Commit | Line | Data |
---|---|---|
c06b540a TT |
1 | /* |
2 | * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved. | |
3 | * | |
4 | * This software is available to you under a choice of one of two | |
5 | * licenses. You may choose to be licensed under the terms of the GNU | |
6 | * General Public License (GPL) Version 2, available from the file | |
7 | * COPYING in the main directory of this source tree, or the BSD-type | |
8 | * license below: | |
9 | * | |
10 | * Redistribution and use in source and binary forms, with or without | |
11 | * modification, are permitted provided that the following conditions | |
12 | * are met: | |
13 | * | |
14 | * Redistributions of source code must retain the above copyright | |
15 | * notice, this list of conditions and the following disclaimer. | |
16 | * | |
17 | * Redistributions in binary form must reproduce the above | |
18 | * copyright notice, this list of conditions and the following | |
19 | * disclaimer in the documentation and/or other materials provided | |
20 | * with the distribution. | |
21 | * | |
22 | * Neither the name of the Network Appliance, Inc. nor the names of | |
23 | * its contributors may be used to endorse or promote products | |
24 | * derived from this software without specific prior written | |
25 | * permission. | |
26 | * | |
27 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS | |
28 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT | |
29 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR | |
30 | * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT | |
31 | * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, | |
32 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT | |
33 | * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, | |
34 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY | |
35 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | |
36 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | |
37 | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
38 | * | |
39 | * Author: Tom Tucker <tom@opengridcomputing.com> | |
40 | */ | |
41 | ||
42 | #include <linux/sunrpc/debug.h> | |
43 | #include <linux/sunrpc/rpc_rdma.h> | |
44 | #include <linux/spinlock.h> | |
45 | #include <asm/unaligned.h> | |
46 | #include <rdma/ib_verbs.h> | |
47 | #include <rdma/rdma_cm.h> | |
48 | #include <linux/sunrpc/svc_rdma.h> | |
49 | ||
50 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT | |
51 | ||
52 | /* Encode an XDR as an array of IB SGE | |
53 | * | |
54 | * Assumptions: | |
55 | * - head[0] is physically contiguous. | |
56 | * - tail[0] is physically contiguous. | |
af901ca1 | 57 | * - pages[] is not physically or virtually contiguous and consists of |
c06b540a TT |
58 | * PAGE_SIZE elements. |
59 | * | |
60 | * Output: | |
61 | * SGE[0] reserved for RCPRDMA header | |
62 | * SGE[1] data from xdr->head[] | |
63 | * SGE[2..sge_count-2] data from xdr->pages[] | |
64 | * SGE[sge_count-1] data from xdr->tail. | |
65 | * | |
34d16e42 TT |
66 | * The max SGE we need is the length of the XDR / pagesize + one for |
67 | * head + one for tail + one for RPCRDMA header. Since RPCSVC_MAXPAGES | |
68 | * reserves a page for both the request and the reply header, and this | |
69 | * array is only concerned with the reply we are assured that we have | |
70 | * on extra page for the RPCRMDA header. | |
