Root/
1 | /* |
2 | * Copyright(c) 2007 Yuri Tikhonov <yur@emcraft.com> |
3 | * Copyright(c) 2009 Intel Corporation |
4 | * |
5 | * This program is free software; you can redistribute it and/or modify it |
6 | * under the terms of the GNU General Public License as published by the Free |
7 | * Software Foundation; either version 2 of the License, or (at your option) |
8 | * any later version. |
9 | * |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for |
13 | * more details. |
14 | * |
15 | * You should have received a copy of the GNU General Public License along with |
16 | * this program; if not, write to the Free Software Foundation, Inc., 59 |
17 | * Temple Place - Suite 330, Boston, MA 02111-1307, USA. |
18 | * |
19 | * The full GNU General Public License is included in this distribution in the |
20 | * file called COPYING. |
21 | */ |
22 | #include <linux/kernel.h> |
23 | #include <linux/interrupt.h> |
24 | #include <linux/dma-mapping.h> |
25 | #include <linux/raid/pq.h> |
26 | #include <linux/async_tx.h> |
27 | |
28 | /** |
29 | * pq_scribble_page - space to hold throwaway P or Q buffer for |
30 | * synchronous gen_syndrome |
31 | */ |
32 | static struct page *pq_scribble_page; |
33 | |
34 | /* the struct page *blocks[] parameter passed to async_gen_syndrome() |
35 | * and async_syndrome_val() contains the 'P' destination address at |
36 | * blocks[disks-2] and the 'Q' destination address at blocks[disks-1] |
37 | * |
38 | * note: these are macros as they are used as lvalues |
39 | */ |
40 | #define P(b, d) (b[d-2]) |
41 | #define Q(b, d) (b[d-1]) |
42 | |
43 | /** |
44 | * do_async_gen_syndrome - asynchronously calculate P and/or Q |
45 | */ |
46 | static __async_inline struct dma_async_tx_descriptor * |
47 | do_async_gen_syndrome(struct dma_chan *chan, struct page **blocks, |
48 | const unsigned char *scfs, unsigned int offset, int disks, |
49 | size_t len, dma_addr_t *dma_src, |
50 | struct async_submit_ctl *submit) |
51 | { |
52 | struct dma_async_tx_descriptor *tx = NULL; |
53 | struct dma_device *dma = chan->device; |
54 | enum dma_ctrl_flags dma_flags = 0; |
55 | enum async_tx_flags flags_orig = submit->flags; |
56 | dma_async_tx_callback cb_fn_orig = submit->cb_fn; |
57 | dma_async_tx_callback cb_param_orig = submit->cb_param; |
58 | int src_cnt = disks - 2; |
59 | unsigned char coefs[src_cnt]; |
60 | unsigned short pq_src_cnt; |
61 | dma_addr_t dma_dest[2]; |
62 | int src_off = 0; |
63 | int idx; |
64 | int i; |
65 | |
66 | /* DMAs use destinations as sources, so use BIDIRECTIONAL mapping */ |
67 | if (P(blocks, disks)) |
68 | dma_dest[0] = dma_map_page(dma->dev, P(blocks, disks), offset, |
69 | len, DMA_BIDIRECTIONAL); |
70 | else |
71 | dma_flags |= DMA_PREP_PQ_DISABLE_P; |
72 | if (Q(blocks, disks)) |
73 | dma_dest[1] = dma_map_page(dma->dev, Q(blocks, disks), offset, |
74 | len, DMA_BIDIRECTIONAL); |
75 | else |
76 | dma_flags |= DMA_PREP_PQ_DISABLE_Q; |
77 | |
78 | /* convert source addresses being careful to collapse 'empty' |
79 | * sources and update the coefficients accordingly |
80 | */ |
81 | for (i = 0, idx = 0; i < src_cnt; i++) { |
82 | if (blocks[i] == NULL) |
83 | continue; |
84 | dma_src[idx] = dma_map_page(dma->dev, blocks[i], offset, len, |
85 | DMA_TO_DEVICE); |
86 | coefs[idx] = scfs[i]; |
87 | idx++; |
88 | } |
89 | src_cnt = idx; |
90 | |
91 | while (src_cnt > 0) { |
92 | submit->flags = flags_orig; |
93 | pq_src_cnt = min(src_cnt, dma_maxpq(dma, dma_flags)); |
94 | /* if we are submitting additional pqs, leave the chain open, |
95 | * clear the callback parameters, and leave the destination |
96 | * buffers mapped |
97 | */ |
98 | if (src_cnt > pq_src_cnt) { |
99 | submit->flags &= ~ASYNC_TX_ACK; |
100 | submit->flags |= ASYNC_TX_FENCE; |
