1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 /* 28 * sun4v LDC Link Layer 29 */ 30 #include <sys/types.h> 31 #include <sys/file.h> 32 #include <sys/errno.h> 33 #include <sys/open.h> 34 #include <sys/cred.h> 35 #include <sys/kmem.h> 36 #include <sys/conf.h> 37 #include <sys/cmn_err.h> 38 #include <sys/ksynch.h> 39 #include <sys/modctl.h> 40 #include <sys/stat.h> /* needed for S_IFBLK and S_IFCHR */ 41 #include <sys/debug.h> 42 #include <sys/cred.h> 43 #include <sys/promif.h> 44 #include <sys/ddi.h> 45 #include <sys/sunddi.h> 46 #include <sys/cyclic.h> 47 #include <sys/machsystm.h> 48 #include <sys/vm.h> 49 #include <sys/cpu.h> 50 #include <sys/intreg.h> 51 #include <sys/machcpuvar.h> 52 #include <sys/mmu.h> 53 #include <sys/pte.h> 54 #include <vm/hat.h> 55 #include <vm/as.h> 56 #include <vm/hat_sfmmu.h> 57 #include <sys/vm_machparam.h> 58 #include <vm/seg_kmem.h> 59 #include <vm/seg_kpm.h> 60 #include <sys/note.h> 61 #include <sys/ivintr.h> 62 #include <sys/hypervisor_api.h> 63 #include <sys/ldc.h> 64 #include <sys/ldc_impl.h> 65 #include <sys/cnex.h> 66 #include <sys/hsvc.h> 67 #include <sys/sdt.h> 68 #include <sys/kldc.h> 69 70 /* Core internal functions */ 71 int i_ldc_h2v_error(int h_error); 72 void i_ldc_reset(ldc_chan_t *ldcp, boolean_t force_reset); 73 74 static int i_ldc_txq_reconf(ldc_chan_t *ldcp); 75 static int i_ldc_rxq_reconf(ldc_chan_t *ldcp, boolean_t force_reset); 76 static void i_ldc_rxq_drain(ldc_chan_t *ldcp); 77 static void i_ldc_reset_state(ldc_chan_t *ldcp); 78 static void i_ldc_debug_enter(void); 79 80 static int i_ldc_get_tx_tail(ldc_chan_t *ldcp, uint64_t *tail); 81 static void i_ldc_get_tx_head(ldc_chan_t *ldcp, uint64_t *head); 82 static int i_ldc_set_tx_tail(ldc_chan_t *ldcp, uint64_t tail); 83 static int i_ldc_set_rx_head(ldc_chan_t *ldcp, uint64_t head); 84 static int i_ldc_send_pkt(ldc_chan_t *ldcp, uint8_t pkttype, uint8_t subtype, 85 uint8_t ctrlmsg); 86 87 static int i_ldc_set_rxdq_head(ldc_chan_t *ldcp, uint64_t head); 88 static void i_ldc_rxdq_copy(ldc_chan_t *ldcp, uint64_t *head); 89 static uint64_t i_ldc_dq_rx_get_state(ldc_chan_t *ldcp, uint64_t *head, 90 uint64_t *tail, uint64_t *link_state); 91 static uint64_t i_ldc_hvq_rx_get_state(ldc_chan_t *ldcp, uint64_t *head, 92 uint64_t *tail, uint64_t *link_state); 93 static int i_ldc_rx_ackpeek(ldc_chan_t *ldcp, uint64_t rx_head, 94 uint64_t rx_tail); 95 static uint_t i_ldc_chkq(ldc_chan_t *ldcp); 96 97 /* Interrupt handling functions */ 98 static uint_t i_ldc_tx_hdlr(caddr_t arg1, caddr_t arg2); 99 static uint_t i_ldc_rx_hdlr(caddr_t arg1, caddr_t arg2); 100 static uint_t i_ldc_rx_process_hvq(ldc_chan_t *ldcp, boolean_t *notify_client, 101 uint64_t *notify_event); 102 static void i_ldc_clear_intr(ldc_chan_t *ldcp, cnex_intrtype_t itype); 103 104 /* Read method functions */ 105 static int i_ldc_read_raw(ldc_chan_t *ldcp, caddr_t target_bufp, size_t *sizep); 106 static int i_ldc_read_packet(ldc_chan_t *ldcp, caddr_t target_bufp, 107 size_t *sizep); 108 static int i_ldc_read_stream(ldc_chan_t *ldcp, caddr_t target_bufp, 109 size_t *sizep); 110 111 /* Write method functions */ 112 static int i_ldc_write_raw(ldc_chan_t *ldcp, caddr_t target_bufp, 113 size_t *sizep); 114 static int i_ldc_write_packet(ldc_chan_t *ldcp, caddr_t target_bufp, 115 size_t *sizep); 116 static int i_ldc_write_stream(ldc_chan_t *ldcp, caddr_t target_bufp, 117 size_t *sizep); 118 119 /* Pkt processing internal functions */ 120 static int i_ldc_check_seqid(ldc_chan_t *ldcp, ldc_msg_t *ldcmsg); 121 static int i_ldc_ctrlmsg(ldc_chan_t *ldcp, ldc_msg_t *ldcmsg); 122 static int i_ldc_process_VER(ldc_chan_t *ldcp, ldc_msg_t *msg); 123 static int i_ldc_process_RTS(ldc_chan_t *ldcp, ldc_msg_t *msg); 124 static int i_ldc_process_RTR(ldc_chan_t *ldcp, ldc_msg_t *msg); 125 static int i_ldc_process_RDX(ldc_chan_t *ldcp, ldc_msg_t *msg); 126 static int i_ldc_process_data_ACK(ldc_chan_t *ldcp, ldc_msg_t *msg); 127 128 /* LDC Version */ 129 static ldc_ver_t ldc_versions[] = { {1, 0} }; 130 131 /* number of supported versions */ 132 #define LDC_NUM_VERS (sizeof (ldc_versions) / sizeof (ldc_versions[0])) 133 134 /* Invalid value for the ldc_chan_t rx_ack_head field */ 135 #define ACKPEEK_HEAD_INVALID ((uint64_t)-1) 136 137 138 /* Module State Pointer */ 139 ldc_soft_state_t *ldcssp; 140 141 static struct modldrv md = { 142 &mod_miscops, /* This is a misc module */ 143 "sun4v LDC module", /* Name of the module */ 144 }; 145 146 static struct modlinkage ml = { 147 MODREV_1, 148 &md, 149 NULL 150 }; 151 152 static uint64_t ldc_sup_minor; /* Supported minor number */ 153 static hsvc_info_t ldc_hsvc = { 154 HSVC_REV_1, NULL, HSVC_GROUP_LDC, 1, 1, "ldc" 155 }; 156 157 /* 158 * The no. of MTU size messages that can be stored in 159 * the LDC Tx queue. The number of Tx queue entries is 160 * then computed as (mtu * mtu_msgs)/sizeof(queue_entry) 161 */ 162 uint64_t ldc_mtu_msgs = LDC_MTU_MSGS; 163 164 /* 165 * The minimum queue length. This is the size of the smallest 166 * LDC queue. If the computed value is less than this default, 167 * the queue length is rounded up to 'ldc_queue_entries'. 168 */ 169 uint64_t ldc_queue_entries = LDC_QUEUE_ENTRIES; 170 171 /* 172 * The length of the reliable-mode data queue in terms of the LDC 173 * receive queue length. i.e., the number of times larger than the 174 * LDC receive queue that the data queue should be. The HV receive 175 * queue is required to be a power of 2 and this implementation 176 * assumes the data queue will also be a power of 2. By making the 177 * multiplier a power of 2, we ensure the data queue will be a 178 * power of 2. We use a multiplier because the receive queue is 179 * sized to be sane relative to the MTU and the same is needed for 180 * the data queue. 181 */ 182 uint64_t ldc_rxdq_multiplier = LDC_RXDQ_MULTIPLIER; 183 184 /* 185 * LDC retry count and delay - when the HV returns EWOULDBLOCK 186 * the operation is retried 'ldc_max_retries' times with a 187 * wait of 'ldc_delay' usecs between each retry. 188 */ 189 int ldc_max_retries = LDC_MAX_RETRIES; 190 clock_t ldc_delay = LDC_DELAY; 191 192 /* 193 * Channels which have a devclass satisfying the following 194 * will be reset when entering the prom or kmdb. 195 * 196 * LDC_DEVCLASS_PROM_RESET(devclass) != 0 197 * 198 * By default, only block device service channels are reset. 199 */ 200 #define LDC_DEVCLASS_BIT(dc) (0x1 << (dc)) 201 #define LDC_DEVCLASS_PROM_RESET(dc) \ 202 (LDC_DEVCLASS_BIT(dc) & ldc_debug_reset_mask) 203 static uint64_t ldc_debug_reset_mask = LDC_DEVCLASS_BIT(LDC_DEV_BLK_SVC) | 204 LDC_DEVCLASS_BIT(LDC_DEV_GENERIC); 205 206 /* 207 * delay between each retry of channel unregistration in 208 * ldc_close(), to wait for pending interrupts to complete. 209 */ 210 clock_t ldc_close_delay = LDC_CLOSE_DELAY; 211 212 #ifdef DEBUG 213 214 /* 215 * Print debug messages 216 * 217 * set ldcdbg to 0x7 for enabling all msgs 218 * 0x4 - Warnings 219 * 0x2 - All debug messages 220 * 0x1 - Minimal debug messages 221 * 222 * set ldcdbgchan to the channel number you want to debug 223 * setting it to -1 prints debug messages for all channels 224 * NOTE: ldcdbgchan has no effect on error messages 225 */ 226 227 int ldcdbg = 0x0; 228 int64_t ldcdbgchan = DBG_ALL_LDCS; 229 uint64_t ldc_inject_err_flag = 0; 230 231 void 232 ldcdebug(int64_t id, const char *fmt, ...) 233 { 234 char buf[512]; 235 va_list ap; 236 237 /* 238 * Do not return if, 239 * caller wants to print it anyway - (id == DBG_ALL_LDCS) 240 * debug channel is set to all LDCs - (ldcdbgchan == DBG_ALL_LDCS) 241 * debug channel = caller specified channel 242 */ 243 if ((id != DBG_ALL_LDCS) && 244 (ldcdbgchan != DBG_ALL_LDCS) && 245 (ldcdbgchan != id)) { 246 return; 247 } 248 249 va_start(ap, fmt); 250 (void) vsprintf(buf, fmt, ap); 251 va_end(ap); 252 253 cmn_err(CE_CONT, "?%s", buf); 254 } 255 256 #define LDC_ERR_RESET 0x1 257 #define LDC_ERR_PKTLOSS 0x2 258 #define LDC_ERR_DQFULL 0x4 259 #define LDC_ERR_DRNGCLEAR 0x8 260 261 static boolean_t 262 ldc_inject_error(ldc_chan_t *ldcp, uint64_t error) 263 { 264 if ((ldcdbgchan != DBG_ALL_LDCS) && (ldcdbgchan != ldcp->id)) 265 return (B_FALSE); 266 267 if ((ldc_inject_err_flag & error) == 0) 268 return (B_FALSE); 269 270 /* clear the injection state */ 271 ldc_inject_err_flag &= ~error; 272 273 return (B_TRUE); 274 } 275 276 #define D1 \ 277 if (ldcdbg & 0x01) \ 278 ldcdebug 279 280 #define D2 \ 281 if (ldcdbg & 0x02) \ 282 ldcdebug 283 284 #define DWARN \ 285 if (ldcdbg & 0x04) \ 286 ldcdebug 287 288 #define DUMP_PAYLOAD(id, addr) \ 289 { \ 290 char buf[65*3]; \ 291 int i; \ 292 uint8_t *src = (uint8_t *)addr; \ 293 for (i = 0; i < 64; i++, src++) \ 294 (void) sprintf(&buf[i * 3], "|%02x", *src); \ 295 (void) sprintf(&buf[i * 3], "|\n"); \ 296 D2((id), "payload: %s", buf); \ 297 } 298 299 #define DUMP_LDC_PKT(c, s, addr) \ 300 { \ 301 ldc_msg_t *msg = (ldc_msg_t *)(addr); \ 302 uint32_t mid = ((c)->mode != LDC_MODE_RAW) ? msg->seqid : 0; \ 303 if (msg->type == LDC_DATA) { \ 304 D2((c)->id, "%s: msg%d (/%x/%x/%x/,env[%c%c,sz=%d])", \ 305 (s), mid, msg->type, msg->stype, msg->ctrl, \ 306 (msg->env & LDC_FRAG_START) ? 'B' : ' ', \ 307 (msg->env & LDC_FRAG_STOP) ? 'E' : ' ', \ 308 (msg->env & LDC_LEN_MASK)); \ 309 } else { \ 310 D2((c)->id, "%s: msg%d (/%x/%x/%x/,env=%x)", (s), \ 311 mid, msg->type, msg->stype, msg->ctrl, msg->env); \ 312 } \ 313 } 314 315 #define LDC_INJECT_RESET(_ldcp) ldc_inject_error(_ldcp, LDC_ERR_RESET) 316 #define LDC_INJECT_PKTLOSS(_ldcp) ldc_inject_error(_ldcp, LDC_ERR_PKTLOSS) 317 #define LDC_INJECT_DQFULL(_ldcp) ldc_inject_error(_ldcp, LDC_ERR_DQFULL) 318 #define LDC_INJECT_DRNGCLEAR(_ldcp) ldc_inject_error(_ldcp, LDC_ERR_DRNGCLEAR) 319 extern void i_ldc_mem_inject_dring_clear(ldc_chan_t *ldcp); 320 321 #else 322 323 #define DBG_ALL_LDCS -1 324 325 #define D1 326 #define D2 327 #define DWARN 328 329 #define DUMP_PAYLOAD(id, addr) 330 #define DUMP_LDC_PKT(c, s, addr) 331 332 #define LDC_INJECT_RESET(_ldcp) (B_FALSE) 333 #define LDC_INJECT_PKTLOSS(_ldcp) (B_FALSE) 334 #define LDC_INJECT_DQFULL(_ldcp) (B_FALSE) 335 #define LDC_INJECT_DRNGCLEAR(_ldcp) (B_FALSE) 336 337 #endif 338 339 /* 340 * dtrace SDT probes to ease tracing of the rx data queue and HV queue 341 * lengths. Just pass the head, tail, and entries values so that the 342 * length can be calculated in a dtrace script when the probe is enabled. 343 */ 344 #define TRACE_RXDQ_LENGTH(ldcp) \ 345 DTRACE_PROBE4(rxdq__size, \ 346 uint64_t, ldcp->id, \ 347 uint64_t, ldcp->rx_dq_head, \ 348 uint64_t, ldcp->rx_dq_tail, \ 349 uint64_t, ldcp->rx_dq_entries) 350 351 #define TRACE_RXHVQ_LENGTH(ldcp, head, tail) \ 352 DTRACE_PROBE4(rxhvq__size, \ 353 uint64_t, ldcp->id, \ 354 uint64_t, head, \ 355 uint64_t, tail, \ 356 uint64_t, ldcp->rx_q_entries) 357 358 /* A dtrace SDT probe to ease tracing of data queue copy operations */ 359 #define TRACE_RXDQ_COPY(ldcp, bytes) \ 360 DTRACE_PROBE2(rxdq__copy, uint64_t, ldcp->id, uint64_t, bytes) \ 361 362 /* The amount of contiguous space at the tail of the queue */ 363 #define Q_CONTIG_SPACE(head, tail, size) \ 364 ((head) <= (tail) ? ((size) - (tail)) : \ 365 ((head) - (tail) - LDC_PACKET_SIZE)) 366 367 #define ZERO_PKT(p) \ 368 bzero((p), sizeof (ldc_msg_t)); 369 370 #define IDX2COOKIE(idx, pg_szc, pg_shift) \ 371 (((pg_szc) << LDC_COOKIE_PGSZC_SHIFT) | ((idx) << (pg_shift))) 372 373 int 374 _init(void) 375 { 376 int status; 377 extern void i_ldc_mem_set_hsvc_vers(uint64_t major, uint64_t minor); 378 379 status = hsvc_register(&ldc_hsvc, &ldc_sup_minor); 380 if (status != 0) { 381 cmn_err(CE_NOTE, "!%s: cannot negotiate hypervisor LDC services" 382 " group: 0x%lx major: %ld minor: %ld errno: %d", 383 ldc_hsvc.hsvc_modname, ldc_hsvc.hsvc_group, 384 ldc_hsvc.hsvc_major, ldc_hsvc.hsvc_minor, status); 385 return (-1); 386 } 387 388 /* Initialize shared memory HV API version checking */ 389 i_ldc_mem_set_hsvc_vers(ldc_hsvc.hsvc_major, ldc_sup_minor); 390 391 /* allocate soft state structure */ 392 ldcssp = kmem_zalloc(sizeof (ldc_soft_state_t), KM_SLEEP); 393 394 /* Link the module into the system */ 395 status = mod_install(&ml); 396 if (status != 0) { 397 kmem_free(ldcssp, sizeof (ldc_soft_state_t)); 398 return (status); 399 } 400 401 /* Initialize the LDC state structure */ 402 mutex_init(&ldcssp->lock, NULL, MUTEX_DRIVER, NULL); 403 404 mutex_enter(&ldcssp->lock); 405 406 /* Create a cache for memory handles */ 407 ldcssp->memhdl_cache = kmem_cache_create("ldc_memhdl_cache", 408 sizeof (ldc_mhdl_t), 0, NULL, NULL, NULL, NULL, NULL, 0); 409 if (ldcssp->memhdl_cache == NULL) { 410 DWARN(DBG_ALL_LDCS, "_init: ldc_memhdl cache create failed\n"); 411 mutex_exit(&ldcssp->lock); 412 return (-1); 413 } 414 415 /* Create cache for memory segment structures */ 416 ldcssp->memseg_cache = kmem_cache_create("ldc_memseg_cache", 417 sizeof (ldc_memseg_t), 0, NULL, NULL, NULL, NULL, NULL, 0); 418 if (ldcssp->memseg_cache == NULL) { 419 DWARN(DBG_ALL_LDCS, "_init: ldc_memseg cache create failed\n"); 420 mutex_exit(&ldcssp->lock); 421 return (-1); 422 } 423 424 425 ldcssp->channel_count = 0; 426 ldcssp->channels_open = 0; 427 ldcssp->chan_list = NULL; 428 ldcssp->dring_list = NULL; 429 430 /* Register debug_enter callback */ 431 kldc_set_debug_cb(&i_ldc_debug_enter); 432 433 mutex_exit(&ldcssp->lock); 434 435 return (0); 436 } 437 438 int 439 _info(struct modinfo *modinfop) 440 { 441 /* Report status of the dynamically loadable driver module */ 442 return (mod_info(&ml, modinfop)); 443 } 444 445 int 446 _fini(void) 447 { 448 int rv, status; 449 ldc_chan_t *tmp_ldcp, *ldcp; 450 ldc_dring_t *tmp_dringp, *dringp; 451 ldc_mem_info_t minfo; 452 453 /* Unlink the driver module from the system */ 454 status = mod_remove(&ml); 455 if (status) { 456 DWARN(DBG_ALL_LDCS, "_fini: mod_remove failed\n"); 457 return (EIO); 458 } 459 460 /* Unregister debug_enter callback */ 461 kldc_set_debug_cb(NULL); 462 463 /* Free descriptor rings */ 464 dringp = ldcssp->dring_list; 465 while (dringp != NULL) { 466 tmp_dringp = dringp->next; 467 468 rv = ldc_mem_dring_info((ldc_dring_handle_t)dringp, &minfo); 469 if (rv == 0 && minfo.status != LDC_UNBOUND) { 470 if (minfo.status == LDC_BOUND) { 471 (void) ldc_mem_dring_unbind( 472 (ldc_dring_handle_t)dringp); 473 } 474 if (minfo.status == LDC_MAPPED) { 475 (void) ldc_mem_dring_unmap( 476 (ldc_dring_handle_t)dringp); 477 } 478 } 479 480 (void) ldc_mem_dring_destroy((ldc_dring_handle_t)dringp); 481 dringp = tmp_dringp; 482 } 483 ldcssp->dring_list = NULL; 484 485 /* close and finalize channels */ 486 ldcp = ldcssp->chan_list; 487 while (ldcp != NULL) { 488 tmp_ldcp = ldcp->next; 489 490 (void) ldc_close((ldc_handle_t)ldcp); 491 (void) ldc_fini((ldc_handle_t)ldcp); 492 493 ldcp = tmp_ldcp; 494 } 495 ldcssp->chan_list = NULL; 496 497 /* Destroy kmem caches */ 498 kmem_cache_destroy(ldcssp->memhdl_cache); 499 kmem_cache_destroy(ldcssp->memseg_cache); 500 501 /* 502 * We have successfully "removed" the driver. 