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 return; 929 930 out: 931 if (ntb_link_is_up(ndev, NULL, NULL) == 1) 932 schedule_delayed_work(&nt->link_work, 933 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 934 } 935 936 static void ntb_qp_link_work(struct work_struct *work) 937 { 938 struct ntb_transport_qp *qp = container_of(work, 939 struct ntb_transport_qp, 940 link_work.work); 941 struct pci_dev *pdev = qp->ndev->pdev; 942 struct ntb_transport_ctx *nt = qp->transport; 943 int val; 944 945 WARN_ON(!nt->link_is_up); 946 947 val = ntb_spad_read(nt->ndev, QP_LINKS); 948 949 ntb_peer_spad_write(nt->ndev, PIDX, QP_LINKS, val | BIT(qp->qp_num)); 950 951 /* query remote spad for qp ready bits */ 952 dev_dbg_ratelimited(&pdev->dev, "Remote QP link status = %x\n", val); 953 954 /* See if the remote side is up */ 955 if (val & BIT(qp->qp_num)) { 956 dev_info(&pdev->dev, "qp %d: Link Up\n", qp->qp_num); 957 qp->link_is_up = true; 958 qp->active = true; 959 960 if (qp->event_handler) 961 qp->event_handler(qp->cb_data, qp->link_is_up); 962 963 if (qp->active) 964 tasklet_schedule(&qp->rxc_db_work); 965 } else if (nt->link_is_up) 966 schedule_delayed_work(&qp->link_work, 967 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 968 } 969 970 static int ntb_transport_init_queue(struct ntb_transport_ctx *nt, 971 unsigned int qp_num) 972 { 973 struct ntb_transport_qp *qp; 974 phys_addr_t mw_base; 975 resource_size_t mw_size; 976 unsigned int num_qps_mw, tx_size; 977 unsigned int mw_num, mw_count, qp_count; 978 u64 qp_offset; 979 980 mw_count = nt->mw_count; 981 qp_count = nt->qp_count; 982 983 mw_num = QP_TO_MW(nt, qp_num); 984 985 qp = &nt->qp_vec[qp_num]; 986 qp->qp_num = qp_num; 987 qp->transport = nt; 988 qp->ndev = nt->ndev; 989 qp->client_ready = false; 990 qp->event_handler = NULL; 991 ntb_qp_link_down_reset(qp); 992 993 if (mw_num < qp_count % mw_count) 994 num_qps_mw = qp_count / mw_count + 1; 995 else 996 num_qps_mw = qp_count / mw_count; 997 998 mw_base = nt->mw_vec[mw_num].phys_addr; 999 mw_size = nt->mw_vec[mw_num].phys_size; 1000 1001 tx_size = (unsigned int)mw_size / num_qps_mw; 1002 qp_offset = tx_size * (qp_num / mw_count); 1003 1004 qp->tx_mw = nt->mw_vec[mw_num].vbase + qp_offset; 1005 if (!qp->tx_mw) 1006 return -EINVAL; 1007 1008 qp->tx_mw_phys = mw_base + qp_offset; 1009 if (!qp->tx_mw_phys) 1010 return -EINVAL; 1011 1012 tx_size -= sizeof(struct ntb_rx_info); 1013 qp->rx_info = qp->tx_mw + tx_size; 1014 1015 /* Due to housekeeping, there must be atleast 2 buffs */ 1016 qp->tx_max_frame = min(transport_mtu, tx_size / 2); 1017 qp->tx_max_entry = tx_size / qp->tx_max_frame; 1018 1019 if (nt->debugfs_node_dir) { 1020 char debugfs_name[4]; 1021 1022 snprintf(debugfs_name, 4, "qp%d", qp_num); 1023 qp->debugfs_dir = debugfs_create_dir(debugfs_name, 1024 nt->debugfs_node_dir); 1025 1026 qp->debugfs_stats = debugfs_create_file("stats", S_IRUSR, 1027 qp->debugfs_dir, qp, 1028 &ntb_qp_debugfs_stats); 1029 } else { 1030 qp->debugfs_dir = NULL; 1031 qp->debugfs_stats = NULL; 1032 } 1033 1034 INIT_DELAYED_WORK(&qp->link_work, ntb_qp_link_work); 1035 INIT_WORK(&qp->link_cleanup, ntb_qp_link_cleanup_work); 1036 1037 spin_lock_init(&qp->ntb_rx_q_lock); 1038 spin_lock_init(&qp->ntb_tx_free_q_lock); 1039 1040 INIT_LIST_HEAD(&qp->rx_post_q); 1041 INIT_LIST_HEAD(&qp->rx_pend_q); 1042 INIT_LIST_HEAD(&qp->rx_free_q); 1043 INIT_LIST_HEAD(&qp->tx_free_q); 1044 1045 tasklet_init(&qp->rxc_db_work, ntb_transport_rxc_db, 1046 (unsigned long)qp); 1047 1048 return 0; 1049 } 1050 1051 static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev) 1052 { 1053 struct ntb_transport_ctx *nt; 1054 struct ntb_transport_mw *mw; 1055 unsigned int mw_count, qp_count, spad_count, max_mw_count_for_spads; 1056 u64 qp_bitmap; 1057 int node; 1058 int rc, i; 1059 1060 mw_count = ntb_peer_mw_count(ndev); 1061 1062 if (!ndev->ops->mw_set_trans) { 1063 dev_err(&ndev->dev, "Inbound MW based NTB API is required\n"); 1064 return -EINVAL; 1065 } 1066 1067 if (ntb_db_is_unsafe(ndev)) 1068 dev_dbg(&ndev->dev, 1069 "doorbell is unsafe, proceed anyway...\n"); 1070 if (ntb_spad_is_unsafe(ndev)) 1071 dev_dbg(&ndev->dev, 1072 "scratchpad is unsafe, proceed anyway...