1 /*
2 * Copyright (c) 2009-2010 Chelsio, 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
33 #include <linux/module.h>
34 #include <linux/moduleparam.h>
35 #include <rdma/ib_umem.h>
36 #include <linux/atomic.h>
37
38 #include "iw_cxgb4.h"
39
40 int use_dsgl = 0;
41 module_param(use_dsgl, int, 0644);
42 MODULE_PARM_DESC(use_dsgl, "Use DSGL for PBL/FastReg (default=0)");
43
44 #define T4_ULPTX_MIN_IO 32
45 #define C4IW_MAX_INLINE_SIZE 96
46 #define T4_ULPTX_MAX_DMA 1024
47 #define C4IW_INLINE_THRESHOLD 128
48
49 static int inline_threshold = C4IW_INLINE_THRESHOLD;
50 module_param(inline_threshold, int, 0644);
51 MODULE_PARM_DESC(inline_threshold, "inline vs dsgl threshold (default=128)");
52
_c4iw_write_mem_dma_aligned(struct c4iw_rdev * rdev,u32 addr,u32 len,dma_addr_t data,int wait)53 static int _c4iw_write_mem_dma_aligned(struct c4iw_rdev *rdev, u32 addr,
54 u32 len, dma_addr_t data, int wait)
55 {
56 struct sk_buff *skb;
57 struct ulp_mem_io *req;
58 struct ulptx_sgl *sgl;
59 u8 wr_len;
60 int ret = 0;
61 struct c4iw_wr_wait wr_wait;
62
63 addr &= 0x7FFFFFF;
64
65 if (wait)
66 c4iw_init_wr_wait(&wr_wait);
67 wr_len = roundup(sizeof(*req) + sizeof(*sgl), 16);
68
69 skb = alloc_skb(wr_len, GFP_KERNEL | __GFP_NOFAIL);
70 if (!skb)
71 return -ENOMEM;
72 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
73
74 req = (struct ulp_mem_io *)__skb_put(skb, wr_len);
75 memset(req, 0, wr_len);
76 INIT_ULPTX_WR(req, wr_len, 0, 0);
77 req->wr.wr_hi = cpu_to_be32(FW_WR_OP(FW_ULPTX_WR) |
78 (wait ? FW_WR_COMPL(1) : 0));
79 req->wr.wr_lo = wait ? (__force __be64)(unsigned long) &wr_wait : 0L;
80 req->wr.wr_mid = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(wr_len, 16)));
81 req->cmd = cpu_to_be32(ULPTX_CMD(ULP_TX_MEM_WRITE));
82 req->cmd |= cpu_to_be32(V_T5_ULP_MEMIO_ORDER(1));
83 req->dlen = cpu_to_be32(ULP_MEMIO_DATA_LEN(len>>5));
84 req->len16 = cpu_to_be32(DIV_ROUND_UP(wr_len-sizeof(req->wr), 16));
85 req->lock_addr = cpu_to_be32(ULP_MEMIO_ADDR(addr));
86
87 sgl = (struct ulptx_sgl *)(req + 1);
88 sgl->cmd_nsge = cpu_to_be32(ULPTX_CMD(ULP_TX_SC_DSGL) |
89 ULPTX_NSGE(1));
90 sgl->len0 = cpu_to_be32(len);
91 sgl->addr0 = cpu_to_be64(data);
92
93 ret = c4iw_ofld_send(rdev, skb);
94 if (ret)
95 return ret;
96 if (wait)
97 ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, 0, __func__);
98 return ret;
99 }
100
_c4iw_write_mem_inline(struct c4iw_rdev * rdev,u32 addr,u32 len,void * data)101 static int _c4iw_write_mem_inline(struct c4iw_rdev *rdev, u32 addr, u32 len,
102 void *data)
103 {
104 struct sk_buff *skb;
105 struct ulp_mem_io *req;
106 struct ulptx_idata *sc;
107 u8 wr_len, *to_dp, *from_dp;
108 int copy_len, num_wqe, i, ret = 0;
109 struct c4iw_wr_wait wr_wait;
110 __be32 cmd = cpu_to_be32(ULPTX_CMD(ULP_TX_MEM_WRITE));
111
112 if (is_t4(rdev->lldi.adapter_type))
113 cmd |= cpu_to_be32(ULP_MEMIO_ORDER(1));
114 else
115 cmd |= cpu_to_be32(V_T5_ULP_MEMIO_IMM(1));
116
117 addr &= 0x7FFFFFF;
118 PDBG("%s addr 0x%x len %u\n", __func__, addr, len);
119 num_wqe = DIV_ROUND_UP(len, C4IW_MAX_INLINE_SIZE);
120 c4iw_init_wr_wait(&wr_wait);
121 for (i = 0; i < num_wqe; i++) {
122
123 copy_len = len > C4IW_MAX_INLINE_SIZE ? C4IW_MAX_INLINE_SIZE :
124 len;
125 wr_len = roundup(sizeof *req + sizeof *sc +
126 roundup(copy_len, T4_ULPTX_MIN_IO), 16);
127
128 skb = alloc_skb(wr_len, GFP_KERNEL);
