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 #include <rdma/ib_user_verbs.h> 38 39 #include "iw_cxgb4.h" 40 41 int use_dsgl = 1; 42 module_param(use_dsgl, int, 0644); 43 MODULE_PARM_DESC(use_dsgl, "Use DSGL for PBL/FastReg (default=1) (DEPRECATED)"); 44 45 #define T4_ULPTX_MIN_IO 32 46 #define C4IW_MAX_INLINE_SIZE 96 47 #define T4_ULPTX_MAX_DMA 1024 48 #define C4IW_INLINE_THRESHOLD 128 49 50 static int inline_threshold = C4IW_INLINE_THRESHOLD; 51 module_param(inline_threshold, int, 0644); 52 MODULE_PARM_DESC(inline_threshold, "inline vs dsgl threshold (default=128)"); 53 54 static int mr_exceeds_hw_limits(struct c4iw_dev *dev, u64 length) 55 { 56 return (is_t4(dev->rdev.lldi.adapter_type) || 57 is_t5(dev->rdev.lldi.adapter_type)) && 58 length >= 8*1024*1024*1024ULL; 59 } 60 61 static int _c4iw_write_mem_dma_aligned(struct c4iw_rdev *rdev, u32 addr, 62 u32 len, dma_addr_t data, 63 struct sk_buff *skb, 64 struct c4iw_wr_wait *wr_waitp) 65 { 66 struct ulp_mem_io *req; 67 struct ulptx_sgl *sgl; 68 u8 wr_len; 69 int ret = 0; 70 71 addr &= 0x7FFFFFF; 72 73 if (wr_waitp) 74 c4iw_init_wr_wait(wr_waitp); 75 wr_len = roundup(sizeof(*req) + sizeof(*sgl), 16); 76 77 if (!skb) { 78 skb = alloc_skb(wr_len, GFP_KERNEL | __GFP_NOFAIL); 79 if (!skb) 80 return -ENOMEM; 81 } 82 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0); 83 84 req = __skb_put_zero(skb, wr_len); 85 INIT_ULPTX_WR(req, wr_len, 0, 0); 86 req->wr.wr_hi = cpu_to_be32(FW_WR_OP_V(FW_ULPTX_WR) | 87 (wr_waitp ? FW_WR_COMPL_F : 0)); 88 req->wr.wr_lo = wr_waitp ? (__force __be64)(unsigned long)wr_waitp : 0L; 89 req->wr.wr_mid = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(wr_len, 16))); 90 req->cmd = cpu_to_be32(ULPTX_CMD_V(ULP_TX_MEM_WRITE) | 91 T5_ULP_MEMIO_ORDER_V(1) | 92 T5_ULP_MEMIO_FID_V(rdev->lldi.rxq_ids[0])); 93 req->dlen = cpu_to_be32(ULP_MEMIO_DATA_LEN_V(len>>5)); 94 req->len16 = cpu_to_be32(DIV_ROUND_UP(wr_len-sizeof(req->wr), 16)); 95 req->lock_addr = cpu_to_be32(ULP_MEMIO_ADDR_V(addr)); 96 97 sgl = (struct ulptx_sgl *)(req + 1); 98 sgl->cmd_nsge = cpu_to_be32(ULPTX_CMD_V(ULP_TX_SC_DSGL) | 99 ULPTX_NSGE_V(1)); 100 sgl->len0 = cpu_to_be32(len); 101 sgl->addr0 = cpu_to_be64(data); 102 103 if (wr_waitp) 104 ret = c4iw_ref_send_wait(rdev, skb, wr_waitp, 0, 0, __func__); 105 else 106 ret = c4iw_ofld_send(rdev, skb); 107 return ret; 108 } 109 110 static int _c4iw_write_mem_inline(struct c4iw_rdev *rdev, u32 addr, u32 len, 111 void *data, struct sk_buff *skb, 112 struct c4iw_wr_wait *wr_waitp) 113 { 114 struct ulp_mem_io *req; 115 struct ulptx_idata *sc; 116 u8 wr_len, *to_dp, *from_dp; 117 int copy_len, num_wqe, i, ret = 0; 118 __be32 cmd = cpu_to_be32(ULPTX_CMD_V(ULP_TX_MEM_WRITE)); 119 120 if (is_t4(rdev->lldi.adapter_type)) 121 cmd |= cpu_to_be32(ULP_MEMIO_ORDER_F); 