c06b540a | 71 | */ |
5eaa65b2 | 72 | static int fast_reg_xdr(struct svcxprt_rdma *xprt, |
afd566ea TT |
73 | struct xdr_buf *xdr, |
74 | struct svc_rdma_req_map *vec) | |
75 | { | |
76 | int sge_no; | |
77 | u32 sge_bytes; | |
78 | u32 page_bytes; | |
79 | u32 page_off; | |
80 | int page_no = 0; | |
81 | u8 *frva; | |
82 | struct svc_rdma_fastreg_mr *frmr; | |
83 | ||
84 | frmr = svc_rdma_get_frmr(xprt); | |
85 | if (IS_ERR(frmr)) | |
86 | return -ENOMEM; | |
87 | vec->frmr = frmr; | |
88 | ||
89 | /* Skip the RPCRDMA header */ | |
90 | sge_no = 1; | |
91 | ||
92 | /* Map the head. */ | |
93 | frva = (void *)((unsigned long)(xdr->head[0].iov_base) & PAGE_MASK); | |
94 | vec->sge[sge_no].iov_base = xdr->head[0].iov_base; | |
95 | vec->sge[sge_no].iov_len = xdr->head[0].iov_len; | |
96 | vec->count = 2; | |
97 | sge_no++; | |
98 | ||
99 | /* Build the FRMR */ | |
100 | frmr->kva = frva; | |
101 | frmr->direction = DMA_TO_DEVICE; | |
102 | frmr->access_flags = 0; | |
103 | frmr->map_len = PAGE_SIZE; | |
104 | frmr->page_list_len = 1; | |
105 | frmr->page_list->page_list[page_no] = | |
106 | ib_dma_map_single(xprt->sc_cm_id->device, | |
107 | (void *)xdr->head[0].iov_base, | |
108 | PAGE_SIZE, DMA_TO_DEVICE); | |
109 | if (ib_dma_mapping_error(xprt->sc_cm_id->device, | |
110 | frmr->page_list->page_list[page_no])) | |
111 | goto fatal_err; | |
112 | atomic_inc(&xprt->sc_dma_used); | |
113 | ||
114 | page_off = xdr->page_base; | |
115 | page_bytes = xdr->page_len + page_off; | |
116 | if (!page_bytes) | |
117 | goto encode_tail; | |
118 | ||
119 | /* Map the pages */ | |
120 | vec->sge[sge_no].iov_base = frva + frmr->map_len + page_off; | |
121 | vec->sge[sge_no].iov_len = page_bytes; | |
122 | sge_no++; | |
123 | while (page_bytes) { | |
124 | struct page *page; | |
125 | ||
126 | page = xdr->pages[page_no++]; | |
127 | sge_bytes = min_t(u32, page_bytes, (PAGE_SIZE - page_off)); | |
128 | page_bytes -= sge_bytes; | |
129 | ||
130 | frmr->page_list->page_list[page_no] = | |
98779be8 SW |
131 | ib_dma_map_single(xprt->sc_cm_id->device, |
132 | page_address(page), | |
afd566ea TT |
133 | PAGE_SIZE, DMA_TO_DEVICE); |
134 | if (ib_dma_mapping_error(xprt->sc_cm_id->device, | |
135 | frmr->page_list->page_list[page_no])) | |
136 | goto fatal_err; | |
137 | ||
138 | atomic_inc(&xprt->sc_dma_used); | |
139 | page_off = 0; /* reset for next time through loop */ | |
140 | frmr->map_len += PAGE_SIZE; | |
141 | frmr->page_list_len++; | |
142 | } | |
143 | vec->count++; | |
144 | ||
145 | encode_tail: | |
146 | /* Map tail */ | |
147 | if (0 == xdr->tail[0].iov_len) | |
148 | goto done; | |
149 | ||
150 | vec->count++; | |
151 | vec->sge[sge_no].iov_len = xdr->tail[0].iov_len; | |
152 | ||
153 | if (((unsigned long)xdr->tail[0].iov_base & PAGE_MASK) == | |
154 | ((unsigned long)xdr->head[0].iov_base & PAGE_MASK)) { | |
155 | /* | |
156 | * If head and tail use the same page, we don't need | |
157 | * to map it again. | |
158 | */ | |
159 | vec->sge[sge_no].iov_base = xdr->tail[0].iov_base; | |
160 | } else { | |
161 | void *va; | |
162 | ||
163 | /* Map another page for the tail */ | |
164 | page_off = (unsigned long)xdr->tail[0].iov_base & ~PAGE_MASK; | |
165 | va = (void *)((unsigned long)xdr->tail[0].iov_base & PAGE_MASK); | |
166 | vec->sge[sge_no].iov_base = frva + frmr->map_len + page_off; | |
