1 /* 2 * This file is provided under a CDDLv1 license. When using or 3 * redistributing this file, you may do so under this license. 4 * In redistributing this file this license must be included 5 * and no other modification of this header file is permitted. 6 * 7 * CDDL LICENSE SUMMARY 8 * 9 * Copyright(c) 1999 - 2008 Intel Corporation. All rights reserved. 10 * 11 * The contents of this file are subject to the terms of Version 12 * 1.0 of the Common Development and Distribution License (the "License"). 13 * 14 * You should have received a copy of the License with this software. 15 * You can obtain a copy of the License at 16 * http://www.opensolaris.org/os/licensing. 17 * See the License for the specific language governing permissions 18 * and limitations under the License. 19 */ 20 21 /* 22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms of the CDDLv1. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 /* 29 * ********************************************************************** 30 * * 31 * Module Name: * 32 * e1000g_tx.c * 33 * * 34 * Abstract: * 35 * This file contains some routines that take care of Transmit, * 36 * make the hardware to send the data pointed by the packet out * 37 * on to the physical medium. * 38 * * 39 * ********************************************************************** 40 */ 41 42 #include "e1000g_sw.h" 43 #include "e1000g_debug.h" 44 45 static boolean_t e1000g_send(struct e1000g *, mblk_t *); 46 static int e1000g_tx_copy(e1000g_tx_ring_t *, 47 p_tx_sw_packet_t, mblk_t *, uint32_t); 48 static int e1000g_tx_bind(e1000g_tx_ring_t *, 49 p_tx_sw_packet_t, mblk_t *); 50 static boolean_t check_cksum_context(e1000g_tx_ring_t *, cksum_data_t *); 51 static int e1000g_fill_tx_ring(e1000g_tx_ring_t *, LIST_DESCRIBER *, 52 cksum_data_t *); 53 static void e1000g_fill_context_descriptor(cksum_data_t *, 54 struct e1000_context_desc *); 55 static int e1000g_fill_tx_desc(e1000g_tx_ring_t *, 56 p_tx_sw_packet_t, uint64_t, size_t); 57 static uint32_t e1000g_fill_82544_desc(uint64_t Address, size_t Length, 58 p_desc_array_t desc_array); 59 static int e1000g_tx_workaround_PCIX_82544(p_tx_sw_packet_t, uint64_t, size_t); 60 static int e1000g_tx_workaround_jumbo_82544(p_tx_sw_packet_t, uint64_t, size_t); 61 static void e1000g_82547_timeout(void *); 62 static void e1000g_82547_tx_move_tail(e1000g_tx_ring_t *); 63 static void e1000g_82547_tx_move_tail_work(e1000g_tx_ring_t *); 64 65 #ifndef E1000G_DEBUG 66 #pragma inline(e1000g_tx_copy) 67 #pragma inline(e1000g_tx_bind) 68 #pragma inline(check_cksum_context) 69 #pragma inline(e1000g_fill_tx_ring) 70 #pragma inline(e1000g_fill_context_descriptor) 71 #pragma inline(e1000g_fill_tx_desc) 72 #pragma inline(e1000g_fill_82544_desc) 73 #pragma inline(e1000g_tx_workaround_PCIX_82544) 74 #pragma inline(e1000g_tx_workaround_jumbo_82544) 75 #pragma inline(e1000g_free_tx_swpkt) 76 #endif 77 78 /* 79 * e1000g_free_tx_swpkt - free up the tx sw packet 80 * 81 * Unbind the previously bound DMA handle for a given 82 * transmit sw packet. And reset the sw packet data. 83 */ 84 void 85 e1000g_free_tx_swpkt(register p_tx_sw_packet_t packet) 86 { 87 switch (packet->data_transfer_type) { 88 case USE_BCOPY: 89 packet->tx_buf->len = 0; 90 break; 91 #ifdef __sparc 92 case USE_DVMA: 93 dvma_unload(packet->tx_dma_handle, 0, -1); 94 break; 95 #endif 96 case USE_DMA: 97 ddi_dma_unbind_handle(packet->tx_dma_handle); 98 break; 99 default: 100 break; 101 } 102 103 /* 104 * The mblk has been stripped off the sw packet 105 * and will be freed in a triggered soft intr. 106 */ 107 ASSERT(packet->mp == NULL); 108 109 packet->data_transfer_type = USE_NONE; 110 packet->num_mblk_frag = 0; 111 packet->num_desc = 0; 112 } 113 114 mblk_t * 115 e1000g_m_tx(void *arg, mblk_t *mp) 116 { 117 struct e1000g *Adapter = (struct e1000g *)arg; 118 mblk_t *next; 119 120 rw_enter(&Adapter->chip_lock, RW_READER); 121 122 if ((Adapter->chip_state != E1000G_START) || 123 (Adapter->link_state != LINK_STATE_UP)) { 124 freemsgchain(mp); 125 mp = NULL; 126 } 127 128 while (mp != NULL) { 129 next = mp->b_next; 130 mp->b_next = NULL; 131 132 if (!e1000g_send(Adapter, mp)) { 133 mp->b_next = next; 134 break; 135 } 136 137 mp = next; 138 } 139 140 rw_exit(&Adapter->chip_lock); 141 return (mp); 142 } 143 144 /* 145 * e1000g_send - send packets onto the wire 146 * 147 * Called from e1000g_m_tx with an mblk ready to send. this 148 * routine sets up the transmit descriptors and sends data to 149 * the wire. It also pushes the just transmitted packet to 150 * the used tx sw packet list. 151 */ 152 static boolean_t 153 e1000g_send(struct e1000g *Adapter, mblk_t *mp) 154 { 155 struct e1000_hw *hw; 156 p_tx_sw_packet_t packet; 157 LIST_DESCRIBER pending_list; 158 size_t len; 159 size_t msg_size; 160 uint32_t frag_count; 161 int desc_count; 162 uint32_t desc_total; 163 uint32_t force_bcopy; 164 mblk_t *nmp; 165 mblk_t *tmp; 166 e1000g_tx_ring_t *tx_ring; 167 cksum_data_t cksum; 168 169 hw = &Adapter->shared; 170 tx_ring = Adapter->tx_ring; 171 172 /* Get the total size and frags number of the message */ 173 force_bcopy = 0; 174 frag_count = 0; 175 msg_size = 0; 176 for (nmp = mp; nmp; nmp = nmp->b_cont) { 177 frag_count++; 178 msg_size += MBLKL(nmp); 179 } 180 181 /* Make sure packet is less than the max frame size */ 182 if (msg_size > Adapter->max_frame_size - ETHERFCSL) { 183 /* 184 * For the over size packet, we'll just drop it. 185 * So we return B_TRUE here. 186 */ 187 E1000G_DEBUGLOG_1(Adapter, E1000G_WARN_LEVEL, 188 "Tx packet out of bound. length = %d \n", msg_size); 189 E1000G_STAT(tx_ring->stat_over_size); 190 freemsg(mp); 191 return (B_TRUE); 192 } 193 194 /* 195 * Check and reclaim tx descriptors. 