129 if (!skb)
130 return -ENOMEM;
131 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
132
133 req = (struct ulp_mem_io *)__skb_put(skb, wr_len);
134 memset(req, 0, wr_len);
135 INIT_ULPTX_WR(req, wr_len, 0, 0);
136
137 if (i == (num_wqe-1)) {
138 req->wr.wr_hi = cpu_to_be32(FW_WR_OP(FW_ULPTX_WR) |
139 FW_WR_COMPL(1));
140 req->wr.wr_lo = (__force __be64)(unsigned long) &wr_wait;
141 } else
142 req->wr.wr_hi = cpu_to_be32(FW_WR_OP(FW_ULPTX_WR));
143 req->wr.wr_mid = cpu_to_be32(
144 FW_WR_LEN16(DIV_ROUND_UP(wr_len, 16)));
145
146 req->cmd = cmd;
147 req->dlen = cpu_to_be32(ULP_MEMIO_DATA_LEN(
148 DIV_ROUND_UP(copy_len, T4_ULPTX_MIN_IO)));
149 req->len16 = cpu_to_be32(DIV_ROUND_UP(wr_len-sizeof(req->wr),
150 16));
151 req->lock_addr = cpu_to_be32(ULP_MEMIO_ADDR(addr + i * 3));
152
153 sc = (struct ulptx_idata *)(req + 1);
154 sc->cmd_more = cpu_to_be32(ULPTX_CMD(ULP_TX_SC_IMM));
155 sc->len = cpu_to_be32(roundup(copy_len, T4_ULPTX_MIN_IO));
156
157 to_dp = (u8 *)(sc + 1);
158 from_dp = (u8 *)data + i * C4IW_MAX_INLINE_SIZE;
159 if (data)
160 memcpy(to_dp, from_dp, copy_len);
161 else
162 memset(to_dp, 0, copy_len);
163 if (copy_len % T4_ULPTX_MIN_IO)
164 memset(to_dp + copy_len, 0, T4_ULPTX_MIN_IO -
165 (copy_len % T4_ULPTX_MIN_IO));
166 ret = c4iw_ofld_send(rdev, skb);
167 if (ret)
168 return ret;
169 len -= C4IW_MAX_INLINE_SIZE;
170 }
171
172 ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, 0, __func__);
173 return ret;
174 }
175
_c4iw_write_mem_dma(struct c4iw_rdev * rdev,u32 addr,u32 len,void * data)176 static int _c4iw_write_mem_dma(struct c4iw_rdev *rdev, u32 addr, u32 len, void *data)
177 {
178 u32 remain = len;
179 u32 dmalen;
180 int ret = 0;
181 dma_addr_t daddr;
182 dma_addr_t save;
183
184 daddr = dma_map_single(&rdev->lldi.pdev->dev, data, len, DMA_TO_DEVICE);
185 if (dma_mapping_error(&rdev->lldi.pdev->dev, daddr))
186 return -1;
187 save = daddr;
188
189 while (remain > inline_threshold) {
190 if (remain < T4_ULPTX_MAX_DMA) {
191 if (remain & ~T4_ULPTX_MIN_IO)
192 dmalen = remain & ~(T4_ULPTX_MIN_IO-1);
193 else
194 dmalen = remain;
195 } else
196 dmalen = T4_ULPTX_MAX_DMA;
197 remain -= dmalen;
198 ret = _c4iw_write_mem_dma_aligned(rdev, addr, dmalen, daddr,
199 !remain);
200 if (ret)
201 goto out;
202 addr += dmalen >> 5;
203 data += dmalen;
204 daddr += dmalen;
205 }
206 if (remain)
207 ret = _c4iw_write_mem_inline(rdev, addr, remain, data);
208 out:
209 dma_unmap_single(&rdev->lldi.pdev->dev, save, len, DMA_TO_DEVICE);
210 return ret;
211 }
212
213 /*
214 * write len bytes of data into addr (32B aligned address)
215 * If data is NULL, clear len byte of memory to zero.
216 */
write_adapter_mem(struct c4iw_rdev * rdev,u32 addr,u32 len,void * data)217 static int write_adapter_mem(struct c4iw_rdev *rdev, u32 addr, u32 len,
218 void *data)
219 {
220 if (is_t5(rdev->lldi.adapter_type) && use_dsgl) {
221 if (len > inline_threshold) {
222 if (_c4iw_write_mem_dma(rdev, addr, len, data)) {
223 printk_ratelimited(KERN_WARNING
224 "%s: dma map"
225 " failure (non fatal)\n",
226 pci_name(rdev->lldi.pdev));
227 return _c4iw_write_mem_inline(rdev, addr, len,
228 data);
229 } else
230 return 0;
231 } else
232 return _c4iw_write_mem_inline(rdev, addr, len, data);
233 } else
234 return _c4iw_write_mem_inline(rdev, addr, len, data);
235 }
236
237 /*
238 * Build and write a TPT entry.
