1 /* 2 * This file is provided under a dual BSD/GPLv2 license. When using or 3 * redistributing this file, you may do so under either license. 4 * 5 * GPL LICENSE SUMMARY 6 * 7 * Copyright(c) 2012 Intel Corporation. All rights reserved. 8 * Copyright (C) 2015 EMC Corporation. All Rights Reserved. 9 * 10 * This program is free software; you can redistribute it and/or modify 11 * it under the terms of version 2 of the GNU General Public License as 12 * published by the Free Software Foundation. 13 * 14 * BSD LICENSE 15 * 16 * Copyright(c) 2012 Intel Corporation. All rights reserved. 17 * Copyright (C) 2015 EMC Corporation. All Rights Reserved. 18 * 19 * Redistribution and use in source and binary forms, with or without 20 * modification, are permitted provided that the following conditions 21 * are met: 22 * 23 * * Redistributions of source code must retain the above copyright 24 * notice, this list of conditions and the following disclaimer. 25 * * Redistributions in binary form must reproduce the above copy 26 * notice, this list of conditions and the following disclaimer in 27 * the documentation and/or other materials provided with the 28 * distribution. 29 * * Neither the name of Intel Corporation nor the names of its 30 * contributors may be used to endorse or promote products derived 31 * from this software without specific prior written permission. 32 * 33 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 34 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 35 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 36 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 37 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 38 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 39 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 40 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 41 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 42 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 43 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 44 * 45 * PCIe NTB Transport Linux driver 46 * 47 * Contact Information: 48 * Jon Mason <jon.mason@intel.com> 49 */ 50 #include <linux/debugfs.h> 51 #include <linux/delay.h> 52 #include <linux/dmaengine.h> 53 #include <linux/dma-mapping.h> 54 #include <linux/errno.h> 55 #include <linux/export.h> 56 #include <linux/interrupt.h> 57 #include <linux/module.h> 58 #include <linux/pci.h> 59 #include <linux/slab.h> 60 #include <linux/types.h> 61 #include <linux/uaccess.h> 62 #include "linux/ntb.h" 63 #include "linux/ntb_transport.h" 64 65 #define NTB_TRANSPORT_VERSION 4 66 #define NTB_TRANSPORT_VER "4" 67 #define NTB_TRANSPORT_NAME "ntb_transport" 68 #define NTB_TRANSPORT_DESC "Software Queue-Pair Transport over NTB" 69 #define NTB_TRANSPORT_MIN_SPADS (MW0_SZ_HIGH + 2) 70 71 MODULE_DESCRIPTION(NTB_TRANSPORT_DESC); 72 MODULE_VERSION(NTB_TRANSPORT_VER); 73 MODULE_LICENSE("Dual BSD/GPL"); 74 MODULE_AUTHOR("Intel Corporation"); 75 76 static unsigned long max_mw_size; 77 module_param(max_mw_size, ulong, 0644); 78 MODULE_PARM_DESC(max_mw_size, "Limit size of large memory windows"); 79 80 static unsigned int transport_mtu = 0x10000; 81 module_param(transport_mtu, uint, 0644); 82 MODULE_PARM_DESC(transport_mtu, "Maximum size of NTB transport packets"); 83 84 static unsigned char max_num_clients; 85 module_param(max_num_clients, byte, 0644); 86 MODULE_PARM_DESC(max_num_clients, "Maximum number of NTB transport clients"); 87 88 static unsigned int copy_bytes = 1024; 89 module_param(copy_bytes, uint, 0644); 90 MODULE_PARM_DESC(copy_bytes, "Threshold under which NTB will use the CPU to copy instead of DMA"); 91 92 static bool use_dma; 93 module_param(use_dma, bool, 0644); 94 MODULE_PARM_DESC(use_dma, "Use DMA engine to perform large data copy"); 95 96 static struct dentry *nt_debugfs_dir; 97 98 /* Only two-ports NTB devices are supported */ 99 #define PIDX NTB_DEF_PEER_IDX 100 101 struct ntb_queue_entry { 102 /* ntb_queue list reference */ 103 struct list_head entry; 104 /* pointers to data to be transferred */ 105 void *cb_data; 106 void *buf; 107 unsigned int len; 108 unsigned int flags; 109 int retries; 110 int errors; 111 unsigned int tx_index; 112 unsigned int rx_index; 113 114 struct ntb_transport_qp *qp; 115 union { 116 struct ntb_payload_header __iomem *tx_hdr; 117 struct ntb_payload_header *rx_hdr; 118 }; 119 }; 120 121 struct ntb_rx_info { 122 unsigned int entry; 123 }; 124 125 struct ntb_transport_qp { 126 struct ntb_transport_ctx *transport; 127 struct ntb_dev *ndev; 128 void *cb_data; 129 struct dma_chan *tx_dma_chan; 130 struct dma_chan *rx_dma_chan; 131 132 bool client_ready; 133 bool link_is_up; 134 bool active; 135 136 u8 qp_num; /* Only 64 QP's are allowed. 0-63 */ 137 u64 qp_bit; 138 139 struct ntb_rx_info __iomem *rx_info; 140 struct ntb_rx_info *remote_rx_info; 141 142 void (*tx_handler)(struct ntb_transport_qp *qp, void *qp_data, 143 void *data, int len); 144 struct list_head tx_free_q; 145 spinlock_t ntb_tx_free_q_lock; 146 void __iomem *tx_mw; 147 dma_addr_t tx_mw_phys; 148 unsigned int tx_index; 149 unsigned int tx_max_entry; 150 unsigned int tx_max_frame; 151 152 void (*rx_handler)(struct ntb_transport_qp *qp, void *qp_data, 153 void *data, int len); 154 struct list_head rx_post_q; 155 struct list_head rx_pend_q; 156 struct list_head rx_free_q; 157 /* ntb_rx_q_lock: synchronize access to rx_XXXX_q */ 158 spinlock_t ntb_rx_q_lock; 159 void *rx_buff; 160 unsigned int rx_index; 161 unsigned int rx_max_entry; 162 unsigned int rx_max_frame; 163 unsigned int rx_alloc_entry; 164 dma_cookie_t last_cookie; 165 struct tasklet_struct rxc_db_work; 166 167 void (*event_handler)(void *data, int status); 168 struct delayed_work link_work; 169 struct work_struct link_cleanup; 170 171 struct dentry *debugfs_dir; 172 struct dentry *debugfs_stats; 173 174 /* Stats */ 175 u64 rx_bytes; 176 u64 rx_pkts; 177 u64 rx_ring_empty; 178 u64 rx_err_no_buf; 179 u64 rx_err_oflow; 180 u64 rx_err_ver; 181 u64 rx_memcpy; 182 u64 rx_async; 183 u64 tx_bytes; 184 u64 tx_pkts; 185 u64 tx_ring_full; 186 u64 tx_err_no_buf; 187 u64 tx_memcpy; 188 u64 tx_async; 189 }; 190 191 struct ntb_transport_mw { 192 phys_addr_t phys_addr; 193 resource_size_t phys_size; 194 resource_size_t xlat_align; 195 resource_size_t xlat_align_size; 196 void __iomem *vbase; 197 size_t xlat_size; 198 size_t buff_size; 199 void *virt_addr; 200 dma_addr_t dma_addr; 201 }; 202 203 struct ntb_transport_client_dev { 204 struct list_head entry; 205 struct ntb_transport_ctx *nt; 206 struct device dev; 207 }; 208 209 struct ntb_transport_ctx { 210 struct list_head entry; 211 struct list_head client_devs; 212 213 struct ntb_dev *ndev; 214 215 struct ntb_transport_mw *mw_vec; 216 struct ntb_transport_qp *qp_vec; 217 unsigned int mw_count; 218 unsigned int qp_count; 219 u64 qp_bitmap; 220 u64 qp_bitmap_free; 221 222 bool link_is_up; 223 struct delayed_work link_work; 224 struct work_struct link_cleanup; 225 226 struct dentry *debugfs_node_dir; 227 }; 228 229 enum { 230 DESC_DONE_FLAG = BIT(0), 231 LINK_DOWN_FLAG = BIT(1), 232 }; 233 234 struct ntb_payload_header { 235 unsigned int ver; 236 unsigned int len; 237 unsigned int flags; 238 }; 239 240 enum { 241 VERSION = 0, 242 QP_LINKS, 243 NUM_QPS, 244 NUM_MWS, 245 MW0_SZ_HIGH, 246 MW0_SZ_LOW, 247 }; 248 249 #define dev_client_dev(__dev) \ 250 container_of((__dev), struct ntb_transport_client_dev, dev) 251 252 #define drv_client(__drv) \ 253 container_of((__drv), struct ntb_transport_client, driver) 254 255 #define QP_TO_MW(nt, qp) ((qp) % nt->mw_count) 256 #define NTB_QP_DEF_NUM_ENTRIES 100 257 #define NTB_LINK_DOWN_TIMEOUT 10 258 259 static void ntb_transport_rxc_db(unsigned long data); 260 static const struct ntb_ctx_ops ntb_transport_ops; 261 static struct ntb_client ntb_transport_client; 262 static int ntb_async_tx_submit(struct ntb_transport_qp *qp, 263 struct ntb_queue_entry *entry); 264 static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset); 265 