\n"); 1073 1074 if (ntb_peer_port_count(ndev) != NTB_DEF_PEER_CNT) 1075 dev_warn(&ndev->dev, "Multi-port NTB devices unsupported\n"); 1076 1077 node = dev_to_node(&ndev->dev); 1078 1079 nt = kzalloc_node(sizeof(*nt), GFP_KERNEL, node); 1080 if (!nt) 1081 return -ENOMEM; 1082 1083 nt->ndev = ndev; 1084 spad_count = ntb_spad_count(ndev); 1085 1086 /* Limit the MW's based on the availability of scratchpads */ 1087 1088 if (spad_count < NTB_TRANSPORT_MIN_SPADS) { 1089 nt->mw_count = 0; 1090 rc = -EINVAL; 1091 goto err; 1092 } 1093 1094 max_mw_count_for_spads = (spad_count - MW0_SZ_HIGH) / 2; 1095 nt->mw_count = min(mw_count, max_mw_count_for_spads); 1096 1097 nt->mw_vec = kzalloc_node(mw_count * sizeof(*nt->mw_vec), 1098 GFP_KERNEL, node); 1099 if (!nt->mw_vec) { 1100 rc = -ENOMEM; 1101 goto err; 1102 } 1103 1104 for (i = 0; i < mw_count; i++) { 1105 mw = &nt->mw_vec[i]; 1106 1107 rc = ntb_mw_get_align(ndev, PIDX, i, &mw->xlat_align, 1108 &mw->xlat_align_size, NULL); 1109 if (rc) 1110 goto err1; 1111 1112 rc = ntb_peer_mw_get_addr(ndev, i, &mw->phys_addr, 1113 &mw->phys_size); 1114 if (rc) 1115 goto err1; 1116 1117 mw->vbase = ioremap_wc(mw->phys_addr, mw->phys_size); 1118 if (!mw->vbase) { 1119 rc = -ENOMEM; 1120 goto err1; 1121 } 1122 1123 mw->buff_size = 0; 1124 mw->xlat_size = 0; 1125 mw->virt_addr = NULL; 1126 mw->dma_addr = 0; 1127 } 1128 1129 qp_bitmap = ntb_db_valid_mask(ndev); 1130 1131 qp_count = ilog2(qp_bitmap); 1132 if (max_num_clients && max_num_clients < qp_count) 1133 qp_count = max_num_clients; 1134 else if (nt->mw_count < qp_count) 1135 qp_count = nt->mw_count; 1136 1137 qp_bitmap &= BIT_ULL(qp_count) - 1; 1138 1139 nt->qp_count = qp_count; 1140 nt->qp_bitmap = qp_bitmap; 1141 nt->qp_bitmap_free = qp_bitmap; 1142 1143 nt->qp_vec = kzalloc_node(qp_count * sizeof(*nt->qp_vec), 1144 GFP_KERNEL, node); 1145 if (!nt->qp_vec) { 1146 rc = -ENOMEM; 1147 goto err1; 1148 } 1149 1150 if (nt_debugfs_dir) { 1151 nt->debugfs_node_dir = 1152 debugfs_create_dir(pci_name(ndev->pdev), 1153 nt_debugfs_dir); 1154 } 1155 1156 for (i = 0; i < qp_count; i++) { 1157 rc = ntb_transport_init_queue(nt, i); 1158 if (rc) 1159 goto err2; 1160 } 1161 1162 INIT_DELAYED_WORK(&nt->link_work, ntb_transport_link_work); 1163 INIT_WORK(&nt->link_cleanup, ntb_transport_link_cleanup_work); 1164 1165 rc = ntb_set_ctx(ndev, nt, &ntb_transport_ops); 1166 if (rc) 1167 goto err2; 1168 1169 INIT_LIST_HEAD(&nt->client_devs); 1170 rc = ntb_bus_init(nt); 1171 if (rc) 1172 goto err3; 1173 1174 nt->link_is_up = false; 1175 ntb_link_enable(ndev, NTB_SPEED_AUTO, NTB_WIDTH_AUTO); 1176 ntb_link_event(ndev); 1177 1178 return 0; 1179 1180 err3: 1181 ntb_clear_ctx(ndev); 1182 err2: 1183 kfree(nt->qp_vec); 1184 err1: 1185 while (i--) { 1186 mw = &nt->mw_vec[i]; 1187 iounmap(mw->vbase); 1188 } 1189 kfree(nt->mw_vec); 1190 err: 1191 kfree(nt); 1192 return rc; 1193 } 1194 1195 static void ntb_transport_free(struct ntb_client *self, struct ntb_dev *ndev) 1196 { 1197 struct ntb_transport_ctx *nt = ndev->ctx; 1198 struct ntb_transport_qp *qp; 1199 u64 qp_bitmap_alloc; 1200 int i; 1201 1202 ntb_transport_link_cleanup(nt); 1203 cancel_work_sync(&nt->link_cleanup); 1204 cancel_delayed_work_sync(&nt->link_work); 1205 1206 qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free; 1207 1208 /* verify that all the qp's are freed */ 1209 for (i = 0; i < nt->qp_count; i++) { 1210 qp = &nt->qp_vec[i]; 1211 if (qp_bitmap_alloc & BIT_ULL(i)) 1212 ntb_transport_free_queue(qp); 1213 debugfs_remove_recursive(qp->debugfs_dir); 1214 } 1215 1216 ntb_link_disable(ndev); 1217 ntb_clear_ctx(ndev); 1218 1219 ntb_bus_remove(nt); 1220 1221 for (i = nt->mw_count; i--; ) { 1222 ntb_free_mw(nt, i); 1223 iounmap(nt->mw_vec[i].vbase); 1224 } 1225 1226 kfree(nt->qp_vec); 1227 kfree(nt->mw_vec); 1228 kfree(nt); 1229 } 1230 1231 static void ntb_complete_rxc(struct ntb_transport_qp *qp) 1232 { 1233 struct ntb_queue_entry *entry; 1234 void *cb_data; 1235 unsigned int len; 1236 unsigned long irqflags; 1237 1238 spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags); 1239 1240 while (!list_empty(&qp->rx_post_q)) { 1241 entry = list_first_entry(&qp->rx_post_q, 1242 struct ntb_queue_entry, entry); 1243 if (!(entry->flags & DESC_DONE_FLAG)) 1244 break; 1245 1246 entry->rx_hdr->flags = 0; 1247 iowrite32(entry->rx_index, &qp->rx_info->entry); 1248 1249 cb_data = entry->cb_data; 1250 len = entry->len; 1251 1252 list_move_tail(&entry->entry, &qp->rx_free_q); 1253 1254 spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags); 1255 1256 if (qp->rx_handler && qp->client_ready) 1257 qp->rx_handler(qp, qp->cb_data, cb_data, len); 1258 1259 spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags); 1260 } 1261 1262 spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags); 1263 } 1264 1265 static void ntb_rx_copy_callback(void *data, 1266 const struct dmaengine_result *res) 1267 { 1268 