196 * This low water mark check should be done all the time as 197 * Transmit interrupt delay can produce Transmit interrupts little 198 * late and that may cause few problems related to reaping Tx 199 * Descriptors... As you may run short of them before getting any 200 * transmit interrupt... 201 */ 202 if (tx_ring->resched_needed || 203 (tx_ring->tbd_avail < Adapter->tx_recycle_thresh)) { 204 (void) e1000g_recycle(tx_ring); 205 E1000G_DEBUG_STAT(tx_ring->stat_recycle); 206 207 if (tx_ring->tbd_avail < DEFAULT_TX_NO_RESOURCE) { 208 E1000G_DEBUG_STAT(tx_ring->stat_lack_desc); 209 goto tx_no_resource; 210 } 211 } 212 213 /* 214 * If there are many frags of the message, then bcopy them 215 * into one tx descriptor buffer will get better performance. 216 */ 217 if ((frag_count >= tx_ring->frags_limit) && 218 (msg_size <= Adapter->tx_buffer_size)) { 219 E1000G_DEBUG_STAT(tx_ring->stat_exceed_frags); 220 force_bcopy |= FORCE_BCOPY_EXCEED_FRAGS; 221 } 222 223 /* 224 * If the message size is less than the minimum ethernet packet size, 225 * we'll use bcopy to send it, and padd it to 60 bytes later. 226 */ 227 if (msg_size < ETHERMIN) { 228 E1000G_DEBUG_STAT(tx_ring->stat_under_size); 229 force_bcopy |= FORCE_BCOPY_UNDER_SIZE; 230 } 231 232 /* Initialize variables */ 233 desc_count = 1; /* The initial value should be greater than 0 */ 234 desc_total = 0; 235 QUEUE_INIT_LIST(&pending_list); 236 237 /* Retrieve checksum info */ 238 hcksum_retrieve(mp, NULL, NULL, &cksum.cksum_start, &cksum.cksum_stuff, 239 NULL, NULL, &cksum.cksum_flags); 240 241 if (((struct ether_vlan_header *)mp->b_rptr)->ether_tpid == 242 htons(ETHERTYPE_VLAN)) 243 cksum.ether_header_size = sizeof (struct ether_vlan_header); 244 else 245 cksum.ether_header_size = sizeof (struct ether_header); 246 247 /* Process each mblk fragment and fill tx descriptors */ 248 packet = NULL; 249 nmp = mp; 250 while (nmp) { 251 tmp = nmp->b_cont; 252 253 len = MBLKL(nmp); 254 /* Check zero length mblks */ 255 if (len == 0) { 256 E1000G_DEBUG_STAT(tx_ring->stat_empty_frags); 257 /* 258 * If there're no packet buffers have been used, 259 * or we just completed processing a buffer, then 260 * skip the empty mblk fragment. 261 * Otherwise, there's still a pending buffer that 262 * needs to be processed (tx_copy). 263 */ 264 if (desc_count > 0) { 265 nmp = tmp; 266 continue; 267 } 268 } 269 270 /* 271 * Get a new TxSwPacket to process mblk buffers. 272 */ 273 if (desc_count > 0) { 274 mutex_enter(&tx_ring->freelist_lock); 275 packet = (p_tx_sw_packet_t) 276 QUEUE_POP_HEAD(&tx_ring->free_list); 277 mutex_exit(&tx_ring->freelist_lock); 278 279 if (packet == NULL) { 280 E1000G_DEBUGLOG_0(Adapter, E1000G_INFO_LEVEL, 281 "No Tx SwPacket available\n"); 282 E1000G_STAT(tx_ring->stat_no_swpkt); 283 goto tx_send_failed; 284 } 285 QUEUE_PUSH_TAIL(&pending_list, &packet->Link); 286 } 287 288 ASSERT(packet); 289 /* 290 * If the size of the fragment is less than the tx_bcopy_thresh 291 * we'll use bcopy; Otherwise, we'll use DMA binding. 292 */ 293 if ((len <= Adapter->tx_bcopy_thresh) || force_bcopy) { 294 desc_count = 295 e1000g_tx_copy(tx_ring, packet, nmp, force_bcopy); 296 E1000G_DEBUG_STAT(tx_ring->stat_copy); 297 } else { 298 desc_count = 299 e1000g_tx_bind(tx_ring, packet, nmp); 300 E1000G_DEBUG_STAT(tx_ring->stat_bind); 301 } 302 303 if (desc_count > 0) 304 desc_total += desc_count; 305 else if (desc_count < 0) 306 goto tx_send_failed; 307 308 nmp = tmp; 309 } 310 311 /* Assign the message to the last sw packet */ 312 ASSERT(packet); 313 ASSERT(packet->mp == NULL); 314 packet->mp = mp; 315 316 /* Try to recycle the tx descriptors again */ 317 if (tx_ring->tbd_avail < (desc_total + 2)) { 318 E1000G_DEBUG_STAT(tx_ring->stat_recycle_retry); 319 (void) e1000g_recycle(tx_ring); 320 } 321 322 mutex_enter(&tx_ring->tx_lock); 323 324 /* 325 * If the number of available tx descriptors is not enough for transmit 326 * (one redundant descriptor and one hw checksum context descriptor are 327 * included), then return failure. 328 */ 329 if (tx_ring->tbd_avail < (desc_total + 2)) { 330 E1000G_DEBUGLOG_0(Adapter, E1000G_INFO_LEVEL, 331 "No Enough Tx descriptors\n"); 332 E1000G_STAT(tx_ring->stat_no_desc); 333 mutex_exit(&tx_ring->tx_lock); 334 goto tx_send_failed; 335 } 336 337 desc_count = e1000g_fill_tx_ring(tx_ring, &pending_list, &cksum); 338 339 mutex_exit(&tx_ring->tx_lock); 340 341 ASSERT(desc_count > 0); 342 343 /* Send successful */ 344 return (B_TRUE); 345 346 tx_send_failed: 347 /* 348 * Enable Transmit interrupts, so that the interrupt routine can 349 * call mac_tx_update() when transmit descriptors become available. 