167 | ||
168 | frmr->page_list->page_list[page_no] = | |
169 | ib_dma_map_single(xprt->sc_cm_id->device, va, PAGE_SIZE, | |
170 | DMA_TO_DEVICE); | |
171 | if (ib_dma_mapping_error(xprt->sc_cm_id->device, | |
172 | frmr->page_list->page_list[page_no])) | |
173 | goto fatal_err; | |
174 | atomic_inc(&xprt->sc_dma_used); | |
175 | frmr->map_len += PAGE_SIZE; | |
176 | frmr->page_list_len++; | |
177 | } | |
178 | ||
179 | done: | |
180 | if (svc_rdma_fastreg(xprt, frmr)) | |
181 | goto fatal_err; | |
182 | ||
183 | return 0; | |
184 | ||
185 | fatal_err: | |
186 | printk("svcrdma: Error fast registering memory for xprt %p\n", xprt); | |
21515e46 | 187 | vec->frmr = NULL; |
afd566ea TT |
188 | svc_rdma_put_frmr(xprt, frmr); |
189 | return -EIO; | |
190 | } | |
191 | ||
192 | static int map_xdr(struct svcxprt_rdma *xprt, | |
193 | struct xdr_buf *xdr, | |
194 | struct svc_rdma_req_map *vec) | |
c06b540a | 195 | { |
c06b540a | 196 | int sge_no; |
c06b540a TT |
197 | u32 sge_bytes; |
198 | u32 page_bytes; | |
34d16e42 | 199 | u32 page_off; |
c06b540a TT |
200 | int page_no; |
201 | ||
34d16e42 TT |
202 | BUG_ON(xdr->len != |
203 | (xdr->head[0].iov_len + xdr->page_len + xdr->tail[0].iov_len)); | |
204 | ||
afd566ea TT |
205 | if (xprt->sc_frmr_pg_list_len) |
206 | return fast_reg_xdr(xprt, xdr, vec); | |
207 | ||
c06b540a TT |
208 | /* Skip the first sge, this is for the RPCRDMA header */ |
209 | sge_no = 1; | |
210 | ||
211 | /* Head SGE */ | |
34d16e42 TT |
212 | vec->sge[sge_no].iov_base = xdr->head[0].iov_base; |
213 | vec->sge[sge_no].iov_len = xdr->head[0].iov_len; | |
c06b540a TT |
214 | sge_no++; |
215 | ||
216 | /* pages SGE */ | |
217 | page_no = 0; | |
218 | page_bytes = xdr->page_len; | |
219 | page_off = xdr->page_base; | |
34d16e42 TT |
220 | while (page_bytes) { |
221 | vec->sge[sge_no].iov_base = | |
222 | page_address(xdr->pages[page_no]) + page_off; | |
223 | sge_bytes = min_t(u32, page_bytes, (PAGE_SIZE - page_off)); | |
c06b540a | 224 | page_bytes -= sge_bytes; |
34d16e42 | 225 | vec->sge[sge_no].iov_len = sge_bytes; |
c06b540a TT |
226 | |
227 | sge_no++; | |
228 | page_no++; | |
229 | page_off = 0; /* reset for next time through loop */ | |
230 | } | |
231 | ||
232 | /* Tail SGE */ | |
34d16e42 TT |
233 | if (xdr->tail[0].iov_len) { |
234 | vec->sge[sge_no].iov_base = xdr->tail[0].iov_base; | |
235 | vec->sge[sge_no].iov_len = xdr->tail[0].iov_len; | |
c06b540a TT |
236 | sge_no++; |
237 | } | |
238 | ||
b1e1e158 | 239 | dprintk("svcrdma: map_xdr: sge_no %d page_no %d " |
2e3c230b | 240 | "page_base %u page_len %u head_len %zu tail_len %zu\n", |
b1e1e158 TT |
241 | sge_no, page_no, xdr->page_base, xdr->page_len, |
242 | xdr->head[0].iov_len, xdr->tail[0].iov_len); | |
243 | ||
34d16e42 | 244 | vec->count = sge_no; |
afd566ea | 245 | return 0; |
c06b540a TT |
246 | } |
247 | ||
c06b540a | 248 | /* Assumptions: |
afd566ea TT |
249 | * - We are using FRMR |
250 | * - or - | |
c06b540a TT |
251 | * - The specified write_len can be represented in sc_max_sge * PAGE_SIZE |
252 | */ | |
253 | static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp, | |
254 | u32 rmr, u64 to, | |
255 | u32 xdr_off, int write_len, | |
34d16e42 | 256 | struct svc_rdma_req_map *vec) |
c06b540a | 257 | { |
c06b540a TT |
258 | struct ib_send_wr write_wr; |
