]>
Commit | Line | Data |
---|---|---|
0e16aafb SJ |
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 | ||
33b58b01 | 24 | #include <asm/vio.h> |
0e16aafb | 25 | |
7011a122 | 26 | #include "nx-842.h" |
0e16aafb SJ |
27 | #include "nx_csbcpb.h" /* struct nx_csbcpb */ |
28 | ||
0e16aafb SJ |
29 | MODULE_LICENSE("GPL"); |
30 | MODULE_AUTHOR("Robert Jennings <rcj@linux.vnet.ibm.com>"); | |
31 | MODULE_DESCRIPTION("842 H/W Compression driver for IBM Power processors"); | |
32 | ||
959e6659 | 33 | static struct nx842_constraints nx842_pseries_constraints = { |
3154de71 | 34 | .alignment = DDE_BUFFER_ALIGN, |
959e6659 | 35 | .multiple = DDE_BUFFER_LAST_MULT, |
3154de71 | 36 | .minimum = DDE_BUFFER_LAST_MULT, |
959e6659 DS |
37 | .maximum = PAGE_SIZE, /* dynamic, max_sync_size */ |
38 | }; | |
39 | ||
b8e04187 | 40 | static int check_constraints(unsigned long buf, unsigned int *len, bool in) |
0e16aafb | 41 | { |
b8e04187 DS |
42 | if (!IS_ALIGNED(buf, nx842_pseries_constraints.alignment)) { |
43 | pr_debug("%s buffer 0x%lx not aligned to 0x%x\n", | |
44 | in ? "input" : "output", buf, | |
45 | nx842_pseries_constraints.alignment); | |
46 | return -EINVAL; | |
47 | } | |
48 | if (*len % nx842_pseries_constraints.multiple) { | |
49 | pr_debug("%s buffer len 0x%x not multiple of 0x%x\n", | |
50 | in ? "input" : "output", *len, | |
51 | nx842_pseries_constraints.multiple); | |
52 | if (in) | |
53 | return -EINVAL; | |
54 | *len = round_down(*len, nx842_pseries_constraints.multiple); | |
55 | } | |
56 | if (*len < nx842_pseries_constraints.minimum) { | |
57 | pr_debug("%s buffer len 0x%x under minimum 0x%x\n", | |
58 | in ? "input" : "output", *len, | |
59 | nx842_pseries_constraints.minimum); | |
60 | return -EINVAL; | |
61 | } | |
62 | if (*len > nx842_pseries_constraints.maximum) { | |
63 | pr_debug("%s buffer len 0x%x over maximum 0x%x\n", | |
64 | in ? "input" : "output", *len, | |
65 | nx842_pseries_constraints.maximum); | |
66 | if (in) | |
67 | return -EINVAL; | |
68 | *len = nx842_pseries_constraints.maximum; | |
69 | } | |
70 | return 0; | |
0e16aafb SJ |
71 | } |
72 | ||
b8e04187 DS |
73 | /* I assume we need to align the CSB? */ |
74 | #define WORKMEM_ALIGN (256) | |
75 | ||
76 | struct nx842_workmem { | |
77 | /* scatterlist */ | |
78 | char slin[4096]; | |
79 | char slout[4096]; | |
80 | /* coprocessor status/parameter block */ | |
81 | struct nx_csbcpb csbcpb; | |
82 | ||
83 | char padding[WORKMEM_ALIGN]; | |
84 | } __aligned(WORKMEM_ALIGN); | |
85 | ||
0e16aafb SJ |
86 | /* Macros for fields within nx_csbcpb */ |
87 | /* Check the valid bit within the csbcpb valid field */ | |
88 | #define NX842_CSBCBP_VALID_CHK(x) (x & BIT_MASK(7)) | |
89 | ||
90 | /* CE macros operate on the completion_extension field bits in the csbcpb. | |
91 | * CE0 0=full completion, 1=partial completion | |
92 | * CE1 0=CE0 indicates completion, 1=termination (output may be modified) | |
93 | * CE2 0=processed_bytes is source bytes, 1=processed_bytes is target bytes */ | |
94 | #define NX842_CSBCPB_CE0(x) (x & BIT_MASK(7)) | |
95 | #define NX842_CSBCPB_CE1(x) (x & BIT_MASK(6)) | |
96 | #define NX842_CSBCPB_CE2(x) (x & BIT_MASK(5)) | |
97 | ||
98 | /* The NX unit accepts data only on 4K page boundaries */ | |
b8e04187 | 99 | #define NX842_HW_PAGE_SIZE (4096) |
0e16aafb SJ |
100 | #define NX842_HW_PAGE_MASK (~(NX842_HW_PAGE_SIZE-1)) |
101 | ||
102 | enum nx842_status { | |
103 | UNAVAILABLE, | |
104 | AVAILABLE | |
105 | }; | |
106 | ||
107 | struct ibm_nx842_counters { | |
108 | atomic64_t comp_complete; | |
109 | atomic64_t comp_failed; | |
110 | atomic64_t decomp_complete; | |
111 | atomic64_t decomp_failed; | |
112 | atomic64_t swdecomp; | |
113 | atomic64_t comp_times[32]; | |
114 | atomic64_t decomp_times[32]; | |
115 | }; | |
116 | ||
117 | static struct nx842_devdata { | |
118 | struct vio_dev *vdev; | |
119 | struct device *dev; | |
120 | struct ibm_nx842_counters *counters; | |
121 | unsigned int max_sg_len; | |
122 | unsigned int max_sync_size; | |
123 | unsigned int max_sync_sg; | |
124 | enum nx842_status status; | |
125 | } __rcu *devdata; | |
126 | static DEFINE_SPINLOCK(devdata_mutex); | |
127 | ||
128 | #define NX842_COUNTER_INC(_x) \ | |
129 | static inline void nx842_inc_##_x( \ | |
130 | const struct nx842_devdata *dev) { \ | |
131 | if (dev) \ | |
132 | atomic64_inc(&dev->counters->_x); \ | |
133 | } | |
134 | NX842_COUNTER_INC(comp_complete); | |
135 | NX842_COUNTER_INC(comp_failed); | |
136 | NX842_COUNTER_INC(decomp_complete); | |
137 | NX842_COUNTER_INC(decomp_failed); | |
138 | NX842_COUNTER_INC(swdecomp); | |
139 | ||
140 | #define NX842_HIST_SLOTS 16 | |
141 | ||
142 | static void ibm_nx842_incr_hist(atomic64_t *times, unsigned int time) | |
143 | { | |
144 | int bucket = fls(time); | |
145 | ||
146 | if (bucket) | |
147 | bucket = min((NX842_HIST_SLOTS - 1), bucket - 1); | |
148 | ||
149 | atomic64_inc(×[bucket]); | |
150 | } | |
151 | ||
152 | /* NX unit operation flags */ | |
153 | #define NX842_OP_COMPRESS 0x0 | |
154 | #define NX842_OP_CRC 0x1 | |
155 | #define NX842_OP_DECOMPRESS 0x2 | |
156 | #define NX842_OP_COMPRESS_CRC (NX842_OP_COMPRESS | NX842_OP_CRC) | |
157 | #define NX842_OP_DECOMPRESS_CRC (NX842_OP_DECOMPRESS | NX842_OP_CRC) | |
158 | #define NX842_OP_ASYNC (1<<23) | |
159 | #define NX842_OP_NOTIFY (1<<22) | |
160 | #define NX842_OP_NOTIFY_INT(x) ((x & 0xff)<<8) | |
161 | ||
162 | static unsigned long nx842_get_desired_dma(struct vio_dev *viodev) | |
163 | { | |
