1/*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2012 Chelsio Communications, Inc. 5 * All rights reserved. 6 * Written by: Navdeep Parhar <np@FreeBSD.org> 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30#include <sys/cdefs.h> 31__FBSDID("$FreeBSD$"); 32 33#include "opt_inet.h" 34 35#include <sys/param.h> 36#include <sys/aio.h> 37#include <sys/file.h> 38#include <sys/systm.h> 39#include <sys/kernel.h> 40#include <sys/ktr.h> 41#include <sys/module.h> 42#include <sys/protosw.h> 43#include <sys/proc.h> 44#include <sys/domain.h> 45#include <sys/socket.h> 46#include <sys/socketvar.h> 47#include <sys/taskqueue.h> 48#include <sys/uio.h> 49#include <netinet/in.h> 50#include <netinet/in_pcb.h> 51#include <netinet/ip.h> 52#include <netinet/tcp_var.h> 53#define TCPSTATES 54#include <netinet/tcp_fsm.h> 55#include <netinet/toecore.h> 56 57#include <vm/vm.h> 58#include <vm/vm_extern.h> 59#include <vm/vm_param.h> 60#include <vm/pmap.h> 61#include <vm/vm_map.h> 62#include <vm/vm_page.h> 63#include <vm/vm_object.h> 64 65#ifdef TCP_OFFLOAD 66#include "common/common.h" 67#include "common/t4_msg.h" 68#include "common/t4_regs.h" 69#include "common/t4_tcb.h" 70#include "tom/t4_tom.h" 71 72/* 73 * Use the 'backend3' field in AIO jobs to store the amount of data 74 * received by the AIO job so far. 75 */ 76#define aio_received backend3 77 78static void aio_ddp_requeue_task(void *context, int pending); 79static void ddp_complete_all(struct toepcb *toep, int error); 80static void t4_aio_cancel_active(struct kaiocb *job); 81static void t4_aio_cancel_queued(struct kaiocb *job); 82 83static TAILQ_HEAD(, pageset) ddp_orphan_pagesets; 84static struct mtx ddp_orphan_pagesets_lock; 85static struct task ddp_orphan_task; 86 87#define MAX_DDP_BUFFER_SIZE (M_TCB_RX_DDP_BUF0_LEN) 88 89/* 90 * A page set holds information about a buffer used for DDP. The page 91 * set holds resources such as the VM pages backing the buffer (either 92 * held or wired) and the page pods associated with the buffer. 93 * Recently used page sets are cached to allow for efficient reuse of 94 * buffers (avoiding the need to re-fault in pages, hold them, etc.). 95 * Note that cached page sets keep the backing pages wired. The 96 * number of wired pages is capped by only allowing for two wired 97 * pagesets per connection. This is not a perfect cap, but is a 98 * trade-off for performance. 99 * 100 * If an application ping-pongs two buffers for a connection via 101 * aio_read(2) then those buffers should remain wired and expensive VM 102 * fault lookups should be avoided after each buffer has been used 103 * once. If an application uses more than two buffers then this will 104 * fall back to doing expensive VM fault lookups for each operation. 105 */ 106static void 107free_pageset(struct tom_data *td, struct pageset *ps) 108{ 109 vm_page_t p; 110 int i; 111 112 if (ps->prsv.prsv_nppods > 0) 113 t4_free_page_pods(&ps->prsv); 114 115 for (i = 0; i < ps->npages; i++) { 116 p = ps->pages[i]; 117 vm_page_unwire(p, PQ_INACTIVE); 118 } 119 mtx_lock(&ddp_orphan_pagesets_lock); 120 TAILQ_INSERT_TAIL(&ddp_orphan_pagesets, ps, link); 121 taskqueue_enqueue(taskqueue_thread, &ddp_orphan_task); 122 mtx_unlock(&ddp_orphan_pagesets_lock); 123} 124 125static void 126ddp_free_orphan_pagesets(void *context, int pending) 127{ 128 struct pageset *ps; 129 130 mtx_lock(&ddp_orphan_pagesets_lock); 131 while (!TAILQ_EMPTY(&ddp_orphan_pagesets)) { 132 ps = TAILQ_FIRST(&ddp_orphan_pagesets); 133 TAILQ_REMOVE(&ddp_orphan_pagesets, ps, link); 134 mtx_unlock(&ddp_orphan_pagesets_lock); 135 if (ps->vm) 136 vmspace_free(ps->vm); 137 free(ps, M_CXGBE); 138 mtx_lock(&ddp_orphan_pagesets_lock); 139 } 140 mtx_unlock(&ddp_orphan_pagesets_lock); 141} 142 143static void 144recycle_pageset(struct toepcb *toep, struct pageset *ps) 145{ 146 147 DDP_ASSERT_LOCKED(toep); 148 if (!(toep->ddp.flags & DDP_DEAD)) { 149 KASSERT(toep->ddp.cached_count + toep->ddp.active_count < 150 nitems(toep->ddp.db), ("too many wired pagesets")); 151 TAILQ_INSERT_HEAD(&toep->ddp.cached_pagesets, ps, link); 152 toep->ddp.cached_count++; 153 } else 154 free_pageset(toep->td, ps); 155} 156 157static void 158ddp_complete_one(struct kaiocb *job, int error) 159{ 160 long copied; 161 162 /* 163 * If this job had copied data out of the socket buffer before 164 * it was cancelled, report it as a short read rather than an 165 * error. 166 */ 167 copied = job->aio_received; 168 if (copied != 0 || error == 0) 169 aio_complete(job, copied, 0); 170 else 171 aio_complete(job, -1, error); 172} 173 174static void 175free_ddp_buffer(struct tom_data *td, struct ddp_buffer *db) 176{ 177 178 if (db->job) { 179 /* 180 * XXX: If we are un-offloading the socket then we 181 * should requeue these on the socket somehow. If we 182 * got a FIN from the remote end, then this completes 183 * any remaining requests with an EOF read. 184 */ 185 if (!aio_clear_cancel_function(db->job)) 186 ddp_complete_one(db->job, 0); 187 } 188 189 if (db->ps) 190 free_pageset(td, db->ps); 191} 192 193void 194ddp_init_toep(struct toepcb *toep) 195{ 196 197 TAILQ_INIT(&toep->ddp.aiojobq); 198 TASK_INIT(&toep->ddp.requeue_task, 0, aio_ddp_requeue_task, toep); 199 toep->ddp.flags = DDP_OK; 200 toep->ddp.active_id = -1; 201 mtx_init(&toep->ddp.lock, "t4 ddp", NULL, MTX_DEF); 202} 203 204void 205ddp_uninit_toep(struct toepcb *toep) 206{ 207 208 mtx_destroy(&toep->ddp.lock); 209} 210 211void 212release_ddp_resources(struct toepcb *toep) 213{ 214 struct pageset *ps; 215 int i; 216 217 DDP_LOCK(toep); 218 toep->ddp.flags |= DDP_DEAD; 219 for (i = 0; i < nitems(toep->ddp.db); i++) { 220 free_ddp_buffer(toep->td, &toep->ddp.db[i]); 221 } 222 while ((ps = TAILQ_FIRST(&toep->ddp.cached_pagesets)) != NULL) { 223 TAILQ_REMOVE(&toep->ddp.cached_pagesets, ps, link); 224 free_pageset(toep->td, ps); 225 } 226 ddp_complete_all(toep, 0); 227 DDP_UNLOCK(toep); 228} 229 230#ifdef INVARIANTS 231void 232ddp_assert_empty(struct toepcb *toep) 233{ 234 int i; 235 236 MPASS(!(toep->ddp.flags & DDP_TASK_ACTIVE)); 237 for (i = 0; i < nitems(toep->ddp.db); i++) { 238 MPASS(toep->ddp.db[i].job == NULL); 239 MPASS(toep->ddp.db[i].ps == NULL); 240 } 241 MPASS(TAILQ_EMPTY(&toep->ddp.cached_pagesets)); 242 MPASS(TAILQ_EMPTY(&toep->ddp.aiojobq)); 243} 244#endif 245 246static void 247complete_ddp_buffer(struct toepcb *toep, struct ddp_buffer *db, 248 unsigned int db_idx) 249{ 250 unsigned int db_flag; 251 252 toep->ddp.active_count--; 253 if (toep->ddp.active_id == db_idx) { 254 if (toep->ddp.active_count == 0) { 255 KASSERT(toep->ddp.db[db_idx ^ 1].job == NULL, 256 ("%s: active_count mismatch", __func__)); 257 toep->ddp.active_id = -1; 258 } else 259 toep->ddp.active_id ^= 1; 260#ifdef VERBOSE_TRACES 261 CTR3(KTR_CXGBE, "%s: tid %u, ddp_active_id = %d", __func__, 262 toep->tid, toep->ddp.active_id); 263#endif 264 } else { 265 KASSERT(toep->ddp.active_count != 0 && 266 toep->ddp.active_id != -1, 267 ("%s: active count mismatch", __func__)); 268 } 269 270 db->cancel_pending = 0; 271 db->job = NULL; 272 recycle_pageset(toep, db->ps); 273 db->ps = NULL; 274 275 db_flag = db_idx == 1 ? DDP_BUF1_ACTIVE : DDP_BUF0_ACTIVE; 276 KASSERT(toep->ddp.flags & db_flag, 277 ("%s: DDP buffer not active. toep %p, ddp_flags 0x%x", 278 __func__, toep, toep->ddp.flags)); 279 toep->ddp.flags &= ~db_flag; 280} 281 282/* XXX: handle_ddp_data code duplication */ 283void 284insert_ddp_data(struct toepcb *toep, uint32_t n) 285{ 286 struct inpcb *inp = toep->inp; 287 struct tcpcb *tp = intotcpcb(inp); 288 struct ddp_buffer *db; 289 struct kaiocb *job; 290 size_t placed; 291 long copied; 292 unsigned int db_flag, db_idx; 293 294 INP_WLOCK_ASSERT(inp); 295 DDP_ASSERT_LOCKED(toep); 296 297 tp->rcv_nxt += n; 298#ifndef USE_DDP_RX_FLOW_CONTROL 299 KASSERT(tp->rcv_wnd >= n, ("%s: negative window size", __func__)); 300 tp->rcv_wnd -= n; 301#endif 302 CTR2(KTR_CXGBE, "%s: placed %u bytes before falling out of DDP", 303 __func__, n); 304 while (toep->ddp.active_count > 0) { 305 MPASS(toep->ddp.active_id != -1); 306 db_idx = toep->ddp.active_id; 307 db_flag = db_idx == 1 ? DDP_BUF1_ACTIVE : DDP_BUF0_ACTIVE; 308 MPASS((toep->ddp.flags & db_flag) != 0); 309 db = &toep->ddp.db[db_idx]; 310 job = db->job; 311 copied = job->aio_received; 312 placed = n; 313 if (placed > job->uaiocb.aio_nbytes - copied) 314 placed = job->uaiocb.aio_nbytes - copied; 315 if (placed > 0) 316 job->msgrcv = 1; 317 if (!aio_clear_cancel_function(job)) { 318 /* 319 * Update the copied length for when 320 * t4_aio_cancel_active() completes this 321 * request. 322 */ 323 job->aio_received += placed; 324 } else if (copied + placed != 0) { 325 CTR4(KTR_CXGBE, 326 "%s: completing %p (copied %ld, placed %lu)", 327 __func__, job, copied, placed); 328 /* XXX: This always completes if there is some data. */ 329 aio_complete(job, copied + placed, 0); 330 } else if (aio_set_cancel_function(job, t4_aio_cancel_queued)) { 331 TAILQ_INSERT_HEAD(&toep->ddp.aiojobq, job, list); 332 toep->ddp.waiting_count++; 333 } else 334 aio_cancel(job); 335 n -= placed; 336 complete_ddp_buffer(toep, db, db_idx); 337 } 338 339 MPASS(n == 0); 340} 341 342/* SET_TCB_FIELD sent as a ULP command looks like this */ 343#define LEN__SET_TCB_FIELD_ULP (sizeof(struct ulp_txpkt) + \ 344 sizeof(struct ulptx_idata) + sizeof(struct cpl_set_tcb_field_core)) 345 346/* RX_DATA_ACK sent as a ULP command looks like this */ 347#define LEN__RX_DATA_ACK_ULP (sizeof(struct ulp_txpkt) + \ 348 sizeof(struct ulptx_idata) + sizeof(struct cpl_rx_data_ack_core)) 349 350static inline void * 351mk_set_tcb_field_ulp(struct ulp_txpkt *ulpmc, struct toepcb *toep, 352 uint64_t word, uint64_t mask, uint64_t val) 353{ 354 struct ulptx_idata *ulpsc; 355 struct cpl_set_tcb_field_core *req; 356 357 ulpmc->cmd_dest = htonl(V_ULPTX_CMD(ULP_TX_PKT) | V_ULP_TXPKT_DEST(0)); 358 ulpmc->len = htobe32(howmany(LEN__SET_TCB_FIELD_ULP, 16)); 359 360 ulpsc = (struct ulptx_idata *)(ulpmc + 1); 361 ulpsc->cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_IMM)); 362 ulpsc->len = htobe32(sizeof(*req)); 363 364 req = (struct cpl_set_tcb_field_core *)(ulpsc + 1); 365 OPCODE_TID(req) = htobe32(MK_OPCODE_TID(CPL_SET_TCB_FIELD, toep->tid)); 366 req->reply_ctrl = htobe16(V_NO_REPLY(1) | 367 V_QUEUENO(toep->ofld_rxq->iq.abs_id)); 368 req->word_cookie = htobe16(V_WORD(word) | V_COOKIE(0)); 369 req->mask = htobe64(mask); 370 req->val = htobe64(val); 371 372 ulpsc = (struct ulptx_idata *)(req + 1); 373 if (LEN__SET_TCB_FIELD_ULP % 16) { 374 ulpsc->cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_NOOP)); 375 ulpsc->len = htobe32(0); 376 return (ulpsc + 1); 377 } 378 return (ulpsc); 379} 380 381static inline void * 382mk_rx_data_ack_ulp(struct ulp_txpkt *ulpmc, struct toepcb *toep) 383{ 384 struct ulptx_idata *ulpsc; 385 struct cpl_rx_data_ack_core *req; 386 387 ulpmc->cmd_dest = htonl(V_ULPTX_CMD(ULP_TX_PKT) | V_ULP_TXPKT_DEST(0)); 388 ulpmc->len = htobe32(howmany(LEN__RX_DATA_ACK_ULP, 16)); 389 390 ulpsc = (struct ulptx_idata *)(ulpmc + 1); 391 ulpsc->cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_IMM)); 392 ulpsc->len = htobe32(sizeof(*req)); 393 394 req = (struct cpl_rx_data_ack_core *)(ulpsc + 1); 395 OPCODE_TID(req) = htobe32(MK_OPCODE_TID(CPL_RX_DATA_ACK, toep->tid)); 396 req->credit_dack = htobe32(F_RX_MODULATE_RX); 397 398 ulpsc = (struct ulptx_idata *)(req + 1); 399 if (LEN__RX_DATA_ACK_ULP % 16) { 400 ulpsc->cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_NOOP)); 401 ulpsc->len = htobe32(0); 402 return (ulpsc + 1); 403 } 404 return (ulpsc); 405} 406 407static struct wrqe * 408mk_update_tcb_for_ddp(struct adapter *sc, struct toepcb *toep, int db_idx, 409 struct pageset *ps, int offset, uint64_t ddp_flags, uint64_t ddp_flags_mask) 410{ 411 struct wrqe *wr; 412 struct work_request_hdr *wrh; 413 struct ulp_txpkt *ulpmc; 414 int len; 415 416 KASSERT(db_idx == 0 || db_idx == 1, 417 ("%s: bad DDP buffer index %d", __func__, db_idx)); 418 419 /* 420 * We'll send a compound work request that has 3 SET_TCB_FIELDs and an 421 * RX_DATA_ACK (with RX_MODULATE to speed up delivery). 422 * 423 * The work request header is 16B and always ends at a 16B boundary. 424 * The ULPTX master commands that follow must all end at 16B boundaries 425 * too so we round up the size to 16. 426 */ 427 len = sizeof(*wrh) + 3 * roundup2(LEN__SET_TCB_FIELD_ULP, 16) + 428 roundup2(LEN__RX_DATA_ACK_ULP, 16); 429 430 wr = alloc_wrqe(len, toep->ctrlq); 431 if (wr == NULL) 432 return (NULL); 433 wrh = wrtod(wr); 434 INIT_ULPTX_WRH(wrh, len, 1, 0); /* atomic */ 435 ulpmc = (struct ulp_txpkt *)(wrh + 1); 436 437 /* Write the buffer's tag */ 438 ulpmc = mk_set_tcb_field_ulp(ulpmc, toep, 439 W_TCB_RX_DDP_BUF0_TAG + db_idx, 440 V_TCB_RX_DDP_BUF0_TAG(M_TCB_RX_DDP_BUF0_TAG), 441 V_TCB_RX_DDP_BUF0_TAG(ps->prsv.prsv_tag)); 442 443 /* Update the current offset in the DDP buffer and its total length */ 444 if (db_idx == 0) 445 ulpmc = mk_set_tcb_field_ulp(ulpmc, toep, 446 W_TCB_RX_DDP_BUF0_OFFSET, 447 V_TCB_RX_DDP_BUF0_OFFSET(M_TCB_RX_DDP_BUF0_OFFSET) | 448 V_TCB_RX_DDP_BUF0_LEN(M_TCB_RX_DDP_BUF0_LEN), 449 V_TCB_RX_DDP_BUF0_OFFSET(offset) | 450 V_TCB_RX_DDP_BUF0_LEN(ps->len)); 451 else 452 ulpmc = mk_set_tcb_field_ulp(ulpmc, toep, 453 W_TCB_RX_DDP_BUF1_OFFSET, 454 V_TCB_RX_DDP_BUF1_OFFSET(M_TCB_RX_DDP_BUF1_OFFSET) | 455 V_TCB_RX_DDP_BUF1_LEN((u64)M_TCB_RX_DDP_BUF1_LEN << 32), 456 V_TCB_RX_DDP_BUF1_OFFSET(offset) | 457 V_TCB_RX_DDP_BUF1_LEN((u64)ps->len << 32)); 458 459 /* Update DDP flags */ 460 ulpmc = mk_set_tcb_field_ulp(ulpmc, toep, W_TCB_RX_DDP_FLAGS, 461 ddp_flags_mask, ddp_flags); 462 463 /* Gratuitous RX_DATA_ACK with RX_MODULATE set to speed up delivery. */ 464 ulpmc = mk_rx_data_ack_ulp(ulpmc, toep); 465 466 return (wr); 467} 468 469static int 470handle_ddp_data(struct toepcb *toep, __be32 ddp_report, __be32 rcv_nxt, int len) 471{ 472 uint32_t report = be32toh(ddp_report); 473 unsigned int db_idx; 474 struct inpcb *inp = toep->inp; 475 struct ddp_buffer *db; 476 struct tcpcb *tp; 477 struct socket *so; 478 struct sockbuf *sb; 479 struct kaiocb *job; 480 long copied; 481 482 db_idx = report & F_DDP_BUF_IDX ? 1 : 0; 483 484 if (__predict_false(!