static int ntb_async_rx_submit(struct ntb_queue_entry *entry, void *offset); 266 static void ntb_memcpy_rx(struct ntb_queue_entry *entry, void *offset); 267 268 269 static int ntb_transport_bus_match(struct device *dev, 270 struct device_driver *drv) 271 { 272 return !strncmp(dev_name(dev), drv->name, strlen(drv->name)); 273 } 274 275 static int ntb_transport_bus_probe(struct device *dev) 276 { 277 const struct ntb_transport_client *client; 278 int rc = -EINVAL; 279 280 get_device(dev); 281 282 client = drv_client(dev->driver); 283 rc = client->probe(dev); 284 if (rc) 285 put_device(dev); 286 287 return rc; 288 } 289 290 static int ntb_transport_bus_remove(struct device *dev) 291 { 292 const struct ntb_transport_client *client; 293 294 client = drv_client(dev->driver); 295 client->remove(dev); 296 297 put_device(dev); 298 299 return 0; 300 } 301 302 static struct bus_type ntb_transport_bus = { 303 .name = "ntb_transport", 304 .match = ntb_transport_bus_match, 305 .probe = ntb_transport_bus_probe, 306 .remove = ntb_transport_bus_remove, 307 }; 308 309 static LIST_HEAD(ntb_transport_list); 310 311 static int ntb_bus_init(struct ntb_transport_ctx *nt) 312 { 313 list_add_tail(&nt->entry, &ntb_transport_list); 314 return 0; 315 } 316 317 static void ntb_bus_remove(struct ntb_transport_ctx *nt) 318 { 319 struct ntb_transport_client_dev *client_dev, *cd; 320 321 list_for_each_entry_safe(client_dev, cd, &nt->client_devs, entry) { 322 dev_err(client_dev->dev.parent, "%s still attached to bus, removing\n", 323 dev_name(&client_dev->dev)); 324 list_del(&client_dev->entry); 325 device_unregister(&client_dev->dev); 326 } 327 328 list_del(&nt->entry); 329 } 330 331 static void ntb_transport_client_release(struct device *dev) 332 { 333 struct ntb_transport_client_dev *client_dev; 334 335 client_dev = dev_client_dev(dev); 336 kfree(client_dev); 337 } 338 339 /** 340 * ntb_transport_unregister_client_dev - Unregister NTB client device 341 * @device_name: Name of NTB client device 342 * 343 * Unregister an NTB client device with the NTB transport layer 344 */ 345 void ntb_transport_unregister_client_dev(char *device_name) 346 { 347 struct ntb_transport_client_dev *client, *cd; 348 struct ntb_transport_ctx *nt; 349 350 list_for_each_entry(nt, &ntb_transport_list, entry) 351 list_for_each_entry_safe(client, cd, &nt->client_devs, entry) 352 if (!strncmp(dev_name(&client->dev), device_name, 353 strlen(device_name))) { 354 list_del(&client->entry); 355 device_unregister(&client->dev); 356 } 357 } 358 EXPORT_SYMBOL_GPL(ntb_transport_unregister_client_dev); 359 360 /** 361 * ntb_transport_register_client_dev - Register NTB client device 362 * @device_name: Name of NTB client device 363 * 364 * Register an NTB client device with the NTB transport layer 365 */ 366 int ntb_transport_register_client_dev(char *device_name) 367 { 368 struct ntb_transport_client_dev *client_dev; 369 struct ntb_transport_ctx *nt; 370 int node; 371 int rc, i = 0; 372 373 if (list_empty(&ntb_transport_list)) 374 return -ENODEV; 375 376 list_for_each_entry(nt, &ntb_transport_list, entry) { 377 struct device *dev; 378 379 node = dev_to_node(&nt->ndev->dev); 380 381 client_dev = kzalloc_node(sizeof(*client_dev), 382 GFP_KERNEL, node); 383 if (!client_dev) { 384 rc = -ENOMEM; 385 goto err; 386 } 387 388 dev = &client_dev->dev; 389 390 /* setup and register client devices */ 391 dev_set_name(dev, "%s%d", device_name, i); 392 dev->bus = &ntb_transport_bus; 393 dev->release = ntb_transport_client_release; 394 dev->parent = &nt->ndev->dev; 395 396 rc = device_register(dev); 397 if (rc) { 398 kfree(client_dev); 399 goto err; 400 } 401 402 list_add_tail(&client_dev->entry, &nt->client_devs); 403 i++; 404 } 405 406 return 0; 407 408 err: 409 ntb_transport_unregister_client_dev(device_name); 410 411 return rc; 412 } 413 EXPORT_SYMBOL_GPL(ntb_transport_register_client_dev); 414 415 /** 416 * ntb_transport_register_client - Register NTB client driver 417 * @drv: NTB client driver to be registered 418 * 419 * Register an NTB client driver with the NTB transport layer 420 * 421 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 422 */ 423 int ntb_transport_register_client(struct ntb_transport_client *drv) 424 { 425 drv->driver.bus = &ntb_transport_bus; 426 427 if (list_empty(&ntb_transport_list)) 428 return -ENODEV; 429 430 return driver_register(&drv->driver); 431 } 432 EXPORT_SYMBOL_GPL(ntb_transport_register_client); 433 434 /** 435 * ntb_transport_unregister_client - Unregister NTB client driver 436 * @drv: NTB client driver to be unregistered 437 * 438 * Unregister an NTB client driver with the NTB transport layer 439 * 440 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 441 */ 442 void ntb_transport_unregister_client(struct ntb_transport_client *drv) 443 { 444 driver_unregister(&drv->driver); 445 } 446 EXPORT_SYMBOL_GPL(ntb_transport_unregister_client); 447 448 static ssize_t debugfs_read(struct file *filp, char __user *ubuf, size_t count, 449 loff_t *offp) 450 { 451 struct ntb_transport_qp *qp; 452 char *buf; 453 ssize_t ret, out_offset, out_count; 454 455 qp = filp->private_data; 456 457 if (!qp || !qp->link_is_up) 458 return 0; 459 460 out_count = 1000; 461 462 buf = kmalloc(out_count, GFP_KERNEL); 463 if (!buf) 464 return -ENOMEM; 465 466 out_offset = 0; 467 out_offset += snprintf(buf + out_offset, out_count - out_offset, 468 "\nNTB QP stats:\n\n"); 469 out_offset += snprintf(buf + out_offset, out_count - out_offset, 470 "rx_bytes - \t%llu\n", qp->rx_bytes); 471 out_offset += snprintf(buf + out_offset, out_count - out_offset, 472 "rx_pkts - \t%llu\n", qp->rx_pkts); 473 out_offset += snprintf(buf + out_offset, out_count - out_offset, 474 "rx_memcpy - \t%llu\n", qp->rx_memcpy); 475 out_offset += snprintf(buf + out_offset, out_count - out_offset, 476 "rx_async - \t%llu\n", qp->rx_async); 477 out_offset += snprintf(buf + out_offset, out_count - out_offset, 478 "rx_ring_empty - %llu\n", qp->rx_ring_empty); 479 out_offset += snprintf(buf + out_offset, out_count - out_offset, 480 "rx_err_no_buf - %llu\n", qp->rx_err_no_buf); 481 out_offset += snprintf(buf + out_offset, out_count - out_offset, 482 "rx_err_oflow - \t%llu\n", qp->rx_err_oflow); 483 out_offset += snprintf(buf + out_offset, out_count - out_offset, 484 "rx_err_ver - \t%llu\n", qp->rx_err_ver); 485 out_offset += snprintf(buf + out_offset, out_count - out_offset, 486 "rx_buff - \t0x%p\n", qp->rx_buff); 487 out_offset += snprintf(buf + out_offset, out_count - out_offset, 488 "rx_index - \t%u\n", qp->rx_index); 489 out_offset += snprintf(buf + out_offset, out_count - out_offset, 490 "rx_max_entry - \t%u\n", qp->rx_max_entry); 491 out_offset += snprintf(buf + out_offset, out_count - out_offset, 492 "rx_alloc_entry - \t%u\n\n", qp->rx_alloc_entry); 493 494 out_offset += snprintf(buf + out_offset, out_count - out_offset, 495 "tx_bytes - \t%llu\n", qp->tx_bytes); 496 out_offset += snprintf(buf + out_offset, out_count - out_offset, 497 "tx_pkts - \t%llu\n", qp->tx_pkts); 498 out_offset += snprintf(buf + out_offset, out_count - out_offset, 499 "tx_memcpy - \t%llu\n", qp->tx_memcpy); 500 out_offset += snprintf(buf + out_offset, out_count - out_offset, 501 "tx_async - \t%llu\n", qp->tx_async); 502 out_offset += snprintf(buf + out_offset, out_count - out_offset, 503 "tx_ring_full - \t%llu\n", qp->tx_ring_full); 504 out_offset += snprintf(buf + out_offset, out_count - out_offset, 505 "tx_err_no_buf - %llu\n", qp->tx_err_no_buf); 506 out_offset += snprintf(buf + out_offset, out_count - out_offset, 507 "tx_mw - \t0x%p\n", qp->tx_mw); 508 out_offset += snprintf(buf + out_offset, out_count - out_offset, 509 "tx_index (H) - \t%u\n", qp->tx_index); 510 out_offset += snprintf(buf + out_offset, out_count - out_offset, 511 "RRI (T) - \t%u\n", 512 qp->remote_rx_info->entry); 513 out_offset += snprintf(buf + out_offset, out_count - out_offset, 514 "tx_max_entry - \t%u\n", qp->tx_max_entry); 515 out_offset += snprintf(buf + out_offset, out_count - out_offset, 516 "free tx - \t%u\n", 517 ntb_transport_tx_free_entry(qp)); 518 519 out_offset += snprintf(buf + out_offset, out_count - out_offset, 520 "\n"); 521 out_offset += snprintf(buf + out_offset, out_count - out_offset, 522 "Using TX DMA - \t%s\n", 523 qp->tx_dma_chan ? "Yes" : "No"); 524 out_offset += snprintf(buf + out_offset, out_count - out_offset, 525 "Using RX DMA - \t%s\n", 526 qp->rx_dma_chan ? "Yes" : "No"); 527 out_offset += snprintf(buf + out_offset, out_count - out_offset, 528 "QP Link - \t%s\n", 529 qp->link_is_up ? "Up" : "Down"); 530 out_offset += snprintf(buf + out_offset, out_count - out_offset, 531 "\n"); 532 533 if (out_offset > out_count) 534 out_offset = out_count; 535 536 ret = simple_read_from_buffer(ubuf, count, offp, buf, out_offset); 537 kfree(buf); 538 return ret; 539 } 540 541 static const struct file_operations ntb_qp_debugfs_stats = { 542 .owner = THIS_MODULE, 543 .open = simple_open, 544 .read = debugfs_read, 545 }; 546 547 static void ntb_list_add(spinlock_t *lock, struct list_head *entry, 548 struct list_head *list) 549 { 550 unsigned long flags; 551 552 spin_lock_irqsave(lock, flags); 553 list_add_tail(entry, list); 554 spin_unlock_irqrestore(lock, flags); 555 } 556 557 static struct ntb_queue_entry *ntb_list_rm(spinlock_t *lock, 558 struct list_head *list) 559 { 560 struct ntb_queue_entry *entry; 561 unsigned long flags; 562 563 spin_lock_irqsave(lock, flags); 564 if (list_empty(list)) { 565 entry = NULL; 566 goto out; 567 } 568 entry = list_first_entry(list, struct ntb_queue_entry, entry); 569 list_del(&entry->entry); 570 571 out: 572 spin_unlock_irqrestore(lock, flags); 573 574 return entry; 575 } 576 577 static struct ntb_queue_entry *ntb_list_mv(spinlock_t *lock, 578 struct list_head *list, 579 struct list_head *to_list) 580 { 581 struct ntb_queue_entry *entry; 582 unsigned long flags; 583 584 spin_lock_irqsave(lock, flags); 585 586 if (list_empty(list)) { 587 entry = NULL; 588 } else { 589 entry = list_first_entry(list, struct ntb_queue_entry, entry); 590 list_move_tail(&entry->entry, to_list); 591 } 592 593 spin_unlock_irqrestore(lock, flags); 594 595 return entry; 596 } 597 598 static int ntb_transport_setup_qp_mw(struct ntb_transport_ctx *nt, 599 unsigned int qp_num) 600 { 601 struct ntb_transport_qp *qp = &nt->qp_vec[qp_num]; 602 struct ntb_transport_mw *mw; 603 struct ntb_dev *ndev = nt->ndev; 604 struct ntb_queue_entry *entry; 605 unsigned int rx_size, num_qps_mw; 606 unsigned int mw_num, mw_count, qp_count; 607 unsigned int i; 608 int node; 609 610 mw_count = nt->mw_count; 611 qp_count = nt->qp_count; 612 613 mw_num = QP_TO_MW(nt, qp_num); 614 mw = &nt->mw_vec[mw_num]; 615 616 if (!mw->virt_addr) 617 return -ENOMEM; 618 619 if (mw_num < qp_count % mw_count) 620 num_qps_mw = qp_count / mw_count + 1; 621 else 622 num_qps_mw = qp_count / mw_count; 623 624 rx_size = (unsigned int)mw->xlat_size / num_qps_mw; 625 qp->rx_buff = mw->virt_addr + rx_size * (qp_num / mw_count); 626 rx_size -= sizeof(struct ntb_rx_info); 627 628 qp->remote_rx_info = qp->rx_buff + rx_size; 629 630 /* Due to housekeeping, there must be atleast 2 buffs */ 631 qp->rx_max_frame = min(transport_mtu, rx_size / 2); 632 qp->rx_max_entry = rx_size / qp->rx_max_frame; 633 qp->rx_index = 0; 634 635 /* 636 * Checking to see if we have more entries than the default. 637 * We should add additional entries if that is the case so we 638 * can be in sync with the transport frames. 639 */ 640 node = dev_to_node(&ndev->dev); 641 for (i = qp->rx_alloc_entry; i < qp->rx_max_entry; i++) { 642 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 643 if (!entry) 644 return -ENOMEM; 645 646 entry->qp = qp; 647 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, 648 &qp->rx_free_q); 649 qp->rx_alloc_entry++; 650 } 651 652 qp->remote_rx_info->entry = qp->rx_max_entry - 1; 653 654 /* setup the hdr offsets with 0's */ 655 for (i = 0; i < qp->rx_max_entry; i++) { 656 void *offset = (qp->rx_buff + qp->rx_max_frame * (i + 1) - 657 sizeof(struct ntb_payload_header)); 658 memset(offset, 0, sizeof(struct ntb_payload_header)); 659 } 660 661 qp->rx_pkts = 0; 662 qp->tx_pkts = 0; 663 qp->tx_index = 0; 664 665 return 0; 666 } 667 668 static void ntb_free_mw(struct ntb_transport_ctx *nt, int num_mw) 669 { 670 struct ntb_transport_mw *mw = &nt->mw_vec[num_mw]; 671 struct pci_dev *pdev = nt->ndev->pdev; 672 673 if (!mw->virt_addr) 674 return; 675 676 ntb_mw_clear_trans(nt->ndev, PIDX, num_mw); 677 dma_free_coherent(&pdev->dev, mw->buff_size, 678 mw->virt_addr, mw->dma_addr); 679 mw->xlat_size = 0; 680 mw->buff_size = 0; 681 mw->virt_addr = NULL; 682 } 683 684 static int ntb_set_mw(struct ntb_transport_ctx *nt, int num_mw, 685 resource_size_t size) 686 { 687 struct ntb_transport_mw *mw = &nt->mw_vec[num_mw]; 688 struct pci_dev *pdev = nt->ndev->pdev; 689 size_t xlat_size, buff_size; 690 int rc; 691 692 if (!size) 693 return -EINVAL; 694 695 xlat_size = round_up(size, mw->xlat_align_size); 696 buff_size = round_up(size, mw->xlat_align); 697 698 /* No need to re-setup */ 699 if (mw->xlat_size == xlat_size) 700 return 0; 701 702 if (mw->buff_size) 703 ntb_free_mw(nt, num_mw); 704 705 /* Alloc memory for receiving data. Must be aligned */ 706 mw->xlat_size = xlat_size; 707 mw->buff_size = buff_size; 708 709 mw->virt_addr = dma_alloc_coherent(&pdev->dev, buff_size, 710 &mw->dma_addr, GFP_KERNEL); 711 if (!mw->virt_addr) { 712 mw->xlat_size = 0; 713 mw->buff_size = 0; 714 dev_err(&pdev->dev, "Unable to alloc MW buff of size %zu\n", 715 buff_size); 716 return -ENOMEM; 717 } 718 719 /* 720 * we must ensure that the memory address allocated is BAR size 721 * aligned in order for the XLAT register to take the value. This 722 * is a requirement of the hardware. It is recommended to setup CMA 723 * for BAR sizes equal or greater than 4MB. 724 */ 725 if (!IS_ALIGNED(mw->dma_addr, mw->xlat_align)) { 726 dev_err(&pdev->dev, "DMA memory %pad is not aligned\n", 727 &mw->dma_addr); 728 ntb_free_mw(nt, num_mw); 729 return -ENOMEM; 730 } 731 732 /* Notify HW the memory location of the receive buffer */ 733 rc = ntb_mw_set_trans(nt->ndev, PIDX, num_mw, mw->dma_addr, 734 mw->xlat_size); 735 if (rc) { 736 dev_err(&pdev->dev, "Unable to set mw%d translation", num_mw); 737 ntb_free_mw(nt, num_mw); 738 return -EIO; 739 } 740 741 return 0; 742 } 743 744 static void ntb_qp_link_down_reset(struct ntb_transport_qp *qp) 745 { 746 qp->link_is_up = false; 747 qp->active = false; 748 749 qp->tx_index = 0; 750 qp->rx_index = 0; 751 qp->rx_bytes = 0; 752 qp->rx_pkts = 0; 753 qp->rx_ring_empty = 0; 754 qp->rx_err_no_buf = 0; 755 qp->rx_err_oflow = 0; 756 qp->rx_err_ver = 0; 757 qp->rx_memcpy = 0; 758 qp->rx_async = 0; 759 qp->tx_bytes = 0; 760 qp->tx_pkts = 0; 761 qp->tx_ring_full = 0; 762 qp->tx_err_no_buf = 0; 763 qp->tx_memcpy = 0; 764 qp->tx_async = 0; 765 } 766 767 static void ntb_qp_link_cleanup(struct ntb_transport_qp *qp) 768 { 769 struct ntb_transport_ctx *nt = qp->transport; 770 struct pci_dev *pdev = nt->ndev->pdev; 771 772 dev_info(&pdev->dev, "qp %d: Link Cleanup\n", qp->qp_num); 773 774 cancel_delayed_work_sync(&qp->link_work); 775 ntb_qp_link_down_reset(qp); 776 777 if (qp->event_handler) 778 qp->event_handler(qp->cb_data, qp->link_is_up); 779 } 780 781 static void ntb_qp_link_cleanup_work(struct work_struct *work) 782 { 783 struct ntb_transport_qp *qp = container_of(work, 784 struct ntb_transport_qp, 785 link_cleanup); 786 struct ntb_transport_ctx *nt = qp->transport; 787 788 ntb_qp_link_cleanup(qp); 789 790 if (nt->link_is_up) 791 schedule_delayed_work(&qp->link_work, 792 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 793 } 794 795 static void ntb_qp_link_down(struct ntb_transport_qp *qp) 796 { 797 schedule_work(&qp->link_cleanup); 798 } 799 800 static void ntb_transport_link_cleanup(struct ntb_transport_ctx *nt) 801 { 802 struct ntb_transport_qp *qp; 803 u64 qp_bitmap_alloc; 804 unsigned int i, count; 805 806 qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free; 807 808 /* Pass along the info to any clients */ 809 for (i = 0; i < nt->qp_count; i++) 810 if (qp_bitmap_alloc & BIT_ULL(i)) { 811 qp = &nt->qp_vec[i]; 812 ntb_qp_link_cleanup(qp); 813 cancel_work_sync(&qp->link_cleanup); 814 cancel_delayed_work_sync(&qp->link_work); 815 } 816 817 if (!nt->link_is_up) 818 cancel_delayed_work_sync(&nt->link_work); 819 820 /* The scratchpad registers keep the values if the remote side 821 * goes down, blast them now to give them a sane value the next 822 * time they are accessed 823 */ 824 count = ntb_spad_count(nt->ndev); 825 for (i = 0; i < count; i++) 826 ntb_spad_write(nt->ndev, i, 0); 827 } 828 829 static void ntb_transport_link_cleanup_work(struct work_struct *work) 830 { 831 struct ntb_transport_ctx *nt = 832 container_of(work, struct ntb_transport_ctx, link_cleanup); 833 834 ntb_transport_link_cleanup(nt); 835 } 836 837 static void ntb_transport_event_callback(void *data) 838 { 839 struct ntb_transport_ctx *nt = data; 840 841 if (ntb_link_is_up(nt->ndev, NULL, NULL) == 1) 842 schedule_delayed_work(&nt->link_work, 0); 843 else 844 schedule_work(&nt->link_cleanup); 845 } 846 847 static void ntb_transport_link_work(struct work_struct *work) 848 { 849 struct ntb_transport_ctx *nt = 850 container_of(work, struct ntb_transport_ctx, link_work.work); 851 struct ntb_dev *ndev = nt->ndev; 852 struct pci_dev *pdev = ndev->pdev; 853 resource_size_t size; 854 u32 val; 855 int rc = 0, i, spad; 856 857 /* send the local info, in the opposite order of the way we read it */ 858 for (i = 0; i < nt->mw_count; i++) { 859 size = nt->mw_vec[i].phys_size; 860 861 if (max_mw_size && size > max_mw_size) 862 size = max_mw_size; 863 864 spad = MW0_SZ_HIGH + (i * 2); 865 ntb_peer_spad_write(ndev, PIDX, spad, upper_32_bits(size)); 866 867 spad = MW0_SZ_LOW + (i * 2); 868 ntb_peer_spad_write(ndev, PIDX, spad, lower_32_bits(size)); 869 } 870 871 ntb_peer_spad_write(ndev, PIDX, NUM_MWS, nt->mw_count); 872 873 ntb_peer_spad_write(ndev, PIDX, NUM_QPS, nt->qp_count); 874 875 ntb_peer_spad_write(ndev, PIDX, VERSION, NTB_TRANSPORT_VERSION); 876 877 /* Query the remote side for its info */ 878 val = ntb_spad_read(ndev, VERSION); 879 dev_dbg(&pdev->dev, "Remote version = %d\n", val); 880 if (val != NTB_TRANSPORT_VERSION) 881 goto out; 882 883 val = ntb_spad_read(ndev, NUM_QPS); 884 dev_dbg(&pdev->dev, "Remote max number of qps = %d\n", val); 885 if (val != nt->qp_count) 886 goto out; 887 888 val = ntb_spad_read(ndev, NUM_MWS); 889 dev_dbg(&pdev->dev, "Remote number of mws = %d\n", val); 890 if (val != nt->mw_count) 891 goto out; 892 893 for (i = 0; i < nt->mw_count; i++) { 894 u64 val64; 895 896 val = ntb_spad_read(ndev, MW0_SZ_HIGH + (i * 2)); 897 val64 = (u64)val << 32; 898 899 val = ntb_spad_read(ndev, MW0_SZ_LOW + (i * 2)); 900 val64 |= val; 901 902 dev_dbg(&pdev->dev, "Remote MW%d size = %#llx\n", i, val64); 903 904 rc = ntb_set_mw(nt, i, val64); 905 if (rc) 906 goto out1; 907 } 908 909 nt->link_is_up = true; 910 911 for (i = 0; i < nt->qp_count; i++) { 912 struct ntb_transport_qp *qp = &nt->qp_vec[i]; 913 914 ntb_transport_setup_qp_mw(nt, i); 915 916 if (qp->client_ready) 917 schedule_delayed_work(&qp->link_work, 0); 918 } 919 920 return; 921 922 out1: 923 for (i = 0; i < nt->mw_count; i++) 924 ntb_free_mw(nt, i); 925 926 /* if there's an actual failure, we should just bail */ 927 if (rc < 0) { 928 ntb_link_disable(ndev); 929 return; 930 } 931 932 out: 933 if (ntb_link_is_up(ndev, NULL, NULL) == 1) 934 schedule_delayed_work(&nt->link_work, 935 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 936 } 937 938 static void ntb_qp_link_work(struct work_struct *work) 939 { 940 struct ntb_transport_qp *qp = container_of(work, 941 struct ntb_transport_qp, 942 link_work.work); 943 struct pci_dev *pdev = qp->ndev->pdev; 944 struct ntb_transport_ctx *nt = qp->transport; 945 int val; 946 947 WARN_ON(!nt->link_is_up); 948 949 val = ntb_spad_read(nt->ndev, QP_LINKS); 950 951 ntb_peer_spad_write(nt->ndev, PIDX, QP_LINKS, val | BIT(qp->qp_num)); 952 953 /* query remote spad for qp ready bits */ 954 dev_dbg_ratelimited(&pdev->dev, "Remote QP link status = %x\n", val); 955 956 /* See if the remote side is up */ 957 if (val & BIT(qp->qp_num)) { 958 dev_info(&pdev->dev, "qp %d: Link Up\n", qp->qp_num); 959 qp->link_is_up = true; 960 qp->active = true; 961 962 if (qp->event_handler) 963 qp->event_handler(qp->cb_data, qp->link_is_up); 964 965 if (qp->active) 966 tasklet_schedule(&qp->rxc_db_work); 967 } else if (nt->link_is_up) 968 schedule_delayed_work(&qp->link_work, 969 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 970 } 971 972 static int ntb_transport_init_queue(struct ntb_transport_ctx *nt, 973 unsigned int qp_num) 974 { 975 struct ntb_transport_qp *qp; 976 phys_addr_t mw_base; 977 resource_size_t mw_size; 978 unsigned int num_qps_mw, tx_size; 979 unsigned int mw_num, mw_count, qp_count; 980 u64 qp_offset; 981 982 mw_count = nt->mw_count; 983 qp_count = nt->qp_count; 984 985 mw_num = QP_TO_MW(nt, qp_num); 986 987 qp = &nt->qp_vec[qp_num]; 988 qp->qp_num = qp_num; 989 qp->transport = nt; 990 qp->ndev = nt->ndev; 991 qp->client_ready = false; 992 qp->event_handler = NULL; 993 ntb_qp_link_down_reset(qp); 994 995 if (mw_num < qp_count % mw_count) 996 num_qps_mw = qp_count / mw_count + 1; 997 else 998 num_qps_mw = qp_count / mw_count; 999 1000 mw_base = nt->mw_vec[mw_num].phys_addr; 1001 mw_size = nt->mw_vec[mw_num].phys_size; 1002 1003 tx_size = (unsigned int)mw_size / num_qps_mw; 1004 qp_offset = tx_size * (qp_num / mw_count); 1005 1006 qp->tx_mw = nt->mw_vec[mw_num].vbase + qp_offset; 1007 if (!qp->tx_mw) 1008 return -EINVAL; 1009 1010 qp->tx_mw_phys = mw_base + qp_offset; 1011 if (!qp->tx_mw_phys) 1012 return -EINVAL; 1013 1014 tx_size -= sizeof(struct ntb_rx_info); 1015 qp->rx_info = qp->tx_mw + tx_size; 1016 1017 /* Due to housekeeping, there must be atleast 2 buffs */ 1018 qp->tx_max_frame = min(transport_mtu, tx_size / 2); 1019 qp->tx_max_entry = tx_size / qp->tx_max_frame; 1020 1021 if (nt->debugfs_node_dir) { 1022 char debugfs_name[4]; 1023 1024 snprintf(debugfs_name, 4, "qp%d", qp_num); 1025 qp->debugfs_dir = debugfs_create_dir(debugfs_name, 1026 nt->debugfs_node_dir); 1027 1028 qp->debugfs_stats = debugfs_create_file("stats", S_IRUSR, 1029 qp->debugfs_dir, qp, 1030 &ntb_qp_debugfs_stats); 1031 } else { 1032 qp->debugfs_dir = NULL; 1033 qp->debugfs_stats = NULL; 1034 } 1035 1036 INIT_DELAYED_WORK(&qp->link_work, ntb_qp_link_work); 1037 INIT_WORK(&qp->link_cleanup, ntb_qp_link_cleanup_work); 1038 1039 spin_lock_init(&qp->ntb_rx_q_lock); 1040 spin_lock_init(&qp->ntb_tx_free_q_lock); 1041 1042 INIT_LIST_HEAD(&qp->rx_post_q); 1043 INIT_LIST_HEAD(&qp->rx_pend_q); 1044 INIT_LIST_HEAD(&qp->rx_free_q); 1045 INIT_LIST_HEAD(&qp->tx_free_q); 1046 1047 tasklet_init(&qp->rxc_db_work, ntb_transport_rxc_db, 1048 (unsigned long)qp); 1049 1050 return 0; 1051 } 1052 1053 static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev) 1054 { 1055 struct ntb_transport_ctx *nt; 1056 struct ntb_transport_mw *mw; 1057 unsigned int mw_count, qp_count, spad_count, max_mw_count_for_spads; 1058 u64 qp_bitmap; 1059 int node; 1060 int rc, i; 1061 1062 mw_count = ntb_peer_mw_count(ndev); 1063 1064 if (!ndev->ops->mw_set_trans) { 1065 dev_err(&ndev->dev, "Inbound MW based NTB API is required\n"); 1066 return -EINVAL; 1067 } 1068 1069 if (ntb_db_is_unsafe(ndev)) 1070 dev_dbg(&ndev->dev, 1071 "doorbell is unsafe, proceed anyway...\n"); 1072 if (ntb_spad_is_unsafe(ndev)) 1073 dev_dbg(&ndev->dev, 1074 "scratchpad is unsafe, proceed anyway...