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 70 MODULE_DESCRIPTION(NTB_TRANSPORT_DESC); 71 MODULE_VERSION(NTB_TRANSPORT_VER); 72 MODULE_LICENSE("Dual BSD/GPL"); 73 MODULE_AUTHOR("Intel Corporation"); 74 75 static unsigned long max_mw_size; 76 module_param(max_mw_size, ulong, 0644); 77 MODULE_PARM_DESC(max_mw_size, "Limit size of large memory windows"); 78 79 static unsigned int transport_mtu = 0x10000; 80 module_param(transport_mtu, uint, 0644); 81 MODULE_PARM_DESC(transport_mtu, "Maximum size of NTB transport packets"); 82 83 static unsigned char max_num_clients; 84 module_param(max_num_clients, byte, 0644); 85 MODULE_PARM_DESC(max_num_clients, "Maximum number of NTB transport clients"); 86 87 static unsigned int copy_bytes = 1024; 88 module_param(copy_bytes, uint, 0644); 89 MODULE_PARM_DESC(copy_bytes, "Threshold under which NTB will use the CPU to copy instead of DMA"); 90 91 static bool use_dma; 92 module_param(use_dma, bool, 0644); 93 MODULE_PARM_DESC(use_dma, "Use DMA engine to perform large data copy"); 94 95 static struct dentry *nt_debugfs_dir; 96 97 struct ntb_queue_entry { 98 /* ntb_queue list reference */ 99 struct list_head entry; 100 /* pointers to data to be transferred */ 101 void *cb_data; 102 void *buf; 103 unsigned int len; 104 unsigned int flags; 105 106 struct ntb_transport_qp *qp; 107 union { 108 struct ntb_payload_header __iomem *tx_hdr; 109 struct ntb_payload_header *rx_hdr; 110 }; 111 unsigned int index; 112 }; 113 114 struct ntb_rx_info { 115 unsigned int entry; 116 }; 117 118 struct ntb_transport_qp { 119 struct ntb_transport_ctx *transport; 120 struct ntb_dev *ndev; 121 void *cb_data; 122 struct dma_chan *tx_dma_chan; 123 struct dma_chan *rx_dma_chan; 124 125 bool client_ready; 126 bool link_is_up; 127 bool active; 128 129 u8 qp_num; /* Only 64 QP's are allowed. 0-63 */ 130 u64 qp_bit; 131 132 struct ntb_rx_info __iomem *rx_info; 133 struct ntb_rx_info *remote_rx_info; 134 135 void (*tx_handler)(struct ntb_transport_qp *qp, void *qp_data, 136 void *data, int len); 137 struct list_head tx_free_q; 138 spinlock_t ntb_tx_free_q_lock; 139 void __iomem *tx_mw; 140 dma_addr_t tx_mw_phys; 141 unsigned int tx_index; 142 unsigned int tx_max_entry; 143 unsigned int tx_max_frame; 144 145 void (*rx_handler)(struct ntb_transport_qp *qp, void *qp_data, 146 void *data, int len); 147 struct list_head rx_post_q; 148 struct list_head rx_pend_q; 149 struct list_head rx_free_q; 150 /* ntb_rx_q_lock: synchronize access to rx_XXXX_q */ 151 spinlock_t ntb_rx_q_lock; 152 void *rx_buff; 153 unsigned int rx_index; 154 unsigned int rx_max_entry; 155 unsigned int rx_max_frame; 156 unsigned int rx_alloc_entry; 157 dma_cookie_t last_cookie; 158 struct tasklet_struct rxc_db_work; 159 160 void (*event_handler)(void *data, int status); 161 struct delayed_work link_work; 162 struct work_struct link_cleanup; 163 164 struct dentry *debugfs_dir; 165 struct dentry *debugfs_stats; 166 167 /* Stats */ 168 u64 rx_bytes; 169 u64 rx_pkts; 170 u64 rx_ring_empty; 171 u64 rx_err_no_buf; 172 u64 rx_err_oflow; 173 u64 rx_err_ver; 174 u64 rx_memcpy; 175 u64 rx_async; 176 u64 dma_rx_prep_err; 177 u64 tx_bytes; 178 u64 tx_pkts; 179 u64 tx_ring_full; 180 u64 tx_err_no_buf; 181 u64 tx_memcpy; 182 u64 tx_async; 183 u64 dma_tx_prep_err; 184 }; 185 186 struct ntb_transport_mw { 187 phys_addr_t phys_addr; 188 resource_size_t phys_size; 189 resource_size_t xlat_align; 190 resource_size_t xlat_align_size; 191 void __iomem *vbase; 192 size_t xlat_size; 193 size_t buff_size; 194 void *virt_addr; 195 dma_addr_t dma_addr; 196 }; 197 198 struct ntb_transport_client_dev { 199 struct list_head entry; 200 struct ntb_transport_ctx *nt; 201 struct device dev; 202 }; 203 204 struct ntb_transport_ctx { 205 struct list_head entry; 206 struct list_head client_devs; 207 208 struct ntb_dev *ndev; 209 210 struct ntb_transport_mw *mw_vec; 211 struct ntb_transport_qp *qp_vec; 212 unsigned int mw_count; 213 unsigned int qp_count; 214 u64 qp_bitmap; 215 u64 qp_bitmap_free; 216 217 bool link_is_up; 218 struct delayed_work link_work; 219 struct work_struct link_cleanup; 220 221 struct dentry *debugfs_node_dir; 222 }; 223 224 enum { 225 DESC_DONE_FLAG = BIT(0), 226 LINK_DOWN_FLAG = BIT(1), 227 }; 228 229 struct ntb_payload_header { 230 unsigned int ver; 231 unsigned int len; 232 unsigned int flags; 233 }; 234 235 enum { 236 VERSION = 0, 237 QP_LINKS, 238 NUM_QPS, 239 NUM_MWS, 240 MW0_SZ_HIGH, 241 MW0_SZ_LOW, 242 MW1_SZ_HIGH, 243 MW1_SZ_LOW, 244 MAX_SPAD, 245 }; 246 247 #define dev_client_dev(__dev) \ 248 container_of((__dev), struct ntb_transport_client_dev, dev) 249 250 #define drv_client(__drv) \ 251 container_of((__drv), struct ntb_transport_client, driver) 252 253 #define QP_TO_MW(nt, qp) ((qp) % nt->mw_count) 254 #define NTB_QP_DEF_NUM_ENTRIES 100 255 #define NTB_LINK_DOWN_TIMEOUT 10 256 #define DMA_RETRIES 20 257 #define DMA_OUT_RESOURCE_TO 50 258 259 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; 262 263 static int ntb_transport_bus_match(struct device *dev, 264 struct device_driver *drv) 265 { 266 return !strncmp(dev_name(dev), drv->name, strlen(drv->name)); 267 } 268 269 static int ntb_transport_bus_probe(struct device *dev) 270 { 271 const struct ntb_transport_client *client; 272 int rc = -EINVAL; 273 274 get_device(dev); 275 276 client = drv_client(dev->driver); 277 rc = client->probe(dev); 278 if (rc) 279 put_device(dev); 280 281 return rc; 282 } 283 284 static int ntb_transport_bus_remove(struct device *dev) 285 { 286 const struct ntb_transport_client *client; 287 288 client = drv_client(dev->driver); 289 client->remove(dev); 290 291 put_device(dev); 292 293 return 0; 294 } 295 296 static struct bus_type ntb_transport_bus = { 297 .name = "ntb_transport", 298 .match = ntb_transport_bus_match, 299 .probe = ntb_transport_bus_probe, 300 .remove = ntb_transport_bus_remove, 301 }; 302 303 static LIST_HEAD(ntb_transport_list); 304 305 static int ntb_bus_init(struct ntb_transport_ctx *nt) 306 { 307 list_add_tail(&nt->entry, &ntb_transport_list); 308 return 0; 309 } 310 311 static void ntb_bus_remove(struct ntb_transport_ctx *nt) 312 { 313 struct ntb_transport_client_dev *client_dev, *cd; 314 315 