239 * IN: stag key, pdid, perm, bind_enabled, zbva, to, len, page_size,
240 * pbl_size and pbl_addr
241 * OUT: stag index
242 */
write_tpt_entry(struct c4iw_rdev * rdev,u32 reset_tpt_entry,u32 * stag,u8 stag_state,u32 pdid,enum fw_ri_stag_type type,enum fw_ri_mem_perms perm,int bind_enabled,u32 zbva,u64 to,u64 len,u8 page_size,u32 pbl_size,u32 pbl_addr)243 static int write_tpt_entry(struct c4iw_rdev *rdev, u32 reset_tpt_entry,
244 u32 *stag, u8 stag_state, u32 pdid,
245 enum fw_ri_stag_type type, enum fw_ri_mem_perms perm,
246 int bind_enabled, u32 zbva, u64 to,
247 u64 len, u8 page_size, u32 pbl_size, u32 pbl_addr)
248 {
249 int err;
250 struct fw_ri_tpte tpt;
251 u32 stag_idx;
252 static atomic_t key;
253
254 if (c4iw_fatal_error(rdev))
255 return -EIO;
256
257 stag_state = stag_state > 0;
258 stag_idx = (*stag) >> 8;
259
260 if ((!reset_tpt_entry) && (*stag == T4_STAG_UNSET)) {
261 stag_idx = c4iw_get_resource(&rdev->resource.tpt_table);
262 if (!stag_idx) {
263 mutex_lock(&rdev->stats.lock);
264 rdev->stats.stag.fail++;
265 mutex_unlock(&rdev->stats.lock);
266 return -ENOMEM;
267 }
268 mutex_lock(&rdev->stats.lock);
269 rdev->stats.stag.cur += 32;
270 if (rdev->stats.stag.cur > rdev->stats.stag.max)
271 rdev->stats.stag.max = rdev->stats.stag.cur;
272 mutex_unlock(&rdev->stats.lock);
273 *stag = (stag_idx << 8) | (atomic_inc_return(&key) & 0xff);
274 }
275 PDBG("%s stag_state 0x%0x type 0x%0x pdid 0x%0x, stag_idx 0x%x\n",
276 __func__, stag_state, type, pdid, stag_idx);
277
278 /* write TPT entry */
279 if (reset_tpt_entry)
280 memset(&tpt, 0, sizeof(tpt));
281 else {
282 tpt.valid_to_pdid = cpu_to_be32(F_FW_RI_TPTE_VALID |
283 V_FW_RI_TPTE_STAGKEY((*stag & M_FW_RI_TPTE_STAGKEY)) |
284 V_FW_RI_TPTE_STAGSTATE(stag_state) |
285 V_FW_RI_TPTE_STAGTYPE(type) | V_FW_RI_TPTE_PDID(pdid));
286 tpt.locread_to_qpid = cpu_to_be32(V_FW_RI_TPTE_PERM(perm) |
287 (bind_enabled ? F_FW_RI_TPTE_MWBINDEN : 0) |
288 V_FW_RI_TPTE_ADDRTYPE((zbva ? FW_RI_ZERO_BASED_TO :
289 FW_RI_VA_BASED_TO))|
290 V_FW_RI_TPTE_PS(page_size));
291 tpt.nosnoop_pbladdr = !pbl_size ? 0 : cpu_to_be32(
292 V_FW_RI_TPTE_PBLADDR(PBL_OFF(rdev, pbl_addr)>>3));
293 tpt.len_lo = cpu_to_be32((u32)(len & 0xffffffffUL));
294 tpt.va_hi = cpu_to_be32((u32)(to >> 32));
295 tpt.va_lo_fbo = cpu_to_be32((u32)(to & 0xffffffffUL));
296 tpt.dca_mwbcnt_pstag = cpu_to_be32(0);
297 tpt.len_hi = cpu_to_be32((u32)(len >> 32));
298 }
299 err = write_adapter_mem(rdev, stag_idx +
300 (rdev->lldi.vr->stag.start >> 5),
301 sizeof(tpt), &tpt);
302
303 if (reset_tpt_entry) {
304 c4iw_put_resource(&rdev->resource.tpt_table, stag_idx);
305 mutex_lock(&rdev->stats.lock);
306 rdev->stats.stag.cur -= 32;
307 mutex_unlock(&rdev->stats.lock);
308 }
309 return err;
310 }
311
write_pbl(struct c4iw_rdev * rdev,__be64 * pbl,u32 pbl_addr,u32 pbl_size)312 static int write_pbl(struct c4iw_rdev *rdev, __be64 *pbl,
313 u32 pbl_addr, u32 pbl_size)
314 {
315 int err;
316
317 PDBG("%s *pdb_addr 0x%x, pbl_base 0x%x, pbl_size %d\n",
318 __func__, pbl_addr, rdev->lldi.vr->pbl.start,
319 pbl_size);
320
321 err = write_adapter_mem(rdev, pbl_addr >> 5, pbl_size << 3, pbl);
322 return err;
323 }
324
dereg_mem(struct c4iw_rdev * rdev,u32 stag,u32 pbl_size,u32 pbl_addr)325 static int dereg_mem(struct c4iw_rdev *rdev, u32 stag, u32 pbl_size,
326 u32 pbl_addr)
327 {
328 return write_tpt_entry(rdev, 1, &stag, 0, 0, 0, 0, 0, 0, 0UL, 0, 0,
329 pbl_size, pbl_addr);
330 }
331
