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 int retries; 106 int errors; 107 unsigned int tx_index; 108 unsigned int rx_index; 109 110 struct ntb_transport_qp *qp; 111 union { 112 struct ntb_payload_header __iomem *tx_hdr; 113 struct ntb_payload_header *rx_hdr; 114 }; 115 }; 116 117 struct ntb_rx_info { 118 unsigned int entry; 119 }; 120 121 struct ntb_transport_qp { 122 struct ntb_transport_ctx *transport; 123 struct ntb_dev *ndev; 124 void *cb_data; 125 struct dma_chan *tx_dma_chan; 126 struct dma_chan *rx_dma_chan; 127 128 bool client_ready; 129 bool link_is_up; 130 bool active; 131 132 u8 qp_num; /* Only 64 QP's are allowed. 0-63 */ 133 u64 qp_bit; 134 135 struct ntb_rx_info __iomem *rx_info; 136 struct ntb_rx_info *remote_rx_info; 137 138 void (*tx_handler)(struct ntb_transport_qp *qp, void *qp_data, 139 void *data, int len); 140 struct list_head tx_free_q; 141 spinlock_t ntb_tx_free_q_lock; 142 void __iomem *tx_mw; 143 dma_addr_t tx_mw_phys; 144 unsigned int tx_index; 145 unsigned int tx_max_entry; 146 unsigned int tx_max_frame; 147 148 void (*rx_handler)(struct ntb_transport_qp *qp, void *qp_data, 149 void *data, int len); 150 struct list_head rx_post_q; 151 struct list_head rx_pend_q; 152 struct list_head rx_free_q; 153 /* ntb_rx_q_lock: synchronize access to rx_XXXX_q */ 154 spinlock_t ntb_rx_q_lock; 155 void *rx_buff; 156 unsigned int rx_index; 157 unsigned int rx_max_entry; 158 unsigned int rx_max_frame; 159 unsigned int rx_alloc_entry; 160 dma_cookie_t last_cookie; 161 struct tasklet_struct rxc_db_work; 162 163 void (*event_handler)(void *data, int status); 164 struct delayed_work link_work; 165 struct work_struct link_cleanup; 166 167 struct dentry *debugfs_dir; 168 struct dentry *debugfs_stats; 169 170 /* Stats */ 171 u64 rx_bytes; 172 u64 rx_pkts; 173 u64 rx_ring_empty; 174 u64 rx_err_no_buf; 175 u64 rx_err_oflow; 176 u64 rx_err_ver; 177 u64 rx_memcpy; 178 u64 rx_async; 179 u64 dma_rx_prep_err; 180 u64 tx_bytes; 181 u64 tx_pkts; 182 u64 tx_ring_full; 183 u64 tx_err_no_buf; 184 u64 tx_memcpy; 185 u64 tx_async; 186 u64 dma_tx_prep_err; 187 }; 188 189 struct ntb_transport_mw { 190 phys_addr_t phys_addr; 191 resource_size_t phys_size; 192 resource_size_t xlat_align; 193 resource_size_t xlat_align_size; 194 void __iomem *vbase; 195 size_t xlat_size; 196 size_t buff_size; 197 void *virt_addr; 198 dma_addr_t dma_addr; 199 }; 200 201 struct ntb_transport_client_dev { 202 struct list_head entry; 203 struct ntb_transport_ctx *nt; 204 struct device dev; 205 }; 206 207 struct ntb_transport_ctx { 208 struct list_head entry; 209 struct list_head client_devs; 210 211 struct ntb_dev *ndev; 212 213 struct ntb_transport_mw *mw_vec; 214 struct ntb_transport_qp *qp_vec; 215 unsigned int mw_count; 216 unsigned int qp_count; 217 u64 qp_bitmap; 218 u64 qp_bitmap_free; 219 220 bool link_is_up; 221 struct delayed_work link_work; 222 struct work_struct link_cleanup; 223 224 struct dentry *debugfs_node_dir; 225 }; 226 227 enum { 228 DESC_DONE_FLAG = BIT(0), 229 LINK_DOWN_FLAG = BIT(1), 230 }; 231 232 struct ntb_payload_header { 233 unsigned int ver; 234 unsigned int len; 235 unsigned int flags; 236 }; 237 238 enum { 239 VERSION = 0, 240 QP_LINKS, 241 NUM_QPS, 242 NUM_MWS, 243 MW0_SZ_HIGH, 244 MW0_SZ_LOW, 245 MW1_SZ_HIGH, 246 MW1_SZ_LOW, 247 MAX_SPAD, 248 }; 249 250 #define dev_client_dev(__dev) \ 251 container_of((__dev), struct ntb_transport_client_dev, dev) 252 253 #define drv_client(__drv) \ 254 container_of((__drv), struct ntb_transport_client, driver) 255 256 #define QP_TO_MW(nt, qp) ((qp) % nt->mw_count) 257 #define NTB_QP_DEF_NUM_ENTRIES 100 258 #define NTB_LINK_DOWN_TIMEOUT 10 259 #define DMA_RETRIES 20 260 #define DMA_OUT_RESOURCE_TO 50 261 262 static void ntb_transport_rxc_db(unsigned long data); 263 static const struct ntb_ctx_ops ntb_transport_ops; 264 static struct ntb_client ntb_transport_client; 265 static int ntb_async_tx_submit(struct ntb_transport_qp *qp, 266 struct ntb_queue_entry *entry); 267 static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset); 268 static int ntb_async_rx_submit(struct ntb_queue_entry *entry, void *offset); 269 static void ntb_memcpy_rx(struct ntb_queue_entry *entry, void *offset); 270 271 272 static int ntb_transport_bus_match(struct device *dev, 273 struct device_driver *drv) 274 { 275 return !strncmp(dev_name(dev), drv->name, strlen(drv->name)); 276 } 277 278 static int ntb_transport_bus_probe(struct device *dev) 279 { 280 const struct ntb_transport_client *client; 281 int rc = -EINVAL; 282 283 get_device(dev); 284 285 client = drv_client(dev->driver); 286 rc = client->probe(dev); 287 if (rc) 288 put_device(dev); 289 290 return rc; 291 } 292 293 static int ntb_transport_bus_remove(struct device *dev) 294 { 295 const struct ntb_transport_client *client; 296 297 client = drv_client(dev->driver); 298 client->remove(dev); 299 300 put_device(dev); 301 302 return 0; 303 } 304 305 static struct bus_type ntb_transport_bus = { 306 .name = "ntb_transport", 307 .match = ntb_transport_bus_match, 308 .probe = ntb_transport_bus_probe, 309 .remove = ntb_transport_bus_remove, 310 }; 311 312 static LIST_HEAD(ntb_transport_list); 313 314 static int ntb_bus_init(struct ntb_transport_ctx *nt) 315 { 316 list_add_tail(&nt->entry, &ntb_transport_list); 317 return 0; 318 } 319 320 static void ntb_bus_remove(struct ntb_transport_ctx *nt) 321 { 322 struct ntb_transport_client_dev *client_dev, *cd; 323 324 list_for_each_entry_safe(client_dev, cd, &nt->client_devs, entry) { 325 dev_err(client_dev->dev.parent, "%s still attached to bus, removing\n", 326 dev_name(&client_dev->dev)); 327 list_del(&client_dev->entry); 328 device_unregister(&client_dev->dev); 329 } 330 331 list_del(&nt->entry); 332 } 333 334 static void ntb_transport_client_release(struct device *dev) 335 { 336 struct ntb_transport_client_dev *client_dev; 337 338 client_dev = dev_client_dev(dev); 339 kfree(client_dev); 340 } 341 342 /** 343 * ntb_transport_unregister_client_dev - Unregister NTB client device 344 * @device_name: Name of NTB client device 345 * 346 * Unregister an NTB client device with the NTB transport layer 347 */ 348 void ntb_transport_unregister_client_dev(char *device_name) 349 { 350 struct ntb_transport_client_dev *client, *cd; 351 struct ntb_transport_ctx *nt; 352 353 list_for_each_entry(nt, &ntb_transport_list, entry) 354 list_for_each_entry_safe(client, cd, &nt->client_devs, entry) 355 if (!strncmp(dev_name(&client->dev), device_name, 356 strlen(device_name))) { 357 list_del(&client->entry); 358 device_unregister(&client->dev); 359 } 360 } 361 EXPORT_SYMBOL_GPL(ntb_transport_unregister_client_dev); 362 363 /** 364 * ntb_transport_register_client_dev - Register NTB client device 365 * @device_name: Name of NTB client device 366 * 367 * Register an NTB client device with the NTB transport layer 368 */ 369 int ntb_transport_register_client_dev(char *device_name) 370 { 371 struct ntb_transport_client_dev *client_dev; 372 struct ntb_transport_ctx *nt; 373 int node; 374 int rc, i = 0; 375 376 if (list_empty(&ntb_transport_list)) 377 return -ENODEV; 378 379 list_for_each_entry(nt, &ntb_transport_list, entry) { 380 struct device *dev; 381 382 node = dev_to_node(&nt->ndev->dev); 383 384 client_dev = kzalloc_node(sizeof(*client_dev), 385 GFP_KERNEL, node); 386 if (!client_dev) { 387 rc = -ENOMEM; 388 goto err; 389 } 390 391 dev = &client_dev->dev; 392 393 /* setup and register client devices */ 394 dev_set_name(dev, "%s%d", device_name, i); 395 dev->bus = &ntb_transport_bus; 