503 * Destroying soft states 504 */ 505 mutex_destroy(&ldcssp->lock); 506 kmem_free(ldcssp, sizeof (ldc_soft_state_t)); 507 508 (void) hsvc_unregister(&ldc_hsvc); 509 510 return (status); 511 } 512 513 /* -------------------------------------------------------------------------- */ 514 515 /* 516 * LDC Link Layer Internal Functions 517 */ 518 519 /* 520 * Translate HV Errors to sun4v error codes 521 */ 522 int 523 i_ldc_h2v_error(int h_error) 524 { 525 switch (h_error) { 526 527 case H_EOK: 528 return (0); 529 530 case H_ENORADDR: 531 return (EFAULT); 532 533 case H_EBADPGSZ: 534 case H_EINVAL: 535 return (EINVAL); 536 537 case H_EWOULDBLOCK: 538 return (EWOULDBLOCK); 539 540 case H_ENOACCESS: 541 case H_ENOMAP: 542 return (EACCES); 543 544 case H_EIO: 545 case H_ECPUERROR: 546 return (EIO); 547 548 case H_ENOTSUPPORTED: 549 return (ENOTSUP); 550 551 case H_ETOOMANY: 552 return (ENOSPC); 553 554 case H_ECHANNEL: 555 return (ECHRNG); 556 default: 557 break; 558 } 559 560 return (EIO); 561 } 562 563 /* 564 * Reconfigure the transmit queue 565 */ 566 static int 567 i_ldc_txq_reconf(ldc_chan_t *ldcp) 568 { 569 int rv; 570 571 ASSERT(MUTEX_HELD(&ldcp->lock)); 572 ASSERT(MUTEX_HELD(&ldcp->tx_lock)); 573 574 rv = hv_ldc_tx_qconf(ldcp->id, ldcp->tx_q_ra, ldcp->tx_q_entries); 575 if (rv) { 576 cmn_err(CE_WARN, 577 "i_ldc_txq_reconf: (0x%lx) cannot set qconf", ldcp->id); 578 return (EIO); 579 } 580 rv = hv_ldc_tx_get_state(ldcp->id, &(ldcp->tx_head), 581 &(ldcp->tx_tail), &(ldcp->link_state)); 582 if (rv) { 583 cmn_err(CE_WARN, 584 "i_ldc_txq_reconf: (0x%lx) cannot get qptrs", ldcp->id); 585 return (EIO); 586 } 587 D1(ldcp->id, "i_ldc_txq_reconf: (0x%llx) h=0x%llx,t=0x%llx," 588 "s=0x%llx\n", ldcp->id, ldcp->tx_head, ldcp->tx_tail, 589 ldcp->link_state); 590 591 return (0); 592 } 593 594 /* 595 * Reconfigure the receive queue 596 */ 597 static int 598 i_ldc_rxq_reconf(ldc_chan_t *ldcp, boolean_t force_reset) 599 { 600 int rv; 601 uint64_t rx_head, rx_tail; 602 603 ASSERT(MUTEX_HELD(&ldcp->lock)); 604 rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail, 605 &(ldcp->link_state)); 606 if (rv) { 607 cmn_err(CE_WARN, 608 "i_ldc_rxq_reconf: (0x%lx) cannot get state", 609 ldcp->id); 610 return (EIO); 611 } 612 613 if (force_reset || (ldcp->tstate & ~TS_IN_RESET) == TS_UP) { 614 rv = hv_ldc_rx_qconf(ldcp->id, ldcp->rx_q_ra, 615 ldcp->rx_q_entries); 616 if (rv) { 617 cmn_err(CE_WARN, 618 "i_ldc_rxq_reconf: (0x%lx) cannot set qconf", 619 ldcp->id); 620 return (EIO); 621 } 622 D1(ldcp->id, "i_ldc_rxq_reconf: (0x%llx) completed q reconf", 623 ldcp->id); 624 } 625 626 return (0); 627 } 628 629 630 /* 631 * Drain the contents of the receive queue 632 */ 633 static void 634 i_ldc_rxq_drain(ldc_chan_t *ldcp) 635 { 636 int rv; 637 uint64_t rx_head, rx_tail; 638 int retries = 0; 639 640 ASSERT(MUTEX_HELD(&ldcp->lock)); 641 rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail, 642 &(ldcp->link_state)); 643 if (rv) { 644 cmn_err(CE_WARN, "i_ldc_rxq_drain: (0x%lx) cannot get state, " 645 "rv = 0x%x", ldcp->id, rv); 646 return; 647 } 648 649 /* If the queue is already empty just return success. */ 650 if (rx_head == rx_tail) 651 return; 652 653 /* 654 * We are draining the queue in order to close the channel. 655 * Call hv_ldc_rx_set_qhead directly instead of i_ldc_set_rx_head 656 * because we do not need to reset the channel if the set 657 * qhead fails. 658 */ 659 if ((rv = hv_ldc_rx_set_qhead(ldcp->id, rx_tail)) == 0) 660 return; 661 662 while ((rv == H_EWOULDBLOCK) && (retries++ < ldc_max_retries)) { 663 drv_usecwait(ldc_delay); 664 if ((rv = hv_ldc_rx_set_qhead(ldcp->id, rx_tail)) == 0) 665 return; 666 } 667 668 cmn_err(CE_WARN, "i_ldc_rxq_drain: (0x%lx) cannot set qhead 0x%lx, " 669 "rv = 0x%x", ldcp->id, rx_tail, rv); 670 } 671 672 673 /* 674 * Reset LDC state structure and its contents 675 */ 676 static void 677 i_ldc_reset_state(ldc_chan_t *ldcp) 678 { 679 ASSERT(MUTEX_HELD(&ldcp->lock)); 680 ldcp->last_msg_snt = LDC_INIT_SEQID; 681 ldcp->last_ack_rcd = 0; 682 ldcp->last_msg_rcd = 0; 683 ldcp->tx_ackd_head = ldcp->tx_head; 684 ldcp->stream_remains = 0; 685 ldcp->next_vidx = 0; 686 ldcp->hstate = 0; 687 ldcp->tstate = TS_OPEN; 688 ldcp->status = LDC_OPEN; 689 ldcp->rx_ack_head = ACKPEEK_HEAD_INVALID; 690 ldcp->rx_dq_head = 0; 691 ldcp->rx_dq_tail = 0; 692 693 if (ldcp->link_state == LDC_CHANNEL_UP || 694 ldcp->link_state == LDC_CHANNEL_RESET) { 695 696 if (ldcp->mode == LDC_MODE_RAW) { 697 ldcp->status = LDC_UP; 698 ldcp->tstate = TS_UP; 699 } else { 700 ldcp->status = LDC_READY; 701 ldcp->tstate |= TS_LINK_READY; 702 } 703 } 704 } 705 706 /* 707 * Reset a LDC channel 708 */ 709 void 710 i_ldc_reset(ldc_chan_t *ldcp, boolean_t force_reset) 711 { 712 DWARN(ldcp->id, "i_ldc_reset: (0x%llx) channel reset\n", ldcp->id); 713 714 ASSERT(MUTEX_HELD(&ldcp->lock)); 715 ASSERT(MUTEX_HELD(&ldcp->tx_lock)); 716 717 /* reconfig Tx and Rx queues */ 718 (void) i_ldc_txq_reconf(ldcp); 719 (void) i_ldc_rxq_reconf(ldcp, force_reset); 720 721 /* Clear Tx and Rx interrupts */ 722 (void) i_ldc_clear_intr(ldcp, CNEX_TX_INTR); 723 (void) i_ldc_clear_intr(ldcp, CNEX_RX_INTR); 724 725 /* Reset channel state */ 726 i_ldc_reset_state(ldcp); 727 728 /* Mark channel in reset */ 729 ldcp->tstate |= TS_IN_RESET; 730 } 731 732 /* 733 * Walk the channel list and reset channels if they are of the right 734 * devclass and their Rx queues have been configured. No locks are 735 * taken because the function is only invoked by the kernel just before 736 * entering the prom or debugger when the system is single-threaded. 737 */ 738 static void 739 i_ldc_debug_enter(void) 740 { 741 ldc_chan_t *ldcp; 742 743 ldcp = ldcssp->chan_list; 744 while (ldcp != NULL) { 745 if (((ldcp->tstate & TS_QCONF_RDY) == TS_QCONF_RDY) && 746 (LDC_DEVCLASS_PROM_RESET(ldcp->devclass) != 0)) { 747 (void) hv_ldc_rx_qconf(ldcp->id, ldcp->rx_q_ra, 748 ldcp->rx_q_entries); 749 } 750 ldcp = ldcp->next; 751 } 752 } 753 754 /* 755 * Clear pending interrupts 756 */ 757 static void 758 i_ldc_clear_intr(ldc_chan_t *ldcp, cnex_intrtype_t itype) 759 { 760 ldc_cnex_t *cinfo = &ldcssp->cinfo; 761 762 ASSERT(MUTEX_HELD(&ldcp->lock)); 763 ASSERT(cinfo->dip != NULL); 764 765 switch (itype) { 766 case CNEX_TX_INTR: 767 /* check Tx interrupt */ 768 if (ldcp->tx_intr_state) 769 ldcp->tx_intr_state = LDC_INTR_NONE; 770 else 771 return; 772 break; 773 774 case CNEX_RX_INTR: 775 /* check Rx interrupt */ 776 if (ldcp->rx_intr_state) 777 ldcp->rx_intr_state = LDC_INTR_NONE; 778 else 779 return; 780 break; 781 } 782 783 (void) cinfo->clr_intr(cinfo->dip, ldcp->id, itype); 784 D2(ldcp->id, 785 "i_ldc_clear_intr: (0x%llx) cleared 0x%x intr\n", 786 ldcp->id, itype); 787 } 788 789 /* 790 * Set the receive queue head 791 * Resets connection and returns an error if it fails. 792 */ 793 static int 794 i_ldc_set_rx_head(ldc_chan_t *ldcp, uint64_t head) 795 { 796 int rv; 797 int retries; 798 799 ASSERT(MUTEX_HELD(&ldcp->lock)); 800 for (retries = 0; retries < ldc_max_retries; retries++) { 801 802 if ((rv = hv_ldc_rx_set_qhead(ldcp->id, head)) == 0) 803 return (0); 804 805 if (rv != H_EWOULDBLOCK) 806 break; 807 808 /* wait for ldc_delay usecs */ 809 drv_usecwait(ldc_delay); 810 } 811 812 cmn_err(CE_WARN, "ldc_set_rx_qhead: (0x%lx) cannot set qhead 0x%lx, " 813 "rv = 0x%x", ldcp->id, head, rv); 814 mutex_enter(&ldcp->tx_lock); 815 i_ldc_reset(ldcp, B_TRUE); 816 mutex_exit(&ldcp->tx_lock); 817 818 return (ECONNRESET); 819 } 820 821 /* 822 * Returns the tx_head to be used for transfer 823 */ 824 static void 825 i_ldc_get_tx_head(ldc_chan_t *ldcp, uint64_t *head) 826 { 827 ldc_msg_t *pkt; 828 829 ASSERT(MUTEX_HELD(&ldcp->tx_lock)); 830 831 /* get current Tx head */ 832 *head = ldcp->tx_head; 833 834 /* 835 * Reliable mode will use the ACKd head instead of the regular tx_head. 836 * Also in Reliable mode, advance ackd_head for all non DATA/INFO pkts, 837 * up to the current location of tx_head. This needs to be done 838 * as the peer will only ACK DATA/INFO pkts. 839 */ 840 if (ldcp->mode == LDC_MODE_RELIABLE) { 841 while (ldcp->tx_ackd_head != ldcp->tx_head) { 842 pkt = (ldc_msg_t *)(ldcp->tx_q_va + ldcp->tx_ackd_head); 843 if ((pkt->type & LDC_DATA) && (pkt->stype & LDC_INFO)) { 844 break; 845 } 846 /* advance ACKd head */ 847 ldcp->tx_ackd_head = 848 (ldcp->tx_ackd_head + LDC_PACKET_SIZE) % 849 (ldcp->tx_q_entries << LDC_PACKET_SHIFT); 850 } 851 *head = ldcp->tx_ackd_head; 852 } 853 } 854 855 /* 856 * Returns the tx_tail to be used for transfer 857 * Re-reads the TX queue ptrs if and only if the 858 * the cached head and tail are equal (queue is full) 859 */ 860 static int 861 i_ldc_get_tx_tail(ldc_chan_t *ldcp, uint64_t *tail) 862 { 863 int rv; 864 uint64_t current_head, new_tail; 865 866 ASSERT(MUTEX_HELD(&ldcp->tx_lock)); 867 /* Read the head and tail ptrs from HV */ 868 rv = hv_ldc_tx_get_state(ldcp->id, 869 &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state); 870 if (rv) { 871 cmn_err(CE_WARN, 872 "i_ldc_get_tx_tail: (0x%lx) cannot read qptrs\n", 873 ldcp->id); 874 return (EIO); 875 } 876 if (ldcp->link_state == LDC_CHANNEL_DOWN) { 877 D1(ldcp->id, "i_ldc_get_tx_tail: (0x%llx) channel not ready\n", 878 ldcp->id); 879 return (ECONNRESET); 880 } 881 882 i_ldc_get_tx_head(ldcp, ¤t_head); 883 884 /* increment the tail */ 885 new_tail = (ldcp->tx_tail + LDC_PACKET_SIZE) % 886 (ldcp->tx_q_entries << LDC_PACKET_SHIFT); 887 888 if (new_tail == current_head) { 889 DWARN(ldcp->id, 890 "i_ldc_get_tx_tail: (0x%llx) TX queue is full\n", 891 ldcp->id); 892 return (EWOULDBLOCK); 893 } 894 895 D2(ldcp->id, "i_ldc_get_tx_tail: (0x%llx) head=0x%llx, tail=0x%llx\n", 896 ldcp->id, ldcp->tx_head, ldcp->tx_tail); 897 898 *tail = ldcp->tx_tail; 899 return (0); 900 } 901 902 /* 903 * Set the tail pointer. If HV returns EWOULDBLOCK, it will back off 904 * and retry ldc_max_retries times before returning an error. 905 * Returns 0, EWOULDBLOCK or EIO 906 */ 907 static int 908 i_ldc_set_tx_tail(ldc_chan_t *ldcp, uint64_t tail) 909 { 910 int rv, retval = EWOULDBLOCK; 911 int retries; 912 913 ASSERT(MUTEX_HELD(&ldcp->tx_lock)); 914 for (retries = 0; retries < ldc_max_retries; retries++) { 915 916 if ((rv = hv_ldc_tx_set_qtail(ldcp->id, tail)) == 0) { 917 retval = 0; 918 break; 919 } 920 if (rv != H_EWOULDBLOCK) { 921 DWARN(ldcp->id, "i_ldc_set_tx_tail: (0x%llx) set " 922 "qtail=0x%llx failed, rv=%d\n", ldcp->id, tail, rv); 923 retval = EIO; 924 break; 925 } 926 927 /* wait for ldc_delay usecs */ 928 drv_usecwait(ldc_delay); 929 } 930 return (retval); 931 } 932 933 /* 934 * Copy a data packet from the HV receive queue to the data queue. 935 * Caller must ensure that the data queue is not already full. 936 * 937 * The *head argument represents the current head pointer for the HV 938 * receive queue. After copying a packet from the HV receive queue, 939 * the *head pointer will be updated. This allows the caller to update 940 * the head pointer in HV using the returned *head value. 941 */ 942 void 943 i_ldc_rxdq_copy(ldc_chan_t *ldcp, uint64_t *head) 944 { 945 uint64_t q_size, dq_size; 946 947 ASSERT(MUTEX_HELD(&ldcp->lock)); 948 949 q_size = ldcp->rx_q_entries << LDC_PACKET_SHIFT; 950 dq_size = ldcp->rx_dq_entries << LDC_PACKET_SHIFT; 951 952 ASSERT(Q_CONTIG_SPACE(ldcp->rx_dq_head, ldcp->rx_dq_tail, 953 dq_size) >= LDC_PACKET_SIZE); 954 955 bcopy((void *)(ldcp->rx_q_va + *head), 956 (void *)(ldcp->rx_dq_va + ldcp->rx_dq_tail), LDC_PACKET_SIZE); 957 TRACE_RXDQ_COPY(ldcp, LDC_PACKET_SIZE); 958 959 /* Update rx head */ 960 *head = (*head + LDC_PACKET_SIZE) % q_size; 961 962 /* Update dq tail */ 963 ldcp->rx_dq_tail = (ldcp->rx_dq_tail + LDC_PACKET_SIZE) % dq_size; 964 } 965 966 /* 967 * Update the Rx data queue head pointer 968 */ 969 static int 970 i_ldc_set_rxdq_head(ldc_chan_t *ldcp, uint64_t head) 971 { 972 ldcp->rx_dq_head = head; 973 return (0); 974 } 975 976 /* 977 * Get the Rx data queue head and tail pointers 978 */ 979 static uint64_t 980 i_ldc_dq_rx_get_state(ldc_chan_t *ldcp, uint64_t *head, uint64_t *tail, 981 uint64_t *link_state) 982 { 983 _NOTE(ARGUNUSED(link_state)) 984 *head = ldcp->rx_dq_head; 985 *tail = ldcp->rx_dq_tail; 986 return (0); 987 } 988 989 /* 990 * Wrapper for the Rx HV queue set head function. Giving the 991 * data queue and HV queue set head functions the same type. 992 */ 993 static uint64_t 994 i_ldc_hvq_rx_get_state(ldc_chan_t *ldcp, uint64_t *head, uint64_t *tail, 995 uint64_t *link_state) 996 { 997 return (i_ldc_h2v_error(hv_ldc_rx_get_state(ldcp->id, head, tail, 998 link_state))); 999 } 1000 1001 /* 1002 * LDC receive interrupt handler 1003 * triggered for channel with data pending to read 1004 * i.e. Rx queue content changes 1005 */ 1006 static uint_t 1007 i_ldc_rx_hdlr(caddr_t arg1, caddr_t arg2) 1008 { 1009 _NOTE(ARGUNUSED(arg2)) 1010 1011 ldc_chan_t *ldcp; 1012 boolean_t notify; 1013 uint64_t event; 1014 int rv, status; 1015 1016 /* Get the channel for which interrupt was received */ 1017 if (arg1 == NULL) { 1018 cmn_err(CE_WARN, "i_ldc_rx_hdlr: invalid arg\n"); 1019 return (DDI_INTR_UNCLAIMED); 1020 } 1021 1022 ldcp = (ldc_chan_t *)arg1; 1023 1024 D1(ldcp->id, "i_ldc_rx_hdlr: (0x%llx) Received intr, ldcp=0x%p\n", 1025 ldcp->id, ldcp); 1026 D1(ldcp->id, "i_ldc_rx_hdlr: (%llx) USR%lx/TS%lx/HS%lx, LSTATE=%lx\n", 1027 ldcp->id, ldcp->status, ldcp->tstate, ldcp->hstate, 1028 ldcp->link_state); 1029 1030 /* Lock channel */ 1031 mutex_enter(&ldcp->lock); 1032 1033 /* Mark the interrupt as being actively handled */ 1034 ldcp->rx_intr_state = LDC_INTR_ACTIVE; 1035 1036 status = i_ldc_rx_process_hvq(ldcp, ¬ify, &event); 1037 1038 if (ldcp->mode != LDC_MODE_RELIABLE) { 1039 /* 1040 * If there are no data packets on the queue, clear 1041 * the interrupt. Otherwise, the ldc_read will clear 1042 * interrupts after draining the queue. To indicate the 1043 * interrupt has not yet been cleared, it is marked 1044 * as pending. 1045 */ 1046 if ((event & LDC_EVT_READ) == 0) { 1047 i_ldc_clear_intr(ldcp, CNEX_RX_INTR); 1048 } else { 1049 ldcp->rx_intr_state = LDC_INTR_PEND; 1050 } 1051 } 1052 1053 /* if callbacks are disabled, do not notify */ 1054 if (notify && ldcp->cb_enabled) { 1055 ldcp->cb_inprogress = B_TRUE; 1056 mutex_exit(&ldcp->lock); 1057 rv = ldcp->cb(event, ldcp->cb_arg); 1058 if (rv) { 1059 DWARN(ldcp->id, 1060 "i_ldc_rx_hdlr: (0x%llx) callback failure", 1061 ldcp->id); 1062 } 1063 mutex_enter(&ldcp->lock); 1064 ldcp->cb_inprogress = B_FALSE; 1065 } 1066 1067 if (ldcp->mode == LDC_MODE_RELIABLE) { 1068 if (status == ENOSPC) { 1069 /* 1070 * Here, ENOSPC indicates the secondary data 1071 * queue is full and the Rx queue is non-empty. 1072 * Much like how reliable and raw modes are 1073 * handled above, since the Rx queue is non- 1074 * empty, we mark the interrupt as pending to 1075 * indicate it has not yet been cleared. 1076 */ 1077 ldcp->rx_intr_state = LDC_INTR_PEND; 1078 } else { 1079 /* 1080 * We have processed all CTRL packets and 1081 * copied all DATA packets to the secondary 1082 * queue. Clear the interrupt. 1083 */ 1084 i_ldc_clear_intr(ldcp, CNEX_RX_INTR); 1085 } 1086 } 1087 1088 mutex_exit(&ldcp->lock); 1089 1090 D1(ldcp->id, "i_ldc_rx_hdlr: (0x%llx) exiting handler", ldcp->id); 1091 1092 return (DDI_INTR_CLAIMED); 1093 } 1094 1095 /* 1096 * Wrapper for the Rx HV queue processing function to be used when 1097 * checking the Rx HV queue for data packets. Unlike the interrupt 1098 * handler code flow, the Rx interrupt is not cleared here and 1099 * callbacks are not made. 1100 */ 1101 static uint_t 1102 i_ldc_chkq(ldc_chan_t *ldcp) 1103 { 1104 boolean_t notify; 1105 uint64_t event; 1106 1107 return (i_ldc_rx_process_hvq(ldcp, ¬ify, &event)); 1108 } 1109 1110 /* 1111 * Send a LDC message 1112 */ 1113 static int 1114 i_ldc_send_pkt(ldc_chan_t *ldcp, uint8_t pkttype, uint8_t subtype, 1115 uint8_t ctrlmsg) 1116 { 1117 int rv; 1118 ldc_msg_t *pkt; 1119 uint64_t tx_tail; 1120 uint32_t curr_seqid; 1121 1122 /* Obtain Tx lock */ 1123 mutex_enter(&ldcp->tx_lock); 1124 1125 curr_seqid = ldcp->last_msg_snt; 1126 1127 /* get the current tail for the message */ 1128 rv = i_ldc_get_tx_tail(ldcp, &tx_tail); 1129 if (rv) { 1130 DWARN(ldcp->id, 1131 "i_ldc_send_pkt: (0x%llx) error sending pkt, " 1132 "type=0x%x,subtype=0x%x,ctrl=0x%x\n", 1133 ldcp->id, pkttype, subtype, ctrlmsg); 1134 mutex_exit(&ldcp->tx_lock); 1135 return (rv); 1136 } 1137 1138 pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail); 1139 ZERO_PKT(pkt); 1140 1141 /* Initialize the packet */ 1142 pkt->type = pkttype; 1143 pkt->stype = subtype; 1144 pkt->ctrl = ctrlmsg; 1145 1146 /* Store ackid/seqid iff it is RELIABLE mode & not a RTS/RTR message */ 1147 if (((ctrlmsg & LDC_CTRL_MASK) != LDC_RTS) && 1148 ((ctrlmsg & LDC_CTRL_MASK) != LDC_RTR)) { 1149 curr_seqid++; 1150 if (ldcp->mode != LDC_MODE_RAW) { 1151 pkt->seqid = curr_seqid; 1152 pkt->ackid = ldcp->last_msg_rcd; 1153 } 1154 } 1155 DUMP_LDC_PKT(ldcp, "i_ldc_send_pkt", (uint64_t)pkt); 1156 1157 /* initiate the send by calling into HV and set the new tail */ 1158 tx_tail = (tx_tail + LDC_PACKET_SIZE) % 1159 (ldcp->tx_q_entries << LDC_PACKET_SHIFT); 1160 1161 rv = i_ldc_set_tx_tail(ldcp, tx_tail); 1162 if (rv) { 1163 DWARN(ldcp->id, 1164 "i_ldc_send_pkt:(0x%llx) error sending pkt, " 1165 "type=0x%x,stype=0x%x,ctrl=0x%x\n", 1166 ldcp->id, pkttype, subtype, ctrlmsg); 1167 mutex_exit(&ldcp->tx_lock); 1168 return (EIO); 1169 } 1170 1171 ldcp->last_msg_snt = curr_seqid; 1172 ldcp->tx_tail = tx_tail; 1173 1174 mutex_exit(&ldcp->tx_lock); 1175 return (0); 1176 } 1177 1178 /* 1179 * Checks if packet was received in right order 1180 * in the case of a reliable link. 1181 * Returns 0 if in order, else EIO 1182 */ 1183 static int 1184 i_ldc_check_seqid(ldc_chan_t *ldcp, ldc_msg_t *msg) 1185 { 1186 /* No seqid checking for RAW mode */ 1187 if (ldcp->mode == LDC_MODE_RAW) 1188 return (0); 1189 1190 /* No seqid checking for version, RTS, RTR message */ 1191 if (msg->ctrl == LDC_VER || 1192 msg->ctrl == LDC_RTS || 1193 msg->ctrl == LDC_RTR) 1194 return (0); 1195 1196 /* Initial seqid to use is sent in RTS/RTR and saved in last_msg_rcd */ 1197 if (msg->seqid != (ldcp->last_msg_rcd + 1)) { 1198 DWARN(ldcp->id, 1199 "i_ldc_check_seqid: (0x%llx) out-of-order pkt, got 0x%x, " 1200 "expecting 0x%x\n", ldcp->id, msg->seqid, 1201 (ldcp->last_msg_rcd + 1)); 1202 return (EIO); 1203 } 1204 1205 #ifdef DEBUG 1206 if (LDC_INJECT_PKTLOSS(ldcp)) { 1207 DWARN(ldcp->id, 1208 "i_ldc_check_seqid: (0x%llx) inject pkt loss\n", ldcp->id); 1209 return (EIO); 1210 } 1211 #endif 1212 1213 return (0); 1214 } 1215 1216 1217 /* 1218 * Process an incoming version ctrl message 1219 */ 1220 static int 1221 i_ldc_process_VER(ldc_chan_t *ldcp, ldc_msg_t *msg) 1222 { 1223 int rv = 0, idx = ldcp->next_vidx; 1224 ldc_msg_t *pkt; 1225 uint64_t tx_tail; 1226 ldc_ver_t *rcvd_ver; 1227 1228 /* get the received version */ 1229 rcvd_ver = (ldc_ver_t *)((uint64_t)msg + LDC_PAYLOAD_VER_OFF); 1230 1231 D2(ldcp->id, "i_ldc_process_VER: (0x%llx) received VER v%u.