259 | struct ib_sge *sge; | |
260 | int xdr_sge_no; | |
261 | int sge_no; | |
262 | int sge_bytes; | |
263 | int sge_off; | |
264 | int bc; | |
265 | struct svc_rdma_op_ctxt *ctxt; | |
c06b540a | 266 | |
34d16e42 | 267 | BUG_ON(vec->count > RPCSVC_MAXPAGES); |
c06b540a | 268 | dprintk("svcrdma: RDMA_WRITE rmr=%x, to=%llx, xdr_off=%d, " |
34d16e42 | 269 | "write_len=%d, vec->sge=%p, vec->count=%lu\n", |
bb50c801 | 270 | rmr, (unsigned long long)to, xdr_off, |
34d16e42 | 271 | write_len, vec->sge, vec->count); |
c06b540a TT |
272 | |
273 | ctxt = svc_rdma_get_context(xprt); | |
34d16e42 TT |
274 | ctxt->direction = DMA_TO_DEVICE; |
275 | sge = ctxt->sge; | |
c06b540a TT |
276 | |
277 | /* Find the SGE associated with xdr_off */ | |
34d16e42 | 278 | for (bc = xdr_off, xdr_sge_no = 1; bc && xdr_sge_no < vec->count; |
c06b540a | 279 | xdr_sge_no++) { |
34d16e42 | 280 | if (vec->sge[xdr_sge_no].iov_len > bc) |
c06b540a | 281 | break; |
34d16e42 | 282 | bc -= vec->sge[xdr_sge_no].iov_len; |
c06b540a TT |
283 | } |
284 | ||
285 | sge_off = bc; | |
286 | bc = write_len; | |
287 | sge_no = 0; | |
288 | ||
289 | /* Copy the remaining SGE */ | |
afd566ea TT |
290 | while (bc != 0) { |
291 | sge_bytes = min_t(size_t, | |
292 | bc, vec->sge[xdr_sge_no].iov_len-sge_off); | |
c06b540a | 293 | sge[sge_no].length = sge_bytes; |
afd566ea TT |
294 | if (!vec->frmr) { |
295 | sge[sge_no].addr = | |
296 | ib_dma_map_single(xprt->sc_cm_id->device, | |
297 | (void *) | |
298 | vec->sge[xdr_sge_no].iov_base + sge_off, | |
299 | sge_bytes, DMA_TO_DEVICE); | |
300 | if (ib_dma_mapping_error(xprt->sc_cm_id->device, | |
301 | sge[sge_no].addr)) | |
302 | goto err; | |
303 | atomic_inc(&xprt->sc_dma_used); | |
304 | sge[sge_no].lkey = xprt->sc_dma_lkey; | |
305 | } else { | |
306 | sge[sge_no].addr = (unsigned long) | |
307 | vec->sge[xdr_sge_no].iov_base + sge_off; | |
308 | sge[sge_no].lkey = vec->frmr->mr->lkey; | |
309 | } | |
310 | ctxt->count++; | |
311 | ctxt->frmr = vec->frmr; | |
c06b540a TT |
312 | sge_off = 0; |
313 | sge_no++; | |
314 | xdr_sge_no++; | |
afd566ea | 315 | BUG_ON(xdr_sge_no > vec->count); |
c06b540a TT |
316 | bc -= sge_bytes; |
317 | } | |
318 | ||
c06b540a TT |
319 | /* Prepare WRITE WR */ |
320 | memset(&write_wr, 0, sizeof write_wr); | |
321 | ctxt->wr_op = IB_WR_RDMA_WRITE; | |
322 | write_wr.wr_id = (unsigned long)ctxt; | |
323 | write_wr.sg_list = &sge[0]; | |
324 | write_wr.num_sge = sge_no; | |
325 | write_wr.opcode = IB_WR_RDMA_WRITE; | |
326 | write_wr.send_flags = IB_SEND_SIGNALED; | |
327 | write_wr.wr.rdma.rkey = rmr; | |
328 | write_wr.wr.rdma.remote_addr = to; | |
329 | ||
330 | /* Post It */ | |
331 | atomic_inc(&rdma_stat_write); | |
34d16e42 TT |
332 | if (svc_rdma_send(xprt, &write_wr)) |
333 | goto err; | |
334 | return 0; | |
335 | err: | |
336 | svc_rdma_put_context(ctxt, 0); | |
337 | /* Fatal error, close transport */ | |
338 | return -EIO; | |
c06b540a TT |
339 | } |
340 | ||
341 | static int send_write_chunks(struct svcxprt_rdma *xprt, | |
342 | struct rpcrdma_msg *rdma_argp, | |
343 | struct rpcrdma_msg *rdma_resp, | |
344 | struct svc_rqst *rqstp, | |
34d16e42 | 345 | struct svc_rdma_req_map *vec) |
c06b540a TT |
346 | { |
347 | u32 xfer_len = rqstp->rq_res.page_len + rqstp->rq_res.tail[0].iov_len; | |
348 | int write_len; | |
349 | int max_write; | |
350 | u32 xdr_off; | |