396 dev->release = ntb_transport_client_release; 397 dev->parent = &nt->ndev->dev; 398 399 rc = device_register(dev); 400 if (rc) { 401 kfree(client_dev); 402 goto err; 403 } 404 405 list_add_tail(&client_dev->entry, &nt->client_devs); 406 i++; 407 } 408 409 return 0; 410 411 err: 412 ntb_transport_unregister_client_dev(device_name); 413 414 return rc; 415 } 416 EXPORT_SYMBOL_GPL(ntb_transport_register_client_dev); 417 418 /** 419 * ntb_transport_register_client - Register NTB client driver 420 * @drv: NTB client driver to be registered 421 * 422 * Register an NTB client driver with the NTB transport layer 423 * 424 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 425 */ 426 int ntb_transport_register_client(struct ntb_transport_client *drv) 427 { 428 drv->driver.bus = &ntb_transport_bus; 429 430 if (list_empty(&ntb_transport_list)) 431 return -ENODEV; 432 433 return driver_register(&drv->driver); 434 } 435 EXPORT_SYMBOL_GPL(ntb_transport_register_client); 436 437 /** 438 * ntb_transport_unregister_client - Unregister NTB client driver 439 * @drv: NTB client driver to be unregistered 440 * 441 * Unregister an NTB client driver with the NTB transport layer 442 * 443 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 444 */ 445 void ntb_transport_unregister_client(struct ntb_transport_client *drv) 446 { 447 driver_unregister(&drv->driver); 448 } 449 EXPORT_SYMBOL_GPL(ntb_transport_unregister_client); 450 451 static ssize_t debugfs_read(struct file *filp, char __user *ubuf, size_t count, 452 loff_t *offp) 453 { 454 struct ntb_transport_qp *qp; 455 char *buf; 456 ssize_t ret, out_offset, out_count; 457 458 qp = filp->private_data; 459 460 if (!qp || !qp->link_is_up) 461 return 0; 462 463 out_count = 1000; 464 465 buf = kmalloc(out_count, GFP_KERNEL); 466 if (!buf) 467 return -ENOMEM; 468 469 out_offset = 0; 470 out_offset += snprintf(buf + out_offset, out_count - out_offset, 471 "\nNTB QP stats:\n\n"); 472 out_offset += snprintf(buf + out_offset, out_count - out_offset, 473 "rx_bytes - \t%llu\n", qp->rx_bytes); 474 out_offset += snprintf(buf + out_offset, out_count - out_offset, 475 "rx_pkts - \t%llu\n", qp->rx_pkts); 476 out_offset += snprintf(buf + out_offset, out_count - out_offset, 477 "rx_memcpy - \t%llu\n", qp->rx_memcpy); 478 out_offset += snprintf(buf + out_offset, out_count - out_offset, 479 "rx_async - \t%llu\n", qp->rx_async); 480 out_offset += snprintf(buf + out_offset, out_count - out_offset, 481 "rx_ring_empty - %llu\n", qp->rx_ring_empty); 482 out_offset += snprintf(buf + out_offset, out_count - out_offset, 483 "rx_err_no_buf - %llu\n", qp->rx_err_no_buf); 484 out_offset += snprintf(buf + out_offset, out_count - out_offset, 485 "rx_err_oflow - \t%llu\n", qp->rx_err_oflow); 486 out_offset += snprintf(buf + out_offset, out_count - out_offset, 487 "rx_err_ver - \t%llu\n", qp->rx_err_ver); 488 out_offset += snprintf(buf + out_offset, out_count - out_offset, 489 "rx_buff - \t0x%p\n", qp->rx_buff); 490 out_offset += snprintf(buf + out_offset, out_count - out_offset, 491 "rx_index - \t%u\n", qp->rx_index); 492 out_offset += snprintf(buf + out_offset, out_count - out_offset, 493 "rx_max_entry - \t%u\n", qp->rx_max_entry); 494 out_offset += snprintf(buf + out_offset, out_count - out_offset, 495 "rx_alloc_entry - \t%u\n\n", qp->rx_alloc_entry); 496 497 out_offset += snprintf(buf + out_offset, out_count - out_offset, 498 "tx_bytes - \t%llu\n", qp->tx_bytes); 499 out_offset += snprintf(buf + out_offset, out_count - out_offset, 500 "tx_pkts - \t%llu\n", qp->tx_pkts); 501 out_offset += snprintf(buf + out_offset, out_count - out_offset, 502 "tx_memcpy - \t%llu\n", qp->tx_memcpy); 503 out_offset += snprintf(buf + out_offset, out_count - out_offset, 504 "tx_async - \t%llu\n", qp->tx_async); 505 out_offset += snprintf(buf + out_offset, out_count - out_offset, 506 "tx_ring_full - \t%llu\n", qp->tx_ring_full); 507 out_offset += snprintf(buf + out_offset, out_count - out_offset, 508 "tx_err_no_buf - %llu\n", qp->tx_err_no_buf); 509 out_offset += snprintf(buf + out_offset, out_count - out_offset, 510 "tx_mw - \t0x%p\n", qp->tx_mw); 511 out_offset += snprintf(buf + out_offset, out_count - out_offset, 512 "tx_index (H) - \t%u\n", qp->tx_index); 513 out_offset += snprintf(buf + out_offset, out_count - out_offset, 514 "RRI (T) - \t%u\n", 515 qp->remote_rx_info->entry); 516 out_offset += snprintf(buf + out_offset, out_count - out_offset, 517 "tx_max_entry - \t%u\n", qp->tx_max_entry); 518 out_offset += snprintf(buf + out_offset, out_count - out_offset, 519 "free tx - \t%u\n", 520 ntb_transport_tx_free_entry(qp)); 521 out_offset += snprintf(buf + out_offset, out_count - out_offset, 522 "DMA tx prep err - \t%llu\n", 523 qp->dma_tx_prep_err); 524 out_offset += snprintf(buf + out_offset, out_count - out_offset, 525 "DMA rx prep err - \t%llu\n", 526 qp->dma_rx_prep_err); 527 528 out_offset += snprintf(buf + out_offset, out_count - out_offset, 529 "\n"); 530 out_offset += snprintf(buf + out_offset, out_count - out_offset, 531 "Using TX DMA - \t%s\n", 532 qp->tx_dma_chan ? "Yes" : "No"); 533 out_offset += snprintf(buf + out_offset, out_count - out_offset, 534 "Using RX DMA - \t%s\n", 535 qp->rx_dma_chan ? "Yes" : "No"); 536 out_offset += snprintf(buf + out_offset, out_count - out_offset, 537 "QP Link - \t%s\n", 538 qp->link_is_up ? "Up" : "Down"); 539 out_offset += snprintf(buf + out_offset, out_count - out_offset, 540 "\n"); 541 542 if (out_offset > out_count) 543 out_offset = out_count; 544 545 ret = simple_read_from_buffer(ubuf, count, offp, buf, out_offset); 546 kfree(buf); 547 return ret; 548 } 549 550 static const struct file_operations ntb_qp_debugfs_stats = { 551 .owner = THIS_MODULE, 552 .open = simple_open, 553 .read = debugfs_read, 554 }; 555 556 static void ntb_list_add(spinlock_t *lock, struct list_head *entry, 557 struct list_head *list) 558 { 559 unsigned long flags; 560 561 spin_lock_irqsave(lock, flags); 562 list_add_tail(entry, list); 563 spin_unlock_irqrestore(lock, flags); 564 } 565 566 static struct ntb_queue_entry *ntb_list_rm(spinlock_t *lock, 567 struct list_head *list) 568 { 569 struct ntb_queue_entry *entry; 570 unsigned long flags; 571 572 spin_lock_irqsave(lock, flags); 573 if (list_empty(list)) { 574 entry = NULL; 575 goto out; 576 } 577 entry = list_first_entry(list, struct ntb_queue_entry, entry); 578 list_del(&entry->entry); 579 580 out: 581 spin_unlock_irqrestore(lock, flags); 582 583 return entry; 584 } 585 586 static struct ntb_queue_entry *ntb_list_mv(spinlock_t *lock, 587 struct list_head *list, 588 struct list_head *to_list) 589 { 590 struct ntb_queue_entry *entry; 591 unsigned long flags; 592 593 spin_lock_irqsave(lock, flags); 594 595 if (list_empty(list)) { 596 entry = NULL; 597 } else { 598 entry = list_first_entry(list, struct ntb_queue_entry, entry); 599 list_move_tail(&entry->entry, to_list); 600 } 601 602 spin_unlock_irqrestore(lock, flags); 603 604 return entry; 605 } 606 607 static int ntb_transport_setup_qp_mw(struct ntb_transport_ctx *nt, 608 unsigned int qp_num) 609 { 610 struct ntb_transport_qp *qp = &nt->qp_vec[qp_num]; 611 struct ntb_transport_mw *mw; 612 struct ntb_dev *ndev = nt->ndev; 613 struct ntb_queue_entry *entry; 614 unsigned int rx_size, num_qps_mw; 615 unsigned int mw_num, mw_count, qp_count; 616 unsigned int i; 617 int node; 618 619 mw_count = nt->mw_count; 620 qp_count = nt->qp_count; 621 622 mw_num = QP_TO_MW(nt, qp_num); 623 mw = &nt->mw_vec[mw_num]; 624 625 if (!mw->virt_addr) 626 return -ENOMEM; 627 628 if (qp_count % mw_count && mw_num + 1 < qp_count / mw_count) 629 num_qps_mw = qp_count / mw_count + 1; 630 else 631 num_qps_mw = qp_count / mw_count; 632 633 rx_size = (unsigned int)mw->xlat_size / num_qps_mw; 634 qp->rx_buff = mw->virt_addr + rx_size * (qp_num / mw_count); 635 rx_size -= sizeof(struct ntb_rx_info); 636 637 qp->remote_rx_info = qp->rx_buff + rx_size; 638 639 /* Due to housekeeping, there must be atleast 2 buffs */ 640 qp->rx_max_frame = min(transport_mtu, rx_size / 2); 641 qp->rx_max_entry = rx_size / qp->rx_max_frame; 642 qp->rx_index = 0; 643 644 /* 645 * Checking to see if we have more entries than the default. 