122 else 123 cmd |= cpu_to_be32(T5_ULP_MEMIO_IMM_F); 124 125 addr &= 0x7FFFFFF; 126 pr_debug("addr 0x%x len %u\n", addr, len); 127 num_wqe = DIV_ROUND_UP(len, C4IW_MAX_INLINE_SIZE); 128 c4iw_init_wr_wait(wr_waitp); 129 for (i = 0; i < num_wqe; i++) { 130 131 copy_len = len > C4IW_MAX_INLINE_SIZE ? C4IW_MAX_INLINE_SIZE : 132 len; 133 wr_len = roundup(sizeof *req + sizeof *sc + 134 roundup(copy_len, T4_ULPTX_MIN_IO), 16); 135 136 if (!skb) { 137 skb = alloc_skb(wr_len, GFP_KERNEL | __GFP_NOFAIL); 138 if (!skb) 139 return -ENOMEM; 140 } 141 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0); 142 143 req = __skb_put_zero(skb, wr_len); 144 INIT_ULPTX_WR(req, wr_len, 0, 0); 145 146 if (i == (num_wqe-1)) { 147 req->wr.wr_hi = cpu_to_be32(FW_WR_OP_V(FW_ULPTX_WR) | 148 FW_WR_COMPL_F); 149 req->wr.wr_lo = (__force __be64)(unsigned long)wr_waitp; 150 } else 151 req->wr.wr_hi = cpu_to_be32(FW_WR_OP_V(FW_ULPTX_WR)); 152 req->wr.wr_mid = cpu_to_be32( 153 FW_WR_LEN16_V(DIV_ROUND_UP(wr_len, 16))); 154 155 req->cmd = cmd; 156 req->dlen = cpu_to_be32(ULP_MEMIO_DATA_LEN_V( 157 DIV_ROUND_UP(copy_len, T4_ULPTX_MIN_IO))); 158 req->len16 = cpu_to_be32(DIV_ROUND_UP(wr_len-sizeof(req->wr), 159 16)); 160 req->lock_addr = cpu_to_be32(ULP_MEMIO_ADDR_V(addr + i * 3)); 161 162 sc = (struct ulptx_idata *)(req + 1); 163 sc->cmd_more = cpu_to_be32(ULPTX_CMD_V(ULP_TX_SC_IMM)); 164 sc->len = cpu_to_be32(roundup(copy_len, T4_ULPTX_MIN_IO)); 165 166 to_dp = (u8 *)(sc + 1); 167 from_dp = (u8 *)data + i * C4IW_MAX_INLINE_SIZE; 168 if (data) 169 memcpy(to_dp, from_dp, copy_len); 170 else 171 memset(to_dp, 0, copy_len); 172 if (copy_len % T4_ULPTX_MIN_IO) 173 memset(to_dp + copy_len, 0, T4_ULPTX_MIN_IO - 174 (copy_len % T4_ULPTX_MIN_IO)); 175 if (i == (num_wqe-1)) 176 ret = c4iw_ref_send_wait(rdev, skb, wr_waitp, 0, 0, 177 __func__); 178 else 179 ret = c4iw_ofld_send(rdev, skb); 180 if (ret) 181 break; 182 skb = NULL; 183 len -= C4IW_MAX_INLINE_SIZE; 184 } 185 186 return ret; 187 } 188 189 static int _c4iw_write_mem_dma(struct c4iw_rdev *rdev, u32 addr, u32 len, 190 void *data, struct sk_buff *skb, 191 struct c4iw_wr_wait *wr_waitp) 192 { 193 u32 remain = len; 194 u32 dmalen; 195 int ret = 0; 196 dma_addr_t daddr; 197 dma_addr_t save; 198 199 daddr = dma_map_single(&rdev->lldi.pdev->dev, data, len, DMA_TO_DEVICE); 200 if (dma_mapping_error(&rdev->lldi.pdev->dev, daddr)) 201 return -1; 202 save = daddr; 203 204 while (remain > inline_threshold) { 205 if (remain < T4_ULPTX_MAX_DMA) { 206 if (remain & ~T4_ULPTX_MIN_IO) 207 dmalen = remain & ~(T4_ULPTX_MIN_IO-1); 208 else 209 dmalen = remain; 210 } else 211 dmalen = T4_ULPTX_MAX_DMA; 212 remain -= dmalen; 213 ret = _c4iw_write_mem_dma_aligned(rdev, addr, dmalen, daddr, 214 skb, remain ? NULL : wr_waitp); 215 if (ret) 216 goto out; 217 addr += dmalen >> 5; 218 data += dmalen; 219 daddr += dmalen; 220 } 221 if (remain) 222 ret = _c4iw_write_mem_inline(rdev, addr, remain, data, skb, 223 wr_waitp); 224 out: 225 dma_unmap_single(&rdev->lldi.pdev->dev, save, len, DMA_TO_DEVICE); 226 return ret; 227 } 228 229 /* 230 * write len bytes of data into addr (32B aligned address) 231 * If data is NULL, clear len byte of memory to zero. 232 */ 233 static int write_adapter_mem(struct c4iw_rdev *rdev, u32 addr, u32 len, 234 void *data, struct sk_buff *skb, 235 struct c4iw_wr_wait *wr_waitp) 236 { 237 int ret; 238 239 if (!rdev->lldi.ulptx_memwrite_dsgl || !use_dsgl) { 240 ret = _c4iw_write_mem_inline(rdev, addr, len, data, skb, 241 wr_waitp); 242 goto out; 243 } 244 245 if (len <= inline_threshold) { 246 ret = _c4iw_write_mem_inline(rdev, addr, len, data, skb, 247 wr_waitp); 248 goto out; 249 } 250 251 ret = _c4iw_write_mem_dma(rdev, addr, len, data, skb, wr_waitp); 252 if (ret) { 253 pr_warn_ratelimited("%s: dma map failure (non fatal)\n", 254 pci_name(rdev->lldi.pdev)); 255 ret = _c4iw_write_mem_inline(rdev, addr, len, data, skb, 256 wr_waitp); 257 } 258 out: 259 return ret; 260 261 } 262 263 /* 264 * Build and write a TPT entry. 265 * IN: stag key, pdid, perm, bind_enabled, zbva, to, len, page_size, 266 * pbl_size and pbl_addr 267 * OUT: stag index 268 */ 269 static int write_tpt_entry(struct c4iw_rdev *rdev, u32 reset_tpt_entry, 270 u32 *stag, u8 stag_state, u32 pdid, 271 enum fw_ri_stag_type type, enum fw_ri_mem_perms perm, 272 int bind_enabled, u32 zbva, u64 to, 273 u64 len, u8 page_size, u32 pbl_size, u32 pbl_addr, 274 struct sk_buff *skb, struct c4iw_wr_wait *wr_waitp) 275 { 276 int err; 277 struct fw_ri_tpte tpt; 278 u32 stag_idx; 279 static atomic_t key; 280 281 if (c4iw_fatal_error(rdev)) 282 return -EIO; 283 284 stag_state = stag_state > 0; 285 stag_idx = (*stag) >> 8; 286 287 if ((!reset_tpt_entry) && (*stag == T4_STAG_UNSET)) { 288 stag_idx = c4iw_get_resource(&rdev->resource.tpt_table); 289 if (!stag_idx) { 290 mutex_lock(&rdev->stats.lock); 291 rdev->stats.stag.fail++; 292 mutex_unlock(&rdev->stats.lock); 293 return -ENOMEM; 294 } 295 mutex_lock(&rdev->stats.lock); 296 rdev->stats.stag.cur += 32; 297 if (rdev->stats.stag.cur > rdev->stats.stag.max) 298 rdev->stats.stag.max = rdev->stats.stag.cur; 299 mutex_unlock(&rdev->stats.lock); 300 *stag = (stag_idx << 8) | (atomic_inc_return(&key) & 0xff); 301 } 302 pr_debug("stag_state 0x%0x type 0x%0x pdid 0x%0x, stag_idx 0x%x\n", 303 stag_state, type, pdid, stag_idx); 304 305 /* write TPT entry */ 306 if (reset_tpt_entry) 307 memset(&tpt, 0, sizeof(tpt)); 308 else { 309 tpt.valid_to_pdid = cpu_to_be32(FW_RI_TPTE_VALID_F | 310 FW_RI_TPTE_STAGKEY_V((*stag & FW_RI_TPTE_STAGKEY_M)) | 311 FW_RI_TPTE_STAGSTATE_V(stag_state) | 312 FW_RI_TPTE_STAGTYPE_V(type) | FW_RI_TPTE_PDID_V(pdid)); 313 tpt.locread_to_qpid = cpu_to_be32(FW_RI_TPTE_PERM_V(perm) | 314 (bind_enabled ? FW_RI_TPTE_MWBINDEN_F : 0) | 315 FW_RI_TPTE_ADDRTYPE_V((zbva ? FW_RI_ZERO_BASED_TO : 316 FW_RI_VA_BASED_TO))| 317 FW_RI_TPTE_PS_V(page_size)); 318 tpt.nosnoop_pbladdr = !pbl_size ? 