struct ntb_queue_entry *entry = data; 1269 1270 /* we need to check DMA results if we are using DMA */ 1271 if (res) { 1272 enum dmaengine_tx_result dma_err = res->result; 1273 1274 switch (dma_err) { 1275 case DMA_TRANS_READ_FAILED: 1276 case DMA_TRANS_WRITE_FAILED: 1277 entry->errors++; 1278 case DMA_TRANS_ABORTED: 1279 { 1280 struct ntb_transport_qp *qp = entry->qp; 1281 void *offset = qp->rx_buff + qp->rx_max_frame * 1282 qp->rx_index; 1283 1284 ntb_memcpy_rx(entry, offset); 1285 qp->rx_memcpy++; 1286 return; 1287 } 1288 1289 case DMA_TRANS_NOERROR: 1290 default: 1291 break; 1292 } 1293 } 1294 1295 entry->flags |= DESC_DONE_FLAG; 1296 1297 ntb_complete_rxc(entry->qp); 1298 } 1299 1300 static void ntb_memcpy_rx(struct ntb_queue_entry *entry, void *offset) 1301 { 1302 void *buf = entry->buf; 1303 size_t len = entry->len; 1304 1305 memcpy(buf, offset, len); 1306 1307 /* Ensure that the data is fully copied out before clearing the flag */ 1308 wmb(); 1309 1310 ntb_rx_copy_callback(entry, NULL); 1311 } 1312 1313 static int ntb_async_rx_submit(struct ntb_queue_entry *entry, void *offset) 1314 { 1315 struct dma_async_tx_descriptor *txd; 1316 struct ntb_transport_qp *qp = entry->qp; 1317 struct dma_chan *chan = qp->rx_dma_chan; 1318 struct dma_device *device; 1319 size_t pay_off, buff_off, len; 1320 struct dmaengine_unmap_data *unmap; 1321 dma_cookie_t cookie; 1322 void *buf = entry->buf; 1323 1324 len = entry->len; 1325 device = chan->device; 1326 pay_off = (size_t)offset & ~PAGE_MASK; 1327 buff_off = (size_t)buf & ~PAGE_MASK; 1328 1329 if (!is_dma_copy_aligned(device, pay_off, buff_off, len)) 1330 goto err; 1331 1332 unmap = dmaengine_get_unmap_data(device->dev, 2, GFP_NOWAIT); 1333 if (!unmap) 1334 goto err; 1335 1336 unmap->len = len; 1337 unmap->addr[0] = dma_map_page(device->dev, virt_to_page(offset), 1338 pay_off, len, DMA_TO_DEVICE); 1339 if (dma_mapping_error(device->dev, unmap->addr[0])) 1340 goto err_get_unmap; 1341 1342 unmap->to_cnt = 1; 1343 1344 unmap->addr[1] = dma_map_page(device->dev, virt_to_page(buf), 1345 buff_off, len, DMA_FROM_DEVICE); 1346 if (dma_mapping_error(device->dev, unmap->addr[1])) 1347 goto err_get_unmap; 1348 1349 unmap->from_cnt = 1; 1350 1351 txd = device->device_prep_dma_memcpy(chan, unmap->addr[1], 1352 unmap->addr[0], len, 1353 DMA_PREP_INTERRUPT); 1354 if (!txd) 1355 goto err_get_unmap; 1356 1357 txd->callback_result = ntb_rx_copy_callback; 1358 txd->callback_param = entry; 1359 dma_set_unmap(txd, unmap); 1360 1361 cookie = dmaengine_submit(txd); 1362 if (dma_submit_error(cookie)) 1363 goto err_set_unmap; 1364 1365 dmaengine_unmap_put(unmap); 1366 1367 qp->last_cookie = cookie; 1368 1369 qp->rx_async++; 1370 1371 return 0; 1372 1373 err_set_unmap: 1374 dmaengine_unmap_put(unmap); 1375 err_get_unmap: 1376 dmaengine_unmap_put(unmap); 1377 err: 1378 return -ENXIO; 1379 } 1380 1381 static void ntb_async_rx(struct ntb_queue_entry *entry, void *offset) 1382 { 1383 struct ntb_transport_qp *qp = entry->qp; 1384 struct dma_chan *chan = qp->rx_dma_chan; 1385 int res; 1386 1387 if (!chan) 1388 goto err; 1389 1390 if (entry->len < copy_bytes) 1391 goto err; 1392 1393 res = ntb_async_rx_submit(entry, offset); 1394 if (res < 0) 1395 goto err; 1396 1397 if (!entry->retries) 1398 qp->rx_async++; 1399 1400 return; 1401 1402 err: 1403 ntb_memcpy_rx(entry, offset); 1404 qp->rx_memcpy++; 1405 } 1406 1407 static int ntb_process_rxc(struct ntb_transport_qp *qp) 1408 { 1409 struct ntb_payload_header *hdr; 1410 struct ntb_queue_entry *entry; 1411 void *offset; 1412 1413 offset = qp->rx_buff + qp->rx_max_frame * qp->rx_index; 1414 hdr = offset + qp->rx_max_frame - sizeof(struct ntb_payload_header); 1415 1416 dev_dbg(&qp->ndev->pdev->dev, "qp %d: RX ver %u len %d flags %x\n", 1417 qp->qp_num, hdr->ver, hdr->len, hdr->flags); 1418 1419 if (!(hdr->flags & DESC_DONE_FLAG)) { 1420 dev_dbg(&qp->ndev->pdev->dev, "done flag not set\n"); 1421 qp->rx_ring_empty++; 1422 return -EAGAIN; 1423 } 1424 1425 if (hdr->flags & LINK_DOWN_FLAG) { 1426 dev_dbg(&qp->ndev->pdev->dev, "link down flag set\n"); 1427 ntb_qp_link_down(qp); 1428 hdr->flags = 0; 1429 return -EAGAIN; 1430 } 1431 1432 if (hdr->ver != (u32)qp->rx_pkts) { 1433 dev_dbg(&qp->ndev->pdev->dev, 1434 "version mismatch, expected %llu - got %u\n", 1435 qp->rx_pkts, hdr->ver); 1436 qp->rx_err_ver++; 1437 return -EIO; 1438 } 1439 1440 entry = ntb_list_mv(&qp->ntb_rx_q_lock, &qp->rx_pend_q, &qp->rx_post_q); 1441 if (!entry) { 1442 dev_dbg(&qp->ndev->pdev->dev, "no receive buffer\n"); 1443 qp->rx_err_no_buf++; 1444 