(report & F_DDP_INV))) 485 CXGBE_UNIMPLEMENTED("DDP buffer still valid"); 486 487 INP_WLOCK(inp); 488 so = inp_inpcbtosocket(inp); 489 sb = &so->so_rcv; 490 DDP_LOCK(toep); 491 492 KASSERT(toep->ddp.active_id == db_idx, 493 ("completed DDP buffer (%d) != active_id (%d) for tid %d", db_idx, 494 toep->ddp.active_id, toep->tid)); 495 db = &toep->ddp.db[db_idx]; 496 job = db->job; 497 498 if (__predict_false(inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT))) { 499 /* 500 * This can happen due to an administrative tcpdrop(8). 501 * Just fail the request with ECONNRESET. 502 */ 503 CTR5(KTR_CXGBE, "%s: tid %u, seq 0x%x, len %d, inp_flags 0x%x", 504 __func__, toep->tid, be32toh(rcv_nxt), len, inp->inp_flags); 505 if (aio_clear_cancel_function(job)) 506 ddp_complete_one(job, ECONNRESET); 507 goto completed; 508 } 509 510 tp = intotcpcb(inp); 511 512 /* 513 * For RX_DDP_COMPLETE, len will be zero and rcv_nxt is the 514 * sequence number of the next byte to receive. The length of 515 * the data received for this message must be computed by 516 * comparing the new and old values of rcv_nxt. 517 * 518 * For RX_DATA_DDP, len might be non-zero, but it is only the 519 * length of the most recent DMA. It does not include the 520 * total length of the data received since the previous update 521 * for this DDP buffer. rcv_nxt is the sequence number of the 522 * first received byte from the most recent DMA. 523 */ 524 len += be32toh(rcv_nxt) - tp->rcv_nxt; 525 tp->rcv_nxt += len; 526 tp->t_rcvtime = ticks; 527#ifndef USE_DDP_RX_FLOW_CONTROL 528 KASSERT(tp->rcv_wnd >= len, ("%s: negative window size", __func__)); 529 tp->rcv_wnd -= len; 530#endif 531#ifdef VERBOSE_TRACES 532 CTR5(KTR_CXGBE, "%s: tid %u, DDP[%d] placed %d bytes (%#x)", __func__, 533 toep->tid, db_idx, len, report); 534#endif 535 536 /* receive buffer autosize */ 537 MPASS(toep->vnet == so->so_vnet); 538 CURVNET_SET(toep->vnet); 539 SOCKBUF_LOCK(sb); 540 if (sb->sb_flags & SB_AUTOSIZE && 541 V_tcp_do_autorcvbuf && 542 sb->sb_hiwat < V_tcp_autorcvbuf_max && 543 len > (sbspace(sb) / 8 * 7)) { 544 struct adapter *sc = td_adapter(toep->td); 545 unsigned int hiwat = sb->sb_hiwat; 546 unsigned int newsize = min(hiwat + sc->tt.autorcvbuf_inc, 547 V_tcp_autorcvbuf_max); 548 549 if (!sbreserve_locked(sb, newsize, so, NULL)) 550 sb->sb_flags &= ~SB_AUTOSIZE; 551 } 552 SOCKBUF_UNLOCK(sb); 553 CURVNET_RESTORE(); 554 555 job->msgrcv = 1; 556 if (db->cancel_pending) { 557 /* 558 * Update the job's length but defer completion to the 559 * TCB_RPL callback. 560 */ 561 job->aio_received += len; 562 goto out; 563 } else if (!aio_clear_cancel_function(job)) { 564 /* 565 * Update the copied length for when 566 * t4_aio_cancel_active() completes this request. 567 */ 568 job->aio_received += len; 569 } else { 570 copied = job->aio_received; 571#ifdef VERBOSE_TRACES 572 CTR5(KTR_CXGBE, 573 "%s: tid %u, completing %p (copied %ld, placed %d)", 574 __func__, toep->tid, job, copied, len); 575#endif 576 aio_complete(job, copied + len, 0); 577 t4_rcvd(&toep->td->tod, tp); 578 } 579 580completed: 581 complete_ddp_buffer(toep, db, db_idx); 582 if (toep->ddp.waiting_count > 0) 583 ddp_queue_toep(toep); 584out: 585 DDP_UNLOCK(toep); 586 INP_WUNLOCK(inp); 587 588 return (0); 589} 590 591void 592handle_ddp_indicate(struct toepcb *toep) 593{ 594 595 DDP_ASSERT_LOCKED(toep); 596 MPASS(toep->ddp.active_count == 0); 597 MPASS((toep->ddp.flags & (DDP_BUF0_ACTIVE | DDP_BUF1_ACTIVE)) == 0); 598 if (toep->ddp.waiting_count == 0) { 599 /* 600 * The pending requests that triggered the request for an 601 * an indicate were cancelled. Those cancels should have 602 * already disabled DDP. Just ignore this as the data is 603 * going into the socket buffer anyway. 604 */ 605 return; 606 } 607 CTR3(KTR_CXGBE, "%s: tid %d indicated (%d waiting)", __func__, 608 toep->tid, toep->ddp.waiting_count); 609 ddp_queue_toep(toep); 610} 611 612CTASSERT(CPL_COOKIE_DDP0 + 1 == CPL_COOKIE_DDP1); 613 614static int 615do_ddp_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 616{ 617 struct adapter *sc = iq->adapter; 618 const struct cpl_set_tcb_rpl *cpl = (const void *)(rss + 1); 619 unsigned int tid = GET_TID(cpl); 620 unsigned int db_idx; 621 struct toepcb *toep; 622 struct inpcb *inp; 623 struct ddp_buffer *db; 624 struct kaiocb *job; 625 long copied; 626 627 if (cpl->status != CPL_ERR_NONE) 628 panic("XXX: tcp_rpl failed: %d", cpl->status); 629 630 toep = lookup_tid(sc, tid); 631 inp = toep->inp; 632 switch (cpl->cookie) { 633 case V_WORD(W_TCB_RX_DDP_FLAGS) | V_COOKIE(CPL_COOKIE_DDP0): 634 case V_WORD(W_TCB_RX_DDP_FLAGS) | V_COOKIE(CPL_COOKIE_DDP1): 635 /* 636 * XXX: This duplicates a lot of code with handle_ddp_data(). 637 */ 638 db_idx = G_COOKIE(cpl->cookie) - CPL_COOKIE_DDP0; 639 MPASS(db_idx < nitems(toep->ddp.db)); 640 INP_WLOCK(inp); 641 DDP_LOCK(toep); 642 db = &toep->ddp.db[db_idx]; 643 644 /* 645 * handle_ddp_data() should leave the job around until 646 * this callback runs once a cancel is pending. 647 */ 648 MPASS(db != NULL); 649 MPASS(db->job != NULL); 650 MPASS(db->cancel_pending); 651 652 /* 653 * XXX: It's not clear what happens if there is data 654 * placed when the buffer is invalidated. I suspect we 655 * need to read the TCB to see how much data was placed. 656 * 657 * For now this just pretends like nothing was placed. 658 * 659 * XXX: Note that if we did check the PCB we would need to 660 * also take care of updating the tp, etc. 661 */ 662 job = db->job; 663 copied = job->aio_received; 664 if (copied == 0) { 665 CTR2(KTR_CXGBE, "%s: cancelling %p", __func__, job); 666 aio_cancel(job); 667 } else { 668 CTR3(KTR_CXGBE, "%s: completing %p (copied %ld)", 669 __func__, job, copied); 670 aio_complete(job, copied, 0); 671 t4_rcvd(&toep->td->tod, intotcpcb(inp)); 672 } 673 674 complete_ddp_buffer(toep, db, db_idx); 675 if (toep->ddp.waiting_count > 0) 676 ddp_queue_toep(toep); 677 DDP_UNLOCK(toep); 678 INP_WUNLOCK(inp); 679 break; 680 default: 681 panic("XXX: unknown tcb_rpl offset %#x, cookie %#x", 682 G_WORD(cpl->cookie), G_COOKIE(cpl->cookie)); 683 } 684 685 return (0); 686} 687 688void 689handle_ddp_close(struct toepcb *toep, struct tcpcb *tp, __be32 rcv_nxt) 690{ 691 struct ddp_buffer *db; 692 struct kaiocb *job; 693 long copied; 694 unsigned int db_flag, db_idx; 695 int len, placed; 696 697 INP_WLOCK_ASSERT(toep->inp); 698 DDP_ASSERT_LOCKED(toep); 699 700 len = be32toh(rcv_nxt) - tp->rcv_nxt; 701 tp->rcv_nxt += len; 702 703 while (toep->ddp.active_count > 0) { 704 MPASS(toep->ddp.active_id != -1); 705 db_idx = toep->ddp.active_id; 706 db_flag = db_idx == 1 ? DDP_BUF1_ACTIVE : DDP_BUF0_ACTIVE; 707 MPASS((toep->ddp.flags & db_flag) != 0); 708 db = &toep->ddp.db[db_idx]; 709 job = db->job; 710 copied = job->aio_received; 711 placed = len; 712 if (placed > job->uaiocb.aio_nbytes - copied) 713 placed = job->uaiocb.aio_nbytes - copied; 714 if (placed > 0) 715 job->msgrcv = 1; 716 if (!aio_clear_cancel_function(job)) { 717 /* 718 * Update the copied length for when 719 * t4_aio_cancel_active() completes this 720 * request. 