allocate_window(struct c4iw_rdev * rdev,u32 * stag,u32 pdid)332 static int allocate_window(struct c4iw_rdev *rdev, u32 * stag, u32 pdid)
333 {
334 *stag = T4_STAG_UNSET;
335 return write_tpt_entry(rdev, 0, stag, 0, pdid, FW_RI_STAG_MW, 0, 0, 0,
336 0UL, 0, 0, 0, 0);
337 }
338
deallocate_window(struct c4iw_rdev * rdev,u32 stag)339 static int deallocate_window(struct c4iw_rdev *rdev, u32 stag)
340 {
341 return write_tpt_entry(rdev, 1, &stag, 0, 0, 0, 0, 0, 0, 0UL, 0, 0, 0,
342 0);
343 }
344
allocate_stag(struct c4iw_rdev * rdev,u32 * stag,u32 pdid,u32 pbl_size,u32 pbl_addr)345 static int allocate_stag(struct c4iw_rdev *rdev, u32 *stag, u32 pdid,
346 u32 pbl_size, u32 pbl_addr)
347 {
348 *stag = T4_STAG_UNSET;
349 return write_tpt_entry(rdev, 0, stag, 0, pdid, FW_RI_STAG_NSMR, 0, 0, 0,
350 0UL, 0, 0, pbl_size, pbl_addr);
351 }
352
finish_mem_reg(struct c4iw_mr * mhp,u32 stag)353 static int finish_mem_reg(struct c4iw_mr *mhp, u32 stag)
354 {
355 u32 mmid;
356
357 mhp->attr.state = 1;
358 mhp->attr.stag = stag;
359 mmid = stag >> 8;
360 mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
361 PDBG("%s mmid 0x%x mhp %p\n", __func__, mmid, mhp);
362 return insert_handle(mhp->rhp, &mhp->rhp->mmidr, mhp, mmid);
363 }
364
register_mem(struct c4iw_dev * rhp,struct c4iw_pd * php,struct c4iw_mr * mhp,int shift)365 static int register_mem(struct c4iw_dev *rhp, struct c4iw_pd *php,
366 struct c4iw_mr *mhp, int shift)
367 {
368 u32 stag = T4_STAG_UNSET;
369 int ret;
370
371 ret = write_tpt_entry(&rhp->rdev, 0, &stag, 1, mhp->attr.pdid,
372 FW_RI_STAG_NSMR, mhp->attr.perms,
373 mhp->attr.mw_bind_enable, mhp->attr.zbva,
374 mhp->attr.va_fbo, mhp->attr.len, shift - 12,
375 mhp->attr.pbl_size, mhp->attr.pbl_addr);
376 if (ret)
377 return ret;
378
379 ret = finish_mem_reg(mhp, stag);
380 if (ret)
381 dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
382 mhp->attr.pbl_addr);
383 return ret;
384 }
385
reregister_mem(struct c4iw_dev * rhp,struct c4iw_pd * php,struct c4iw_mr * mhp,int shift,int npages)386 static int reregister_mem(struct c4iw_dev *rhp, struct c4iw_pd *php,
387 struct c4iw_mr *mhp, int shift, int npages)
388 {
389 u32 stag;
390 int ret;
391
392 if (npages > mhp->attr.pbl_size)
393 return -ENOMEM;
394
395 stag = mhp->attr.stag;
396 ret = write_tpt_entry(&rhp->rdev, 0, &stag, 1, mhp->attr.pdid,
397 FW_RI_STAG_NSMR, mhp->attr.perms,
398 mhp->attr.mw_bind_enable, mhp->attr.zbva,
399 mhp->attr.va_fbo, mhp->attr.len, shift - 12,
400 mhp->attr.pbl_size, mhp->attr.pbl_addr);
401 if (ret)
402 return ret;
403
404 ret = finish_mem_reg(mhp, stag);
405 if (ret)
406 dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
407 mhp->attr.pbl_addr);
408
409 return ret;
410 }
411
alloc_pbl(struct c4iw_mr * mhp,int npages)412 static int alloc_pbl(struct c4iw_mr *mhp, int npages)
413 {
414 mhp->attr.pbl_addr = c4iw_pblpool_alloc(&mhp->rhp->rdev,
415 npages << 3);
416
417 if (!mhp->attr.pbl_addr)
418 return -ENOMEM;
419
420 mhp->attr.pbl_size = npages;
421
422 return 0;
423 }
424
build_phys_page_list(struct ib_phys_buf * buffer_list,int num_phys_buf,u64 * iova_start,u64 * total_size,int * npages,int * shift,__be64 ** page_list)425 static int build_phys_page_list(struct ib_phys_buf *buffer_list,
426 int num_phys_buf, u64 *iova_start,
427 u64 *total_size, int *npages,
428 int *shift, __be64 **page_list)
429 {
430 u64 mask;
431 int i, j, n;
432
433 mask = 0;
434 *total_size = 0;
435 for (i = 0; i < num_phys_buf; ++i) {
436 if (i != 0 && buffer_list[i].addr & ~PAGE_MASK)