646 * We should add additional entries if that is the case so we 647 * can be in sync with the transport frames. 648 */ 649 node = dev_to_node(&ndev->dev); 650 for (i = qp->rx_alloc_entry; i < qp->rx_max_entry; i++) { 651 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 652 if (!entry) 653 return -ENOMEM; 654 655 entry->qp = qp; 656 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, 657 &qp->rx_free_q); 658 qp->rx_alloc_entry++; 659 } 660 661 qp->remote_rx_info->entry = qp->rx_max_entry - 1; 662 663 /* setup the hdr offsets with 0's */ 664 for (i = 0; i < qp->rx_max_entry; i++) { 665 void *offset = (qp->rx_buff + qp->rx_max_frame * (i + 1) - 666 sizeof(struct ntb_payload_header)); 667 memset(offset, 0, sizeof(struct ntb_payload_header)); 668 } 669 670 qp->rx_pkts = 0; 671 qp->tx_pkts = 0; 672 qp->tx_index = 0; 673 674 return 0; 675 } 676 677 static void ntb_free_mw(struct ntb_transport_ctx *nt, int num_mw) 678 { 679 struct ntb_transport_mw *mw = &nt->mw_vec[num_mw]; 680 struct pci_dev *pdev = nt->ndev->pdev; 681 682 if (!mw->virt_addr) 683 return; 684 685 ntb_mw_clear_trans(nt->ndev, num_mw); 686 dma_free_coherent(&pdev->dev, mw->buff_size, 687 mw->virt_addr, mw->dma_addr); 688 mw->xlat_size = 0; 689 mw->buff_size = 0; 690 mw->virt_addr = NULL; 691 } 692 693 static int ntb_set_mw(struct ntb_transport_ctx *nt, int num_mw, 694 resource_size_t size) 695 { 696 struct ntb_transport_mw *mw = &nt->mw_vec[num_mw]; 697 struct pci_dev *pdev = nt->ndev->pdev; 698 size_t xlat_size, buff_size; 699 int rc; 700 701 if (!size) 702 return -EINVAL; 703 704 xlat_size = round_up(size, mw->xlat_align_size); 705 buff_size = round_up(size, mw->xlat_align); 706 707 /* No need to re-setup */ 708 if (mw->xlat_size == xlat_size) 709 return 0; 710 711 if (mw->buff_size) 712 ntb_free_mw(nt, num_mw); 713 714 /* Alloc memory for receiving data. Must be aligned */ 715 mw->xlat_size = xlat_size; 716 mw->buff_size = buff_size; 717 718 mw->virt_addr = dma_alloc_coherent(&pdev->dev, buff_size, 719 &mw->dma_addr, GFP_KERNEL); 720 if (!mw->virt_addr) { 721 mw->xlat_size = 0; 722 mw->buff_size = 0; 723 dev_err(&pdev->dev, "Unable to alloc MW buff of size %zu\n", 724 buff_size); 725 return -ENOMEM; 726 } 727 728 /* 729 * we must ensure that the memory address allocated is BAR size 730 * aligned in order for the XLAT register to take the value. This 731 * is a requirement of the hardware. It is recommended to setup CMA 732 * for BAR sizes equal or greater than 4MB. 733 */ 734 if (!IS_ALIGNED(mw->dma_addr, mw->xlat_align)) { 735 dev_err(&pdev->dev, "DMA memory %pad is not aligned\n", 736 &mw->dma_addr); 737 ntb_free_mw(nt, num_mw); 738 return -ENOMEM; 739 } 740 741 /* Notify HW the memory location of the receive buffer */ 742 rc = ntb_mw_set_trans(nt->ndev, num_mw, mw->dma_addr, mw->xlat_size); 743 if (rc) { 744 dev_err(&pdev->dev, "Unable to set mw%d translation", num_mw); 745 ntb_free_mw(nt, num_mw); 746 return -EIO; 747 } 748 749 return 0; 750 } 751 752 static void ntb_qp_link_down_reset(struct ntb_transport_qp *qp) 753 { 754 qp->link_is_up = false; 755 qp->active = false; 756 757 qp->tx_index = 0; 758 qp->rx_index = 0; 759 qp->rx_bytes = 0; 760 qp->rx_pkts = 0; 761 qp->rx_ring_empty = 0; 762 qp->rx_err_no_buf = 0; 763 qp->rx_err_oflow = 0; 764 qp->rx_err_ver = 0; 765 qp->rx_memcpy = 0; 766 qp->rx_async = 0; 767 qp->tx_bytes = 0; 768 qp->tx_pkts = 0; 769 qp->tx_ring_full = 0; 770 qp->tx_err_no_buf = 0; 771 qp->tx_memcpy = 0; 772 qp->tx_async = 0; 773 qp->dma_tx_prep_err = 0; 774 qp->dma_rx_prep_err = 0; 775 } 776 777 static void ntb_qp_link_cleanup(struct ntb_transport_qp *qp) 778 { 779 struct ntb_transport_ctx *nt = qp->transport; 780 struct pci_dev *pdev = nt->ndev->pdev; 781 782 dev_info(&pdev->dev, "qp %d: Link Cleanup\n", qp->qp_num); 783 784 cancel_delayed_work_sync(&qp->link_work); 785 ntb_qp_link_down_reset(qp); 786 787 if (qp->event_handler) 788 qp->event_handler(qp->cb_data, qp->link_is_up); 789 } 790 791 static void ntb_qp_link_cleanup_work(struct work_struct *work) 792 { 793 struct ntb_transport_qp *qp = container_of(work, 794 struct ntb_transport_qp, 795 link_cleanup); 796 struct ntb_transport_ctx *nt = qp->transport; 797 798 ntb_qp_link_cleanup(qp); 799 800 if (nt->link_is_up) 801 schedule_delayed_work(&qp->link_work, 802 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 803 } 804 805 static void ntb_qp_link_down(struct ntb_transport_qp *qp) 806 { 807 schedule_work(&qp->link_cleanup); 808 } 809 810 static void ntb_transport_link_cleanup(struct ntb_transport_ctx *nt) 811 { 812 struct ntb_transport_qp *qp; 813 u64 qp_bitmap_alloc; 814 int i; 815 816 qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free; 817 818 /* Pass along the info to any clients */ 819 for (i = 0; i < nt->qp_count; i++) 820 if (qp_bitmap_alloc & BIT_ULL(i)) { 821 qp = &nt->qp_vec[i]; 822 ntb_qp_link_cleanup(qp); 823 cancel_work_sync(&qp->link_cleanup); 824 cancel_delayed_work_sync(&qp->link_work); 825 } 826 827 if (!nt->link_is_up) 828 cancel_delayed_work_sync(&nt->link_work); 829 830 /* The scratchpad registers keep the values if the remote side 831 * goes down, blast them now to give them a sane value the next 832 * time they are accessed 833 */ 834 for (i = 0; i < MAX_SPAD; i++) 835 ntb_spad_write(nt->ndev, i, 0); 836 } 837 838 static void ntb_transport_link_cleanup_work(struct work_struct *work) 839 { 840 struct ntb_transport_ctx *nt = 841 container_of(work, struct ntb_transport_ctx, link_cleanup); 842 843 ntb_transport_link_cleanup(nt); 844 } 845 846 static void ntb_transport_event_callback(void *data) 847 { 848 struct ntb_transport_ctx *nt = data; 849 850 if (ntb_link_is_up(nt->ndev, NULL, NULL) == 1) 851 schedule_delayed_work(&nt->link_work, 0); 852 else 853 schedule_work(&nt->link_cleanup); 854 } 855 856 static void ntb_transport_link_work(struct work_struct *work) 857 { 858 struct ntb_transport_ctx *nt = 859 container_of(work, struct ntb_transport_ctx, link_work.work); 860 struct ntb_dev *ndev = nt->ndev; 861 struct pci_dev *pdev = ndev->pdev; 862 resource_size_t size; 863 u32 val; 864 int rc = 0, i, spad; 865 866 /* send the local info, in the opposite order of the way we read it */ 867 for (i = 0; i < nt->mw_count; i++) { 868 size = nt->mw_vec[i].phys_size; 869 870 if (max_mw_size && size > max_mw_size) 871 size = max_mw_size; 872 873 spad = MW0_SZ_HIGH + (i * 2); 874 ntb_peer_spad_write(ndev, spad, upper_32_bits(size)); 875 876 spad = MW0_SZ_LOW + (i * 2); 877 ntb_peer_spad_write(ndev, spad, lower_32_bits(size)); 878 } 879 880 ntb_peer_spad_write(ndev, NUM_MWS, nt->mw_count); 881 882 ntb_peer_spad_write(ndev, NUM_QPS, nt->qp_count); 883 884 ntb_peer_spad_write(ndev, VERSION, NTB_TRANSPORT_VERSION); 885 886 /* Query the remote side for its info */ 887 val = ntb_spad_read(ndev, VERSION); 888 dev_dbg(&pdev->dev, "Remote version = %d\n", val); 889 if (val != NTB_TRANSPORT_VERSION) 