%u\n", 1232 ldcp->id, rcvd_ver->major, rcvd_ver->minor); 1233 1234 /* Obtain Tx lock */ 1235 mutex_enter(&ldcp->tx_lock); 1236 1237 switch (msg->stype) { 1238 case LDC_INFO: 1239 1240 if ((ldcp->tstate & ~TS_IN_RESET) == TS_VREADY) { 1241 (void) i_ldc_txq_reconf(ldcp); 1242 i_ldc_reset_state(ldcp); 1243 mutex_exit(&ldcp->tx_lock); 1244 return (EAGAIN); 1245 } 1246 1247 /* get the current tail and pkt for the response */ 1248 rv = i_ldc_get_tx_tail(ldcp, &tx_tail); 1249 if (rv != 0) { 1250 DWARN(ldcp->id, 1251 "i_ldc_process_VER: (0x%llx) err sending " 1252 "version ACK/NACK\n", ldcp->id); 1253 i_ldc_reset(ldcp, B_TRUE); 1254 mutex_exit(&ldcp->tx_lock); 1255 return (ECONNRESET); 1256 } 1257 1258 pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail); 1259 ZERO_PKT(pkt); 1260 1261 /* initialize the packet */ 1262 pkt->type = LDC_CTRL; 1263 pkt->ctrl = LDC_VER; 1264 1265 for (;;) { 1266 1267 D1(ldcp->id, "i_ldc_process_VER: got %u.%u chk %u.%u\n", 1268 rcvd_ver->major, rcvd_ver->minor, 1269 ldc_versions[idx].major, ldc_versions[idx].minor); 1270 1271 if (rcvd_ver->major == ldc_versions[idx].major) { 1272 /* major version match - ACK version */ 1273 pkt->stype = LDC_ACK; 1274 1275 /* 1276 * lower minor version to the one this endpt 1277 * supports, if necessary 1278 */ 1279 if (rcvd_ver->minor > ldc_versions[idx].minor) 1280 rcvd_ver->minor = 1281 ldc_versions[idx].minor; 1282 bcopy(rcvd_ver, pkt->udata, sizeof (*rcvd_ver)); 1283 1284 break; 1285 } 1286 1287 if (rcvd_ver->major > ldc_versions[idx].major) { 1288 1289 D1(ldcp->id, "i_ldc_process_VER: using next" 1290 " lower idx=%d, v%u.%u\n", idx, 1291 ldc_versions[idx].major, 1292 ldc_versions[idx].minor); 1293 1294 /* nack with next lower version */ 1295 pkt->stype = LDC_NACK; 1296 bcopy(&ldc_versions[idx], pkt->udata, 1297 sizeof (ldc_versions[idx])); 1298 ldcp->next_vidx = idx; 1299 break; 1300 } 1301 1302 /* next major version */ 1303 idx++; 1304 1305 D1(ldcp->id, "i_ldc_process_VER: inc idx %x\n", idx); 1306 1307 if (idx == LDC_NUM_VERS) { 1308 /* no version match - send NACK */ 1309 pkt->stype = LDC_NACK; 1310 bzero(pkt->udata, sizeof (ldc_ver_t)); 1311 ldcp->next_vidx = 0; 1312 break; 1313 } 1314 } 1315 1316 /* initiate the send by calling into HV and set the new tail */ 1317 tx_tail = (tx_tail + LDC_PACKET_SIZE) % 1318 (ldcp->tx_q_entries << LDC_PACKET_SHIFT); 1319 1320 rv = i_ldc_set_tx_tail(ldcp, tx_tail); 1321 if (rv == 0) { 1322 ldcp->tx_tail = tx_tail; 1323 if (pkt->stype == LDC_ACK) { 1324 D2(ldcp->id, "i_ldc_process_VER: (0x%llx) sent" 1325 " version ACK\n", ldcp->id); 1326 /* Save the ACK'd version */ 1327 ldcp->version.major = rcvd_ver->major; 1328 ldcp->version.minor = rcvd_ver->minor; 1329 ldcp->hstate |= TS_RCVD_VER; 1330 ldcp->tstate |= TS_VER_DONE; 1331 D1(DBG_ALL_LDCS, 1332 "(0x%llx) Sent ACK, " 1333 "Agreed on version v%u.%u\n", 1334 ldcp->id, rcvd_ver->major, rcvd_ver->minor); 1335 } 1336 } else { 1337 DWARN(ldcp->id, 1338 "i_ldc_process_VER: (0x%llx) error sending " 1339 "ACK/NACK\n", ldcp->id); 1340 i_ldc_reset(ldcp, B_TRUE); 1341 mutex_exit(&ldcp->tx_lock); 1342 return (ECONNRESET); 1343 } 1344 1345 break; 1346 1347 case LDC_ACK: 1348 if ((ldcp->tstate & ~TS_IN_RESET) == TS_VREADY) { 1349 if (ldcp->version.major != rcvd_ver->major || 1350 ldcp->version.minor != rcvd_ver->minor) { 1351 1352 /* mismatched version - reset connection */ 1353 DWARN(ldcp->id, 1354 "i_ldc_process_VER: (0x%llx) recvd" 1355 " ACK ver != sent ACK ver\n", ldcp->id); 1356 i_ldc_reset(ldcp, B_TRUE); 1357 mutex_exit(&ldcp->tx_lock); 1358 return (ECONNRESET); 1359 } 1360 } else { 1361 /* SUCCESS - we have agreed on a version */ 1362 ldcp->version.major = rcvd_ver->major; 1363 ldcp->version.minor = rcvd_ver->minor; 1364 ldcp->tstate |= TS_VER_DONE; 1365 } 1366 1367 D1(ldcp->id, "(0x%llx) Got ACK, Agreed on version v%u.%u\n", 1368 ldcp->id, rcvd_ver->major, rcvd_ver->minor); 1369 1370 /* initiate RTS-RTR-RDX handshake */ 1371 rv = i_ldc_get_tx_tail(ldcp, &tx_tail); 1372 if (rv) { 1373 DWARN(ldcp->id, 1374 "i_ldc_process_VER: (0x%llx) cannot send RTS\n", 1375 ldcp->id); 1376 i_ldc_reset(ldcp, B_TRUE); 1377 mutex_exit(&ldcp->tx_lock); 1378 return (ECONNRESET); 1379 } 1380 1381 pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail); 1382 ZERO_PKT(pkt); 1383 1384 pkt->type = LDC_CTRL; 1385 pkt->stype = LDC_INFO; 1386 pkt->ctrl = LDC_RTS; 1387 pkt->env = ldcp->mode; 1388 if (ldcp->mode != LDC_MODE_RAW) 1389 pkt->seqid = LDC_INIT_SEQID; 1390 1391 ldcp->last_msg_rcd = LDC_INIT_SEQID; 1392 1393 DUMP_LDC_PKT(ldcp, "i_ldc_process_VER snd rts", (uint64_t)pkt); 1394 1395 /* initiate the send by calling into HV and set the new tail */ 1396 tx_tail = (tx_tail + LDC_PACKET_SIZE) % 1397 (ldcp->tx_q_entries << LDC_PACKET_SHIFT); 1398 1399 rv = i_ldc_set_tx_tail(ldcp, tx_tail); 1400 if (rv) { 1401 D2(ldcp->id, 1402 "i_ldc_process_VER: (0x%llx) no listener\n", 1403 ldcp->id); 1404 i_ldc_reset(ldcp, B_TRUE); 1405 mutex_exit(&ldcp->tx_lock); 1406 return (ECONNRESET); 1407 } 1408 1409 ldcp->tx_tail = tx_tail; 1410 ldcp->hstate |= TS_SENT_RTS; 1411 1412 break; 1413 1414 case LDC_NACK: 1415 /* check if version in NACK is zero */ 1416 if (rcvd_ver->major == 0 && rcvd_ver->minor == 0) { 1417 /* version handshake failure */ 1418 DWARN(DBG_ALL_LDCS, 1419 "i_ldc_process_VER: (0x%llx) no version match\n", 1420 ldcp->id); 1421 i_ldc_reset(ldcp, B_TRUE); 1422 mutex_exit(&ldcp->tx_lock); 1423 return (ECONNRESET); 1424 } 1425 1426 /* get the current tail and pkt for the response */ 1427 rv = i_ldc_get_tx_tail(ldcp, &tx_tail); 1428 if (rv != 0) { 1429 cmn_err(CE_NOTE, 1430 "i_ldc_process_VER: (0x%lx) err sending " 1431 "version ACK/NACK\n", ldcp->id); 1432 i_ldc_reset(ldcp, B_TRUE); 1433 mutex_exit(&ldcp->tx_lock); 1434 return (ECONNRESET); 1435 } 1436 1437 pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail); 1438 ZERO_PKT(pkt); 1439 1440 /* initialize the packet */ 1441 pkt->type = LDC_CTRL; 1442 pkt->ctrl = LDC_VER; 1443 pkt->stype = LDC_INFO; 1444 1445 /* check ver in NACK msg has a match */ 1446 for (;;) { 1447 if (rcvd_ver->major == ldc_versions[idx].major) { 1448 /* 1449 * major version match - resubmit request 1450 * if lower minor version to the one this endpt 1451 * supports, if necessary 1452 */ 1453 if (rcvd_ver->minor > ldc_versions[idx].minor) 1454 rcvd_ver->minor = 1455 ldc_versions[idx].minor; 1456 bcopy(rcvd_ver, pkt->udata, sizeof (*rcvd_ver)); 1457 break; 1458 } 1459 1460 if (rcvd_ver->major > ldc_versions[idx].major) { 1461 1462 D1(ldcp->id, "i_ldc_process_VER: using next" 1463 " lower idx=%d, v%u.%u\n", idx, 1464 ldc_versions[idx].major, 1465 ldc_versions[idx].minor); 1466 1467 /* send next lower version */ 1468 bcopy(&ldc_versions[idx], pkt->udata, 1469 sizeof (ldc_versions[idx])); 1470 ldcp->next_vidx = idx; 1471 break; 1472 } 1473 1474 /* next version */ 1475 idx++; 1476 1477 D1(ldcp->id, "i_ldc_process_VER: inc idx %x\n", idx); 1478 1479 if (idx == LDC_NUM_VERS) { 1480 /* no version match - terminate */ 1481 ldcp->next_vidx = 0; 1482 mutex_exit(&ldcp->tx_lock); 1483 return (ECONNRESET); 1484 } 1485 } 1486 1487 /* initiate the send by calling into HV and set the new tail */ 1488 tx_tail = (tx_tail + LDC_PACKET_SIZE) % 1489 (ldcp->tx_q_entries << LDC_PACKET_SHIFT); 1490 1491 rv = i_ldc_set_tx_tail(ldcp, tx_tail); 1492 if (rv == 0) { 1493 D2(ldcp->id, "i_ldc_process_VER: (0x%llx) sent version" 1494 "INFO v%u.%u\n", ldcp->id, ldc_versions[idx].major, 1495 ldc_versions[idx].minor); 1496 ldcp->tx_tail = tx_tail; 1497 } else { 1498 cmn_err(CE_NOTE, 1499 "i_ldc_process_VER: (0x%lx) error sending version" 1500 "INFO\n", ldcp->id); 1501 i_ldc_reset(ldcp, B_TRUE); 1502 mutex_exit(&ldcp->tx_lock); 1503 return (ECONNRESET); 1504 } 1505 1506 break; 1507 } 1508 1509 mutex_exit(&ldcp->tx_lock); 1510 return (rv); 1511 } 1512 1513 1514 /* 1515 * Process an incoming RTS ctrl message 1516 */ 1517 static int 1518 i_ldc_process_RTS(ldc_chan_t *ldcp, ldc_msg_t *msg) 1519 { 1520 int rv = 0; 1521 ldc_msg_t *pkt; 1522 uint64_t tx_tail; 1523 boolean_t sent_NACK = B_FALSE; 1524 1525 D2(ldcp->id, "i_ldc_process_RTS: (0x%llx) received RTS\n", ldcp->id); 1526 1527 switch (msg->stype) { 1528 case LDC_NACK: 1529 DWARN(ldcp->id, 1530 "i_ldc_process_RTS: (0x%llx) RTS NACK received\n", 1531 ldcp->id); 1532 1533 /* Reset the channel -- as we cannot continue */ 1534 mutex_enter(&ldcp->tx_lock); 1535 i_ldc_reset(ldcp, B_TRUE); 1536 mutex_exit(&ldcp->tx_lock); 1537 rv = ECONNRESET; 1538 break; 1539 1540 case LDC_INFO: 1541 1542 /* check mode */ 1543 if (ldcp->mode != (ldc_mode_t)msg->env) { 1544 cmn_err(CE_NOTE, 1545 "i_ldc_process_RTS: (0x%lx) mode mismatch\n", 1546 ldcp->id); 1547 /* 1548 * send NACK in response to MODE message 1549 * get the current tail for the response 1550 */ 1551 rv = i_ldc_send_pkt(ldcp, LDC_CTRL, LDC_NACK, LDC_RTS); 1552 if (rv) { 1553 /* if cannot send NACK - reset channel */ 1554 mutex_enter(&ldcp->tx_lock); 1555 i_ldc_reset(ldcp, B_TRUE); 1556 mutex_exit(&ldcp->tx_lock); 1557 rv = ECONNRESET; 1558 break; 1559 } 1560 sent_NACK = B_TRUE; 1561 } 1562 break; 1563 default: 1564 DWARN(ldcp->id, "i_ldc_process_RTS: (0x%llx) unexp ACK\n", 1565 ldcp->id); 1566 mutex_enter(&ldcp->tx_lock); 1567 i_ldc_reset(ldcp, B_TRUE); 1568 mutex_exit(&ldcp->tx_lock); 1569 rv = ECONNRESET; 1570 break; 1571 } 1572 1573 /* 1574 * If either the connection was reset (when rv != 0) or 1575 * a NACK was sent, we return. In the case of a NACK 1576 * we dont want to consume the packet that came in but 1577 * not record that we received the RTS 1578 */ 1579 if (rv || sent_NACK) 1580 return (rv); 1581 1582 /* record RTS received */ 1583 ldcp->hstate |= TS_RCVD_RTS; 1584 1585 /* store initial SEQID info */ 1586 ldcp->last_msg_snt = msg->seqid; 1587 1588 /* Obtain Tx lock */ 1589 mutex_enter(&ldcp->tx_lock); 1590 1591 /* get the current tail for the response */ 1592 rv = i_ldc_get_tx_tail(ldcp, &tx_tail); 1593 if (rv != 0) { 1594 cmn_err(CE_NOTE, 1595 "i_ldc_process_RTS: (0x%lx) err sending RTR\n", 1596 ldcp->id); 1597 i_ldc_reset(ldcp, B_TRUE); 1598 mutex_exit(&ldcp->tx_lock); 1599 return (ECONNRESET); 1600 } 1601 1602 pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail); 1603 ZERO_PKT(pkt); 1604 1605 /* initialize the packet */ 1606 pkt->type = LDC_CTRL; 1607 pkt->stype = LDC_INFO; 1608 pkt->ctrl = LDC_RTR; 1609 pkt->env = ldcp->mode; 1610 if (ldcp->mode != LDC_MODE_RAW) 1611 pkt->seqid = LDC_INIT_SEQID; 1612 1613 ldcp->last_msg_rcd = msg->seqid; 1614 1615 /* initiate the send by calling into HV and set the new tail */ 1616 tx_tail = (tx_tail + LDC_PACKET_SIZE) % 1617 (ldcp->tx_q_entries << LDC_PACKET_SHIFT); 1618 1619 rv = i_ldc_set_tx_tail(ldcp, tx_tail); 1620 if (rv == 0) { 1621 D2(ldcp->id, 1622 "i_ldc_process_RTS: (0x%llx) sent RTR\n", ldcp->id); 1623 DUMP_LDC_PKT(ldcp, "i_ldc_process_RTS sent rtr", (uint64_t)pkt); 1624 1625 ldcp->tx_tail = tx_tail; 1626 ldcp->hstate |= TS_SENT_RTR; 1627 1628 } else { 1629 cmn_err(CE_NOTE, 1630 "i_ldc_process_RTS: (0x%lx) error sending RTR\n", 1631 ldcp->id); 1632 i_ldc_reset(ldcp, B_TRUE); 1633 mutex_exit(&ldcp->tx_lock); 1634 return (ECONNRESET); 1635 } 1636 1637 mutex_exit(&ldcp->tx_lock); 1638 return (0); 1639 } 1640 1641 /* 1642 * Process an incoming RTR ctrl message 1643 */ 1644 static int 1645 i_ldc_process_RTR(ldc_chan_t *ldcp, ldc_msg_t *msg) 1646 { 1647 int rv = 0; 1648 boolean_t sent_NACK = B_FALSE; 1649 1650 D2(ldcp->id, "i_ldc_process_RTR: (0x%llx) received RTR\n", ldcp->id); 1651 1652 switch (msg->stype) { 1653 case LDC_NACK: 1654 /* RTR NACK received */ 1655 DWARN(ldcp->id, 1656 "i_ldc_process_RTR: (0x%llx) RTR NACK received\n", 1657 ldcp->id); 1658 1659 /* Reset the channel -- as we cannot continue */ 1660 mutex_enter(&ldcp->tx_lock); 1661 i_ldc_reset(ldcp, B_TRUE); 1662 mutex_exit(&ldcp->tx_lock); 1663 rv = ECONNRESET; 1664 1665 break; 1666 1667 case LDC_INFO: 1668 1669 /* check mode */ 1670 if (ldcp->mode != (ldc_mode_t)msg->env) { 1671 DWARN(ldcp->id, 1672 "i_ldc_process_RTR: (0x%llx) mode mismatch, " 1673 "expecting 0x%x, got 0x%x\n", 1674 ldcp->id, ldcp->mode, (ldc_mode_t)msg->env); 1675 /* 1676 * send NACK in response to MODE message 1677 * get the current tail for the response 1678 */ 1679 rv = i_ldc_send_pkt(ldcp, LDC_CTRL, LDC_NACK, LDC_RTR); 1680 if (rv) { 1681 /* if cannot send NACK - reset channel */ 1682 mutex_enter(&ldcp->tx_lock); 1683 i_ldc_reset(ldcp, B_TRUE); 1684 mutex_exit(&ldcp->tx_lock); 1685 rv = ECONNRESET; 1686 break; 1687 } 1688 sent_NACK = B_TRUE; 1689 } 1690 break; 1691 1692 default: 1693 DWARN(ldcp->id, "i_ldc_process_RTR: (0x%llx) unexp ACK\n", 1694 ldcp->id); 1695 1696 /* Reset the channel -- as we cannot continue */ 1697 mutex_enter(&ldcp->tx_lock); 1698 i_ldc_reset(ldcp, B_TRUE); 1699 mutex_exit(&ldcp->tx_lock); 1700 rv = ECONNRESET; 1701 break; 1702 } 1703 1704 /* 1705 * If either the connection was reset (when rv != 0) or 1706 * a NACK was sent, we return. In the case of a NACK 1707 * we dont want to consume the packet that came in but 1708 * not record that we received the RTR 1709 */ 1710 if (rv || sent_NACK) 1711 return (rv); 1712 1713 ldcp->last_msg_snt = msg->seqid; 1714 ldcp->hstate |= TS_RCVD_RTR; 1715 1716 rv = i_ldc_send_pkt(ldcp, LDC_CTRL, LDC_INFO, LDC_RDX); 1717 if (rv) { 1718 cmn_err(CE_NOTE, 1719 "i_ldc_process_RTR: (0x%lx) cannot send RDX\n", 1720 ldcp->id); 1721 mutex_enter(&ldcp->tx_lock); 1722 i_ldc_reset(ldcp, B_TRUE); 1723 mutex_exit(&ldcp->tx_lock); 1724 return (ECONNRESET); 1725 } 1726 D2(ldcp->id, 1727 "i_ldc_process_RTR: (0x%llx) sent RDX\n", ldcp->id); 1728 1729 ldcp->hstate |= TS_SENT_RDX; 1730 ldcp->tstate |= TS_HSHAKE_DONE; 1731 if ((ldcp->tstate & TS_IN_RESET) == 0) 1732 ldcp->status = LDC_UP; 1733 1734 D1(ldcp->id, "(0x%llx) Handshake Complete\n", ldcp->id); 1735 1736 return (0); 1737 } 1738 1739 1740 /* 1741 * Process an incoming RDX ctrl message 1742 */ 1743 static int 1744 i_ldc_process_RDX(ldc_chan_t *ldcp, ldc_msg_t *msg) 1745 { 1746 int rv = 0; 1747 1748 D2(ldcp->id, "i_ldc_process_RDX: (0x%llx) received RDX\n", ldcp->id); 1749 1750 switch (msg->stype) { 1751 case LDC_NACK: 1752 /* RDX NACK received */ 1753 DWARN(ldcp->id, 1754 "i_ldc_process_RDX: (0x%llx) RDX NACK received\n", 1755 ldcp->id); 1756 1757 /* Reset the channel -- as we cannot continue */ 1758 mutex_enter(&ldcp->tx_lock); 1759 i_ldc_reset(ldcp, B_TRUE); 