164 | /* No use of DMA mappings within the driver. */ | |
165 | return 0; | |
166 | } | |
167 | ||
168 | struct nx842_slentry { | |
c47d6302 DS |
169 | __be64 ptr; /* Real address (use __pa()) */ |
170 | __be64 len; | |
0e16aafb SJ |
171 | }; |
172 | ||
173 | /* pHyp scatterlist entry */ | |
174 | struct nx842_scatterlist { | |
175 | int entry_nr; /* number of slentries */ | |
176 | struct nx842_slentry *entries; /* ptr to array of slentries */ | |
177 | }; | |
178 | ||
179 | /* Does not include sizeof(entry_nr) in the size */ | |
180 | static inline unsigned long nx842_get_scatterlist_size( | |
181 | struct nx842_scatterlist *sl) | |
182 | { | |
183 | return sl->entry_nr * sizeof(struct nx842_slentry); | |
184 | } | |
185 | ||
186 | static int nx842_build_scatterlist(unsigned long buf, int len, | |
187 | struct nx842_scatterlist *sl) | |
188 | { | |
c47d6302 | 189 | unsigned long entrylen; |
0e16aafb SJ |
190 | struct nx842_slentry *entry; |
191 | ||
192 | sl->entry_nr = 0; | |
193 | ||
194 | entry = sl->entries; | |
195 | while (len) { | |
c47d6302 DS |
196 | entry->ptr = cpu_to_be64(nx842_get_pa((void *)buf)); |
197 | entrylen = min_t(int, len, | |
198 | LEN_ON_SIZE(buf, NX842_HW_PAGE_SIZE)); | |
199 | entry->len = cpu_to_be64(entrylen); | |
200 | ||
201 | len -= entrylen; | |
202 | buf += entrylen; | |
203 | ||
0e16aafb SJ |
204 | sl->entry_nr++; |
205 | entry++; | |
206 | } | |
207 | ||
208 | return 0; | |
209 | } | |
210 | ||
0e16aafb SJ |
211 | static int nx842_validate_result(struct device *dev, |
212 | struct cop_status_block *csb) | |
213 | { | |
214 | /* The csb must be valid after returning from vio_h_cop_sync */ | |
215 | if (!NX842_CSBCBP_VALID_CHK(csb->valid)) { | |
216 | dev_err(dev, "%s: cspcbp not valid upon completion.\n", | |
217 | __func__); | |
218 | dev_dbg(dev, "valid:0x%02x cs:0x%02x cc:0x%02x ce:0x%02x\n", | |
219 | csb->valid, | |
220 | csb->crb_seq_number, | |
221 | csb->completion_code, | |
222 | csb->completion_extension); | |
223 | dev_dbg(dev, "processed_bytes:%d address:0x%016lx\n", | |
c47d6302 DS |
224 | be32_to_cpu(csb->processed_byte_count), |
225 | (unsigned long)be64_to_cpu(csb->address)); | |
0e16aafb SJ |
226 | return -EIO; |
227 | } | |
228 | ||
229 | /* Check return values from the hardware in the CSB */ | |
230 | switch (csb->completion_code) { | |
231 | case 0: /* Completed without error */ | |
232 | break; | |
233 | case 64: /* Target bytes > Source bytes during compression */ | |
234 | case 13: /* Output buffer too small */ | |
235 | dev_dbg(dev, "%s: Compression output larger than input\n", | |
236 | __func__); | |
237 | return -ENOSPC; | |
238 | case 66: /* Input data contains an illegal template field */ | |
239 | case 67: /* Template indicates data past the end of the input stream */ | |
240 | dev_dbg(dev, "%s: Bad data for decompression (code:%d)\n", | |
241 | __func__, csb->completion_code); | |
242 | return -EINVAL; | |
243 | default: | |
244 | dev_dbg(dev, "%s: Unspecified error (code:%d)\n", | |
245 | __func__, csb->completion_code); | |
246 | return -EIO; | |
247 | } | |
248 | ||
249 | /* Hardware sanity check */ | |
250 | if (!NX842_CSBCPB_CE2(csb->completion_extension)) { | |
251 | dev_err(dev, "%s: No error returned by hardware, but " | |
252 | "data returned is unusable, contact support.\n" | |
253 | "(Additional info: csbcbp->processed bytes " | |
254 | "does not specify processed bytes for the " | |
255 | "target buffer.)\n", __func__); | |
256 | return -EIO; | |
257 | } | |
258 | ||
259 | return 0; | |
260 | } | |
261 | ||
262 | /** | |
7011a122 | 263 | * nx842_pseries_compress - Compress data using the 842 algorithm |
0e16aafb SJ |
264 | * |
265 | * Compression provide by the NX842 coprocessor on IBM Power systems. | |
266 | * The input buffer is compressed and the result is stored in the | |
267 | * provided output buffer. | |
268 | * | |
269 | * Upon return from this function @outlen contains the length of the | |
270 | * compressed data. If there is an error then @outlen will be 0 and an | |
271 | * error will be specified by the return code from this function. | |
272 | * | |
b8e04187 DS |
273 | * @in: Pointer to input buffer |
274 | * @inlen: Length of input buffer | |
0e16aafb SJ |
275 | * @out: Pointer to output buffer |
276 | * @outlen: Length of output buffer | |
277 | * @wrkmem: ptr to buffer for working memory, size determined by | |
2c6f6eab | 278 | * nx842_pseries_driver.workmem_size |
0e16aafb SJ |
279 | * |
280 | * Returns: | |
281 | * 0 Success, output of length @outlen stored in the buffer at @out | |
282 | * -ENOMEM Unable to allocate internal buffers | |
283 | * -ENOSPC Output buffer is to small | |
0e16aafb SJ |
284 | * -EIO Internal error |
285 | * -ENODEV Hardware unavailable | |
286 | */ | |
7011a122 DS |
287 | static int nx842_pseries_compress(const unsigned char *in, unsigned int inlen, |
288 | unsigned char *out, unsigned int *outlen, | |
289 | void *wmem) | |
0e16aafb | 290 | { |
0e16aafb SJ |
291 | struct nx842_devdata *local_devdata; |
292 | struct device *dev = NULL; | |
293 | struct nx842_workmem *workmem; | |
294 | struct nx842_scatterlist slin, slout; | |
295 | struct nx_csbcpb *csbcpb; | |
b8e04187 DS |
296 | int ret = 0, max_sync_size; |
297 | unsigned long inbuf, outbuf; | |
0e16aafb SJ |
298 | struct vio_pfo_op op = { |
299 | .done = NULL, | |
300 | .handle = 0, | |
301 | .timeout = 0, | |
302 | }; | |
b8e04187 | 303 | unsigned long start = get_tb(); |
0e16aafb | 304 | |
0e16aafb | 305 | inbuf = (unsigned long)in; |
b8e04187 DS |