890 goto out; 891 892 val = ntb_spad_read(ndev, NUM_QPS); 893 dev_dbg(&pdev->dev, "Remote max number of qps = %d\n", val); 894 if (val != nt->qp_count) 895 goto out; 896 897 val = ntb_spad_read(ndev, NUM_MWS); 898 dev_dbg(&pdev->dev, "Remote number of mws = %d\n", val); 899 if (val != nt->mw_count) 900 goto out; 901 902 for (i = 0; i < nt->mw_count; i++) { 903 u64 val64; 904 905 val = ntb_spad_read(ndev, MW0_SZ_HIGH + (i * 2)); 906 val64 = (u64)val << 32; 907 908 val = ntb_spad_read(ndev, MW0_SZ_LOW + (i * 2)); 909 val64 |= val; 910 911 dev_dbg(&pdev->dev, "Remote MW%d size = %#llx\n", i, val64); 912 913 rc = ntb_set_mw(nt, i, val64); 914 if (rc) 915 goto out1; 916 } 917 918 nt->link_is_up = true; 919 920 for (i = 0; i < nt->qp_count; i++) { 921 struct ntb_transport_qp *qp = &nt->qp_vec[i]; 922 923 ntb_transport_setup_qp_mw(nt, i); 924 925 if (qp->client_ready) 926 schedule_delayed_work(&qp->link_work, 0); 927 } 928 929 return; 930 931 out1: 932 for (i = 0; i < nt->mw_count; i++) 933 ntb_free_mw(nt, i); 934 935 /* if there's an actual failure, we should just bail */ 936 if (rc < 0) { 937 ntb_link_disable(ndev); 938 return; 939 } 940 941 out: 942 if (ntb_link_is_up(ndev, NULL, NULL) == 1) 943 schedule_delayed_work(&nt->link_work, 944 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 945 } 946 947 static void ntb_qp_link_work(struct work_struct *work) 948 { 949 struct ntb_transport_qp *qp = container_of(work, 950 struct ntb_transport_qp, 951 link_work.work); 952 struct pci_dev *pdev = qp->ndev->pdev; 953 struct ntb_transport_ctx *nt = qp->transport; 954 int val; 955 956 WARN_ON(!nt->link_is_up); 957 958 val = ntb_spad_read(nt->ndev, QP_LINKS); 959 960 ntb_peer_spad_write(nt->ndev, QP_LINKS, val | BIT(qp->qp_num)); 961 962 /* query remote spad for qp ready bits */ 963 ntb_peer_spad_read(nt->ndev, QP_LINKS); 964 dev_dbg_ratelimited(&pdev->dev, "Remote QP link status = %x\n", val); 965 966 /* See if the remote side is up */ 967 if (val & BIT(qp->qp_num)) { 968 dev_info(&pdev->dev, "qp %d: Link Up\n", qp->qp_num); 969 qp->link_is_up = true; 970 qp->active = true; 971 972 if (qp->event_handler) 973 qp->event_handler(qp->cb_data, qp->link_is_up); 974 975 if (qp->active) 976 tasklet_schedule(&qp->rxc_db_work); 977 } else if (nt->link_is_up) 978 schedule_delayed_work(&qp->link_work, 979 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 980 } 981 982 static int ntb_transport_init_queue(struct ntb_transport_ctx *nt, 983 unsigned int qp_num) 984 { 985 struct ntb_transport_qp *qp; 986 phys_addr_t mw_base; 987 resource_size_t mw_size; 988 unsigned int num_qps_mw, tx_size; 989 unsigned int mw_num, mw_count, qp_count; 990 u64 qp_offset; 991 992 mw_count = nt->mw_count; 993 qp_count = nt->qp_count; 994 995 mw_num = QP_TO_MW(nt, qp_num); 996 997 qp = &nt->qp_vec[qp_num]; 998 qp->qp_num = qp_num; 999 qp->transport = nt; 1000 qp->ndev = nt->ndev; 1001 qp->client_ready = false; 1002 qp->event_handler = NULL; 1003 ntb_qp_link_down_reset(qp); 1004 1005 if (qp_count % mw_count && mw_num + 1 < qp_count / mw_count) 1006 num_qps_mw = qp_count / mw_count + 1; 1007 else 1008 num_qps_mw = qp_count / mw_count; 1009 1010 mw_base = nt->mw_vec[mw_num].phys_addr; 1011 mw_size = nt->mw_vec[mw_num].phys_size; 1012 1013 tx_size = (unsigned int)mw_size / num_qps_mw; 1014 qp_offset = tx_size * (qp_num / mw_count); 1015 1016 qp->tx_mw = nt->mw_vec[mw_num].vbase + qp_offset; 1017 if (!qp->tx_mw) 1018 return -EINVAL; 1019 1020 qp->tx_mw_phys = mw_base + qp_offset; 1021 if (!qp->tx_mw_phys) 1022 return -EINVAL; 1023 1024 tx_size -= sizeof(struct ntb_rx_info); 1025 qp->rx_info = qp->tx_mw + tx_size; 1026 1027 /* Due to housekeeping, there must be atleast 2 buffs */ 1028 qp->tx_max_frame = min(transport_mtu, tx_size / 2); 1029 qp->tx_max_entry = tx_size / qp->tx_max_frame; 1030 1031 if (nt->debugfs_node_dir) { 1032 char debugfs_name[4]; 1033 1034 snprintf(debugfs_name, 4, "qp%d", qp_num); 1035 qp->debugfs_dir = debugfs_create_dir(debugfs_name, 1036 nt->debugfs_node_dir); 1037 1038 qp->debugfs_stats = debugfs_create_file("stats", S_IRUSR, 1039 qp->debugfs_dir, qp, 1040 &ntb_qp_debugfs_stats); 1041 } else { 1042 qp->debugfs_dir = NULL; 1043 qp->debugfs_stats = NULL; 1044 } 1045 1046 INIT_DELAYED_WORK(&qp->link_work, ntb_qp_link_work); 1047 INIT_WORK(&qp->link_cleanup, ntb_qp_link_cleanup_work); 1048 1049 spin_lock_init(&qp->ntb_rx_q_lock); 1050 spin_lock_init(&qp->ntb_tx_free_q_lock); 1051 1052 INIT_LIST_HEAD(&qp->rx_post_q); 1053 INIT_LIST_HEAD(&qp->rx_pend_q); 1054 INIT_LIST_HEAD(&qp->rx_free_q); 1055 INIT_LIST_HEAD(&qp->tx_free_q); 1056 1057 tasklet_init(&qp->rxc_db_work, ntb_transport_rxc_db, 1058 (unsigned long)qp); 1059 1060 return 0; 1061 } 1062 1063 static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev) 1064 { 1065 struct ntb_transport_ctx *nt; 1066 struct ntb_transport_mw *mw; 1067 unsigned int mw_count, qp_count; 1068 u64 qp_bitmap; 1069 int node; 1070 int rc, i; 1071 1072 mw_count = ntb_mw_count(ndev); 1073 if (ntb_spad_count(ndev) < (NUM_MWS + 1 + mw_count * 2)) { 1074 dev_err(&ndev->dev, "Not enough scratch pad registers for %s", 1075 NTB_TRANSPORT_NAME); 1076 return -EIO; 1077 } 1078 1079 if (ntb_db_is_unsafe(ndev)) 1080 dev_dbg(&ndev->dev, 1081 "doorbell is unsafe, proceed anyway...\n"); 1082 if (ntb_spad_is_unsafe(ndev)) 1083 dev_dbg(&ndev->dev, 1084 "scratchpad is unsafe, proceed anyway...\n"); 1085 1086 node = dev_to_node(&ndev->dev); 1087 1088 nt = kzalloc_node(sizeof(*nt), GFP_KERNEL, node); 1089 if (!nt) 1090 return -ENOMEM; 1091 1092 nt->ndev = ndev; 1093 1094 nt->mw_count = mw_count; 1095 1096 nt->mw_vec = kzalloc_node(mw_count * sizeof(*nt->mw_vec), 1097 GFP_KERNEL, node); 1098 if (!nt->mw_vec) { 1099 rc = -ENOMEM; 1100 goto err; 1101 } 1102 1103 for (i = 0; i < mw_count; i++) { 1104 mw = &nt->mw_vec[i]; 1105 1106 rc = ntb_mw_get_range(ndev, i, &mw->phys_addr, &mw->phys_size, 1107 &mw->xlat_align, &mw->xlat_align_size); 1108 if (rc) 1109 goto err1; 1110 1111 mw->vbase = ioremap_wc(mw->phys_addr, mw->phys_size); 1112 if (!mw->vbase) { 1113 rc = -ENOMEM; 1114 goto err1; 1115 } 1116 1117 mw->buff_size = 0; 1118 mw->xlat_size = 0; 1119 mw->virt_addr = NULL; 1120 mw->dma_addr = 0; 1121 } 1122 1123 qp_bitmap = ntb_db_valid_mask(ndev); 1124 1125 qp_count = ilog2(qp_bitmap); 1126 if (max_num_clients && max_num_clients < qp_count) 1127 qp_count = max_num_clients; 1128 else if (mw_count < qp_count) 1129 qp_count = mw_count; 1130 1131 qp_bitmap &= BIT_ULL(qp_count) - 1; 1132 1133 nt->qp_count = qp_count; 1134 nt->qp_bitmap = qp_bitmap; 1135 nt->qp_bitmap_free = qp_bitmap; 1136 1137 nt->qp_vec = kzalloc_node(qp_count * sizeof(*nt->qp_vec), 1138 GFP_KERNEL, node); 1139 if (!nt->qp_vec) { 1140 rc = -ENOMEM; 1141 goto err1; 1142 } 1143 1144 if (nt_debugfs_dir) { 1145 nt->debugfs_node_dir = 1146 debugfs_create_dir(pci_name(ndev->pdev), 1147 nt_debugfs_dir); 1148 } 1149 1150 for (i = 0; i < qp_count; i++) { 1151 rc = ntb_transport_init_queue(nt, i); 1152 if (rc) 1153 goto err2; 1154 } 1155 1156 INIT_DELAYED_WORK(&nt->link_work, ntb_transport_link_work); 1157 INIT_WORK(&nt->link_cleanup, ntb_transport_link_cleanup_work); 1158 1159 rc = ntb_set_ctx(ndev, nt, &ntb_transport_ops); 1160 if (rc) 1161 goto err2; 1162 1163 INIT_LIST_HEAD(&nt->client_devs); 1164 rc = ntb_bus_init(nt); 1165 if (rc) 1166 goto err3; 1167 1168 nt->link_is_up = false; 1169 ntb_link_enable(ndev, NTB_SPEED_AUTO, NTB_WIDTH_AUTO); 1170 ntb_link_event(ndev); 1171 1172 return 0; 1173 1174 err3: 1175 ntb_clear_ctx(ndev); 