101 | dma_flags |= DMA_COMPL_SKIP_DEST_UNMAP; |
102 | submit->cb_fn = NULL; |
103 | submit->cb_param = NULL; |
104 | } else { |
105 | dma_flags &= ~DMA_COMPL_SKIP_DEST_UNMAP; |
106 | submit->cb_fn = cb_fn_orig; |
107 | submit->cb_param = cb_param_orig; |
108 | if (cb_fn_orig) |
109 | dma_flags |= DMA_PREP_INTERRUPT; |
110 | } |
111 | if (submit->flags & ASYNC_TX_FENCE) |
112 | dma_flags |= DMA_PREP_FENCE; |
113 | |
114 | /* Since we have clobbered the src_list we are committed |
115 | * to doing this asynchronously. Drivers force forward |
116 | * progress in case they can not provide a descriptor |
117 | */ |
118 | for (;;) { |
119 | tx = dma->device_prep_dma_pq(chan, dma_dest, |
120 | &dma_src[src_off], |
121 | pq_src_cnt, |
122 | &coefs[src_off], len, |
123 | dma_flags); |
124 | if (likely(tx)) |
125 | break; |
126 | async_tx_quiesce(&submit->depend_tx); |
127 | dma_async_issue_pending(chan); |
128 | } |
129 | |
130 | async_tx_submit(chan, tx, submit); |
131 | submit->depend_tx = tx; |
132 | |
133 | /* drop completed sources */ |
134 | src_cnt -= pq_src_cnt; |
135 | src_off += pq_src_cnt; |
136 | |
137 | dma_flags |= DMA_PREP_CONTINUE; |
138 | } |
139 | |
140 | return tx; |
141 | } |
142 | |
143 | /** |
144 | * do_sync_gen_syndrome - synchronously calculate a raid6 syndrome |
145 | */ |
146 | static void |
147 | do_sync_gen_syndrome(struct page **blocks, unsigned int offset, int disks, |
148 | size_t len, struct async_submit_ctl *submit) |
149 | { |
150 | void **srcs; |
151 | int i; |
152 | |
153 | if (submit->scribble) |
154 | srcs = submit->scribble; |
155 | else |
156 | srcs = (void **) blocks; |
157 | |
158 | for (i = 0; i < disks; i++) { |
159 | if (blocks[i] == NULL) { |
160 | BUG_ON(i > disks - 3); /* P or Q can't be zero */ |
161 | srcs[i] = (void*)raid6_empty_zero_page; |
162 | } else |
163 | srcs[i] = page_address(blocks[i]) + offset; |
164 | } |
165 | raid6_call.gen_syndrome(disks, len, srcs); |
166 | async_tx_sync_epilog(submit); |
167 | } |
168 | |
169 | /** |
170 | * async_gen_syndrome - asynchronously calculate a raid6 syndrome |
171 | * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1 |
172 | * @offset: common offset into each block (src and dest) to start transaction |
173 | * @disks: number of blocks (including missing P or Q, see below) |
174 | * @len: length of operation in bytes |
175 | * @submit: submission/completion modifiers |
176 | * |
177 | * General note: This routine assumes a field of GF(2^8) with a |
178 | * primitive polynomial of 0x11d and a generator of {02}. |
179 | * |
180 | * 'disks' note: callers can optionally omit either P or Q (but not |
181 | * both) from the calculation by setting blocks[disks-2] or |
182 | * blocks[disks-1] to NULL. When P or Q is omitted 'len' must be <= |
183 | * PAGE_SIZE as a temporary buffer of this size is used in the |
184 | * synchronous path. 'disks' always accounts for both destination |
185 | * buffers. If any source buffers (blocks[i] where i < disks - 2) are |
186 | * set to NULL those buffers will be replaced with the raid6_zero_page |
187 | * in the synchronous path and omitted in the hardware-asynchronous |
188 | * path. |
189 | * |
190 | * 'blocks' note: if submit->scribble is NULL then the contents of |
191 | * 'blocks' may be overwritten to perform address conversions |
192 | * (dma_map_page() or page_address()). |
193 | */ |
194 | struct dma_async_tx_descriptor * |
195 | async_gen_syndrome(struct page **blocks, unsigned int offset, int disks, |
196 | size_t len, struct async_submit_ctl *submit) |
197 | { |
198 | int src_cnt = disks - 2; |
199 | struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ, |
200 | &P(blocks, disks), 2, |
201 | blocks, src_cnt, len); |