437 return -EINVAL;
438 if (i != 0 && i != num_phys_buf - 1 &&
439 (buffer_list[i].size & ~PAGE_MASK))
440 return -EINVAL;
441 *total_size += buffer_list[i].size;
442 if (i > 0)
443 mask |= buffer_list[i].addr;
444 else
445 mask |= buffer_list[i].addr & PAGE_MASK;
446 if (i != num_phys_buf - 1)
447 mask |= buffer_list[i].addr + buffer_list[i].size;
448 else
449 mask |= (buffer_list[i].addr + buffer_list[i].size +
450 PAGE_SIZE - 1) & PAGE_MASK;
451 }
452
453 if (*total_size > 0xFFFFFFFFULL)
454 return -ENOMEM;
455
456 /* Find largest page shift we can use to cover buffers */
457 for (*shift = PAGE_SHIFT; *shift < 27; ++(*shift))
458 if ((1ULL << *shift) & mask)
459 break;
460
461 buffer_list[0].size += buffer_list[0].addr & ((1ULL << *shift) - 1);
462 buffer_list[0].addr &= ~0ull << *shift;
463
464 *npages = 0;
465 for (i = 0; i < num_phys_buf; ++i)
466 *npages += (buffer_list[i].size +
467 (1ULL << *shift) - 1) >> *shift;
468
469 if (!*npages)
470 return -EINVAL;
471
472 *page_list = kmalloc(sizeof(u64) * *npages, GFP_KERNEL);
473 if (!*page_list)
474 return -ENOMEM;
475
476 n = 0;
477 for (i = 0; i < num_phys_buf; ++i)
478 for (j = 0;
479 j < (buffer_list[i].size + (1ULL << *shift) - 1) >> *shift;
480 ++j)
481 (*page_list)[n++] = cpu_to_be64(buffer_list[i].addr +
482 ((u64) j << *shift));
483
484 PDBG("%s va 0x%llx mask 0x%llx shift %d len %lld pbl_size %d\n",
485 __func__, (unsigned long long)*iova_start,
486 (unsigned long long)mask, *shift, (unsigned long long)*total_size,
487 *npages);
488
489 return 0;
490
491 }
492
c4iw_reregister_phys_mem(struct ib_mr * mr,int mr_rereg_mask,struct ib_pd * pd,struct ib_phys_buf * buffer_list,int num_phys_buf,int acc,u64 * iova_start)493 int c4iw_reregister_phys_mem(struct ib_mr *mr, int mr_rereg_mask,
494 struct ib_pd *pd, struct ib_phys_buf *buffer_list,
495 int num_phys_buf, int acc, u64 *iova_start)
496 {
497
498 struct c4iw_mr mh, *mhp;
499 struct c4iw_pd *php;
500 struct c4iw_dev *rhp;
501 __be64 *page_list = NULL;
502 int shift = 0;
503 u64 total_size;
504 int npages;
505 int ret;
506
507 PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
508
509 /* There can be no memory windows */
510 if (atomic_read(&mr->usecnt))
511 return -EINVAL;
512
513 mhp = to_c4iw_mr(mr);
514 rhp = mhp->rhp;
515 php = to_c4iw_pd(mr->pd);
516
517 /* make sure we are on the same adapter */
518 if (rhp != php->rhp)
519 return -EINVAL;
520
521 memcpy(&mh, mhp, sizeof *mhp);
522
523 if (mr_rereg_mask & IB_MR_REREG_PD)
524 php = to_c4iw_pd(pd);
525 if (mr_rereg_mask & IB_MR_REREG_ACCESS) {
526 mh.attr.perms = c4iw_ib_to_tpt_access(acc);
527 mh.attr.mw_bind_enable = (acc & IB_ACCESS_MW_BIND) ==
528 IB_ACCESS_MW_BIND;
529 }
530 if (mr_rereg_mask & IB_MR_REREG_TRANS) {
531 ret = build_phys_page_list(buffer_list, num_phys_buf,
532 iova_start,
533 &total_size, &npages,
534 &shift, &page_list);
535 if (ret)
536 return ret;
537 }
538
539 ret = reregister_mem(rhp, php, &mh, shift, npages);
540 kfree(page_list);
541 if (ret)
542 return ret;
543 if (mr_rereg_mask & IB_MR_REREG_PD)
544 mhp->attr.pdid = php->pdid;
545 if (mr_rereg_mask & IB_MR_REREG_ACCESS)
546 mhp->attr.perms = c4iw_ib_to_tpt_access(acc);
547 if (mr_rereg_mask & IB_MR_REREG_TRANS) {
548 mhp->attr.zbva = 0;
549 mhp->attr.va_fbo = *iova_start;
550 mhp->attr.page_size = shift - 12;
551 mhp->attr.len = (u32) total_size;
552 mhp->attr.pbl_size = npages;
553 }
554
555 return 0;
556 }
557