1760 mutex_exit(&ldcp->tx_lock); 1761 rv = ECONNRESET; 1762 1763 break; 1764 1765 case LDC_INFO: 1766 1767 /* 1768 * if channel is UP and a RDX received after data transmission 1769 * has commenced it is an error 1770 */ 1771 if ((ldcp->tstate == TS_UP) && (ldcp->hstate & TS_RCVD_RDX)) { 1772 DWARN(DBG_ALL_LDCS, 1773 "i_ldc_process_RDX: (0x%llx) unexpected RDX" 1774 " - LDC reset\n", ldcp->id); 1775 mutex_enter(&ldcp->tx_lock); 1776 i_ldc_reset(ldcp, B_TRUE); 1777 mutex_exit(&ldcp->tx_lock); 1778 return (ECONNRESET); 1779 } 1780 1781 ldcp->hstate |= TS_RCVD_RDX; 1782 ldcp->tstate |= TS_HSHAKE_DONE; 1783 if ((ldcp->tstate & TS_IN_RESET) == 0) 1784 ldcp->status = LDC_UP; 1785 1786 D1(DBG_ALL_LDCS, "(0x%llx) Handshake Complete\n", ldcp->id); 1787 break; 1788 1789 default: 1790 DWARN(ldcp->id, "i_ldc_process_RDX: (0x%llx) unexp ACK\n", 1791 ldcp->id); 1792 1793 /* Reset the channel -- as we cannot continue */ 1794 mutex_enter(&ldcp->tx_lock); 1795 i_ldc_reset(ldcp, B_TRUE); 1796 mutex_exit(&ldcp->tx_lock); 1797 rv = ECONNRESET; 1798 break; 1799 } 1800 1801 return (rv); 1802 } 1803 1804 /* 1805 * Process an incoming ACK for a data packet 1806 */ 1807 static int 1808 i_ldc_process_data_ACK(ldc_chan_t *ldcp, ldc_msg_t *msg) 1809 { 1810 int rv; 1811 uint64_t tx_head; 1812 ldc_msg_t *pkt; 1813 1814 /* Obtain Tx lock */ 1815 mutex_enter(&ldcp->tx_lock); 1816 1817 /* 1818 * Read the current Tx head and tail 1819 */ 1820 rv = hv_ldc_tx_get_state(ldcp->id, 1821 &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state); 1822 if (rv != 0) { 1823 cmn_err(CE_WARN, 1824 "i_ldc_process_data_ACK: (0x%lx) cannot read qptrs\n", 1825 ldcp->id); 1826 1827 /* Reset the channel -- as we cannot continue */ 1828 i_ldc_reset(ldcp, B_TRUE); 1829 mutex_exit(&ldcp->tx_lock); 1830 return (ECONNRESET); 1831 } 1832 1833 /* 1834 * loop from where the previous ACK location was to the 1835 * current head location. This is how far the HV has 1836 * actually send pkts. Pkts between head and tail are 1837 * yet to be sent by HV. 1838 */ 1839 tx_head = ldcp->tx_ackd_head; 1840 for (;;) { 1841 pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_head); 1842 tx_head = (tx_head + LDC_PACKET_SIZE) % 1843 (ldcp->tx_q_entries << LDC_PACKET_SHIFT); 1844 1845 if (pkt->seqid == msg->ackid) { 1846 D2(ldcp->id, 1847 "i_ldc_process_data_ACK: (0x%llx) found packet\n", 1848 ldcp->id); 1849 ldcp->last_ack_rcd = msg->ackid; 1850 ldcp->tx_ackd_head = tx_head; 1851 break; 1852 } 1853 if (tx_head == ldcp->tx_head) { 1854 /* could not find packet */ 1855 DWARN(ldcp->id, 1856 "i_ldc_process_data_ACK: (0x%llx) invalid ACKid\n", 1857 ldcp->id); 1858 1859 /* Reset the channel -- as we cannot continue */ 1860 i_ldc_reset(ldcp, B_TRUE); 1861 mutex_exit(&ldcp->tx_lock); 1862 return (ECONNRESET); 1863 } 1864 } 1865 1866 mutex_exit(&ldcp->tx_lock); 1867 return (0); 1868 } 1869 1870 /* 1871 * Process incoming control message 1872 * Return 0 - session can continue 1873 * EAGAIN - reprocess packet - state was changed 1874 * ECONNRESET - channel was reset 1875 */ 1876 static int 1877 i_ldc_ctrlmsg(ldc_chan_t *ldcp, ldc_msg_t *msg) 1878 { 1879 int rv = 0; 1880 1881 D1(ldcp->id, "i_ldc_ctrlmsg: (%llx) tstate = %lx, hstate = %lx\n", 1882 ldcp->id, ldcp->tstate, ldcp->hstate); 1883 1884 switch (ldcp->tstate & ~TS_IN_RESET) { 1885 1886 case TS_OPEN: 1887 case TS_READY: 1888 1889 switch (msg->ctrl & LDC_CTRL_MASK) { 1890 case LDC_VER: 1891 /* process version message */ 1892 rv = i_ldc_process_VER(ldcp, msg); 1893 break; 1894 default: 1895 DWARN(ldcp->id, 1896 "i_ldc_ctrlmsg: (0x%llx) unexp ctrl 0x%x " 1897 "tstate=0x%x\n", ldcp->id, 1898 (msg->ctrl & LDC_CTRL_MASK), ldcp->tstate); 1899 break; 1900 } 1901 1902 break; 1903 1904 case TS_VREADY: 1905 1906 switch (msg->ctrl & LDC_CTRL_MASK) { 1907 case LDC_VER: 1908 /* process version message */ 1909 rv = i_ldc_process_VER(ldcp, msg); 1910 break; 1911 case LDC_RTS: 1912 /* process RTS message */ 1913 rv = i_ldc_process_RTS(ldcp, msg); 1914 break; 1915 case LDC_RTR: 1916 /* process RTR message */ 1917 rv = i_ldc_process_RTR(ldcp, msg); 1918 break; 1919 case LDC_RDX: 1920 /* process RDX message */ 1921 rv = i_ldc_process_RDX(ldcp, msg); 1922 break; 1923 default: 1924 DWARN(ldcp->id, 1925 "i_ldc_ctrlmsg: (0x%llx) unexp ctrl 0x%x " 1926 "tstate=0x%x\n", ldcp->id, 1927 (msg->ctrl & LDC_CTRL_MASK), ldcp->tstate); 1928 break; 1929 } 1930 1931 break; 1932 1933 case TS_UP: 1934 1935 switch (msg->ctrl & LDC_CTRL_MASK) { 1936 case LDC_VER: 1937 DWARN(ldcp->id, 1938 "i_ldc_ctrlmsg: (0x%llx) unexpected VER " 1939 "- LDC reset\n", ldcp->id); 1940 /* peer is redoing version negotiation */ 1941 mutex_enter(&ldcp->tx_lock); 1942 (void) i_ldc_txq_reconf(ldcp); 1943 i_ldc_reset_state(ldcp); 1944 mutex_exit(&ldcp->tx_lock); 1945 rv = EAGAIN; 1946 break; 1947 1948 case LDC_RDX: 1949 /* process RDX message */ 1950 rv = i_ldc_process_RDX(ldcp, msg); 1951 break; 1952 1953 default: 1954 DWARN(ldcp->id, 1955 "i_ldc_ctrlmsg: (0x%llx) unexp ctrl 0x%x " 1956 "tstate=0x%x\n", ldcp->id, 1957 (msg->ctrl & LDC_CTRL_MASK), ldcp->tstate); 1958 break; 1959 } 1960 } 1961 1962 return (rv); 1963 } 1964 1965 /* 1966 * Register channel with the channel nexus 1967 */ 1968 static int 1969 i_ldc_register_channel(ldc_chan_t *ldcp) 1970 { 1971 int rv = 0; 1972 ldc_cnex_t *cinfo = &ldcssp->cinfo; 1973 1974 if (cinfo->dip == NULL) { 1975 DWARN(ldcp->id, 1976 "i_ldc_register_channel: cnex has not registered\n"); 1977 return (EAGAIN); 1978 } 1979 1980 rv = cinfo->reg_chan(cinfo->dip, ldcp->id, ldcp->devclass); 1981 if (rv) { 1982 DWARN(ldcp->id, 1983 "i_ldc_register_channel: cannot register channel\n"); 1984 return (rv); 1985 } 1986 1987 rv = cinfo->add_intr(cinfo->dip, ldcp->id, CNEX_TX_INTR, 1988 i_ldc_tx_hdlr, ldcp, NULL); 1989 if (rv) { 1990 DWARN(ldcp->id, 1991 "i_ldc_register_channel: cannot add Tx interrupt\n"); 1992 (void) cinfo->unreg_chan(cinfo->dip, ldcp->id); 1993 return (rv); 1994 } 1995 1996 rv = cinfo->add_intr(cinfo->dip, ldcp->id, CNEX_RX_INTR, 1997 i_ldc_rx_hdlr, ldcp, NULL); 1998 if (rv) { 1999 DWARN(ldcp->id, 2000 "i_ldc_register_channel: cannot add Rx interrupt\n"); 2001 (void) cinfo->rem_intr(cinfo->dip, ldcp->id, CNEX_TX_INTR); 2002 (void) cinfo->unreg_chan(cinfo->dip, ldcp->id); 2003 return (rv); 2004 } 2005 2006 ldcp->tstate |= TS_CNEX_RDY; 2007 2008 return (0); 2009 } 2010 2011 /* 2012 * Unregister a channel with the channel nexus 2013 */ 2014 static int 2015 i_ldc_unregister_channel(ldc_chan_t *ldcp) 2016 { 2017 int rv = 0; 2018 ldc_cnex_t *cinfo = &ldcssp->cinfo; 2019 2020 if (cinfo->dip == NULL) { 2021 DWARN(ldcp->id, 2022 "i_ldc_unregister_channel: cnex has not registered\n"); 2023 return (EAGAIN); 2024 } 2025 2026 if (ldcp->tstate & TS_CNEX_RDY) { 2027 2028 /* Remove the Rx interrupt */ 2029 rv = cinfo->rem_intr(cinfo->dip, ldcp->id, CNEX_RX_INTR); 2030 if (rv) { 2031 if (rv != EAGAIN) { 2032 DWARN(ldcp->id, 2033 "i_ldc_unregister_channel: err removing " 2034 "Rx intr\n"); 2035 return (rv); 2036 } 2037 2038 /* 2039 * If interrupts are pending and handler has 2040 * finished running, clear interrupt and try 2041 * again 2042 */ 2043 if (ldcp->rx_intr_state != LDC_INTR_PEND) 2044 return (rv); 2045 2046 (void) i_ldc_clear_intr(ldcp, CNEX_RX_INTR); 2047 rv = cinfo->rem_intr(cinfo->dip, ldcp->id, 2048 CNEX_RX_INTR); 2049 if (rv) { 2050 DWARN(ldcp->id, "i_ldc_unregister_channel: " 2051 "err removing Rx interrupt\n"); 2052 return (rv); 2053 } 2054 } 2055 2056 /* Remove the Tx interrupt */ 2057 rv = cinfo->rem_intr(cinfo->dip, ldcp->id, CNEX_TX_INTR); 2058 if (rv) { 2059 DWARN(ldcp->id, 2060 "i_ldc_unregister_channel: err removing Tx intr\n"); 2061 return (rv); 2062 } 2063 2064 /* Unregister the channel */ 2065 rv = cinfo->unreg_chan(ldcssp->cinfo.dip, ldcp->id); 2066 if (rv) { 2067 DWARN(ldcp->id, 2068 "i_ldc_unregister_channel: cannot unreg channel\n"); 2069 return (rv); 2070 } 2071 2072 ldcp->tstate &= ~TS_CNEX_RDY; 2073 } 2074 2075 return (0); 2076 } 2077 2078 2079 /* 2080 * LDC transmit interrupt handler 2081 * triggered for chanel up/down/reset events 2082 * and Tx queue content changes 2083 */ 2084 static uint_t 2085 i_ldc_tx_hdlr(caddr_t arg1, caddr_t arg2) 2086 { 2087 _NOTE(ARGUNUSED(arg2)) 2088 2089 int rv; 2090 ldc_chan_t *ldcp; 2091 boolean_t notify_client = B_FALSE; 2092 uint64_t notify_event = 0, link_state; 2093 2094 /* Get the channel for which interrupt was received */ 2095 ASSERT(arg1 != NULL); 2096 ldcp = (ldc_chan_t *)arg1; 2097 2098 D1(ldcp->id, "i_ldc_tx_hdlr: (0x%llx) Received intr, ldcp=0x%p\n", 2099 ldcp->id, ldcp); 2100 2101 /* Lock channel */ 2102 mutex_enter(&ldcp->lock); 2103 2104 /* Obtain Tx lock */ 2105 mutex_enter(&ldcp->tx_lock); 2106 2107 /* mark interrupt as pending */ 2108 ldcp->tx_intr_state = LDC_INTR_ACTIVE; 2109 2110 /* save current link state */ 2111 link_state = ldcp->link_state; 2112 2113 rv = hv_ldc_tx_get_state(ldcp->id, &ldcp->tx_head, &ldcp->tx_tail, 2114 &ldcp->link_state); 2115 if (rv) { 2116 cmn_err(CE_WARN, 2117 "i_ldc_tx_hdlr: (0x%lx) cannot read queue ptrs rv=0x%d\n", 2118 ldcp->id, rv); 2119 i_ldc_clear_intr(ldcp, CNEX_TX_INTR); 2120 mutex_exit(&ldcp->tx_lock); 2121 mutex_exit(&ldcp->lock); 2122 return (DDI_INTR_CLAIMED); 2123 } 2124 2125 /* 2126 * reset the channel state if the channel went down 2127 * (other side unconfigured queue) or channel was reset 2128 * (other side reconfigured its queue) 2129 */ 2130 if (link_state != ldcp->link_state && 2131 ldcp->link_state == LDC_CHANNEL_DOWN) { 2132 D1(ldcp->id, "i_ldc_tx_hdlr: channel link down\n", ldcp->id); 2133 i_ldc_reset(ldcp, B_FALSE); 2134 notify_client = B_TRUE; 2135 notify_event = LDC_EVT_DOWN; 2136 } 2137 2138 if (link_state != ldcp->link_state && 2139 ldcp->link_state == LDC_CHANNEL_RESET) { 2140 D1(ldcp->id, "i_ldc_tx_hdlr: channel link reset\n", ldcp->id); 2141 i_ldc_reset(ldcp, B_FALSE); 2142 notify_client = B_TRUE; 2143 notify_event = LDC_EVT_RESET; 2144 } 2145 2146 if (link_state != ldcp->link_state && 2147 (ldcp->tstate & ~TS_IN_RESET) == TS_OPEN && 2148 ldcp->link_state == LDC_CHANNEL_UP) { 2149 D1(ldcp->id, "i_ldc_tx_hdlr: channel link up\n", ldcp->id); 2150 notify_client = B_TRUE; 2151 notify_event = LDC_EVT_RESET; 2152 ldcp->tstate |= TS_LINK_READY; 2153 ldcp->status = LDC_READY; 2154 } 2155 2156 /* if callbacks are disabled, do not notify */ 2157 if (!ldcp->cb_enabled) 2158 notify_client = B_FALSE; 2159 2160 i_ldc_clear_intr(ldcp, CNEX_TX_INTR); 2161 mutex_exit(&ldcp->tx_lock); 2162 2163 if (notify_client) { 2164 ldcp->cb_inprogress = B_TRUE; 2165 mutex_exit(&ldcp->lock); 2166 rv = ldcp->cb(notify_event, ldcp->cb_arg); 2167 if (rv) { 2168 DWARN(ldcp->id, "i_ldc_tx_hdlr: (0x%llx) callback " 2169 "failure", ldcp->id); 2170 } 2171 mutex_enter(&ldcp->lock); 2172 ldcp->cb_inprogress = B_FALSE; 2173 } 2174 2175 mutex_exit(&ldcp->lock); 2176 2177 D1(ldcp->id, "i_ldc_tx_hdlr: (0x%llx) exiting handler", ldcp->id); 2178 2179 return (DDI_INTR_CLAIMED); 2180 } 2181 2182 /* 2183 * Process the Rx HV queue. 2184 * 2185 * Returns 0 if data packets were found and no errors were encountered, 2186 * otherwise returns an error. In either case, the *notify argument is 2187 * set to indicate whether or not the client callback function should 2188 * be invoked. The *event argument is set to contain the callback event. 2189 * 2190 * Depending on the channel mode, packets are handled differently: 2191 * 2192 * RAW MODE 2193 * For raw mode channels, when a data packet is encountered, 2194 * processing stops and all packets are left on the queue to be removed 2195 * and processed by the ldc_read code path. 2196 * 2197 * UNRELIABLE MODE 2198 * For unreliable mode, when a data packet is encountered, processing 2199 * stops, and all packets are left on the queue to be removed and 2200 * processed by the ldc_read code path. Control packets are processed 2201 * inline if they are encountered before any data packets. 2202 * 2203 * RELIABLE MODE 2204 * For reliable mode channels, all packets on the receive queue 2205 * are processed: data packets are copied to the data queue and 2206 * control packets are processed inline. Packets are only left on 2207 * the receive queue when the data queue is full. 2208 */ 2209 static uint_t 2210 i_ldc_rx_process_hvq(ldc_chan_t *ldcp, boolean_t *notify_client, 2211 uint64_t *notify_event) 2212 { 2213 int rv; 2214 uint64_t rx_head, rx_tail; 2215 ldc_msg_t *msg; 2216 uint64_t link_state, first_fragment = 0; 2217 boolean_t trace_length = B_TRUE; 2218 2219 ASSERT(MUTEX_HELD(&ldcp->lock)); 2220 *notify_client = B_FALSE; 2221 *notify_event = 0; 2222 2223 /* 2224 * Read packet(s) from the queue 2225 */ 2226 for (;;) { 2227 2228 link_state = ldcp->link_state; 2229 rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail, 2230 &ldcp->link_state); 2231 if (rv) { 2232 cmn_err(CE_WARN, 2233 "i_ldc_rx_process_hvq: (0x%lx) cannot read " 2234 "queue ptrs, rv=0x%d\n", ldcp->id, rv); 2235 i_ldc_clear_intr(ldcp, CNEX_RX_INTR); 2236 return (EIO); 2237 } 2238 2239 /* 2240 * reset the channel state if the channel went down 2241 * (other side unconfigured queue) or channel was reset 2242 * (other side reconfigured its queue) 2243 */ 2244 2245 if (link_state != ldcp->link_state) { 2246 2247 switch (ldcp->link_state) { 2248 case LDC_CHANNEL_DOWN: 2249 D1(ldcp->id, "i_ldc_rx_process_hvq: channel " 2250 "link down\n", ldcp->id); 2251 mutex_enter(&ldcp->tx_lock); 2252 i_ldc_reset(ldcp, B_FALSE); 2253 mutex_exit(&ldcp->tx_lock); 2254 *notify_client = B_TRUE; 2255 *notify_event = LDC_EVT_DOWN; 2256 goto loop_exit; 2257 2258 case LDC_CHANNEL_UP: 2259 D1(ldcp->id, "i_ldc_rx_process_hvq: " 2260 "channel link up\n", ldcp->id); 2261 2262 if ((ldcp->tstate & ~TS_IN_RESET) == TS_OPEN) { 2263 *notify_client = B_TRUE; 2264 *notify_event = LDC_EVT_RESET; 2265 ldcp->tstate |= TS_LINK_READY; 2266 ldcp->status = LDC_READY; 2267 } 2268 break; 2269 2270 case LDC_CHANNEL_RESET: 2271 default: 2272 #ifdef DEBUG 2273 force_reset: 2274 #endif 2275 D1(ldcp->id, "i_ldc_rx_process_hvq: channel " 2276 "link reset\n", ldcp->id); 2277 mutex_enter(&ldcp->tx_lock); 2278 i_ldc_reset(ldcp, B_FALSE); 2279 mutex_exit(&ldcp->tx_lock); 2280 *notify_client = B_TRUE; 2281 *notify_event = LDC_EVT_RESET; 2282 break; 2283 } 2284 } 2285 2286 #ifdef DEBUG 2287 if (LDC_INJECT_RESET(ldcp)) 2288 goto force_reset; 2289 if (LDC_INJECT_DRNGCLEAR(ldcp)) 2290 i_ldc_mem_inject_dring_clear(ldcp); 2291 #endif 2292 if (trace_length) { 2293 TRACE_RXHVQ_LENGTH(ldcp, rx_head, rx_tail); 2294 trace_length = B_FALSE; 2295 } 2296 2297 if (rx_head == rx_tail) { 2298 D2(ldcp->id, "i_ldc_rx_process_hvq: (0x%llx) " 2299 "No packets\n", ldcp->id); 2300 break; 2301 } 2302 2303 D2(ldcp->id, "i_ldc_rx_process_hvq: head=0x%llx, " 2304 "tail=0x%llx\n", rx_head, rx_tail); 2305 DUMP_LDC_PKT(ldcp, "i_ldc_rx_process_hvq rcd", 2306 ldcp->rx_q_va + rx_head); 2307 2308 /* get the message */ 2309 msg = (ldc_msg_t *)(ldcp->rx_q_va + rx_head); 2310 2311 /* if channel is in RAW mode or data pkt, notify and return */ 2312 if (ldcp->mode == LDC_MODE_RAW) { 2313 *notify_client = B_TRUE; 2314 *notify_event |= LDC_EVT_READ; 2315 break; 2316 } 2317 2318 if ((msg->type & LDC_DATA) && (msg->stype & LDC_INFO)) { 2319 2320 /* discard packet if channel is not up */ 2321 if ((ldcp->tstate & ~TS_IN_RESET) != TS_UP) { 2322 2323 /* move the head one position */ 2324 rx_head = (rx_head + LDC_PACKET_SIZE) % 2325 (ldcp->rx_q_entries << LDC_PACKET_SHIFT); 2326 2327 if (rv = i_ldc_set_rx_head(ldcp, rx_head)) 2328 break; 2329 2330 continue; 2331 } else { 2332 uint64_t dq_head, dq_tail; 2333 2334 /* process only RELIABLE mode data packets */ 2335 if (ldcp->mode != LDC_MODE_RELIABLE) { 2336 if ((ldcp->tstate & TS_IN_RESET) == 0) 2337 *notify_client = B_TRUE; 2338 *notify_event |= LDC_EVT_READ; 2339 break; 2340 } 2341 2342 /* don't process packet if queue full */ 2343 (void) i_ldc_dq_rx_get_state(ldcp, &dq_head, 2344 &dq_tail, NULL); 2345 dq_tail = (dq_tail + LDC_PACKET_SIZE) % 2346 (ldcp->rx_dq_entries << LDC_PACKET_SHIFT); 2347 if (dq_tail == dq_head || 2348 LDC_INJECT_DQFULL(ldcp)) { 2349 rv = ENOSPC; 2350 break; 2351 } 2352 } 2353 } 2354 