return -EAGAIN; 1445 } 1446 1447 entry->rx_hdr = hdr; 1448 entry->rx_index = qp->rx_index; 1449 1450 if (hdr->len > entry->len) { 1451 dev_dbg(&qp->ndev->pdev->dev, 1452 "receive buffer overflow! Wanted %d got %d\n", 1453 hdr->len, entry->len); 1454 qp->rx_err_oflow++; 1455 1456 entry->len = -EIO; 1457 entry->flags |= DESC_DONE_FLAG; 1458 1459 ntb_complete_rxc(qp); 1460 } else { 1461 dev_dbg(&qp->ndev->pdev->dev, 1462 "RX OK index %u ver %u size %d into buf size %d\n", 1463 qp->rx_index, hdr->ver, hdr->len, entry->len); 1464 1465 qp->rx_bytes += hdr->len; 1466 qp->rx_pkts++; 1467 1468 entry->len = hdr->len; 1469 1470 ntb_async_rx(entry, offset); 1471 } 1472 1473 qp->rx_index++; 1474 qp->rx_index %= qp->rx_max_entry; 1475 1476 return 0; 1477 } 1478 1479 static void ntb_transport_rxc_db(unsigned long data) 1480 { 1481 struct ntb_transport_qp *qp = (void *)data; 1482 int rc, i; 1483 1484 dev_dbg(&qp->ndev->pdev->dev, "%s: doorbell %d received\n", 1485 __func__, qp->qp_num); 1486 1487 /* Limit the number of packets processed in a single interrupt to 1488 * provide fairness to others 1489 */ 1490 for (i = 0; i < qp->rx_max_entry; i++) { 1491 rc = ntb_process_rxc(qp); 1492 if (rc) 1493 break; 1494 } 1495 1496 if (i && qp->rx_dma_chan) 1497 dma_async_issue_pending(qp->rx_dma_chan); 1498 1499 if (i == qp->rx_max_entry) { 1500 /* there is more work to do */ 1501 if (qp->active) 1502 tasklet_schedule(&qp->rxc_db_work); 1503 } else if (ntb_db_read(qp->ndev) & BIT_ULL(qp->qp_num)) { 1504 /* the doorbell bit is set: clear it */ 1505 ntb_db_clear(qp->ndev, BIT_ULL(qp->qp_num)); 1506 /* ntb_db_read ensures ntb_db_clear write is committed */ 1507 ntb_db_read(qp->ndev); 1508 1509 /* an interrupt may have arrived between finishing 1510 * ntb_process_rxc and clearing the doorbell bit: 1511 * there might be some more work to do. 1512 */ 1513 if (qp->active) 1514 tasklet_schedule(&qp->rxc_db_work); 1515 } 1516 } 1517 1518 static void ntb_tx_copy_callback(void *data, 1519 const struct dmaengine_result *res) 1520 { 1521 struct ntb_queue_entry *entry = data; 1522 struct ntb_transport_qp *qp = entry->qp; 1523 struct ntb_payload_header __iomem *hdr = entry->tx_hdr; 1524 1525 /* we need to check DMA results if we are using DMA */ 1526 if (res) { 1527 enum dmaengine_tx_result dma_err = res->result; 1528 1529 switch (dma_err) { 1530 case DMA_TRANS_READ_FAILED: 1531 case DMA_TRANS_WRITE_FAILED: 1532 entry->errors++; 1533 case DMA_TRANS_ABORTED: 1534 { 1535 void __iomem *offset = 1536 qp->tx_mw + qp->tx_max_frame * 1537 entry->tx_index; 1538 1539 /* resubmit via CPU */ 1540 ntb_memcpy_tx(entry, offset); 1541 qp->tx_memcpy++; 1542 return; 1543 } 1544 1545 case DMA_TRANS_NOERROR: 1546 default: 1547 break; 1548 } 1549 } 1550 1551 iowrite32(entry->flags | DESC_DONE_FLAG, &hdr->flags); 1552 1553 ntb_peer_db_set(qp->ndev, BIT_ULL(qp->qp_num)); 1554 1555 /* The entry length can only be zero if the packet is intended to be a 1556 * "link down" or similar. Since no payload is being sent in these 1557 * cases, there is nothing to add to the completion queue. 1558 */ 1559 if (entry->len > 0) { 1560 qp->tx_bytes += entry->len; 1561 1562 if (qp->tx_handler) 1563 qp->tx_handler(qp, qp->cb_data, entry->cb_data, 1564 entry->len); 1565 } 1566 1567 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, &qp->tx_free_q); 1568 } 1569 1570 static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset) 1571 { 1572 #ifdef ARCH_HAS_NOCACHE_UACCESS 1573 /* 1574 * Using non-temporal mov to improve performance on non-cached 1575 * writes, even though we aren't actually copying from user space. 1576 */ 1577 __copy_from_user_inatomic_nocache(offset, entry->buf, entry->len); 1578 #else 1579 memcpy_toio(offset, entry->buf, entry->len); 1580 #endif 1581 1582 /* Ensure that the data is fully copied out before setting the flags */ 1583 wmb(); 1584 1585 ntb_tx_copy_callback(entry, NULL); 1586 } 1587 1588 static int ntb_async_tx_submit(struct ntb_transport_qp *qp, 1589 struct ntb_queue_entry *entry) 1590 { 1591 struct dma_async_tx_descriptor *txd; 1592 struct dma_chan *chan = qp->tx_dma_chan; 1593 struct dma_device *device; 1594 size_t len = entry->len; 1595 void *buf = entry->buf; 1596 size_t dest_off, buff_off; 1597 struct dmaengine_unmap_data *unmap; 1598 dma_addr_t dest; 1599 dma_cookie_t cookie; 1600 1601 device = chan->device; 1602 dest = qp->tx_mw_phys + qp->tx_max_frame * entry->tx_index; 1603 buff_off = (size_t)buf & ~PAGE_MASK; 1604 dest_off = (size_t)dest & ~PAGE_MASK; 1605 1606 if (!is_dma_copy_aligned(device, buff_off, dest_off, len)) 