c4iw_register_phys_mem(struct ib_pd * pd,struct ib_phys_buf * buffer_list,int num_phys_buf,int acc,u64 * iova_start)558 struct ib_mr *c4iw_register_phys_mem(struct ib_pd *pd,
559 struct ib_phys_buf *buffer_list,
560 int num_phys_buf, int acc, u64 *iova_start)
561 {
562 __be64 *page_list;
563 int shift;
564 u64 total_size;
565 int npages;
566 struct c4iw_dev *rhp;
567 struct c4iw_pd *php;
568 struct c4iw_mr *mhp;
569 int ret;
570
571 PDBG("%s ib_pd %p\n", __func__, pd);
572 php = to_c4iw_pd(pd);
573 rhp = php->rhp;
574
575 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
576 if (!mhp)
577 return ERR_PTR(-ENOMEM);
578
579 mhp->rhp = rhp;
580
581 /* First check that we have enough alignment */
582 if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
583 ret = -EINVAL;
584 goto err;
585 }
586
587 if (num_phys_buf > 1 &&
588 ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
589 ret = -EINVAL;
590 goto err;
591 }
592
593 ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
594 &total_size, &npages, &shift,
595 &page_list);
596 if (ret)
597 goto err;
598
599 ret = alloc_pbl(mhp, npages);
600 if (ret) {
601 kfree(page_list);
602 goto err;
603 }
604
605 ret = write_pbl(&mhp->rhp->rdev, page_list, mhp->attr.pbl_addr,
606 npages);
607 kfree(page_list);
608 if (ret)
609 goto err_pbl;
610
611 mhp->attr.pdid = php->pdid;
612 mhp->attr.zbva = 0;
613
614 mhp->attr.perms = c4iw_ib_to_tpt_access(acc);
615 mhp->attr.va_fbo = *iova_start;
616 mhp->attr.page_size = shift - 12;
617
618 mhp->attr.len = (u32) total_size;
619 mhp->attr.pbl_size = npages;
620 ret = register_mem(rhp, php, mhp, shift);
621 if (ret)
622 goto err_pbl;
623
624 return &mhp->ibmr;
625
626 err_pbl:
627 c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr,
628 mhp->attr.pbl_size << 3);
629
630 err:
631 kfree(mhp);
632 return ERR_PTR(ret);
633
634 }
635
c4iw_get_dma_mr(struct ib_pd * pd,int acc)636 struct ib_mr *c4iw_get_dma_mr(struct ib_pd *pd, int acc)
637 {
638 struct c4iw_dev *rhp;
639 struct c4iw_pd *php;
640 struct c4iw_mr *mhp;
641 int ret;
642 u32 stag = T4_STAG_UNSET;
643
644 PDBG("%s ib_pd %p\n", __func__, pd);
645 php = to_c4iw_pd(pd);
646 rhp = php->rhp;
647
648 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
649 if (!mhp)
650 return ERR_PTR(-ENOMEM);
651
652 mhp->rhp = rhp;
653 mhp->attr.pdid = php->pdid;
654 mhp->attr.perms = c4iw_ib_to_tpt_access(acc);
655 mhp->attr.mw_bind_enable = (acc&IB_ACCESS_MW_BIND) == IB_ACCESS_MW_BIND;
656 mhp->attr.zbva = 0;
657 mhp->attr.va_fbo = 0;
658 mhp->attr.page_size = 0;
659 mhp->attr.len = ~0UL;
660 mhp->attr.pbl_size = 0;
661
662 ret = write_tpt_entry(&rhp->rdev, 0, &stag, 1, php->pdid,
663 FW_RI_STAG_NSMR, mhp->attr.perms,
664 mhp->attr.mw_bind_enable, 0, 0, ~0UL, 0, 0, 0);
665 if (ret)
666 goto err1;
667
668 ret = finish_mem_reg(mhp, stag);
669 if (ret)
670 goto err2;
671 return &mhp->ibmr;
672 err2:
673 dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
674 mhp->attr.pbl_addr);
675 err1:
676 kfree(mhp);
677 return ERR_PTR(ret);
678 }
679
c4iw_reg_user_mr(struct ib_pd * pd,u64 start,u64 length,u64 virt,int acc,struct ib_udata * udata)680 struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
681 u64 virt, int acc, struct ib_udata *udata)
682 {
683 __be64 *pages;
684 int shift, n, len;
685 int i, k, entry;
686 int err = 0;
687 struct scatterlist *sg;
688 struct c4iw_dev *rhp;
689 struct c4iw_pd *php;
690 struct c4iw_mr *mhp;
691
692 PDBG("%s ib_pd %p\n", __func__, pd);
693
694 if (length == ~0ULL)