721 */ 722 job->aio_received += placed; 723 } else { 724 CTR4(KTR_CXGBE, "%s: tid %d completed buf %d len %d", 725 __func__, toep->tid, db_idx, placed); 726 aio_complete(job, copied + placed, 0); 727 } 728 len -= placed; 729 complete_ddp_buffer(toep, db, db_idx); 730 } 731 732 MPASS(len == 0); 733 ddp_complete_all(toep, 0); 734} 735 736#define DDP_ERR (F_DDP_PPOD_MISMATCH | F_DDP_LLIMIT_ERR | F_DDP_ULIMIT_ERR |\ 737 F_DDP_PPOD_PARITY_ERR | F_DDP_PADDING_ERR | F_DDP_OFFSET_ERR |\ 738 F_DDP_INVALID_TAG | F_DDP_COLOR_ERR | F_DDP_TID_MISMATCH |\ 739 F_DDP_INVALID_PPOD | F_DDP_HDRCRC_ERR | F_DDP_DATACRC_ERR) 740 741extern cpl_handler_t t4_cpl_handler[]; 742 743static int 744do_rx_data_ddp(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 745{ 746 struct adapter *sc = iq->adapter; 747 const struct cpl_rx_data_ddp *cpl = (const void *)(rss + 1); 748 unsigned int tid = GET_TID(cpl); 749 uint32_t vld; 750 struct toepcb *toep = lookup_tid(sc, tid); 751 752 KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 753 KASSERT(toep->tid == tid, ("%s: toep tid/atid mismatch", __func__)); 754 KASSERT(!(toep->flags & TPF_SYNQE), 755 ("%s: toep %p claims to be a synq entry", __func__, toep)); 756 757 vld = be32toh(cpl->ddpvld); 758 if (__predict_false(vld & DDP_ERR)) { 759 panic("%s: DDP error 0x%x (tid %d, toep %p)", 760 __func__, vld, tid, toep); 761 } 762 763 if (ulp_mode(toep) == ULP_MODE_ISCSI) { 764 t4_cpl_handler[CPL_RX_ISCSI_DDP](iq, rss, m); 765 return (0); 766 } 767 768 handle_ddp_data(toep, cpl->u.ddp_report, cpl->seq, be16toh(cpl->len)); 769 770 return (0); 771} 772 773static int 774do_rx_ddp_complete(struct sge_iq *iq, const struct rss_header *rss, 775 struct mbuf *m) 776{ 777 struct adapter *sc = iq->adapter; 778 const struct cpl_rx_ddp_complete *cpl = (const void *)(rss + 1); 779 unsigned int tid = GET_TID(cpl); 780 struct toepcb *toep = lookup_tid(sc, tid); 781 782 KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 783 KASSERT(toep->tid == tid, ("%s: toep tid/atid mismatch", __func__)); 784 KASSERT(!(toep->flags & TPF_SYNQE), 785 ("%s: toep %p claims to be a synq entry", __func__, toep)); 786 787 handle_ddp_data(toep, cpl->ddp_report, cpl->rcv_nxt, 0); 788 789 return (0); 790} 791 792static void 793enable_ddp(struct adapter *sc, struct toepcb *toep) 794{ 795 796 KASSERT((toep->ddp.flags & (DDP_ON | DDP_OK | DDP_SC_REQ)) == DDP_OK, 797 ("%s: toep %p has bad ddp_flags 0x%x", 798 __func__, toep, toep->ddp.flags)); 799 800 CTR3(KTR_CXGBE, "%s: tid %u (time %u)", 801 __func__, toep->tid, time_uptime); 802 803 DDP_ASSERT_LOCKED(toep); 804 toep->ddp.flags |= DDP_SC_REQ; 805 t4_set_tcb_field(sc, toep->ctrlq, toep, W_TCB_RX_DDP_FLAGS, 806 V_TF_DDP_OFF(1) | V_TF_DDP_INDICATE_OUT(1) | 807 V_TF_DDP_BUF0_INDICATE(1) | V_TF_DDP_BUF1_INDICATE(1) | 808 V_TF_DDP_BUF0_VALID(1) | V_TF_DDP_BUF1_VALID(1), 809 V_TF_DDP_BUF0_INDICATE(1) | V_TF_DDP_BUF1_INDICATE(1), 0, 0); 810 t4_set_tcb_field(sc, toep->ctrlq, toep, W_TCB_T_FLAGS, 811 V_TF_RCV_COALESCE_ENABLE(1), 0, 0, 0); 812} 813 814static int 815calculate_hcf(int n1, int n2) 816{ 817 int a, b, t; 818 819 if (n1 <= n2) { 820 a = n1; 821 b = n2; 822 } else { 823 a = n2; 824 b = n1; 825 } 826 827 while (a != 0) { 828 t = a; 829 a = b % a; 830 b = t; 831 } 832 833 return (b); 834} 835 836static inline int 837pages_to_nppods(int npages, int ddp_page_shift) 838{ 839 840 MPASS(ddp_page_shift >= PAGE_SHIFT); 841 842 return (howmany(npages >> (ddp_page_shift - PAGE_SHIFT), PPOD_PAGES)); 843} 844 845static int 846alloc_page_pods(struct ppod_region *pr, u_int nppods, u_int pgsz_idx, 847 struct ppod_reservation *prsv) 848{ 849 vmem_addr_t addr; /* relative to start of region */ 850 851 if (vmem_alloc(pr->pr_arena, PPOD_SZ(nppods), M_NOWAIT | M_FIRSTFIT, 852 &addr) != 0) 853 return (ENOMEM); 854 855 CTR5(KTR_CXGBE, "%-17s arena %p, addr 0x%08x, nppods %d, pgsz %d", 856 __func__, pr->pr_arena, (uint32_t)addr & pr->pr_tag_mask, 857 nppods, 1 << pr->pr_page_shift[pgsz_idx]); 858 859 /* 860 * The hardware tagmask includes an extra invalid bit but the arena was 861 * seeded with valid values only. An allocation out of this arena will 862 * fit inside the tagmask but won't have the invalid bit set. 863 */ 864 MPASS((addr & pr->pr_tag_mask) == addr); 865 MPASS((addr & pr->pr_invalid_bit) == 0); 866 867 prsv->prsv_pr = pr; 868 prsv->prsv_tag = V_PPOD_PGSZ(pgsz_idx) | addr; 869 prsv->prsv_nppods = nppods; 870 871 return (0); 872} 873 874int 875t4_alloc_page_pods_for_ps(struct ppod_region *pr, struct pageset *ps) 876{ 877 int i, hcf, seglen, idx, nppods; 878 struct ppod_reservation *prsv = &ps->prsv; 879 880 KASSERT(prsv->prsv_nppods == 0, 881 ("%s: page pods already allocated", __func__)); 882 883 /* 884 * The DDP page size is unrelated to the VM page size. We combine 885 * contiguous physical pages into larger segments to get the best DDP 886 * page size possible. This is the largest of the four sizes in 887 * A_ULP_RX_TDDP_PSZ that evenly divides the HCF of the segment sizes in 888 * the page list. 889 */ 890 hcf = 0; 891 for (i = 0; i < ps->npages; i++) { 892 seglen = PAGE_SIZE; 893 while (i < ps->npages - 1 && 894 ps->pages[i]->phys_addr + PAGE_SIZE == 895 ps->pages[i + 1]->phys_addr) { 896 seglen += PAGE_SIZE; 897 i++; 898 } 899 900 hcf = calculate_hcf(hcf, seglen); 901 if (hcf < (1 << pr->pr_page_shift[1])) { 902 idx = 0; 903 goto have_pgsz; /* give up, short circuit */ 904 } 905 } 906 907#define PR_PAGE_MASK(x) ((1 << pr->pr_page_shift[(x)]) - 1) 908 MPASS((hcf & PR_PAGE_MASK(0)) == 0); /* PAGE_SIZE is >= 4K everywhere */ 909 for (idx = nitems(pr->pr_page_shift) - 1; idx > 0; idx--) { 910 if ((hcf & PR_PAGE_MASK(idx)) == 0) 911 break; 912 } 913#undef PR_PAGE_MASK 914 915have_pgsz: 916 MPASS(idx <= M_PPOD_PGSZ); 917 918 nppods = pages_to_nppods(ps->npages, pr->pr_page_shift[idx]); 919 if (alloc_page_pods(pr, nppods, idx, prsv) != 0) 920 return (0); 921 MPASS(prsv->prsv_nppods > 0); 922 923 return (1); 924} 925 926int 927t4_alloc_page_pods_for_buf(struct ppod_region *pr, vm_offset_t buf, int len, 928 struct ppod_reservation *prsv) 929{ 930 int hcf, seglen, idx, npages, nppods; 931 uintptr_t start_pva, end_pva, pva, p1; 932 933 MPASS(buf > 0); 934 MPASS(len > 0); 935 936 /* 937 * The DDP page size is unrelated to the VM page size. We combine 938 * contiguous physical pages into larger segments to get the best DDP 939 * page size possible. This is the largest of the four sizes in 940 * A_ULP_RX_ISCSI_PSZ that evenly divides the HCF of the segment sizes 941 * in the page list. 942 */ 943 hcf = 0; 944 start_pva = trunc_page(buf); 945 end_pva = trunc_page(buf + len - 1); 946 pva = start_pva; 947 while (pva <= end_pva) { 948 seglen = PAGE_SIZE; 949 p1 = pmap_kextract(pva); 950 pva += PAGE_SIZE; 951 while (pva <= end_pva && p1 + seglen == pmap_kextract(pva)) { 952 seglen += PAGE_SIZE; 953 pva += PAGE_SIZE; 954 } 955 956 hcf = calculate_hcf(hcf, seglen); 957 if (hcf < (1 << pr->pr_page_shift[1])) { 958 idx = 0; 959 goto have_pgsz; /* give up, short circuit */ 960 } 961 } 962 963#define PR_PAGE_MASK(x) ((1 << pr->pr_page_shift[(x)]) - 1) 964 MPASS((hcf & PR_PAGE_MASK(0)) == 0); /* PAGE_SIZE is >= 4K everywhere */ 965 for (idx = nitems(pr->pr_page_shift) - 1; idx > 0; idx--) { 966 if ((hcf & PR_PAGE_MASK(idx)) == 0) 967 break; 968 } 969#undef PR_PAGE_MASK 