350 */ 351 tx_ring->resched_needed = B_TRUE; 352 if (!Adapter->tx_intr_enable) 353 e1000g_mask_tx_interrupt(Adapter); 354 355 /* Free pending TxSwPackets */ 356 packet = (p_tx_sw_packet_t)QUEUE_GET_HEAD(&pending_list); 357 while (packet) { 358 packet->mp = NULL; 359 e1000g_free_tx_swpkt(packet); 360 packet = (p_tx_sw_packet_t) 361 QUEUE_GET_NEXT(&pending_list, &packet->Link); 362 } 363 364 /* Return pending TxSwPackets to the "Free" list */ 365 mutex_enter(&tx_ring->freelist_lock); 366 QUEUE_APPEND(&tx_ring->free_list, &pending_list); 367 mutex_exit(&tx_ring->freelist_lock); 368 369 E1000G_STAT(tx_ring->stat_send_fail); 370 371 /* Message will be scheduled for re-transmit */ 372 return (B_FALSE); 373 374 tx_no_resource: 375 /* 376 * Enable Transmit interrupts, so that the interrupt routine can 377 * call mac_tx_update() when transmit descriptors become available. 378 */ 379 tx_ring->resched_needed = B_TRUE; 380 if (!Adapter->tx_intr_enable) 381 e1000g_mask_tx_interrupt(Adapter); 382 383 /* Message will be scheduled for re-transmit */ 384 return (B_FALSE); 385 } 386 387 static boolean_t 388 check_cksum_context(e1000g_tx_ring_t *tx_ring, cksum_data_t *cksum) 389 { 390 boolean_t cksum_load; 391 cksum_data_t *last; 392 393 cksum_load = B_FALSE; 394 last = &tx_ring->cksum_data; 395 396 if (cksum->cksum_flags != 0) { 397 if ((cksum->ether_header_size != last->ether_header_size) || 398 (cksum->cksum_flags != last->cksum_flags) || 399 (cksum->cksum_stuff != last->cksum_stuff) || 400 (cksum->cksum_start != last->cksum_start)) { 401 402 cksum_load = B_TRUE; 403 } 404 } 405 406 return (cksum_load); 407 } 408 409 static int 410 e1000g_fill_tx_ring(e1000g_tx_ring_t *tx_ring, LIST_DESCRIBER *pending_list, 411 cksum_data_t *cksum) 412 { 413 struct e1000g *Adapter; 414 struct e1000_hw *hw; 415 p_tx_sw_packet_t first_packet; 416 p_tx_sw_packet_t packet; 417 boolean_t cksum_load; 418 struct e1000_tx_desc *first_data_desc; 419 struct e1000_tx_desc *next_desc; 420 struct e1000_tx_desc *descriptor; 421 int desc_count; 422 int i; 423 424 Adapter = tx_ring->adapter; 425 hw = &Adapter->shared; 426 427 desc_count = 0; 428 first_packet = NULL; 429 first_data_desc = NULL; 430 descriptor = NULL; 431 432 next_desc = tx_ring->tbd_next; 433 434 /* IP Head/TCP/UDP checksum offload */ 435 cksum_load = check_cksum_context(tx_ring, cksum); 436 437 if (cksum_load) { 438 first_packet = (p_tx_sw_packet_t)QUEUE_GET_HEAD(pending_list); 439 440 descriptor = next_desc; 441 442 e1000g_fill_context_descriptor(cksum, 443 (struct e1000_context_desc *)descriptor); 444 445 /* Check the wrap-around case */ 446 if (descriptor == tx_ring->tbd_last) 447 next_desc = tx_ring->tbd_first; 448 else 449 next_desc++; 450 451 desc_count++; 452 } 453 454 first_data_desc = next_desc; 455 456 packet = (p_tx_sw_packet_t)QUEUE_GET_HEAD(pending_list); 457 while (packet) { 458 ASSERT(packet->num_desc); 459 460 for (i = 0; i < packet->num_desc; i++) { 461 ASSERT(tx_ring->tbd_avail > 0); 462 463 descriptor = next_desc; 464 descriptor->buffer_addr = 465 packet->desc[i].address; 466 descriptor->lower.data = 467 packet->desc[i].length; 468 469 /* Zero out status */ 470 descriptor->upper.data = 0; 471 472 descriptor->lower.data |= 473 E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D; 474 /* must set RS on every outgoing descriptor */ 475 descriptor->lower.data |= 476 E1000_TXD_CMD_RS; 477 478 /* Check the wrap-around case */ 479 if (descriptor == tx_ring->tbd_last) 480 next_desc = tx_ring->tbd_first; 481 else 482 next_desc++; 483 484 desc_count++; 485 } 486 487 if (first_packet != NULL) { 488 /* 489 * Count the checksum context descriptor for 490 * the first SwPacket. 491 */ 492 first_packet->num_desc++; 493 first_packet = NULL; 494 } 495 496 packet = (p_tx_sw_packet_t) 497 QUEUE_GET_NEXT(pending_list, &packet->Link); 498 } 499 500 ASSERT(descriptor); 501 502 if (cksum->cksum_flags) { 503 if (cksum->cksum_flags & HCK_IPV4_HDRCKSUM) 504 ((struct e1000_data_desc *)first_data_desc)-> 505 upper.fields.popts |= E1000_TXD_POPTS_IXSM; 506 if (cksum->cksum_flags & HCK_PARTIALCKSUM) 507 ((struct e1000_data_desc *)first_data_desc)-> 508 upper.fields.popts |= E1000_TXD_POPTS_TXSM; 509 } 510 511 /* 512 * Last Descriptor of Packet needs End Of Packet (EOP), Report 513 * Status (RS) and append Ethernet CRC (IFCS) bits set. 514 */ 515 if (Adapter->tx_intr_delay) { 516 descriptor->lower.data |= E1000_TXD_CMD_IDE | 517 E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS; 518 } else { 519 descriptor->lower.data |= 520 E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS; 521 } 522 523 /* 524 * Sync the Tx descriptors DMA buffer 525 */ 526 (void) ddi_dma_sync(tx_ring->tbd_dma_handle, 527 0, 0, DDI_DMA_SYNC_FORDEV); 528 529 tx_ring->tbd_next = next_desc; 530 531 /* 532 * Advance the Transmit Descriptor Tail (Tdt), this tells the 533 * FX1000 that this frame is available to transmit. 