695 return ERR_PTR(-EINVAL);
696
697 if ((length + start) < start)
698 return ERR_PTR(-EINVAL);
699
700 php = to_c4iw_pd(pd);
701 rhp = php->rhp;
702 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
703 if (!mhp)
704 return ERR_PTR(-ENOMEM);
705
706 mhp->rhp = rhp;
707
708 mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
709 if (IS_ERR(mhp->umem)) {
710 err = PTR_ERR(mhp->umem);
711 kfree(mhp);
712 return ERR_PTR(err);
713 }
714
715 shift = ffs(mhp->umem->page_size) - 1;
716
717 n = mhp->umem->nmap;
718 err = alloc_pbl(mhp, n);
719 if (err)
720 goto err;
721
722 pages = (__be64 *) __get_free_page(GFP_KERNEL);
723 if (!pages) {
724 err = -ENOMEM;
725 goto err_pbl;
726 }
727
728 i = n = 0;
729
730 for_each_sg(mhp->umem->sg_head.sgl, sg, mhp->umem->nmap, entry) {
731 len = sg_dma_len(sg) >> shift;
732 for (k = 0; k < len; ++k) {
733 pages[i++] = cpu_to_be64(sg_dma_address(sg) +
734 mhp->umem->page_size * k);
735 if (i == PAGE_SIZE / sizeof *pages) {
736 err = write_pbl(&mhp->rhp->rdev,
737 pages,
738 mhp->attr.pbl_addr + (n << 3), i);
739 if (err)
740 goto pbl_done;
741 n += i;
742 i = 0;
743 }
744 }
745 }
746
747 if (i)
748 err = write_pbl(&mhp->rhp->rdev, pages,
749 mhp->attr.pbl_addr + (n << 3), i);
750
751 pbl_done:
752 free_page((unsigned long) pages);
753 if (err)
754 goto err_pbl;
755
756 mhp->attr.pdid = php->pdid;
757 mhp->attr.zbva = 0;
758 mhp->attr.perms = c4iw_ib_to_tpt_access(acc);
759 mhp->attr.va_fbo = virt;
760 mhp->attr.page_size = shift - 12;
761 mhp->attr.len = length;
762
763 err = register_mem(rhp, php, mhp, shift);
764 if (err)
765 goto err_pbl;
766
767 return &mhp->ibmr;
768
769 err_pbl:
770 c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr,
771 mhp->attr.pbl_size << 3);
772
773 err:
774 ib_umem_release(mhp->umem);
775 kfree(mhp);
776 return ERR_PTR(err);
777 }
778
c4iw_alloc_mw(struct ib_pd * pd,enum ib_mw_type type)779 struct ib_mw *c4iw_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
780 {
781 struct c4iw_dev *rhp;
782 struct c4iw_pd *php;
783 struct c4iw_mw *mhp;
784 u32 mmid;
785 u32 stag = 0;
786 int ret;
787
788 if (type != IB_MW_TYPE_1)
789 return ERR_PTR(-EINVAL);
790
791 php = to_c4iw_pd(pd);
792 rhp = php->rhp;
793 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
794 if (!mhp)
795 return ERR_PTR(-ENOMEM);
796 ret = allocate_window(&rhp->rdev, &stag, php->pdid);
797 if (ret) {
798 kfree(mhp);
799 return ERR_PTR(ret);
800 }
801 mhp->rhp = rhp;
802 mhp->attr.pdid = php->pdid;
803 mhp->attr.type = FW_RI_STAG_MW;
804 mhp->attr.stag = stag;
805 mmid = (stag) >> 8;
806 mhp->ibmw.rkey = stag;
807 if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
808 deallocate_window(&rhp->rdev, mhp->attr.stag);
809 kfree(mhp);
810 return ERR_PTR(-ENOMEM);
811 }
812 PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
813 return &(mhp->ibmw);
814 }
815
c4iw_dealloc_mw(struct ib_mw * mw)816 int c4iw_dealloc_mw(struct ib_mw *mw)
817 {
818 struct c4iw_dev *rhp;
819 struct c4iw_mw *mhp;
820 u32 mmid;
821
822 mhp = to_c4iw_mw(mw);
823 rhp = mhp->rhp;
824 mmid = (mw->rkey) >> 8;
825 remove_handle(rhp, &rhp->mmidr, mmid);
826 deallocate_window(&rhp->rdev, mhp->attr.stag);
827 kfree(mhp);
828 PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
829 return 0;
830 }
831
c4iw_alloc_fast_reg_mr(struct ib_pd * pd,int pbl_depth)832 struct ib_mr *c4iw_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth)
833 {
834 struct c4iw_dev *rhp;
835 struct c4iw_pd *php;