970 971have_pgsz: 972 MPASS(idx <= M_PPOD_PGSZ); 973 974 npages = 1; 975 npages += (end_pva - start_pva) >> pr->pr_page_shift[idx]; 976 nppods = howmany(npages, PPOD_PAGES); 977 if (alloc_page_pods(pr, nppods, idx, prsv) != 0) 978 return (ENOMEM); 979 MPASS(prsv->prsv_nppods > 0); 980 981 return (0); 982} 983 984void 985t4_free_page_pods(struct ppod_reservation *prsv) 986{ 987 struct ppod_region *pr = prsv->prsv_pr; 988 vmem_addr_t addr; 989 990 MPASS(prsv != NULL); 991 MPASS(prsv->prsv_nppods != 0); 992 993 addr = prsv->prsv_tag & pr->pr_tag_mask; 994 MPASS((addr & pr->pr_invalid_bit) == 0); 995 996 CTR4(KTR_CXGBE, "%-17s arena %p, addr 0x%08x, nppods %d", __func__, 997 pr->pr_arena, addr, prsv->prsv_nppods); 998 999 vmem_free(pr->pr_arena, addr, PPOD_SZ(prsv->prsv_nppods)); 1000 prsv->prsv_nppods = 0; 1001} 1002 1003#define NUM_ULP_TX_SC_IMM_PPODS (256 / PPOD_SIZE) 1004 1005int 1006t4_write_page_pods_for_ps(struct adapter *sc, struct sge_wrq *wrq, int tid, 1007 struct pageset *ps) 1008{ 1009 struct wrqe *wr; 1010 struct ulp_mem_io *ulpmc; 1011 struct ulptx_idata *ulpsc; 1012 struct pagepod *ppod; 1013 int i, j, k, n, chunk, len, ddp_pgsz, idx; 1014 u_int ppod_addr; 1015 uint32_t cmd; 1016 struct ppod_reservation *prsv = &ps->prsv; 1017 struct ppod_region *pr = prsv->prsv_pr; 1018 1019 KASSERT(!(ps->flags & PS_PPODS_WRITTEN), 1020 ("%s: page pods already written", __func__)); 1021 MPASS(prsv->prsv_nppods > 0); 1022 1023 cmd = htobe32(V_ULPTX_CMD(ULP_TX_MEM_WRITE)); 1024 if (is_t4(sc)) 1025 cmd |= htobe32(F_ULP_MEMIO_ORDER); 1026 else 1027 cmd |= htobe32(F_T5_ULP_MEMIO_IMM); 1028 ddp_pgsz = 1 << pr->pr_page_shift[G_PPOD_PGSZ(prsv->prsv_tag)]; 1029 ppod_addr = pr->pr_start + (prsv->prsv_tag & pr->pr_tag_mask); 1030 for (i = 0; i < prsv->prsv_nppods; ppod_addr += chunk) { 1031 1032 /* How many page pods are we writing in this cycle */ 1033 n = min(prsv->prsv_nppods - i, NUM_ULP_TX_SC_IMM_PPODS); 1034 chunk = PPOD_SZ(n); 1035 len = roundup2(sizeof(*ulpmc) + sizeof(*ulpsc) + chunk, 16); 1036 1037 wr = alloc_wrqe(len, wrq); 1038 if (wr == NULL) 1039 return (ENOMEM); /* ok to just bail out */ 1040 ulpmc = wrtod(wr); 1041 1042 INIT_ULPTX_WR(ulpmc, len, 0, 0); 1043 ulpmc->cmd = cmd; 1044 ulpmc->dlen = htobe32(V_ULP_MEMIO_DATA_LEN(chunk / 32)); 1045 ulpmc->len16 = htobe32(howmany(len - sizeof(ulpmc->wr), 16)); 1046 ulpmc->lock_addr = htobe32(V_ULP_MEMIO_ADDR(ppod_addr >> 5)); 1047 1048 ulpsc = (struct ulptx_idata *)(ulpmc + 1); 1049 ulpsc->cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_IMM)); 1050 ulpsc->len = htobe32(chunk); 1051 1052 ppod = (struct pagepod *)(ulpsc + 1); 1053 for (j = 0; j < n; i++, j++, ppod++) { 1054 ppod->vld_tid_pgsz_tag_color = htobe64(F_PPOD_VALID | 1055 V_PPOD_TID(tid) | prsv->prsv_tag); 1056 ppod->len_offset = htobe64(V_PPOD_LEN(ps->len) | 1057 V_PPOD_OFST(ps->offset)); 1058 ppod->rsvd = 0; 1059 idx = i * PPOD_PAGES * (ddp_pgsz / PAGE_SIZE); 1060 for (k = 0; k < nitems(ppod->addr); k++) { 1061 if (idx < ps->npages) { 1062 ppod->addr[k] = 1063 htobe64(ps->pages[idx]->phys_addr); 1064 idx += ddp_pgsz / PAGE_SIZE; 1065 } else 1066 ppod->addr[k] = 0; 1067#if 0 1068 CTR5(KTR_CXGBE, 1069 "%s: tid %d ppod[%d]->addr[%d] = %p", 1070 __func__, toep->tid, i, k, 1071 htobe64(ppod->addr[k])); 1072#endif 1073 } 1074 1075 } 1076 1077 t4_wrq_tx(sc, wr); 1078 } 1079 ps->flags |= PS_PPODS_WRITTEN; 1080 1081 return (0); 1082} 1083 1084int 1085t4_write_page_pods_for_buf(struct adapter *sc, struct sge_wrq *wrq, int tid, 1086 struct ppod_reservation *prsv, vm_offset_t buf, int buflen) 1087{ 1088 struct wrqe *wr; 1089 struct ulp_mem_io *ulpmc; 1090 struct ulptx_idata *ulpsc; 1091 struct pagepod *ppod; 1092 int i, j, k, n, chunk, len, ddp_pgsz; 1093 u_int ppod_addr, offset; 1094 uint32_t cmd; 1095 struct ppod_region *pr = prsv->prsv_pr; 1096 uintptr_t end_pva, pva, pa; 1097 1098 cmd = htobe32(V_ULPTX_CMD(ULP_TX_MEM_WRITE)); 1099 if (is_t4(sc)) 1100 cmd |= htobe32(F_ULP_MEMIO_ORDER); 1101 else 1102 cmd |= htobe32(F_T5_ULP_MEMIO_IMM); 1103 ddp_pgsz = 1 << pr->pr_page_shift[G_PPOD_PGSZ(prsv->prsv_tag)]; 1104 offset = buf & PAGE_MASK; 1105 ppod_addr = pr->pr_start + (prsv->prsv_tag & pr->pr_tag_mask); 1106 pva = trunc_page(buf); 1107 end_pva = trunc_page(buf + buflen - 1); 1108 for (i = 0; i < prsv->prsv_nppods; ppod_addr += chunk) { 1109 1110 /* How many page pods are we writing in this cycle */ 1111 n = min(prsv->prsv_nppods - i, NUM_ULP_TX_SC_IMM_PPODS); 1112 MPASS(n > 0); 1113 chunk = PPOD_SZ(n); 1114 len = roundup2(sizeof(*ulpmc) + sizeof(*ulpsc) + chunk, 16); 1115 1116 wr = alloc_wrqe(len, wrq); 1117 if (wr == NULL) 1118 return (ENOMEM); /* ok to just bail out */ 1119 ulpmc = wrtod(wr); 1120 1121 INIT_ULPTX_WR(ulpmc, len, 0, 0); 1122 ulpmc->cmd = cmd; 1123 ulpmc->dlen = htobe32(V_ULP_MEMIO_DATA_LEN(chunk / 32)); 1124 ulpmc->len16 = htobe32(howmany(len - sizeof(ulpmc->wr), 16)); 1125 ulpmc->lock_addr = htobe32(V_ULP_MEMIO_ADDR(ppod_addr >> 5)); 1126 1127 ulpsc = (struct ulptx_idata *)(ulpmc + 1); 1128 ulpsc->cmd_more = htobe32(V_ULPTX_CMD(ULP_TX_SC_IMM)); 1129 ulpsc->len = htobe32(chunk); 1130 1131 ppod = (struct pagepod *)(ulpsc + 1); 1132 for (j = 0; j < n; i++, j++, ppod++) { 1133 ppod->vld_tid_pgsz_tag_color = htobe64(F_PPOD_VALID | 1134 V_PPOD_TID(tid) | 1135 (prsv->prsv_tag & ~V_PPOD_PGSZ(M_PPOD_PGSZ))); 1136 ppod->len_offset = htobe64(V_PPOD_LEN(buflen) | 1137 V_PPOD_OFST(offset)); 1138 ppod->rsvd = 0; 1139 1140 for (k = 0; k < nitems(ppod->addr); k++) { 1141 if (pva > end_pva) 1142 ppod->addr[k] = 0; 1143 else { 1144 pa = pmap_kextract(pva); 1145 ppod->addr[k] = htobe64(pa); 1146 pva += ddp_pgsz; 1147 } 1148#if 0 1149 CTR5(KTR_CXGBE, 1150 "%s: tid %d ppod[%d]->addr[%d] = %p", 1151 __func__, tid, i, k, 1152 htobe64(ppod->addr[k])); 1153#endif 1154 } 1155 1156 /* 1157 * Walk back 1 segment so that the first address in the 1158 * next pod is the same as the last one in the current 1159 * pod. 1160 */ 1161 pva -= ddp_pgsz; 1162 } 1163 1164 t4_wrq_tx(sc, wr); 1165 } 1166 1167 MPASS(pva <= end_pva); 1168 1169 return (0); 1170} 1171 1172/* 1173 * Prepare a pageset for DDP. This sets up page pods. 1174 */ 1175static int 1176prep_pageset(struct adapter *sc, struct toepcb *toep, struct pageset *ps) 1177{ 1178 struct tom_data *td = sc->tom_softc; 1179 1180 if (ps->prsv.prsv_nppods == 0 && 1181 !t4_alloc_page_pods_for_ps(&td->pr, ps)) { 1182 return (0); 1183 } 1184 if (!