1176 err2: 1177 kfree(nt->qp_vec); 1178 err1: 1179 while (i--) { 1180 mw = &nt->mw_vec[i]; 1181 iounmap(mw->vbase); 1182 } 1183 kfree(nt->mw_vec); 1184 err: 1185 kfree(nt); 1186 return rc; 1187 } 1188 1189 static void ntb_transport_free(struct ntb_client *self, struct ntb_dev *ndev) 1190 { 1191 struct ntb_transport_ctx *nt = ndev->ctx; 1192 struct ntb_transport_qp *qp; 1193 u64 qp_bitmap_alloc; 1194 int i; 1195 1196 ntb_transport_link_cleanup(nt); 1197 cancel_work_sync(&nt->link_cleanup); 1198 cancel_delayed_work_sync(&nt->link_work); 1199 1200 qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free; 1201 1202 /* verify that all the qp's are freed */ 1203 for (i = 0; i < nt->qp_count; i++) { 1204 qp = &nt->qp_vec[i]; 1205 if (qp_bitmap_alloc & BIT_ULL(i)) 1206 ntb_transport_free_queue(qp); 1207 debugfs_remove_recursive(qp->debugfs_dir); 1208 } 1209 1210 ntb_link_disable(ndev); 1211 ntb_clear_ctx(ndev); 1212 1213 ntb_bus_remove(nt); 1214 1215 for (i = nt->mw_count; i--; ) { 1216 ntb_free_mw(nt, i); 1217 iounmap(nt->mw_vec[i].vbase); 1218 } 1219 1220 kfree(nt->qp_vec); 1221 kfree(nt->mw_vec); 1222 kfree(nt); 1223 } 1224 1225 static void ntb_complete_rxc(struct ntb_transport_qp *qp) 1226 { 1227 struct ntb_queue_entry *entry; 1228 void *cb_data; 1229 unsigned int len; 1230 unsigned long irqflags; 1231 1232 spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags); 1233 1234 while (!list_empty(&qp->rx_post_q)) { 1235 entry = list_first_entry(&qp->rx_post_q, 1236 struct ntb_queue_entry, entry); 1237 if (!(entry->flags & DESC_DONE_FLAG)) 1238 break; 1239 1240 entry->rx_hdr->flags = 0; 1241 iowrite32(entry->rx_index, &qp->rx_info->entry); 1242 1243 cb_data = entry->cb_data; 1244 len = entry->len; 1245 1246 list_move_tail(&entry->entry, &qp->rx_free_q); 1247 1248 spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags); 1249 1250 if (qp->rx_handler && qp->client_ready) 1251 qp->rx_handler(qp, qp->cb_data, cb_data, len); 1252 1253 spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags); 1254 } 1255 1256 spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags); 1257 } 1258 1259 static void ntb_rx_copy_callback(void *data, 1260 const struct dmaengine_result *res) 1261 { 1262 struct ntb_queue_entry *entry = data; 1263 1264 /* we need to check DMA results if we are using DMA */ 1265 if (res) { 1266 enum dmaengine_tx_result dma_err = res->result; 1267 1268 switch (dma_err) { 1269 case DMA_TRANS_READ_FAILED: 1270 case DMA_TRANS_WRITE_FAILED: 1271 entry->errors++; 1272 case DMA_TRANS_ABORTED: 1273 { 1274 struct ntb_transport_qp *qp = entry->qp; 1275 void *offset = qp->rx_buff + qp->rx_max_frame * 1276 qp->rx_index; 1277 1278 ntb_memcpy_rx(entry, offset); 1279 qp->rx_memcpy++; 1280 return; 1281 } 1282 1283 case DMA_TRANS_NOERROR: 1284 default: 1285 break; 1286 } 1287 } 1288 1289 entry->flags |= DESC_DONE_FLAG; 1290 1291 ntb_complete_rxc(entry->qp); 1292 } 1293 1294 static void ntb_memcpy_rx(struct ntb_queue_entry *entry, void *offset) 1295 { 1296 void *buf = entry->buf; 1297 size_t len = entry->len; 1298 1299 memcpy(buf, offset, len); 1300 1301 /* Ensure that the data is fully copied out before clearing the flag */ 1302 wmb(); 1303 1304 ntb_rx_copy_callback(entry, NULL); 1305 } 1306 1307 static int ntb_async_rx_submit(struct ntb_queue_entry *entry, void *offset) 1308 { 1309 struct dma_async_tx_descriptor *txd; 1310 struct ntb_transport_qp *qp = entry->qp; 1311 struct dma_chan *chan = qp->rx_dma_chan; 1312 struct dma_device *device; 1313 size_t pay_off, buff_off, len; 1314 struct dmaengine_unmap_data *unmap; 1315 dma_cookie_t cookie; 1316 void *buf = entry->buf; 1317 int retries = 0; 1318 1319 len = entry->len; 1320 device = chan->device; 1321 pay_off = (size_t)offset & ~PAGE_MASK; 1322 buff_off = (size_t)buf & ~PAGE_MASK; 1323 1324 if (!is_dma_copy_aligned(device, pay_off, buff_off, len)) 1325 goto err; 1326 1327 unmap = dmaengine_get_unmap_data(device->dev, 2, GFP_NOWAIT); 1328 if (!unmap) 1329 goto err; 1330 1331 unmap->len = len; 1332 unmap->addr[0] = dma_map_page(device->dev, virt_to_page(offset), 1333 pay_off, len, DMA_TO_DEVICE); 1334 if (dma_mapping_error(device->dev, unmap->addr[0])) 1335 goto err_get_unmap; 1336 1337 unmap->to_cnt = 1; 1338 1339 unmap->addr[1] = dma_map_page(device->dev, virt_to_page(buf), 1340 buff_off, len, DMA_FROM_DEVICE); 1341 if (dma_mapping_error(device->dev, unmap->addr[1])) 1342 goto err_get_unmap; 1343 1344 unmap->from_cnt = 1; 1345 1346 for (retries = 0; retries < DMA_RETRIES; retries++) { 1347 txd = device->device_prep_dma_memcpy(chan, 1348 unmap->addr[1], 1349 unmap->addr[0], len, 1350 DMA_PREP_INTERRUPT); 1351 if (txd) 1352 break; 1353 1354 set_current_state(TASK_INTERRUPTIBLE); 1355 schedule_timeout(DMA_OUT_RESOURCE_TO); 1356 } 1357 1358 if (!txd) { 1359 qp->dma_rx_prep_err++; 1360 goto err_get_unmap; 1361 } 1362 1363 txd->callback_result = ntb_rx_copy_callback; 1364 txd->callback_param = entry; 1365 dma_set_unmap(txd, unmap); 1366 1367 cookie = dmaengine_submit(txd); 1368 if (dma_submit_error(cookie)) 1369 goto err_set_unmap; 1370 1371 dmaengine_unmap_put(unmap); 1372 1373 qp->last_cookie = cookie; 1374 1375 qp->rx_async++; 1376 1377 return 0; 1378 1379 err_set_unmap: 1380 dmaengine_unmap_put(unmap); 1381 err_get_unmap: 1382 dmaengine_unmap_put(unmap); 1383 err: 1384 return -ENXIO; 1385 } 1386 1387 static void ntb_async_rx(struct ntb_queue_entry *entry, void *offset) 1388 { 1389 struct ntb_transport_qp *qp = entry->qp; 1390 struct dma_chan *chan = qp->rx_dma_chan; 1391 int res; 1392 1393 if (!chan) 1394 goto err; 1395 1396 if (entry->len < copy_bytes) 1397 goto err; 1398 1399 res = ntb_async_rx_submit(entry, offset); 1400 if (res < 0) 1401 goto err; 1402 1403 if (!entry->retries) 1404 qp->rx_async++; 1405 1406 return; 1407 1408 err: 1409 ntb_memcpy_rx(entry, offset); 1410 qp->rx_memcpy++; 1411 } 1412 1413 static int ntb_process_rxc(struct ntb_transport_qp *qp) 1414 { 1415 struct ntb_payload_header *hdr; 1416 struct ntb_queue_entry *entry; 1417 void *offset; 1418 1419 offset = qp->rx_buff + qp->rx_max_frame * qp->rx_index; 1420 hdr = offset + qp->rx_max_frame - sizeof(struct ntb_payload_header); 1421 1422 dev_dbg(&qp->ndev->pdev->dev, "qp %d: RX ver %u len %d flags %x\n", 1423 qp->qp_num, hdr->ver, hdr->len, hdr->flags); 1424 1425 if (!(hdr->flags & DESC_DONE_FLAG)) { 1426 dev_dbg(&qp->ndev->pdev->dev, "done flag not set\n"); 1427 qp->rx_ring_empty++; 1428 return -EAGAIN; 1429 } 1430 1431 if (hdr->flags & LINK_DOWN_FLAG) { 1432 dev_dbg(&qp->ndev->pdev->dev, "link down flag set\n"); 1433 ntb_qp_link_down(qp); 1434 hdr->flags = 0; 1435 return -EAGAIN; 1436 } 1437 1438 if (hdr->ver != (u32)qp->rx_pkts) { 1439 dev_dbg(&qp->ndev->pdev->dev, 1440 "version mismatch, expected %llu - got %u\n", 1441 qp->rx_pkts, hdr->ver); 1442 qp->rx_err_ver++; 1443 return -EIO; 1444 } 1445 1446 entry = ntb_list_mv(&qp->ntb_rx_q_lock, &qp->rx_pend_q, &qp->rx_post_q); 1447 if (!entry) { 1448 dev_dbg(&qp->ndev->pdev->dev, "no receive buffer\n"); 1449 qp->rx_err_no_buf++; 1450 return -EAGAIN; 1451 } 1452 1453 entry->rx_hdr = hdr; 1454 entry->rx_index = qp->rx_index; 1455 1456 if (hdr->len > entry->len) { 1457 dev_dbg(&qp->ndev->pdev->dev, 1458 "receive buffer overflow! Wanted %d got %d\n", 1459 hdr->len, entry->len); 1460 qp->rx_err_oflow++; 1461 1462 entry->len = -EIO; 1463 entry->flags |= DESC_DONE_FLAG; 1464 1465 ntb_complete_rxc(qp); 1466 } else { 1467 dev_dbg(&qp->ndev->pdev->dev, 1468 "RX OK index %u ver %u size %d into buf size %d\n", 1469 