list_for_each_entry_safe(client_dev, cd, &nt->client_devs, entry) { 316 dev_err(client_dev->dev.parent, "%s still attached to bus, removing\n", 317 dev_name(&client_dev->dev)); 318 list_del(&client_dev->entry); 319 device_unregister(&client_dev->dev); 320 } 321 322 list_del(&nt->entry); 323 } 324 325 static void ntb_transport_client_release(struct device *dev) 326 { 327 struct ntb_transport_client_dev *client_dev; 328 329 client_dev = dev_client_dev(dev); 330 kfree(client_dev); 331 } 332 333 /** 334 * ntb_transport_unregister_client_dev - Unregister NTB client device 335 * @device_name: Name of NTB client device 336 * 337 * Unregister an NTB client device with the NTB transport layer 338 */ 339 void ntb_transport_unregister_client_dev(char *device_name) 340 { 341 struct ntb_transport_client_dev *client, *cd; 342 struct ntb_transport_ctx *nt; 343 344 list_for_each_entry(nt, &ntb_transport_list, entry) 345 list_for_each_entry_safe(client, cd, &nt->client_devs, entry) 346 if (!strncmp(dev_name(&client->dev), device_name, 347 strlen(device_name))) { 348 list_del(&client->entry); 349 device_unregister(&client->dev); 350 } 351 } 352 EXPORT_SYMBOL_GPL(ntb_transport_unregister_client_dev); 353 354 /** 355 * ntb_transport_register_client_dev - Register NTB client device 356 * @device_name: Name of NTB client device 357 * 358 * Register an NTB client device with the NTB transport layer 359 */ 360 int ntb_transport_register_client_dev(char *device_name) 361 { 362 struct ntb_transport_client_dev *client_dev; 363 struct ntb_transport_ctx *nt; 364 int node; 365 int rc, i = 0; 366 367 if (list_empty(&ntb_transport_list)) 368 return -ENODEV; 369 370 list_for_each_entry(nt, &ntb_transport_list, entry) { 371 struct device *dev; 372 373 node = dev_to_node(&nt->ndev->dev); 374 375 client_dev = kzalloc_node(sizeof(*client_dev), 376 GFP_KERNEL, node); 377 if (!client_dev) { 378 rc = -ENOMEM; 379 goto err; 380 } 381 382 dev = &client_dev->dev; 383 384 /* setup and register client devices */ 385 dev_set_name(dev, "%s%d", device_name, i); 386 dev->bus = &ntb_transport_bus; 387 dev->release = ntb_transport_client_release; 388 dev->parent = &nt->ndev->dev; 389 390 rc = device_register(dev); 391 if (rc) { 392 kfree(client_dev); 393 goto err; 394 } 395 396 list_add_tail(&client_dev->entry, &nt->client_devs); 397 i++; 398 } 399 400 return 0; 401 402 err: 403 ntb_transport_unregister_client_dev(device_name); 404 405 return rc; 406 } 407 EXPORT_SYMBOL_GPL(ntb_transport_register_client_dev); 408 409 /** 410 * ntb_transport_register_client - Register NTB client driver 411 * @drv: NTB client driver to be registered 412 * 413 * Register an NTB client driver with the NTB transport layer 414 * 415 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 416 */ 417 int ntb_transport_register_client(struct ntb_transport_client *drv) 418 { 419 drv->driver.bus = &ntb_transport_bus; 420 421 if (list_empty(&ntb_transport_list)) 422 return -ENODEV; 423 424 return driver_register(&drv->driver); 425 } 426 EXPORT_SYMBOL_GPL(ntb_transport_register_client); 427 428 /** 429 * ntb_transport_unregister_client - Unregister NTB client driver 430 * @drv: NTB client driver to be unregistered 431 * 432 * Unregister an NTB client driver with the NTB transport layer 433 * 434 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 435 */ 436 void ntb_transport_unregister_client(struct ntb_transport_client *drv) 437 { 438 driver_unregister(&drv->driver); 439 } 440 EXPORT_SYMBOL_GPL(ntb_transport_unregister_client); 441 442 static ssize_t debugfs_read(struct file *filp, char __user *ubuf, size_t count, 443 loff_t *offp) 444 { 445 struct ntb_transport_qp *qp; 446 char *buf; 447 ssize_t ret, out_offset, out_count; 448 449 qp = filp->private_data; 450 451 if (!qp || !qp->link_is_up) 452 return 0; 453 454 out_count = 1000; 455 456 buf = kmalloc(out_count, GFP_KERNEL); 457 if (!buf) 458 return -ENOMEM; 459 460 out_offset = 0; 461 out_offset += snprintf(buf + out_offset, out_count - out_offset, 462 "\nNTB QP stats:\n\n"); 463 out_offset += snprintf(buf + out_offset, out_count - out_offset, 464 "rx_bytes - \t%llu\n", qp->rx_bytes); 465 out_offset += snprintf(buf + out_offset, out_count - out_offset, 466 "rx_pkts - \t%llu\n", qp->rx_pkts); 467 out_offset += snprintf(buf + out_offset, out_count - out_offset, 468 "rx_memcpy - \t%llu\n", qp->rx_memcpy); 469 out_offset += snprintf(buf + out_offset, out_count - out_offset, 470 "rx_async - \t%llu\n", qp->rx_async); 471 out_offset += snprintf(buf + out_offset, out_count - out_offset, 472 "rx_ring_empty - %llu\n", qp->rx_ring_empty); 473 out_offset += snprintf(buf + out_offset, out_count - out_offset, 474 "rx_err_no_buf - %llu\n", qp->rx_err_no_buf); 475 out_offset += snprintf(buf + out_offset, out_count - out_offset, 476 "rx_err_oflow - \t%llu\n", qp->rx_err_oflow); 477 out_offset += snprintf(buf + out_offset, out_count - out_offset, 478 "rx_err_ver - \t%llu\n", qp->rx_err_ver); 479 out_offset += snprintf(buf + out_offset, out_count - out_offset, 480 "rx_buff - \t0x%p\n", qp->rx_buff); 481 out_offset += snprintf(buf + out_offset, out_count - out_offset, 482 "rx_index - \t%u\n", qp->rx_index); 483 out_offset += snprintf(buf + out_offset, out_count - out_offset, 484 "rx_max_entry - \t%u\n", qp->rx_max_entry); 485 out_offset += snprintf(buf + out_offset, out_count - out_offset, 486 "rx_alloc_entry - \t%u\n\n", qp->rx_alloc_entry); 487 488 out_offset += snprintf(buf + out_offset, out_count - out_offset, 489 "tx_bytes - \t%llu\n", qp->tx_bytes); 490 out_offset += snprintf(buf + out_offset, out_count - out_offset, 491 "tx_pkts - \t%llu\n", qp->tx_pkts); 492 out_offset += snprintf(buf + out_offset, out_count - out_offset, 493 "tx_memcpy - \t%llu\n", qp->tx_memcpy); 494 out_offset += snprintf(buf + out_offset, out_count - out_offset, 495 "tx_async - \t%llu\n", qp->tx_async); 496 out_offset += snprintf(buf + out_offset, out_count - out_offset, 497 "tx_ring_full - \t%llu\n", qp->tx_ring_full); 498 out_offset += snprintf(buf + out_offset, out_count - out_offset, 499 "tx_err_no_buf - %llu\n", qp->tx_err_no_buf); 500 out_offset += snprintf(buf + out_offset, out_count - out_offset, 501 "tx_mw - \t0x%p\n", qp->tx_mw); 502 out_offset += snprintf(buf + out_offset, out_count - out_offset, 503 "tx_index (H) - \t%u\n", qp->tx_index); 504 out_offset += snprintf(buf + out_offset, out_count - out_offset, 505 "RRI (T) - \t%u\n", 506 qp->remote_rx_info->entry); 507 out_offset += snprintf(buf + out_offset, out_count - out_offset, 508 "tx_max_entry - \t%u\n", qp->tx_max_entry); 509 out_offset += snprintf(buf + out_offset, out_count - out_offset, 510 "free tx - \t%u\n", 511 ntb_transport_tx_free_entry(qp)); 512 out_offset += snprintf(buf + out_offset, out_count - out_offset, 513 "DMA tx prep err - \t%llu\n", 514 qp->dma_tx_prep_err); 515 out_offset += snprintf(buf + out_offset, out_count - out_offset, 516 "DMA rx prep err - \t%llu\n", 517 qp->dma_rx_prep_err); 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 (qp_count % mw_count && mw_num + 1 < 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, 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, num_mw, mw->dma_addr, mw->xlat_size); 734 if (rc) { 735 dev_err(&pdev->dev, "Unable to set mw%d translation", num_mw); 736 ntb_free_mw(nt, num_mw); 737 return -EIO; 738 } 739 740 return 0; 741 } 742 743 static void ntb_qp_link_down_reset(struct ntb_transport_qp *qp) 744 { 745 qp->link_is_up = false; 746 qp->active = false; 747 748 qp->tx_index = 0; 749 qp->rx_index = 0; 750 qp->rx_bytes = 0; 751 qp->rx_pkts = 0; 752 qp->rx_ring_empty = 0; 753 qp->rx_err_no_buf = 0; 754 qp->rx_err_oflow = 0; 755 qp->rx_err_ver = 0; 756 qp->rx_memcpy = 0; 757 qp->rx_async = 0; 758 qp->tx_bytes = 0; 759 qp->tx_pkts = 0; 760 qp->tx_ring_full = 0; 761 qp->tx_err_no_buf = 0; 762 qp->tx_memcpy = 0; 763 qp->tx_async = 0; 764 qp->dma_tx_prep_err = 0; 765 qp->dma_rx_prep_err = 0; 766 } 767 768 static void ntb_qp_link_cleanup(struct ntb_transport_qp *qp) 769 { 770 struct ntb_transport_ctx *nt = qp->transport; 771 struct pci_dev *pdev = nt->ndev->pdev; 772 773 dev_info(&pdev->dev, "qp %d: Link Cleanup\n", qp->qp_num); 774 775 cancel_delayed_work_sync(&qp->link_work); 776 ntb_qp_link_down_reset(qp); 777 778 if (qp->event_handler) 779 qp->event_handler(qp->cb_data, qp->link_is_up); 780 } 781 782 static void ntb_qp_link_cleanup_work(struct work_struct *work) 783 { 784 struct ntb_transport_qp *qp = container_of(work, 785 struct ntb_transport_qp, 786 link_cleanup); 787 struct ntb_transport_ctx *nt = qp->transport; 788 789 ntb_qp_link_cleanup(qp); 790 791 if (nt->link_is_up) 792 schedule_delayed_work(&qp->link_work, 793 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 794 } 795 796 static void ntb_qp_link_down(struct ntb_transport_qp *qp) 797 { 798 schedule_work(&qp->link_cleanup); 799 } 800 801 static void ntb_transport_link_cleanup(struct ntb_transport_ctx *nt) 802 { 803 struct ntb_transport_qp *qp; 804 u64 qp_bitmap_alloc; 805 int i; 806 807 qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free; 808 809 /* Pass along the info to any clients */ 810 for (i = 0; i < nt->qp_count; i++) 811 if (qp_bitmap_alloc & BIT_ULL(i)) { 812 qp = &nt->qp_vec[i]; 813 ntb_qp_link_cleanup(qp); 814 cancel_work_sync(&qp->link_cleanup); 815 cancel_delayed_work_sync(&qp->link_work); 816 } 817 818 if (!nt->link_is_up) 819 cancel_delayed_work_sync(&nt->link_work); 820 821 /* The scratchpad registers keep the values if the remote side 822 * goes down, blast them now to give them a sane value the next 823 * time they are accessed 824 */ 825 for (i = 0; i < MAX_SPAD; 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, spad, upper_32_bits(size)); 866 867 spad = MW0_SZ_LOW + (i * 2); 868 ntb_peer_spad_write(ndev, spad, lower_32_bits(size)); 869 } 870 871 ntb_peer_spad_write(ndev, NUM_MWS, nt->mw_count); 872 873 ntb_peer_spad_write(ndev, NUM_QPS, nt->qp_count); 874 875 ntb_peer_spad_write(ndev, 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, QP_LINKS, val | BIT(qp->qp_num)); 952 953 /* query remote spad for qp ready bits */ 954 ntb_peer_spad_read(nt->ndev, QP_LINKS); 955 dev_dbg_ratelimited(&pdev->dev, "Remote QP link status = %x\n", val); 956 957 /* See if the remote side is up */ 958 if (val & BIT(qp->qp_num)) { 959 dev_info(&pdev->dev, "qp %d: Link Up\n", qp->qp_num); 960 qp->link_is_up = true; 961 qp->active = true; 962 963 if (qp->event_handler) 964 qp->event_handler(qp->cb_data, qp->link_is_up); 965 966 if (qp->active) 967 tasklet_schedule(&qp->rxc_db_work); 968 } else if (nt->link_is_up) 969 schedule_delayed_work(&qp->link_work, 970 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 971 } 972 973 static int ntb_transport_init_queue(struct ntb_transport_ctx *nt, 974 unsigned int qp_num) 975 { 976 struct ntb_transport_qp *qp; 977 phys_addr_t mw_base; 978 resource_size_t mw_size; 979 unsigned int num_qps_mw, tx_size; 980 unsigned int mw_num, mw_count, qp_count; 981 u64 qp_offset; 982 983 mw_count = nt->mw_count; 984 qp_count = nt->qp_count; 985 986 mw_num = QP_TO_MW(nt, qp_num); 987 988 qp = &nt->qp_vec[qp_num]; 989 qp->qp_num = qp_num; 990 qp->transport = nt; 991 qp->ndev = nt->ndev; 992 qp->client_ready = false; 993 qp->event_handler = NULL; 994 ntb_qp_link_down_reset(qp); 995 996 if (qp_count % mw_count && mw_num + 1 < qp_count / mw_count) 997 num_qps_mw = qp_count / mw_count + 1; 998 else 999 num_qps_mw = qp_count / mw_count; 1000 1001 mw_base = nt->mw_vec[mw_num].phys_addr; 1002 mw_size = nt->mw_vec[mw_num].phys_size; 1003 1004 tx_size = (unsigned int)mw_size / num_qps_mw; 1005 qp_offset = tx_size * (qp_num / mw_count); 1006 1007 qp->tx_mw = nt->mw_vec[mw_num].vbase + qp_offset; 1008 if (!qp->tx_mw) 1009 return -EINVAL; 1010 1011 qp->tx_mw_phys = mw_base + qp_offset; 1012 if (!qp->tx_mw_phys) 1013 return -EINVAL; 1014 1015 tx_size -= sizeof(struct ntb_rx_info); 1016 qp->rx_info = qp->tx_mw + tx_size; 1017 1018 /* Due to housekeeping, there must be atleast 2 buffs */ 1019 qp->tx_max_frame = min(transport_mtu, tx_size / 2); 1020 qp->tx_max_entry = tx_size / qp->tx_max_frame; 1021 1022 if (nt->debugfs_node_dir) { 1023 char debugfs_name[4]; 1024 1025 snprintf(debugfs_name, 4, "qp%d", qp_num); 1026 qp->debugfs_dir = debugfs_create_dir(debugfs_name, 1027 nt->debugfs_node_dir); 1028 1029 qp->debugfs_stats = debugfs_create_file("stats", S_IRUSR, 1030 qp->debugfs_dir, qp, 1031 &ntb_qp_debugfs_stats); 1032 } else { 1033 qp->debugfs_dir = NULL; 1034 