2355 /* Check the sequence ID for the message received */ 2356 rv = i_ldc_check_seqid(ldcp, msg); 2357 if (rv != 0) { 2358 2359 DWARN(ldcp->id, "i_ldc_rx_process_hvq: (0x%llx) " 2360 "seqid error, q_ptrs=0x%lx,0x%lx", ldcp->id, 2361 rx_head, rx_tail); 2362 2363 /* Reset last_msg_rcd to start of message */ 2364 if (first_fragment != 0) { 2365 ldcp->last_msg_rcd = first_fragment - 1; 2366 first_fragment = 0; 2367 } 2368 2369 /* 2370 * Send a NACK due to seqid mismatch 2371 */ 2372 rv = i_ldc_send_pkt(ldcp, msg->type, LDC_NACK, 2373 (msg->ctrl & LDC_CTRL_MASK)); 2374 2375 if (rv) { 2376 cmn_err(CE_NOTE, "i_ldc_rx_process_hvq: " 2377 "(0x%lx) err sending CTRL/DATA NACK msg\n", 2378 ldcp->id); 2379 2380 /* if cannot send NACK - reset channel */ 2381 mutex_enter(&ldcp->tx_lock); 2382 i_ldc_reset(ldcp, B_TRUE); 2383 mutex_exit(&ldcp->tx_lock); 2384 2385 *notify_client = B_TRUE; 2386 *notify_event = LDC_EVT_RESET; 2387 break; 2388 } 2389 2390 /* purge receive queue */ 2391 (void) i_ldc_set_rx_head(ldcp, rx_tail); 2392 break; 2393 } 2394 2395 /* record the message ID */ 2396 ldcp->last_msg_rcd = msg->seqid; 2397 2398 /* process control messages */ 2399 if (msg->type & LDC_CTRL) { 2400 /* save current internal state */ 2401 uint64_t tstate = ldcp->tstate; 2402 2403 rv = i_ldc_ctrlmsg(ldcp, msg); 2404 if (rv == EAGAIN) { 2405 /* re-process pkt - state was adjusted */ 2406 continue; 2407 } 2408 if (rv == ECONNRESET) { 2409 *notify_client = B_TRUE; 2410 *notify_event = LDC_EVT_RESET; 2411 break; 2412 } 2413 2414 /* 2415 * control message processing was successful 2416 * channel transitioned to ready for communication 2417 */ 2418 if (rv == 0 && ldcp->tstate == TS_UP && 2419 (tstate & ~TS_IN_RESET) != 2420 (ldcp->tstate & ~TS_IN_RESET)) { 2421 *notify_client = B_TRUE; 2422 *notify_event = LDC_EVT_UP; 2423 } 2424 } 2425 2426 /* process data NACKs */ 2427 if ((msg->type & LDC_DATA) && (msg->stype & LDC_NACK)) { 2428 DWARN(ldcp->id, 2429 "i_ldc_rx_process_hvq: (0x%llx) received DATA/NACK", 2430 ldcp->id); 2431 mutex_enter(&ldcp->tx_lock); 2432 i_ldc_reset(ldcp, B_TRUE); 2433 mutex_exit(&ldcp->tx_lock); 2434 *notify_client = B_TRUE; 2435 *notify_event = LDC_EVT_RESET; 2436 break; 2437 } 2438 2439 /* process data ACKs */ 2440 if ((msg->type & LDC_DATA) && (msg->stype & LDC_ACK)) { 2441 if (rv = i_ldc_process_data_ACK(ldcp, msg)) { 2442 *notify_client = B_TRUE; 2443 *notify_event = LDC_EVT_RESET; 2444 break; 2445 } 2446 } 2447 2448 if ((msg->type & LDC_DATA) && (msg->stype & LDC_INFO)) { 2449 ASSERT(ldcp->mode == LDC_MODE_RELIABLE); 2450 2451 /* 2452 * Copy the data packet to the data queue. Note 2453 * that the copy routine updates the rx_head pointer. 2454 */ 2455 i_ldc_rxdq_copy(ldcp, &rx_head); 2456 2457 if ((ldcp->tstate & TS_IN_RESET) == 0) 2458 *notify_client = B_TRUE; 2459 *notify_event |= LDC_EVT_READ; 2460 } else { 2461 rx_head = (rx_head + LDC_PACKET_SIZE) % 2462 (ldcp->rx_q_entries << LDC_PACKET_SHIFT); 2463 } 2464 2465 /* move the head one position */ 2466 if (rv = i_ldc_set_rx_head(ldcp, rx_head)) { 2467 *notify_client = B_TRUE; 2468 *notify_event = LDC_EVT_RESET; 2469 break; 2470 } 2471 2472 } /* for */ 2473 2474 loop_exit: 2475 2476 if (ldcp->mode == LDC_MODE_RELIABLE) { 2477 /* ACK data packets */ 2478 if ((*notify_event & 2479 (LDC_EVT_READ | LDC_EVT_RESET)) == LDC_EVT_READ) { 2480 int ack_rv; 2481 ack_rv = i_ldc_send_pkt(ldcp, LDC_DATA, LDC_ACK, 0); 2482 if (ack_rv && ack_rv != EWOULDBLOCK) { 2483 cmn_err(CE_NOTE, 2484 "i_ldc_rx_process_hvq: (0x%lx) cannot " 2485 "send ACK\n", ldcp->id); 2486 2487 mutex_enter(&ldcp->tx_lock); 2488 i_ldc_reset(ldcp, B_FALSE); 2489 mutex_exit(&ldcp->tx_lock); 2490 2491 *notify_client = B_TRUE; 2492 *notify_event = LDC_EVT_RESET; 2493 goto skip_ackpeek; 2494 } 2495 } 2496 2497 /* 2498 * If we have no more space on the data queue, make sure 2499 * there are no ACKs on the rx queue waiting to be processed. 2500 */ 2501 if (rv == ENOSPC) { 2502 if (i_ldc_rx_ackpeek(ldcp, rx_head, rx_tail) != 0) { 2503 ldcp->rx_ack_head = ACKPEEK_HEAD_INVALID; 2504 *notify_client = B_TRUE; 2505 *notify_event = LDC_EVT_RESET; 2506 } 2507 return (rv); 2508 } else { 2509 ldcp->rx_ack_head = ACKPEEK_HEAD_INVALID; 2510 } 2511 } 2512 2513 skip_ackpeek: 2514 2515 /* Return, indicating whether or not data packets were found */ 2516 if ((*notify_event & (LDC_EVT_READ | LDC_EVT_RESET)) == LDC_EVT_READ) 2517 return (0); 2518 2519 return (ENOMSG); 2520 } 2521 2522 /* 2523 * Process any ACK packets on the HV receive queue. 2524 * 2525 * This function is only used by RELIABLE mode channels when the 2526 * secondary data queue fills up and there are packets remaining on 2527 * the HV receive queue. 2528 */ 2529 int 2530 i_ldc_rx_ackpeek(ldc_chan_t *ldcp, uint64_t rx_head, uint64_t rx_tail) 2531 { 2532 int rv = 0; 2533 ldc_msg_t *msg; 2534 2535 if (ldcp->rx_ack_head == ACKPEEK_HEAD_INVALID) 2536 ldcp->rx_ack_head = rx_head; 2537 2538 while (ldcp->rx_ack_head != rx_tail) { 2539 msg = (ldc_msg_t *)(ldcp->rx_q_va + ldcp->rx_ack_head); 2540 2541 if ((msg->type & LDC_DATA) && (msg->stype & LDC_ACK)) { 2542 if (rv = i_ldc_process_data_ACK(ldcp, msg)) 2543 break; 2544 msg->stype &= ~LDC_ACK; 2545 } 2546 2547 ldcp->rx_ack_head = 2548 (ldcp->rx_ack_head + LDC_PACKET_SIZE) % 2549 (ldcp->rx_q_entries << LDC_PACKET_SHIFT); 2550 } 2551 return (rv); 2552 } 2553 2554 /* -------------------------------------------------------------------------- */ 2555 2556 /* 2557 * LDC API functions 2558 */ 2559 2560 /* 2561 * Initialize the channel. Allocate internal structure and memory for 2562 * TX/RX queues, and initialize locks. 2563 */ 2564 int 2565 ldc_init(uint64_t id, ldc_attr_t *attr, ldc_handle_t *handle) 2566 { 2567 ldc_chan_t *ldcp; 2568 int rv, exit_val; 2569 uint64_t ra_base, nentries; 2570 uint64_t qlen; 2571 2572 exit_val = EINVAL; /* guarantee an error if exit on failure */ 2573 2574 if (attr == NULL) { 2575 DWARN(id, "ldc_init: (0x%llx) invalid attr\n", id); 2576 return (EINVAL); 2577 } 2578 if (handle == NULL) { 2579 DWARN(id, "ldc_init: (0x%llx) invalid handle\n", id); 2580 return (EINVAL); 2581 } 2582 2583 /* check if channel is valid */ 2584 rv = hv_ldc_tx_qinfo(id, &ra_base, &nentries); 2585 if (rv == H_ECHANNEL) { 2586 DWARN(id, "ldc_init: (0x%llx) invalid channel id\n", id); 2587 return (EINVAL); 2588 } 2589 2590 /* check if the channel has already been initialized */ 2591 mutex_enter(&ldcssp->lock); 2592 ldcp = ldcssp->chan_list; 2593 while (ldcp != NULL) { 2594 if (ldcp->id == id) { 2595 DWARN(id, "ldc_init: (0x%llx) already initialized\n", 2596 id); 2597 mutex_exit(&ldcssp->lock); 2598 return (EADDRINUSE); 2599 } 2600 ldcp = ldcp->next; 2601 } 2602 mutex_exit(&ldcssp->lock); 2603 2604 ASSERT(ldcp == NULL); 2605 2606 *handle = 0; 2607 2608 /* Allocate an ldcp structure */ 2609 ldcp = kmem_zalloc(sizeof (ldc_chan_t), KM_SLEEP); 2610 2611 /* 2612 * Initialize the channel and Tx lock 2613 * 2614 * The channel 'lock' protects the entire channel and 2615 * should be acquired before initializing, resetting, 2616 * destroying or reading from a channel. 2617 * 2618 * The 'tx_lock' should be acquired prior to transmitting 2619 * data over the channel. The lock should also be acquired 2620 * prior to channel reconfiguration (in order to prevent 2621 * concurrent writes). 2622 * 2623 * ORDERING: When both locks are being acquired, to prevent 2624 * deadlocks, the channel lock should be always acquired prior 2625 * to the tx_lock. 2626 */ 2627 mutex_init(&ldcp->lock, NULL, MUTEX_DRIVER, NULL); 2628 mutex_init(&ldcp->tx_lock, NULL, MUTEX_DRIVER, NULL); 2629 2630 /* Initialize the channel */ 2631 ldcp->id = id; 2632 ldcp->cb = NULL; 2633 ldcp->cb_arg = NULL; 2634 ldcp->cb_inprogress = B_FALSE; 2635 ldcp->cb_enabled = B_FALSE; 2636 ldcp->next = NULL; 2637 2638 /* Read attributes */ 2639 ldcp->mode = attr->mode; 2640 ldcp->devclass = attr->devclass; 2641 ldcp->devinst = attr->instance; 2642 ldcp->mtu = (attr->mtu > 0) ? attr->mtu : LDC_DEFAULT_MTU; 2643 2644 D1(ldcp->id, 2645 "ldc_init: (0x%llx) channel attributes, class=0x%x, " 2646 "instance=0x%llx, mode=%d, mtu=%d\n", 2647 ldcp->id, ldcp->devclass, ldcp->devinst, ldcp->mode, ldcp->mtu); 2648 2649 ldcp->next_vidx = 0; 2650 ldcp->tstate = TS_IN_RESET; 2651 ldcp->hstate = 0; 2652 ldcp->last_msg_snt = LDC_INIT_SEQID; 2653 ldcp->last_ack_rcd = 0; 2654 ldcp->last_msg_rcd = 0; 2655 ldcp->rx_ack_head = ACKPEEK_HEAD_INVALID; 2656 2657 ldcp->stream_bufferp = NULL; 2658 ldcp->exp_dring_list = NULL; 2659 ldcp->imp_dring_list = NULL; 2660 ldcp->mhdl_list = NULL; 2661 2662 ldcp->tx_intr_state = LDC_INTR_NONE; 2663 ldcp->rx_intr_state = LDC_INTR_NONE; 2664 2665 /* Initialize payload size depending on whether channel is reliable */ 2666 switch (ldcp->mode) { 2667 case LDC_MODE_RAW: 2668 ldcp->pkt_payload = LDC_PAYLOAD_SIZE_RAW; 2669 ldcp->read_p = i_ldc_read_raw; 2670 ldcp->write_p = i_ldc_write_raw; 2671 break; 2672 case LDC_MODE_UNRELIABLE: 2673 ldcp->pkt_payload = LDC_PAYLOAD_SIZE_UNRELIABLE; 2674 ldcp->read_p = i_ldc_read_packet; 2675 ldcp->write_p = i_ldc_write_packet; 2676 break; 2677 case LDC_MODE_RELIABLE: 2678 ldcp->pkt_payload = LDC_PAYLOAD_SIZE_RELIABLE; 2679 2680 ldcp->stream_remains = 0; 2681 ldcp->stream_offset = 0; 2682 ldcp->stream_bufferp = kmem_alloc(ldcp->mtu, KM_SLEEP); 2683 ldcp->read_p = i_ldc_read_stream; 2684 ldcp->write_p = i_ldc_write_stream; 2685 break; 2686 default: 2687 exit_val = EINVAL; 2688 goto cleanup_on_exit; 2689 } 2690 2691 /* 2692 * qlen is (mtu * ldc_mtu_msgs) / pkt_payload. If this 2693 * value is smaller than default length of ldc_queue_entries, 2694 * qlen is set to ldc_queue_entries. Ensure that computed 2695 * length is a power-of-two value. 2696 */ 2697 qlen = (ldcp->mtu * ldc_mtu_msgs) / ldcp->pkt_payload; 2698 if (!ISP2(qlen)) { 2699 uint64_t tmp = 1; 2700 while (qlen) { 2701 qlen >>= 1; tmp <<= 1; 2702 } 2703 qlen = tmp; 2704 } 2705 2706 ldcp->rx_q_entries = 2707 (qlen < ldc_queue_entries) ? ldc_queue_entries : qlen; 2708 ldcp->tx_q_entries = ldcp->rx_q_entries; 2709 2710 D1(ldcp->id, "ldc_init: queue length = 0x%llx\n", ldcp->rx_q_entries); 2711 2712 /* Create a transmit queue */ 2713 ldcp->tx_q_va = (uint64_t) 2714 contig_mem_alloc(ldcp->tx_q_entries << LDC_PACKET_SHIFT); 2715 if (ldcp->tx_q_va == NULL) { 2716 cmn_err(CE_WARN, 2717 "ldc_init: (0x%lx) TX queue allocation failed\n", 2718 ldcp->id); 2719 exit_val = ENOMEM; 2720 goto cleanup_on_exit; 2721 } 2722 ldcp->tx_q_ra = va_to_pa((caddr_t)ldcp->tx_q_va); 2723 2724 D2(ldcp->id, "ldc_init: txq_va=0x%llx, txq_ra=0x%llx, entries=0x%llx\n", 2725 ldcp->tx_q_va, ldcp->tx_q_ra, ldcp->tx_q_entries); 2726 2727 ldcp->tstate |= TS_TXQ_RDY; 2728 2729 /* Create a receive queue */ 2730 ldcp->rx_q_va = (uint64_t) 2731 contig_mem_alloc(ldcp->rx_q_entries << LDC_PACKET_SHIFT); 2732 if (ldcp->rx_q_va == NULL) { 2733 cmn_err(CE_WARN, 2734 "ldc_init: (0x%lx) RX queue allocation failed\n", 2735 ldcp->id); 2736 exit_val = ENOMEM; 2737 goto cleanup_on_exit; 2738 } 2739 ldcp->rx_q_ra = va_to_pa((caddr_t)ldcp->rx_q_va); 2740 2741 D2(ldcp->id, "ldc_init: rxq_va=0x%llx, rxq_ra=0x%llx, entries=0x%llx\n", 2742 ldcp->rx_q_va, ldcp->rx_q_ra, ldcp->rx_q_entries); 2743 2744 ldcp->tstate |= TS_RXQ_RDY; 2745 2746 /* Setup a separate read data queue */ 2747 if (ldcp->mode == LDC_MODE_RELIABLE) { 2748 ldcp->readq_get_state = i_ldc_dq_rx_get_state; 2749 ldcp->readq_set_head = i_ldc_set_rxdq_head; 2750 2751 /* Make sure the data queue multiplier is a power of 2 */ 2752 if (!ISP2(ldc_rxdq_multiplier)) { 2753 D1(ldcp->id, "ldc_init: (0x%llx) ldc_rxdq_multiplier " 2754 "not a power of 2, resetting", ldcp->id); 2755 ldc_rxdq_multiplier = LDC_RXDQ_MULTIPLIER; 2756 } 2757 2758 ldcp->rx_dq_entries = ldc_rxdq_multiplier * ldcp->rx_q_entries; 2759 ldcp->rx_dq_va = (uint64_t) 2760 kmem_alloc(ldcp->rx_dq_entries << LDC_PACKET_SHIFT, 2761 KM_SLEEP); 2762 if (ldcp->rx_dq_va == NULL) { 2763 cmn_err(CE_WARN, 2764 "ldc_init: (0x%lx) RX data queue " 2765 "allocation failed\n", ldcp->id); 2766 exit_val = ENOMEM; 2767 goto cleanup_on_exit; 2768 } 2769 2770 ldcp->rx_dq_head = ldcp->rx_dq_tail = 0; 2771 2772 D2(ldcp->id, "ldc_init: rx_dq_va=0x%llx, " 2773 "rx_dq_entries=0x%llx\n", ldcp->rx_dq_va, 2774 ldcp->rx_dq_entries); 2775 } else { 2776 ldcp->readq_get_state = i_ldc_hvq_rx_get_state; 2777 ldcp->readq_set_head = i_ldc_set_rx_head; 2778 } 2779 2780 /* Init descriptor ring and memory handle list lock */ 2781 mutex_init(&ldcp->exp_dlist_lock, NULL, MUTEX_DRIVER, NULL); 2782 mutex_init(&ldcp->imp_dlist_lock, NULL, MUTEX_DRIVER, NULL); 2783 mutex_init(&ldcp->mlist_lock, NULL, MUTEX_DRIVER, NULL); 2784 2785 /* mark status as INITialized */ 2786 ldcp->status = LDC_INIT; 2787 2788 /* Add to channel list */ 2789 mutex_enter(&ldcssp->lock); 2790 ldcp->next = ldcssp->chan_list; 2791 ldcssp->chan_list = ldcp; 2792 ldcssp->channel_count++; 2793 mutex_exit(&ldcssp->lock); 2794 2795 /* set the handle */ 2796 *handle = (ldc_handle_t)ldcp; 2797 2798 D1(ldcp->id, "ldc_init: (0x%llx) channel initialized\n", ldcp->id); 2799 2800 return (0); 2801 2802 cleanup_on_exit: 2803 2804 if (ldcp->mode == LDC_MODE_RELIABLE && ldcp->stream_bufferp) 2805 kmem_free(ldcp->stream_bufferp, ldcp->mtu); 2806 2807 if (ldcp->tstate & TS_TXQ_RDY) 2808 contig_mem_free((caddr_t)ldcp->tx_q_va, 2809 (ldcp->tx_q_entries << LDC_PACKET_SHIFT)); 2810 2811 if (ldcp->tstate & TS_RXQ_RDY) 2812 contig_mem_free((caddr_t)ldcp->rx_q_va, 2813 (ldcp->rx_q_entries << LDC_PACKET_SHIFT)); 2814 2815 mutex_destroy(&ldcp->tx_lock); 2816 mutex_destroy(&ldcp->lock); 2817 2818 if (ldcp) 2819 kmem_free(ldcp, sizeof (ldc_chan_t)); 2820 2821 return (exit_val); 2822 } 2823 2824 /* 2825 * Finalizes the LDC connection. It will return EBUSY if the 2826 * channel is open. A ldc_close() has to be done prior to 2827 * a ldc_fini operation. It frees TX/RX queues, associated 2828 * with the channel 2829 */ 2830 int 2831 ldc_fini(ldc_handle_t handle) 2832 { 2833 ldc_chan_t *ldcp; 2834 ldc_chan_t *tmp_ldcp; 2835 uint64_t id; 2836 2837 if (handle == NULL) { 2838 DWARN(DBG_ALL_LDCS, "ldc_fini: invalid channel handle\n"); 2839 return (EINVAL); 2840 } 2841 ldcp = (ldc_chan_t *)handle; 2842 id = ldcp->id; 2843 2844 mutex_enter(&ldcp->lock); 2845 2846 if ((ldcp->tstate & ~TS_IN_RESET) > TS_INIT) { 2847 DWARN(ldcp->id, "ldc_fini: (0x%llx) channel is open\n", 2848 ldcp->id); 2849 mutex_exit(&ldcp->lock); 2850 return (EBUSY); 2851 } 2852 2853 /* Remove from the channel list */ 2854 mutex_enter(&ldcssp->lock); 2855 tmp_ldcp = ldcssp->chan_list; 2856 if (tmp_ldcp == ldcp) { 2857 ldcssp->chan_list = ldcp->next; 2858 ldcp->next = NULL; 2859 } else { 2860 while (tmp_ldcp != NULL) { 2861 if (tmp_ldcp->next == ldcp) { 2862 tmp_ldcp->next = ldcp->next; 2863 ldcp->next = NULL; 2864 break; 2865 } 2866 tmp_ldcp = tmp_ldcp->next; 2867 } 2868 if (tmp_ldcp == NULL) { 2869 DWARN(DBG_ALL_LDCS, "ldc_fini: invalid channel hdl\n"); 2870 mutex_exit(&ldcssp->lock); 2871 mutex_exit(&ldcp->lock); 2872 return (EINVAL); 2873 } 2874 } 2875 2876 ldcssp->channel_count--; 2877 2878 mutex_exit(&ldcssp->lock); 2879 2880 /* Free the map table for this channel */ 2881 if (ldcp->mtbl) { 2882 (void) hv_ldc_set_map_table(ldcp->id, NULL, NULL); 2883 if (ldcp->mtbl->contigmem) 2884 contig_mem_free(ldcp->mtbl->table, ldcp->mtbl->size); 2885 else 2886 kmem_free(ldcp->mtbl->table, ldcp->mtbl->size); 2887 mutex_destroy(&ldcp->mtbl->lock); 2888 kmem_free(ldcp->mtbl, sizeof (ldc_mtbl_t)); 2889 } 2890 2891 /* Destroy descriptor ring and memory handle list lock */ 2892 mutex_destroy(&ldcp->exp_dlist_lock); 2893 mutex_destroy(&ldcp->imp_dlist_lock); 2894 mutex_destroy(&ldcp->mlist_lock); 2895 2896 /* Free the stream buffer for RELIABLE_MODE */ 2897 if (ldcp->mode == LDC_MODE_RELIABLE && ldcp->stream_bufferp) 2898 kmem_free(ldcp->stream_bufferp, ldcp->mtu); 2899 2900 /* Free the RX queue */ 2901 contig_mem_free((caddr_t)ldcp->rx_q_va, 2902 (ldcp->rx_q_entries << LDC_PACKET_SHIFT)); 2903 ldcp->tstate &= ~TS_RXQ_RDY; 2904 2905 /* Free the RX data queue */ 2906 if (ldcp->mode == LDC_MODE_RELIABLE) { 2907 kmem_free((caddr_t)ldcp->rx_dq_va, 2908 (ldcp->rx_dq_entries << LDC_PACKET_SHIFT)); 2909 } 2910 2911 /* Free the TX queue */ 2912 contig_mem_free((caddr_t)ldcp->tx_q_va, 2913 (ldcp->tx_q_entries << LDC_PACKET_SHIFT)); 2914 ldcp->tstate &= ~TS_TXQ_RDY; 2915 2916 mutex_exit(&ldcp->lock); 2917 2918 /* Destroy mutex */ 2919 mutex_destroy(&ldcp->tx_lock); 2920 mutex_destroy(&ldcp->lock); 2921 2922 /* free channel structure */ 2923 kmem_free(ldcp, sizeof (ldc_chan_t)); 2924 2925 D1(id, "ldc_fini: (0x%llx) channel finalized\n", id); 2926 2927 return (0); 2928 } 2929 2930 /* 2931 * Open the LDC channel for use. It registers the TX/RX queues 2932 * with the Hypervisor. It also specifies the interrupt number 2933 * and target CPU for this channel 2934 */ 2935 int 2936 ldc_open(ldc_handle_t handle) 2937 { 2938 ldc_chan_t *ldcp; 2939 int rv; 2940 2941 if (handle == NULL) { 2942 DWARN(DBG_ALL_LDCS, "ldc_open: invalid channel handle\n"); 2943 return (EINVAL); 2944 } 2945 2946 ldcp = (ldc_chan_t *)handle; 2947 2948 mutex_enter(&ldcp->lock); 2949 2950 if (ldcp->tstate < TS_INIT) { 2951 DWARN(ldcp->id, 2952 "ldc_open: (0x%llx) channel not initialized\n", ldcp->id); 2953 mutex_exit(&ldcp->lock); 2954 return (EFAULT); 2955 } 2956 if ((ldcp->tstate & ~TS_IN_RESET) >= TS_OPEN) { 2957 DWARN(ldcp->id, 2958 "ldc_open: (0x%llx) channel is already open\n", ldcp->id); 2959 mutex_exit(&ldcp->lock); 2960 return (EFAULT); 2961 } 2962 2963 /* 2964 * Unregister/Register the tx queue with the hypervisor 2965 */ 2966 rv = hv_ldc_tx_qconf(ldcp->id, NULL, NULL); 2967 if (rv) { 2968 cmn_err(CE_WARN, 2969 "ldc_open: (0x%lx) channel tx queue unconf failed\n", 2970 ldcp->id); 2971 mutex_exit(&ldcp->lock); 2972 return (EIO); 2973 } 2974 2975 rv = hv_ldc_tx_qconf(ldcp->id, ldcp->tx_q_ra, ldcp->tx_q_entries); 2976 if (rv) { 2977 cmn_err(CE_WARN, 2978 "ldc_open: (0x%lx) channel tx queue conf failed\n", 2979 ldcp->id); 2980 mutex_exit(&ldcp->lock); 2981 return (EIO); 2982 } 2983 2984 D2(ldcp->id, "ldc_open: (0x%llx) registered tx queue with LDC\n", 2985 ldcp->id); 2986 2987 /* 2988 * Unregister/Register the rx queue with the hypervisor 2989 */ 2990 rv = hv_ldc_rx_qconf(ldcp->id, NULL, NULL); 2991 if (rv) { 2992 cmn_err(CE_WARN, 2993 "ldc_open: (0x%lx) channel rx queue unconf failed\n", 2994 ldcp->id); 2995 mutex_exit(&ldcp->lock); 2996 return (EIO); 2997 } 2998 2999 rv = hv_ldc_rx_qconf(ldcp->id, ldcp->rx_q_ra, ldcp->rx_q_entries); 3000 if (rv) { 3001 cmn_err(CE_WARN, 3002 "ldc_open: (0x%lx) channel rx queue conf failed\n", 3003 ldcp->id); 3004 mutex_exit(&ldcp->lock); 3005 return (EIO); 3006 } 3007 3008 D2(ldcp->id, "ldc_open: (0x%llx) registered rx queue with LDC\n", 3009 ldcp->id); 3010 3011 ldcp->tstate |= TS_QCONF_RDY; 3012 3013 /* Register the channel with the channel nexus */ 3014 rv = i_ldc_register_channel(ldcp); 3015 if (rv && rv != EAGAIN) { 3016 cmn_err(CE_WARN, 3017 "ldc_open: (0x%lx) channel register failed\n", ldcp->id); 3018 ldcp->tstate &= ~TS_QCONF_RDY; 3019 (void) hv_ldc_tx_qconf(ldcp->id, NULL, NULL); 3020 (void) hv_ldc_rx_qconf(ldcp->id, NULL, NULL); 3021 mutex_exit(&ldcp->lock); 3022 return (EIO); 3023 } 3024 3025 /* mark channel in OPEN state */ 3026 ldcp->status = LDC_OPEN; 3027 3028 /* Read channel state */ 3029 rv = hv_ldc_tx_get_state(ldcp->id, 3030 &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state); 3031 if (rv) { 3032 cmn_err(CE_WARN, 3033 "ldc_open: (0x%lx) cannot read channel state\n", 3034 ldcp->id); 3035 (void) i_ldc_unregister_channel(ldcp); 3036 ldcp->tstate &= ~TS_QCONF_RDY; 3037 (void) hv_ldc_tx_qconf(ldcp->id, NULL, NULL); 3038 (void) hv_ldc_rx_qconf(ldcp->id, NULL, NULL); 3039 mutex_exit(&ldcp->lock); 3040 return (EIO); 3041 } 3042 3043 /* 3044 * set the ACKd head to current head location for reliable 3045 */ 3046 ldcp->tx_ackd_head = ldcp->tx_head; 3047 3048 /* mark channel ready if HV report link is UP (peer alloc'd Rx queue) */ 3049 if (ldcp->link_state == LDC_CHANNEL_UP || 3050 ldcp->link_state == LDC_CHANNEL_RESET) { 3051 ldcp->tstate |= TS_LINK_READY; 3052 ldcp->status = LDC_READY; 3053 } 3054 3055 /* 3056 * if channel is being opened in RAW mode - no handshake is needed 3057 * switch the channel READY and UP state 3058 */ 3059 if (ldcp->mode == LDC_MODE_RAW) { 3060 ldcp->tstate = TS_UP; /* set bits associated with LDC UP */ 3061 ldcp->status = LDC_UP; 3062 } 3063 3064 mutex_exit(&ldcp->lock); 3065 3066 /* 3067 * Increment number of open channels 3068 */ 3069 mutex_enter(&ldcssp->lock); 3070 ldcssp->channels_open++; 3071 mutex_exit(&ldcssp->lock); 3072 3073 D1(ldcp->id, 3074 "ldc_open: (0x%llx) channel (0x%p) open for use " 3075 "(tstate=0x%x, status=0x%x)\n", 3076 ldcp->id, ldcp, ldcp->tstate, ldcp->status); 3077 3078 return (0); 3079 } 3080 3081 /* 3082 * Close the LDC connection. It will return EBUSY if there 3083 * are memory segments or descriptor rings either bound to or 3084 * mapped over the channel 3085 */ 3086 int 3087 ldc_close(ldc_handle_t handle) 3088 { 3089 ldc_chan_t *ldcp; 3090 int rv = 0, retries = 0; 3091 boolean_t chk_done = B_FALSE; 3092 3093 if (handle == NULL) { 3094 DWARN(DBG_ALL_LDCS, "ldc_close: invalid channel handle\n"); 3095 return (EINVAL); 3096 } 3097 ldcp = (ldc_chan_t *)handle; 3098 3099 mutex_enter(&ldcp->lock); 3100 3101 /* return error if channel is not open */ 3102 if ((ldcp->tstate & ~TS_IN_RESET) < TS_OPEN) { 3103 DWARN(ldcp->id, 3104 "ldc_close: (0x%llx) channel is not open\n", ldcp->id); 3105 mutex_exit(&ldcp->lock); 3106 return (EFAULT); 3107 } 3108 3109 /* if any memory handles, drings, are bound or mapped cannot close */ 3110 if (ldcp->mhdl_list != NULL) { 3111 DWARN(ldcp->id, 3112 "ldc_close: (0x%llx) channel has bound memory handles\n", 3113 ldcp->id); 3114 mutex_exit(&ldcp->lock); 3115 return (EBUSY); 3116 } 3117 if (ldcp->exp_dring_list != NULL) { 3118 DWARN(ldcp->id, 3119 "ldc_close: (0x%llx) channel has bound descriptor rings\n", 3120 ldcp->id); 3121 mutex_exit(&ldcp->lock); 3122 return (EBUSY); 3123 } 3124 if (ldcp->imp_dring_list != NULL) { 3125 DWARN(ldcp->id, 3126 "ldc_close: (0x%llx) channel has mapped descriptor rings\n", 3127 ldcp->id); 3128 mutex_exit(&ldcp->lock); 3129 return (EBUSY); 3130 } 3131 3132 if (ldcp->cb_inprogress) { 3133 DWARN(ldcp->id, "ldc_close: (0x%llx) callback active\n", 3134 ldcp->id); 3135 mutex_exit(&ldcp->lock); 3136 return (EWOULDBLOCK); 3137 } 3138 3139 /* Obtain Tx lock */ 3140 mutex_enter(&ldcp->tx_lock); 3141 3142 /* 3143 * Wait for pending transmits to complete i.e Tx queue to drain 3144 * if there are pending pkts - wait 1 ms and retry again 3145 */ 3146 for (;;) { 3147 3148 rv = hv_ldc_tx_get_state(ldcp->id, 3149 &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state); 3150 if (rv) { 3151 cmn_err(CE_WARN, 3152 "ldc_close: (0x%lx) cannot read qptrs\n", ldcp->id); 3153 mutex_exit(&ldcp->tx_lock); 3154 mutex_exit(&ldcp->lock); 3155 return (EIO); 3156 } 3157 3158 if (ldcp->tx_head == ldcp->tx_tail || 3159 ldcp->link_state != LDC_CHANNEL_UP) { 3160 break; 3161 } 3162 3163 if (chk_done) { 3164 DWARN(ldcp->id, 3165 "ldc_close: (0x%llx) Tx queue drain timeout\n", 3166 ldcp->id); 3167 break; 3168 } 3169 3170 /* wait for one ms and try again */ 3171 delay(drv_usectohz(1000)); 3172 chk_done = B_TRUE; 3173 } 3174 3175 /* 3176 * Drain the Tx and Rx queues as we are closing the 3177 * channel. We dont care about any pending packets. 3178 * We have to also drain the queue prior to clearing 3179 * pending interrupts, otherwise the HV will trigger 3180 * an interrupt the moment the interrupt state is 3181 * cleared. 3182 */ 3183 (void) i_ldc_txq_reconf(ldcp); 3184 i_ldc_rxq_drain(ldcp); 3185 3186 /* 3187 * Unregister the channel with the nexus 3188 */ 3189 while ((rv = i_ldc_unregister_channel(ldcp)) != 0) { 3190 3191 mutex_exit(&ldcp->tx_lock); 3192 mutex_exit(&ldcp->lock); 3193 3194 /* if any error other than EAGAIN return back */ 3195 if (rv != EAGAIN || retries >= ldc_max_retries) { 3196 cmn_err(CE_WARN, 3197 "ldc_close: (0x%lx) unregister failed, %d\n", 3198 ldcp->id, rv); 3199 return (rv); 3200 } 3201 3202 /* 3203 * As there could be pending interrupts we need 3204 * to wait and try again 3205 */ 3206 drv_usecwait(ldc_close_delay); 3207 mutex_enter(&ldcp->lock); 3208 mutex_enter(&ldcp->tx_lock); 3209 retries++; 3210 } 3211 3212 ldcp->tstate &= ~TS_QCONF_RDY; 3213 3214 /* 3215 * Unregister queues 3216 */ 3217 rv = hv_ldc_tx_qconf(ldcp->id, NULL, NULL); 3218 if (rv) { 3219 cmn_err(CE_WARN, 3220 "ldc_close: (0x%lx) channel TX queue unconf failed\n", 3221 ldcp->id); 3222 mutex_exit(&ldcp->tx_lock); 3223 mutex_exit(&ldcp->lock); 3224 return (EIO); 3225 } 3226 rv = hv_ldc_rx_qconf(ldcp->id, NULL, NULL); 3227 if (rv) { 3228 cmn_err(CE_WARN, 3229 "ldc_close: (0x%lx) channel RX queue unconf failed\n", 3230 ldcp->id); 3231 mutex_exit(&ldcp->tx_lock); 3232 mutex_exit(&ldcp->lock); 3233 return (EIO); 3234 } 3235 3236 /* Reset channel state information */ 3237 i_ldc_reset_state(ldcp); 3238 3239 /* Mark channel as down and in initialized state */ 3240 ldcp->tx_ackd_head = 0; 3241 ldcp->tx_head = 0; 3242 ldcp->tstate = TS_IN_RESET|TS_INIT; 3243 ldcp->status = LDC_INIT; 3244 3245 mutex_exit(&ldcp->tx_lock); 3246 mutex_exit(&ldcp->lock); 3247 3248 /* Decrement number of open channels */ 3249 mutex_enter(&ldcssp->lock); 3250 ldcssp->channels_open--; 3251 mutex_exit(&ldcssp->lock); 3252 3253 D1(ldcp->id, "ldc_close: (0x%llx) channel closed\n", ldcp->id); 3254 3255 return (0); 3256 } 3257 3258 /* 3259 * Register channel callback 3260 */ 3261 int 3262 ldc_reg_callback(ldc_handle_t handle, 3263 uint_t(*cb)(uint64_t event, caddr_t arg), caddr_t arg) 3264 { 3265 ldc_chan_t *ldcp; 3266 3267 if (handle == NULL) { 3268 DWARN(DBG_ALL_LDCS, 3269 "ldc_reg_callback: invalid channel handle\n"); 3270 return (EINVAL); 3271 } 3272 if (((uint64_t)cb) < KERNELBASE) { 3273 DWARN(DBG_ALL_LDCS, "ldc_reg_callback: invalid callback\n"); 3274 return (EINVAL); 3275 } 3276 ldcp = (ldc_chan_t *)handle; 3277 3278 mutex_enter(&ldcp->lock); 3279 3280 if (ldcp->cb) { 3281 DWARN(ldcp->id, "ldc_reg_callback: (0x%llx) callback exists\n", 3282 ldcp->id); 3283 mutex_exit(&ldcp->lock); 3284 return (EIO); 3285 } 3286 if (ldcp->cb_inprogress) { 3287 DWARN(ldcp->id, "ldc_reg_callback: (0x%llx) callback active\n", 3288 ldcp->id); 3289 mutex_exit(&ldcp->lock); 3290 return (EWOULDBLOCK); 3291 } 3292 3293 ldcp->cb = cb; 3294 ldcp->cb_arg = arg; 3295 ldcp->cb_enabled = B_TRUE; 3296 3297 D1(ldcp->id, 3298 "ldc_reg_callback: (0x%llx) registered callback for channel\n", 3299 ldcp->id); 3300 3301 mutex_exit(&ldcp->lock); 3302 3303 return (0); 3304 } 3305 3306 /* 3307 * Unregister channel callback 3308 */ 3309 int 3310 ldc_unreg_callback(ldc_handle_t handle) 3311 { 3312 ldc_chan_t *ldcp; 3313 3314 if (handle == NULL) { 3315 DWARN(DBG_ALL_LDCS, 3316 "ldc_unreg_callback: invalid channel handle\n"); 3317 return (EINVAL); 3318 } 3319 ldcp = (ldc_chan_t *)handle; 3320 3321 mutex_enter(&ldcp->lock); 3322 3323 if (ldcp->cb == NULL) { 3324 DWARN(ldcp->id, 3325 "ldc_unreg_callback: (0x%llx) no callback exists\n", 3326 ldcp->id); 3327 mutex_exit(&ldcp->lock); 3328 return (EIO); 3329 } 3330 if (ldcp->cb_inprogress) { 3331 DWARN(ldcp->id, 3332 "ldc_unreg_callback: (0x%llx) callback active\n", 3333 ldcp->id); 3334 mutex_exit(&ldcp->lock); 3335 return (EWOULDBLOCK); 3336 } 3337 3338 ldcp->cb = NULL; 3339 ldcp->cb_arg = NULL; 3340 ldcp->cb_enabled = B_FALSE; 3341 3342 D1(ldcp->id, 3343 "ldc_unreg_callback: (0x%llx) unregistered callback for channel\n", 3344 ldcp->id); 3345 3346 mutex_exit(&ldcp->lock); 3347 3348 return (0); 3349 } 3350 3351 3352 /* 3353 * Bring a channel up by initiating a handshake with the peer 3354 * This call is asynchronous. It will complete at a later point 3355 * in time when the peer responds back with an RTR. 3356 */ 3357 int 3358 ldc_up(ldc_handle_t handle) 3359 { 3360 int rv; 3361 ldc_chan_t *ldcp; 3362 ldc_msg_t *ldcmsg; 3363 uint64_t tx_tail, tstate, link_state; 3364 3365 if (handle == NULL) { 3366 DWARN(DBG_ALL_LDCS, "ldc_up: invalid channel handle\n"); 3367 return (EINVAL); 3368 } 3369 ldcp = (ldc_chan_t *)handle; 3370 3371 mutex_enter(&ldcp->lock); 3372 3373 D1(ldcp->id, "ldc_up: (0x%llx) doing channel UP\n", ldcp->id); 3374 3375 /* clear the reset state */ 3376 tstate = ldcp->tstate; 3377 ldcp->tstate &= ~TS_IN_RESET; 3378 3379 if (ldcp->tstate == TS_UP) { 3380 DWARN(ldcp->id, 3381 "ldc_up: (0x%llx) channel is already in UP state\n", 3382 ldcp->id); 3383 3384 /* mark channel as up */ 3385 ldcp->status = LDC_UP; 3386 3387 /* 3388 * if channel was in reset state and there was 3389 * pending data clear interrupt state. this will 3390 * trigger an interrupt, causing the RX handler to 3391 * to invoke the client's callback 3392 */ 3393 if ((tstate & TS_IN_RESET) && 3394 ldcp->rx_intr_state == LDC_INTR_PEND) { 3395 D1(ldcp->id, 3396 "ldc_up: (0x%llx) channel has pending data, " 3397 "clearing interrupt\n", ldcp->id); 3398 i_ldc_clear_intr(ldcp, CNEX_RX_INTR); 3399 } 3400 3401 mutex_exit(&ldcp->lock); 3402 return (0); 3403 } 3404 3405 /* if the channel is in RAW mode - mark it as UP, if READY */ 3406 if (ldcp->mode == LDC_MODE_RAW && ldcp->tstate >= TS_READY) { 3407 ldcp->tstate = TS_UP; 3408 mutex_exit(&ldcp->lock); 3409 return (0); 3410 } 3411 3412 /* Don't start another handshake if there is one in progress */ 3413 if (ldcp->hstate) { 3414 D1(ldcp->id, 3415 "ldc_up: (0x%llx) channel handshake in progress\n", 3416 ldcp->id); 3417 mutex_exit(&ldcp->lock); 3418 return (0); 3419 } 3420 3421 mutex_enter(&ldcp->tx_lock); 3422 3423 /* save current link state */ 3424 link_state = ldcp->link_state; 3425 3426 /* get the current tail for the LDC msg */ 3427 rv = i_ldc_get_tx_tail(ldcp, &tx_tail); 3428 if (rv) { 3429 D1(ldcp->id, "ldc_up: (0x%llx) cannot initiate handshake\n", 3430 ldcp->id); 3431 mutex_exit(&ldcp->tx_lock); 3432 mutex_exit(&ldcp->lock); 3433 return (ECONNREFUSED); 3434 } 3435 3436 /* 3437 * If i_ldc_get_tx_tail() changed link_state to either RESET or UP, 3438 * from a previous state of DOWN, then mark the channel as 3439 * being ready for handshake. 