qp->rx_index, hdr->ver, hdr->len, entry->len); 1470 1471 qp->rx_bytes += hdr->len; 1472 qp->rx_pkts++; 1473 1474 entry->len = hdr->len; 1475 1476 ntb_async_rx(entry, offset); 1477 } 1478 1479 qp->rx_index++; 1480 qp->rx_index %= qp->rx_max_entry; 1481 1482 return 0; 1483 } 1484 1485 static void ntb_transport_rxc_db(unsigned long data) 1486 { 1487 struct ntb_transport_qp *qp = (void *)data; 1488 int rc, i; 1489 1490 dev_dbg(&qp->ndev->pdev->dev, "%s: doorbell %d received\n", 1491 __func__, qp->qp_num); 1492 1493 /* Limit the number of packets processed in a single interrupt to 1494 * provide fairness to others 1495 */ 1496 for (i = 0; i < qp->rx_max_entry; i++) { 1497 rc = ntb_process_rxc(qp); 1498 if (rc) 1499 break; 1500 } 1501 1502 if (i && qp->rx_dma_chan) 1503 dma_async_issue_pending(qp->rx_dma_chan); 1504 1505 if (i == qp->rx_max_entry) { 1506 /* there is more work to do */ 1507 if (qp->active) 1508 tasklet_schedule(&qp->rxc_db_work); 1509 } else if (ntb_db_read(qp->ndev) & BIT_ULL(qp->qp_num)) { 1510 /* the doorbell bit is set: clear it */ 1511 ntb_db_clear(qp->ndev, BIT_ULL(qp->qp_num)); 1512 /* ntb_db_read ensures ntb_db_clear write is committed */ 1513 ntb_db_read(qp->ndev); 1514 1515 /* an interrupt may have arrived between finishing 1516 * ntb_process_rxc and clearing the doorbell bit: 1517 * there might be some more work to do. 1518 */ 1519 if (qp->active) 1520 tasklet_schedule(&qp->rxc_db_work); 1521 } 1522 } 1523 1524 static void ntb_tx_copy_callback(void *data, 1525 const struct dmaengine_result *res) 1526 { 1527 struct ntb_queue_entry *entry = data; 1528 struct ntb_transport_qp *qp = entry->qp; 1529 struct ntb_payload_header __iomem *hdr = entry->tx_hdr; 1530 1531 /* we need to check DMA results if we are using DMA */ 1532 if (res) { 1533 enum dmaengine_tx_result dma_err = res->result; 1534 1535 switch (dma_err) { 1536 case DMA_TRANS_READ_FAILED: 1537 case DMA_TRANS_WRITE_FAILED: 1538 entry->errors++; 1539 case DMA_TRANS_ABORTED: 1540 { 1541 void __iomem *offset = 1542 qp->tx_mw + qp->tx_max_frame * 1543 entry->tx_index; 1544 1545 /* resubmit via CPU */ 1546 ntb_memcpy_tx(entry, offset); 1547 qp->tx_memcpy++; 1548 return; 1549 } 1550 1551 case DMA_TRANS_NOERROR: 1552 default: 1553 break; 1554 } 1555 } 1556 1557 iowrite32(entry->flags | DESC_DONE_FLAG, &hdr->flags); 1558 1559 ntb_peer_db_set(qp->ndev, BIT_ULL(qp->qp_num)); 1560 1561 /* The entry length can only be zero if the packet is intended to be a 1562 * "link down" or similar. Since no payload is being sent in these 1563 * cases, there is nothing to add to the completion queue. 1564 */ 1565 if (entry->len > 0) { 1566 qp->tx_bytes += entry->len; 1567 1568 if (qp->tx_handler) 1569 qp->tx_handler(qp, qp->cb_data, entry->cb_data, 1570 entry->len); 1571 } 1572 1573 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, &qp->tx_free_q); 1574 } 1575 1576 static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset) 1577 { 1578 #ifdef ARCH_HAS_NOCACHE_UACCESS 1579 /* 1580 * Using non-temporal mov to improve performance on non-cached 1581 * writes, even though we aren't actually copying from user space. 1582 */ 1583 __copy_from_user_inatomic_nocache(offset, entry->buf, entry->len); 1584 #else 1585 memcpy_toio(offset, entry->buf, entry->len); 1586 #endif 1587 1588 /* Ensure that the data is fully copied out before setting the flags */ 1589 wmb(); 1590 1591 ntb_tx_copy_callback(entry, NULL); 1592 } 1593 1594 static int ntb_async_tx_submit(struct ntb_transport_qp *qp, 1595 struct ntb_queue_entry *entry) 1596 { 1597 struct dma_async_tx_descriptor *txd; 1598 struct dma_chan *chan = qp->tx_dma_chan; 1599 struct dma_device *device; 1600 size_t len = entry->len; 1601 void *buf = entry->buf; 1602 size_t dest_off, buff_off; 1603 struct dmaengine_unmap_data *unmap; 1604 dma_addr_t dest; 1605 dma_cookie_t cookie; 1606 int retries = 0; 1607 1608 device = chan->device; 1609 dest = qp->tx_mw_phys + qp->tx_max_frame * entry->tx_index; 1610 buff_off = (size_t)buf & ~PAGE_MASK; 1611 dest_off = (size_t)dest & ~PAGE_MASK; 1612 1613 if (!is_dma_copy_aligned(device, buff_off, dest_off, len)) 1614 goto err; 1615 1616 unmap = dmaengine_get_unmap_data(device->dev, 1, GFP_NOWAIT); 1617 if (!unmap) 1618 goto err; 1619 1620 unmap->len = len; 1621 unmap->addr[0] = dma_map_page(device->dev, virt_to_page(buf), 1622 buff_off, len, DMA_TO_DEVICE); 1623 if (dma_mapping_error(device->dev, unmap->addr[0])) 1624 goto err_get_unmap; 1625 1626 unmap->to_cnt = 1; 1627 1628 for (retries = 0; retries < DMA_RETRIES; retries++) { 1629 txd = device->device_prep_dma_memcpy(chan, dest, 1630 unmap->addr[0], len, 1631 DMA_PREP_INTERRUPT); 1632 if (txd) 1633 break; 1634 1635 set_current_state(TASK_INTERRUPTIBLE); 1636 schedule_timeout(DMA_OUT_RESOURCE_TO); 1637 } 1638 1639 if (!txd) { 1640 qp->dma_tx_prep_err++; 1641 goto err_get_unmap; 1642 } 1643 1644 txd->callback_result = ntb_tx_copy_callback; 1645 txd->callback_param = entry; 1646 dma_set_unmap(txd, unmap); 1647 1648 cookie = dmaengine_submit(txd); 1649 if (dma_submit_error(cookie)) 1650 goto err_set_unmap; 1651 1652 dmaengine_unmap_put(unmap); 1653 1654 dma_async_issue_pending(chan); 1655 1656 return 0; 1657 err_set_unmap: 1658 dmaengine_unmap_put(unmap); 1659 err_get_unmap: 1660 dmaengine_unmap_put(unmap); 1661 err: 1662 return -ENXIO; 1663 } 1664 1665 static void ntb_async_tx(struct ntb_transport_qp *qp, 1666 struct ntb_queue_entry *entry) 1667 { 1668 struct ntb_payload_header __iomem *hdr; 1669 struct dma_chan *chan = qp->tx_dma_chan; 1670 void __iomem *offset; 1671 int res; 1672 1673 entry->tx_index = qp->tx_index; 1674 offset = qp->tx_mw + qp->tx_max_frame * entry->tx_index; 1675 hdr = offset + qp->tx_max_frame - sizeof(struct ntb_payload_header); 1676 entry->tx_hdr = hdr; 1677 1678 iowrite32(entry->len, &hdr->len); 1679 iowrite32((u32)qp->tx_pkts, &hdr->ver); 1680 1681 if (!chan) 1682 goto err; 1683 1684 if (entry->len < copy_bytes) 1685 goto err; 1686 1687 res = ntb_async_tx_submit(qp, entry); 1688 if (res < 0) 1689 goto err; 1690 1691 if (!entry->retries) 1692 qp->tx_async++; 1693 1694 return; 1695 1696 err: 1697 ntb_memcpy_tx(entry, offset); 1698 qp->tx_memcpy++; 1699 } 1700 1701 static int ntb_process_tx(struct ntb_transport_qp *qp, 1702 struct ntb_queue_entry *entry) 1703 { 1704 if (qp->tx_index == qp->remote_rx_info->entry) { 1705 qp->tx_ring_full++; 1706 return -EAGAIN; 1707 } 1708 1709 if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) { 1710 if (qp->tx_handler) 1711 qp->tx_handler(qp, qp->cb_data, NULL, -EIO); 1712 1713 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1714 &qp->tx_free_q); 1715 return 0; 1716 } 1717 1718 ntb_async_tx(qp, entry); 1719 1720 qp->tx_index++; 1721 qp->tx_index %= qp->tx_max_entry; 1722 1723 qp->tx_pkts++; 1724 1725 return 0; 1726 } 1727 1728 static void ntb_send_link_down(struct ntb_transport_qp *qp) 1729 { 1730 struct pci_dev *pdev = qp->ndev->pdev; 1731 struct ntb_queue_entry *entry; 1732 int i, rc; 1733 1734 if (!qp->link_is_up) 1735 return; 1736 1737 dev_info(&pdev->dev, "qp %d: Send Link Down\n", qp->qp_num); 1738 1739 for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) { 1740 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); 1741 if (entry) 1742 break; 1743 msleep(100); 1744 } 1745 1746 if (!entry) 1747 return; 1748 1749 entry->cb_data = NULL; 1750 entry->buf = NULL; 1751 entry->len = 0; 1752 entry->flags = LINK_DOWN_FLAG; 1753 1754 rc = ntb_process_tx(qp, entry); 1755 if (rc) 1756 dev_err(&pdev->dev, "ntb: QP%d unable to send linkdown msg\n", 1757 qp->qp_num); 1758 1759 ntb_qp_link_down_reset(qp); 1760 } 1761 1762 static bool ntb_dma_filter_fn(struct dma_chan *chan, void *node) 1763 { 1764 return dev_to_node(&chan->dev->device) == (int)(unsigned long)node; 1765 } 1766 1767 /** 1768 * ntb_transport_create_queue - Create a new NTB transport layer queue 1769 * @rx_handler: receive callback function 1770 * @tx_handler: transmit callback function 1771 * @event_handler: event callback function 1772 * 1773 * Create a new NTB transport layer queue and provide the queue with a callback 1774 * routine for both transmit and receive. The receive callback routine will be 1775 * used to pass up data when the transport has received it on the queue. The 1776 * transmit callback routine will be called when the transport has completed the 1777 * transmission of the data on the queue and the data is ready to be freed. 