351 | int chunk_off; | |
352 | int chunk_no; | |
353 | struct rpcrdma_write_array *arg_ary; | |
354 | struct rpcrdma_write_array *res_ary; | |
355 | int ret; | |
356 | ||
357 | arg_ary = svc_rdma_get_write_array(rdma_argp); | |
358 | if (!arg_ary) | |
359 | return 0; | |
360 | res_ary = (struct rpcrdma_write_array *) | |
361 | &rdma_resp->rm_body.rm_chunks[1]; | |
362 | ||
afd566ea TT |
363 | if (vec->frmr) |
364 | max_write = vec->frmr->map_len; | |
365 | else | |
366 | max_write = xprt->sc_max_sge * PAGE_SIZE; | |
c06b540a TT |
367 | |
368 | /* Write chunks start at the pagelist */ | |
369 | for (xdr_off = rqstp->rq_res.head[0].iov_len, chunk_no = 0; | |
370 | xfer_len && chunk_no < arg_ary->wc_nchunks; | |
371 | chunk_no++) { | |
372 | struct rpcrdma_segment *arg_ch; | |
373 | u64 rs_offset; | |
374 | ||
375 | arg_ch = &arg_ary->wc_array[chunk_no].wc_target; | |
376 | write_len = min(xfer_len, arg_ch->rs_length); | |
377 | ||
378 | /* Prepare the response chunk given the length actually | |
379 | * written */ | |
380 | rs_offset = get_unaligned(&(arg_ch->rs_offset)); | |
381 | svc_rdma_xdr_encode_array_chunk(res_ary, chunk_no, | |
382 | arg_ch->rs_handle, | |
383 | rs_offset, | |
384 | write_len); | |
385 | chunk_off = 0; | |
386 | while (write_len) { | |
387 | int this_write; | |
388 | this_write = min(write_len, max_write); | |
389 | ret = send_write(xprt, rqstp, | |
390 | arg_ch->rs_handle, | |
391 | rs_offset + chunk_off, | |
392 | xdr_off, | |
393 | this_write, | |
34d16e42 | 394 | vec); |
c06b540a TT |
395 | if (ret) { |
396 | dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n", | |
397 | ret); | |
398 | return -EIO; | |
399 | } | |
400 | chunk_off += this_write; | |
401 | xdr_off += this_write; | |
402 | xfer_len -= this_write; | |
403 | write_len -= this_write; | |
404 | } | |
405 | } | |
406 | /* Update the req with the number of chunks actually used */ | |
407 | svc_rdma_xdr_encode_write_list(rdma_resp, chunk_no); | |
408 | ||
409 | return rqstp->rq_res.page_len + rqstp->rq_res.tail[0].iov_len; | |
410 | } | |
411 | ||
412 | static int send_reply_chunks(struct svcxprt_rdma *xprt, | |
413 | struct rpcrdma_msg *rdma_argp, | |
414 | struct rpcrdma_msg *rdma_resp, | |
415 | struct svc_rqst *rqstp, | |
34d16e42 | 416 | struct svc_rdma_req_map *vec) |
c06b540a TT |
417 | { |
418 | u32 xfer_len = rqstp->rq_res.len; | |
419 | int write_len; | |
420 | int max_write; | |
421 | u32 xdr_off; | |
422 | int chunk_no; | |
423 | int chunk_off; | |
424 | struct rpcrdma_segment *ch; | |
425 | struct rpcrdma_write_array *arg_ary; | |
426 | struct rpcrdma_write_array *res_ary; | |
427 | int ret; | |
428 | ||
429 | arg_ary = svc_rdma_get_reply_array(rdma_argp); | |
430 | if (!arg_ary) | |
431 | return 0; | |
432 | /* XXX: need to fix when reply lists occur with read-list and or | |
433 | * write-list */ | |
434 | res_ary = (struct rpcrdma_write_array *) | |
435 | &rdma_resp->rm_body.rm_chunks[2]; | |
436 | ||
afd566ea TT |
437 | if (vec->frmr) |
438 | max_write = vec->frmr->map_len; | |
439 | else | |
440 | max_write = xprt->sc_max_sge * PAGE_SIZE; | |
c06b540a TT |
441 | |
442 | /* xdr offset starts at RPC message */ | |
443 | for (xdr_off = 0, chunk_no = 0; | |
444 | xfer_len && chunk_no < arg_ary->wc_nchunks; | |
445 | chunk_no++) { | |
446 | u64 rs_offset; | |