1607 goto err; 1608 1609 unmap = dmaengine_get_unmap_data(device->dev, 1, GFP_NOWAIT); 1610 if (!unmap) 1611 goto err; 1612 1613 unmap->len = len; 1614 unmap->addr[0] = dma_map_page(device->dev, virt_to_page(buf), 1615 buff_off, len, DMA_TO_DEVICE); 1616 if (dma_mapping_error(device->dev, unmap->addr[0])) 1617 goto err_get_unmap; 1618 1619 unmap->to_cnt = 1; 1620 1621 txd = device->device_prep_dma_memcpy(chan, dest, unmap->addr[0], len, 1622 DMA_PREP_INTERRUPT); 1623 if (!txd) 1624 goto err_get_unmap; 1625 1626 txd->callback_result = ntb_tx_copy_callback; 1627 txd->callback_param = entry; 1628 dma_set_unmap(txd, unmap); 1629 1630 cookie = dmaengine_submit(txd); 1631 if (dma_submit_error(cookie)) 1632 goto err_set_unmap; 1633 1634 dmaengine_unmap_put(unmap); 1635 1636 dma_async_issue_pending(chan); 1637 1638 return 0; 1639 err_set_unmap: 1640 dmaengine_unmap_put(unmap); 1641 err_get_unmap: 1642 dmaengine_unmap_put(unmap); 1643 err: 1644 return -ENXIO; 1645 } 1646 1647 static void ntb_async_tx(struct ntb_transport_qp *qp, 1648 struct ntb_queue_entry *entry) 1649 { 1650 struct ntb_payload_header __iomem *hdr; 1651 struct dma_chan *chan = qp->tx_dma_chan; 1652 void __iomem *offset; 1653 int res; 1654 1655 entry->tx_index = qp->tx_index; 1656 offset = qp->tx_mw + qp->tx_max_frame * entry->tx_index; 1657 hdr = offset + qp->tx_max_frame - sizeof(struct ntb_payload_header); 1658 entry->tx_hdr = hdr; 1659 1660 iowrite32(entry->len, &hdr->len); 1661 iowrite32((u32)qp->tx_pkts, &hdr->ver); 1662 1663 if (!chan) 1664 goto err; 1665 1666 if (entry->len < copy_bytes) 1667 goto err; 1668 1669 res = ntb_async_tx_submit(qp, entry); 1670 if (res < 0) 1671 goto err; 1672 1673 if (!entry->retries) 1674 qp->tx_async++; 1675 1676 return; 1677 1678 err: 1679 ntb_memcpy_tx(entry, offset); 1680 qp->tx_memcpy++; 1681 } 1682 1683 static int ntb_process_tx(struct ntb_transport_qp *qp, 1684 struct ntb_queue_entry *entry) 1685 { 1686 if (qp->tx_index == qp->remote_rx_info->entry) { 1687 qp->tx_ring_full++; 1688 return -EAGAIN; 1689 } 1690 1691 if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) { 1692 if (qp->tx_handler) 1693 qp->tx_handler(qp, qp->cb_data, NULL, -EIO); 1694 1695 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1696 &qp->tx_free_q); 1697 return 0; 1698 } 1699 1700 ntb_async_tx(qp, entry); 1701 1702 qp->tx_index++; 1703 qp->tx_index %= qp->tx_max_entry; 1704 1705 qp->tx_pkts++; 1706 1707 return 0; 1708 } 1709 1710 static void ntb_send_link_down(struct ntb_transport_qp *qp) 1711 { 1712 struct pci_dev *pdev = qp->ndev->pdev; 1713 struct ntb_queue_entry *entry; 1714 int i, rc; 1715 1716 if (!qp->link_is_up) 1717 return; 1718 1719 dev_info(&pdev->dev, "qp %d: Send Link Down\n", qp->qp_num); 1720 1721 for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) { 1722 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); 1723 if (entry) 1724 break; 1725 msleep(100); 1726 } 1727 1728 if (!entry) 1729 return; 1730 1731 entry->cb_data = NULL; 1732 entry->buf = NULL; 1733 entry->len = 0; 1734 entry->flags = LINK_DOWN_FLAG; 1735 1736 rc = ntb_process_tx(qp, entry); 1737 if (rc) 1738 dev_err(&pdev->dev, "ntb: QP%d unable to send linkdown msg\n", 1739 qp->qp_num); 1740 1741 ntb_qp_link_down_reset(qp); 1742 } 1743 1744 static bool ntb_dma_filter_fn(struct dma_chan *chan, void *node) 1745 { 1746 return dev_to_node(&chan->dev->device) == (int)(unsigned long)node; 1747 } 1748 1749 /** 1750 * ntb_transport_create_queue - Create a new NTB transport layer queue 1751 * @rx_handler: receive callback function 1752 * @tx_handler: transmit callback function 1753 * @event_handler: event callback function 1754 * 1755 * Create a new NTB transport layer queue and provide the queue with a callback 1756 * routine for both transmit and receive. The receive callback routine will be 1757 * used to pass up data when the transport has received it on the queue. The 1758 * transmit callback routine will be called when the transport has completed the 1759 * transmission of the data on the queue and the data is ready to be freed. 1760 * 1761 * RETURNS: pointer to newly created ntb_queue, NULL on error. 