0 : cpu_to_be32( 319 FW_RI_TPTE_PBLADDR_V(PBL_OFF(rdev, pbl_addr)>>3)); 320 tpt.len_lo = cpu_to_be32((u32)(len & 0xffffffffUL)); 321 tpt.va_hi = cpu_to_be32((u32)(to >> 32)); 322 tpt.va_lo_fbo = cpu_to_be32((u32)(to & 0xffffffffUL)); 323 tpt.dca_mwbcnt_pstag = cpu_to_be32(0); 324 tpt.len_hi = cpu_to_be32((u32)(len >> 32)); 325 } 326 err = write_adapter_mem(rdev, stag_idx + 327 (rdev->lldi.vr->stag.start >> 5), 328 sizeof(tpt), &tpt, skb, wr_waitp); 329 330 if (reset_tpt_entry) { 331 c4iw_put_resource(&rdev->resource.tpt_table, stag_idx); 332 mutex_lock(&rdev->stats.lock); 333 rdev->stats.stag.cur -= 32; 334 mutex_unlock(&rdev->stats.lock); 335 } 336 return err; 337 } 338 339 static int write_pbl(struct c4iw_rdev *rdev, __be64 *pbl, 340 u32 pbl_addr, u32 pbl_size, struct c4iw_wr_wait *wr_waitp) 341 { 342 int err; 343 344 pr_debug("*pdb_addr 0x%x, pbl_base 0x%x, pbl_size %d\n", 345 pbl_addr, rdev->lldi.vr->pbl.start, 346 pbl_size); 347 348 err = write_adapter_mem(rdev, pbl_addr >> 5, pbl_size << 3, pbl, NULL, 349 wr_waitp); 350 return err; 351 } 352 353 static int dereg_mem(struct c4iw_rdev *rdev, u32 stag, u32 pbl_size, 354 u32 pbl_addr, struct sk_buff *skb, 355 struct c4iw_wr_wait *wr_waitp) 356 { 357 return write_tpt_entry(rdev, 1, &stag, 0, 0, 0, 0, 0, 0, 0UL, 0, 0, 358 pbl_size, pbl_addr, skb, wr_waitp); 359 } 360 361 static int allocate_window(struct c4iw_rdev *rdev, u32 *stag, u32 pdid, 362 struct c4iw_wr_wait *wr_waitp) 363 { 364 *stag = T4_STAG_UNSET; 365 return write_tpt_entry(rdev, 0, stag, 0, pdid, FW_RI_STAG_MW, 0, 0, 0, 366 0UL, 0, 0, 0, 0, NULL, wr_waitp); 367 } 368 369 static int deallocate_window(struct c4iw_rdev *rdev, u32 stag, 370 struct sk_buff *skb, 371 struct c4iw_wr_wait *wr_waitp) 372 { 373 return write_tpt_entry(rdev, 1, &stag, 0, 0, 0, 0, 0, 0, 0UL, 0, 0, 0, 374 0, skb, wr_waitp); 375 } 376 377 static int allocate_stag(struct c4iw_rdev *rdev, u32 *stag, u32 pdid, 378 u32 pbl_size, u32 pbl_addr, 379 struct c4iw_wr_wait *wr_waitp) 380 { 381 *stag = T4_STAG_UNSET; 382 return write_tpt_entry(rdev, 0, stag, 0, pdid, FW_RI_STAG_NSMR, 0, 0, 0, 383 0UL, 0, 0, pbl_size, pbl_addr, NULL, wr_waitp); 384 } 385 386 static int finish_mem_reg(struct c4iw_mr *mhp, u32 stag) 387 { 388 u32 mmid; 389 390 mhp->attr.state = 1; 391 mhp->attr.stag = stag; 392 mmid = stag >> 8; 393 mhp->ibmr.rkey = mhp->ibmr.lkey = stag; 394 pr_debug("mmid 0x%x mhp %p\n", mmid, mhp); 395 return insert_handle(mhp->rhp, &mhp->rhp->mmidr, mhp, mmid); 396 } 397 398 static int register_mem(struct c4iw_dev *rhp, struct c4iw_pd *php, 399 struct c4iw_mr *mhp, int shift) 400 { 401 u32 stag = T4_STAG_UNSET; 402 int ret; 403 404 ret = write_tpt_entry(&rhp->rdev, 0, &stag, 1, mhp->attr.pdid, 405 FW_RI_STAG_NSMR, mhp->attr.len ? 