qp->debugfs_stats = NULL; 1035 } 1036 1037 INIT_DELAYED_WORK(&qp->link_work, ntb_qp_link_work); 1038 INIT_WORK(&qp->link_cleanup, ntb_qp_link_cleanup_work); 1039 1040 spin_lock_init(&qp->ntb_rx_q_lock); 1041 spin_lock_init(&qp->ntb_tx_free_q_lock); 1042 1043 INIT_LIST_HEAD(&qp->rx_post_q); 1044 INIT_LIST_HEAD(&qp->rx_pend_q); 1045 INIT_LIST_HEAD(&qp->rx_free_q); 1046 INIT_LIST_HEAD(&qp->tx_free_q); 1047 1048 tasklet_init(&qp->rxc_db_work, ntb_transport_rxc_db, 1049 (unsigned long)qp); 1050 1051 return 0; 1052 } 1053 1054 static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev) 1055 { 1056 struct ntb_transport_ctx *nt; 1057 struct ntb_transport_mw *mw; 1058 unsigned int mw_count, qp_count; 1059 u64 qp_bitmap; 1060 int node; 1061 int rc, i; 1062 1063 mw_count = ntb_mw_count(ndev); 1064 if (ntb_spad_count(ndev) < (NUM_MWS + 1 + mw_count * 2)) { 1065 dev_err(&ndev->dev, "Not enough scratch pad registers for %s", 1066 NTB_TRANSPORT_NAME); 1067 return -EIO; 1068 } 1069 1070 if (ntb_db_is_unsafe(ndev)) 1071 dev_dbg(&ndev->dev, 1072 "doorbell is unsafe, proceed anyway...\n"); 1073 if (ntb_spad_is_unsafe(ndev)) 1074 dev_dbg(&ndev->dev, 1075 "scratchpad is unsafe, proceed anyway...\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 1085 nt->mw_count = mw_count; 1086 1087 nt->mw_vec = kzalloc_node(mw_count * sizeof(*nt->mw_vec), 1088 GFP_KERNEL, node); 1089 if (!nt->mw_vec) { 1090 rc = -ENOMEM; 1091 goto err; 1092 } 1093 1094 for (i = 0; i < mw_count; i++) { 1095 mw = &nt->mw_vec[i]; 1096 1097 rc = ntb_mw_get_range(ndev, i, &mw->phys_addr, &mw->phys_size, 1098 &mw->xlat_align, &mw->xlat_align_size); 1099 if (rc) 1100 goto err1; 1101 1102 mw->vbase = ioremap_wc(mw->phys_addr, mw->phys_size); 1103 if (!mw->vbase) { 1104 rc = -ENOMEM; 1105 goto err1; 1106 } 1107 1108 mw->buff_size = 0; 1109 mw->xlat_size = 0; 1110 mw->virt_addr = NULL; 1111 mw->dma_addr = 0; 1112 } 1113 1114 qp_bitmap = ntb_db_valid_mask(ndev); 1115 1116 qp_count = ilog2(qp_bitmap); 1117 if (max_num_clients && max_num_clients < qp_count) 1118 qp_count = max_num_clients; 1119 else if (mw_count < qp_count) 1120 qp_count = mw_count; 1121 1122 qp_bitmap &= BIT_ULL(qp_count) - 1; 1123 1124 nt->qp_count = qp_count; 1125 nt->qp_bitmap = qp_bitmap; 1126 nt->qp_bitmap_free = qp_bitmap; 1127 1128 nt->qp_vec = kzalloc_node(qp_count * sizeof(*nt->qp_vec), 1129 GFP_KERNEL, node); 1130 if (!nt->qp_vec) { 1131 rc = -ENOMEM; 1132 goto err1; 1133 } 1134 1135 if (nt_debugfs_dir) { 1136 nt->debugfs_node_dir = 1137 debugfs_create_dir(pci_name(ndev->pdev), 1138 nt_debugfs_dir); 1139 } 1140 1141 for (i = 0; i < qp_count; i++) { 1142 rc = ntb_transport_init_queue(nt, i); 1143 if (rc) 1144 goto err2; 1145 } 1146 1147 INIT_DELAYED_WORK(&nt->link_work, ntb_transport_link_work); 1148 INIT_WORK(&nt->link_cleanup, ntb_transport_link_cleanup_work); 1149 1150 rc = ntb_set_ctx(ndev, nt, &ntb_transport_ops); 1151 if (rc) 1152 goto err2; 1153 1154 INIT_LIST_HEAD(&nt->client_devs); 1155 rc = ntb_bus_init(nt); 1156 if (rc) 1157 goto err3; 1158 1159 nt->link_is_up = false; 1160 ntb_link_enable(ndev, NTB_SPEED_AUTO, NTB_WIDTH_AUTO); 1161 ntb_link_event(ndev); 1162 1163 return 0; 1164 1165 err3: 1166 ntb_clear_ctx(ndev); 1167 err2: 1168 kfree(nt->qp_vec); 1169 err1: 1170 while (i--) { 1171 mw = &nt->mw_vec[i]; 1172 iounmap(mw->vbase); 1173 } 1174 kfree(nt->mw_vec); 1175 err: 1176 kfree(nt); 1177 return rc; 1178 } 1179 1180 static void ntb_transport_free(struct ntb_client *self, struct ntb_dev *ndev) 1181 { 1182 struct ntb_transport_ctx *nt = ndev->ctx; 1183 struct ntb_transport_qp *qp; 1184 u64 qp_bitmap_alloc; 1185 int i; 1186 1187 ntb_transport_link_cleanup(nt); 1188 cancel_work_sync(&nt->link_cleanup); 1189 cancel_delayed_work_sync(&nt->link_work); 1190 1191 qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free; 1192 1193 /* verify that all the qp's are freed */ 1194 for (i = 0; i < nt->qp_count; i++) { 1195 qp = &nt->qp_vec[i]; 1196 if (qp_bitmap_alloc & BIT_ULL(i)) 1197 ntb_transport_free_queue(qp); 1198 debugfs_remove_recursive(qp->debugfs_dir); 1199 } 1200 1201 ntb_link_disable(ndev); 1202 ntb_clear_ctx(ndev); 1203 1204 ntb_bus_remove(nt); 1205 1206 for (i = nt->mw_count; i--; ) { 1207 ntb_free_mw(nt, i); 1208 iounmap(nt->mw_vec[i].vbase); 1209 } 1210 1211 kfree(nt->qp_vec); 1212 kfree(nt->mw_vec); 1213 kfree(nt); 1214 } 1215 1216 static void ntb_complete_rxc(struct ntb_transport_qp *qp) 1217 { 1218 struct ntb_queue_entry *entry; 1219 void *cb_data; 1220 unsigned int len; 1221 unsigned long irqflags; 1222 1223 spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags); 1224 1225 while (!list_empty(&qp->rx_post_q)) { 1226 entry = list_first_entry(&qp->rx_post_q, 1227 struct ntb_queue_entry, entry); 1228 if (!(entry->flags & DESC_DONE_FLAG)) 1229 break; 1230 1231 entry->rx_hdr->flags = 0; 1232 iowrite32(entry->index, &qp->rx_info->entry); 1233 1234 cb_data = entry->cb_data; 1235 len = entry->len; 1236 1237 list_move_tail(&entry->entry, &qp->rx_free_q); 1238 1239 spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags); 1240 1241 if (qp->rx_handler && qp->client_ready) 1242 qp->rx_handler(qp, qp->cb_data, cb_data, len); 1243 1244 spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags); 1245 } 1246 1247 spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags); 1248 } 1249 1250 static void ntb_rx_copy_callback(void *data) 1251 { 1252 struct ntb_queue_entry *entry = data; 1253 1254 entry->flags |= DESC_DONE_FLAG; 1255 1256 ntb_complete_rxc(entry->qp); 1257 } 1258 1259 static void ntb_memcpy_rx(struct ntb_queue_entry *entry, void *offset) 1260 { 1261 void *buf = entry->buf; 1262 size_t len = entry->len; 1263 1264 memcpy(buf, offset, len); 1265 1266 /* Ensure that the data is fully copied out before clearing the flag */ 1267 wmb(); 1268 1269 ntb_rx_copy_callback(entry); 1270 } 1271 1272 static void ntb_async_rx(struct ntb_queue_entry *entry, void *offset) 1273 { 1274 struct dma_async_tx_descriptor *txd; 1275 struct ntb_transport_qp *qp = entry->qp; 1276 struct dma_chan *chan = qp->rx_dma_chan; 1277 struct dma_device *device; 1278 size_t pay_off, buff_off, len; 1279 struct dmaengine_unmap_data *unmap; 1280 dma_cookie_t cookie; 1281 void *buf = entry->buf; 