202 | struct dma_device *device = chan ? chan->device : NULL; |
203 | dma_addr_t *dma_src = NULL; |
204 | |
205 | BUG_ON(disks > 255 || !(P(blocks, disks) || Q(blocks, disks))); |
206 | |
207 | if (submit->scribble) |
208 | dma_src = submit->scribble; |
209 | else if (sizeof(dma_addr_t) <= sizeof(struct page *)) |
210 | dma_src = (dma_addr_t *) blocks; |
211 | |
212 | if (dma_src && device && |
213 | (src_cnt <= dma_maxpq(device, 0) || |
214 | dma_maxpq(device, DMA_PREP_CONTINUE) > 0) && |
215 | is_dma_pq_aligned(device, offset, 0, len)) { |
216 | /* run the p+q asynchronously */ |
217 | pr_debug("%s: (async) disks: %d len: %zu\n", |
218 | __func__, disks, len); |
219 | return do_async_gen_syndrome(chan, blocks, raid6_gfexp, offset, |
220 | disks, len, dma_src, submit); |
221 | } |
222 | |
223 | /* run the pq synchronously */ |
224 | pr_debug("%s: (sync) disks: %d len: %zu\n", __func__, disks, len); |
225 | |
226 | /* wait for any prerequisite operations */ |
227 | async_tx_quiesce(&submit->depend_tx); |
228 | |
229 | if (!P(blocks, disks)) { |
230 | P(blocks, disks) = pq_scribble_page; |
231 | BUG_ON(len + offset > PAGE_SIZE); |
232 | } |
233 | if (!Q(blocks, disks)) { |
234 | Q(blocks, disks) = pq_scribble_page; |
235 | BUG_ON(len + offset > PAGE_SIZE); |
236 | } |
237 | do_sync_gen_syndrome(blocks, offset, disks, len, submit); |
238 | |
239 | return NULL; |
240 | } |
241 | EXPORT_SYMBOL_GPL(async_gen_syndrome); |
242 | |
243 | static inline struct dma_chan * |
244 | pq_val_chan(struct async_submit_ctl *submit, struct page **blocks, int disks, size_t len) |
245 | { |
246 | #ifdef CONFIG_ASYNC_TX_DISABLE_PQ_VAL_DMA |
247 | return NULL; |
248 | #endif |
249 | return async_tx_find_channel(submit, DMA_PQ_VAL, NULL, 0, blocks, |
250 | disks, len); |
251 | } |
252 | |
253 | /** |
254 | * async_syndrome_val - asynchronously validate a raid6 syndrome |
255 | * @blocks: source blocks from idx 0..disks-3, P @ disks-2 and Q @ disks-1 |
256 | * @offset: common offset into each block (src and dest) to start transaction |
257 | * @disks: number of blocks (including missing P or Q, see below) |
258 | * @len: length of operation in bytes |
259 | * @pqres: on val failure SUM_CHECK_P_RESULT and/or SUM_CHECK_Q_RESULT are set |
260 | * @spare: temporary result buffer for the synchronous case |
261 | * @submit: submission / completion modifiers |
262 | * |
263 | * The same notes from async_gen_syndrome apply to the 'blocks', |
264 | * and 'disks' parameters of this routine. The synchronous path |
265 | * requires a temporary result buffer and submit->scribble to be |
266 | * specified. |
267 | */ |
268 | struct dma_async_tx_descriptor * |
269 | async_syndrome_val(struct page **blocks, unsigned int offset, int disks, |
270 | size_t len, enum sum_check_flags *pqres, struct page *spare, |
271 | struct async_submit_ctl *submit) |
272 | { |
273 | struct dma_chan *chan = pq_val_chan(submit, blocks, disks, len); |
274 | struct dma_device *device = chan ? chan->device : NULL; |
275 | struct dma_async_tx_descriptor *tx; |
276 | unsigned char coefs[disks-2]; |
277 | enum dma_ctrl_flags dma_flags = submit->cb_fn ? DMA_PREP_INTERRUPT : 0; |
278 | dma_addr_t *dma_src = NULL; |
279 | int src_cnt = 0; |
280 | |
281 | BUG_ON(disks < 4); |
282 | |
283 | if (submit->scribble) |
284 | dma_src = submit->scribble; |
285 | else if (sizeof(dma_addr_t) <= sizeof(struct page *)) |
286 | dma_src = (dma_addr_t *) blocks; |
287 | |
288 | if (dma_src && device && disks <= dma_maxpq(device, 0) && |
289 | is_dma_pq_aligned(device, offset, 0, len)) { |
290 | struct device *dev = device->dev; |
291 | dma_addr_t *pq = &dma_src[disks-2]; |
292 | int i; |
293 | |
294 | pr_debug("%s: (async) disks: %d len: %zu\n", |
295 | __func__, disks, len); |
296 | if (!P(blocks, disks)) |