406 mhp->attr.perms : 0, 407 mhp->attr.mw_bind_enable, mhp->attr.zbva, 408 mhp->attr.va_fbo, mhp->attr.len ? 409 mhp->attr.len : -1, shift - 12, 410 mhp->attr.pbl_size, mhp->attr.pbl_addr, NULL, 411 mhp->wr_waitp); 412 if (ret) 413 return ret; 414 415 ret = finish_mem_reg(mhp, stag); 416 if (ret) { 417 dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size, 418 mhp->attr.pbl_addr, mhp->dereg_skb, mhp->wr_waitp); 419 mhp->dereg_skb = NULL; 420 } 421 return ret; 422 } 423 424 static int alloc_pbl(struct c4iw_mr *mhp, int npages) 425 { 426 mhp->attr.pbl_addr = c4iw_pblpool_alloc(&mhp->rhp->rdev, 427 npages << 3); 428 429 if (!mhp->attr.pbl_addr) 430 return -ENOMEM; 431 432 mhp->attr.pbl_size = npages; 433 434 return 0; 435 } 436 437 struct ib_mr *c4iw_get_dma_mr(struct ib_pd *pd, int acc) 438 { 439 struct c4iw_dev *rhp; 440 struct c4iw_pd *php; 441 struct c4iw_mr *mhp; 442 int ret; 443 u32 stag = T4_STAG_UNSET; 444 445 pr_debug("ib_pd %p\n", pd); 446 php = to_c4iw_pd(pd); 447 rhp = php->rhp; 448 449 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL); 450 if (!mhp) 451 return ERR_PTR(-ENOMEM); 452 mhp->wr_waitp = c4iw_alloc_wr_wait(GFP_KERNEL); 453 if (!mhp->wr_waitp) { 454 ret = -ENOMEM; 455 goto err_free_mhp; 456 } 457 c4iw_init_wr_wait(mhp->wr_waitp); 458 459 mhp->dereg_skb = alloc_skb(SGE_MAX_WR_LEN, GFP_KERNEL); 460 if (!mhp->dereg_skb) { 461 ret = -ENOMEM; 462 goto err_free_wr_wait; 463 } 464 465 mhp->rhp = rhp; 466 mhp->attr.pdid = php->pdid; 467 mhp->attr.perms = c4iw_ib_to_tpt_access(acc); 468 mhp->attr.mw_bind_enable = (acc&IB_ACCESS_MW_BIND) == IB_ACCESS_MW_BIND; 469 mhp->attr.zbva = 0; 470 mhp->attr.va_fbo = 0; 471 mhp->attr.page_size = 0; 472 mhp->attr.len = ~0ULL; 473 mhp->attr.pbl_size = 0; 474 475 ret = write_tpt_entry(&rhp->rdev, 0, &stag, 1, php->pdid, 476 FW_RI_STAG_NSMR, mhp->attr.perms, 477 mhp->attr.mw_bind_enable, 0, 0, ~0ULL, 0, 0, 0, 478 NULL, mhp->wr_waitp); 479 if (ret) 480 goto err_free_skb; 481 482 ret = finish_mem_reg(mhp, stag); 483 if (ret) 484 goto err_dereg_mem; 485 return &mhp->ibmr; 486 err_dereg_mem: 487 dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size, 488 mhp->attr.pbl_addr, mhp->dereg_skb, mhp->wr_waitp); 489 err_free_wr_wait: 490 c4iw_put_wr_wait(mhp->wr_waitp); 491 err_free_skb: 492 kfree_skb(mhp->dereg_skb); 493 err_free_mhp: 494 kfree(mhp); 495 return ERR_PTR(ret); 496 } 497 498 struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, 499 u64 virt, int acc, struct