Merge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jlbec...
[pandora-kernel.git] / drivers / infiniband / ulp / iser / iser_initiator.c
1 /*
2  * Copyright (c) 2004, 2005, 2006 Voltaire, 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
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *        copyright notice, this list of conditions and the following
16  *        disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer in the documentation and/or other materials
21  *        provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/kernel.h>
33 #include <linux/slab.h>
34 #include <linux/mm.h>
35 #include <linux/scatterlist.h>
36 #include <linux/kfifo.h>
37 #include <scsi/scsi_cmnd.h>
38 #include <scsi/scsi_host.h>
39
40 #include "iscsi_iser.h"
41
42 /* Register user buffer memory and initialize passive rdma
43  *  dto descriptor. Total data size is stored in
44  *  iser_task->data[ISER_DIR_IN].data_len
45  */
46 static int iser_prepare_read_cmd(struct iscsi_task *task,
47                                  unsigned int edtl)
48
49 {
50         struct iscsi_iser_task *iser_task = task->dd_data;
51         struct iser_regd_buf *regd_buf;
52         int err;
53         struct iser_hdr *hdr = &iser_task->desc.iser_header;
54         struct iser_data_buf *buf_in = &iser_task->data[ISER_DIR_IN];
55
56         err = iser_dma_map_task_data(iser_task,
57                                      buf_in,
58                                      ISER_DIR_IN,
59                                      DMA_FROM_DEVICE);
60         if (err)
61                 return err;
62
63         if (edtl > iser_task->data[ISER_DIR_IN].data_len) {
64                 iser_err("Total data length: %ld, less than EDTL: "
65                          "%d, in READ cmd BHS itt: %d, conn: 0x%p\n",
66                          iser_task->data[ISER_DIR_IN].data_len, edtl,
67                          task->itt, iser_task->iser_conn);
68                 return -EINVAL;
69         }
70
71         err = iser_reg_rdma_mem(iser_task,ISER_DIR_IN);
72         if (err) {
73                 iser_err("Failed to set up Data-IN RDMA\n");
74                 return err;
75         }
76         regd_buf = &iser_task->rdma_regd[ISER_DIR_IN];
77
78         hdr->flags    |= ISER_RSV;
79         hdr->read_stag = cpu_to_be32(regd_buf->reg.rkey);
80         hdr->read_va   = cpu_to_be64(regd_buf->reg.va);
81
82         iser_dbg("Cmd itt:%d READ tags RKEY:%#.4X VA:%#llX\n",
83                  task->itt, regd_buf->reg.rkey,
84                  (unsigned long long)regd_buf->reg.va);
85
86         return 0;
87 }
88
89 /* Register user buffer memory and initialize passive rdma
90  *  dto descriptor. Total data size is stored in
91  *  task->data[ISER_DIR_OUT].data_len
92  */
93 static int
94 iser_prepare_write_cmd(struct iscsi_task *task,
95                        unsigned int imm_sz,
96                        unsigned int unsol_sz,
97                        unsigned int edtl)
98 {
99         struct iscsi_iser_task *iser_task = task->dd_data;
100         struct iser_regd_buf *regd_buf;
101         int err;
102         struct iser_hdr *hdr = &iser_task->desc.iser_header;
103         struct iser_data_buf *buf_out = &iser_task->data[ISER_DIR_OUT];
104         struct ib_sge *tx_dsg = &iser_task->desc.tx_sg[1];
105
106         err = iser_dma_map_task_data(iser_task,
107                                      buf_out,
108                                      ISER_DIR_OUT,
109                                      DMA_TO_DEVICE);
110         if (err)
111                 return err;
112
113         if (edtl > iser_task->data[ISER_DIR_OUT].data_len) {