447 | ch = &arg_ary->wc_array[chunk_no].wc_target; | |
448 | write_len = min(xfer_len, ch->rs_length); | |
449 | ||
c06b540a TT |
450 | /* Prepare the reply chunk given the length actually |
451 | * written */ | |
452 | rs_offset = get_unaligned(&(ch->rs_offset)); | |
453 | svc_rdma_xdr_encode_array_chunk(res_ary, chunk_no, | |
454 | ch->rs_handle, rs_offset, | |
455 | write_len); | |
456 | chunk_off = 0; | |
457 | while (write_len) { | |
458 | int this_write; | |
459 | ||
460 | this_write = min(write_len, max_write); | |
461 | ret = send_write(xprt, rqstp, | |
462 | ch->rs_handle, | |
463 | rs_offset + chunk_off, | |
464 | xdr_off, | |
465 | this_write, | |
34d16e42 | 466 | vec); |
c06b540a TT |
467 | if (ret) { |
468 | dprintk("svcrdma: RDMA_WRITE failed, ret=%d\n", | |
469 | ret); | |
470 | return -EIO; | |
471 | } | |
472 | chunk_off += this_write; | |
473 | xdr_off += this_write; | |
474 | xfer_len -= this_write; | |
475 | write_len -= this_write; | |
476 | } | |
477 | } | |
478 | /* Update the req with the number of chunks actually used */ | |
479 | svc_rdma_xdr_encode_reply_array(res_ary, chunk_no); | |
480 | ||
481 | return rqstp->rq_res.len; | |
482 | } | |
483 | ||
484 | /* This function prepares the portion of the RPCRDMA message to be | |
485 | * sent in the RDMA_SEND. This function is called after data sent via | |
486 | * RDMA has already been transmitted. There are three cases: | |
487 | * - The RPCRDMA header, RPC header, and payload are all sent in a | |
488 | * single RDMA_SEND. This is the "inline" case. | |
489 | * - The RPCRDMA header and some portion of the RPC header and data | |
490 | * are sent via this RDMA_SEND and another portion of the data is | |
491 | * sent via RDMA. | |
492 | * - The RPCRDMA header [NOMSG] is sent in this RDMA_SEND and the RPC | |
493 | * header and data are all transmitted via RDMA. | |
494 | * In all three cases, this function prepares the RPCRDMA header in | |
495 | * sge[0], the 'type' parameter indicates the type to place in the | |
496 | * RPCRDMA header, and the 'byte_count' field indicates how much of | |
497 | * the XDR to include in this RDMA_SEND. | |
498 | */ | |
499 | static int send_reply(struct svcxprt_rdma *rdma, | |
500 | struct svc_rqst *rqstp, | |
501 | struct page *page, | |
502 | struct rpcrdma_msg *rdma_resp, | |
503 | struct svc_rdma_op_ctxt *ctxt, | |
34d16e42 | 504 | struct svc_rdma_req_map *vec, |
c06b540a TT |
505 | int byte_count) |
506 | { | |
507 | struct ib_send_wr send_wr; | |
afd566ea | 508 | struct ib_send_wr inv_wr; |
c06b540a TT |
509 | int sge_no; |
510 | int sge_bytes; | |
511 | int page_no; | |
512 | int ret; | |
513 | ||
0e7f011a TT |
514 | /* Post a recv buffer to handle another request. */ |
515 | ret = svc_rdma_post_recv(rdma); | |
516 | if (ret) { | |
517 | printk(KERN_INFO | |
518 | "svcrdma: could not post a receive buffer, err=%d." | |
519 | "Closing transport %p.\n", ret, rdma); | |
520 | set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags); | |
21515e46 | 521 | svc_rdma_put_frmr(rdma, vec->frmr); |
5ac461a6 TT |
522 | svc_rdma_put_context(ctxt, 0); |
523 | return -ENOTCONN; | |
0e7f011a TT |
524 | } |
525 | ||
c06b540a TT |
526 | /* Prepare the context */ |
527 | ctxt->pages[0] = page; | |
528 | ctxt->count = 1; | |
afd566ea TT |
529 | ctxt->frmr = vec->frmr; |
530 | if (vec->frmr) | |