1778 * 1779 * RETURNS: pointer to newly created ntb_queue, NULL on error. 1780 */ 1781 struct ntb_transport_qp * 1782 ntb_transport_create_queue(void *data, struct device *client_dev, 1783 const struct ntb_queue_handlers *handlers) 1784 { 1785 struct ntb_dev *ndev; 1786 struct pci_dev *pdev; 1787 struct ntb_transport_ctx *nt; 1788 struct ntb_queue_entry *entry; 1789 struct ntb_transport_qp *qp; 1790 u64 qp_bit; 1791 unsigned int free_queue; 1792 dma_cap_mask_t dma_mask; 1793 int node; 1794 int i; 1795 1796 ndev = dev_ntb(client_dev->parent); 1797 pdev = ndev->pdev; 1798 nt = ndev->ctx; 1799 1800 node = dev_to_node(&ndev->dev); 1801 1802 free_queue = ffs(nt->qp_bitmap); 1803 if (!free_queue) 1804 goto err; 1805 1806 /* decrement free_queue to make it zero based */ 1807 free_queue--; 1808 1809 qp = &nt->qp_vec[free_queue]; 1810 qp_bit = BIT_ULL(qp->qp_num); 1811 1812 nt->qp_bitmap_free &= ~qp_bit; 1813 1814 qp->cb_data = data; 1815 qp->rx_handler = handlers->rx_handler; 1816 qp->tx_handler = handlers->tx_handler; 1817 qp->event_handler = handlers->event_handler; 1818 1819 dma_cap_zero(dma_mask); 1820 dma_cap_set(DMA_MEMCPY, dma_mask); 1821 1822 if (use_dma) { 1823 qp->tx_dma_chan = 1824 dma_request_channel(dma_mask, ntb_dma_filter_fn, 1825 (void *)(unsigned long)node); 1826 if (!qp->tx_dma_chan) 1827 dev_info(&pdev->dev, "Unable to allocate TX DMA channel\n"); 1828 1829 qp->rx_dma_chan = 1830 dma_request_channel(dma_mask, ntb_dma_filter_fn, 1831 (void *)(unsigned long)node); 1832 if (!qp->rx_dma_chan) 1833 dev_info(&pdev->dev, "Unable to allocate RX DMA channel\n"); 1834 } else { 1835 qp->tx_dma_chan = NULL; 1836 qp->rx_dma_chan = NULL; 1837 } 1838 1839 dev_dbg(&pdev->dev, "Using %s memcpy for TX\n", 1840 qp->tx_dma_chan ? "DMA" : "CPU"); 1841 1842 dev_dbg(&pdev->dev, "Using %s memcpy for RX\n", 1843 qp->rx_dma_chan ? "DMA" : "CPU"); 1844 1845 for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) { 1846 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 1847 if (!entry) 1848 goto err1; 1849 1850 entry->qp = qp; 1851 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, 1852 &qp->rx_free_q); 1853 } 1854 qp->rx_alloc_entry = NTB_QP_DEF_NUM_ENTRIES; 1855 1856 for (i = 0; i < qp->tx_max_entry; i++) { 1857 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 1858 if (!entry) 1859 goto err2; 1860 1861 entry->qp = qp; 1862 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1863 &qp->tx_free_q); 1864 } 1865 1866 ntb_db_clear(qp->ndev, qp_bit); 1867 ntb_db_clear_mask(qp->ndev, qp_bit); 1868 1869 dev_info(&pdev->dev, "NTB Transport QP %d created\n", qp->qp_num); 1870 1871 return qp; 1872 1873 err2: 1874 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) 1875 kfree(entry); 1876 err1: 1877 qp->rx_alloc_entry = 0; 1878 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q))) 1879 kfree(entry); 1880 if (qp->tx_dma_chan) 1881 dma_release_channel(qp->tx_dma_chan); 1882 if (qp->rx_dma_chan) 1883 dma_release_channel(qp->rx_dma_chan); 1884 nt->qp_bitmap_free |= qp_bit; 1885 err: 1886 return NULL; 1887 } 1888 EXPORT_SYMBOL_GPL(ntb_transport_create_queue); 1889 1890 /** 1891 * ntb_transport_free_queue - Frees NTB transport queue 1892 * @qp: NTB queue to be freed 1893 * 1894 * Frees NTB transport queue 1895 */ 1896 void ntb_transport_free_queue(struct ntb_transport_qp *qp) 1897 { 1898 struct pci_dev *pdev; 1899 struct ntb_queue_entry *entry; 1900 u64 qp_bit; 1901 1902 if (!qp) 1903 return; 1904 1905 pdev = qp->ndev->pdev; 1906 1907 qp->active = false; 1908 1909 if (qp->tx_dma_chan) { 1910 struct dma_chan *chan = qp->tx_dma_chan; 1911 /* Putting the dma_chan to NULL will force any new traffic to be 1912 * processed by the CPU instead of the DAM engine 1913 */ 1914 qp->tx_dma_chan = NULL; 1915 1916 /* Try to be nice and wait for any queued DMA engine 1917 * transactions to process before smashing it with a rock 1918 */ 1919 dma_sync_wait(chan, qp->last_cookie); 1920 dmaengine_terminate_all(chan); 1921 dma_release_channel(chan); 1922 } 1923 1924 if (qp->rx_dma_chan) { 1925 struct dma_chan *chan = qp->rx_dma_chan; 1926 /* Putting the dma_chan to NULL will force any new traffic to be 1927 * processed by the CPU instead of the DAM engine 1928 */ 1929 qp->rx_dma_chan = NULL; 1930 1931 /* Try to be nice and wait for any queued DMA engine 1932 * transactions to process before smashing it with a rock 1933 */ 1934 dma_sync_wait(chan, qp->last_cookie); 1935 dmaengine_terminate_all(chan); 1936 dma_release_channel(chan); 1937 } 1938 1939 qp_bit = BIT_ULL(qp->qp_num); 1940 1941 ntb_db_set_mask(qp->ndev, qp_bit); 1942 tasklet_kill(&qp->rxc_db_work); 1943 1944 cancel_delayed_work_sync(&qp->link_work); 1945 1946 qp->cb_data = NULL; 1947 qp->rx_handler = NULL; 1948 qp->tx_handler = NULL; 1949 qp->event_handler = NULL; 1950 1951 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q))) 1952 kfree(entry); 1953 1954 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q))) { 1955 dev_warn(&pdev->dev, "Freeing item from non-empty rx_pend_q\n"); 1956 kfree(entry); 1957 } 1958 1959 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_post_q))) { 1960 dev_warn(&pdev->dev, "Freeing item from non-empty rx_post_q\n"); 1961 kfree(entry); 1962 } 1963 1964 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) 1965 kfree(entry); 1966 1967 qp->transport->qp_bitmap_free |= qp_bit; 1968 1969 dev_info(&pdev->dev, "NTB Transport QP %d freed\n", qp->qp_num); 1970 } 1971 EXPORT_SYMBOL_GPL(ntb_transport_free_queue); 1972 1973 /** 1974 * ntb_transport_rx_remove - Dequeues enqueued rx packet 1975 * @qp: NTB queue to be freed 1976 * @len: pointer to variable to write enqueued buffers length 1977 * 1978 * Dequeues unused buffers from receive queue. Should only be used during 1979 * shutdown of qp. 1980 * 1981 * RETURNS: NULL error value on error, or void* for success. 