114                 iser_err("Total data length: %ld, less than EDTL: %d, "
115                          "in WRITE cmd BHS itt: %d, conn: 0x%p\n",
116                          iser_task->data[ISER_DIR_OUT].data_len,
117                          edtl, task->itt, task->conn);
118                 return -EINVAL;
119         }
120
121         err = iser_reg_rdma_mem(iser_task,ISER_DIR_OUT);
122         if (err != 0) {
123                 iser_err("Failed to register write cmd RDMA mem\n");
124                 return err;
125         }
126
127         regd_buf = &iser_task->rdma_regd[ISER_DIR_OUT];
128
129         if (unsol_sz < edtl) {
130                 hdr->flags     |= ISER_WSV;
131                 hdr->write_stag = cpu_to_be32(regd_buf->reg.rkey);
132                 hdr->write_va   = cpu_to_be64(regd_buf->reg.va + unsol_sz);
133
134                 iser_dbg("Cmd itt:%d, WRITE tags, RKEY:%#.4X "
135                          "VA:%#llX + unsol:%d\n",
136                          task->itt, regd_buf->reg.rkey,
137                          (unsigned long long)regd_buf->reg.va, unsol_sz);
138         }
139
140         if (imm_sz > 0) {
141                 iser_dbg("Cmd itt:%d, WRITE, adding imm.data sz: %d\n",
142                          task->itt, imm_sz);
143                 tx_dsg->addr   = regd_buf->reg.va;
144                 tx_dsg->length = imm_sz;
145                 tx_dsg->lkey   = regd_buf->reg.lkey;
146                 iser_task->desc.num_sge = 2;
147         }
148
149         return 0;
150 }
151
152 /* creates a new tx descriptor and adds header regd buffer */
153 static void iser_create_send_desc(struct iser_conn      *ib_conn,
154                                   struct iser_tx_desc   *tx_desc)
155 {
156         struct iser_device *device = ib_conn->device;
157
158         ib_dma_sync_single_for_cpu(device->ib_device,
159                 tx_desc->dma_addr, ISER_HEADERS_LEN, DMA_TO_DEVICE);
160
161         memset(&tx_desc->iser_header, 0, sizeof(struct iser_hdr));
162         tx_desc->iser_header.flags = ISER_VER;
163
164         tx_desc->num_sge = 1;
165
166         if (tx_desc->tx_sg[0].lkey != device->mr->lkey) {
167                 tx_desc->tx_sg[0].lkey = device->mr->lkey;
168                 iser_dbg("sdesc %p lkey mismatch, fixing\n", tx_desc);
169         }
170 }
171
172
173 static int iser_alloc_rx_descriptors(struct iser_conn *ib_conn)
174 {
175         int i, j;
176         u64 dma_addr;
177         struct iser_rx_desc *rx_desc;
178         struct ib_sge       *rx_sg;
179         struct iser_device  *device = ib_conn->device;
180
181         ib_conn->rx_descs = kmalloc(ISER_QP_MAX_RECV_DTOS *
182                                 sizeof(struct iser_rx_desc), GFP_KERNEL);
183         if (!ib_conn->rx_descs)
184                 goto rx_desc_alloc_fail;
185
186         rx_desc = ib_conn->rx_descs;
187
188         for (i = 0; i < ISER_QP_MAX_RECV_DTOS; i++, rx_desc++)  {
189                 dma_addr = ib_dma_map_single(device->ib_device, (void *)rx_desc,
190                                         ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
191                 if (ib_dma_mapping_error(device->ib_device, dma_addr))
192                         goto rx_desc_dma_map_failed;
193
194                 rx_desc->dma_addr = dma_addr;
195
196                 rx_sg = &rx_desc->rx_sg;
197                 rx_sg->addr   = rx_desc->dma_addr;
198                 rx_sg->length = ISER_RX_PAYLOAD_SIZE;
199                 rx_sg->lkey   = device->mr->lkey;
200         }
201
202         ib_conn->rx_desc_head = 0;
203         return 0;
204
205 rx_desc_dma_map_failed:
206         rx_desc = ib_conn->rx_descs;
207         for (j = 0; j < i; j++, rx_desc++)