1762 */ 1763 struct ntb_transport_qp * 1764 ntb_transport_create_queue(void *data, struct device *client_dev, 1765 const struct ntb_queue_handlers *handlers) 1766 { 1767 struct ntb_dev *ndev; 1768 struct pci_dev *pdev; 1769 struct ntb_transport_ctx *nt; 1770 struct ntb_queue_entry *entry; 1771 struct ntb_transport_qp *qp; 1772 u64 qp_bit; 1773 unsigned int free_queue; 1774 dma_cap_mask_t dma_mask; 1775 int node; 1776 int i; 1777 1778 ndev = dev_ntb(client_dev->parent); 1779 pdev = ndev->pdev; 1780 nt = ndev->ctx; 1781 1782 node = dev_to_node(&ndev->dev); 1783 1784 free_queue = ffs(nt->qp_bitmap_free); 1785 if (!free_queue) 1786 goto err; 1787 1788 /* decrement free_queue to make it zero based */ 1789 free_queue--; 1790 1791 qp = &nt->qp_vec[free_queue]; 1792 qp_bit = BIT_ULL(qp->qp_num); 1793 1794 nt->qp_bitmap_free &= ~qp_bit; 1795 1796 qp->cb_data = data; 1797 qp->rx_handler = handlers->rx_handler; 1798 qp->tx_handler = handlers->tx_handler; 1799 qp->event_handler = handlers->event_handler; 1800 1801 dma_cap_zero(dma_mask); 1802 dma_cap_set(DMA_MEMCPY, dma_mask); 1803 1804 if (use_dma) { 1805 qp->tx_dma_chan = 1806 dma_request_channel(dma_mask, ntb_dma_filter_fn, 1807 (void *)(unsigned long)node); 1808 if (!qp->tx_dma_chan) 1809 dev_info(&pdev->dev, "Unable to allocate TX DMA channel\n"); 1810 1811 qp->rx_dma_chan = 1812 dma_request_channel(dma_mask, ntb_dma_filter_fn, 1813 (void *)(unsigned long)node); 1814 if (!qp->rx_dma_chan) 1815 dev_info(&pdev->dev, "Unable to allocate RX DMA channel\n"); 1816 } else { 1817 qp->tx_dma_chan = NULL; 1818 qp->rx_dma_chan = NULL; 1819 } 1820 1821 dev_dbg(&pdev->dev, "Using %s memcpy for TX\n", 1822 qp->tx_dma_chan ? "DMA" : "CPU"); 1823 1824 dev_dbg(&pdev->dev, "Using %s memcpy for RX\n", 1825 qp->rx_dma_chan ? "DMA" : "CPU"); 1826 1827 for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) { 1828 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 1829 if (!entry) 1830 goto err1; 1831 1832 entry->qp = qp; 1833 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, 1834 &qp->rx_free_q); 1835 } 1836 qp->rx_alloc_entry = NTB_QP_DEF_NUM_ENTRIES; 1837 1838 for (i = 0; i < qp->tx_max_entry; i++) { 1839 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 1840 if (!entry) 1841 goto err2; 1842 1843 entry->qp = qp; 1844 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1845 &qp->tx_free_q); 1846 } 1847 1848 ntb_db_clear(qp->ndev, qp_bit); 1849 ntb_db_clear_mask(qp->ndev, qp_bit); 1850 1851 dev_info(&pdev->dev, "NTB Transport QP %d created\n", qp->qp_num); 1852 1853 return qp; 1854 1855 err2: 1856 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) 1857 kfree(entry); 1858 err1: 1859 qp->rx_alloc_entry = 0; 1860 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q))) 1861 kfree(entry); 1862 if (qp->tx_dma_chan) 1863 dma_release_channel(qp->tx_dma_chan); 1864 if (qp->rx_dma_chan) 1865 dma_release_channel(qp->rx_dma_chan); 1866 nt->qp_bitmap_free |= qp_bit; 1867 err: 1868 return NULL; 1869 } 1870 EXPORT_SYMBOL_GPL(ntb_transport_create_queue); 1871 1872 /** 1873 * ntb_transport_free_queue - Frees NTB transport queue 1874 * @qp: NTB queue to be freed 1875 * 1876 * Frees NTB transport queue 1877 */ 1878 void ntb_transport_free_queue(struct ntb_transport_qp *qp) 1879 { 1880 struct pci_dev *pdev; 1881 struct ntb_queue_entry *entry; 1882 u64 qp_bit; 1883 1884 if (!qp) 1885 return; 1886 1887 pdev = qp->ndev->pdev; 1888 1889 qp->active = false; 1890 1891 if (qp->tx_dma_chan) { 1892 struct dma_chan *chan = qp->tx_dma_chan; 1893 /* Putting the dma_chan to NULL will force any new traffic to be 1894 * processed by the CPU instead of the DAM engine 1895 */ 1896 qp->tx_dma_chan = NULL; 1897 1898 /* Try to be nice and wait for any queued DMA engine 1899 * transactions to process before smashing it with a rock 1900 */ 1901 dma_sync_wait(chan, qp->last_cookie); 1902 dmaengine_terminate_all(chan); 1903 dma_release_channel(chan); 1904 } 1905 1906 if (qp->rx_dma_chan) { 1907 struct dma_chan *chan = qp->rx_dma_chan; 1908 /* Putting the dma_chan to NULL will force any new traffic to be 1909 * processed by the CPU instead of the DAM engine 1910 */ 1911 qp->rx_dma_chan = NULL; 1912 1913 /* Try to be nice and wait for any queued DMA engine 1914 * transactions to process before smashing it with a rock 1915 */ 1916 dma_sync_wait(chan, qp->last_cookie); 1917 dmaengine_terminate_all(chan); 1918 dma_release_channel(chan); 1919 } 1920 1921 qp_bit = BIT_ULL(qp->qp_num); 1922 1923 ntb_db_set_mask(qp->ndev, qp_bit); 1924 tasklet_kill(&qp->rxc_db_work); 1925 1926 cancel_delayed_work_sync(&qp->link_work); 1927 1928 qp->cb_data = NULL; 1929 qp->rx_handler = NULL; 1930 qp->tx_handler = NULL; 1931 qp->event_handler = NULL; 1932 1933 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q))) 1934 kfree(entry); 1935 1936 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q))) { 1937 dev_warn(&pdev->dev, "Freeing item from non-empty rx_pend_q\n"); 1938 kfree(entry); 