3440 */ 3441 if ((link_state == LDC_CHANNEL_DOWN) && 3442 (link_state != ldcp->link_state)) { 3443 3444 ASSERT((ldcp->link_state == LDC_CHANNEL_RESET) || 3445 (ldcp->link_state == LDC_CHANNEL_UP)); 3446 3447 if (ldcp->mode == LDC_MODE_RAW) { 3448 ldcp->status = LDC_UP; 3449 ldcp->tstate = TS_UP; 3450 mutex_exit(&ldcp->tx_lock); 3451 mutex_exit(&ldcp->lock); 3452 return (0); 3453 } else { 3454 ldcp->status = LDC_READY; 3455 ldcp->tstate |= TS_LINK_READY; 3456 } 3457 3458 } 3459 3460 ldcmsg = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail); 3461 ZERO_PKT(ldcmsg); 3462 3463 ldcmsg->type = LDC_CTRL; 3464 ldcmsg->stype = LDC_INFO; 3465 ldcmsg->ctrl = LDC_VER; 3466 ldcp->next_vidx = 0; 3467 bcopy(&ldc_versions[0], ldcmsg->udata, sizeof (ldc_versions[0])); 3468 3469 DUMP_LDC_PKT(ldcp, "ldc_up snd ver", (uint64_t)ldcmsg); 3470 3471 /* initiate the send by calling into HV and set the new tail */ 3472 tx_tail = (tx_tail + LDC_PACKET_SIZE) % 3473 (ldcp->tx_q_entries << LDC_PACKET_SHIFT); 3474 3475 rv = i_ldc_set_tx_tail(ldcp, tx_tail); 3476 if (rv) { 3477 DWARN(ldcp->id, 3478 "ldc_up: (0x%llx) cannot initiate handshake rv=%d\n", 3479 ldcp->id, rv); 3480 mutex_exit(&ldcp->tx_lock); 3481 mutex_exit(&ldcp->lock); 3482 return (rv); 3483 } 3484 3485 ldcp->hstate |= TS_SENT_VER; 3486 ldcp->tx_tail = tx_tail; 3487 D1(ldcp->id, "ldc_up: (0x%llx) channel up initiated\n", ldcp->id); 3488 3489 mutex_exit(&ldcp->tx_lock); 3490 mutex_exit(&ldcp->lock); 3491 3492 return (rv); 3493 } 3494 3495 3496 /* 3497 * Bring a channel down by resetting its state and queues 3498 */ 3499 int 3500 ldc_down(ldc_handle_t handle) 3501 { 3502 ldc_chan_t *ldcp; 3503 3504 if (handle == NULL) { 3505 DWARN(DBG_ALL_LDCS, "ldc_down: invalid channel handle\n"); 3506 return (EINVAL); 3507 } 3508 ldcp = (ldc_chan_t *)handle; 3509 mutex_enter(&ldcp->lock); 3510 mutex_enter(&ldcp->tx_lock); 3511 i_ldc_reset(ldcp, B_TRUE); 3512 mutex_exit(&ldcp->tx_lock); 3513 mutex_exit(&ldcp->lock); 3514 3515 return (0); 3516 } 3517 3518 /* 3519 * Get the current channel status 3520 */ 3521 int 3522 ldc_status(ldc_handle_t handle, ldc_status_t *status) 3523 { 3524 ldc_chan_t *ldcp; 3525 3526 if (handle == NULL || status == NULL) { 3527 DWARN(DBG_ALL_LDCS, "ldc_status: invalid argument\n"); 3528 return (EINVAL); 3529 } 3530 ldcp = (ldc_chan_t *)handle; 3531 3532 *status = ((ldc_chan_t *)handle)->status; 3533 3534 D1(ldcp->id, 3535 "ldc_status: (0x%llx) returned status %d\n", ldcp->id, *status); 3536 return (0); 3537 } 3538 3539 3540 /* 3541 * Set the channel's callback mode - enable/disable callbacks 3542 */ 3543 int 3544 ldc_set_cb_mode(ldc_handle_t handle, ldc_cb_mode_t cmode) 3545 { 3546 ldc_chan_t *ldcp; 3547 3548 if (handle == NULL) { 3549 DWARN(DBG_ALL_LDCS, 3550 "ldc_set_intr_mode: invalid channel handle\n"); 3551 return (EINVAL); 3552 } 3553 ldcp = (ldc_chan_t *)handle; 3554 3555 /* 3556 * Record no callbacks should be invoked 3557 */ 3558 mutex_enter(&ldcp->lock); 3559 3560 switch (cmode) { 3561 case LDC_CB_DISABLE: 3562 if (!ldcp->cb_enabled) { 3563 DWARN(ldcp->id, 3564 "ldc_set_cb_mode: (0x%llx) callbacks disabled\n", 3565 ldcp->id); 3566 break; 3567 } 3568 ldcp->cb_enabled = B_FALSE; 3569 3570 D1(ldcp->id, "ldc_set_cb_mode: (0x%llx) disabled callbacks\n", 3571 ldcp->id); 3572 break; 3573 3574 case LDC_CB_ENABLE: 3575 if (ldcp->cb_enabled) { 3576 DWARN(ldcp->id, 3577 "ldc_set_cb_mode: (0x%llx) callbacks enabled\n", 3578 ldcp->id); 3579 break; 3580 } 3581 ldcp->cb_enabled = B_TRUE; 3582 3583 D1(ldcp->id, "ldc_set_cb_mode: (0x%llx) enabled callbacks\n", 3584 ldcp->id); 3585 break; 3586 } 3587 3588 mutex_exit(&ldcp->lock); 3589 3590 return (0); 3591 } 3592 3593 /* 3594 * Check to see if there are packets on the incoming queue 3595 * Will return hasdata = B_FALSE if there are no packets 3596 */ 3597 int 3598 ldc_chkq(ldc_handle_t handle, boolean_t *hasdata) 3599 { 3600 int rv; 3601 uint64_t rx_head, rx_tail; 3602 ldc_chan_t *ldcp; 3603 3604 if (handle == NULL) { 3605 DWARN(DBG_ALL_LDCS, "ldc_chkq: invalid channel handle\n"); 3606 return (EINVAL); 3607 } 3608 ldcp = (ldc_chan_t *)handle; 3609 3610 *hasdata = B_FALSE; 3611 3612 mutex_enter(&ldcp->lock); 3613 3614 if (ldcp->tstate != TS_UP) { 3615 D1(ldcp->id, 3616 "ldc_chkq: (0x%llx) channel is not up\n", ldcp->id); 3617 mutex_exit(&ldcp->lock); 3618 return (ECONNRESET); 3619 } 3620 3621 /* Read packet(s) from the queue */ 3622 rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail, 3623 &ldcp->link_state); 3624 if (rv != 0) { 3625 cmn_err(CE_WARN, 3626 "ldc_chkq: (0x%lx) unable to read queue ptrs", ldcp->id); 3627 mutex_exit(&ldcp->lock); 3628 return (EIO); 3629 } 3630 3631 /* reset the channel state if the channel went down */ 3632 if (ldcp->link_state == LDC_CHANNEL_DOWN || 3633 ldcp->link_state == LDC_CHANNEL_RESET) { 3634 mutex_enter(&ldcp->tx_lock); 3635 i_ldc_reset(ldcp, B_FALSE); 3636 mutex_exit(&ldcp->tx_lock); 3637 mutex_exit(&ldcp->lock); 3638 return (ECONNRESET); 3639 } 3640 3641 switch (ldcp->mode) { 3642 case LDC_MODE_RAW: 3643 /* 3644 * In raw mode, there are no ctrl packets, so checking 3645 * if the queue is non-empty is sufficient. 3646 */ 3647 *hasdata = (rx_head != rx_tail); 3648 break; 3649 3650 case LDC_MODE_UNRELIABLE: 3651 /* 3652 * In unreliable mode, if the queue is non-empty, we need 3653 * to check if it actually contains unread data packets. 3654 * The queue may just contain ctrl packets. 3655 */ 3656 if (rx_head != rx_tail) { 3657 *hasdata = (i_ldc_chkq(ldcp) == 0); 3658 /* 3659 * If no data packets were found on the queue, 3660 * all packets must have been control packets 3661 * which will now have been processed, leaving 3662 * the queue empty. If the interrupt state 3663 * is pending, we need to clear the interrupt 3664 * here. 3665 */ 3666 if (*hasdata == B_FALSE && 3667 ldcp->rx_intr_state == LDC_INTR_PEND) { 3668 i_ldc_clear_intr(ldcp, CNEX_RX_INTR); 3669 } 3670 } 3671 break; 3672 3673 case LDC_MODE_RELIABLE: 3674 /* 3675 * In reliable mode, first check for 'stream_remains' > 0. 3676 * Otherwise, if the data queue head and tail pointers 3677 * differ, there must be data to read. 3678 */ 3679 if (ldcp->stream_remains > 0) 3680 *hasdata = B_TRUE; 3681 else 3682 *hasdata = (ldcp->rx_dq_head != ldcp->rx_dq_tail); 3683 break; 3684 3685 default: 3686 cmn_err(CE_WARN, "ldc_chkq: (0x%lx) unexpected channel mode " 3687 "(0x%x)", ldcp->id, ldcp->mode); 3688 mutex_exit(&ldcp->lock); 3689 return (EIO); 3690 } 3691 3692 mutex_exit(&ldcp->lock); 3693 3694 return (0); 3695 } 3696 3697 3698 /* 3699 * Read 'size' amount of bytes or less. If incoming buffer 3700 * is more than 'size', ENOBUFS is returned. 3701 * 3702 * On return, size contains the number of bytes read. 3703 */ 3704 int 3705 ldc_read(ldc_handle_t handle, caddr_t bufp, size_t *sizep) 3706 { 3707 ldc_chan_t *ldcp; 3708 uint64_t rx_head = 0, rx_tail = 0; 3709 int rv = 0, exit_val; 3710 3711 if (handle == NULL) { 3712 DWARN(DBG_ALL_LDCS, "ldc_read: invalid channel handle\n"); 3713 return (EINVAL); 3714 } 3715 3716 ldcp = (ldc_chan_t *)handle; 3717 3718 /* channel lock */ 3719 mutex_enter(&ldcp->lock); 3720 3721 if (ldcp->tstate != TS_UP) { 3722 DWARN(ldcp->id, 3723 "ldc_read: (0x%llx) channel is not in UP state\n", 3724 ldcp->id); 3725 exit_val = ECONNRESET; 3726 } else if (ldcp->mode == LDC_MODE_RELIABLE) { 3727 TRACE_RXDQ_LENGTH(ldcp); 3728 exit_val = ldcp->read_p(ldcp, bufp, sizep); 3729 3730 /* 3731 * For reliable mode channels, the interrupt 3732 * state is only set to pending during 3733 * interrupt handling when the secondary data 3734 * queue became full, leaving unprocessed 3735 * packets on the Rx queue. If the interrupt 3736 * state is pending and space is now available 3737 * on the data queue, clear the interrupt. 3738 */ 3739 if (ldcp->rx_intr_state == LDC_INTR_PEND && 3740 Q_CONTIG_SPACE(ldcp->rx_dq_head, ldcp->rx_dq_tail, 3741 ldcp->rx_dq_entries << LDC_PACKET_SHIFT) >= 3742 LDC_PACKET_SIZE) { 3743 /* data queue is not full */ 3744 i_ldc_clear_intr(ldcp, CNEX_RX_INTR); 3745 } 3746 3747 mutex_exit(&ldcp->lock); 3748 return (exit_val); 3749 } else { 3750 exit_val = ldcp->read_p(ldcp, bufp, sizep); 3751 } 3752 3753 /* 3754 * if queue has been drained - clear interrupt 3755 */ 3756 rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail, 3757 &ldcp->link_state); 3758 if (rv != 0) { 3759 cmn_err(CE_WARN, "ldc_read: (0x%lx) unable to read queue ptrs", 3760 ldcp->id); 3761 mutex_enter(&ldcp->tx_lock); 3762 i_ldc_reset(ldcp, B_TRUE); 3763 mutex_exit(&ldcp->tx_lock); 3764 mutex_exit(&ldcp->lock); 3765 return (ECONNRESET); 3766 } 3767 3768 if (exit_val == 0) { 3769 if (ldcp->link_state == LDC_CHANNEL_DOWN || 3770 ldcp->link_state == LDC_CHANNEL_RESET) { 3771 mutex_enter(&ldcp->tx_lock); 3772 i_ldc_reset(ldcp, B_FALSE); 3773 exit_val = ECONNRESET; 3774 mutex_exit(&ldcp->tx_lock); 3775 } 3776 if ((rv == 0) && 3777 (ldcp->rx_intr_state == LDC_INTR_PEND) && 3778 (rx_head == rx_tail)) { 3779 i_ldc_clear_intr(ldcp, CNEX_RX_INTR); 3780 } 3781 } 3782 3783 mutex_exit(&ldcp->lock); 3784 return (exit_val); 3785 } 3786 3787 /* 3788 * Basic raw mondo read - 3789 * no interpretation of mondo contents at all. 3790 * 3791 * Enter and exit with ldcp->lock held by caller 3792 */ 3793 static int 3794 i_ldc_read_raw(ldc_chan_t *ldcp, caddr_t target_bufp, size_t *sizep) 3795 { 3796 uint64_t q_size_mask; 3797 ldc_msg_t *msgp; 3798 uint8_t *msgbufp; 3799 int rv = 0, space; 3800 uint64_t rx_head, rx_tail; 3801 3802 space = *sizep; 3803 3804 if (space < LDC_PAYLOAD_SIZE_RAW) 3805 return (ENOBUFS); 3806 3807 ASSERT(mutex_owned(&ldcp->lock)); 3808 3809 /* compute mask for increment */ 3810 q_size_mask = (ldcp->rx_q_entries-1)<<LDC_PACKET_SHIFT; 3811 3812 /* 3813 * Read packet(s) from the queue 3814 */ 3815 rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail, 3816 &ldcp->link_state); 3817 if (rv != 0) { 3818 cmn_err(CE_WARN, 3819 "ldc_read_raw: (0x%lx) unable to read queue ptrs", 3820 ldcp->id); 3821 return (EIO); 3822 } 3823 D1(ldcp->id, "ldc_read_raw: (0x%llx) rxh=0x%llx," 3824 " rxt=0x%llx, st=0x%llx\n", 3825 ldcp->id, rx_head, rx_tail, ldcp->link_state); 3826 3827 /* reset the channel state if the channel went down */ 3828 if (ldcp->link_state == LDC_CHANNEL_DOWN || 3829 ldcp->link_state == LDC_CHANNEL_RESET) { 3830 mutex_enter(&ldcp->tx_lock); 3831 i_ldc_reset(ldcp, B_FALSE); 3832 mutex_exit(&ldcp->tx_lock); 3833 return (ECONNRESET); 3834 } 3835 3836 /* 3837 * Check for empty queue 3838 */ 3839 if (rx_head == rx_tail) { 3840 *sizep = 0; 3841 return (0); 3842 } 3843 3844 /* get the message */ 3845 msgp = (ldc_msg_t *)(ldcp->rx_q_va + rx_head); 3846 3847 /* if channel is in RAW mode, copy data and return */ 3848 msgbufp = (uint8_t *)&(msgp->raw[0]); 3849 3850 bcopy(msgbufp, target_bufp, LDC_PAYLOAD_SIZE_RAW); 3851 3852 DUMP_PAYLOAD(ldcp->id, msgbufp); 3853 3854 *sizep = LDC_PAYLOAD_SIZE_RAW; 3855 3856 rx_head = (rx_head + LDC_PACKET_SIZE) & q_size_mask; 3857 rv = i_ldc_set_rx_head(ldcp, rx_head); 3858 3859 return (rv); 3860 } 3861 3862 /* 3863 * Process LDC mondos to build larger packets 3864 * with either un-reliable or reliable delivery. 3865 * 3866 * Enter and exit with ldcp->lock held by caller 3867 */ 3868 static int 3869 i_ldc_read_packet(ldc_chan_t *ldcp, caddr_t target_bufp, size_t *sizep) 3870 { 3871 int rv = 0; 3872 uint64_t rx_head = 0, rx_tail = 0; 3873 uint64_t curr_head = 0; 3874 ldc_msg_t *msg; 3875 caddr_t target; 3876 size_t len = 0, bytes_read = 0; 3877 int retries = 0; 3878 uint64_t q_va, q_size_mask; 3879 uint64_t first_fragment = 0; 3880 3881 target = target_bufp; 3882 3883 ASSERT(mutex_owned(&ldcp->lock)); 3884 3885 /* check if the buffer and size are valid */ 3886 if (target_bufp == NULL || *sizep == 0) { 3887 DWARN(ldcp->id, "ldc_read: (0x%llx) invalid buffer/size\n", 3888 ldcp->id); 3889 return (EINVAL); 3890 } 3891 3892 /* Set q_va and compute increment mask for the appropriate queue */ 3893 if (ldcp->mode == LDC_MODE_RELIABLE) { 3894 q_va = ldcp->rx_dq_va; 3895 q_size_mask = (ldcp->rx_dq_entries-1)<<LDC_PACKET_SHIFT; 3896 } else { 3897 q_va = ldcp->rx_q_va; 3898 q_size_mask = (ldcp->rx_q_entries-1)<<LDC_PACKET_SHIFT; 3899 } 3900 3901 /* 3902 * Read packet(s) from the queue 3903 */ 3904 rv = ldcp->readq_get_state(ldcp, &curr_head, &rx_tail, 3905 &ldcp->link_state); 3906 if (rv != 0) { 3907 cmn_err(CE_WARN, "ldc_read: (0x%lx) unable to read queue ptrs", 3908 ldcp->id); 3909 mutex_enter(&ldcp->tx_lock); 3910 i_ldc_reset(ldcp, B_TRUE); 3911 mutex_exit(&ldcp->tx_lock); 3912 return (ECONNRESET); 3913 } 3914 D1(ldcp->id, "ldc_read: (0x%llx) chd=0x%llx, tl=0x%llx, st=0x%llx\n", 3915 ldcp->id, curr_head, rx_tail, ldcp->link_state); 3916 3917 /* reset the channel state if the channel went down */ 3918 if (ldcp->link_state != LDC_CHANNEL_UP) 3919 goto channel_is_reset; 3920 3921 for (;;) { 3922 3923 if (curr_head == rx_tail) { 3924 /* 3925 * If a data queue is being used, check the Rx HV 3926 * queue. This will copy over any new data packets 3927 * that have arrived. 3928 */ 3929 if (ldcp->mode == LDC_MODE_RELIABLE) 3930 (void) i_ldc_chkq(ldcp); 3931 3932 rv = ldcp->readq_get_state(ldcp, 3933 &rx_head, &rx_tail, &ldcp->link_state); 3934 if (rv != 0) { 3935 cmn_err(CE_WARN, 3936 "ldc_read: (0x%lx) cannot read queue ptrs", 3937 ldcp->id); 3938 mutex_enter(&ldcp->tx_lock); 3939 i_ldc_reset(ldcp, B_TRUE); 3940 mutex_exit(&ldcp->tx_lock); 3941 return (ECONNRESET); 3942 } 3943 3944 if (ldcp->link_state != LDC_CHANNEL_UP) 3945 goto channel_is_reset; 3946 3947 if (curr_head == rx_tail) { 3948 3949 /* If in the middle of a fragmented xfer */ 3950 if (first_fragment != 0) { 3951 3952 /* wait for ldc_delay usecs */ 3953 drv_usecwait(ldc_delay); 3954 3955 if (++retries < ldc_max_retries) 3956 continue; 3957 3958 *sizep = 0; 3959 if (ldcp->mode != LDC_MODE_RELIABLE) 3960 ldcp->last_msg_rcd = 3961 first_fragment - 1; 3962 DWARN(DBG_ALL_LDCS, "ldc_read: " 3963 "(0x%llx) read timeout", ldcp->id); 3964 return (EAGAIN); 3965 } 3966 *sizep = 0; 3967 break; 3968 } 3969 } 3970 retries = 0; 3971 3972 D2(ldcp->id, 3973 "ldc_read: (0x%llx) chd=0x%llx, rxhd=0x%llx, rxtl=0x%llx\n", 3974 ldcp->id, curr_head, rx_head, rx_tail); 3975 3976 /* get the message */ 3977 msg = (ldc_msg_t *)(q_va + curr_head); 3978 3979 DUMP_LDC_PKT(ldcp, "ldc_read received pkt", 3980 ldcp->rx_q_va + curr_head); 3981 3982 /* Check the message ID for the message received */ 3983 if (ldcp->mode != LDC_MODE_RELIABLE) { 3984 if ((rv = i_ldc_check_seqid(ldcp, msg)) != 0) { 3985 3986 DWARN(ldcp->id, "ldc_read: (0x%llx) seqid " 3987 "error, q_ptrs=0x%lx,0x%lx", 3988 ldcp->id, rx_head, rx_tail); 3989 3990 /* throw away data */ 3991 bytes_read = 0; 3992 3993 /* Reset last_msg_rcd to start of message */ 3994 if (first_fragment != 0) { 3995 ldcp->last_msg_rcd = first_fragment - 1; 3996 first_fragment = 0; 3997 } 3998 /* 3999 * Send a NACK -- invalid seqid 4000 * get the current tail for the response 4001 */ 4002 rv = i_ldc_send_pkt(ldcp, msg->type, LDC_NACK, 4003 (msg->ctrl & LDC_CTRL_MASK)); 4004 if (rv) { 4005 cmn_err(CE_NOTE, 4006 "ldc_read: (0x%lx) err sending " 4007 "NACK msg\n", ldcp->id); 4008 4009 /* if cannot send NACK - reset chan */ 4010 mutex_enter(&ldcp->tx_lock); 4011 i_ldc_reset(ldcp, B_FALSE); 4012 mutex_exit(&ldcp->tx_lock); 4013 rv = ECONNRESET; 4014 break; 4015 } 4016 4017 /* purge receive queue */ 4018 rv = i_ldc_set_rx_head(ldcp, rx_tail); 4019 4020 break; 4021 } 4022 4023 /* 4024 * Process any messages of type CTRL messages 4025 * Future implementations should try to pass these 4026 * to LDC link by resetting the intr state. 4027 * 4028 * NOTE: not done as a switch() as type can be 4029 * both ctrl+data 4030 */ 4031 if (msg->type & LDC_CTRL) { 4032 if (rv = i_ldc_ctrlmsg(ldcp, msg)) { 4033 if (rv == EAGAIN) 4034 continue; 4035 rv = i_ldc_set_rx_head(ldcp, rx_tail); 4036 *sizep = 0; 4037 bytes_read = 0; 4038 break; 4039 } 4040 } 4041 4042 /* process data ACKs */ 4043 if ((msg->type & LDC_DATA) && (msg->stype & LDC_ACK)) { 4044 if (rv = i_ldc_process_data_ACK(ldcp, msg)) { 4045 *sizep = 0; 4046 bytes_read = 0; 4047 break; 4048 } 4049 } 4050 4051 /* process data NACKs */ 4052 if ((msg->type & LDC_DATA) && (msg->stype & LDC_NACK)) { 4053 DWARN(ldcp->id, 4054 "ldc_read: (0x%llx) received DATA/NACK", 4055 ldcp->id); 4056 mutex_enter(&ldcp->tx_lock); 4057 i_ldc_reset(ldcp, B_TRUE); 4058 mutex_exit(&ldcp->tx_lock); 4059 return (ECONNRESET); 4060 } 4061 } 4062 4063 /* process data messages */ 4064 if ((msg->type & LDC_DATA) && (msg->stype & LDC_INFO)) { 4065 4066 uint8_t *msgbuf = (uint8_t *)( 4067 (ldcp->mode == LDC_MODE_RELIABLE) ? 4068 msg->rdata : msg->udata); 4069 4070 D2(ldcp->id, 4071 "ldc_read: (0x%llx) received data msg\n", ldcp->id); 4072 4073 /* get the packet length */ 4074 len = (msg->env & LDC_LEN_MASK); 4075 4076 /* 4077 * FUTURE OPTIMIZATION: 4078 * dont need to set q head for every 4079 * packet we read just need to do this when 4080 * we are done or need to wait for more 4081 * mondos to make a full packet - this is 4082 * currently expensive. 4083 */ 4084 4085 if (first_fragment == 0) { 4086 4087 /* 4088 * first packets should always have the start 4089 * bit set (even for a single packet). If not 4090 * throw away the packet 4091 */ 4092 if (!