\n"); 1075 1076 if (ntb_peer_port_count(ndev) != NTB_DEF_PEER_CNT) 1077 dev_warn(&ndev->dev, "Multi-port NTB devices unsupported\n"); 1078 1079 node = dev_to_node(&ndev->dev); 1080 1081 nt = kzalloc_node(sizeof(*nt), GFP_KERNEL, node); 1082 if (!nt) 1083 return -ENOMEM; 1084 1085 nt->ndev = ndev; 1086 spad_count = ntb_spad_count(ndev); 1087 1088 /* Limit the MW's based on the availability of scratchpads */ 1089 1090 if (spad_count < NTB_TRANSPORT_MIN_SPADS) { 1091 nt->mw_count = 0; 1092 rc = -EINVAL; 1093 goto err; 1094 } 1095 1096 max_mw_count_for_spads = (spad_count - MW0_SZ_HIGH) / 2; 1097 nt->mw_count = min(mw_count, max_mw_count_for_spads); 1098 1099 nt->mw_vec = kzalloc_node(mw_count * sizeof(*nt->mw_vec), 1100 GFP_KERNEL, node); 1101 if (!nt->mw_vec) { 1102 rc = -ENOMEM; 1103 goto err; 1104 } 1105 1106 for (i = 0; i < mw_count; i++) { 1107 mw = &nt->mw_vec[i]; 1108 1109 rc = ntb_mw_get_align(ndev, PIDX, i, &mw->xlat_align, 1110 &mw->xlat_align_size, NULL); 1111 if (rc) 1112 goto err1; 1113 1114 rc = ntb_peer_mw_get_addr(ndev, i, &mw->phys_addr, 1115 &mw->phys_size); 1116 if (rc) 1117 goto err1; 1118 1119 mw->vbase = ioremap_wc(mw->phys_addr, mw->phys_size); 1120 if (!mw->vbase) { 1121 rc = -ENOMEM; 1122 goto err1; 1123 } 1124 1125 mw->buff_size = 0; 1126 mw->xlat_size = 0; 1127 mw->virt_addr = NULL; 1128 mw->dma_addr = 0; 1129 } 1130 1131 qp_bitmap = ntb_db_valid_mask(ndev); 1132 1133 qp_count = ilog2(qp_bitmap); 1134 if (max_num_clients && max_num_clients < qp_count) 1135 qp_count = max_num_clients; 1136 else if (nt->mw_count < qp_count) 1137 qp_count = nt->mw_count; 1138 1139 qp_bitmap &= BIT_ULL(qp_count) - 1; 1140 1141 nt->qp_count = qp_count; 1142 nt->qp_bitmap = qp_bitmap; 1143 nt->qp_bitmap_free = qp_bitmap; 1144 1145 nt->qp_vec = kzalloc_node(qp_count * sizeof(*nt->qp_vec), 1146 GFP_KERNEL, node); 1147 if (!nt->qp_vec) { 1148 rc = -ENOMEM; 1149 goto err1; 1150 } 1151 1152 if (nt_debugfs_dir) { 1153 nt->debugfs_node_dir = 1154 debugfs_create_dir(pci_name(ndev->pdev), 1155 nt_debugfs_dir); 1156 } 1157 1158 for (i = 0; i < qp_count; i++) { 1159 rc = ntb_transport_init_queue(nt, i); 1160 if (rc) 1161 goto err2; 1162 } 1163 1164 INIT_DELAYED_WORK(&nt->link_work, ntb_transport_link_work); 1165 INIT_WORK(&nt->link_cleanup, ntb_transport_link_cleanup_work); 1166 1167 rc = ntb_set_ctx(ndev, nt, &ntb_transport_ops); 1168 if (rc) 1169 goto err2; 1170 1171 INIT_LIST_HEAD(&nt->client_devs); 1172 rc = ntb_bus_init(nt); 1173 if (rc) 1174 goto err3; 1175 1176 nt->link_is_up = false; 1177 ntb_link_enable(ndev, NTB_SPEED_AUTO, NTB_WIDTH_AUTO); 1178 ntb_link_event(ndev); 1179 1180 return 0; 1181 1182 err3: 1183 ntb_clear_ctx(ndev); 1184 err2: 1185 kfree(nt->qp_vec); 1186 err1: 1187 while (i--) { 1188 mw = &nt->mw_vec[i]; 1189 iounmap(mw->vbase); 1190 } 1191 kfree(nt->mw_vec); 1192 err: 1193 kfree(nt); 1194 return rc; 1195 } 1196 1197 static void ntb_transport_free(struct ntb_client *self, struct ntb_dev *ndev) 1198 { 1199 struct ntb_transport_ctx *nt = ndev->ctx; 1200 struct ntb_transport_qp *qp; 1201 u64 qp_bitmap_alloc; 1202 int i; 1203 1204 ntb_transport_link_cleanup(nt); 1205 cancel_work_sync(&nt->link_cleanup); 1206 cancel_delayed_work_sync(&nt->link_work); 1207 1208 qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free; 1209 1210 /* verify that all the qp's are freed */ 1211 for (i = 0; i < nt->qp_count; i++) { 1212 qp = &nt->qp_vec[i]; 1213 if (qp_bitmap_alloc & BIT_ULL(i)) 1214 ntb_transport_free_queue(qp); 1215 debugfs_remove_recursive(qp->debugfs_dir); 1216 } 1217 1218 ntb_link_disable(ndev); 1219 ntb_clear_ctx(ndev); 1220 1221 ntb_bus_remove(nt); 1222 1223 for (i = nt->mw_count; i--; ) { 1224 ntb_free_mw(nt, i); 1225 iounmap(nt->mw_vec[i].vbase); 1226 } 1227 1228 kfree(nt->qp_vec); 1229 kfree(nt->mw_vec); 1230 kfree(nt); 1231 } 1232 1233 static void ntb_complete_rxc(struct ntb_transport_qp *qp) 1234 { 1235 struct ntb_queue_entry *entry; 1236 void *cb_data; 1237 unsigned int len; 1238 unsigned long irqflags; 1239 1240 spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags); 1241 1242 while (!list_empty(&qp->rx_post_q)) { 1243 entry = list_first_entry(&qp->rx_post_q, 1244 struct ntb_queue_entry, entry); 1245 if (!(entry->flags & DESC_DONE_FLAG)) 1246 break; 1247 1248 entry->rx_hdr->flags = 0; 1249 iowrite32(entry->rx_index, &qp->rx_info->entry); 1250 1251 cb_data = entry->cb_data; 1252 len = entry->len; 1253 1254 list_move_tail(&entry->entry, &qp->rx_free_q); 1255 1256 spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags); 1257 1258 if (qp->rx_handler && qp->client_ready) 1259 qp->rx_handler(qp, qp->cb_data, cb_data, len); 1260 1261 spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags); 1262 } 1263 1264 spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags); 1265 } 1266 1267 static void ntb_rx_copy_callback(void *data, 1268 const struct dmaengine_result *res) 1269 { 1270 struct ntb_queue_entry *entry = data; 1271 1272 /* we need to check DMA results if we are using DMA */ 1273 if (res) { 1274 enum dmaengine_tx_result dma_err = res->result; 1275 1276 switch (dma_err) { 1277 case DMA_TRANS_READ_FAILED: 1278 case DMA_TRANS_WRITE_FAILED: 1279 entry->errors++; 1280 case DMA_TRANS_ABORTED: 1281 { 1282 struct ntb_transport_qp *qp = entry->qp; 1283 void *offset = qp->rx_buff + qp->rx_max_frame * 1284 qp->rx_index; 1285 1286 ntb_memcpy_rx(entry, offset); 1287 qp->rx_memcpy++; 1288 return; 1289 } 1290 1291 case DMA_TRANS_NOERROR: 1292 default: 1293 break; 1294 } 1295 } 1296 1297 entry->flags |= DESC_DONE_FLAG; 1298 1299 ntb_complete_rxc(entry->qp); 1300 } 1301 1302 static void ntb_memcpy_rx(struct ntb_queue_entry *entry, void *offset) 1303 { 1304 void *buf = entry->buf; 1305 size_t len = entry->len; 1306 1307 memcpy(buf, offset, len); 1308 1309 /* Ensure that the data is fully copied out before clearing the flag */ 1310 wmb(); 1311 1312 ntb_rx_copy_callback(entry, NULL); 1313 } 1314 1315 static int ntb_async_rx_submit(struct ntb_queue_entry *entry, void *offset) 1316 { 1317 struct dma_async_tx_descriptor *txd; 1318 struct ntb_transport_qp *qp = entry->qp; 1319 struct dma_chan *chan = qp->rx_dma_chan; 1320 struct dma_device *device; 1321 size_t pay_off, buff_off, len; 1322 struct dmaengine_unmap_data *unmap; 1323 dma_cookie_t cookie; 1324 void *buf = entry->buf; 1325 1326 len = entry->len; 1327 device = chan->device; 1328 pay_off = (size_t)offset & ~PAGE_MASK; 1329 buff_off = (size_t)buf & ~PAGE_MASK; 1330 1331 if (!is_dma_copy_aligned(device, pay_off, buff_off, len)) 1332 goto err; 1333 1334 unmap = dmaengine_get_unmap_data(device->dev, 2, GFP_NOWAIT); 1335 if (!unmap) 1336 goto err; 1337 1338 unmap->len = len; 1339 unmap->addr[0] = dma_map_page(device->dev, virt_to_page(offset), 1340 pay_off, len, DMA_TO_DEVICE); 1341 if (dma_mapping_error(device->dev, unmap->addr[0])) 1342 goto err_get_unmap; 1343 1344 unmap->to_cnt = 1; 1345 1346 unmap->addr[1] = dma_map_page(device->dev, virt_to_page(buf), 1347 buff_off, len, DMA_FROM_DEVICE); 1348 if (dma_mapping_error(device->dev, unmap->addr[1])) 1349 goto err_get_unmap; 1350 1351 unmap->from_cnt = 1; 1352 1353 txd = device->device_prep_dma_memcpy(chan, unmap->addr[1], 1354 unmap->addr[0], len, 1355 DMA_PREP_INTERRUPT); 1356 if (!txd) 1357 goto err_get_unmap; 1358 1359 txd->callback_result = ntb_rx_copy_callback; 1360 txd->callback_param = entry; 1361 dma_set_unmap(txd, unmap); 1362 1363 cookie = dmaengine_submit(txd); 1364 if (dma_submit_error(cookie)) 1365 goto err_set_unmap; 1366 1367 dmaengine_unmap_put(unmap); 1368 1369 qp->last_cookie = cookie; 1370 1371 qp->rx_async++; 1372 1373 return 0; 1374 1375 err_set_unmap: 1376 dmaengine_unmap_put(unmap); 1377 err_get_unmap: 1378 dmaengine_unmap_put(unmap); 1379 err: 1380 return -ENXIO; 1381 } 1382 1383 static void ntb_async_rx(struct ntb_queue_entry *entry, void *offset) 1384 { 1385 struct ntb_transport_qp *qp = entry->qp; 1386 struct dma_chan *chan = qp->rx_dma_chan; 1387 int res; 1388 1389 if (!chan) 1390 goto err; 1391 1392 if (entry->len < copy_bytes) 1393 goto err; 1394 1395 res = ntb_async_rx_submit(entry, offset); 1396 if (res < 0) 1397 goto err; 1398 1399 if (!entry->retries) 1400 qp->rx_async++; 1401 1402 return; 1403 1404 err: 1405 ntb_memcpy_rx(entry, offset); 1406 qp->rx_memcpy++; 1407 } 1408 1409 static int ntb_process_rxc(struct ntb_transport_qp *qp) 1410 { 1411 struct ntb_payload_header *hdr; 1412 struct ntb_queue_entry *entry; 1413 void *offset; 1414 1415 offset = qp->rx_buff + qp->rx_max_frame * qp->rx_index; 1416 hdr = offset + qp->rx_max_frame - sizeof(struct ntb_payload_header); 1417 1418 dev_dbg(&qp->ndev->pdev->dev, "qp %d: RX ver %u len %d flags %x\n", 1419 qp->qp_num, hdr->ver, hdr->len, hdr->flags); 1420 1421 if (!