534 */ 535 if (hw->mac.type == e1000_82547) 536 e1000g_82547_tx_move_tail(tx_ring); 537 else 538 E1000_WRITE_REG(hw, E1000_TDT(0), 539 (uint32_t)(next_desc - tx_ring->tbd_first)); 540 541 if (e1000g_check_acc_handle(Adapter->osdep.reg_handle) != DDI_FM_OK) { 542 ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED); 543 Adapter->chip_state = E1000G_ERROR; 544 } 545 546 /* Put the pending SwPackets to the "Used" list */ 547 mutex_enter(&tx_ring->usedlist_lock); 548 QUEUE_APPEND(&tx_ring->used_list, pending_list); 549 tx_ring->tbd_avail -= desc_count; 550 mutex_exit(&tx_ring->usedlist_lock); 551 552 /* Store the cksum data */ 553 if (cksum_load) 554 tx_ring->cksum_data = *cksum; 555 556 return (desc_count); 557 } 558 559 560 /* 561 * e1000g_tx_setup - setup tx data structures 562 * 563 * This routine initializes all of the transmit related 564 * structures. This includes the Transmit descriptors, 565 * and the tx_sw_packet structures. 566 */ 567 void 568 e1000g_tx_setup(struct e1000g *Adapter) 569 { 570 struct e1000_hw *hw; 571 p_tx_sw_packet_t packet; 572 UINT i; 573 uint32_t buf_high; 574 uint32_t buf_low; 575 uint32_t reg_tipg; 576 uint32_t reg_tctl; 577 uint32_t reg_tarc; 578 uint16_t speed, duplex; 579 int size; 580 e1000g_tx_ring_t *tx_ring; 581 582 hw = &Adapter->shared; 583 tx_ring = Adapter->tx_ring; 584 585 /* init the lists */ 586 /* 587 * Here we don't need to protect the lists using the 588 * usedlist_lock and freelist_lock, for they have 589 * been protected by the chip_lock. 590 */ 591 QUEUE_INIT_LIST(&tx_ring->used_list); 592 QUEUE_INIT_LIST(&tx_ring->free_list); 593 594 /* Go through and set up each SW_Packet */ 595 packet = tx_ring->packet_area; 596 for (i = 0; i < Adapter->tx_freelist_num; i++, packet++) { 597 /* Initialize this tx_sw_apcket area */ 598 e1000g_free_tx_swpkt(packet); 599 /* Add this tx_sw_packet to the free list */ 600 QUEUE_PUSH_TAIL(&tx_ring->free_list, 601 &packet->Link); 602 } 603 604 /* Setup TX descriptor pointers */ 605 tx_ring->tbd_next = tx_ring->tbd_first; 606 tx_ring->tbd_oldest = tx_ring->tbd_first; 607 608 /* 609 * Setup Hardware TX Registers 610 */ 611 /* Setup the Transmit Control Register (TCTL). */ 612 reg_tctl = E1000_TCTL_PSP | E1000_TCTL_EN | 613 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT) | 614 (E1000_COLLISION_DISTANCE << E1000_COLD_SHIFT) | 615 E1000_TCTL_RTLC; 616 617 /* Enable the MULR bit */ 618 if (hw->bus.type == e1000_bus_type_pci_express) 619 reg_tctl |= E1000_TCTL_MULR; 620 621 E1000_WRITE_REG(hw, E1000_TCTL, reg_tctl); 622 623 if ((hw->mac.type == e1000_82571) || (hw->mac.type == e1000_82572)) { 624 e1000_get_speed_and_duplex(hw, &speed, &duplex); 625 626 reg_tarc = E1000_READ_REG(hw, E1000_TARC(0)); 627 reg_tarc |= (1 << 25); 628 if (speed == SPEED_1000) 629 reg_tarc |= (1 << 21); 630 E1000_WRITE_REG(hw, E1000_TARC(0), reg_tarc); 631 632 reg_tarc = E1000_READ_REG(hw, E1000_TARC(1)); 633 reg_tarc |= (1 << 25); 634 if (reg_tctl & E1000_TCTL_MULR) 635 reg_tarc &= ~(1 << 28); 636 else 637 reg_tarc |= (1 << 28); 638 E1000_WRITE_REG(hw, E1000_TARC(1), reg_tarc); 639 640 } else if (hw->mac.type == e1000_80003es2lan) { 641 reg_tarc = E1000_READ_REG(hw, E1000_TARC(0)); 642 reg_tarc |= 1; 643 if (hw->phy.media_type == e1000_media_type_internal_serdes) 644 reg_tarc |= (1 << 20); 645 E1000_WRITE_REG(hw, E1000_TARC(0), reg_tarc); 646 647 reg_tarc = E1000_READ_REG(hw, E1000_TARC(1)); 648 reg_tarc |= 1; 649 E1000_WRITE_REG(hw, E1000_TARC(1), reg_tarc); 650 } 651 652 /* Setup HW Base and Length of Tx descriptor area */ 653 size = (Adapter->tx_desc_num * sizeof (struct e1000_tx_desc)); 654 E1000_WRITE_REG(hw, E1000_TDLEN(0), size); 655 size = E1000_READ_REG(hw, E1000_TDLEN(0)); 656 657 buf_low = (uint32_t)tx_ring->tbd_dma_addr; 658 buf_high = (uint32_t)(tx_ring->tbd_dma_addr >> 32); 659 660 E1000_WRITE_REG(hw, E1000_TDBAL(0), buf_low); 661 E1000_WRITE_REG(hw, E1000_TDBAH(0), buf_high); 662 663 /* Setup our HW Tx Head & Tail descriptor pointers */ 664 E1000_WRITE_REG(hw, E1000_TDH(0), 0); 665 E1000_WRITE_REG(hw, E1000_TDT(0), 0); 666 667 /* Set the default values for the Tx Inter Packet Gap timer */ 668 if ((hw->mac.type == e1000_82542) && 669 ((hw->revision_id == E1000_REVISION_2) || 670 (hw->revision_id == E1000_REVISION_3))) { 671 reg_tipg = DEFAULT_82542_TIPG_IPGT; 672 reg_tipg |= 673 DEFAULT_82542_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT; 674 reg_tipg |= 675 DEFAULT_82542_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT; 676 } else { 677 if (hw->phy.media_type == e1000_media_type_fiber) 678 reg_tipg = DEFAULT_82543_TIPG_IPGT_FIBER; 679 else 680 reg_tipg = DEFAULT_82543_TIPG_IPGT_COPPER; 681 reg_tipg |= 682 DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT; 683 reg_tipg |= 684 DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT; 685 } 686 E1000_WRITE_REG(hw, E1000_TIPG, reg_tipg); 687 688 /* Setup Transmit Interrupt Delay Value */ 689 E1000_WRITE_REG(hw, E1000_TIDV, Adapter->tx_intr_delay); 690 E1000G_DEBUGLOG_1(Adapter, E1000G_INFO_LEVEL, 691 "E1000_TIDV: 0x%x\n", Adapter->tx_intr_delay); 692 693 if (hw->mac.type >= e1000_82540) { 694 E1000_WRITE_REG(&Adapter->shared, E1000_TADV, 695 Adapter->tx_intr_abs_delay); 696 E1000G_DEBUGLOG_1(Adapter, E1000G_INFO_LEVEL, 697 "E1000_TADV: 0x%x\n", Adapter->tx_intr_abs_delay); 698 } 699 700 tx_ring->tbd_avail = Adapter->tx_desc_num; 701 702 /* For TCP/UDP checksum offload */ 703 tx_ring->cksum_data.cksum_stuff = 0; 704 tx_ring->cksum_data.cksum_start = 0; 705 tx_ring->cksum_data.cksum_flags = 0; 706 tx_ring->cksum_data.ether_header_size = 0; 707 } 708 709 /* 710 * e1000g_recycle - recycle the tx descriptors and tx sw packets 711 */ 712 int 713 e1000g_recycle(e1000g_tx_ring_t *tx_ring) 714 { 715 struct e1000g *Adapter; 716 LIST_DESCRIBER pending_list; 717 p_tx_sw_packet_t packet; 718 mblk_t *mp; 719 mblk_t *nmp; 720 struct e1000_tx_desc *descriptor; 721 int desc_count; 722 int is_intr; 723 724 /* 725 * This function will examine each TxSwPacket in the 'used' queue 726 * if the e1000g is done with it then the associated resources (Tx 727 * Descriptors) will be "freed" and the TxSwPacket will be 728 * returned to the 'free' queue. 