306 | if (check_constraints(inbuf, &inlen, true)) |
307 | return -EINVAL; | |
308 | ||
309 | outbuf = (unsigned long)out; | |
310 | if (check_constraints(outbuf, outlen, false)) | |
0e16aafb SJ |
311 | return -EINVAL; |
312 | ||
313 | rcu_read_lock(); | |
314 | local_devdata = rcu_dereference(devdata); | |
315 | if (!local_devdata || !local_devdata->dev) { | |
316 | rcu_read_unlock(); | |
317 | return -ENODEV; | |
318 | } | |
319 | max_sync_size = local_devdata->max_sync_size; | |
320 | dev = local_devdata->dev; | |
321 | ||
0e16aafb | 322 | /* Init scatterlist */ |
b8e04187 | 323 | workmem = PTR_ALIGN(wmem, WORKMEM_ALIGN); |
0e16aafb SJ |
324 | slin.entries = (struct nx842_slentry *)workmem->slin; |
325 | slout.entries = (struct nx842_slentry *)workmem->slout; | |
326 | ||
327 | /* Init operation */ | |
328 | op.flags = NX842_OP_COMPRESS; | |
329 | csbcpb = &workmem->csbcpb; | |
330 | memset(csbcpb, 0, sizeof(*csbcpb)); | |
0ba3e101 | 331 | op.csbcpb = nx842_get_pa(csbcpb); |
0e16aafb | 332 | |
b8e04187 DS |
333 | if ((inbuf & NX842_HW_PAGE_MASK) == |
334 | ((inbuf + inlen - 1) & NX842_HW_PAGE_MASK)) { | |
335 | /* Create direct DDE */ | |
336 | op.in = nx842_get_pa((void *)inbuf); | |
337 | op.inlen = inlen; | |
338 | } else { | |
339 | /* Create indirect DDE (scatterlist) */ | |
340 | nx842_build_scatterlist(inbuf, inlen, &slin); | |
341 | op.in = nx842_get_pa(slin.entries); | |
342 | op.inlen = -nx842_get_scatterlist_size(&slin); | |
343 | } | |
0e16aafb | 344 | |
b8e04187 DS |
345 | if ((outbuf & NX842_HW_PAGE_MASK) == |
346 | ((outbuf + *outlen - 1) & NX842_HW_PAGE_MASK)) { | |
347 | /* Create direct DDE */ | |
348 | op.out = nx842_get_pa((void *)outbuf); | |
349 | op.outlen = *outlen; | |
350 | } else { | |
351 | /* Create indirect DDE (scatterlist) */ | |
352 | nx842_build_scatterlist(outbuf, *outlen, &slout); | |
353 | op.out = nx842_get_pa(slout.entries); | |
0e16aafb | 354 | op.outlen = -nx842_get_scatterlist_size(&slout); |
b8e04187 | 355 | } |
0e16aafb | 356 | |
c47d6302 DS |
357 | dev_dbg(dev, "%s: op.in %lx op.inlen %ld op.out %lx op.outlen %ld\n", |
358 | __func__, (unsigned long)op.in, (long)op.inlen, | |
359 | (unsigned long)op.out, (long)op.outlen); | |
360 | ||
b8e04187 DS |
361 | /* Send request to pHyp */ |
362 | ret = vio_h_cop_sync(local_devdata->vdev, &op); | |
0e16aafb | 363 | |
b8e04187 DS |
364 | /* Check for pHyp error */ |
365 | if (ret) { | |
366 | dev_dbg(dev, "%s: vio_h_cop_sync error (ret=%d, hret=%ld)\n", | |
367 | __func__, ret, op.hcall_err); | |
368 | ret = -EIO; | |
369 | goto unlock; | |
0e16aafb SJ |
370 | } |
371 | ||
b8e04187 DS |
372 | /* Check for hardware error */ |
373 | ret = nx842_validate_result(dev, &csbcpb->csb); | |
374 | if (ret) | |
375 | goto unlock; | |
376 | ||
c47d6302 | 377 | *outlen = be32_to_cpu(csbcpb->csb.processed_byte_count); |
b8e04187 | 378 | dev_dbg(dev, "%s: processed_bytes=%d\n", __func__, *outlen); |
0e16aafb SJ |
379 | |
380 | unlock: | |
381 | if (ret) | |
382 | nx842_inc_comp_failed(local_devdata); | |
383 | else { | |
384 | nx842_inc_comp_complete(local_devdata); | |
385 | ibm_nx842_incr_hist(local_devdata->counters->comp_times, | |
b8e04187 | 386 | (get_tb() - start) / tb_ticks_per_usec); |
0e16aafb SJ |
387 | } |
388 | rcu_read_unlock(); | |
389 | return ret; | |
390 | } | |
0e16aafb | 391 | |
0e16aafb | 392 | /** |
7011a122 | 393 | * nx842_pseries_decompress - Decompress data using the 842 algorithm |
0e16aafb SJ |
394 | * |
395 | * Decompression provide by the NX842 coprocessor on IBM Power systems. | |
396 | * The input buffer is decompressed and the result is stored in the | |
397 | * provided output buffer. The size allocated to the output buffer is | |
398 | * provided by the caller of this function in @outlen. Upon return from | |
399 | * this function @outlen contains the length of the decompressed data. | |
400 | * If there is an error then @outlen will be 0 and an error will be | |
401 | * specified by the return code from this function. | |
402 | * | |
b8e04187 | 403 | * @in: Pointer to input buffer |
0e16aafb | 404 | * @inlen: Length of input buffer |
b8e04187 DS |
405 | * @out: Pointer to output buffer |
406 | * @outlen: Length of output buffer | |
0e16aafb | 407 | * @wrkmem: ptr to buffer for working memory, size determined by |
2c6f6eab | 408 | * nx842_pseries_driver.workmem_size |
0e16aafb SJ |
409 | * |
410 | * Returns: | |
411 | * 0 Success, output of length @outlen stored in the buffer at @out | |
412 | * -ENODEV Hardware decompression device is unavailable | |
413 | * -ENOMEM Unable to allocate internal buffers | |
414 | * -ENOSPC Output buffer is to small | |
415 | * -EINVAL Bad input data encountered when attempting decompress | |
416 | * -EIO Internal error | |
417 | */ | |
7011a122 DS |
418 | static int nx842_pseries_decompress(const unsigned char *in, unsigned int inlen, |
419 | unsigned char *out, unsigned int *outlen, | |
420 | void *wmem) | |
0e16aafb | 421 | { |
0e16aafb SJ |
422 | struct nx842_devdata *local_devdata; |
423 | struct device *dev = NULL; | |
424 | struct nx842_workmem *workmem; | |
425 | struct nx842_scatterlist slin, slout; | |
426 | struct nx_csbcpb *csbcpb; | |
b8e04187 | 427 | int ret = 0, max_sync_size; |
0e16aafb SJ |
428 | unsigned long inbuf, outbuf; |
429 | struct vio_pfo_op op = { | |
430 | .done = NULL, | |
431 | .handle = 0, | |
432 | .timeout = 0, | |
433 | }; | |
b8e04187 | 434 | unsigned long start = get_tb(); |
0e16aafb SJ |
435 | |
436 | /* Ensure page alignment and size */ | |