1282 int retries = 0; 1283 1284 len = entry->len; 1285 1286 if (!chan) 1287 goto err; 1288 1289 if (len < copy_bytes) 1290 goto err; 1291 1292 device = chan->device; 1293 pay_off = (size_t)offset & ~PAGE_MASK; 1294 buff_off = (size_t)buf & ~PAGE_MASK; 1295 1296 if (!is_dma_copy_aligned(device, pay_off, buff_off, len)) 1297 goto err; 1298 1299 unmap = dmaengine_get_unmap_data(device->dev, 2, GFP_NOWAIT); 1300 if (!unmap) 1301 goto err; 1302 1303 unmap->len = len; 1304 unmap->addr[0] = dma_map_page(device->dev, virt_to_page(offset), 1305 pay_off, len, DMA_TO_DEVICE); 1306 if (dma_mapping_error(device->dev, unmap->addr[0])) 1307 goto err_get_unmap; 1308 1309 unmap->to_cnt = 1; 1310 1311 unmap->addr[1] = dma_map_page(device->dev, virt_to_page(buf), 1312 buff_off, len, DMA_FROM_DEVICE); 1313 if (dma_mapping_error(device->dev, unmap->addr[1])) 1314 goto err_get_unmap; 1315 1316 unmap->from_cnt = 1; 1317 1318 for (retries = 0; retries < DMA_RETRIES; retries++) { 1319 txd = device->device_prep_dma_memcpy(chan, unmap->addr[1], 1320 unmap->addr[0], len, 1321 DMA_PREP_INTERRUPT); 1322 if (txd) 1323 break; 1324 1325 set_current_state(TASK_INTERRUPTIBLE); 1326 schedule_timeout(DMA_OUT_RESOURCE_TO); 1327 } 1328 1329 if (!txd) { 1330 qp->dma_rx_prep_err++; 1331 goto err_get_unmap; 1332 } 1333 1334 txd->callback = ntb_rx_copy_callback; 1335 txd->callback_param = entry; 1336 dma_set_unmap(txd, unmap); 1337 1338 cookie = dmaengine_submit(txd); 1339 if (dma_submit_error(cookie)) 1340 goto err_set_unmap; 1341 1342 dmaengine_unmap_put(unmap); 1343 1344 qp->last_cookie = cookie; 1345 1346 qp->rx_async++; 1347 1348 return; 1349 1350 err_set_unmap: 1351 dmaengine_unmap_put(unmap); 1352 err_get_unmap: 1353 dmaengine_unmap_put(unmap); 1354 err: 1355 ntb_memcpy_rx(entry, offset); 1356 qp->rx_memcpy++; 1357 } 1358 1359 static int ntb_process_rxc(struct ntb_transport_qp *qp) 1360 { 1361 struct ntb_payload_header *hdr; 1362 struct ntb_queue_entry *entry; 1363 void *offset; 1364 1365 offset = qp->rx_buff + qp->rx_max_frame * qp->rx_index; 1366 hdr = offset + qp->rx_max_frame - sizeof(struct ntb_payload_header); 1367 1368 dev_dbg(&qp->ndev->pdev->dev, "qp %d: RX ver %u len %d flags %x\n", 1369 qp->qp_num, hdr->ver, hdr->len, hdr->flags); 1370 1371 if (!(hdr->flags & DESC_DONE_FLAG)) { 1372 dev_dbg(&qp->ndev->pdev->dev, "done flag not set\n"); 1373 qp->rx_ring_empty++; 1374 return -EAGAIN; 1375 } 1376 1377 if (hdr->flags & LINK_DOWN_FLAG) { 1378 dev_dbg(&qp->ndev->pdev->dev, "link down flag set\n"); 1379 ntb_qp_link_down(qp); 1380 hdr->flags = 0; 1381 return -EAGAIN; 1382 } 1383 1384 if (hdr->ver != (u32)qp->rx_pkts) { 1385 dev_dbg(&qp->ndev->pdev->dev, 1386 "version mismatch, expected %llu - got %u\n", 1387 qp->rx_pkts, hdr->ver); 1388 qp->rx_err_ver++; 1389 return -EIO; 1390 } 1391 1392 entry = ntb_list_mv(&qp->ntb_rx_q_lock, &qp->rx_pend_q, &qp->rx_post_q); 1393 if (!entry) { 1394 dev_dbg(&qp->ndev->pdev->dev, "no receive buffer\n"); 1395 qp->rx_err_no_buf++; 1396 return -EAGAIN; 1397 } 1398 1399 entry->rx_hdr = hdr; 1400 entry->index = qp->rx_index; 1401 1402 if (hdr->len > entry->len) { 1403 dev_dbg(&qp->ndev->pdev->dev, 1404 "receive buffer overflow! Wanted %d got %d\n", 1405 hdr->len, entry->len); 1406 qp->rx_err_oflow++; 1407 1408 entry->len = -EIO; 1409 entry->flags |= DESC_DONE_FLAG; 1410 1411 ntb_complete_rxc(qp); 1412 } else { 1413 dev_dbg(&qp->ndev->pdev->dev, 1414 "RX OK index %u ver %u size %d into buf size %d\n", 1415 qp->rx_index, hdr->ver, hdr->len, entry->len); 1416 1417 qp->rx_bytes += hdr->len; 1418 qp->rx_pkts++; 1419 1420 entry->len = hdr->len; 1421 1422 ntb_async_rx(entry, offset); 1423 } 1424 1425 qp->rx_index++; 1426 qp->rx_index %= qp->rx_max_entry; 1427 1428 return 0; 1429 } 1430 1431 static void ntb_transport_rxc_db(unsigned long data) 1432 { 1433 struct ntb_transport_qp *qp = (void *)data; 1434 int rc, i; 1435 1436 dev_dbg(&qp->ndev->pdev->dev, "%s: doorbell %d received\n", 1437 __func__, qp->qp_num); 1438 1439 /* Limit the number of packets processed in a single interrupt to 1440 * provide fairness to others 1441 */ 1442 for (i = 0; i < qp->rx_max_entry; i++) { 1443 rc = ntb_process_rxc(qp); 1444 if (rc) 1445 break; 1446 } 1447 1448 if (i && qp->rx_dma_chan) 1449 dma_async_issue_pending(qp->rx_dma_chan); 1450 1451 if (i == qp->rx_max_entry) { 1452 /* there is more work to do */ 1453 if (qp->active) 1454 tasklet_schedule(&qp->rxc_db_work); 1455 } else if (ntb_db_read(qp->ndev) & BIT_ULL(qp->qp_num)) { 1456 /* the doorbell bit is set: clear it */ 1457 ntb_db_clear(qp->ndev, BIT_ULL(qp->qp_num)); 1458 /* ntb_db_read ensures ntb_db_clear write is committed */ 1459 ntb_db_read(qp->ndev); 1460 1461 /* an interrupt may have arrived between finishing 1462 * ntb_process_rxc and clearing the doorbell bit: 1463 * there might be some more work to do. 1464 */ 1465 if (qp->active) 1466 tasklet_schedule(&qp->rxc_db_work); 1467 } 1468 } 1469 1470 static void ntb_tx_copy_callback(void *data) 1471 { 1472 struct ntb_queue_entry *entry = data; 1473 struct ntb_transport_qp *qp = entry->qp; 1474 struct ntb_payload_header __iomem *hdr = entry->tx_hdr; 1475 1476 iowrite32(entry->flags | DESC_DONE_FLAG, &hdr->flags); 1477 1478 ntb_peer_db_set(qp->ndev, BIT_ULL(qp->qp_num)); 1479 1480 /* The entry length can only be zero if the packet is intended to be a 1481 * "link down" or similar. Since no payload is being sent in these 1482 * cases, there is nothing to add to the completion queue. 1483 */ 1484 if (entry->len > 0) { 1485 qp->tx_bytes += entry->len; 1486 1487 if (qp->tx_handler) 1488 qp->tx_handler(qp, qp->cb_data, entry->cb_data, 1489 entry->len); 1490 } 1491 1492 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, &qp->tx_free_q); 1493 } 1494 1495 static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset) 1496 { 1497 #ifdef ARCH_HAS_NOCACHE_UACCESS 1498 /* 1499 * Using non-temporal mov to improve performance on non-cached 1500 * writes, even though we aren't actually copying from user space. 