(ps->flags & PS_PPODS_WRITTEN) && 1185 t4_write_page_pods_for_ps(sc, toep->ctrlq, toep->tid, ps) != 0) { 1186 return (0); 1187 } 1188 1189 return (1); 1190} 1191 1192int 1193t4_init_ppod_region(struct ppod_region *pr, struct t4_range *r, u_int psz, 1194 const char *name) 1195{ 1196 int i; 1197 1198 MPASS(pr != NULL); 1199 MPASS(r->size > 0); 1200 1201 pr->pr_start = r->start; 1202 pr->pr_len = r->size; 1203 pr->pr_page_shift[0] = 12 + G_HPZ0(psz); 1204 pr->pr_page_shift[1] = 12 + G_HPZ1(psz); 1205 pr->pr_page_shift[2] = 12 + G_HPZ2(psz); 1206 pr->pr_page_shift[3] = 12 + G_HPZ3(psz); 1207 1208 /* The SGL -> page pod algorithm requires the sizes to be in order. */ 1209 for (i = 1; i < nitems(pr->pr_page_shift); i++) { 1210 if (pr->pr_page_shift[i] <= pr->pr_page_shift[i - 1]) 1211 return (ENXIO); 1212 } 1213 1214 pr->pr_tag_mask = ((1 << fls(r->size)) - 1) & V_PPOD_TAG(M_PPOD_TAG); 1215 pr->pr_alias_mask = V_PPOD_TAG(M_PPOD_TAG) & ~pr->pr_tag_mask; 1216 if (pr->pr_tag_mask == 0 || pr->pr_alias_mask == 0) 1217 return (ENXIO); 1218 pr->pr_alias_shift = fls(pr->pr_tag_mask); 1219 pr->pr_invalid_bit = 1 << (pr->pr_alias_shift - 1); 1220 1221 pr->pr_arena = vmem_create(name, 0, pr->pr_len, PPOD_SIZE, 0, 1222 M_FIRSTFIT | M_NOWAIT); 1223 if (pr->pr_arena == NULL) 1224 return (ENOMEM); 1225 1226 return (0); 1227} 1228 1229void 1230t4_free_ppod_region(struct ppod_region *pr) 1231{ 1232 1233 MPASS(pr != NULL); 1234 1235 if (pr->pr_arena) 1236 vmem_destroy(pr->pr_arena); 1237 bzero(pr, sizeof(*pr)); 1238} 1239 1240static int 1241pscmp(struct pageset *ps, struct vmspace *vm, vm_offset_t start, int npages, 1242 int pgoff, int len) 1243{ 1244 1245 if (ps->start != start || ps->npages != npages || 1246 ps->offset != pgoff || ps->len != len) 1247 return (1); 1248 1249 return (ps->vm != vm || ps->vm_timestamp != vm->vm_map.timestamp); 1250} 1251 1252static int 1253hold_aio(struct toepcb *toep, struct kaiocb *job, struct pageset **pps) 1254{ 1255 struct vmspace *vm; 1256 vm_map_t map; 1257 vm_offset_t start, end, pgoff; 1258 struct pageset *ps; 1259 int n; 1260 1261 DDP_ASSERT_LOCKED(toep); 1262 1263 /* 1264 * The AIO subsystem will cancel and drain all requests before 1265 * permitting a process to exit or exec, so p_vmspace should 1266 * be stable here. 1267 */ 1268 vm = job->userproc->p_vmspace; 1269 map = &vm->vm_map; 1270 start = (uintptr_t)job->uaiocb.aio_buf; 1271 pgoff = start & PAGE_MASK; 1272 end = round_page(start + job->uaiocb.aio_nbytes); 1273 start = trunc_page(start); 1274 1275 if (end - start > MAX_DDP_BUFFER_SIZE) { 1276 /* 1277 * Truncate the request to a short read. 1278 * Alternatively, we could DDP in chunks to the larger 1279 * buffer, but that would be quite a bit more work. 1280 * 1281 * When truncating, round the request down to avoid 1282 * crossing a cache line on the final transaction. 1283 */ 1284 end = rounddown2(start + MAX_DDP_BUFFER_SIZE, CACHE_LINE_SIZE); 1285#ifdef VERBOSE_TRACES 1286 CTR4(KTR_CXGBE, "%s: tid %d, truncating size from %lu to %lu", 1287 __func__, toep->tid, (unsigned long)job->uaiocb.aio_nbytes, 1288 (unsigned long)(end - (start + pgoff))); 1289 job->uaiocb.aio_nbytes = end - (start + pgoff); 1290#endif 1291 end = round_page(end); 1292 } 1293 1294 n = atop(end - start); 1295 1296 /* 1297 * Try to reuse a cached pageset. 1298 */ 1299 TAILQ_FOREACH(ps, &toep->ddp.cached_pagesets, link) { 1300 if (pscmp(ps, vm, start, n, pgoff, 1301 job->uaiocb.aio_nbytes) == 0) { 1302 TAILQ_REMOVE(&toep->ddp.cached_pagesets, ps, link); 1303 toep->ddp.cached_count--; 1304 *pps = ps; 1305 return (0); 1306 } 1307 } 1308 1309 /* 1310 * If there are too many cached pagesets to create a new one, 1311 * free a pageset before creating a new one. 1312 */ 1313 KASSERT(toep->ddp.active_count + toep->ddp.cached_count <= 1314 nitems(toep->ddp.db), ("%s: too many wired pagesets", __func__)); 1315 if (toep->ddp.active_count + toep->ddp.cached_count == 1316 nitems(toep->ddp.db)) { 1317 KASSERT(toep->ddp.cached_count > 0, 1318 ("no cached pageset to free")); 1319 ps = TAILQ_LAST(&toep->ddp.cached_pagesets, pagesetq); 1320 TAILQ_REMOVE(&toep->ddp.cached_pagesets, ps, link); 1321 toep->ddp.cached_count--; 1322 free_pageset(toep->td, ps); 1323 } 1324 DDP_UNLOCK(toep); 1325 1326 /* Create a new pageset. */ 1327 ps = malloc(sizeof(*ps) + n * sizeof(vm_page_t), M_CXGBE, M_WAITOK | 1328 M_ZERO); 1329 ps->pages = (vm_page_t *)(ps + 1); 1330 ps->vm_timestamp = map->timestamp; 1331 ps->npages = vm_fault_quick_hold_pages(map, start, end - start, 1332 VM_PROT_WRITE, ps->pages, n); 1333 1334 DDP_LOCK(toep); 1335 if (ps->npages < 0) { 1336 free(ps, M_CXGBE); 1337 return (EFAULT); 1338 } 1339 1340 KASSERT(ps->npages == n, ("hold_aio: page count mismatch: %d vs %d", 1341 ps->npages, n)); 1342 1343 ps->offset = pgoff; 1344 ps->len = job->uaiocb.aio_nbytes; 1345 refcount_acquire(&vm->vm_refcnt); 1346 ps->vm = vm; 1347 ps->start = start; 1348 1349 CTR5(KTR_CXGBE, "%s: tid %d, new pageset %p for job %p, npages %d", 1350 __func__, toep->tid, ps, job, ps->npages); 1351 *pps = ps; 1352 return (0); 1353} 1354 1355static void 1356ddp_complete_all(struct toepcb *toep, int error) 1357{ 1358 struct kaiocb *job; 1359 1360 DDP_ASSERT_LOCKED(toep); 1361 while (!TAILQ_EMPTY(&toep->ddp.aiojobq)) { 1362 job = TAILQ_FIRST(&toep->ddp.aiojobq); 1363 TAILQ_REMOVE(&toep->ddp.aiojobq, job, list); 1364 toep->ddp.waiting_count--; 1365 if (aio_clear_cancel_function(job)) 1366 ddp_complete_one(job, error); 1367 } 1368} 1369 1370static void 1371aio_ddp_cancel_one(struct kaiocb *job) 1372{ 1373 long copied; 1374 1375 /* 1376 * If this job had copied data out of the socket buffer before 1377 * it was cancelled, report it as a short read rather than an 1378 * error. 1379 */ 1380 copied = job->aio_received; 1381 if (copied != 0) 1382 aio_complete(job, copied, 0); 1383 else 1384 aio_cancel(job); 1385} 1386 1387/* 1388 * Called when the main loop wants to requeue a job to retry it later. 1389 * Deals with the race of the job being cancelled while it was being 1390 * examined. 1391 */ 1392static void 1393aio_ddp_requeue_one(struct toepcb *toep, struct kaiocb *job) 1394{ 1395 1396 DDP_ASSERT_LOCKED(toep); 1397 if (!(toep->ddp.flags & DDP_DEAD) && 1398 aio_set_cancel_function(job, t4_aio_cancel_queued)) { 1399 TAILQ_INSERT_HEAD(&toep->ddp.aiojobq, job, list); 1400 toep->ddp.waiting_count++; 1401 } else 1402 aio_ddp_cancel_one(job); 1403} 1404 1405static void 1406aio_ddp_requeue(struct toepcb *toep) 1407{ 1408 struct adapter *sc = td_adapter(toep->td); 1409 struct socket *so; 1410 struct sockbuf *sb; 1411 struct inpcb *inp; 1412 struct kaiocb *job; 1413 struct ddp_buffer *db; 1414 size_t copied, offset, resid; 1415 struct pageset *ps; 1416 struct mbuf *m; 1417 uint64_t ddp_flags, ddp_flags_mask; 1418 struct wrqe *wr; 1419 int buf_flag, db_idx, error; 1420 1421 DDP_ASSERT_LOCKED(toep); 1422 1423restart: 1424 if (toep->ddp.flags & DDP_DEAD) { 1425 MPASS(toep->ddp.waiting_count == 0); 1426 MPASS(toep->ddp.active_count == 0); 1427 return; 1428 } 1429 1430 if (toep->ddp.waiting_count == 0 || 1431 toep->ddp.active_count == nitems(toep->ddp.db)) { 1432 return; 1433 } 1434 1435 job = TAILQ_FIRST(&toep->ddp.aiojobq); 1436 so = job->fd_file->f_data; 1437 sb = &so->so_rcv; 1438 SOCKBUF_LOCK(sb); 1439 1440 /* We will never get anything unless we are or were connected. */ 1441 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 1442 SOCKBUF_UNLOCK(sb); 1443 ddp_complete_all(toep, ENOTCONN); 1444 return; 1445 } 1446 1447 KASSERT(toep->ddp.active_count == 0 || sbavail(sb) == 0, 1448 ("%s: pending sockbuf data and DDP is active", __func__)); 1449 1450 /* Abort if socket has reported problems. */ 1451 /* XXX: Wait for any queued DDP's to finish and/or flush them? */ 1452 if (so->so_error && sbavail(sb) == 0) { 1453 toep->ddp.waiting_count--; 1454 TAILQ_REMOVE(&toep->ddp.aiojobq, job, list); 1455 if (!aio_clear_cancel_function(job)) { 1456 SOCKBUF_UNLOCK(sb); 1457 goto restart; 1458 } 1459 1460 /* 1461 * If this job has previously copied some data, report 1462 * a short read and leave the error to be reported by 1463 * a future request. 