208                 ib_dma_unmap_single(device->ib_device, rx_desc->dma_addr,
209                         ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
210         kfree(ib_conn->rx_descs);
211         ib_conn->rx_descs = NULL;
212 rx_desc_alloc_fail:
213         iser_err("failed allocating rx descriptors / data buffers\n");
214         return -ENOMEM;
215 }
216
217 void iser_free_rx_descriptors(struct iser_conn *ib_conn)
218 {
219         int i;
220         struct iser_rx_desc *rx_desc;
221         struct iser_device *device = ib_conn->device;
222
223         if (ib_conn->login_buf) {
224                 if (ib_conn->login_req_dma)
225                         ib_dma_unmap_single(device->ib_device,
226                                 ib_conn->login_req_dma,
227                                 ISCSI_DEF_MAX_RECV_SEG_LEN, DMA_TO_DEVICE);
228                 if (ib_conn->login_resp_dma)
229                         ib_dma_unmap_single(device->ib_device,
230                                 ib_conn->login_resp_dma,
231                                 ISER_RX_LOGIN_SIZE, DMA_FROM_DEVICE);
232                 kfree(ib_conn->login_buf);
233         }
234
235         if (!ib_conn->rx_descs)
236                 return;
237
238         rx_desc = ib_conn->rx_descs;
239         for (i = 0; i < ISER_QP_MAX_RECV_DTOS; i++, rx_desc++)
240                 ib_dma_unmap_single(device->ib_device, rx_desc->dma_addr,
241                         ISER_RX_PAYLOAD_SIZE, DMA_FROM_DEVICE);
242         kfree(ib_conn->rx_descs);
243 }
244
245 /**
246  *  iser_conn_set_full_featured_mode - (iSER API)
247  */
248 int iser_conn_set_full_featured_mode(struct iscsi_conn *conn)
249 {
250         struct iscsi_iser_conn *iser_conn = conn->dd_data;
251
252         iser_dbg("Initially post: %d\n", ISER_MIN_POSTED_RX);
253
254         /* Check that there is no posted recv or send buffers left - */
255         /* they must be consumed during the login phase */
256         BUG_ON(iser_conn->ib_conn->post_recv_buf_count != 0);
257         BUG_ON(atomic_read(&iser_conn->ib_conn->post_send_buf_count) != 0);
258
259         if (iser_alloc_rx_descriptors(iser_conn->ib_conn))
260                 return -ENOMEM;
261
262         /* Initial post receive buffers */
263         if (iser_post_recvm(iser_conn->ib_conn, ISER_MIN_POSTED_RX))
264                 return -ENOMEM;
265
266         return 0;
267 }
268
269 /**
270  * iser_send_command - send command PDU
271  */
272 int iser_send_command(struct iscsi_conn *conn,
273                       struct iscsi_task *task)
274 {
275         struct iscsi_iser_conn *iser_conn = conn->dd_data;
276         struct iscsi_iser_task *iser_task = task->dd_data;
277         unsigned long edtl;
278         int err;
279         struct iser_data_buf *data_buf;
280         struct iscsi_scsi_req *hdr = (struct iscsi_scsi_req *)task->hdr;
281         struct scsi_cmnd *sc  =  task->sc;
282         struct iser_tx_desc *tx_desc = &iser_task->desc;
283
284         edtl = ntohl(hdr->data_length);
285
286         /* build the tx desc regd header and add it to the tx desc dto */
287         tx_desc->type = ISCSI_TX_SCSI_COMMAND;
288         iser_create_send_desc(iser_conn->ib_conn, tx_desc);
289
290         if (hdr->flags & ISCSI_FLAG_CMD_READ)
291                 data_buf = &iser_task->data[ISER_DIR_IN];
292         else
293                 data_buf = &iser_task->data[ISER_DIR_OUT];
294
295         if (scsi_sg_count(sc)) { /* using a scatter list */
296                 data_buf->buf  = scsi_sglist(sc);
297                 data_buf->size = scsi_sg_count(sc);
298         }
299
300         data_buf->data_len = scsi_bufflen(sc);