(hdr->flags & DESC_DONE_FLAG)) { 1422 dev_dbg(&qp->ndev->pdev->dev, "done flag not set\n"); 1423 qp->rx_ring_empty++; 1424 return -EAGAIN; 1425 } 1426 1427 if (hdr->flags & LINK_DOWN_FLAG) { 1428 dev_dbg(&qp->ndev->pdev->dev, "link down flag set\n"); 1429 ntb_qp_link_down(qp); 1430 hdr->flags = 0; 1431 return -EAGAIN; 1432 } 1433 1434 if (hdr->ver != (u32)qp->rx_pkts) { 1435 dev_dbg(&qp->ndev->pdev->dev, 1436 "version mismatch, expected %llu - got %u\n", 1437 qp->rx_pkts, hdr->ver); 1438 qp->rx_err_ver++; 1439 return -EIO; 1440 } 1441 1442 entry = ntb_list_mv(&qp->ntb_rx_q_lock, &qp->rx_pend_q, &qp->rx_post_q); 1443 if (!entry) { 1444 dev_dbg(&qp->ndev->pdev->dev, "no receive buffer\n"); 1445 qp->rx_err_no_buf++; 1446 return -EAGAIN; 1447 } 1448 1449 entry->rx_hdr = hdr; 1450 entry->rx_index = qp->rx_index; 1451 1452 if (hdr->len > entry->len) { 1453 dev_dbg(&qp->ndev->pdev->dev, 1454 "receive buffer overflow! Wanted %d got %d\n", 1455 hdr->len, entry->len); 1456 qp->rx_err_oflow++; 1457 1458 entry->len = -EIO; 1459 entry->flags |= DESC_DONE_FLAG; 1460 1461 ntb_complete_rxc(qp); 1462 } else { 1463 dev_dbg(&qp->ndev->pdev->dev, 1464 "RX OK index %u ver %u size %d into buf size %d\n", 1465 qp->rx_index, hdr->ver, hdr->len, entry->len); 1466 1467 qp->rx_bytes += hdr->len; 1468 qp->rx_pkts++; 1469 1470 entry->len = hdr->len; 1471 1472 ntb_async_rx(entry, offset); 1473 } 1474 1475 qp->rx_index++; 1476 qp->rx_index %= qp->rx_max_entry; 1477 1478 return 0; 1479 } 1480 1481 static void ntb_transport_rxc_db(unsigned long data) 1482 { 1483 struct ntb_transport_qp *qp = (void *)data; 1484 int rc, i; 1485 1486 dev_dbg(&qp->ndev->pdev->dev, "%s: doorbell %d received\n", 1487 __func__, qp->qp_num); 1488 1489 /* Limit the number of packets processed in a single interrupt to 1490 * provide fairness to others 1491 */ 1492 for (i = 0; i < qp->rx_max_entry; i++) { 1493 rc = ntb_process_rxc(qp); 1494 if (rc) 1495 break; 1496 } 1497 1498 if (i && qp->rx_dma_chan) 1499 dma_async_issue_pending(qp->rx_dma_chan); 1500 1501 if (i == qp->rx_max_entry) { 1502 /* there is more work to do */ 1503 if (qp->active) 1504 tasklet_schedule(&qp->rxc_db_work); 1505 } else if (ntb_db_read(qp->ndev) & BIT_ULL(qp->qp_num)) { 1506 /* the doorbell bit is set: clear it */ 1507 ntb_db_clear(qp->ndev, BIT_ULL(qp->qp_num)); 1508 /* ntb_db_read ensures ntb_db_clear write is committed */ 1509 ntb_db_read(qp->ndev); 1510 1511 /* an interrupt may have arrived between finishing 1512 * ntb_process_rxc and clearing the doorbell bit: 1513 * there might be some more work to do. 1514 */ 1515 if (qp->active) 1516 tasklet_schedule(&qp->rxc_db_work); 1517 } 1518 } 1519 1520 static void ntb_tx_copy_callback(void *data, 1521 const struct dmaengine_result *res) 1522 { 1523 struct ntb_queue_entry *entry = data; 1524 struct ntb_transport_qp *qp = entry->qp; 1525 struct ntb_payload_header __iomem *hdr = entry->tx_hdr; 1526 1527 /* we need to check DMA results if we are using DMA */ 1528 if (res) { 1529 enum dmaengine_tx_result dma_err = res->result; 1530 1531 switch (dma_err) { 1532 case DMA_TRANS_READ_FAILED: 1533 case DMA_TRANS_WRITE_FAILED: 1534 entry->errors++; 1535 case DMA_TRANS_ABORTED: 1536 { 1537 void __iomem *offset = 1538 qp->tx_mw + qp->tx_max_frame * 1539 entry->tx_index; 1540 1541 /* resubmit via CPU */ 1542 ntb_memcpy_tx(entry, offset); 1543 qp->tx_memcpy++; 1544 return; 1545 } 1546 1547 case DMA_TRANS_NOERROR: 1548 default: 1549 break; 1550 } 1551 } 1552 1553 iowrite32(entry->flags | DESC_DONE_FLAG, &hdr->flags); 1554 1555 ntb_peer_db_set(qp->ndev, BIT_ULL(qp->qp_num)); 1556 1557 /* The entry length can only be zero if the packet is intended to be a 1558 * "link down" or similar. Since no payload is being sent in these 1559 * cases, there is nothing to add to the completion queue. 1560 */ 1561 if (entry->len > 0) { 1562 qp->tx_bytes += entry->len; 1563 1564 if (qp->tx_handler) 1565 qp->tx_handler(qp, qp->cb_data, entry->cb_data, 1566 entry->len); 1567 } 1568 1569 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, &qp->tx_free_q); 1570 } 1571 1572 static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset) 1573 { 1574 #ifdef ARCH_HAS_NOCACHE_UACCESS 1575 /* 1576 * Using non-temporal mov to improve performance on non-cached 1577 * writes, even though we aren't actually copying from user space. 1578 */ 1579 __copy_from_user_inatomic_nocache(offset, entry->buf, entry->len); 1580 #else 1581 memcpy_toio(offset, entry->buf, entry->len); 1582 #endif 1583 1584 /* Ensure that the data is fully copied out before setting the flags */ 1585 wmb(); 1586 1587 ntb_tx_copy_callback(entry, NULL); 1588 } 1589 1590 static int ntb_async_tx_submit(struct ntb_transport_qp *qp, 1591 struct ntb_queue_entry *entry) 1592 { 1593 struct dma_async_tx_descriptor *txd; 1594 struct dma_chan *chan = qp->tx_dma_chan; 1595 struct dma_device *device; 1596 size_t len = entry->len; 1597 void *buf = entry->buf; 1598 size_t dest_off, buff_off; 1599 struct dmaengine_unmap_data *unmap; 1600 dma_addr_t dest; 1601 dma_cookie_t cookie; 1602 1603 device = chan->device; 1604 dest = qp->tx_mw_phys + qp->tx_max_frame * entry->tx_index; 1605 buff_off = (size_t)buf & ~PAGE_MASK; 1606 dest_off = (size_t)dest & ~PAGE_MASK; 1607 1608 if (!is_dma_copy_aligned(device, buff_off, dest_off, len)) 1609 goto err; 1610 1611 unmap = dmaengine_get_unmap_data(device->dev, 1, GFP_NOWAIT); 1612 if (!unmap) 1613 goto err; 1614 1615 unmap->len = len; 1616 unmap->addr[0] = dma_map_page(device->dev, virt_to_page(buf), 1617 buff_off, len, DMA_TO_DEVICE); 1618 if (dma_mapping_error(device->dev, unmap->addr[0])) 1619 goto err_get_unmap; 1620 1621 unmap->to_cnt = 1; 1622 1623 txd = device->device_prep_dma_memcpy(chan, dest, unmap->addr[0], len, 1624 DMA_PREP_INTERRUPT); 1625 if (!txd) 1626 goto err_get_unmap; 1627 1628 txd->callback_result = ntb_tx_copy_callback; 1629 txd->callback_param = entry; 1630 dma_set_unmap(txd, unmap); 1631 1632 cookie = dmaengine_submit(txd); 1633 if (dma_submit_error(cookie)) 1634 goto err_set_unmap; 1635 1636 dmaengine_unmap_put(unmap); 1637 1638 dma_async_issue_pending(chan); 1639 1640 return 0; 1641 err_set_unmap: 1642 dmaengine_unmap_put(unmap); 1643 err_get_unmap: 1644 dmaengine_unmap_put(unmap); 1645 err: 1646 return -ENXIO; 1647 } 1648 1649 static void ntb_async_tx(struct ntb_transport_qp *qp, 1650 struct ntb_queue_entry *entry) 1651 { 1652 struct ntb_payload_header __iomem *hdr; 1653 struct dma_chan *chan = qp->tx_dma_chan; 1654 void __iomem *offset; 1655 int res; 1656 1657 entry->tx_index = qp->tx_index; 1658 offset = qp->tx_mw + qp->tx_max_frame * entry->tx_index; 1659 hdr = offset + qp->tx_max_frame - sizeof(struct ntb_payload_header); 1660 entry->tx_hdr = hdr; 1661 1662 iowrite32(entry->len, &hdr->len); 1663 iowrite32((u32)qp->tx_pkts, &hdr->ver); 1664 1665 if (!chan) 1666 goto err; 1667 1668 if (entry->len < copy_bytes) 1669 goto err; 1670 1671 res = ntb_async_tx_submit(qp, entry); 1672 if (res < 0) 1673 goto err; 1674 1675 if (!entry->retries) 1676 qp->tx_async++; 1677 1678 return; 1679 1680 err: 1681 ntb_memcpy_tx(entry, offset); 1682 qp->tx_memcpy++; 1683 } 1684 1685 static int ntb_process_tx(struct ntb_transport_qp *qp, 1686 struct ntb_queue_entry *entry) 1687 { 1688 if (qp->tx_index == qp->remote_rx_info->entry) { 1689 qp->tx_ring_full++; 1690 return -EAGAIN; 1691 } 1692 1693 if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) { 1694 if (qp->tx_handler) 1695 qp->tx_handler(qp, qp->cb_data, NULL, -EIO); 1696 1697 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1698 &qp->tx_free_q); 