1464 */ 1465 copied = job->aio_received; 1466 if (copied != 0) { 1467 SOCKBUF_UNLOCK(sb); 1468 aio_complete(job, copied, 0); 1469 goto restart; 1470 } 1471 error = so->so_error; 1472 so->so_error = 0; 1473 SOCKBUF_UNLOCK(sb); 1474 aio_complete(job, -1, error); 1475 goto restart; 1476 } 1477 1478 /* 1479 * Door is closed. If there is pending data in the socket buffer, 1480 * deliver it. If there are pending DDP requests, wait for those 1481 * to complete. Once they have completed, return EOF reads. 1482 */ 1483 if (sb->sb_state & SBS_CANTRCVMORE && sbavail(sb) == 0) { 1484 SOCKBUF_UNLOCK(sb); 1485 if (toep->ddp.active_count != 0) 1486 return; 1487 ddp_complete_all(toep, 0); 1488 return; 1489 } 1490 1491 /* 1492 * If DDP is not enabled and there is no pending socket buffer 1493 * data, try to enable DDP. 1494 */ 1495 if (sbavail(sb) == 0 && (toep->ddp.flags & DDP_ON) == 0) { 1496 SOCKBUF_UNLOCK(sb); 1497 1498 /* 1499 * Wait for the card to ACK that DDP is enabled before 1500 * queueing any buffers. Currently this waits for an 1501 * indicate to arrive. This could use a TCB_SET_FIELD_RPL 1502 * message to know that DDP was enabled instead of waiting 1503 * for the indicate which would avoid copying the indicate 1504 * if no data is pending. 1505 * 1506 * XXX: Might want to limit the indicate size to the size 1507 * of the first queued request. 1508 */ 1509 if ((toep->ddp.flags & DDP_SC_REQ) == 0) 1510 enable_ddp(sc, toep); 1511 return; 1512 } 1513 SOCKBUF_UNLOCK(sb); 1514 1515 /* 1516 * If another thread is queueing a buffer for DDP, let it 1517 * drain any work and return. 1518 */ 1519 if (toep->ddp.queueing != NULL) 1520 return; 1521 1522 /* Take the next job to prep it for DDP. */ 1523 toep->ddp.waiting_count--; 1524 TAILQ_REMOVE(&toep->ddp.aiojobq, job, list); 1525 if (!aio_clear_cancel_function(job)) 1526 goto restart; 1527 toep->ddp.queueing = job; 1528 1529 /* NB: This drops DDP_LOCK while it holds the backing VM pages. */ 1530 error = hold_aio(toep, job, &ps); 1531 if (error != 0) { 1532 ddp_complete_one(job, error); 1533 toep->ddp.queueing = NULL; 1534 goto restart; 1535 } 1536 1537 SOCKBUF_LOCK(sb); 1538 if (so->so_error && sbavail(sb) == 0) { 1539 copied = job->aio_received; 1540 if (copied != 0) { 1541 SOCKBUF_UNLOCK(sb); 1542 recycle_pageset(toep, ps); 1543 aio_complete(job, copied, 0); 1544 toep->ddp.queueing = NULL; 1545 goto restart; 1546 } 1547 1548 error = so->so_error; 1549 so->so_error = 0; 1550 SOCKBUF_UNLOCK(sb); 1551 recycle_pageset(toep, ps); 1552 aio_complete(job, -1, error); 1553 toep->ddp.queueing = NULL; 1554 goto restart; 1555 } 1556 1557 if (sb->sb_state & SBS_CANTRCVMORE && sbavail(sb) == 0) { 1558 SOCKBUF_UNLOCK(sb); 1559 recycle_pageset(toep, ps); 1560 if (toep->ddp.active_count != 0) { 1561 /* 1562 * The door is closed, but there are still pending 1563 * DDP buffers. Requeue. These jobs will all be 1564 * completed once those buffers drain. 1565 */ 1566 aio_ddp_requeue_one(toep, job); 1567 toep->ddp.queueing = NULL; 1568 return; 1569 } 1570 ddp_complete_one(job, 0); 1571 ddp_complete_all(toep, 0); 1572 toep->ddp.queueing = NULL; 1573 return; 1574 } 1575 1576sbcopy: 1577 /* 1578 * If the toep is dead, there shouldn't be any data in the socket 1579 * buffer, so the above case should have handled this. 1580 */ 1581 MPASS(!(toep->ddp.flags & DDP_DEAD)); 1582 1583 /* 1584 * If there is pending data in the socket buffer (either 1585 * from before the requests were queued or a DDP indicate), 1586 * copy those mbufs out directly. 1587 */ 1588 copied = 0; 1589 offset = ps->offset + job->aio_received; 1590 MPASS(job->aio_received <= job->uaiocb.aio_nbytes); 1591 resid = job->uaiocb.aio_nbytes - job->aio_received; 1592 m = sb->sb_mb; 1593 KASSERT(m == NULL || toep->ddp.active_count == 0, 1594 ("%s: sockbuf data with active DDP", __func__)); 1595 while (m != NULL && resid > 0) { 1596 struct iovec iov[1]; 1597 struct uio uio; 1598 int error; 1599 1600 iov[0].iov_base = mtod(m, void *); 1601 iov[0].iov_len = m->m_len; 1602 if (iov[0].iov_len > resid) 1603 iov[0].iov_len = resid; 1604 uio.uio_iov = iov; 1605 uio.uio_iovcnt = 1; 1606 uio.uio_offset = 0; 1607 uio.uio_resid = iov[0].iov_len; 1608 uio.uio_segflg = UIO_SYSSPACE; 1609 uio.uio_rw = UIO_WRITE; 1610 error = uiomove_fromphys(ps->pages, offset + copied, 1611 uio.uio_resid, &uio); 1612 MPASS(error == 0 && uio.uio_resid == 0); 1613 copied += uio.uio_offset; 1614 resid -= uio.uio_offset; 1615 m = m->m_next; 1616 } 1617 if (copied != 0) { 1618 sbdrop_locked(sb, copied); 1619 job->aio_received += copied; 1620 job->msgrcv = 1; 1621 copied = job->aio_received; 1622 inp = sotoinpcb(so); 1623 if (!INP_TRY_WLOCK(inp)) { 1624 /* 1625 * The reference on the socket file descriptor in 1626 * the AIO job should keep 'sb' and 'inp' stable. 1627 * Our caller has a reference on the 'toep' that 1628 * keeps it stable. 1629 */ 1630 SOCKBUF_UNLOCK(sb); 1631 DDP_UNLOCK(toep); 1632 INP_WLOCK(inp); 1633 DDP_LOCK(toep); 1634 SOCKBUF_LOCK(sb); 1635 1636 /* 1637 * If the socket has been closed, we should detect 1638 * that and complete this request if needed on 1639 * the next trip around the loop. 1640 */ 1641 } 1642 t4_rcvd_locked(&toep->td->tod, intotcpcb(inp)); 1643 INP_WUNLOCK(inp); 1644 if (resid == 0 || toep->ddp.flags & DDP_DEAD) { 1645 /* 1646 * We filled the entire buffer with socket 1647 * data, DDP is not being used, or the socket 1648 * is being shut down, so complete the 1649 * request. 1650 */ 1651 SOCKBUF_UNLOCK(sb); 1652 recycle_pageset(toep, ps); 1653 aio_complete(job, copied, 0); 1654 toep->ddp.queueing = NULL; 1655 goto restart; 1656 } 1657 1658 /* 1659 * If DDP is not enabled, requeue this request and restart. 1660 * This will either enable DDP or wait for more data to 1661 * arrive on the socket buffer. 1662 */ 1663 if ((toep->ddp.flags & (DDP_ON | DDP_SC_REQ)) != DDP_ON) { 1664 SOCKBUF_UNLOCK(sb); 1665 recycle_pageset(toep, ps); 1666 aio_ddp_requeue_one(toep, job); 1667 toep->ddp.queueing = NULL; 1668 goto restart; 1669 } 1670 1671 /* 1672 * An indicate might have arrived and been added to 1673 * the socket buffer while it was unlocked after the 1674 * copy to lock the INP. If so, restart the copy. 1675 */ 1676 if (sbavail(sb) != 0) 1677 goto sbcopy; 1678 } 1679 SOCKBUF_UNLOCK(sb); 1680 1681 if (prep_pageset(sc, toep, ps) == 0) { 1682 recycle_pageset(toep, ps); 1683 aio_ddp_requeue_one(toep, job); 1684 toep->ddp.queueing = NULL; 1685 1686 /* 1687 * XXX: Need to retry this later. Mostly need a trigger 1688 * when page pods are freed up. 1689 */ 1690 printf("%s: prep_pageset failed\n", __func__); 1691 return; 1692 } 1693 1694 /* Determine which DDP buffer to use. */ 1695 if (toep->ddp.db[0].job == NULL) { 1696 db_idx = 0; 1697 } else { 1698 MPASS(toep->ddp.db[1].job == NULL); 1699 db_idx = 1; 1700 } 1701 1702 ddp_flags = 0; 1703 ddp_flags_mask = 0; 1704 if (db_idx == 0) { 1705 