ib_udata *udata) 500 { 501 __be64 *pages; 502 int shift, n, len; 503 int i, k, entry; 504 int err = -ENOMEM; 505 struct scatterlist *sg; 506 struct c4iw_dev *rhp; 507 struct c4iw_pd *php; 508 struct c4iw_mr *mhp; 509 510 pr_debug("ib_pd %p\n", pd); 511 512 if (length == ~0ULL) 513 return ERR_PTR(-EINVAL); 514 515 if ((length + start) < start) 516 return ERR_PTR(-EINVAL); 517 518 php = to_c4iw_pd(pd); 519 rhp = php->rhp; 520 521 if (mr_exceeds_hw_limits(rhp, length)) 522 return ERR_PTR(-EINVAL); 523 524 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL); 525 if (!mhp) 526 return ERR_PTR(-ENOMEM); 527 mhp->wr_waitp = c4iw_alloc_wr_wait(GFP_KERNEL); 528 if (!mhp->wr_waitp) 529 goto err_free_mhp; 530 531 mhp->dereg_skb = alloc_skb(SGE_MAX_WR_LEN, GFP_KERNEL); 532 if (!mhp->dereg_skb) 533 goto err_free_wr_wait; 534 535 mhp->rhp = rhp; 536 537 mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0); 538 if (IS_ERR(mhp->umem)) 539 goto err_free_skb; 540 541 shift = mhp->umem->page_shift; 542 543 n = mhp->umem->nmap; 544 err = alloc_pbl(mhp, n); 545 if (err) 546 goto err_umem_release; 547 548 pages = (__be64 *) __get_free_page(GFP_KERNEL); 549 if (!pages) { 550 err = -ENOMEM; 551 goto err_pbl_free; 552 } 553 554 i = n = 0; 555 556 for_each_sg(mhp->umem->sg_head.sgl, sg, mhp->umem->nmap, entry) { 557 len = sg_dma_len(sg) >> shift; 558 for (k = 0; k < len; ++k) { 559 pages[i++] = cpu_to_be64(sg_dma_address(sg) + 560 (k << shift)); 561 if (i == PAGE_SIZE / sizeof *pages) { 562 err = write_pbl(&mhp->rhp->rdev, 563 pages, 564 mhp->attr.pbl_addr + (n << 3), i, 565 mhp->wr_waitp); 566 if (err) 567 goto pbl_done; 568 n += i; 569 i = 0; 570 } 571 } 572 } 573 574 if (i) 575 err = write_pbl(&mhp->rhp->rdev, pages, 576 mhp->attr.pbl_addr + (n << 3), i, 577 mhp->wr_waitp); 578 579 pbl_done: 580 free_page((unsigned long) pages); 581 if (err) 582 goto err_pbl_free; 583 584 mhp->attr.pdid = php->pdid; 585 mhp->attr.zbva = 0; 586 mhp->attr.perms = c4iw_ib_to_tpt_access(acc); 587 mhp->attr.va_fbo = virt; 588 mhp->attr.page_size = shift - 12; 589 mhp->attr.len = length; 590 591 err = register_mem(rhp, php, mhp, shift); 592 if (err) 593 goto err_pbl_free; 594 595 return &mhp->ibmr; 596 597 err_pbl_free: 598 c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr, 599 mhp->attr.pbl_size << 3); 600 err_umem_release: 601 ib_umem_release(mhp->umem); 602 err_free_skb: 603 kfree_skb(mhp->dereg_skb); 604 err_free_wr_wait: 605 c4iw_put_wr_wait(mhp->wr_waitp); 606 err_free_mhp: 607 kfree(mhp); 608 return ERR_PTR(err); 609 } 610 611 struct ib_mw *c4iw_alloc_mw(struct ib_pd *pd, enum ib_mw_type type, 612 struct ib_udata *udata) 613 { 614 struct c4iw_dev *rhp; 615 struct c4iw_pd *php; 616 struct c4iw_mw *mhp; 617 u32 mmid; 618 u32 stag = 0; 619 int ret; 620 621 if (type != IB_MW_TYPE_1) 622 return ERR_PTR(-EINVAL); 623 624 php = to_c4iw_pd(pd); 625 rhp = php->rhp; 626 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL); 627 if (!mhp) 628 return ERR_PTR(-ENOMEM); 629 630 mhp->wr_waitp = c4iw_alloc_wr_wait(GFP_KERNEL); 631 if (!mhp->wr_waitp) { 632 ret = -ENOMEM; 633 goto free_mhp; 634 } 635 636 mhp->dereg_skb = alloc_skb(SGE_MAX_WR_LEN, GFP_KERNEL); 637 if (!mhp->dereg_skb) { 638 ret = -ENOMEM; 639 goto free_wr_wait; 640 } 641 642 ret = allocate_window(&rhp->rdev, &stag, php->pdid, mhp->wr_waitp); 643 if (ret) 644 goto free_skb; 645 mhp->rhp = rhp; 646 mhp->attr.pdid = php->pdid; 647 mhp->attr.type = FW_RI_STAG_MW; 648 mhp->attr.stag = stag; 649 mmid = (stag) >> 8; 650 mhp->ibmw.rkey = stag; 651 if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) { 652 ret = -ENOMEM; 653 goto dealloc_win; 654 } 655 pr_debug("mmid 0x%x mhp %p stag 0x%x\n", mmid, mhp, stag); 656 return &(mhp->ibmw); 657 658 dealloc_win: 659 deallocate_window(&rhp->rdev, mhp->attr.stag, mhp->dereg_skb, 660 mhp->wr_waitp); 661 free_skb: 662 kfree_skb(mhp->dereg_skb); 663 free_wr_wait: 664 c4iw_put_wr_wait(mhp->wr_waitp); 665 free_mhp: 666 kfree(mhp); 667 return ERR_PTR(ret); 668 } 669 670 int c4iw_dealloc_mw(struct ib_mw *mw) 671 { 672 struct c4iw_dev *rhp; 673 struct c4iw_mw *mhp; 674 u32 mmid; 675 676 mhp = to_c4iw_mw(mw); 677 rhp = mhp->rhp; 678 mmid = (mw->rkey) >> 8; 679 remove_handle(rhp, &rhp->mmidr, mmid); 680 deallocate_window(&rhp->rdev, mhp->attr.stag, mhp->dereg_skb, 681 mhp->wr_waitp); 682 kfree_skb(mhp->dereg_skb); 683 c4iw_put_wr_wait(mhp->wr_waitp); 684 kfree(mhp); 685 pr_debug("ib_mw %p mmid 0x%x ptr %p\n", mw, mmid, mhp); 686 return 0; 687 } 688 689 struct ib_mr *c4iw_alloc_mr(struct ib_pd *pd, 690 enum ib_mr_type mr_type, 691 u32 max_num_sg) 692 { 693 struct c4iw_dev *rhp; 694 struct c4iw_pd *php; 695 struct c4iw_mr *mhp; 696 u32 mmid; 697 u32 stag = 0; 698 int ret = 0; 699 int length = roundup(max_num_sg * sizeof(u64), 32); 700 701 php = to_c4iw_pd(pd); 702 rhp = php->rhp; 703 704 if (mr_type != IB_MR_TYPE_MEM_REG || 705 max_num_sg > t4_max_fr_depth(rhp->rdev.lldi.ulptx_memwrite_dsgl && 706 use_dsgl)) 707 return ERR_PTR(-EINVAL); 708 709 mhp = kzalloc(sizeof(*mhp), GFP_KERNEL); 710 if (!mhp) { 711 ret = -ENOMEM; 712 goto err; 713 } 714 715 mhp->wr_waitp = c4iw_alloc_wr_wait(GFP_KERNEL); 716 if (!mhp->wr_waitp) { 717 ret = -ENOMEM; 718 goto err_free_mhp; 719 } 720 c4iw_init_wr_wait(mhp->wr_waitp); 721 722 mhp->mpl = dma_alloc_coherent(&rhp->rdev.lldi.pdev->dev, 723 length, &mhp->mpl_addr, GFP_KERNEL); 724 if (!mhp->mpl) { 725 ret = -ENOMEM; 726 goto err_free_wr_wait; 727 } 728 mhp->max_mpl_len = length; 729 730 mhp->rhp = rhp; 731 ret = alloc_pbl(mhp, max_num_sg); 732 if (ret) 733 goto err_free_dma; 734 