836 struct c4iw_mr *mhp;
837 u32 mmid;
838 u32 stag = 0;
839 int ret = 0;
840
841 php = to_c4iw_pd(pd);
842 rhp = php->rhp;
843 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
844 if (!mhp) {
845 ret = -ENOMEM;
846 goto err;
847 }
848
849 mhp->rhp = rhp;
850 ret = alloc_pbl(mhp, pbl_depth);
851 if (ret)
852 goto err1;
853 mhp->attr.pbl_size = pbl_depth;
854 ret = allocate_stag(&rhp->rdev, &stag, php->pdid,
855 mhp->attr.pbl_size, mhp->attr.pbl_addr);
856 if (ret)
857 goto err2;
858 mhp->attr.pdid = php->pdid;
859 mhp->attr.type = FW_RI_STAG_NSMR;
860 mhp->attr.stag = stag;
861 mhp->attr.state = 1;
862 mmid = (stag) >> 8;
863 mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
864 if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
865 ret = -ENOMEM;
866 goto err3;
867 }
868
869 PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
870 return &(mhp->ibmr);
871 err3:
872 dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size,
873 mhp->attr.pbl_addr);
874 err2:
875 c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr,
876 mhp->attr.pbl_size << 3);
877 err1:
878 kfree(mhp);
879 err:
880 return ERR_PTR(ret);
881 }
882
c4iw_alloc_fastreg_pbl(struct ib_device * device,int page_list_len)883 struct ib_fast_reg_page_list *c4iw_alloc_fastreg_pbl(struct ib_device *device,
884 int page_list_len)
885 {
886 struct c4iw_fr_page_list *c4pl;
887 struct c4iw_dev *dev = to_c4iw_dev(device);
888 dma_addr_t dma_addr;
889 int pll_len = roundup(page_list_len * sizeof(u64), 32);
890
891 c4pl = kmalloc(sizeof(*c4pl), GFP_KERNEL);
892 if (!c4pl)
893 return ERR_PTR(-ENOMEM);
894
895 c4pl->ibpl.page_list = dma_alloc_coherent(&dev->rdev.lldi.pdev->dev,
896 pll_len, &dma_addr,
897 GFP_KERNEL);
898 if (!c4pl->ibpl.page_list) {
899 kfree(c4pl);
900 return ERR_PTR(-ENOMEM);
901 }
902 dma_unmap_addr_set(c4pl, mapping, dma_addr);
903 c4pl->dma_addr = dma_addr;
904 c4pl->dev = dev;
905 c4pl->pll_len = pll_len;
906
907 PDBG("%s c4pl %p pll_len %u page_list %p dma_addr %pad\n",
908 __func__, c4pl, c4pl->pll_len, c4pl->ibpl.page_list,
909 &c4pl->dma_addr);
910
911 return &c4pl->ibpl;
912 }
913
c4iw_free_fastreg_pbl(struct ib_fast_reg_page_list * ibpl)914 void c4iw_free_fastreg_pbl(struct ib_fast_reg_page_list *ibpl)
915 {
916 struct c4iw_fr_page_list *c4pl = to_c4iw_fr_page_list(ibpl);
917
918 PDBG("%s c4pl %p pll_len %u page_list %p dma_addr %pad\n",
919 __func__, c4pl, c4pl->pll_len, c4pl->ibpl.page_list,
920 &c4pl->dma_addr);
921
922 dma_free_coherent(&c4pl->dev->rdev.lldi.pdev->dev,
923 c4pl->pll_len,
924 c4pl->ibpl.page_list, dma_unmap_addr(c4pl, mapping));
925 kfree(c4pl);
926 }
927
c4iw_dereg_mr(struct ib_mr * ib_mr)928 int c4iw_dereg_mr(struct ib_mr *ib_mr)
929 {
930 struct c4iw_dev *rhp;
931 struct c4iw_mr *mhp;
932 u32 mmid;
933
934 PDBG("%s ib_mr %p\n", __func__, ib_mr);
935 /* There can be no memory windows */
936 if (atomic_read(&ib_mr->usecnt))
937 return -EINVAL;
938
939 mhp = to_c4iw_mr(ib_mr);
940 rhp = mhp->rhp;
941 mmid = mhp->attr.stag >> 8;
942 remove_handle(rhp, &rhp->mmidr, mmid);
943 dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
944 mhp->attr.pbl_addr);
945 if (mhp->attr.pbl_size)
946 c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr,
947 mhp->attr.pbl_size << 3);
948 if (mhp->kva)
949 kfree((void *) (unsigned long) mhp->kva);
950 if (mhp->umem)
951 ib_umem_release(mhp->umem);
952 PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
953 kfree(mhp);
954 return 0;
955 }
956