b8e04187 DS |
437 | inbuf = (unsigned long)in; |
438 | if (check_constraints(inbuf, &inlen, true)) | |
439 | return -EINVAL; | |
440 | ||
0e16aafb | 441 | outbuf = (unsigned long)out; |
b8e04187 | 442 | if (check_constraints(outbuf, outlen, false)) |
0e16aafb SJ |
443 | return -EINVAL; |
444 | ||
445 | rcu_read_lock(); | |
446 | local_devdata = rcu_dereference(devdata); | |
b8e04187 DS |
447 | if (!local_devdata || !local_devdata->dev) { |
448 | rcu_read_unlock(); | |
449 | return -ENODEV; | |
0e16aafb | 450 | } |
b8e04187 DS |
451 | max_sync_size = local_devdata->max_sync_size; |
452 | dev = local_devdata->dev; | |
453 | ||
454 | workmem = PTR_ALIGN(wmem, WORKMEM_ALIGN); | |
0e16aafb SJ |
455 | |
456 | /* Init scatterlist */ | |
457 | slin.entries = (struct nx842_slentry *)workmem->slin; | |
458 | slout.entries = (struct nx842_slentry *)workmem->slout; | |
459 | ||
460 | /* Init operation */ | |
461 | op.flags = NX842_OP_DECOMPRESS; | |
462 | csbcpb = &workmem->csbcpb; | |
463 | memset(csbcpb, 0, sizeof(*csbcpb)); | |
0ba3e101 | 464 | op.csbcpb = nx842_get_pa(csbcpb); |
0e16aafb | 465 | |
b8e04187 DS |
466 | if ((inbuf & NX842_HW_PAGE_MASK) == |
467 | ((inbuf + inlen - 1) & NX842_HW_PAGE_MASK)) { | |
468 | /* Create direct DDE */ | |
469 | op.in = nx842_get_pa((void *)inbuf); | |
470 | op.inlen = inlen; | |
471 | } else { | |
472 | /* Create indirect DDE (scatterlist) */ | |
473 | nx842_build_scatterlist(inbuf, inlen, &slin); | |
474 | op.in = nx842_get_pa(slin.entries); | |
475 | op.inlen = -nx842_get_scatterlist_size(&slin); | |
476 | } | |
0e16aafb | 477 | |
b8e04187 DS |
478 | if ((outbuf & NX842_HW_PAGE_MASK) == |
479 | ((outbuf + *outlen - 1) & NX842_HW_PAGE_MASK)) { | |
480 | /* Create direct DDE */ | |
481 | op.out = nx842_get_pa((void *)outbuf); | |
482 | op.outlen = *outlen; | |
483 | } else { | |
484 | /* Create indirect DDE (scatterlist) */ | |
485 | nx842_build_scatterlist(outbuf, *outlen, &slout); | |
486 | op.out = nx842_get_pa(slout.entries); | |
487 | op.outlen = -nx842_get_scatterlist_size(&slout); | |
488 | } | |
0e16aafb | 489 | |
c47d6302 DS |
490 | dev_dbg(dev, "%s: op.in %lx op.inlen %ld op.out %lx op.outlen %ld\n", |
491 | __func__, (unsigned long)op.in, (long)op.inlen, | |
492 | (unsigned long)op.out, (long)op.outlen); | |
493 | ||
b8e04187 DS |
494 | /* Send request to pHyp */ |
495 | ret = vio_h_cop_sync(local_devdata->vdev, &op); | |
0e16aafb | 496 | |
b8e04187 DS |
497 | /* Check for pHyp error */ |
498 | if (ret) { | |
499 | dev_dbg(dev, "%s: vio_h_cop_sync error (ret=%d, hret=%ld)\n", | |
500 | __func__, ret, op.hcall_err); | |
501 | goto unlock; | |
0e16aafb SJ |
502 | } |
503 | ||
b8e04187 DS |
504 | /* Check for hardware error */ |
505 | ret = nx842_validate_result(dev, &csbcpb->csb); | |
506 | if (ret) | |
507 | goto unlock; | |
508 | ||
c47d6302 | 509 | *outlen = be32_to_cpu(csbcpb->csb.processed_byte_count); |
0e16aafb SJ |
510 | |
511 | unlock: | |
512 | if (ret) | |
513 | /* decompress fail */ | |
514 | nx842_inc_decomp_failed(local_devdata); | |
515 | else { | |
0e16aafb SJ |
516 | nx842_inc_decomp_complete(local_devdata); |
517 | ibm_nx842_incr_hist(local_devdata->counters->decomp_times, | |
b8e04187 | 518 | (get_tb() - start) / tb_ticks_per_usec); |
0e16aafb SJ |
519 | } |
520 | ||
521 | rcu_read_unlock(); | |
522 | return ret; | |
523 | } | |
0e16aafb SJ |
524 | |
525 | /** | |
526 | * nx842_OF_set_defaults -- Set default (disabled) values for devdata | |
527 | * | |
528 | * @devdata - struct nx842_devdata to update | |
529 | * | |
530 | * Returns: | |
531 | * 0 on success | |
532 | * -ENOENT if @devdata ptr is NULL | |
533 | */ | |
534 | static int nx842_OF_set_defaults(struct nx842_devdata *devdata) | |
535 | { | |
536 | if (devdata) { | |
537 | devdata->max_sync_size = 0; | |
538 | devdata->max_sync_sg = 0; | |
539 | devdata->max_sg_len = 0; | |
540 | devdata->status = UNAVAILABLE; | |
541 | return 0; | |
542 | } else | |
543 | return -ENOENT; | |
544 | } | |
545 | ||
546 | /** | |
547 | * nx842_OF_upd_status -- Update the device info from OF status prop | |
548 | * | |
549 | * The status property indicates if the accelerator is enabled. If the | |
550 | * device is in the OF tree it indicates that the hardware is present. | |
551 | * The status field indicates if the device is enabled when the status | |
552 | * is 'okay'. Otherwise the device driver will be disabled. | |
553 | * | |
554 | * @devdata - struct nx842_devdata to update | |
555 | * @prop - struct property point containing the maxsyncop for the update | |
556 | * | |
557 | * Returns: | |
558 | * 0 - Device is available | |
fa9a9a08 | 559 | * -ENODEV - Device is not available |
0e16aafb SJ |
560 | */ |
561 | static int nx842_OF_upd_status(struct nx842_devdata *devdata, | |
562 | struct property *prop) { | |
563 | int ret = 0; | |
564 | const char *status = (const char *)prop->value; | |
565 | ||
566 | if (!strncmp(status, "okay", (size_t)prop->length)) { | |
567 | devdata->status = AVAILABLE; | |
568 | } else { | |
7abd75bf NA |
569 | /* |
570 | * Caller will log that the device is disabled, so only | |
571 | * output if there is an unexpected status. | |
572 | */ | |
573 | if (strncmp(status, "disabled", (size_t)prop->length)) { | |
574 | dev_info(devdata->dev, "%s: status '%s' is not 'okay'\n", | |
0e16aafb | 575 | __func__, status); |
7abd75bf | 576 | } |
0e16aafb | 577 | devdata->status = UNAVAILABLE; |
fa9a9a08 | 578 | ret = -ENODEV; |
0e16aafb SJ |
579 | } |
580 | ||
581 | return ret; | |