1939 } 1940 1941 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_post_q))) { 1942 dev_warn(&pdev->dev, "Freeing item from non-empty rx_post_q\n"); 1943 kfree(entry); 1944 } 1945 1946 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) 1947 kfree(entry); 1948 1949 qp->transport->qp_bitmap_free |= qp_bit; 1950 1951 dev_info(&pdev->dev, "NTB Transport QP %d freed\n", qp->qp_num); 1952 } 1953 EXPORT_SYMBOL_GPL(ntb_transport_free_queue); 1954 1955 /** 1956 * ntb_transport_rx_remove - Dequeues enqueued rx packet 1957 * @qp: NTB queue to be freed 1958 * @len: pointer to variable to write enqueued buffers length 1959 * 1960 * Dequeues unused buffers from receive queue. Should only be used during 1961 * shutdown of qp. 1962 * 1963 * RETURNS: NULL error value on error, or void* for success. 1964 */ 1965 void *ntb_transport_rx_remove(struct ntb_transport_qp *qp, unsigned int *len) 1966 { 1967 struct ntb_queue_entry *entry; 1968 void *buf; 1969 1970 if (!qp || qp->client_ready) 1971 return NULL; 1972 1973 entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q); 1974 if (!entry) 1975 return NULL; 1976 1977 buf = entry->cb_data; 1978 *len = entry->len; 1979 1980 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_free_q); 1981 1982 return buf; 1983 } 1984 EXPORT_SYMBOL_GPL(ntb_transport_rx_remove); 1985 1986 /** 1987 * ntb_transport_rx_enqueue - Enqueue a new NTB queue entry 1988 * @qp: NTB transport layer queue the entry is to be enqueued on 1989 * @cb: per buffer pointer for callback function to use 1990 * @data: pointer to data buffer that incoming packets will be copied into 1991 * @len: length of the data buffer 1992 * 1993 * Enqueue a new receive buffer onto the transport queue into which a NTB 1994 * payload can be received into. 1995 * 1996 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 1997 */ 1998 int ntb_transport_rx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, 1999 unsigned int len) 2000 { 2001 struct ntb_queue_entry *entry; 2002 2003 if (!qp) 2004 return -EINVAL; 2005 2006 entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q); 2007 if (!entry) 2008 return -ENOMEM; 2009 2010 entry->cb_data = cb; 2011 entry->buf = data; 2012 entry->len = len; 2013 entry->flags = 0; 2014 entry->retries = 0; 2015 entry->errors = 0; 2016 entry->rx_index = 0; 2017 2018 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_pend_q); 2019 2020 if (qp->active) 2021 tasklet_schedule(&qp->rxc_db_work); 2022 2023 return 0; 2024 } 2025 EXPORT_SYMBOL_GPL(ntb_transport_rx_enqueue); 2026 2027 /** 2028 * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry 2029 * @qp: NTB transport layer queue the entry is to be enqueued on 2030 * @cb: per buffer pointer for callback function to use 2031 * @data: pointer to data buffer that will be sent 2032 * @len: length of the data buffer 2033 * 2034 * Enqueue a new transmit buffer onto the transport queue from which a NTB 2035 * payload will be transmitted. This assumes that a lock is being held to 2036 * serialize access to the qp. 2037 * 2038 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 2039 */ 2040 int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, 2041 unsigned int len) 2042 { 2043 struct ntb_queue_entry *entry; 2044 int rc; 2045 2046 if (!qp || !qp->link_is_up || !len) 2047 return -EINVAL; 2048 2049 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); 2050 if (!entry) { 2051 qp->tx_err_no_buf++; 2052 return -EBUSY; 2053 } 2054 2055 entry->cb_data = cb; 2056 entry->buf = data; 2057 entry->len = len; 2058 entry->flags = 0; 2059 entry->errors = 0; 2060 entry->retries = 0; 2061 entry->tx_index = 0; 2062 2063 rc = ntb_process_tx(qp, entry); 2064 if (rc) 2065 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 2066 &qp->tx_free_q); 2067 2068 return rc; 2069 } 2070 EXPORT_SYMBOL_GPL(ntb_transport_tx_enqueue); 2071 2072 /** 2073 * ntb_transport_link_up - Notify NTB transport of client readiness to use queue 2074 * @qp: NTB transport layer queue to be enabled 2075 * 2076 * Notify NTB transport layer of client readiness to use queue 2077 */ 2078 void ntb_transport_link_up(struct ntb_transport_qp *qp) 2079 { 2080 if (!qp) 2081 return; 2082 2083 qp->client_ready = true; 2084 2085 if (qp->transport->link_is_up) 2086 schedule_delayed_work(&qp->link_work, 0); 2087 } 2088 EXPORT_SYMBOL_GPL(ntb_transport_link_up); 2089 2090 /** 2091 * ntb_transport_link_down - Notify NTB transport to no longer enqueue data 2092 * @qp: NTB transport layer queue to be disabled 2093 * 2094 * Notify NTB transport layer of client's desire to no longer receive data on 2095 * transport queue specified. It is the client's responsibility to ensure all 2096 * entries on queue are purged or otherwise handled appropriately. 