(msg->env & LDC_FRAG_START)) { 4093 4094 DWARN(DBG_ALL_LDCS, 4095 "ldc_read: (0x%llx) not start - " 4096 "frag=%x\n", ldcp->id, 4097 (msg->env) & LDC_FRAG_MASK); 4098 4099 /* toss pkt, inc head, cont reading */ 4100 bytes_read = 0; 4101 target = target_bufp; 4102 curr_head = 4103 (curr_head + LDC_PACKET_SIZE) 4104 & q_size_mask; 4105 if (rv = ldcp->readq_set_head(ldcp, 4106 curr_head)) 4107 break; 4108 4109 continue; 4110 } 4111 4112 first_fragment = msg->seqid; 4113 } else { 4114 /* check to see if this is a pkt w/ START bit */ 4115 if (msg->env & LDC_FRAG_START) { 4116 DWARN(DBG_ALL_LDCS, 4117 "ldc_read:(0x%llx) unexpected pkt" 4118 " env=0x%x discarding %d bytes," 4119 " lastmsg=%d, currentmsg=%d\n", 4120 ldcp->id, msg->env&LDC_FRAG_MASK, 4121 bytes_read, ldcp->last_msg_rcd, 4122 msg->seqid); 4123 4124 /* throw data we have read so far */ 4125 bytes_read = 0; 4126 target = target_bufp; 4127 first_fragment = msg->seqid; 4128 4129 if (rv = ldcp->readq_set_head(ldcp, 4130 curr_head)) 4131 break; 4132 } 4133 } 4134 4135 /* copy (next) pkt into buffer */ 4136 if (len <= (*sizep - bytes_read)) { 4137 bcopy(msgbuf, target, len); 4138 target += len; 4139 bytes_read += len; 4140 } else { 4141 /* 4142 * there is not enough space in the buffer to 4143 * read this pkt. throw message away & continue 4144 * reading data from queue 4145 */ 4146 DWARN(DBG_ALL_LDCS, 4147 "ldc_read: (0x%llx) buffer too small, " 4148 "head=0x%lx, expect=%d, got=%d\n", ldcp->id, 4149 curr_head, *sizep, bytes_read+len); 4150 4151 first_fragment = 0; 4152 target = target_bufp; 4153 bytes_read = 0; 4154 4155 /* throw away everything received so far */ 4156 if (rv = ldcp->readq_set_head(ldcp, curr_head)) 4157 break; 4158 4159 /* continue reading remaining pkts */ 4160 continue; 4161 } 4162 } 4163 4164 /* set the message id */ 4165 if (ldcp->mode != LDC_MODE_RELIABLE) 4166 ldcp->last_msg_rcd = msg->seqid; 4167 4168 /* move the head one position */ 4169 curr_head = (curr_head + LDC_PACKET_SIZE) & q_size_mask; 4170 4171 if (msg->env & LDC_FRAG_STOP) { 4172 4173 /* 4174 * All pkts that are part of this fragmented transfer 4175 * have been read or this was a single pkt read 4176 * or there was an error 4177 */ 4178 4179 /* set the queue head */ 4180 if (rv = ldcp->readq_set_head(ldcp, curr_head)) 4181 bytes_read = 0; 4182 4183 *sizep = bytes_read; 4184 4185 break; 4186 } 4187 4188 /* advance head if it is a CTRL packet or a DATA ACK packet */ 4189 if ((msg->type & LDC_CTRL) || 4190 ((msg->type & LDC_DATA) && (msg->stype & LDC_ACK))) { 4191 4192 /* set the queue head */ 4193 if (rv = ldcp->readq_set_head(ldcp, curr_head)) { 4194 bytes_read = 0; 4195 break; 4196 } 4197 4198 D2(ldcp->id, "ldc_read: (0x%llx) set ACK qhead 0x%llx", 4199 ldcp->id, curr_head); 4200 } 4201 4202 } /* for (;;) */ 4203 4204 D2(ldcp->id, "ldc_read: (0x%llx) end size=%d", ldcp->id, *sizep); 4205 4206 return (rv); 4207 4208 channel_is_reset: 4209 mutex_enter(&ldcp->tx_lock); 4210 i_ldc_reset(ldcp, B_FALSE); 4211 mutex_exit(&ldcp->tx_lock); 4212 return (ECONNRESET); 4213 } 4214 4215 /* 4216 * Fetch and buffer incoming packets so we can hand them back as 4217 * a basic byte stream. 4218 * 4219 * Enter and exit with ldcp->lock held by caller 4220 */ 4221 static int 4222 i_ldc_read_stream(ldc_chan_t *ldcp, caddr_t target_bufp, size_t *sizep) 4223 { 4224 int rv; 4225 size_t size; 4226 4227 ASSERT(mutex_owned(&ldcp->lock)); 4228 4229 D2(ldcp->id, "i_ldc_read_stream: (0x%llx) buffer size=%d", 4230 ldcp->id, *sizep); 4231 4232 if (ldcp->stream_remains == 0) { 4233 size = ldcp->mtu; 4234 rv = i_ldc_read_packet(ldcp, 4235 (caddr_t)ldcp->stream_bufferp, &size); 4236 D2(ldcp->id, "i_ldc_read_stream: read packet (0x%llx) size=%d", 4237 ldcp->id, size); 4238 4239 if (rv != 0) 4240 return (rv); 4241 4242 ldcp->stream_remains = size; 4243 ldcp->stream_offset = 0; 4244 } 4245 4246 size = MIN(ldcp->stream_remains, *sizep); 4247 4248 bcopy(ldcp->stream_bufferp + ldcp->stream_offset, target_bufp, size); 4249 ldcp->stream_offset += size; 4250 ldcp->stream_remains -= size; 4251 4252 D2(ldcp->id, "i_ldc_read_stream: (0x%llx) fill from buffer size=%d", 4253 ldcp->id, size); 4254 4255 *sizep = size; 4256 return (0); 4257 } 4258 4259 /* 4260 * Write specified amount of bytes to the channel 4261 * in multiple pkts of pkt_payload size. Each 4262 * packet is tagged with an unique packet ID in 4263 * the case of a reliable link. 4264 * 4265 * On return, size contains the number of bytes written. 4266 */ 4267 int 4268 ldc_write(ldc_handle_t handle, caddr_t buf, size_t *sizep) 4269 { 4270 ldc_chan_t *ldcp; 4271 int rv = 0; 4272 4273 if (handle == NULL) { 4274 DWARN(DBG_ALL_LDCS, "ldc_write: invalid channel handle\n"); 4275 return (EINVAL); 4276 } 4277 ldcp = (ldc_chan_t *)handle; 4278 4279 /* check if writes can occur */ 4280 if (!mutex_tryenter(&ldcp->tx_lock)) { 4281 /* 4282 * Could not get the lock - channel could 4283 * be in the process of being unconfigured 4284 * or reader has encountered an error 4285 */ 4286 return (EAGAIN); 4287 } 4288 4289 /* check if non-zero data to write */ 4290 if (buf == NULL || sizep == NULL) { 4291 DWARN(ldcp->id, "ldc_write: (0x%llx) invalid data write\n", 4292 ldcp->id); 4293 mutex_exit(&ldcp->tx_lock); 4294 return (EINVAL); 4295 } 4296 4297 if (*sizep == 0) { 4298 DWARN(ldcp->id, "ldc_write: (0x%llx) write size of zero\n", 4299 ldcp->id); 4300 mutex_exit(&ldcp->tx_lock); 4301 return (0); 4302 } 4303 4304 /* Check if channel is UP for data exchange */ 4305 if (ldcp->tstate != TS_UP) { 4306 DWARN(ldcp->id, 4307 "ldc_write: (0x%llx) channel is not in UP state\n", 4308 ldcp->id); 4309 *sizep = 0; 4310 rv = ECONNRESET; 4311 } else { 4312 rv = ldcp->write_p(ldcp, buf, sizep); 4313 } 4314 4315 mutex_exit(&ldcp->tx_lock); 4316 4317 return (rv); 4318 } 4319 4320 /* 4321 * Write a raw packet to the channel 4322 * On return, size contains the number of bytes written. 4323 */ 4324 static int 4325 i_ldc_write_raw(ldc_chan_t *ldcp, caddr_t buf, size_t *sizep) 4326 { 4327 ldc_msg_t *ldcmsg; 4328 uint64_t tx_head, tx_tail, new_tail; 4329 int rv = 0; 4330 size_t size; 4331 4332 ASSERT(MUTEX_HELD(&ldcp->tx_lock)); 4333 ASSERT(ldcp->mode == LDC_MODE_RAW); 4334 4335 size = *sizep; 4336 4337 /* 4338 * Check to see if the packet size is less than or 4339 * equal to packet size support in raw mode 4340 */ 4341 if (size > ldcp->pkt_payload) { 4342 DWARN(ldcp->id, 4343 "ldc_write: (0x%llx) invalid size (0x%llx) for RAW mode\n", 4344 ldcp->id, *sizep); 4345 *sizep = 0; 4346 return (EMSGSIZE); 4347 } 4348 4349 /* get the qptrs for the tx queue */ 4350 rv = hv_ldc_tx_get_state(ldcp->id, 4351 &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state); 4352 if (rv != 0) { 4353 cmn_err(CE_WARN, 4354 "ldc_write: (0x%lx) cannot read queue ptrs\n", ldcp->id); 4355 *sizep = 0; 4356 return (EIO); 4357 } 4358 4359 if (ldcp->link_state == LDC_CHANNEL_DOWN || 4360 ldcp->link_state == LDC_CHANNEL_RESET) { 4361 DWARN(ldcp->id, 4362 "ldc_write: (0x%llx) channel down/reset\n", ldcp->id); 4363 4364 *sizep = 0; 4365 if (mutex_tryenter(&ldcp->lock)) { 4366 i_ldc_reset(ldcp, B_FALSE); 4367 mutex_exit(&ldcp->lock); 4368 } else { 4369 /* 4370 * Release Tx lock, and then reacquire channel 4371 * and Tx lock in correct order 4372 */ 4373 mutex_exit(&ldcp->tx_lock); 4374 mutex_enter(&ldcp->lock); 4375 mutex_enter(&ldcp->tx_lock); 4376 i_ldc_reset(ldcp, B_FALSE); 4377 mutex_exit(&ldcp->lock); 4378 } 4379 return (ECONNRESET); 4380 } 4381 4382 tx_tail = ldcp->tx_tail; 4383 tx_head = ldcp->tx_head; 4384 new_tail = (tx_tail + LDC_PACKET_SIZE) & 4385 ((ldcp->tx_q_entries-1) << LDC_PACKET_SHIFT); 4386 4387 if (new_tail == tx_head) { 4388 DWARN(DBG_ALL_LDCS, 4389 "ldc_write: (0x%llx) TX queue is full\n", ldcp->id); 4390 *sizep = 0; 4391 return (EWOULDBLOCK); 4392 } 4393 4394 D2(ldcp->id, "ldc_write: (0x%llx) start xfer size=%d", 4395 ldcp->id, size); 4396 4397 /* Send the data now */ 4398 ldcmsg = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail); 4399 4400 /* copy the data into pkt */ 4401 bcopy((uint8_t *)buf, ldcmsg, size); 4402 4403 /* increment tail */ 4404 tx_tail = new_tail; 4405 4406 /* 4407 * All packets have been copied into the TX queue 4408 * update the tail ptr in the HV 4409 */ 4410 rv = i_ldc_set_tx_tail(ldcp, tx_tail); 4411 if (rv) { 4412 if (rv == EWOULDBLOCK) { 4413 DWARN(ldcp->id, "ldc_write: (0x%llx) write timed out\n", 4414 ldcp->id); 4415 *sizep = 0; 4416 return (EWOULDBLOCK); 4417 } 4418 4419 *sizep = 0; 4420 if (mutex_tryenter(&ldcp->lock)) { 4421 i_ldc_reset(ldcp, B_FALSE); 4422 mutex_exit(&ldcp->lock); 4423 } else { 4424 /* 4425 * Release Tx lock, and then reacquire channel 4426 * and Tx lock in correct order 4427 */ 4428 mutex_exit(&ldcp->tx_lock); 4429 mutex_enter(&ldcp->lock); 4430 mutex_enter(&ldcp->tx_lock); 4431 i_ldc_reset(ldcp, B_FALSE); 4432 mutex_exit(&ldcp->lock); 4433 } 4434 return (ECONNRESET); 4435 } 4436 4437 ldcp->tx_tail = tx_tail; 4438 *sizep = size; 4439 4440 D2(ldcp->id, "ldc_write: (0x%llx) end xfer size=%d", ldcp->id, size); 4441 4442 return (rv); 4443 } 4444 4445 4446 /* 4447 * Write specified amount of bytes to the channel 4448 * in multiple pkts of pkt_payload size. Each 4449 * packet is tagged with an unique packet ID in 4450 * the case of a reliable link. 4451 * 4452 * On return, size contains the number of bytes written. 4453 * This function needs to ensure that the write size is < MTU size 4454 */ 4455 static int 4456 i_ldc_write_packet(ldc_chan_t *ldcp, caddr_t buf, size_t *size) 4457 { 4458 ldc_msg_t *ldcmsg; 4459 uint64_t tx_head, tx_tail, new_tail, start; 4460 uint64_t txq_size_mask, numavail; 4461 uint8_t *msgbuf, *source = (uint8_t *)buf; 4462 size_t len, bytes_written = 0, remaining; 4463 int rv; 4464 uint32_t curr_seqid; 4465 4466 ASSERT(MUTEX_HELD(&ldcp->tx_lock)); 4467 4468 ASSERT(ldcp->mode == LDC_MODE_RELIABLE || 4469 ldcp->mode == LDC_MODE_UNRELIABLE); 4470 4471 /* compute mask for increment */ 4472 txq_size_mask = (ldcp->tx_q_entries - 1) << LDC_PACKET_SHIFT; 4473 4474 /* get the qptrs for the tx queue */ 4475 rv = hv_ldc_tx_get_state(ldcp->id, 4476 &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state); 4477 if (rv != 0) { 4478 cmn_err(CE_WARN, 4479 "ldc_write: (0x%lx) cannot read queue ptrs\n", ldcp->id); 4480 *size = 0; 4481 return (EIO); 4482 } 4483 4484 if (ldcp->link_state == LDC_CHANNEL_DOWN || 4485 ldcp->link_state == LDC_CHANNEL_RESET) { 4486 DWARN(ldcp->id, 4487 "ldc_write: (0x%llx) channel down/reset\n", ldcp->id); 4488 *size = 0; 4489 if (mutex_tryenter(&ldcp->lock)) { 4490 i_ldc_reset(ldcp, B_FALSE); 4491 mutex_exit(&ldcp->lock); 4492 } else { 4493 /* 4494 * Release Tx lock, and then reacquire channel 4495 * and Tx lock in correct order 4496 */ 4497 mutex_exit(&ldcp->tx_lock); 4498 mutex_enter(&ldcp->lock); 4499 mutex_enter(&ldcp->tx_lock); 4500 i_ldc_reset(ldcp, B_FALSE); 4501 mutex_exit(&ldcp->lock); 4502 } 4503 return (ECONNRESET); 4504 } 4505 4506 tx_tail = ldcp->tx_tail; 4507 new_tail = (tx_tail + LDC_PACKET_SIZE) % 4508 (ldcp->tx_q_entries << LDC_PACKET_SHIFT); 4509 4510 /* 4511 * Check to see if the queue is full. The check is done using 4512 * the appropriate head based on the link mode. 4513 */ 4514 i_ldc_get_tx_head(ldcp, &tx_head); 4515 4516 if (new_tail == tx_head) { 4517 DWARN(DBG_ALL_LDCS, 4518 "ldc_write: (0x%llx) TX queue is full\n", ldcp->id); 4519 *size = 0; 4520 return (EWOULDBLOCK); 4521 } 4522 4523 /* 4524 * Make sure that the LDC Tx queue has enough space 4525 */ 4526 numavail = (tx_head >> LDC_PACKET_SHIFT) - (tx_tail >> LDC_PACKET_SHIFT) 4527 + ldcp->tx_q_entries - 1; 4528 numavail %= ldcp->tx_q_entries; 4529 4530 if (*size > (numavail * ldcp->pkt_payload)) { 4531 DWARN(DBG_ALL_LDCS, 4532 "ldc_write: (0x%llx) TX queue has no space\n", ldcp->id); 4533 return (EWOULDBLOCK); 4534 } 4535 4536 D2(ldcp->id, "ldc_write: (0x%llx) start xfer size=%d", 4537 ldcp->id, *size); 4538 4539 /* Send the data now */ 4540 bytes_written = 0; 4541 curr_seqid = ldcp->last_msg_snt; 4542 start = tx_tail; 4543 4544 while (*size > bytes_written) { 4545 4546 ldcmsg = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail); 4547 4548 msgbuf = (uint8_t *)((ldcp->mode == LDC_MODE_RELIABLE) ? 4549 ldcmsg->rdata : ldcmsg->udata); 4550 4551 ldcmsg->type = LDC_DATA; 4552 ldcmsg->stype = LDC_INFO; 4553 ldcmsg->ctrl = 0; 4554 4555 remaining = *size - bytes_written; 4556 len = min(ldcp->pkt_payload, remaining); 4557 ldcmsg->env = (uint8_t)len; 4558 4559 curr_seqid++; 4560 ldcmsg->seqid = curr_seqid; 4561 4562 /* copy the data into pkt */ 4563 bcopy(source, msgbuf, len); 4564 4565 source += len; 4566 bytes_written += len; 4567 4568 /* increment tail */ 4569 tx_tail = (tx_tail + LDC_PACKET_SIZE) & txq_size_mask; 4570 4571 ASSERT(tx_tail != tx_head); 4572 } 4573 4574 /* Set the start and stop bits */ 4575 ldcmsg->env |= LDC_FRAG_STOP; 4576 ldcmsg = (ldc_msg_t *)(ldcp->tx_q_va + start); 4577 ldcmsg->env |= LDC_FRAG_START; 4578 4579 /* 4580 * All packets have been copied into the TX queue 4581 * update the tail ptr in the HV 4582 */ 4583 rv = i_ldc_set_tx_tail(ldcp, tx_tail); 4584 if (rv == 0) { 4585 ldcp->tx_tail = tx_tail; 4586 ldcp->last_msg_snt = curr_seqid; 4587 *size = bytes_written; 4588 } else { 4589 int rv2; 4590 4591 if (rv != EWOULDBLOCK) { 4592 *size = 0; 4593 if (mutex_tryenter(&ldcp->lock)) { 4594 i_ldc_reset(ldcp, B_FALSE); 4595 mutex_exit(&ldcp->lock); 4596 } else { 4597 /* 4598 * Release Tx lock, and then reacquire channel 4599 * and Tx lock in correct order 4600 */ 4601 mutex_exit(&ldcp->tx_lock); 4602 mutex_enter(&ldcp->lock); 4603 mutex_enter(&ldcp->tx_lock); 4604 i_ldc_reset(ldcp, B_FALSE); 4605 mutex_exit(&ldcp->lock); 4606 } 4607 return (ECONNRESET); 4608 } 4609 4610 D1(ldcp->id, "hv_tx_set_tail returns 0x%x (head 0x%x, " 4611 "old tail 0x%x, new tail 0x%x, qsize=0x%x)\n", 4612 rv, ldcp->tx_head, ldcp->tx_tail, tx_tail, 4613 (ldcp->tx_q_entries << LDC_PACKET_SHIFT)); 4614 4615 rv2 = hv_ldc_tx_get_state(ldcp->id, 4616 &tx_head, &tx_tail, &ldcp->link_state); 4617 4618 D1(ldcp->id, "hv_ldc_tx_get_state returns 0x%x " 4619 "(head 0x%x, tail 0x%x state 0x%x)\n", 4620 rv2, tx_head, tx_tail, ldcp->link_state); 4621 4622 *size = 0; 4623 } 4624 4625 D2(ldcp->id, "ldc_write: (0x%llx) end xfer size=%d", ldcp->id, *size); 4626 4627 return (rv); 4628 } 4629 4630 /* 4631 * Write specified amount of bytes to the channel 4632 * in multiple pkts of pkt_payload size. Each 4633 * packet is tagged with an unique packet ID in 4634 * the case of a reliable link. 4635 * 4636 * On return, size contains the number of bytes written. 4637 * This function needs to ensure that the write size is < MTU size 4638 */ 4639 static int 4640 i_ldc_write_stream(ldc_chan_t *ldcp, caddr_t buf, size_t *sizep) 4641 { 4642 ASSERT(MUTEX_HELD(&ldcp->tx_lock)); 4643 ASSERT(ldcp->mode == LDC_MODE_RELIABLE); 4644 4645 /* Truncate packet to max of MTU size */ 4646 if (*sizep > ldcp->mtu) *sizep = ldcp->mtu; 4647 return (i_ldc_write_packet(ldcp, buf, sizep)); 4648 } 4649 4650 4651 /* 4652 * Interfaces for channel nexus to register/unregister with LDC module 4653 * The nexus will register functions to be used to register individual 4654 * channels with the nexus and enable interrupts for the channels 4655 */ 4656 int 4657 ldc_register(ldc_cnex_t *cinfo) 4658 { 4659 ldc_chan_t *ldcp; 4660 4661 if (cinfo == NULL || cinfo->dip == NULL || 4662 cinfo->reg_chan == NULL || cinfo->unreg_chan == NULL || 4663 cinfo->add_intr == NULL || cinfo->rem_intr == NULL || 4664 cinfo->clr_intr == NULL) { 4665 4666 DWARN(DBG_ALL_LDCS, "ldc_register: invalid nexus info\n"); 4667 return (EINVAL); 4668 } 4669 4670 mutex_enter(&ldcssp->lock); 4671 4672 /* nexus registration */ 4673 ldcssp->cinfo.dip = cinfo->dip; 4674 ldcssp->cinfo.reg_chan = cinfo->reg_chan; 4675 ldcssp->cinfo.unreg_chan = cinfo->unreg_chan; 4676 ldcssp->cinfo.add_intr = cinfo->add_intr; 4677 ldcssp->cinfo.rem_intr = cinfo->rem_intr; 4678 ldcssp->cinfo.clr_intr = cinfo->clr_intr; 4679 4680 /* register any channels that might have been previously initialized */ 4681 ldcp = ldcssp->chan_list; 4682 while (ldcp) { 4683 if ((ldcp->tstate & TS_QCONF_RDY) && 4684 (ldcp->tstate & TS_CNEX_RDY) == 0) 4685 (void) i_ldc_register_channel(ldcp); 4686 4687 ldcp = ldcp->next; 4688 } 4689 4690 mutex_exit(&ldcssp->lock); 4691 4692 return (0); 4693 } 4694 4695 int 4696 ldc_unregister(ldc_cnex_t *cinfo) 4697 { 4698 if (cinfo == NULL || cinfo->dip == NULL) { 4699 DWARN(DBG_ALL_LDCS, "ldc_unregister: invalid nexus info\n"); 4700 return (EINVAL); 4701 } 4702 4703 mutex_enter(&ldcssp->lock); 4704 4705 if (cinfo->dip != ldcssp->cinfo.dip) { 4706 DWARN(DBG_ALL_LDCS, "ldc_unregister: invalid dip\n"); 4707 mutex_exit(&ldcssp->lock); 4708 return (EINVAL); 4709 } 4710 4711 /* nexus unregister */ 4712 ldcssp->cinfo.dip = NULL; 4713 ldcssp->cinfo.reg_chan = NULL; 4714 ldcssp->cinfo.unreg_chan = NULL; 4715 ldcssp->cinfo.add_intr = NULL; 4716 ldcssp->cinfo.rem_intr = NULL; 4717 ldcssp->cinfo.clr_intr = NULL; 4718 4719 mutex_exit(&ldcssp->lock); 4720 4721 return (0); 4722 } 4723