582 | } | |
583 | ||
584 | /** | |
585 | * nx842_OF_upd_maxsglen -- Update the device info from OF maxsglen prop | |
586 | * | |
587 | * Definition of the 'ibm,max-sg-len' OF property: | |
588 | * This field indicates the maximum byte length of a scatter list | |
589 | * for the platform facility. It is a single cell encoded as with encode-int. | |
590 | * | |
591 | * Example: | |
592 | * # od -x ibm,max-sg-len | |
593 | * 0000000 0000 0ff0 | |
594 | * | |
595 | * In this example, the maximum byte length of a scatter list is | |
596 | * 0x0ff0 (4,080). | |
597 | * | |
598 | * @devdata - struct nx842_devdata to update | |
599 | * @prop - struct property point containing the maxsyncop for the update | |
600 | * | |
601 | * Returns: | |
602 | * 0 on success | |
603 | * -EINVAL on failure | |
604 | */ | |
605 | static int nx842_OF_upd_maxsglen(struct nx842_devdata *devdata, | |
606 | struct property *prop) { | |
607 | int ret = 0; | |
c47d6302 | 608 | const unsigned int maxsglen = of_read_number(prop->value, 1); |
0e16aafb | 609 | |
c47d6302 | 610 | if (prop->length != sizeof(maxsglen)) { |
0e16aafb SJ |
611 | dev_err(devdata->dev, "%s: unexpected format for ibm,max-sg-len property\n", __func__); |
612 | dev_dbg(devdata->dev, "%s: ibm,max-sg-len is %d bytes long, expected %lu bytes\n", __func__, | |
c47d6302 | 613 | prop->length, sizeof(maxsglen)); |
0e16aafb SJ |
614 | ret = -EINVAL; |
615 | } else { | |
c47d6302 DS |
616 | devdata->max_sg_len = min_t(unsigned int, |
617 | maxsglen, NX842_HW_PAGE_SIZE); | |
0e16aafb SJ |
618 | } |
619 | ||
620 | return ret; | |
621 | } | |
622 | ||
623 | /** | |
624 | * nx842_OF_upd_maxsyncop -- Update the device info from OF maxsyncop prop | |
625 | * | |
626 | * Definition of the 'ibm,max-sync-cop' OF property: | |
627 | * Two series of cells. The first series of cells represents the maximums | |
628 | * that can be synchronously compressed. The second series of cells | |
629 | * represents the maximums that can be synchronously decompressed. | |
630 | * 1. The first cell in each series contains the count of the number of | |
631 | * data length, scatter list elements pairs that follow – each being | |
632 | * of the form | |
633 | * a. One cell data byte length | |
634 | * b. One cell total number of scatter list elements | |
635 | * | |
636 | * Example: | |
637 | * # od -x ibm,max-sync-cop | |
638 | * 0000000 0000 0001 0000 1000 0000 01fe 0000 0001 | |
639 | * 0000020 0000 1000 0000 01fe | |
640 | * | |
641 | * In this example, compression supports 0x1000 (4,096) data byte length | |
642 | * and 0x1fe (510) total scatter list elements. Decompression supports | |
643 | * 0x1000 (4,096) data byte length and 0x1f3 (510) total scatter list | |
644 | * elements. | |
645 | * | |
646 | * @devdata - struct nx842_devdata to update | |
647 | * @prop - struct property point containing the maxsyncop for the update | |
648 | * | |
649 | * Returns: | |
650 | * 0 on success | |
651 | * -EINVAL on failure | |
652 | */ | |
653 | static int nx842_OF_upd_maxsyncop(struct nx842_devdata *devdata, | |
654 | struct property *prop) { | |
655 | int ret = 0; | |
c47d6302 DS |
656 | unsigned int comp_data_limit, decomp_data_limit; |
657 | unsigned int comp_sg_limit, decomp_sg_limit; | |
0e16aafb | 658 | const struct maxsynccop_t { |
c47d6302 DS |
659 | __be32 comp_elements; |
660 | __be32 comp_data_limit; | |
661 | __be32 comp_sg_limit; | |
662 | __be32 decomp_elements; | |
663 | __be32 decomp_data_limit; | |
664 | __be32 decomp_sg_limit; | |
0e16aafb SJ |
665 | } *maxsynccop; |
666 | ||
667 | if (prop->length != sizeof(*maxsynccop)) { | |
668 | dev_err(devdata->dev, "%s: unexpected format for ibm,max-sync-cop property\n", __func__); | |
669 | dev_dbg(devdata->dev, "%s: ibm,max-sync-cop is %d bytes long, expected %lu bytes\n", __func__, prop->length, | |
670 | sizeof(*maxsynccop)); | |
671 | ret = -EINVAL; | |
672 | goto out; | |
673 | } | |
674 | ||
675 | maxsynccop = (const struct maxsynccop_t *)prop->value; | |
c47d6302 DS |
676 | comp_data_limit = be32_to_cpu(maxsynccop->comp_data_limit); |
677 | comp_sg_limit = be32_to_cpu(maxsynccop->comp_sg_limit); | |
678 | decomp_data_limit = be32_to_cpu(maxsynccop->decomp_data_limit); | |
679 | decomp_sg_limit = be32_to_cpu(maxsynccop->decomp_sg_limit); | |
0e16aafb SJ |
680 | |
681 | /* Use one limit rather than separate limits for compression and | |
682 | * decompression. Set a maximum for this so as not to exceed the | |
683 | * size that the header can support and round the value down to | |
684 | * the hardware page size (4K) */ | |
c47d6302 | 685 | devdata->max_sync_size = min(comp_data_limit, decomp_data_limit); |
0e16aafb SJ |
686 | |
687 | devdata->max_sync_size = min_t(unsigned int, devdata->max_sync_size, | |
b8e04187 | 688 | 65536); |
0e16aafb | 689 | |
b8e04187 | 690 | if (devdata->max_sync_size < 4096) { |
0e16aafb SJ |
691 | dev_err(devdata->dev, "%s: hardware max data size (%u) is " |
692 | "less than the driver minimum, unable to use " | |
693 | "the hardware device\n", | |
694 | __func__, devdata->max_sync_size); | |
695 | ret = -EINVAL; | |
696 | goto out; | |
697 | } | |
698 | ||
959e6659 DS |
699 | nx842_pseries_constraints.maximum = devdata->max_sync_size; |
700 | ||
c47d6302 | 701 | devdata->max_sync_sg = min(comp_sg_limit, decomp_sg_limit); |
0e16aafb SJ |
702 | if (devdata->max_sync_sg < 1) { |
703 | dev_err(devdata->dev, "%s: hardware max sg size (%u) is " | |
704 | "less than the driver minimum, unable to use " | |
705 | "the hardware device\n", | |