1699 return 0; 1700 } 1701 1702 ntb_async_tx(qp, entry); 1703 1704 qp->tx_index++; 1705 qp->tx_index %= qp->tx_max_entry; 1706 1707 qp->tx_pkts++; 1708 1709 return 0; 1710 } 1711 1712 static void ntb_send_link_down(struct ntb_transport_qp *qp) 1713 { 1714 struct pci_dev *pdev = qp->ndev->pdev; 1715 struct ntb_queue_entry *entry; 1716 int i, rc; 1717 1718 if (!qp->link_is_up) 1719 return; 1720 1721 dev_info(&pdev->dev, "qp %d: Send Link Down\n", qp->qp_num); 1722 1723 for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) { 1724 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); 1725 if (entry) 1726 break; 1727 msleep(100); 1728 } 1729 1730 if (!entry) 1731 return; 1732 1733 entry->cb_data = NULL; 1734 entry->buf = NULL; 1735 entry->len = 0; 1736 entry->flags = LINK_DOWN_FLAG; 1737 1738 rc = ntb_process_tx(qp, entry); 1739 if (rc) 1740 dev_err(&pdev->dev, "ntb: QP%d unable to send linkdown msg\n", 1741 qp->qp_num); 1742 1743 ntb_qp_link_down_reset(qp); 1744 } 1745 1746 static bool ntb_dma_filter_fn(struct dma_chan *chan, void *node) 1747 { 1748 return dev_to_node(&chan->dev->device) == (int)(unsigned long)node; 1749 } 1750 1751 /** 1752 * ntb_transport_create_queue - Create a new NTB transport layer queue 1753 * @rx_handler: receive callback function 1754 * @tx_handler: transmit callback function 1755 * @event_handler: event callback function 1756 * 1757 * Create a new NTB transport layer queue and provide the queue with a callback 1758 * routine for both transmit and receive. The receive callback routine will be 1759 * used to pass up data when the transport has received it on the queue. The 1760 * transmit callback routine will be called when the transport has completed the 1761 * transmission of the data on the queue and the data is ready to be freed. 1762 * 1763 * RETURNS: pointer to newly created ntb_queue, NULL on error. 1764 */ 1765 struct ntb_transport_qp * 1766 ntb_transport_create_queue(void *data, struct device *client_dev, 1767 const struct ntb_queue_handlers *handlers) 1768 { 1769 struct ntb_dev *ndev; 1770 struct pci_dev *pdev; 1771 struct ntb_transport_ctx *nt; 1772 struct ntb_queue_entry *entry; 1773 struct ntb_transport_qp *qp; 1774 u64 qp_bit; 1775 unsigned int free_queue; 1776 dma_cap_mask_t dma_mask; 1777 int node; 1778 int i; 1779 1780 ndev = dev_ntb(client_dev->parent); 1781 pdev = ndev->pdev; 1782 nt = ndev->ctx; 1783 1784 node = dev_to_node(&ndev->dev); 1785 1786 free_queue = ffs(nt->qp_bitmap_free); 1787 if (!free_queue) 1788 goto err; 1789 1790 /* decrement free_queue to make it zero based */ 1791 free_queue--; 1792 1793 qp = &nt->qp_vec[free_queue]; 1794 qp_bit = BIT_ULL(qp->qp_num); 1795 1796 nt->qp_bitmap_free &= ~qp_bit; 1797 1798 qp->cb_data = data; 1799 qp->rx_handler = handlers->rx_handler; 1800 qp->tx_handler = handlers->tx_handler; 1801 qp->event_handler = handlers->event_handler; 1802 1803 dma_cap_zero(dma_mask); 1804 dma_cap_set(DMA_MEMCPY, dma_mask); 1805 1806 if (use_dma) { 1807 qp->tx_dma_chan = 1808 dma_request_channel(dma_mask, ntb_dma_filter_fn, 1809 (void *)(unsigned long)node); 1810 if (!qp->tx_dma_chan) 1811 dev_info(&pdev->dev, "Unable to allocate TX DMA channel\n"); 1812 1813 qp->rx_dma_chan = 1814 dma_request_channel(dma_mask, ntb_dma_filter_fn, 1815 (void *)(unsigned long)node); 1816 if (!qp->rx_dma_chan) 1817 dev_info(&pdev->dev, "Unable to allocate RX DMA channel\n"); 1818 } else { 1819 qp->tx_dma_chan = NULL; 1820 qp->rx_dma_chan = NULL; 1821 } 1822 1823 dev_dbg(&pdev->dev, "Using %s memcpy for TX\n", 1824 qp->tx_dma_chan ? "DMA" : "CPU"); 1825 1826 dev_dbg(&pdev->dev, "Using %s memcpy for RX\n", 1827 qp->rx_dma_chan ? "DMA" : "CPU"); 1828 1829 for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) { 1830 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 1831 if (!entry) 1832 goto err1; 1833 1834 entry->qp = qp; 1835 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, 1836 &qp->rx_free_q); 1837 } 1838 qp->rx_alloc_entry = NTB_QP_DEF_NUM_ENTRIES; 1839 1840 for (i = 0; i < qp->tx_max_entry; i++) { 1841 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 1842 if (!entry) 1843 goto err2; 1844 1845 entry->qp = qp; 1846 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1847 &qp->tx_free_q); 1848 } 1849 1850 ntb_db_clear(qp->ndev, qp_bit); 1851 ntb_db_clear_mask(qp->ndev, qp_bit); 1852 1853 dev_info(&pdev->dev, "NTB Transport QP %d created\n", qp->qp_num); 1854 1855 return qp; 1856 1857 err2: 1858 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) 1859 kfree(entry); 1860 err1: 1861 qp->rx_alloc_entry = 0; 1862 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q))) 1863 kfree(entry); 1864 if (qp->tx_dma_chan) 1865 dma_release_channel(qp->tx_dma_chan); 1866 if (qp->rx_dma_chan) 1867 dma_release_channel(qp->rx_dma_chan); 1868 nt->qp_bitmap_free |= qp_bit; 1869 err: 1870 return NULL; 1871 } 1872 EXPORT_SYMBOL_GPL(ntb_transport_create_queue); 1873 1874 /** 1875 * ntb_transport_free_queue - Frees NTB transport queue 1876 * @qp: NTB queue to be freed 1877 * 1878 * Frees NTB transport queue 1879 */ 1880 void ntb_transport_free_queue(struct ntb_transport_qp *qp) 1881 { 1882 struct pci_dev *pdev; 1883 struct ntb_queue_entry *entry; 1884 u64 qp_bit; 1885 1886 if (!qp) 1887 return; 1888 1889 pdev = qp->ndev->pdev; 1890 1891 qp->active = false; 1892 1893 if (qp->tx_dma_chan) { 1894 struct dma_chan *chan = qp->tx_dma_chan; 1895 /* Putting the dma_chan to NULL will force any new traffic to be 1896 * processed by the CPU instead of the DAM engine 1897 */ 1898 qp->tx_dma_chan = NULL; 1899 1900 /* Try to be nice and wait for any queued DMA engine 1901 * transactions to process before smashing it with a rock 1902 */ 1903 dma_sync_wait(chan, qp->last_cookie); 1904 dmaengine_terminate_all(chan); 1905 dma_release_channel(chan); 1906 } 1907 1908 if (qp->rx_dma_chan) { 1909 struct dma_chan *chan = qp->rx_dma_chan; 1910 /* Putting the dma_chan to NULL will force any new traffic to be 1911 * processed by the CPU instead of the DAM engine 1912 */ 1913 qp->rx_dma_chan = NULL; 1914 1915 /* Try to be nice and wait for any queued DMA engine 1916 * transactions to process before smashing it with a rock 1917 */ 1918 dma_sync_wait(chan, qp->last_cookie); 1919 dmaengine_terminate_all(chan); 1920 dma_release_channel(chan); 1921 } 1922 1923 qp_bit = BIT_ULL(qp->qp_num); 1924 1925 ntb_db_set_mask(qp->ndev, qp_bit); 1926 tasklet_kill(&qp->rxc_db_work); 1927 1928 cancel_delayed_work_sync(&qp->link_work); 1929 1930 qp->cb_data = NULL; 1931 qp->rx_handler = NULL; 1932 qp->tx_handler = NULL; 1933 qp->event_handler = NULL; 1934 1935 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q))) 1936 kfree(entry); 1937 1938 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q))) { 1939 dev_warn(&pdev->dev, "Freeing item from non-empty rx_pend_q\n"); 1940 kfree(entry); 1941 } 1942 1943 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_post_q))) { 1944 dev_warn(&pdev->dev, "Freeing item from non-empty rx_post_q\n"); 1945 kfree(entry); 1946 } 1947 1948 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) 1949 kfree(entry); 1950 1951 qp->transport->qp_bitmap_free |= qp_bit; 1952 1953 dev_info(&pdev->dev, "NTB Transport QP %d freed\n", qp->qp_num); 1954 } 1955 EXPORT_SYMBOL_GPL(ntb_transport_free_queue); 1956 1957 /** 1958 * ntb_transport_rx_remove - Dequeues enqueued rx packet 1959 * @qp: NTB queue to be freed 1960 * @len: pointer to variable to write enqueued buffers length 1961 * 1962 * Dequeues unused buffers from receive queue. Should only be used during 1963 * shutdown of qp. 1964 * 1965 * RETURNS: NULL error value on error, or void* for success. 