ddp_flags |= V_TF_DDP_BUF0_VALID(1); 1706 if (so->so_state & SS_NBIO) 1707 ddp_flags |= V_TF_DDP_BUF0_FLUSH(1); 1708 ddp_flags_mask |= V_TF_DDP_PSH_NO_INVALIDATE0(1) | 1709 V_TF_DDP_PUSH_DISABLE_0(1) | V_TF_DDP_PSHF_ENABLE_0(1) | 1710 V_TF_DDP_BUF0_FLUSH(1) | V_TF_DDP_BUF0_VALID(1); 1711 buf_flag = DDP_BUF0_ACTIVE; 1712 } else { 1713 ddp_flags |= V_TF_DDP_BUF1_VALID(1); 1714 if (so->so_state & SS_NBIO) 1715 ddp_flags |= V_TF_DDP_BUF1_FLUSH(1); 1716 ddp_flags_mask |= V_TF_DDP_PSH_NO_INVALIDATE1(1) | 1717 V_TF_DDP_PUSH_DISABLE_1(1) | V_TF_DDP_PSHF_ENABLE_1(1) | 1718 V_TF_DDP_BUF1_FLUSH(1) | V_TF_DDP_BUF1_VALID(1); 1719 buf_flag = DDP_BUF1_ACTIVE; 1720 } 1721 MPASS((toep->ddp.flags & buf_flag) == 0); 1722 if ((toep->ddp.flags & (DDP_BUF0_ACTIVE | DDP_BUF1_ACTIVE)) == 0) { 1723 MPASS(db_idx == 0); 1724 MPASS(toep->ddp.active_id == -1); 1725 MPASS(toep->ddp.active_count == 0); 1726 ddp_flags_mask |= V_TF_DDP_ACTIVE_BUF(1); 1727 } 1728 1729 /* 1730 * The TID for this connection should still be valid. If DDP_DEAD 1731 * is set, SBS_CANTRCVMORE should be set, so we shouldn't be 1732 * this far anyway. Even if the socket is closing on the other 1733 * end, the AIO job holds a reference on this end of the socket 1734 * which will keep it open and keep the TCP PCB attached until 1735 * after the job is completed. 1736 */ 1737 wr = mk_update_tcb_for_ddp(sc, toep, db_idx, ps, job->aio_received, 1738 ddp_flags, ddp_flags_mask); 1739 if (wr == NULL) { 1740 recycle_pageset(toep, ps); 1741 aio_ddp_requeue_one(toep, job); 1742 toep->ddp.queueing = NULL; 1743 1744 /* 1745 * XXX: Need a way to kick a retry here. 1746 * 1747 * XXX: We know the fixed size needed and could 1748 * preallocate this using a blocking request at the 1749 * start of the task to avoid having to handle this 1750 * edge case. 1751 */ 1752 printf("%s: mk_update_tcb_for_ddp failed\n", __func__); 1753 return; 1754 } 1755 1756 if (!aio_set_cancel_function(job, t4_aio_cancel_active)) { 1757 free_wrqe(wr); 1758 recycle_pageset(toep, ps); 1759 aio_ddp_cancel_one(job); 1760 toep->ddp.queueing = NULL; 1761 goto restart; 1762 } 1763 1764#ifdef VERBOSE_TRACES 1765 CTR6(KTR_CXGBE, 1766 "%s: tid %u, scheduling %p for DDP[%d] (flags %#lx/%#lx)", __func__, 1767 toep->tid, job, db_idx, ddp_flags, ddp_flags_mask); 1768#endif 1769 /* Give the chip the go-ahead. */ 1770 t4_wrq_tx(sc, wr); 1771 db = &toep->ddp.db[db_idx]; 1772 db->cancel_pending = 0; 1773 db->job = job; 1774 db->ps = ps; 1775 toep->ddp.queueing = NULL; 1776 toep->ddp.flags |= buf_flag; 1777 toep->ddp.active_count++; 1778 if (toep->ddp.active_count == 1) { 1779 MPASS(toep->ddp.active_id == -1); 1780 toep->ddp.active_id = db_idx; 1781 CTR2(KTR_CXGBE, "%s: ddp_active_id = %d", __func__, 1782 toep->ddp.active_id); 1783 } 1784 goto restart; 1785} 1786 1787void 1788ddp_queue_toep(struct toepcb *toep) 1789{ 1790 1791 DDP_ASSERT_LOCKED(toep); 1792 if (toep->ddp.flags & DDP_TASK_ACTIVE) 1793 return; 1794 toep->ddp.flags |= DDP_TASK_ACTIVE; 1795 hold_toepcb(toep); 1796 soaio_enqueue(&toep->ddp.requeue_task); 1797} 1798 1799static void 1800aio_ddp_requeue_task(void *context, int pending) 1801{ 1802 struct toepcb *toep = context; 1803 1804 DDP_LOCK(toep); 1805 aio_ddp_requeue(toep); 1806 toep->ddp.flags &= ~DDP_TASK_ACTIVE; 1807 DDP_UNLOCK(toep); 1808 1809 free_toepcb(toep); 1810} 1811 1812static void 1813t4_aio_cancel_active(struct kaiocb *job) 1814{ 1815 struct socket *so = job->fd_file->f_data; 1816 struct tcpcb *tp = so_sototcpcb(so); 1817 struct toepcb *toep = tp->t_toe; 1818 struct adapter *sc = td_adapter(toep->td); 1819 uint64_t valid_flag; 1820 int i; 1821 1822 DDP_LOCK(toep); 1823 if (aio_cancel_cleared(job)) { 1824 DDP_UNLOCK(toep); 1825 aio_ddp_cancel_one(job); 1826 return; 1827 } 1828 1829 for (i = 0; i < nitems(toep->ddp.db); i++) { 1830 if (toep->ddp.db[i].job == job) { 1831 /* Should only ever get one cancel request for a job. */ 1832 MPASS(toep->ddp.db[i].cancel_pending == 0); 1833 1834 /* 1835 * Invalidate this buffer. It will be 1836 * cancelled or partially completed once the 1837 * card ACKs the invalidate. 1838 */ 1839 valid_flag = i == 0 ? V_TF_DDP_BUF0_VALID(1) : 1840 V_TF_DDP_BUF1_VALID(1); 1841 t4_set_tcb_field(sc, toep->ctrlq, toep, 1842 W_TCB_RX_DDP_FLAGS, valid_flag, 0, 1, 1843 CPL_COOKIE_DDP0 + i); 1844 toep->ddp.db[i].cancel_pending = 1; 1845 CTR2(KTR_CXGBE, "%s: request %p marked pending", 1846 __func__, job); 1847 break; 1848 } 1849 } 1850 DDP_UNLOCK(toep); 1851} 1852 1853static void 1854t4_aio_cancel_queued(struct kaiocb *job) 1855{ 1856 struct socket *so = job->fd_file->f_data; 1857 struct tcpcb *tp = so_sototcpcb(so); 1858 struct toepcb *toep = tp->t_toe; 1859 1860 DDP_LOCK(toep); 1861 if (!aio_cancel_cleared(job)) { 1862 TAILQ_REMOVE(&toep->ddp.aiojobq, job, list); 1863 toep->ddp.waiting_count--; 1864 if (toep->ddp.waiting_count == 0) 1865 ddp_queue_toep(toep); 1866 } 1867 CTR2(KTR_CXGBE, "%s: request %p cancelled", __func__, job); 1868 DDP_UNLOCK(toep); 1869 1870 aio_ddp_cancel_one(job); 1871} 1872 1873int 1874t4_aio_queue_ddp(struct socket *so, struct kaiocb *job) 1875{ 1876 struct tcpcb *tp = so_sototcpcb(so); 1877 struct toepcb *toep = tp->t_toe; 1878 1879 1880 /* Ignore writes. */ 1881 if (job->uaiocb.aio_lio_opcode != LIO_READ) 1882 return (EOPNOTSUPP); 1883 1884 DDP_LOCK(toep); 1885 1886 /* 1887 * XXX: Think about possibly returning errors for ENOTCONN, 1888 * etc. Perhaps the caller would only queue the request 1889 * if it failed with EOPNOTSUPP? 1890 */ 1891 1892#ifdef VERBOSE_TRACES 1893 CTR3(KTR_CXGBE, "%s: queueing %p for tid %u", __func__, job, toep->tid); 1894#endif 1895 if (!aio_set_cancel_function(job, t4_aio_cancel_queued)) 1896 panic("new job was cancelled"); 1897 TAILQ_INSERT_TAIL(&toep->ddp.aiojobq, job, list); 1898 toep->ddp.waiting_count++; 1899 toep->ddp.flags |= DDP_OK; 1900 1901 /* 1902 * Try to handle this request synchronously. If this has 1903 * to block because the task is running, it will just bail 1904 * and let the task handle it instead. 1905 */ 1906 aio_ddp_requeue(toep); 1907 DDP_UNLOCK(toep); 1908 return (0); 1909} 1910 1911void 1912t4_ddp_mod_load(void) 1913{ 1914 1915 t4_register_shared_cpl_handler(CPL_SET_TCB_RPL, do_ddp_tcb_rpl, 1916 CPL_COOKIE_DDP0); 1917 t4_register_shared_cpl_handler(CPL_SET_TCB_RPL, do_ddp_tcb_rpl, 1918 CPL_COOKIE_DDP1); 1919 t4_register_cpl_handler(CPL_RX_DATA_DDP, do_rx_data_ddp); 1920 t4_register_cpl_handler(CPL_RX_DDP_COMPLETE, do_rx_ddp_complete); 1921 TAILQ_INIT(&ddp_orphan_pagesets); 1922 mtx_init(&ddp_orphan_pagesets_lock, "ddp orphans", NULL, MTX_DEF); 1923 TASK_INIT(&ddp_orphan_task, 0, ddp_free_orphan_pagesets, NULL); 1924} 1925 1926void 1927t4_ddp_mod_unload(void) 1928{ 1929 1930 taskqueue_drain(taskqueue_thread, &ddp_orphan_task); 1931 MPASS(TAILQ_EMPTY(&ddp_orphan_pagesets)); 1932 mtx_destroy(&ddp_orphan_pagesets_lock); 1933 t4_register_shared_cpl_handler(CPL_SET_TCB_RPL, NULL, CPL_COOKIE_DDP0); 1934 t4_register_shared_cpl_handler(CPL_SET_TCB_RPL, NULL, CPL_COOKIE_DDP1); 1935 t4_register_cpl_handler(CPL_RX_DATA_DDP, NULL); 1936 t4_register_cpl_handler(CPL_RX_DDP_COMPLETE, NULL); 1937} 1938#endif 1939