1501 */ 1502 __copy_from_user_inatomic_nocache(offset, entry->buf, entry->len); 1503 #else 1504 memcpy_toio(offset, entry->buf, entry->len); 1505 #endif 1506 1507 /* Ensure that the data is fully copied out before setting the flags */ 1508 wmb(); 1509 1510 ntb_tx_copy_callback(entry); 1511 } 1512 1513 static void ntb_async_tx(struct ntb_transport_qp *qp, 1514 struct ntb_queue_entry *entry) 1515 { 1516 struct ntb_payload_header __iomem *hdr; 1517 struct dma_async_tx_descriptor *txd; 1518 struct dma_chan *chan = qp->tx_dma_chan; 1519 struct dma_device *device; 1520 size_t dest_off, buff_off; 1521 struct dmaengine_unmap_data *unmap; 1522 dma_addr_t dest; 1523 dma_cookie_t cookie; 1524 void __iomem *offset; 1525 size_t len = entry->len; 1526 void *buf = entry->buf; 1527 int retries = 0; 1528 1529 offset = qp->tx_mw + qp->tx_max_frame * qp->tx_index; 1530 hdr = offset + qp->tx_max_frame - sizeof(struct ntb_payload_header); 1531 entry->tx_hdr = hdr; 1532 1533 iowrite32(entry->len, &hdr->len); 1534 iowrite32((u32)qp->tx_pkts, &hdr->ver); 1535 1536 if (!chan) 1537 goto err; 1538 1539 if (len < copy_bytes) 1540 goto err; 1541 1542 device = chan->device; 1543 dest = qp->tx_mw_phys + qp->tx_max_frame * qp->tx_index; 1544 buff_off = (size_t)buf & ~PAGE_MASK; 1545 dest_off = (size_t)dest & ~PAGE_MASK; 1546 1547 if (!is_dma_copy_aligned(device, buff_off, dest_off, len)) 1548 goto err; 1549 1550 unmap = dmaengine_get_unmap_data(device->dev, 1, GFP_NOWAIT); 1551 if (!unmap) 1552 goto err; 1553 1554 unmap->len = len; 1555 unmap->addr[0] = dma_map_page(device->dev, virt_to_page(buf), 1556 buff_off, len, DMA_TO_DEVICE); 1557 if (dma_mapping_error(device->dev, unmap->addr[0])) 1558 goto err_get_unmap; 1559 1560 unmap->to_cnt = 1; 1561 1562 for (retries = 0; retries < DMA_RETRIES; retries++) { 1563 txd = device->device_prep_dma_memcpy(chan, dest, unmap->addr[0], 1564 len, DMA_PREP_INTERRUPT); 1565 if (txd) 1566 break; 1567 1568 set_current_state(TASK_INTERRUPTIBLE); 1569 schedule_timeout(DMA_OUT_RESOURCE_TO); 1570 } 1571 1572 if (!txd) { 1573 qp->dma_tx_prep_err++; 1574 goto err_get_unmap; 1575 } 1576 1577 txd->callback = ntb_tx_copy_callback; 1578 txd->callback_param = entry; 1579 dma_set_unmap(txd, unmap); 1580 1581 cookie = dmaengine_submit(txd); 1582 if (dma_submit_error(cookie)) 1583 goto err_set_unmap; 1584 1585 dmaengine_unmap_put(unmap); 1586 1587 dma_async_issue_pending(chan); 1588 qp->tx_async++; 1589 1590 return; 1591 err_set_unmap: 1592 dmaengine_unmap_put(unmap); 1593 err_get_unmap: 1594 dmaengine_unmap_put(unmap); 1595 err: 1596 ntb_memcpy_tx(entry, offset); 1597 qp->tx_memcpy++; 1598 } 1599 1600 static int ntb_process_tx(struct ntb_transport_qp *qp, 1601 struct ntb_queue_entry *entry) 1602 { 1603 if (qp->tx_index == qp->remote_rx_info->entry) { 1604 qp->tx_ring_full++; 1605 return -EAGAIN; 1606 } 1607 1608 if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) { 1609 if (qp->tx_handler) 1610 qp->tx_handler(qp, qp->cb_data, NULL, -EIO); 1611 1612 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1613 &qp->tx_free_q); 1614 return 0; 1615 } 1616 1617 ntb_async_tx(qp, entry); 1618 1619 qp->tx_index++; 1620 qp->tx_index %= qp->tx_max_entry; 1621 1622 qp->tx_pkts++; 1623 1624 return 0; 1625 } 1626 1627 static void ntb_send_link_down(struct ntb_transport_qp *qp) 1628 { 1629 struct pci_dev *pdev = qp->ndev->pdev; 1630 struct ntb_queue_entry *entry; 1631 int i, rc; 1632 1633 if (!qp->link_is_up) 1634 return; 1635 1636 dev_info(&pdev->dev, "qp %d: Send Link Down\n", qp->qp_num); 1637 1638 for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) { 1639 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); 1640 if (entry) 1641 break; 1642 msleep(100); 1643 } 1644 1645 if (!entry) 1646 return; 1647 1648 entry->cb_data = NULL; 1649 entry->buf = NULL; 1650 entry->len = 0; 1651 entry->flags = LINK_DOWN_FLAG; 1652 1653 rc = ntb_process_tx(qp, entry); 1654 if (rc) 1655 dev_err(&pdev->dev, "ntb: QP%d unable to send linkdown msg\n", 1656 qp->qp_num); 1657 1658 ntb_qp_link_down_reset(qp); 1659 } 1660 1661 static bool ntb_dma_filter_fn(struct dma_chan *chan, void *node) 1662 { 1663 return dev_to_node(&chan->dev->device) == (int)(unsigned long)node; 1664 } 1665 1666 /** 1667 * ntb_transport_create_queue - Create a new NTB transport layer queue 1668 * @rx_handler: receive callback function 1669 * @tx_handler: transmit callback function 1670 * @event_handler: event callback function 1671 * 1672 * Create a new NTB transport layer queue and provide the queue with a callback 1673 * routine for both transmit and receive. The receive callback routine will be 1674 * used to pass up data when the transport has received it on the queue. The 1675 * transmit callback routine will be called when the transport has completed the 1676 * transmission of the data on the queue and the data is ready to be freed. 1677 * 1678 * RETURNS: pointer to newly created ntb_queue, NULL on error. 1679 */ 1680 struct ntb_transport_qp * 1681 ntb_transport_create_queue(void *data, struct device *client_dev, 1682 const struct ntb_queue_handlers *handlers) 1683 { 1684 struct ntb_dev *ndev; 1685 struct pci_dev *pdev; 1686 struct ntb_transport_ctx *nt; 1687 struct ntb_queue_entry *entry; 1688 struct ntb_transport_qp *qp; 1689 u64 qp_bit; 1690 unsigned int free_queue; 1691 dma_cap_mask_t dma_mask; 1692 int node; 1693 int i; 1694 1695 ndev = dev_ntb(client_dev->parent); 1696 pdev = ndev->pdev; 1697 nt = ndev->ctx; 1698 1699 node = dev_to_node(&ndev->dev); 1700 1701 free_queue = ffs(nt->qp_bitmap); 1702 if (!free_queue) 1703 goto err; 1704 1705 /* decrement free_queue to make it zero based */ 1706 free_queue--; 1707 1708 qp = &nt->qp_vec[free_queue]; 1709 qp_bit = BIT_ULL(qp->qp_num); 1710 1711 nt->qp_bitmap_free &= ~qp_bit; 1712 1713 qp->cb_data = data; 1714 qp->rx_handler = handlers->rx_handler; 1715 qp->tx_handler = handlers->tx_handler; 1716 qp->event_handler = handlers->event_handler; 1717 1718 dma_cap_zero(dma_mask); 1719 dma_cap_set(DMA_MEMCPY, dma_mask); 1720 1721 if (use_dma) { 1722 qp->tx_dma_chan = 1723 dma_request_channel(dma_mask, ntb_dma_filter_fn, 1724 (void *)(unsigned long)node); 1725 if (!qp->tx_dma_chan) 1726 dev_info(&pdev->dev, "Unable to allocate TX DMA channel\n"); 1727 1728 qp->rx_dma_chan = 1729 dma_request_channel(dma_mask, ntb_dma_filter_fn, 1730 (void *)(unsigned long)node); 1731 if (!qp->rx_dma_chan) 1732 dev_info(&pdev->dev, "Unable to