531 | set_bit(RDMACTXT_F_FAST_UNREG, &ctxt->flags); | |
532 | else | |
533 | clear_bit(RDMACTXT_F_FAST_UNREG, &ctxt->flags); | |
c06b540a TT |
534 | |
535 | /* Prepare the SGE for the RPCRDMA Header */ | |
98779be8 SW |
536 | ctxt->sge[0].lkey = rdma->sc_dma_lkey; |
537 | ctxt->sge[0].length = svc_rdma_xdr_get_reply_hdr_len(rdma_resp); | |
c06b540a | 538 | ctxt->sge[0].addr = |
98779be8 SW |
539 | ib_dma_map_single(rdma->sc_cm_id->device, page_address(page), |
540 | ctxt->sge[0].length, DMA_TO_DEVICE); | |
afd566ea TT |
541 | if (ib_dma_mapping_error(rdma->sc_cm_id->device, ctxt->sge[0].addr)) |
542 | goto err; | |
543 | atomic_inc(&rdma->sc_dma_used); | |
544 | ||
c06b540a | 545 | ctxt->direction = DMA_TO_DEVICE; |
afd566ea | 546 | |
c06b540a | 547 | /* Determine how many of our SGE are to be transmitted */ |
34d16e42 TT |
548 | for (sge_no = 1; byte_count && sge_no < vec->count; sge_no++) { |
549 | sge_bytes = min_t(size_t, vec->sge[sge_no].iov_len, byte_count); | |
c06b540a | 550 | byte_count -= sge_bytes; |
afd566ea TT |
551 | if (!vec->frmr) { |
552 | ctxt->sge[sge_no].addr = | |
553 | ib_dma_map_single(rdma->sc_cm_id->device, | |
554 | vec->sge[sge_no].iov_base, | |
555 | sge_bytes, DMA_TO_DEVICE); | |
556 | if (ib_dma_mapping_error(rdma->sc_cm_id->device, | |
557 | ctxt->sge[sge_no].addr)) | |
558 | goto err; | |
559 | atomic_inc(&rdma->sc_dma_used); | |
560 | ctxt->sge[sge_no].lkey = rdma->sc_dma_lkey; | |
561 | } else { | |
562 | ctxt->sge[sge_no].addr = (unsigned long) | |
563 | vec->sge[sge_no].iov_base; | |
564 | ctxt->sge[sge_no].lkey = vec->frmr->mr->lkey; | |
565 | } | |
34d16e42 | 566 | ctxt->sge[sge_no].length = sge_bytes; |
c06b540a TT |
567 | } |
568 | BUG_ON(byte_count != 0); | |
569 | ||
570 | /* Save all respages in the ctxt and remove them from the | |
571 | * respages array. They are our pages until the I/O | |
572 | * completes. | |
573 | */ | |
574 | for (page_no = 0; page_no < rqstp->rq_resused; page_no++) { | |
575 | ctxt->pages[page_no+1] = rqstp->rq_respages[page_no]; | |
576 | ctxt->count++; | |
577 | rqstp->rq_respages[page_no] = NULL; | |
afd566ea TT |
578 | /* |
579 | * If there are more pages than SGE, terminate SGE | |
580 | * list so that svc_rdma_unmap_dma doesn't attempt to | |
581 | * unmap garbage. | |
582 | */ | |
34d16e42 TT |
583 | if (page_no+1 >= sge_no) |
584 | ctxt->sge[page_no+1].length = 0; | |
c06b540a | 585 | } |
c06b540a TT |
586 | BUG_ON(sge_no > rdma->sc_max_sge); |
587 | memset(&send_wr, 0, sizeof send_wr); | |
588 | ctxt->wr_op = IB_WR_SEND; | |
589 | send_wr.wr_id = (unsigned long)ctxt; | |
590 | send_wr.sg_list = ctxt->sge; | |
591 | send_wr.num_sge = sge_no; | |
592 | send_wr.opcode = IB_WR_SEND; | |
593 | send_wr.send_flags = IB_SEND_SIGNALED; | |
afd566ea TT |
594 | if (vec->frmr) { |
595 | /* Prepare INVALIDATE WR */ | |
596 | memset(&inv_wr, 0, sizeof inv_wr); | |
597 | inv_wr.opcode = IB_WR_LOCAL_INV; | |
598 | inv_wr.send_flags = IB_SEND_SIGNALED; | |
599 | inv_wr.ex.invalidate_rkey = | |
600 | vec->frmr->mr->lkey; | |
601 | send_wr.next = &inv_wr; | |
602 | } | |
c06b540a TT |
603 | |
604 | ret = svc_rdma_send(rdma, &send_wr); | |
605 | if (ret) | |
afd566ea | 606 | goto err; |
c06b540a | 607 | |
afd566ea TT |
608 | return 0; |
609 | ||
610 | err: | |
21515e46 | 611 | svc_rdma_unmap_dma(ctxt); |
afd566ea TT |