301
302         if (hdr->flags & ISCSI_FLAG_CMD_READ) {
303                 err = iser_prepare_read_cmd(task, edtl);
304                 if (err)
305                         goto send_command_error;
306         }
307         if (hdr->flags & ISCSI_FLAG_CMD_WRITE) {
308                 err = iser_prepare_write_cmd(task,
309                                              task->imm_count,
310                                              task->imm_count +
311                                              task->unsol_r2t.data_length,
312                                              edtl);
313                 if (err)
314                         goto send_command_error;
315         }
316
317         iser_task->status = ISER_TASK_STATUS_STARTED;
318
319         err = iser_post_send(iser_conn->ib_conn, tx_desc);
320         if (!err)
321                 return 0;
322
323 send_command_error:
324         iser_err("conn %p failed task->itt %d err %d\n",conn, task->itt, err);
325         return err;
326 }
327
328 /**
329  * iser_send_data_out - send data out PDU
330  */
331 int iser_send_data_out(struct iscsi_conn *conn,
332                        struct iscsi_task *task,
333                        struct iscsi_data *hdr)
334 {
335         struct iscsi_iser_conn *iser_conn = conn->dd_data;
336         struct iscsi_iser_task *iser_task = task->dd_data;
337         struct iser_tx_desc *tx_desc = NULL;
338         struct iser_regd_buf *regd_buf;
339         unsigned long buf_offset;
340         unsigned long data_seg_len;
341         uint32_t itt;
342         int err = 0;
343         struct ib_sge *tx_dsg;
344
345         itt = (__force uint32_t)hdr->itt;
346         data_seg_len = ntoh24(hdr->dlength);
347         buf_offset   = ntohl(hdr->offset);
348
349         iser_dbg("%s itt %d dseg_len %d offset %d\n",
350                  __func__,(int)itt,(int)data_seg_len,(int)buf_offset);
351
352         tx_desc = kmem_cache_zalloc(ig.desc_cache, GFP_ATOMIC);
353         if (tx_desc == NULL) {
354                 iser_err("Failed to alloc desc for post dataout\n");
355                 return -ENOMEM;
356         }
357
358         tx_desc->type = ISCSI_TX_DATAOUT;
359         tx_desc->iser_header.flags = ISER_VER;
360         memcpy(&tx_desc->iscsi_header, hdr, sizeof(struct iscsi_hdr));
361
362         /* build the tx desc */
363         iser_initialize_task_headers(task, tx_desc);
364
365         regd_buf = &iser_task->rdma_regd[ISER_DIR_OUT];
366         tx_dsg = &tx_desc->tx_sg[1];
367         tx_dsg->addr    = regd_buf->reg.va + buf_offset;
368         tx_dsg->length  = data_seg_len;
369         tx_dsg->lkey    = regd_buf->reg.lkey;
370         tx_desc->num_sge = 2;
371
372         if (buf_offset + data_seg_len > iser_task->data[ISER_DIR_OUT].data_len) {
373                 iser_err("Offset:%ld & DSL:%ld in Data-Out "
374                          "inconsistent with total len:%ld, itt:%d\n",
375                          buf_offset, data_seg_len,
376                          iser_task->data[ISER_DIR_OUT].data_len, itt);
377                 err = -EINVAL;
378                 goto send_data_out_error;
379         }
380         iser_dbg("data-out itt: %d, offset: %ld, sz: %ld\n",
381                  itt, buf_offset, data_seg_len);
382
383
384         err = iser_post_send(iser_conn->ib_conn, tx_desc);
385         if (!err)
386                 return 0;
387
388 send_data_out_error:
389         kmem_cache_free(ig.desc_cache, tx_desc);
390         iser_err("conn %p failed err %d\n",conn, err);
391         return err;
392 }
393
394 int iser_send_control(struct iscsi_conn *conn,
395                       struct iscsi_task *task)
396 {