allocate RX DMA channel\n"); 1733 } else { 1734 qp->tx_dma_chan = NULL; 1735 qp->rx_dma_chan = NULL; 1736 } 1737 1738 dev_dbg(&pdev->dev, "Using %s memcpy for TX\n", 1739 qp->tx_dma_chan ? "DMA" : "CPU"); 1740 1741 dev_dbg(&pdev->dev, "Using %s memcpy for RX\n", 1742 qp->rx_dma_chan ? "DMA" : "CPU"); 1743 1744 for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) { 1745 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 1746 if (!entry) 1747 goto err1; 1748 1749 entry->qp = qp; 1750 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, 1751 &qp->rx_free_q); 1752 } 1753 qp->rx_alloc_entry = NTB_QP_DEF_NUM_ENTRIES; 1754 1755 for (i = 0; i < qp->tx_max_entry; i++) { 1756 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 1757 if (!entry) 1758 goto err2; 1759 1760 entry->qp = qp; 1761 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1762 &qp->tx_free_q); 1763 } 1764 1765 ntb_db_clear(qp->ndev, qp_bit); 1766 ntb_db_clear_mask(qp->ndev, qp_bit); 1767 1768 dev_info(&pdev->dev, "NTB Transport QP %d created\n", qp->qp_num); 1769 1770 return qp; 1771 1772 err2: 1773 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) 1774 kfree(entry); 1775 err1: 1776 qp->rx_alloc_entry = 0; 1777 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q))) 1778 kfree(entry); 1779 if (qp->tx_dma_chan) 1780 dma_release_channel(qp->tx_dma_chan); 1781 if (qp->rx_dma_chan) 1782 dma_release_channel(qp->rx_dma_chan); 1783 nt->qp_bitmap_free |= qp_bit; 1784 err: 1785 return NULL; 1786 } 1787 EXPORT_SYMBOL_GPL(ntb_transport_create_queue); 1788 1789 /** 1790 * ntb_transport_free_queue - Frees NTB transport queue 1791 * @qp: NTB queue to be freed 1792 * 1793 * Frees NTB transport queue 1794 */ 1795 void ntb_transport_free_queue(struct ntb_transport_qp *qp) 1796 { 1797 struct pci_dev *pdev; 1798 struct ntb_queue_entry *entry; 1799 u64 qp_bit; 1800 1801 if (!qp) 1802 return; 1803 1804 pdev = qp->ndev->pdev; 1805 1806 qp->active = false; 1807 1808 if (qp->tx_dma_chan) { 1809 struct dma_chan *chan = qp->tx_dma_chan; 1810 /* Putting the dma_chan to NULL will force any new traffic to be 1811 * processed by the CPU instead of the DAM engine 1812 */ 1813 qp->tx_dma_chan = NULL; 1814 1815 /* Try to be nice and wait for any queued DMA engine 1816 * transactions to process before smashing it with a rock 1817 */ 1818 dma_sync_wait(chan, qp->last_cookie); 1819 dmaengine_terminate_all(chan); 1820 dma_release_channel(chan); 1821 } 1822 1823 if (qp->rx_dma_chan) { 1824 struct dma_chan *chan = qp->rx_dma_chan; 1825 /* Putting the dma_chan to NULL will force any new traffic to be 1826 * processed by the CPU instead of the DAM engine 1827 */ 1828 qp->rx_dma_chan = NULL; 1829 1830 /* Try to be nice and wait for any queued DMA engine 1831 * transactions to process before smashing it with a rock 1832 */ 1833 dma_sync_wait(chan, qp->last_cookie); 1834 dmaengine_terminate_all(chan); 1835 dma_release_channel(chan); 1836 } 1837 1838 qp_bit = BIT_ULL(qp->qp_num); 1839 1840 ntb_db_set_mask(qp->ndev, qp_bit); 1841 tasklet_kill(&qp->rxc_db_work); 1842 1843 cancel_delayed_work_sync(&qp->link_work); 1844 1845 qp->cb_data = NULL; 1846 qp->rx_handler = NULL; 1847 qp->tx_handler = NULL; 1848 qp->event_handler = NULL; 1849 1850 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q))) 1851 kfree(entry); 1852 1853 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q))) { 1854 dev_warn(&pdev->dev, "Freeing item from non-empty rx_pend_q\n"); 1855 kfree(entry); 1856 } 1857 1858 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_post_q))) { 1859 dev_warn(&pdev->dev, "Freeing item from non-empty rx_post_q\n"); 1860 kfree(entry); 1861 } 1862 1863 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) 1864 kfree(entry); 1865 1866 qp->transport->qp_bitmap_free |= qp_bit; 1867 1868 dev_info(&pdev->dev, "NTB Transport QP %d freed\n", qp->qp_num); 1869 } 1870 EXPORT_SYMBOL_GPL(ntb_transport_free_queue); 1871 1872 /** 1873 * ntb_transport_rx_remove - Dequeues enqueued rx packet 1874 * @qp: NTB queue to be freed 1875 * @len: pointer to variable to write enqueued buffers length 1876 * 1877 * Dequeues unused buffers from receive queue. Should only be used during 1878 * shutdown of qp. 1879 * 1880 * RETURNS: NULL error value on error, or void* for success. 1881 */ 1882 void *ntb_transport_rx_remove(struct ntb_transport_qp *qp, unsigned int *len) 1883 { 1884 struct ntb_queue_entry *entry; 1885 void *buf; 1886 1887 if (!qp || qp->client_ready) 1888 return NULL; 1889 1890 entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q); 1891 if (!entry) 1892 return NULL; 1893 1894 buf = entry->cb_data; 1895 *len = entry->len; 1896 1897 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_free_q); 1898 1899 return buf; 1900 } 1901 EXPORT_SYMBOL_GPL(ntb_transport_rx_remove); 1902 1903 /** 1904 * ntb_transport_rx_enqueue - Enqueue a new NTB queue entry 1905 * @qp: NTB transport layer queue the entry is to be enqueued on 1906 * @cb: per buffer pointer for callback function to use 1907 * @data: pointer to data buffer that incoming packets will be copied into 1908 * @len: length of the data buffer 1909 * 1910 * Enqueue a new receive buffer onto the transport queue into which a NTB 1911 * payload can be received into. 1912 * 1913 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 1914 */ 1915 int ntb_transport_rx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, 1916 unsigned int len) 1917 { 1918 struct ntb_queue_entry *entry; 1919 1920 if (!qp) 1921 return -EINVAL; 1922 1923 entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q); 1924 if (!entry) 1925 return -ENOMEM; 1926 1927 entry->cb_data = cb; 1928 entry->buf = data; 1929 entry->len = len; 1930 entry->flags = 0; 1931 1932 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_pend_q); 1933 1934 if (qp->active) 1935 tasklet_schedule(&qp->rxc_db_work); 1936 1937 return 0; 1938 } 1939 EXPORT_SYMBOL_GPL(ntb_transport_rx_enqueue); 1940 1941 /** 1942 * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry 1943 * @qp: NTB transport layer queue the entry is to be enqueued on 1944 * @cb: per buffer pointer for callback function to use 1945 * @data: pointer to data buffer that will be sent 1946 * @len: length of the data