706 | __func__, devdata->max_sync_sg); | |
707 | ret = -EINVAL; | |
708 | goto out; | |
709 | } | |
710 | ||
711 | out: | |
712 | return ret; | |
713 | } | |
714 | ||
715 | /** | |
716 | * | |
717 | * nx842_OF_upd -- Handle OF properties updates for the device. | |
718 | * | |
719 | * Set all properties from the OF tree. Optionally, a new property | |
720 | * can be provided by the @new_prop pointer to overwrite an existing value. | |
721 | * The device will remain disabled until all values are valid, this function | |
722 | * will return an error for updates unless all values are valid. | |
723 | * | |
724 | * @new_prop: If not NULL, this property is being updated. If NULL, update | |
725 | * all properties from the current values in the OF tree. | |
726 | * | |
727 | * Returns: | |
728 | * 0 - Success | |
729 | * -ENOMEM - Could not allocate memory for new devdata structure | |
730 | * -EINVAL - property value not found, new_prop is not a recognized | |
731 | * property for the device or property value is not valid. | |
732 | * -ENODEV - Device is not available | |
733 | */ | |
734 | static int nx842_OF_upd(struct property *new_prop) | |
735 | { | |
736 | struct nx842_devdata *old_devdata = NULL; | |
737 | struct nx842_devdata *new_devdata = NULL; | |
738 | struct device_node *of_node = NULL; | |
739 | struct property *status = NULL; | |
740 | struct property *maxsglen = NULL; | |
741 | struct property *maxsyncop = NULL; | |
742 | int ret = 0; | |
743 | unsigned long flags; | |
744 | ||
745 | spin_lock_irqsave(&devdata_mutex, flags); | |
746 | old_devdata = rcu_dereference_check(devdata, | |
747 | lockdep_is_held(&devdata_mutex)); | |
748 | if (old_devdata) | |
749 | of_node = old_devdata->dev->of_node; | |
750 | ||
751 | if (!old_devdata || !of_node) { | |
752 | pr_err("%s: device is not available\n", __func__); | |
753 | spin_unlock_irqrestore(&devdata_mutex, flags); | |
754 | return -ENODEV; | |
755 | } | |
756 | ||
757 | new_devdata = kzalloc(sizeof(*new_devdata), GFP_NOFS); | |
758 | if (!new_devdata) { | |
759 | dev_err(old_devdata->dev, "%s: Could not allocate memory for device data\n", __func__); | |
760 | ret = -ENOMEM; | |
761 | goto error_out; | |
762 | } | |
763 | ||
764 | memcpy(new_devdata, old_devdata, sizeof(*old_devdata)); | |
765 | new_devdata->counters = old_devdata->counters; | |
766 | ||
767 | /* Set ptrs for existing properties */ | |
768 | status = of_find_property(of_node, "status", NULL); | |
769 | maxsglen = of_find_property(of_node, "ibm,max-sg-len", NULL); | |
770 | maxsyncop = of_find_property(of_node, "ibm,max-sync-cop", NULL); | |
771 | if (!status || !maxsglen || !maxsyncop) { | |
772 | dev_err(old_devdata->dev, "%s: Could not locate device properties\n", __func__); | |
773 | ret = -EINVAL; | |
774 | goto error_out; | |
775 | } | |
776 | ||
259092a3 GL |
777 | /* |
778 | * If this is a property update, there are only certain properties that | |
779 | * we care about. Bail if it isn't in the below list | |
780 | */ | |
781 | if (new_prop && (strncmp(new_prop->name, "status", new_prop->length) || | |
782 | strncmp(new_prop->name, "ibm,max-sg-len", new_prop->length) || | |
783 | strncmp(new_prop->name, "ibm,max-sync-cop", new_prop->length))) | |
784 | goto out; | |
0e16aafb SJ |
785 | |
786 | /* Perform property updates */ | |
787 | ret = nx842_OF_upd_status(new_devdata, status); | |
788 | if (ret) | |
789 | goto error_out; | |
790 | ||
791 | ret = nx842_OF_upd_maxsglen(new_devdata, maxsglen); | |
792 | if (ret) | |
793 | goto error_out; | |
794 | ||
795 | ret = nx842_OF_upd_maxsyncop(new_devdata, maxsyncop); | |
796 | if (ret) | |
797 | goto error_out; | |
798 | ||
799 | out: | |
800 | dev_info(old_devdata->dev, "%s: max_sync_size new:%u old:%u\n", | |
801 | __func__, new_devdata->max_sync_size, | |
802 | old_devdata->max_sync_size); | |
803 | dev_info(old_devdata->dev, "%s: max_sync_sg new:%u old:%u\n", | |
804 | __func__, new_devdata->max_sync_sg, | |
805 | old_devdata->max_sync_sg); | |
806 | dev_info(old_devdata->dev, "%s: max_sg_len new:%u old:%u\n", | |
807 | __func__, new_devdata->max_sg_len, | |
808 | old_devdata->max_sg_len); | |
809 | ||
810 | rcu_assign_pointer(devdata, new_devdata); | |
811 | spin_unlock_irqrestore(&devdata_mutex, flags); | |
812 | synchronize_rcu(); | |
813 | dev_set_drvdata(new_devdata->dev, new_devdata); | |
814 | kfree(old_devdata); | |
815 | return 0; | |
816 | ||
817 | error_out: | |
818 | if (new_devdata) { | |
819 | dev_info(old_devdata->dev, "%s: device disabled\n", __func__); | |
820 | nx842_OF_set_defaults(new_devdata); | |
821 | rcu_assign_pointer(devdata, new_devdata); | |
822 | spin_unlock_irqrestore(&devdata_mutex, flags); | |
823 | synchronize_rcu(); | |
824 | dev_set_drvdata(new_devdata->dev, new_devdata); | |
825 | kfree(old_devdata); | |
826 | } else { | |
827 | dev_err(old_devdata->dev, "%s: could not update driver from hardware\n", __func__); | |
828 | spin_unlock_irqrestore(&devdata_mutex, flags); | |
829 | } | |
830 | ||
831 | if (!ret) | |
832 | ret = -EINVAL; | |
833 | return ret; | |
834 | } | |
835 | ||
836 | /** | |
837 | * nx842_OF_notifier - Process updates to OF properties for the device | |
838 | * | |
839 | * @np: notifier block | |
840 | * @action: notifier action | |
841 | * @update: struct pSeries_reconfig_prop_update pointer if action is | |
842 | * PSERIES_UPDATE_PROPERTY | |
843 | * | |
844 | * Returns: | |
845 | * NOTIFY_OK on success | |
846 | * NOTIFY_BAD encoded with error number on failure, use | |