297 | dma_flags |= DMA_PREP_PQ_DISABLE_P; |
298 | else |
299 | pq[0] = dma_map_page(dev, P(blocks, disks), |
300 | offset, len, |
301 | DMA_TO_DEVICE); |
302 | if (!Q(blocks, disks)) |
303 | dma_flags |= DMA_PREP_PQ_DISABLE_Q; |
304 | else |
305 | pq[1] = dma_map_page(dev, Q(blocks, disks), |
306 | offset, len, |
307 | DMA_TO_DEVICE); |
308 | |
309 | if (submit->flags & ASYNC_TX_FENCE) |
310 | dma_flags |= DMA_PREP_FENCE; |
311 | for (i = 0; i < disks-2; i++) |
312 | if (likely(blocks[i])) { |
313 | dma_src[src_cnt] = dma_map_page(dev, blocks[i], |
314 | offset, len, |
315 | DMA_TO_DEVICE); |
316 | coefs[src_cnt] = raid6_gfexp[i]; |
317 | src_cnt++; |
318 | } |
319 | |
320 | for (;;) { |
321 | tx = device->device_prep_dma_pq_val(chan, pq, dma_src, |
322 | src_cnt, |
323 | coefs, |
324 | len, pqres, |
325 | dma_flags); |
326 | if (likely(tx)) |
327 | break; |
328 | async_tx_quiesce(&submit->depend_tx); |
329 | dma_async_issue_pending(chan); |
330 | } |
331 | async_tx_submit(chan, tx, submit); |
332 | |
333 | return tx; |
334 | } else { |
335 | struct page *p_src = P(blocks, disks); |
336 | struct page *q_src = Q(blocks, disks); |
337 | enum async_tx_flags flags_orig = submit->flags; |
338 | dma_async_tx_callback cb_fn_orig = submit->cb_fn; |
339 | void *scribble = submit->scribble; |
340 | void *cb_param_orig = submit->cb_param; |
341 | void *p, *q, *s; |
342 | |
343 | pr_debug("%s: (sync) disks: %d len: %zu\n", |
344 | __func__, disks, len); |
345 | |
346 | /* caller must provide a temporary result buffer and |
347 | * allow the input parameters to be preserved |
348 | */ |
349 | BUG_ON(!spare || !scribble); |
350 | |
351 | /* wait for any prerequisite operations */ |
352 | async_tx_quiesce(&submit->depend_tx); |
353 | |
354 | /* recompute p and/or q into the temporary buffer and then |
355 | * check to see the result matches the current value |
356 | */ |
357 | tx = NULL; |
358 | *pqres = 0; |
359 | if (p_src) { |
360 | init_async_submit(submit, ASYNC_TX_XOR_ZERO_DST, NULL, |
361 | NULL, NULL, scribble); |
362 | tx = async_xor(spare, blocks, offset, disks-2, len, submit); |
363 | async_tx_quiesce(&tx); |
364 | p = page_address(p_src) + offset; |
365 | s = page_address(spare) + offset; |
366 | *pqres |= !!memcmp(p, s, len) << SUM_CHECK_P; |
367 | } |
368 | |
369 | if (q_src) { |
370 | P(blocks, disks) = NULL; |
371 | Q(blocks, disks) = spare; |
372 | init_async_submit(submit, 0, NULL, NULL, NULL, scribble); |
373 | tx = async_gen_syndrome(blocks, offset, disks, len, submit); |
374 | async_tx_quiesce(&tx); |
375 | q = page_address(q_src) + offset; |
376 | s = page_address(spare) + offset; |
377 | *pqres |= !!memcmp(q, s, len) << SUM_CHECK_Q; |
378 | } |
379 | |
380 | /* restore P, Q and submit */ |
381 | P(blocks, disks) = p_src; |
382 | Q(blocks, disks) = q_src; |
383 | |
384 | submit->cb_fn = cb_fn_orig; |
385 | submit->cb_param = cb_param_orig; |
386 | submit->flags = flags_orig; |
387 | async_tx_sync_epilog(submit); |
388 | |
389 | return NULL; |
390 | } |
391 | } |
392 | EXPORT_SYMBOL_GPL(async_syndrome_val); |
393 | |
394 | static int __init async_pq_init(void) |
395 | { |
396 | pq_scribble_page = alloc_page(GFP_KERNEL); |
397 | |
398 | if (pq_scribble_page) |
399 | return 0; |
400 | |
401 | pr_err("%s: failed to allocate required spare page\n", __func__); |
402 | |
403 | return -ENOMEM; |
404 | } |
405 | |
406 | static void __exit async_pq_exit(void) |
407 | { |
408 | put_page(pq_scribble_page); |
409 | } |
410 | |
411 | module_init(async_pq_init); |
412 | module_exit(async_pq_exit); |
413 | |
414 | MODULE_DESCRIPTION("asynchronous raid6 syndrome generation/validation"); |
415 | MODULE_LICENSE("GPL"); |
416 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9