2097 */ 2098 void ntb_transport_link_down(struct ntb_transport_qp *qp) 2099 { 2100 int val; 2101 2102 if (!qp) 2103 return; 2104 2105 qp->client_ready = false; 2106 2107 val = ntb_spad_read(qp->ndev, QP_LINKS); 2108 2109 ntb_peer_spad_write(qp->ndev, PIDX, QP_LINKS, val & ~BIT(qp->qp_num)); 2110 2111 if (qp->link_is_up) 2112 ntb_send_link_down(qp); 2113 else 2114 cancel_delayed_work_sync(&qp->link_work); 2115 } 2116 EXPORT_SYMBOL_GPL(ntb_transport_link_down); 2117 2118 /** 2119 * ntb_transport_link_query - Query transport link state 2120 * @qp: NTB transport layer queue to be queried 2121 * 2122 * Query connectivity to the remote system of the NTB transport queue 2123 * 2124 * RETURNS: true for link up or false for link down 2125 */ 2126 bool ntb_transport_link_query(struct ntb_transport_qp *qp) 2127 { 2128 if (!qp) 2129 return false; 2130 2131 return qp->link_is_up; 2132 } 2133 EXPORT_SYMBOL_GPL(ntb_transport_link_query); 2134 2135 /** 2136 * ntb_transport_qp_num - Query the qp number 2137 * @qp: NTB transport layer queue to be queried 2138 * 2139 * Query qp number of the NTB transport queue 2140 * 2141 * RETURNS: a zero based number specifying the qp number 2142 */ 2143 unsigned char ntb_transport_qp_num(struct ntb_transport_qp *qp) 2144 { 2145 if (!qp) 2146 return 0; 2147 2148 return qp->qp_num; 2149 } 2150 EXPORT_SYMBOL_GPL(ntb_transport_qp_num); 2151 2152 /** 2153 * ntb_transport_max_size - Query the max payload size of a qp 2154 * @qp: NTB transport layer queue to be queried 2155 * 2156 * Query the maximum payload size permissible on the given qp 2157 * 2158 * RETURNS: the max payload size of a qp 2159 */ 2160 unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp) 2161 { 2162 unsigned int max_size; 2163 unsigned int copy_align; 2164 struct dma_chan *rx_chan, *tx_chan; 2165 2166 if (!qp) 2167 return 0; 2168 2169 rx_chan = qp->rx_dma_chan; 2170 tx_chan = qp->tx_dma_chan; 2171 2172 copy_align = max(rx_chan ? rx_chan->device->copy_align : 0, 2173 tx_chan ? tx_chan->device->copy_align : 0); 2174 2175 /* If DMA engine usage is possible, try to find the max size for that */ 2176 max_size = qp->tx_max_frame - sizeof(struct ntb_payload_header); 2177 max_size = round_down(max_size, 1 << copy_align); 2178 2179 return max_size; 2180 } 2181 EXPORT_SYMBOL_GPL(ntb_transport_max_size); 2182 2183 unsigned int ntb_transport_tx_free_entry(struct ntb_transport_qp *qp) 2184 { 2185 unsigned int head = qp->tx_index; 2186 unsigned int tail = qp->remote_rx_info->entry; 2187 2188 return tail > head ? tail - head : qp->tx_max_entry + tail - head; 2189 } 2190 EXPORT_SYMBOL_GPL(ntb_transport_tx_free_entry); 2191 2192 static void ntb_transport_doorbell_callback(void *data, int vector) 2193 { 2194 struct ntb_transport_ctx *nt = data; 2195 struct ntb_transport_qp *qp; 2196 u64 db_bits; 2197 unsigned int qp_num; 2198 2199 db_bits = (nt->qp_bitmap & ~nt->qp_bitmap_free & 2200 ntb_db_vector_mask(nt->ndev, vector)); 2201 2202 while (db_bits) { 2203 qp_num = __ffs(db_bits); 2204 qp = &nt->qp_vec[qp_num]; 2205 2206 if (qp->active) 2207 tasklet_schedule(&qp->rxc_db_work); 2208 2209 db_bits &= ~BIT_ULL(qp_num); 2210 } 2211 } 2212 2213 static const struct ntb_ctx_ops ntb_transport_ops = { 2214 .link_event = ntb_transport_event_callback, 2215 .db_event = ntb_transport_doorbell_callback, 2216 }; 2217 2218 static struct ntb_client ntb_transport_client = { 2219 .ops = { 2220 .probe = ntb_transport_probe, 2221 .remove = ntb_transport_free, 2222 }, 2223 }; 2224 2225 static int __init ntb_transport_init(void) 2226 { 2227 int rc; 2228 2229 pr_info("%s, version %s\n", NTB_TRANSPORT_DESC, NTB_TRANSPORT_VER); 2230 2231 if (debugfs_initialized()) 2232 nt_debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL); 2233 2234 rc = bus_register(&ntb_transport_bus); 2235 if (rc) 2236 goto err_bus; 2237 2238 rc = ntb_register_client(&ntb_transport_client); 2239 if (rc) 2240 goto err_client; 2241 2242 return 0; 2243 2244 err_client: 2245 bus_unregister(&ntb_transport_bus); 2246 err_bus: 2247 debugfs_remove_recursive(nt_debugfs_dir); 2248 return rc; 2249 } 2250 module_init(ntb_transport_init); 2251 2252 static void __exit ntb_transport_exit(void) 2253 { 2254 ntb_unregister_client(&ntb_transport_client); 2255 bus_unregister(&ntb_transport_bus); 2256 debugfs_remove_recursive(nt_debugfs_dir); 2257 } 2258 module_exit(ntb_transport_exit); 2259