1982 */ 1983 void *ntb_transport_rx_remove(struct ntb_transport_qp *qp, unsigned int *len) 1984 { 1985 struct ntb_queue_entry *entry; 1986 void *buf; 1987 1988 if (!qp || qp->client_ready) 1989 return NULL; 1990 1991 entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q); 1992 if (!entry) 1993 return NULL; 1994 1995 buf = entry->cb_data; 1996 *len = entry->len; 1997 1998 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_free_q); 1999 2000 return buf; 2001 } 2002 EXPORT_SYMBOL_GPL(ntb_transport_rx_remove); 2003 2004 /** 2005 * ntb_transport_rx_enqueue - Enqueue a new NTB queue entry 2006 * @qp: NTB transport layer queue the entry is to be enqueued on 2007 * @cb: per buffer pointer for callback function to use 2008 * @data: pointer to data buffer that incoming packets will be copied into 2009 * @len: length of the data buffer 2010 * 2011 * Enqueue a new receive buffer onto the transport queue into which a NTB 2012 * payload can be received into. 2013 * 2014 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 2015 */ 2016 int ntb_transport_rx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, 2017 unsigned int len) 2018 { 2019 struct ntb_queue_entry *entry; 2020 2021 if (!qp) 2022 return -EINVAL; 2023 2024 entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q); 2025 if (!entry) 2026 return -ENOMEM; 2027 2028 entry->cb_data = cb; 2029 entry->buf = data; 2030 entry->len = len; 2031 entry->flags = 0; 2032 entry->retries = 0; 2033 entry->errors = 0; 2034 entry->rx_index = 0; 2035 2036 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_pend_q); 2037 2038 if (qp->active) 2039 tasklet_schedule(&qp->rxc_db_work); 2040 2041 return 0; 2042 } 2043 EXPORT_SYMBOL_GPL(ntb_transport_rx_enqueue); 2044 2045 /** 2046 * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry 2047 * @qp: NTB transport layer queue the entry is to be enqueued on 2048 * @cb: per buffer pointer for callback function to use 2049 * @data: pointer to data buffer that will be sent 2050 * @len: length of the data buffer 2051 * 2052 * Enqueue a new transmit buffer onto the transport queue from which a NTB 2053 * payload will be transmitted. This assumes that a lock is being held to 2054 * serialize access to the qp. 2055 * 2056 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 2057 */ 2058 int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, 2059 unsigned int len) 2060 { 2061 struct ntb_queue_entry *entry; 2062 int rc; 2063 2064 if (!qp || !qp->link_is_up || !len) 2065 return -EINVAL; 2066 2067 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); 2068 if (!entry) { 2069 qp->tx_err_no_buf++; 2070 return -EBUSY; 2071 } 2072 2073 entry->cb_data = cb; 2074 entry->buf = data; 2075 entry->len = len; 2076 entry->flags = 0; 2077 entry->errors = 0; 2078 entry->retries = 0; 2079 entry->tx_index = 0; 2080 2081 rc = ntb_process_tx(qp, entry); 2082 if (rc) 2083 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 2084 &qp->tx_free_q); 2085 2086 return rc; 2087 } 2088 EXPORT_SYMBOL_GPL(ntb_transport_tx_enqueue); 2089 2090 /** 2091 * ntb_transport_link_up - Notify NTB transport of client readiness to use queue 2092 * @qp: NTB transport layer queue to be enabled 2093 * 2094 * Notify NTB transport layer of client readiness to use queue 2095 */ 2096 void ntb_transport_link_up(struct ntb_transport_qp *qp) 2097 { 2098 if (!qp) 2099 return; 2100 2101 qp->client_ready = true; 2102 2103 if (qp->transport->link_is_up) 2104 schedule_delayed_work(&qp->link_work, 0); 2105 } 2106 EXPORT_SYMBOL_GPL(ntb_transport_link_up); 2107 2108 /** 2109 * ntb_transport_link_down - Notify NTB transport to no longer enqueue data 2110 * @qp: NTB transport layer queue to be disabled 2111 * 2112 * Notify NTB transport layer of client's desire to no longer receive data on 2113 * transport queue specified. It is the client's responsibility to ensure all 2114 * entries on queue are purged or otherwise handled appropriately. 2115 */ 2116 void ntb_transport_link_down(struct ntb_transport_qp *qp) 2117 { 2118 int val; 2119 2120 if (!qp) 2121 return; 2122 2123 qp->client_ready = false; 2124 2125 val = ntb_spad_read(qp->ndev, QP_LINKS); 2126 2127 ntb_peer_spad_write(qp->ndev, QP_LINKS, 2128 val & ~BIT(qp->qp_num)); 2129 2130 if (qp->link_is_up) 2131 ntb_send_link_down(qp); 2132 else 2133 cancel_delayed_work_sync(&qp->link_work); 2134 } 2135 EXPORT_SYMBOL_GPL(ntb_transport_link_down); 2136 2137 /** 2138 * ntb_transport_link_query - Query transport link state 2139 * @qp: NTB transport layer queue to be queried 2140 * 2141 * Query connectivity to the remote system of the NTB transport queue 2142 * 2143 * RETURNS: true for link up or false for link down 2144 */ 2145 bool ntb_transport_link_query(struct ntb_transport_qp *qp) 2146 { 2147 if (!qp) 2148 return false; 2149 2150 return qp->link_is_up; 2151 } 2152 EXPORT_SYMBOL_GPL(ntb_transport_link_query); 2153 2154 /** 2155 * ntb_transport_qp_num - Query the qp number 2156 * @qp: NTB transport layer queue to be queried 2157 * 2158 * Query qp number of the NTB transport queue 2159 * 2160 * RETURNS: a zero based number specifying the qp number 2161 */ 2162 unsigned char ntb_transport_qp_num(struct ntb_transport_qp *qp) 2163 { 2164 if (!qp) 2165 return 0; 2166 2167 return qp->qp_num; 2168 } 2169 EXPORT_SYMBOL_GPL(ntb_transport_qp_num); 2170 2171 /** 2172 * ntb_transport_max_size - Query the max payload size of a qp 2173 * @qp: NTB transport layer queue to be queried 2174 * 2175 * Query the maximum payload size permissible on the given qp 2176 * 2177 * RETURNS: the max payload size of a qp 2178 */ 2179 unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp) 2180 { 2181 unsigned int max_size; 2182 unsigned int copy_align; 2183 struct dma_chan *rx_chan, *tx_chan; 2184 2185 if (!qp) 2186 return 0; 2187 2188 rx_chan = qp->rx_dma_chan; 2189 tx_chan = qp->tx_dma_chan; 2190 2191 copy_align = max(rx_chan ? rx_chan->device->copy_align : 0, 2192 tx_chan ? tx_chan->device->copy_align : 0); 2193 2194 /* If DMA engine usage is possible, try to find the max size for that */ 2195 max_size = qp->tx_max_frame - sizeof(struct ntb_payload_header); 2196 max_size = round_down(max_size, 1 << copy_align); 2197 2198 return max_size; 2199 } 2200 EXPORT_SYMBOL_GPL(ntb_transport_max_size); 2201 2202 unsigned int ntb_transport_tx_free_entry(struct ntb_transport_qp *qp) 2203 { 2204 unsigned int head = qp->tx_index; 2205 unsigned int tail = qp->remote_rx_info->entry; 2206 2207 return tail > head ? tail - head : qp->tx_max_entry + tail - head; 2208 } 2209 EXPORT_SYMBOL_GPL(ntb_transport_tx_free_entry); 2210 2211 static void ntb_transport_doorbell_callback(void *data, int vector) 2212 { 2213 struct ntb_transport_ctx *nt = data; 2214 struct ntb_transport_qp *qp; 2215 u64 db_bits; 2216 unsigned int qp_num; 2217 2218 db_bits = (nt->qp_bitmap & ~nt->qp_bitmap_free & 2219 ntb_db_vector_mask(nt->ndev, vector)); 2220 2221 while (db_bits) { 2222 qp_num = __ffs(db_bits); 2223 qp = &nt->qp_vec[qp_num]; 2224 2225 if (qp->active) 2226 tasklet_schedule(&qp->rxc_db_work); 2227 2228 db_bits &= ~BIT_ULL(qp_num); 2229 } 2230 } 2231 2232 static const struct ntb_ctx_ops ntb_transport_ops = { 2233 .link_event = ntb_transport_event_callback, 2234 .db_event = ntb_transport_doorbell_callback, 2235 }; 2236 2237 static struct ntb_client ntb_transport_client = { 2238 .ops = { 2239 .probe = ntb_transport_probe, 2240 .remove = ntb_transport_free, 2241 }, 2242 }; 2243 2244 static int __init ntb_transport_init(void) 2245 { 2246 int rc; 2247 2248 pr_info("%s, version %s\n", NTB_TRANSPORT_DESC, NTB_TRANSPORT_VER); 2249 2250 if (debugfs_initialized()) 2251 nt_debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL); 2252 2253 rc = bus_register(&ntb_transport_bus); 2254 if (rc) 2255 goto err_bus; 2256 2257 rc = ntb_register_client(&ntb_transport_client); 2258 if (rc) 2259 goto err_client; 2260 2261 return 0; 2262 2263 err_client: 2264 bus_unregister(&ntb_transport_bus); 2265 err_bus: 2266 debugfs_remove_recursive(nt_debugfs_dir); 2267 return rc; 2268 } 2269 module_init(ntb_transport_init); 2270 2271 static void __exit ntb_transport_exit(void) 2272 { 2273 debugfs_remove_recursive(nt_debugfs_dir); 2274 2275 ntb_unregister_client(&ntb_transport_client); 2276 bus_unregister(&ntb_transport_bus); 2277 } 2278 module_exit(ntb_transport_exit); 2279