1966 */ 1967 void *ntb_transport_rx_remove(struct ntb_transport_qp *qp, unsigned int *len) 1968 { 1969 struct ntb_queue_entry *entry; 1970 void *buf; 1971 1972 if (!qp || qp->client_ready) 1973 return NULL; 1974 1975 entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q); 1976 if (!entry) 1977 return NULL; 1978 1979 buf = entry->cb_data; 1980 *len = entry->len; 1981 1982 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_free_q); 1983 1984 return buf; 1985 } 1986 EXPORT_SYMBOL_GPL(ntb_transport_rx_remove); 1987 1988 /** 1989 * ntb_transport_rx_enqueue - Enqueue a new NTB queue entry 1990 * @qp: NTB transport layer queue the entry is to be enqueued on 1991 * @cb: per buffer pointer for callback function to use 1992 * @data: pointer to data buffer that incoming packets will be copied into 1993 * @len: length of the data buffer 1994 * 1995 * Enqueue a new receive buffer onto the transport queue into which a NTB 1996 * payload can be received into. 1997 * 1998 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 1999 */ 2000 int ntb_transport_rx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, 2001 unsigned int len) 2002 { 2003 struct ntb_queue_entry *entry; 2004 2005 if (!qp) 2006 return -EINVAL; 2007 2008 entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q); 2009 if (!entry) 2010 return -ENOMEM; 2011 2012 entry->cb_data = cb; 2013 entry->buf = data; 2014 entry->len = len; 2015 entry->flags = 0; 2016 entry->retries = 0; 2017 entry->errors = 0; 2018 entry->rx_index = 0; 2019 2020 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_pend_q); 2021 2022 if (qp->active) 2023 tasklet_schedule(&qp->rxc_db_work); 2024 2025 return 0; 2026 } 2027 EXPORT_SYMBOL_GPL(ntb_transport_rx_enqueue); 2028 2029 /** 2030 * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry 2031 * @qp: NTB transport layer queue the entry is to be enqueued on 2032 * @cb: per buffer pointer for callback function to use 2033 * @data: pointer to data buffer that will be sent 2034 * @len: length of the data buffer 2035 * 2036 * Enqueue a new transmit buffer onto the transport queue from which a NTB 2037 * payload will be transmitted. This assumes that a lock is being held to 2038 * serialize access to the qp. 2039 * 2040 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 2041 */ 2042 int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, 2043 unsigned int len) 2044 { 2045 struct ntb_queue_entry *entry; 2046 int rc; 2047 2048 if (!qp || !qp->link_is_up || !len) 2049 return -EINVAL; 2050 2051 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); 2052 if (!entry) { 2053 qp->tx_err_no_buf++; 2054 return -EBUSY; 2055 } 2056 2057 entry->cb_data = cb; 2058 entry->buf = data; 2059 entry->len = len; 2060 entry->flags = 0; 2061 entry->errors = 0; 2062 entry->retries = 0; 2063 entry->tx_index = 0; 2064 2065 rc = ntb_process_tx(qp, entry); 2066 if (rc) 2067 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 2068 &qp->tx_free_q); 2069 2070 return rc; 2071 } 2072 EXPORT_SYMBOL_GPL(ntb_transport_tx_enqueue); 2073 2074 /** 2075 * ntb_transport_link_up - Notify NTB transport of client readiness to use queue 2076 * @qp: NTB transport layer queue to be enabled 2077 * 2078 * Notify NTB transport layer of client readiness to use queue 2079 */ 2080 void ntb_transport_link_up(struct ntb_transport_qp *qp) 2081 { 2082 if (!qp) 2083 return; 2084 2085 qp->client_ready = true; 2086 2087 if (qp->transport->link_is_up) 2088 schedule_delayed_work(&qp->link_work, 0); 2089 } 2090 EXPORT_SYMBOL_GPL(ntb_transport_link_up); 2091 2092 /** 2093 * ntb_transport_link_down - Notify NTB transport to no longer enqueue data 2094 * @qp: NTB transport layer queue to be disabled 2095 * 2096 * Notify NTB transport layer of client's desire to no longer receive data on 2097 * transport queue specified. It is the client's responsibility to ensure all 2098 * entries on queue are purged or otherwise handled appropriately. 2099 */ 2100 void ntb_transport_link_down(struct ntb_transport_qp *qp) 2101 { 2102 int val; 2103 2104 if (!qp) 2105 return; 2106 2107 qp->client_ready = false; 2108 2109 val = ntb_spad_read(qp->ndev, QP_LINKS); 2110 2111 ntb_peer_spad_write(qp->ndev, PIDX, QP_LINKS, val & ~BIT(qp->qp_num)); 2112 2113 if (qp->link_is_up) 2114 ntb_send_link_down(qp); 2115 else 2116 cancel_delayed_work_sync(&qp->link_work); 2117 } 2118 EXPORT_SYMBOL_GPL(ntb_transport_link_down); 2119 2120 /** 2121 * ntb_transport_link_query - Query transport link state 2122 * @qp: NTB transport layer queue to be queried 2123 * 2124 * Query connectivity to the remote system of the NTB transport queue 2125 * 2126 * RETURNS: true for link up or false for link down 2127 */ 2128 bool ntb_transport_link_query(struct ntb_transport_qp *qp) 2129 { 2130 if (!qp) 2131 return false; 2132 2133 return qp->link_is_up; 2134 } 2135 EXPORT_SYMBOL_GPL(ntb_transport_link_query); 2136 2137 /** 2138 * ntb_transport_qp_num - Query the qp number 2139 * @qp: NTB transport layer queue to be queried 2140 * 2141 * Query qp number of the NTB transport queue 2142 * 2143 * RETURNS: a zero based number specifying the qp number 2144 */ 2145 unsigned char ntb_transport_qp_num(struct ntb_transport_qp *qp) 2146 { 2147 if (!qp) 2148 return 0; 2149 2150 return qp->qp_num; 2151 } 2152 EXPORT_SYMBOL_GPL(ntb_transport_qp_num); 2153 2154 /** 2155 * ntb_transport_max_size - Query the max payload size of a qp 2156 * @qp: NTB transport layer queue to be queried 2157 * 2158 * Query the maximum payload size permissible on the given qp 2159 * 2160 * RETURNS: the max payload size of a qp 2161 */ 2162 unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp) 2163 { 2164 unsigned int max_size; 2165 unsigned int copy_align; 2166 struct dma_chan *rx_chan, *tx_chan; 2167 2168 if (!qp) 2169 return 0; 2170 2171 rx_chan = qp->rx_dma_chan; 2172 tx_chan = qp->tx_dma_chan; 2173 2174 copy_align = max(rx_chan ? rx_chan->device->copy_align : 0, 2175 tx_chan ? tx_chan->device->copy_align : 0); 2176 2177 /* If DMA engine usage is possible, try to find the max size for that */ 2178 max_size = qp->tx_max_frame - sizeof(struct ntb_payload_header); 2179 max_size = round_down(max_size, 1 << copy_align); 2180 2181 return max_size; 2182 } 2183 EXPORT_SYMBOL_GPL(ntb_transport_max_size); 2184 2185 unsigned int ntb_transport_tx_free_entry(struct ntb_transport_qp *qp) 2186 { 2187 unsigned int head = qp->tx_index; 2188 unsigned int tail = qp->remote_rx_info->entry; 2189 2190 return tail > head ? tail - head : qp->tx_max_entry + tail - head; 2191 } 2192 EXPORT_SYMBOL_GPL(ntb_transport_tx_free_entry); 2193 2194 static void ntb_transport_doorbell_callback(void *data, int vector) 2195 { 2196 struct ntb_transport_ctx *nt = data; 2197 struct ntb_transport_qp *qp; 2198 u64 db_bits; 2199 unsigned int qp_num; 2200 2201 db_bits = (nt->qp_bitmap & ~nt->qp_bitmap_free & 2202 ntb_db_vector_mask(nt->ndev, vector)); 2203 2204 while (db_bits) { 2205 qp_num = __ffs(db_bits); 2206 qp = &nt->qp_vec[qp_num]; 2207 2208 if (qp->active) 2209 tasklet_schedule(&qp->rxc_db_work); 2210 2211 db_bits &= ~BIT_ULL(qp_num); 2212 } 2213 } 2214 2215 static const struct ntb_ctx_ops ntb_transport_ops = { 2216 .link_event = ntb_transport_event_callback, 2217 .db_event = ntb_transport_doorbell_callback, 2218 }; 2219 2220 static struct ntb_client ntb_transport_client = { 2221 .ops = { 2222 .probe = ntb_transport_probe, 2223 .remove = ntb_transport_free, 2224 }, 2225 }; 2226 2227 static int __init ntb_transport_init(void) 2228 { 2229 int rc; 2230 2231 pr_info("%s, version %s\n", NTB_TRANSPORT_DESC, NTB_TRANSPORT_VER); 2232 2233 if (debugfs_initialized()) 2234 nt_debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL); 2235 2236 rc = bus_register(&ntb_transport_bus); 2237 if (rc) 2238 goto err_bus; 2239 2240 rc = ntb_register_client(&ntb_transport_client); 2241 if (rc) 2242 goto err_client; 2243 2244 return 0; 2245 2246 err_client: 2247 bus_unregister(&ntb_transport_bus); 2248 err_bus: 2249 debugfs_remove_recursive(nt_debugfs_dir); 2250 return rc; 2251 } 2252 module_init(ntb_transport_init); 2253 2254 static void __exit ntb_transport_exit(void) 2255 { 2256 ntb_unregister_client(&ntb_transport_client); 2257 bus_unregister(&ntb_transport_bus); 2258 debugfs_remove_recursive(nt_debugfs_dir); 2259 } 2260 module_exit(ntb_transport_exit); 2261