612 | svc_rdma_put_frmr(rdma, vec->frmr); |
613 | svc_rdma_put_context(ctxt, 1); | |
614 | return -EIO; | |
c06b540a TT |
615 | } |
616 | ||
617 | void svc_rdma_prep_reply_hdr(struct svc_rqst *rqstp) | |
618 | { | |
619 | } | |
620 | ||
621 | /* | |
622 | * Return the start of an xdr buffer. | |
623 | */ | |
624 | static void *xdr_start(struct xdr_buf *xdr) | |
625 | { | |
626 | return xdr->head[0].iov_base - | |
627 | (xdr->len - | |
628 | xdr->page_len - | |
629 | xdr->tail[0].iov_len - | |
630 | xdr->head[0].iov_len); | |
631 | } | |
632 | ||
633 | int svc_rdma_sendto(struct svc_rqst *rqstp) | |
634 | { | |
635 | struct svc_xprt *xprt = rqstp->rq_xprt; | |
636 | struct svcxprt_rdma *rdma = | |
637 | container_of(xprt, struct svcxprt_rdma, sc_xprt); | |
638 | struct rpcrdma_msg *rdma_argp; | |
639 | struct rpcrdma_msg *rdma_resp; | |
640 | struct rpcrdma_write_array *reply_ary; | |
641 | enum rpcrdma_proc reply_type; | |
642 | int ret; | |
643 | int inline_bytes; | |
c06b540a TT |
644 | struct page *res_page; |
645 | struct svc_rdma_op_ctxt *ctxt; | |
34d16e42 | 646 | struct svc_rdma_req_map *vec; |
c06b540a TT |
647 | |
648 | dprintk("svcrdma: sending response for rqstp=%p\n", rqstp); | |
649 | ||
650 | /* Get the RDMA request header. */ | |
651 | rdma_argp = xdr_start(&rqstp->rq_arg); | |
652 | ||
34d16e42 | 653 | /* Build an req vec for the XDR */ |
c06b540a TT |
654 | ctxt = svc_rdma_get_context(rdma); |
655 | ctxt->direction = DMA_TO_DEVICE; | |
34d16e42 | 656 | vec = svc_rdma_get_req_map(); |
afd566ea TT |
657 | ret = map_xdr(rdma, &rqstp->rq_res, vec); |
658 | if (ret) | |
659 | goto err0; | |
c06b540a TT |
660 | inline_bytes = rqstp->rq_res.len; |
661 | ||
662 | /* Create the RDMA response header */ | |
663 | res_page = svc_rdma_get_page(); | |
664 | rdma_resp = page_address(res_page); | |
665 | reply_ary = svc_rdma_get_reply_array(rdma_argp); | |
666 | if (reply_ary) | |
667 | reply_type = RDMA_NOMSG; | |
668 | else | |
669 | reply_type = RDMA_MSG; | |
670 | svc_rdma_xdr_encode_reply_header(rdma, rdma_argp, | |
671 | rdma_resp, reply_type); | |
672 | ||
673 | /* Send any write-chunk data and build resp write-list */ | |
674 | ret = send_write_chunks(rdma, rdma_argp, rdma_resp, | |
34d16e42 | 675 | rqstp, vec); |
c06b540a TT |
676 | if (ret < 0) { |
677 | printk(KERN_ERR "svcrdma: failed to send write chunks, rc=%d\n", | |
678 | ret); | |
afd566ea | 679 | goto err1; |
c06b540a TT |
680 | } |
681 | inline_bytes -= ret; | |
682 | ||
683 | /* Send any reply-list data and update resp reply-list */ | |
684 | ret = send_reply_chunks(rdma, rdma_argp, rdma_resp, | |
34d16e42 | 685 | rqstp, vec); |
c06b540a TT |
686 | if (ret < 0) { |
687 | printk(KERN_ERR "svcrdma: failed to send reply chunks, rc=%d\n", | |
688 | ret); | |
afd566ea | 689 | goto err1; |
c06b540a TT |
690 | } |
691 | inline_bytes -= ret; | |
692 | ||
34d16e42 | 693 | ret = send_reply(rdma, rqstp, res_page, rdma_resp, ctxt, vec, |
c06b540a | 694 | inline_bytes); |
34d16e42 | 695 | svc_rdma_put_req_map(vec); |
c06b540a TT |
696 | dprintk("svcrdma: send_reply returns %d\n", ret); |
697 | return ret; | |
afd566ea TT |
698 | |
699 | err1: | |
700 | put_page(res_page); | |
701 | err0: | |
34d16e42 | 702 | svc_rdma_put_req_map(vec); |
c06b540a | 703 | svc_rdma_put_context(ctxt, 0); |
c06b540a TT |
704 | return ret; |
705 | } |