397         struct iscsi_iser_conn *iser_conn = conn->dd_data;
398         struct iscsi_iser_task *iser_task = task->dd_data;
399         struct iser_tx_desc *mdesc = &iser_task->desc;
400         unsigned long data_seg_len;
401         int err = 0;
402         struct iser_device *device;
403         struct iser_conn *ib_conn = iser_conn->ib_conn;
404
405         /* build the tx desc regd header and add it to the tx desc dto */
406         mdesc->type = ISCSI_TX_CONTROL;
407         iser_create_send_desc(iser_conn->ib_conn, mdesc);
408
409         device = iser_conn->ib_conn->device;
410
411         data_seg_len = ntoh24(task->hdr->dlength);
412
413         if (data_seg_len > 0) {
414                 struct ib_sge *tx_dsg = &mdesc->tx_sg[1];
415                 if (task != conn->login_task) {
416                         iser_err("data present on non login task!!!\n");
417                         goto send_control_error;
418                 }
419
420                 ib_dma_sync_single_for_cpu(device->ib_device,
421                         ib_conn->login_req_dma, task->data_count,
422                         DMA_TO_DEVICE);
423
424                 memcpy(iser_conn->ib_conn->login_req_buf, task->data,
425                                                         task->data_count);
426
427                 ib_dma_sync_single_for_device(device->ib_device,
428                         ib_conn->login_req_dma, task->data_count,
429                         DMA_TO_DEVICE);
430
431                 tx_dsg->addr    = iser_conn->ib_conn->login_req_dma;
432                 tx_dsg->length  = task->data_count;
433                 tx_dsg->lkey    = device->mr->lkey;
434                 mdesc->num_sge = 2;
435         }
436
437         if (task == conn->login_task) {
438                 err = iser_post_recvl(iser_conn->ib_conn);
439                 if (err)
440                         goto send_control_error;
441         }
442
443         err = iser_post_send(iser_conn->ib_conn, mdesc);
444         if (!err)
445                 return 0;
446
447 send_control_error:
448         iser_err("conn %p failed err %d\n",conn, err);
449         return err;
450 }
451
452 /**
453  * iser_rcv_dto_completion - recv DTO completion
454  */
455 void iser_rcv_completion(struct iser_rx_desc *rx_desc,
456                          unsigned long rx_xfer_len,
457                          struct iser_conn *ib_conn)
458 {
459         struct iscsi_iser_conn *conn = ib_conn->iser_conn;
460         struct iscsi_hdr *hdr;
461         u64 rx_dma;
462         int rx_buflen, outstanding, count, err;
463
464         /* differentiate between login to all other PDUs */
465         if ((char *)rx_desc == ib_conn->login_resp_buf) {
466                 rx_dma = ib_conn->login_resp_dma;
467                 rx_buflen = ISER_RX_LOGIN_SIZE;
468         } else {
469                 rx_dma = rx_desc->dma_addr;
470                 rx_buflen = ISER_RX_PAYLOAD_SIZE;
471         }
472
473         ib_dma_sync_single_for_cpu(ib_conn->device->ib_device, rx_dma,
474                         rx_buflen, DMA_FROM_DEVICE);
475
476         hdr = &rx_desc->iscsi_header;
477
478         iser_dbg("op 0x%x itt 0x%x dlen %d\n", hdr->opcode,
479                         hdr->itt, (int)(rx_xfer_len - ISER_HEADERS_LEN));
480
481         iscsi_iser_recv(conn->iscsi_conn, hdr,
482                 rx_desc->data, rx_xfer_len - ISER_HEADERS_LEN);
483
484         ib_dma_sync_single_for_device(ib_conn->device->ib_device, rx_dma,
485                         rx_buflen, DMA_FROM_DEVICE);
486
487         /* decrementing conn->post_recv_buf_count only --after-- freeing the   *