729 */ 730 Adapter = tx_ring->adapter; 731 732 packet = (p_tx_sw_packet_t)QUEUE_GET_HEAD(&tx_ring->used_list); 733 if (packet == NULL) { 734 tx_ring->recycle_fail = 0; 735 tx_ring->stall_watchdog = 0; 736 return (0); 737 } 738 739 is_intr = servicing_interrupt(); 740 741 if (is_intr) 742 mutex_enter(&tx_ring->usedlist_lock); 743 else if (mutex_tryenter(&tx_ring->usedlist_lock) == 0) 744 return (0); 745 746 desc_count = 0; 747 QUEUE_INIT_LIST(&pending_list); 748 749 /* Sync the Tx descriptor DMA buffer */ 750 (void) ddi_dma_sync(tx_ring->tbd_dma_handle, 751 0, 0, DDI_DMA_SYNC_FORKERNEL); 752 if (e1000g_check_dma_handle( 753 tx_ring->tbd_dma_handle) != DDI_FM_OK) { 754 mutex_exit(&tx_ring->usedlist_lock); 755 ddi_fm_service_impact(Adapter->dip, DDI_SERVICE_DEGRADED); 756 Adapter->chip_state = E1000G_ERROR; 757 return (0); 758 } 759 760 /* 761 * While there are still TxSwPackets in the used queue check them 762 */ 763 while (packet = 764 (p_tx_sw_packet_t)QUEUE_GET_HEAD(&tx_ring->used_list)) { 765 766 /* 767 * Get hold of the next descriptor that the e1000g will 768 * report status back to (this will be the last descriptor 769 * of a given sw packet). We only want to free the 770 * sw packet (and it resources) if the e1000g is done 771 * with ALL of the descriptors. If the e1000g is done 772 * with the last one then it is done with all of them. 773 */ 774 ASSERT(packet->num_desc); 775 descriptor = tx_ring->tbd_oldest + (packet->num_desc - 1); 776 777 /* Check for wrap case */ 778 if (descriptor > tx_ring->tbd_last) 779 descriptor -= Adapter->tx_desc_num; 780 781 /* 782 * If the descriptor done bit is set free TxSwPacket and 783 * associated resources 784 */ 785 if (descriptor->upper.fields.status & E1000_TXD_STAT_DD) { 786 QUEUE_POP_HEAD(&tx_ring->used_list); 787 QUEUE_PUSH_TAIL(&pending_list, &packet->Link); 788 789 if (descriptor == tx_ring->tbd_last) 790 tx_ring->tbd_oldest = 791 tx_ring->tbd_first; 792 else 793 tx_ring->tbd_oldest = 794 descriptor + 1; 795 796 desc_count += packet->num_desc; 797 798 if (is_intr && (desc_count >= Adapter->tx_recycle_num)) 799 break; 800 } else { 801 /* 802 * Found a sw packet that the e1000g is not done 803 * with then there is no reason to check the rest 804 * of the queue. 805 */ 806 break; 807 } 808 } 809 810 tx_ring->tbd_avail += desc_count; 811 Adapter->tx_pkt_cnt += desc_count; 812 813 mutex_exit(&tx_ring->usedlist_lock); 814 815 if (desc_count == 0) { 816 tx_ring->recycle_fail++; 817 E1000G_DEBUG_STAT(tx_ring->stat_recycle_none); 818 return (0); 819 } 820 821 tx_ring->recycle_fail = 0; 822 tx_ring->stall_watchdog = 0; 823 824 mp = NULL; 825 nmp = NULL; 826 packet = (p_tx_sw_packet_t)QUEUE_GET_HEAD(&pending_list); 827 ASSERT(packet != NULL); 828 while (packet != NULL) { 829 if (packet->mp != NULL) { 830 ASSERT(packet->mp->b_next == NULL); 831 /* Assemble the message chain */ 832 if (mp == NULL) { 833 mp = packet->mp; 834 nmp = packet->mp; 835 } else { 836 nmp->b_next = packet->mp; 837 nmp = packet->mp; 838 } 839 /* Disconnect the message from the sw packet */ 840 packet->mp = NULL; 841 } 842 843 /* Free the TxSwPackets */ 844 e1000g_free_tx_swpkt(packet); 845 846 packet = (p_tx_sw_packet_t) 847 QUEUE_GET_NEXT(&pending_list, &packet->Link); 848 } 849 850 /* Return the TxSwPackets back to the FreeList */ 851 mutex_enter(&tx_ring->freelist_lock); 852 QUEUE_APPEND(&tx_ring->free_list, &pending_list); 853 mutex_exit(&tx_ring->freelist_lock); 854 855 if (mp != NULL) 856 freemsgchain(mp); 857 858 return (desc_count); 859 } 860 /* 861 * 82544 Coexistence issue workaround: 862 * There are 2 issues. 863 * 1. If a 32 bit split completion happens from P64H2 and another 864 * agent drives a 64 bit request/split completion after ONLY 865 * 1 idle clock (BRCM/Emulex/Adaptec fiber channel cards) then 866 * 82544 has a problem where in to clock all the data in, it 867 * looks at REQ64# signal and since it has changed so fast (i.e. 1 868 * idle clock turn around), it will fail to clock all the data in. 869 * Data coming from certain ending addresses has exposure to this issue. 870 * 871 * To detect this issue, following equation can be used... 872 * SIZE[3:0] + ADDR[2:0] = SUM[3:0]. 873 * If SUM[3:0] is in between 1 to 4, we will have this issue. 