buffer 1947 * 1948 * Enqueue a new transmit buffer onto the transport queue from which a NTB 1949 * payload will be transmitted. This assumes that a lock is being held to 1950 * serialize access to the qp. 1951 * 1952 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 1953 */ 1954 int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, 1955 unsigned int len) 1956 { 1957 struct ntb_queue_entry *entry; 1958 int rc; 1959 1960 if (!qp || !qp->link_is_up || !len) 1961 return -EINVAL; 1962 1963 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); 1964 if (!entry) { 1965 qp->tx_err_no_buf++; 1966 return -EBUSY; 1967 } 1968 1969 entry->cb_data = cb; 1970 entry->buf = data; 1971 entry->len = len; 1972 entry->flags = 0; 1973 1974 rc = ntb_process_tx(qp, entry); 1975 if (rc) 1976 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1977 &qp->tx_free_q); 1978 1979 return rc; 1980 } 1981 EXPORT_SYMBOL_GPL(ntb_transport_tx_enqueue); 1982 1983 /** 1984 * ntb_transport_link_up - Notify NTB transport of client readiness to use queue 1985 * @qp: NTB transport layer queue to be enabled 1986 * 1987 * Notify NTB transport layer of client readiness to use queue 1988 */ 1989 void ntb_transport_link_up(struct ntb_transport_qp *qp) 1990 { 1991 if (!qp) 1992 return; 1993 1994 qp->client_ready = true; 1995 1996 if (qp->transport->link_is_up) 1997 schedule_delayed_work(&qp->link_work, 0); 1998 } 1999 EXPORT_SYMBOL_GPL(ntb_transport_link_up); 2000 2001 /** 2002 * ntb_transport_link_down - Notify NTB transport to no longer enqueue data 2003 * @qp: NTB transport layer queue to be disabled 2004 * 2005 * Notify NTB transport layer of client's desire to no longer receive data on 2006 * transport queue specified. It is the client's responsibility to ensure all 2007 * entries on queue are purged or otherwise handled appropriately. 2008 */ 2009 void ntb_transport_link_down(struct ntb_transport_qp *qp) 2010 { 2011 int val; 2012 2013 if (!qp) 2014 return; 2015 2016 qp->client_ready = false; 2017 2018 val = ntb_spad_read(qp->ndev, QP_LINKS); 2019 2020 ntb_peer_spad_write(qp->ndev, QP_LINKS, 2021 val & ~BIT(qp->qp_num)); 2022 2023 if (qp->link_is_up) 2024 ntb_send_link_down(qp); 2025 else 2026 cancel_delayed_work_sync(&qp->link_work); 2027 } 2028 EXPORT_SYMBOL_GPL(ntb_transport_link_down); 2029 2030 /** 2031 * ntb_transport_link_query - Query transport link state 2032 * @qp: NTB transport layer queue to be queried 2033 * 2034 * Query connectivity to the remote system of the NTB transport queue 2035 * 2036 * RETURNS: true for link up or false for link down 2037 */ 2038 bool ntb_transport_link_query(struct ntb_transport_qp *qp) 2039 { 2040 if (!qp) 2041 return false; 2042 2043 return qp->link_is_up; 2044 } 2045 EXPORT_SYMBOL_GPL(ntb_transport_link_query); 2046 2047 /** 2048 * ntb_transport_qp_num - Query the qp number 2049 * @qp: NTB transport layer queue to be queried 2050 * 2051 * Query qp number of the NTB transport queue 2052 * 2053 * RETURNS: a zero based number specifying the qp number 2054 */ 2055 unsigned char ntb_transport_qp_num(struct ntb_transport_qp *qp) 2056 { 2057 if (!qp) 2058 return 0; 2059 2060 return qp->qp_num; 2061 } 2062 EXPORT_SYMBOL_GPL(ntb_transport_qp_num); 2063 2064 /** 2065 * ntb_transport_max_size - Query the max payload size of a qp 2066 * @qp: NTB transport layer queue to be queried 2067 * 2068 * Query the maximum payload size permissible on the given qp 2069 * 2070 * RETURNS: the max payload size of a qp 2071 */ 2072 unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp) 2073 { 2074 unsigned int max_size; 2075 unsigned int copy_align; 2076 struct dma_chan *rx_chan, *tx_chan; 2077 2078 if (!qp) 2079 return 0; 2080 2081 rx_chan = qp->rx_dma_chan; 2082 tx_chan = qp->tx_dma_chan; 2083 2084 copy_align = max(rx_chan ? rx_chan->device->copy_align : 0, 2085 tx_chan ? tx_chan->device->copy_align : 0); 2086 2087 /* If DMA engine usage is possible, try to find the max size for that */ 2088 max_size = qp->tx_max_frame - sizeof(struct ntb_payload_header); 2089 max_size = round_down(max_size, 1 << copy_align); 2090 2091 return max_size; 2092 } 2093 EXPORT_SYMBOL_GPL(ntb_transport_max_size); 2094 2095 unsigned int ntb_transport_tx_free_entry(struct ntb_transport_qp *qp) 2096 { 2097 unsigned int head = qp->tx_index; 2098 unsigned int tail = qp->remote_rx_info->entry; 2099 2100 return tail > head ? tail - head : qp->tx_max_entry + tail - head; 2101 } 2102 EXPORT_SYMBOL_GPL(ntb_transport_tx_free_entry); 2103 2104 static void ntb_transport_doorbell_callback(void *data, int vector) 2105 { 2106 struct ntb_transport_ctx *nt = data; 2107 struct ntb_transport_qp *qp; 2108 u64 db_bits; 2109 unsigned int qp_num; 2110 2111 db_bits = (nt->qp_bitmap & ~nt->qp_bitmap_free & 2112 ntb_db_vector_mask(nt->ndev, vector)); 2113 2114 while (db_bits) { 2115 qp_num = __ffs(db_bits); 2116 qp = &nt->qp_vec[qp_num]; 2117 2118 if (qp->active) 2119 tasklet_schedule(&qp->rxc_db_work); 2120 2121 db_bits &= ~BIT_ULL(qp_num); 2122 } 2123 } 2124 2125 static const struct ntb_ctx_ops ntb_transport_ops = { 2126 .link_event = ntb_transport_event_callback, 2127 .db_event = ntb_transport_doorbell_callback, 2128 }; 2129 2130 static struct ntb_client ntb_transport_client = { 2131 .ops = { 2132 .probe = ntb_transport_probe, 2133 .remove = ntb_transport_free, 2134 }, 2135 }; 2136 2137 static int __init ntb_transport_init(void) 2138 { 2139 int rc; 2140 2141 pr_info("%s, version %s\n", NTB_TRANSPORT_DESC, NTB_TRANSPORT_VER); 2142 2143 if (debugfs_initialized()) 2144 nt_debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL); 2145 2146 rc = bus_register(&ntb_transport_bus); 2147 if (rc) 2148 goto err_bus; 2149 2150 rc = ntb_register_client(&ntb_transport_client); 2151 if (rc) 2152 goto err_client; 2153 2154 return 0; 2155 2156 err_client: 2157 bus_unregister(&ntb_transport_bus); 2158 err_bus: 2159 debugfs_remove_recursive(nt_debugfs_dir); 2160 return rc; 2161 } 2162 module_init(ntb_transport_init); 2163 2164 static void __exit ntb_transport_exit(void) 2165 { 2166 debugfs_remove_recursive(nt_debugfs_dir); 2167 2168 ntb_unregister_client(&ntb_transport_client); 2169 bus_unregister(&ntb_transport_bus); 2170 } 2171 module_exit(ntb_transport_exit); 2172