519 }
1520
1521 opindex = 0;
1522 prevbuf = origbuf;
1523 origbuf = outbuf;
1524 switch (tmpl) {
1525 case SW842_TMPL_REPEAT:
1526 if (prevbuf == NULL) {
1527 ret = -EINVAL;
1528 goto out;
1529 }
1530
1531 repeat_count = sw842_get_repeat_count(&inbuf,
1532 &inbit) + 1;
1533
1534 /* Did the repeat count advance past the end of input */
1535 if (inbuf > inbuf_end) {
1536 ret = -EINVAL;
1537 goto out;
1538 }
1539
1540 for (i = 0; i < repeat_count; i++) {
1541 /* Would this overflow the output buffer */
1542 if ((outbuf + 8) > outbuf_end) {
1543 ret = -ENOSPC;
1544 goto out;
1545 }
1546
1547 memcpy(outbuf, prevbuf, 8);
1548 sw842_copy_to_fifo(outbuf, fifo);
1549 outbuf += 8;
1550 }
1551 break;
1552
1553 case SW842_TMPL_ZEROS:
1554 /* Would this overflow the output buffer */
1555 if ((outbuf + 8) > outbuf_end) {
1556 ret = -ENOSPC;
1557 goto out;
1558 }
1559
1560 memset(outbuf, 0, 8);
1561 sw842_copy_to_fifo(outbuf, fifo);
1562 outbuf += 8;
1563 break;
1564
1565 default:
1566 if (tmpl > 25) {
1567 ret = -EINVAL;
1568 goto out;
1569 }
1570
1571 /* Does this go past the end of the input buffer */
1572 if ((inbuf + 2) > inbuf_end) {
1573 ret = -EINVAL;
1574 goto out;
1575 }
1576
1577 /* Would this overflow the output buffer */
1578 if ((outbuf + 8) > outbuf_end) {
1579 ret = -ENOSPC;
1580 goto out;
1581 }
1582
1583 while (opindex < 4 &&
1584 (op = sw842_tmpl_ops[tmpl][opindex++])
1585 != NULL) {
1586 ret = (*op)(&inbuf, &inbit, &outbuf, fifo);
1587 if (ret) {
1588 ret = -EINVAL;
1589 goto out;
1590 }
1591 sw842_copy_to_fifo(origbuf, fifo);
1592 }
1593 }
1594 }
1595
1596 out:
1597 if (!ret)
1598 *destlen = (unsigned int)(outbuf - dst);
1599 else
1600 *destlen = 0;
1601
1602 return ret;
1603 }
This page took 0.063931 seconds and 5 git commands to generate.