847 | * notifier_to_errno() to decode this value | |
848 | */ | |
1cf3d8b3 | 849 | static int nx842_OF_notifier(struct notifier_block *np, unsigned long action, |
f5242e5a | 850 | void *data) |
0e16aafb | 851 | { |
f5242e5a | 852 | struct of_reconfig_data *upd = data; |
0e16aafb SJ |
853 | struct nx842_devdata *local_devdata; |
854 | struct device_node *node = NULL; | |
855 | ||
0e16aafb SJ |
856 | rcu_read_lock(); |
857 | local_devdata = rcu_dereference(devdata); | |
858 | if (local_devdata) | |
859 | node = local_devdata->dev->of_node; | |
860 | ||
861 | if (local_devdata && | |
1cf3d8b3 NF |
862 | action == OF_RECONFIG_UPDATE_PROPERTY && |
863 | !strcmp(upd->dn->name, node->name)) { | |
0e16aafb | 864 | rcu_read_unlock(); |
1cf3d8b3 | 865 | nx842_OF_upd(upd->prop); |
0e16aafb SJ |
866 | } else |
867 | rcu_read_unlock(); | |
868 | ||
869 | return NOTIFY_OK; | |
870 | } | |
871 | ||
872 | static struct notifier_block nx842_of_nb = { | |
873 | .notifier_call = nx842_OF_notifier, | |
874 | }; | |
875 | ||
876 | #define nx842_counter_read(_name) \ | |
877 | static ssize_t nx842_##_name##_show(struct device *dev, \ | |
878 | struct device_attribute *attr, \ | |
879 | char *buf) { \ | |
880 | struct nx842_devdata *local_devdata; \ | |
881 | int p = 0; \ | |
882 | rcu_read_lock(); \ | |
883 | local_devdata = rcu_dereference(devdata); \ | |
884 | if (local_devdata) \ | |
885 | p = snprintf(buf, PAGE_SIZE, "%ld\n", \ | |
886 | atomic64_read(&local_devdata->counters->_name)); \ | |
887 | rcu_read_unlock(); \ | |
888 | return p; \ | |
889 | } | |
890 | ||
891 | #define NX842DEV_COUNTER_ATTR_RO(_name) \ | |
892 | nx842_counter_read(_name); \ | |
893 | static struct device_attribute dev_attr_##_name = __ATTR(_name, \ | |
894 | 0444, \ | |
895 | nx842_##_name##_show,\ | |
896 | NULL); | |
897 | ||
898 | NX842DEV_COUNTER_ATTR_RO(comp_complete); | |
899 | NX842DEV_COUNTER_ATTR_RO(comp_failed); | |
900 | NX842DEV_COUNTER_ATTR_RO(decomp_complete); | |
901 | NX842DEV_COUNTER_ATTR_RO(decomp_failed); | |
902 | NX842DEV_COUNTER_ATTR_RO(swdecomp); | |
903 | ||
904 | static ssize_t nx842_timehist_show(struct device *, | |
905 | struct device_attribute *, char *); | |
906 | ||
907 | static struct device_attribute dev_attr_comp_times = __ATTR(comp_times, 0444, | |
908 | nx842_timehist_show, NULL); | |
909 | static struct device_attribute dev_attr_decomp_times = __ATTR(decomp_times, | |
910 | 0444, nx842_timehist_show, NULL); | |
911 | ||
912 | static ssize_t nx842_timehist_show(struct device *dev, | |
913 | struct device_attribute *attr, char *buf) { | |
914 | char *p = buf; | |
915 | struct nx842_devdata *local_devdata; | |
916 | atomic64_t *times; | |
917 | int bytes_remain = PAGE_SIZE; | |
918 | int bytes; | |
919 | int i; | |
920 | ||
921 | rcu_read_lock(); | |
922 | local_devdata = rcu_dereference(devdata); | |
923 | if (!local_devdata) { | |
924 | rcu_read_unlock(); | |
925 | return 0; | |
926 | } | |
927 | ||
928 | if (attr == &dev_attr_comp_times) | |
929 | times = local_devdata->counters->comp_times; | |
930 | else if (attr == &dev_attr_decomp_times) | |
931 | times = local_devdata->counters->decomp_times; | |
932 | else { | |
933 | rcu_read_unlock(); | |
934 | return 0; | |
935 | } | |
936 | ||
937 | for (i = 0; i < (NX842_HIST_SLOTS - 2); i++) { | |
938 | bytes = snprintf(p, bytes_remain, "%u-%uus:\t%ld\n", | |
939 | i ? (2<<(i-1)) : 0, (2<<i)-1, | |
940 | atomic64_read(×[i])); | |
941 | bytes_remain -= bytes; | |
942 | p += bytes; | |
943 | } | |
944 | /* The last bucket holds everything over | |
945 | * 2<<(NX842_HIST_SLOTS - 2) us */ | |
946 | bytes = snprintf(p, bytes_remain, "%uus - :\t%ld\n", | |
947 | 2<<(NX842_HIST_SLOTS - 2), | |
948 | atomic64_read(×[(NX842_HIST_SLOTS - 1)])); | |
949 | p += bytes; | |
950 | ||
951 | rcu_read_unlock(); | |
952 | return p - buf; | |
953 | } | |
954 | ||
955 | static struct attribute *nx842_sysfs_entries[] = { | |
956 | &dev_attr_comp_complete.attr, | |
957 | &dev_attr_comp_failed.attr, | |
958 | &dev_attr_decomp_complete.attr, | |
959 | &dev_attr_decomp_failed.attr, | |
960 | &dev_attr_swdecomp.attr, | |
961 | &dev_attr_comp_times.attr, | |
962 | &dev_attr_decomp_times.attr, | |
963 | NULL, | |
964 | }; | |
965 | ||
966 | static struct attribute_group nx842_attribute_group = { | |
967 | .name = NULL, /* put in device directory */ | |
968 | .attrs = nx842_sysfs_entries, | |
969 | }; | |
970 | ||
7011a122 | 971 | static struct nx842_driver nx842_pseries_driver = { |
3e648cbe | 972 | .name = KBUILD_MODNAME, |
7011a122 | 973 | .owner = THIS_MODULE, |
2c6f6eab | 974 | .workmem_size = sizeof(struct nx842_workmem), |
959e6659 | 975 | .constraints = &nx842_pseries_constraints, |
7011a122 DS |
976 | .compress = nx842_pseries_compress, |
977 | .decompress = nx842_pseries_decompress, | |
978 | }; | |
979 | ||
039af967 DS |
980 | static int nx842_probe(struct vio_dev *viodev, |
981 | const struct vio_device_id *id) | |
0e16aafb SJ |
982 | { |
983 | struct nx842_devdata *old_devdata, *new_devdata = NULL; | |
984 | unsigned long flags; | |
985 | int ret = 0; | |
986 | ||
987 | spin_lock_irqsave(&devdata_mutex, flags); | |
988 | old_devdata = rcu_dereference_check(devdata, | |
989 | lockdep_is_held(&devdata_mutex)); | |
990 | ||
991 | if (old_devdata && old_devdata->vdev != NULL) { | |
992 | dev_err(&viodev->dev, "%s: Attempt to register more than one instance of the hardware\n", __func__); | |
993 | ret = -1; | |
994 | goto error_unlock; | |
995 | } | |
996 | ||