mhp->attr.pbl_size = max_num_sg; 735 ret = allocate_stag(&rhp->rdev, &stag, php->pdid, 736 mhp->attr.pbl_size, mhp->attr.pbl_addr, 737 mhp->wr_waitp); 738 if (ret) 739 goto err_free_pbl; 740 mhp->attr.pdid = php->pdid; 741 mhp->attr.type = FW_RI_STAG_NSMR; 742 mhp->attr.stag = stag; 743 mhp->attr.state = 0; 744 mmid = (stag) >> 8; 745 mhp->ibmr.rkey = mhp->ibmr.lkey = stag; 746 if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) { 747 ret = -ENOMEM; 748 goto err_dereg; 749 } 750 751 pr_debug("mmid 0x%x mhp %p stag 0x%x\n", mmid, mhp, stag); 752 return &(mhp->ibmr); 753 err_dereg: 754 dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size, 755 mhp->attr.pbl_addr, mhp->dereg_skb, mhp->wr_waitp); 756 err_free_pbl: 757 c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr, 758 mhp->attr.pbl_size << 3); 759 err_free_dma: 760 dma_free_coherent(&mhp->rhp->rdev.lldi.pdev->dev, 761 mhp->max_mpl_len, mhp->mpl, mhp->mpl_addr); 762 err_free_wr_wait: 763 c4iw_put_wr_wait(mhp->wr_waitp); 764 err_free_mhp: 765 kfree(mhp); 766 err: 767 return ERR_PTR(ret); 768 } 769 770 static int c4iw_set_page(struct ib_mr *ibmr, u64 addr) 771 { 772 struct c4iw_mr *mhp = to_c4iw_mr(ibmr); 773 774 if (unlikely(mhp->mpl_len == mhp->max_mpl_len)) 775 return -ENOMEM; 776 777 mhp->mpl[mhp->mpl_len++] = addr; 778 779 return 0; 780 } 781 782 int c4iw_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg, int sg_nents, 783 unsigned int *sg_offset) 784 { 785 struct c4iw_mr *mhp = to_c4iw_mr(ibmr); 786 787 mhp->mpl_len = 0; 788 789 return ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, c4iw_set_page); 790 } 791 792 int c4iw_dereg_mr(struct ib_mr *ib_mr) 793 { 794 struct c4iw_dev *rhp; 795 struct c4iw_mr *mhp; 796 u32 mmid; 797 798 pr_debug("ib_mr %p\n", ib_mr); 799 800 mhp = to_c4iw_mr(ib_mr); 801 rhp = mhp->rhp; 802 mmid = mhp->attr.stag >> 8; 803 remove_handle(rhp, &rhp->mmidr, mmid); 804 if (mhp->mpl) 805 dma_free_coherent(&mhp->rhp->rdev.lldi.pdev->dev, 806 mhp->max_mpl_len, mhp->mpl, mhp->mpl_addr); 807 dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size, 808 mhp->attr.pbl_addr, mhp->dereg_skb, mhp->wr_waitp); 809 if (mhp->attr.pbl_size) 810 c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr, 811 mhp->attr.pbl_size << 3); 812 if (mhp->kva) 813 kfree((void *) (unsigned long) mhp->kva); 814 if (mhp->umem) 815 ib_umem_release(mhp->umem); 816 pr_debug("mmid 0x%x ptr %p\n", mmid, mhp); 817 c4iw_put_wr_wait(mhp->wr_waitp); 818 kfree(mhp); 819 return 0; 820 } 821 822 void c4iw_invalidate_mr(struct c4iw_dev *rhp, u32 rkey) 823 { 824 struct c4iw_mr *mhp; 825 unsigned long flags; 826 827 spin_lock_irqsave(&rhp->lock, flags); 828 mhp = get_mhp(rhp, rkey >> 8); 829 if (mhp) 830 mhp->attr.state = 0; 831 spin_unlock_irqrestore(&rhp->lock, flags); 832 } 833