874 * 875 * ROOT CAUSE: 876 * The erratum involves the 82544 PCIX elasticity FIFO implementations as 877 * 64-bit FIFO's and flushing of the final partial-bytes corresponding 878 * to the end of a requested read burst. Under a specific burst condition 879 * of ending-data alignment and 32-byte split-completions, the final 880 * byte(s) of split-completion data require an extra clock cycle to flush 881 * into 64-bit FIFO orientation. An incorrect logic dependency on the 882 * REQ64# signal occurring during during this clock cycle may cause the 883 * residual byte(s) to be lost, thereby rendering the internal DMA client 884 * forever awaiting the final byte(s) for an outbound data-fetch. The 885 * erratum is confirmed to *only* occur if certain subsequent external 886 * 64-bit PCIX bus transactions occur immediately (minimum possible bus 887 * turn- around) following the odd-aligned 32-bit split-completion 888 * containing the final byte(s). Intel has confirmed that this has been 889 * seen only with chipset/bridges which have the capability to provide 890 * 32-bit split-completion data, and in the presence of newer PCIX bus 891 * agents which fully-optimize the inter-transaction turn-around (zero 892 * additional initiator latency when pre-granted bus ownership). 893 * 894 * This issue does not exist in PCI bus mode, when any agent is operating 895 * in 32 bit only mode or on chipsets that do not do 32 bit split 896 * completions for 64 bit read requests (Serverworks chipsets). P64H2 does 897 * 32 bit split completions for any read request that has bit 2 set to 1 898 * for the requested address and read request size is more than 8 bytes. 899 * 900 * 2. Another issue is related to 82544 driving DACs under the similar 901 * scenario (32 bit split completion followed by 64 bit transaction with 902 * only 1 cycle turnaround). This issue is still being root caused. We 903 * think that both of these issues can be avoided if following workaround 904 * is implemented. It seems DAC issues is related to ending addresses being 905 * 0x9, 0xA, 0xB, 0xC and hence ending up at odd boundaries in elasticity 906 * FIFO which does not get flushed due to REQ64# dependency. We will only 907 * know the full story after it has been simulated successfully by HW team. 908 * 909 * WORKAROUND: 910 * Make sure we do not have ending address as 1,2,3,4(Hang) or 9,a,b,c(DAC) 911 */ 912 static uint32_t 913 e1000g_fill_82544_desc(uint64_t address, 914 size_t length, p_desc_array_t desc_array) 915 { 916 /* 917 * Since issue is sensitive to length and address. 918 * Let us first check the address... 919 */ 920 uint32_t safe_terminator; 921 922 if (length <= 4) { 923 desc_array->descriptor[0].address = address; 924 desc_array->descriptor[0].length = length; 925 desc_array->elements = 1; 926 return (desc_array->elements); 927 } 928 safe_terminator = 929 (uint32_t)((((uint32_t)address & 0x7) + 930 (length & 0xF)) & 0xF); 931 /* 932 * if it does not fall between 0x1 to 0x4 and 0x9 to 0xC then 933 * return 934 */ 935 if (safe_terminator == 0 || 936 (safe_terminator > 4 && safe_terminator < 9) || 937 (safe_terminator > 0xC && safe_terminator <= 0xF)) { 938 desc_array->descriptor[0].address = address; 939 desc_array->descriptor[0].length = length; 940 desc_array->elements = 1; 941 return (desc_array->elements); 942 } 943 944 desc_array->descriptor[0].address = address; 945 desc_array->descriptor[0].length = length - 4; 946 desc_array->descriptor[1].address = address + (length - 4); 947 desc_array->descriptor[1].length = 4; 948 desc_array->elements = 2; 949 return (desc_array->elements); 950 } 951 952 static int 953 e1000g_tx_copy(e1000g_tx_ring_t *tx_ring, p_tx_sw_packet_t packet, 954 mblk_t *mp, uint32_t force_bcopy) 955 { 956 size_t len; 957 size_t len1; 958 dma_buffer_t *tx_buf; 959 mblk_t *nmp; 960 boolean_t finished; 961 int desc_count; 962 963 desc_count = 0; 964 tx_buf = packet->tx_buf; 965 len = MBLKL(mp); 966 967 ASSERT((tx_buf->len + len) <= tx_buf->size); 968 969 if (len > 0) { 970 bcopy(mp->b_rptr, 971 tx_buf->address + tx_buf->len, 972 len); 973 tx_buf->len += len; 974 975 packet->num_mblk_frag++; 976 } 977 978 nmp = mp->b_cont; 979 if (nmp == NULL) { 980 finished = B_TRUE; 981 } else { 982 len1 = MBLKL(nmp); 983 if ((tx_buf->len + len1) > tx_buf->size) 984 finished = B_TRUE; 985 else if (force_bcopy) 986 finished = B_FALSE; 987 else if (len1 > tx_ring->adapter->tx_bcopy_thresh) 988 finished = B_TRUE; 989 else 990 finished = B_FALSE; 991 } 992 993 if (finished) { 994 E1000G_DEBUG_STAT_COND(tx_ring->stat_multi_copy, 995 (tx_buf->len > len)); 996 997 /* 998 * If the packet is smaller than 64 bytes, which is the 999 * minimum ethernet packet size, pad the packet to make 1000 * it at least 60 bytes. The hardware will add 4 bytes 1001 * for CRC. 1002 */ 1003 if (force_bcopy & FORCE_BCOPY_UNDER_SIZE) { 1004 ASSERT(tx_buf->len < ETHERMIN); 1005 1006 bzero(tx_buf->address + tx_buf->len, 1007 ETHERMIN - tx_buf->len); 1008 tx_buf->len = ETHERMIN; 1009 } 1010 1011 #ifdef __sparc 1012 if (packet->dma_type == USE_DVMA) 1013 dvma_sync(tx_buf->dma_handle, 0, DDI_DMA_SYNC_FORDEV); 1014 else 1015 (void) ddi_dma_sync(tx_buf->dma_handle, 0, 1016 tx_buf->len, DDI_DMA_SYNC_FORDEV); 1017 #else 1018 (void) ddi_dma_sync(tx_buf->dma_handle, 0, 1019 tx_buf->len, DDI_DMA_SYNC_FORDEV); 1020 #endif 1021 1022 packet->data_transfer_type = USE_BCOPY; 1023 1024 desc_count = e1000g_fill_tx_desc(tx_ring, 1025 packet, 1026 tx_buf->dma_address, 1027 tx_buf->len); 1028 1029 if (desc_count <= 0) 1030 return (-1); 1031 } 1032 1033 return (desc_count); 1034 } 1035 1036 static int 1037 e1000g_tx_bind(e1000g_tx_ring_t *tx_ring, p_tx_sw_packet_t packet, mblk_t *mp) 1038 { 1039 int j; 1040 int mystat; 1041 size_t len; 1042 ddi_dma_cookie_t dma_cookie; 1043 uint_t ncookies; 1044 int desc_count; 1045 uint32_t desc_total; 1046 1047 desc_total = 0; 1048 len = MBLKL(mp); 1049 1050 /* 1051 * ddi_dma_addr_bind_handle() allocates DMA resources for a 1052 * memory object such that a device can perform DMA to or from 1053 * the object. DMA resources are allocated considering the 1054 * device's DMA attributes as expressed by ddi_dma_attr(9S) 1055 * (see ddi_dma_alloc_handle(9F)). 1056 * 1057 * ddi_dma_addr_bind_handle() fills in the first DMA cookie 1058 * pointed to by cookiep with the appropriate address, length, 1059 * and bus type. *ccountp is set to the number of DMA cookies 1060 * representing this DMA object. Subsequent DMA cookies must be 1061 * retrieved by calling ddi_dma_nextcookie(9F) the number of 1062 * times specified by *countp - 1. 1063 */ 1064 switch (packet->dma_type) { 1065 #ifdef __sparc 1066 case USE_DVMA: 1067 dvma_kaddr_load(packet->tx_dma_handle, 1068 (caddr_t)mp->b_rptr, len, 0, &dma_cookie); 1069 1070 dvma_sync(packet->tx_dma_handle, 0, 1071 DDI_DMA_SYNC_FORDEV); 1072 1073 ncookies = 1; 1074 packet->data_transfer_type = USE_DVMA; 1075 break; 1076 #endif 1077 case USE_DMA: 1078 if ((mystat = ddi_dma_addr_bind_handle( 1079 packet->tx_dma_handle, NULL, 1080 (caddr_t)mp->b_rptr, len, 1081 DDI_DMA_WRITE | DDI_DMA_STREAMING, 1082 DDI_DMA_DONTWAIT, 0, &dma_cookie, 1083 &ncookies)) != DDI_DMA_MAPPED) { 1084 1085 e1000g_log(tx_ring->adapter, CE_WARN, 1086 "Couldn't bind mblk buffer to Tx DMA handle: " 1087 "return: %X, Pkt: %X\n", 1088 mystat, packet); 1089 return (-1); 1090 } 1091 1092 /* 1093 * An implicit ddi_dma_sync() is done when the 1094 * ddi_dma_addr_bind_handle() is called. So we 1095 * don't need to explicitly call ddi_dma_sync() 1096 * here any more. 1097 */ 1098 ASSERT(ncookies); 1099 E1000G_DEBUG_STAT_COND(tx_ring->stat_multi_cookie, 1100 (ncookies > 1)); 1101 1102 /* 1103 * The data_transfer_type value must be set after the handle 1104 * has been bound, for it will be used in e1000g_free_tx_swpkt() 1105 * to decide whether we need to unbind the handle. 1106 */ 1107 packet->data_transfer_type = USE_DMA; 1108 break; 1109 default: 1110 ASSERT(B_FALSE); 1111 break; 1112 } 1113 1114 packet->num_mblk_frag++; 1115 1116 /* 1117 * Each address could span thru multpile cookie.. 1118 * Each cookie will have one descriptor 1119 */ 1120 for (j = ncookies; j != 0; j--) { 1121 1122 desc_count = e1000g_fill_tx_desc(tx_ring, 1123 packet, 1124 dma_cookie.dmac_laddress, 1125 dma_cookie.dmac_size); 1126 1127 if (desc_count <= 0) 1128 return (-1); 1129 1130 desc_total += desc_count; 1131 1132 /* 1133 * ddi_dma_nextcookie() retrieves subsequent DMA 1134 * cookies for a DMA object. 1135 * ddi_dma_nextcookie() fills in the 1136 * ddi_dma_cookie(9S) structure pointed to by 1137 * cookiep. The ddi_dma_cookie(9S) structure 1138 * must be allocated prior to calling 1139 * ddi_dma_nextcookie(). The DMA cookie count 1140 * returned by ddi_dma_buf_bind_handle(9F), 1141 * ddi_dma_addr_bind_handle(9F), or 1142 * ddi_dma_getwin(9F) indicates the number of DMA 1143 * cookies a DMA object consists of. If the 1144 * resulting cookie count, N, is larger than 1, 1145 * ddi_dma_nextcookie() must be called N-1 times 1146 * to retrieve all DMA cookies. 1147 */ 1148 if (j > 1) { 1149 ddi_dma_nextcookie(packet->tx_dma_handle, 1150 &dma_cookie); 1151 } 1152 } 1153 1154 return (desc_total); 1155 } 1156 1157 static void 1158 e1000g_fill_context_descriptor(cksum_data_t *cksum, 1159 struct e1000_context_desc *cksum_desc) 1160 { 1161 if (cksum->cksum_flags & HCK_IPV4_HDRCKSUM) { 1162 cksum_desc->lower_setup.ip_fields.ipcss = 1163 cksum->ether_header_size; 1164 cksum_desc->lower_setup.ip_fields.ipcso = 1165 cksum->ether_header_size + 1166 offsetof(struct ip, ip_sum); 1167 cksum_desc->lower_setup.ip_fields.ipcse = 1168 cksum->ether_header_size + 1169 sizeof (struct ip) - 1; 1170 } else 1171 cksum_desc->lower_setup.ip_config = 0; 1172 1173 if (cksum->cksum_flags & HCK_PARTIALCKSUM) { 1174 /* 1175 * The packet with same protocol has the following 1176 * stuff and start offset: 1177 * | Protocol | Stuff | Start | Checksum 1178 * | | Offset | Offset | Enable 1179 * | IPv4 + TCP | 0x24 | 0x14 | Yes 1180 * | IPv4 + UDP | 0x1A | 0x14 | Yes 1181 * | IPv6 + TCP | 0x20 | 0x10 | No 1182 * | IPv6 + UDP | 0x14 | 0x10 | No 1183 */ 1184 cksum_desc->upper_setup.tcp_fields.tucss = 1185 cksum->cksum_start + cksum->ether_header_size; 1186 cksum_desc->upper_setup.tcp_fields.tucso = 1187 cksum->cksum_stuff + cksum->ether_header_size; 1188 cksum_desc->upper_setup.tcp_fields.tucse = 0; 1189 } else 1190 cksum_desc->upper_setup.tcp_config = 0; 1191 1192 cksum_desc->cmd_and_length = E1000_TXD_CMD_DEXT; 1193 1194 /* 1195 * Zero out the options for TCP Segmentation Offload, 1196 * since we don't support it in this version 1197 */ 1198 cksum_desc->tcp_seg_setup.data = 0; 1199 } 1200 1201 static int 1202 e1000g_fill_tx_desc(e1000g_tx_ring_t *tx_ring, 1203 p_tx_sw_packet_t packet, uint64_t address, size_t size) 1204 { 1205 struct e1000_hw *hw = &tx_ring->adapter->shared; 1206 p_sw_desc_t desc; 1207 1208 if (hw->mac.type == e1000_82544) { 1209 if (hw->bus.type == e1000_bus_type_pcix) 1210 return (e1000g_tx_workaround_PCIX_82544(packet, 1211 address, size)); 1212 1213 if (size > JUMBO_FRAG_LENGTH) 1214 return (e1000g_tx_workaround_jumbo_82544(packet, 1215 address, size)); 1216 } 1217 1218 ASSERT(packet->num_desc < MAX_TX_DESC_PER_PACKET); 1219 1220 desc = &packet->desc[packet->num_desc]; 1221 desc->address = address; 1222 desc->length = size; 1223 1224 packet->num_desc++; 1225 1226 return (1); 1227 } 1228 1229 static int 1230 e1000g_tx_workaround_PCIX_82544(p_tx_sw_packet_t packet, 1231 uint64_t address, size_t size) 1232 { 1233 p_sw_desc_t desc; 1234 int desc_count; 1235 long size_left; 1236 size_t len; 1237 uint32_t counter; 1238 uint32_t array_elements; 1239 desc_array_t desc_array; 1240 1241 /* 1242 * Coexist Workaround for cordova: RP: 07/04/03 1243 * 1244 * RP: ERRATA: Workaround ISSUE: 1245 * 8kb_buffer_Lockup CONTROLLER: Cordova Breakup 1246 * Eachbuffer in to 8kb pieces until the 1247 * remainder is < 8kb 1248 */ 1249 size_left = size; 1250 desc_count = 0; 1251 1252 while (size_left > 0) { 1253 if (size_left > MAX_TX_BUF_SIZE) 1254 len = MAX_TX_BUF_SIZE; 1255 else 1256 len = size_left; 1257 1258 array_elements = e1000g_fill_82544_desc(address, 1259 len, &desc_array); 1260 1261 for (counter = 0; counter < array_elements; counter++) { 1262 ASSERT(packet->num_desc < MAX_TX_DESC_PER_PACKET); 1263 /* 1264 * Put in the buffer address 1265 */ 1266 desc = &packet->desc[packet->num_desc]; 1267 1268 desc->address = 1269 desc_array.descriptor[counter].address; 1270 desc->length = 1271 desc_array.descriptor[counter].length; 1272 1273 packet->num_desc++; 1274 desc_count++; 1275 } /* for */ 1276 1277 /* 1278 * Update the buffer address and length 1279 */ 1280 address += MAX_TX_BUF_SIZE; 1281 size_left -= MAX_TX_BUF_SIZE; 1282 } /* while */ 1283 1284 return (desc_count); 1285 } 1286 1287 static int 1288 e1000g_tx_workaround_jumbo_82544(p_tx_sw_packet_t packet, 1289 uint64_t address, size_t size) 1290 { 1291 p_sw_desc_t desc; 1292 int desc_count; 1293 long size_left; 1294 uint32_t offset; 1295 1296 /* 1297 * Workaround for Jumbo Frames on Cordova 1298 * PSD 06/01/2001 1299 */ 1300 size_left = size; 1301 desc_count = 0; 1302 offset = 0; 1303 while (size_left > 0) { 1304 ASSERT(packet->num_desc < MAX_TX_DESC_PER_PACKET); 1305 1306 desc = &packet->desc[packet->num_desc]; 1307 1308 desc->address = address + offset; 1309 1310 if (size_left > JUMBO_FRAG_LENGTH) 1311 desc->length = JUMBO_FRAG_LENGTH; 1312 else 1313 desc->length = size_left; 1314 1315 packet->num_desc++; 1316 desc_count++; 1317 1318 offset += desc->length; 1319 size_left -= JUMBO_FRAG_LENGTH; 1320 } 1321 1322 return (desc_count); 1323 } 1324 1325 #pragma inline(e1000g_82547_tx_move_tail_work) 1326 1327 static void 1328 e1000g_82547_tx_move_tail_work(e1000g_tx_ring_t *tx_ring) 1329 { 1330 struct e1000_hw *hw; 1331 uint16_t hw_tdt; 1332 uint16_t sw_tdt; 1333 struct e1000_tx_desc *tx_desc; 1334 uint16_t length = 0; 1335 boolean_t eop = B_FALSE; 1336 struct e1000g *Adapter; 1337 1338 Adapter = tx_ring->adapter; 1339 hw = &Adapter->shared; 1340 1341 hw_tdt = E1000_READ_REG(hw, E1000_TDT(0)); 1342 sw_tdt = tx_ring->tbd_next - tx_ring->tbd_first; 1343 1344 while (hw_tdt != sw_tdt) { 1345 tx_desc = &(tx_ring->tbd_first[hw_tdt]); 1346 length += tx_desc->lower.flags.length; 1347 eop = tx_desc->lower.data & E1000_TXD_CMD_EOP; 1348 if (++hw_tdt == Adapter->tx_desc_num) 1349 hw_tdt = 0; 1350 1351 if (eop) { 1352 if ((Adapter->link_duplex == HALF_DUPLEX) && 1353 (e1000_fifo_workaround_82547(hw, length) 1354 != E1000_SUCCESS)) { 1355 if (tx_ring->timer_enable_82547) { 1356 ASSERT(tx_ring->timer_id_82547 == 0); 1357 tx_ring->timer_id_82547 = 1358 timeout(e1000g_82547_timeout, 1359 (void *)tx_ring, 1360 drv_usectohz(10000)); 1361 } 1362 return; 1363 1364 } else { 1365 E1000_WRITE_REG(hw, E1000_TDT(0), hw_tdt); 1366 e1000_update_tx_fifo_head_82547(hw, length); 1367 length = 0; 1368 } 1369 } 1370 } 1371 } 1372 1373 static void 1374 e1000g_82547_timeout(void *arg) 1375 { 1376 e1000g_tx_ring_t *tx_ring; 1377 1378 tx_ring = (e1000g_tx_ring_t *)arg; 1379 1380 mutex_enter(&tx_ring->tx_lock); 1381 1382 tx_ring->timer_id_82547 = 0; 1383 e1000g_82547_tx_move_tail_work(tx_ring); 1384 1385 mutex_exit(&tx_ring->tx_lock); 1386 } 1387 1388 static void 1389 e1000g_82547_tx_move_tail(e1000g_tx_ring_t *tx_ring) 1390 { 1391 timeout_id_t tid; 1392 1393 ASSERT(MUTEX_HELD(&tx_ring->tx_lock)); 1394 1395 tid = tx_ring->timer_id_82547; 1396 tx_ring->timer_id_82547 = 0; 1397 if (tid != 0) { 1398 tx_ring->timer_enable_82547 = B_FALSE; 1399 mutex_exit(&tx_ring->tx_lock); 1400 1401 (void) untimeout(tid); 1402 1403 mutex_enter(&tx_ring->tx_lock); 1404 } 1405 tx_ring->timer_enable_82547 = B_TRUE; 1406 e1000g_82547_tx_move_tail_work(tx_ring); 1407 } 1408