997 | dev_set_drvdata(&viodev->dev, NULL); | |
998 | ||
999 | new_devdata = kzalloc(sizeof(*new_devdata), GFP_NOFS); | |
1000 | if (!new_devdata) { | |
1001 | dev_err(&viodev->dev, "%s: Could not allocate memory for device data\n", __func__); | |
1002 | ret = -ENOMEM; | |
1003 | goto error_unlock; | |
1004 | } | |
1005 | ||
1006 | new_devdata->counters = kzalloc(sizeof(*new_devdata->counters), | |
1007 | GFP_NOFS); | |
1008 | if (!new_devdata->counters) { | |
1009 | dev_err(&viodev->dev, "%s: Could not allocate memory for performance counters\n", __func__); | |
1010 | ret = -ENOMEM; | |
1011 | goto error_unlock; | |
1012 | } | |
1013 | ||
1014 | new_devdata->vdev = viodev; | |
1015 | new_devdata->dev = &viodev->dev; | |
1016 | nx842_OF_set_defaults(new_devdata); | |
1017 | ||
1018 | rcu_assign_pointer(devdata, new_devdata); | |
1019 | spin_unlock_irqrestore(&devdata_mutex, flags); | |
1020 | synchronize_rcu(); | |
1021 | kfree(old_devdata); | |
1022 | ||
1cf3d8b3 | 1023 | of_reconfig_notifier_register(&nx842_of_nb); |
0e16aafb SJ |
1024 | |
1025 | ret = nx842_OF_upd(NULL); | |
1026 | if (ret && ret != -ENODEV) { | |
1027 | dev_err(&viodev->dev, "could not parse device tree. %d\n", ret); | |
1028 | ret = -1; | |
1029 | goto error; | |
1030 | } | |
1031 | ||
1032 | rcu_read_lock(); | |
cda43576 | 1033 | dev_set_drvdata(&viodev->dev, rcu_dereference(devdata)); |
0e16aafb SJ |
1034 | rcu_read_unlock(); |
1035 | ||
1036 | if (sysfs_create_group(&viodev->dev.kobj, &nx842_attribute_group)) { | |
1037 | dev_err(&viodev->dev, "could not create sysfs device attributes\n"); | |
1038 | ret = -1; | |
1039 | goto error; | |
1040 | } | |
1041 | ||
1042 | return 0; | |
1043 | ||
1044 | error_unlock: | |
1045 | spin_unlock_irqrestore(&devdata_mutex, flags); | |
1046 | if (new_devdata) | |
1047 | kfree(new_devdata->counters); | |
1048 | kfree(new_devdata); | |
1049 | error: | |
1050 | return ret; | |
1051 | } | |
1052 | ||
039af967 | 1053 | static int nx842_remove(struct vio_dev *viodev) |
0e16aafb SJ |
1054 | { |
1055 | struct nx842_devdata *old_devdata; | |
1056 | unsigned long flags; | |
1057 | ||
1058 | pr_info("Removing IBM Power 842 compression device\n"); | |
1059 | sysfs_remove_group(&viodev->dev.kobj, &nx842_attribute_group); | |
1060 | ||
1061 | spin_lock_irqsave(&devdata_mutex, flags); | |
1062 | old_devdata = rcu_dereference_check(devdata, | |
1063 | lockdep_is_held(&devdata_mutex)); | |
1cf3d8b3 | 1064 | of_reconfig_notifier_unregister(&nx842_of_nb); |
7ded6e3d | 1065 | RCU_INIT_POINTER(devdata, NULL); |
0e16aafb SJ |
1066 | spin_unlock_irqrestore(&devdata_mutex, flags); |
1067 | synchronize_rcu(); | |
1068 | dev_set_drvdata(&viodev->dev, NULL); | |
1069 | if (old_devdata) | |
1070 | kfree(old_devdata->counters); | |
1071 | kfree(old_devdata); | |
7011a122 | 1072 | |
0e16aafb SJ |
1073 | return 0; |
1074 | } | |
1075 | ||
b8e04187 | 1076 | static struct vio_device_id nx842_vio_driver_ids[] = { |
3e648cbe | 1077 | {"ibm,compression-v1", "ibm,compression"}, |
0e16aafb SJ |
1078 | {"", ""}, |
1079 | }; | |
1080 | ||
b8e04187 | 1081 | static struct vio_driver nx842_vio_driver = { |
3e648cbe | 1082 | .name = KBUILD_MODNAME, |
0e16aafb | 1083 | .probe = nx842_probe, |
039af967 | 1084 | .remove = nx842_remove, |
0e16aafb | 1085 | .get_desired_dma = nx842_get_desired_dma, |
b8e04187 | 1086 | .id_table = nx842_vio_driver_ids, |
0e16aafb SJ |
1087 | }; |
1088 | ||
ec13bcbe | 1089 | static int __init nx842_pseries_init(void) |
0e16aafb SJ |
1090 | { |
1091 | struct nx842_devdata *new_devdata; | |
3e648cbe DS |
1092 | int ret; |
1093 | ||
3e648cbe DS |
1094 | if (!of_find_compatible_node(NULL, NULL, "ibm,compression")) |
1095 | return -ENODEV; | |
1096 | ||
0e16aafb SJ |
1097 | RCU_INIT_POINTER(devdata, NULL); |
1098 | new_devdata = kzalloc(sizeof(*new_devdata), GFP_KERNEL); | |
1099 | if (!new_devdata) { | |
1100 | pr_err("Could not allocate memory for device data\n"); | |
1101 | return -ENOMEM; | |
1102 | } | |
1103 | new_devdata->status = UNAVAILABLE; | |
1104 | RCU_INIT_POINTER(devdata, new_devdata); | |
1105 | ||
3e648cbe DS |
1106 | ret = vio_register_driver(&nx842_vio_driver); |
1107 | if (ret) { | |
1108 | pr_err("Could not register VIO driver %d\n", ret); | |
1109 | ||
1110 | kfree(new_devdata); | |
1111 | return ret; | |
1112 | } | |
1113 | ||
1114 | if (!nx842_platform_driver_set(&nx842_pseries_driver)) { | |
1115 | vio_unregister_driver(&nx842_vio_driver); | |
1116 | kfree(new_devdata); | |
1117 | return -EEXIST; | |
1118 | } | |
1119 | ||
1120 | return 0; | |
0e16aafb SJ |
1121 | } |
1122 | ||
ec13bcbe | 1123 | module_init(nx842_pseries_init); |
0e16aafb | 1124 | |
ec13bcbe | 1125 | static void __exit nx842_pseries_exit(void) |
0e16aafb SJ |
1126 | { |
1127 | struct nx842_devdata *old_devdata; | |
1128 | unsigned long flags; | |
1129 | ||
3e648cbe | 1130 | nx842_platform_driver_unset(&nx842_pseries_driver); |
0e16aafb SJ |
1131 | spin_lock_irqsave(&devdata_mutex, flags); |
1132 | old_devdata = rcu_dereference_check(devdata, | |
1133 | lockdep_is_held(&devdata_mutex)); | |
7ded6e3d | 1134 | RCU_INIT_POINTER(devdata, NULL); |
0e16aafb SJ |
1135 | spin_unlock_irqrestore(&devdata_mutex, flags); |
1136 | synchronize_rcu(); | |
b8e04187 | 1137 | if (old_devdata && old_devdata->dev) |
0e16aafb SJ |
1138 | dev_set_drvdata(old_devdata->dev, NULL); |
1139 | kfree(old_devdata); | |
b8e04187 | 1140 | vio_unregister_driver(&nx842_vio_driver); |
0e16aafb SJ |
1141 | } |
1142 | ||
ec13bcbe | 1143 | module_exit(nx842_pseries_exit); |
0e16aafb | 1144 |