488          * task eliminates the need to worry on tasks which are completed in   *
489          * parallel to the execution of iser_conn_term. So the code that waits *
490          * for the posted rx bufs refcount to become zero handles everything   */
491         conn->ib_conn->post_recv_buf_count--;
492
493         if (rx_dma == ib_conn->login_resp_dma)
494                 return;
495
496         outstanding = ib_conn->post_recv_buf_count;
497         if (outstanding + ISER_MIN_POSTED_RX <= ISER_QP_MAX_RECV_DTOS) {
498                 count = min(ISER_QP_MAX_RECV_DTOS - outstanding,
499                                                 ISER_MIN_POSTED_RX);
500                 err = iser_post_recvm(ib_conn, count);
501                 if (err)
502                         iser_err("posting %d rx bufs err %d\n", count, err);
503         }
504 }
505
506 void iser_snd_completion(struct iser_tx_desc *tx_desc,
507                         struct iser_conn *ib_conn)
508 {
509         struct iscsi_task *task;
510         struct iser_device *device = ib_conn->device;
511
512         if (tx_desc->type == ISCSI_TX_DATAOUT) {
513                 ib_dma_unmap_single(device->ib_device, tx_desc->dma_addr,
514                                         ISER_HEADERS_LEN, DMA_TO_DEVICE);
515                 kmem_cache_free(ig.desc_cache, tx_desc);
516         }
517
518         atomic_dec(&ib_conn->post_send_buf_count);
519
520         if (tx_desc->type == ISCSI_TX_CONTROL) {
521                 /* this arithmetic is legal by libiscsi dd_data allocation */
522                 task = (void *) ((long)(void *)tx_desc -
523                                   sizeof(struct iscsi_task));
524                 if (task->hdr->itt == RESERVED_ITT)
525                         iscsi_put_task(task);
526         }
527 }
528
529 void iser_task_rdma_init(struct iscsi_iser_task *iser_task)
530
531 {
532         iser_task->status = ISER_TASK_STATUS_INIT;
533
534         iser_task->dir[ISER_DIR_IN] = 0;
535         iser_task->dir[ISER_DIR_OUT] = 0;
536
537         iser_task->data[ISER_DIR_IN].data_len  = 0;
538         iser_task->data[ISER_DIR_OUT].data_len = 0;
539
540         memset(&iser_task->rdma_regd[ISER_DIR_IN], 0,
541                sizeof(struct iser_regd_buf));
542         memset(&iser_task->rdma_regd[ISER_DIR_OUT], 0,
543                sizeof(struct iser_regd_buf));
544 }
545
546 void iser_task_rdma_finalize(struct iscsi_iser_task *iser_task)
547 {
548         int is_rdma_aligned = 1;
549         struct iser_regd_buf *regd;
550
551         /* if we were reading, copy back to unaligned sglist,
552          * anyway dma_unmap and free the copy
553          */
554         if (iser_task->data_copy[ISER_DIR_IN].copy_buf != NULL) {
555                 is_rdma_aligned = 0;
556                 iser_finalize_rdma_unaligned_sg(iser_task, ISER_DIR_IN);
557         }
558         if (iser_task->data_copy[ISER_DIR_OUT].copy_buf != NULL) {
559                 is_rdma_aligned = 0;
560                 iser_finalize_rdma_unaligned_sg(iser_task, ISER_DIR_OUT);
561         }
562
563         if (iser_task->dir[ISER_DIR_IN]) {
564                 regd = &iser_task->rdma_regd[ISER_DIR_IN];
565                 if (regd->reg.is_fmr)
566                         iser_unreg_mem(&regd->reg);
567         }
568
569         if (iser_task->dir[ISER_DIR_OUT]) {
570                 regd = &iser_task->rdma_regd[ISER_DIR_OUT];
571                 if (regd->reg.is_fmr)
572                         iser_unreg_mem(&regd->reg);
573         }
574
575        /* if the data was unaligned, it was already unmapped and then copied */
576        if (is_rdma_aligned)
577                 iser_dma_unmap_task_data(iser_task);
578 }