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; 127 128 u8 qp_num; /* Only 64 QP's are allowed. 0-63 */ 129 u64 qp_bit; 130 131 struct ntb_rx_info __iomem *rx_info; 132 struct ntb_rx_info *remote_rx_info; 133 134 void (*tx_handler)(struct ntb_transport_qp *qp, void *qp_data, 135 void *data, int len); 136 struct list_head tx_free_q; 137 spinlock_t ntb_tx_free_q_lock; 138 void __iomem *tx_mw; 139 dma_addr_t tx_mw_phys; 140 unsigned int tx_index; 141 unsigned int tx_max_entry; 142 unsigned int tx_max_frame; 143 144 void (*rx_handler)(struct ntb_transport_qp *qp, void *qp_data, 145 void *data, int len); 146 struct list_head rx_post_q; 147 struct list_head rx_pend_q; 148 struct list_head rx_free_q; 149 /* ntb_rx_q_lock: synchronize access to rx_XXXX_q */ 150 spinlock_t ntb_rx_q_lock; 151 void *rx_buff; 152 unsigned int rx_index; 153 unsigned int rx_max_entry; 154 unsigned int rx_max_frame; 155 dma_cookie_t last_cookie; 156 struct tasklet_struct rxc_db_work; 157 158 void (*event_handler)(void *data, int status); 159 struct delayed_work link_work; 160 struct work_struct link_cleanup; 161 162 struct dentry *debugfs_dir; 163 struct dentry *debugfs_stats; 164 165 /* Stats */ 166 u64 rx_bytes; 167 u64 rx_pkts; 168 u64 rx_ring_empty; 169 u64 rx_err_no_buf; 170 u64 rx_err_oflow; 171 u64 rx_err_ver; 172 u64 rx_memcpy; 173 u64 rx_async; 174 u64 tx_bytes; 175 u64 tx_pkts; 176 u64 tx_ring_full; 177 u64 tx_err_no_buf; 178 u64 tx_memcpy; 179 u64 tx_async; 180 }; 181 182 struct ntb_transport_mw { 183 phys_addr_t phys_addr; 184 resource_size_t phys_size; 185 resource_size_t xlat_align; 186 resource_size_t xlat_align_size; 187 void __iomem *vbase; 188 size_t xlat_size; 189 size_t buff_size; 190 void *virt_addr; 191 dma_addr_t dma_addr; 192 }; 193 194 struct ntb_transport_client_dev { 195 struct list_head entry; 196 struct ntb_transport_ctx *nt; 197 struct device dev; 198 }; 199 200 struct ntb_transport_ctx { 201 struct list_head entry; 202 struct list_head client_devs; 203 204 struct ntb_dev *ndev; 205 206 struct ntb_transport_mw *mw_vec; 207 struct ntb_transport_qp *qp_vec; 208 unsigned int mw_count; 209 unsigned int qp_count; 210 u64 qp_bitmap; 211 u64 qp_bitmap_free; 212 213 bool link_is_up; 214 struct delayed_work link_work; 215 struct work_struct link_cleanup; 216 217 struct dentry *debugfs_node_dir; 218 }; 219 220 enum { 221 DESC_DONE_FLAG = BIT(0), 222 LINK_DOWN_FLAG = BIT(1), 223 }; 224 225 struct ntb_payload_header { 226 unsigned int ver; 227 unsigned int len; 228 unsigned int flags; 229 }; 230 231 enum { 232 VERSION = 0, 233 QP_LINKS, 234 NUM_QPS, 235 NUM_MWS, 236 MW0_SZ_HIGH, 237 MW0_SZ_LOW, 238 MW1_SZ_HIGH, 239 MW1_SZ_LOW, 240 MAX_SPAD, 241 }; 242 243 #define dev_client_dev(__dev) \ 244 container_of((__dev), struct ntb_transport_client_dev, dev) 245 246 #define drv_client(__drv) \ 247 container_of((__drv), struct ntb_transport_client, driver) 248 249 #define QP_TO_MW(nt, qp) ((qp) % nt->mw_count) 250 #define NTB_QP_DEF_NUM_ENTRIES 100 251 #define NTB_LINK_DOWN_TIMEOUT 10 252 253 static void ntb_transport_rxc_db(unsigned long data); 254 static const struct ntb_ctx_ops ntb_transport_ops; 255 static struct ntb_client ntb_transport_client; 256 257 static int ntb_transport_bus_match(struct device *dev, 258 struct device_driver *drv) 259 { 260 return !strncmp(dev_name(dev), drv->name, strlen(drv->name)); 261 } 262 263 static int ntb_transport_bus_probe(struct device *dev) 264 { 265 const struct ntb_transport_client *client; 266 int rc = -EINVAL; 267 268 get_device(dev); 269 270 client = drv_client(dev->driver); 271 rc = client->probe(dev); 272 if (rc) 273 put_device(dev); 274 275 return rc; 276 } 277 278 static int ntb_transport_bus_remove(struct device *dev) 279 { 280 const struct ntb_transport_client *client; 281 282 client = drv_client(dev->driver); 283 client->remove(dev); 284 285 put_device(dev); 286 287 return 0; 288 } 289 290 static struct bus_type ntb_transport_bus = { 291 .name = "ntb_transport", 292 .match = ntb_transport_bus_match, 293 .probe = ntb_transport_bus_probe, 294 .remove = ntb_transport_bus_remove, 295 }; 296 297 static LIST_HEAD(ntb_transport_list); 298 299 static int ntb_bus_init(struct ntb_transport_ctx *nt) 300 { 301 list_add_tail(&nt->entry, &ntb_transport_list); 302 return 0; 303 } 304 305 static void ntb_bus_remove(struct ntb_transport_ctx *nt) 306 { 307 struct ntb_transport_client_dev *client_dev, *cd; 308 309 list_for_each_entry_safe(client_dev, cd, &nt->client_devs, entry) { 310 dev_err(client_dev->dev.parent, "%s still attached to bus, removing\n", 311 dev_name(&client_dev->dev)); 312 list_del(&client_dev->entry); 313 device_unregister(&client_dev->dev); 314 } 315 316 list_del(&nt->entry); 317 } 318 319 static void ntb_transport_client_release(struct device *dev) 320 { 321 struct ntb_transport_client_dev *client_dev; 322 323 client_dev = dev_client_dev(dev); 324 kfree(client_dev); 325 } 326 327 /** 328 * ntb_transport_unregister_client_dev - Unregister NTB client device 329 * @device_name: Name of NTB client device 330 * 331 * Unregister an NTB client device with the NTB transport layer 332 */ 333 void ntb_transport_unregister_client_dev(char *device_name) 334 { 335 struct ntb_transport_client_dev *client, *cd; 336 struct ntb_transport_ctx *nt; 337 338 list_for_each_entry(nt, &ntb_transport_list, entry) 339 list_for_each_entry_safe(client, cd, &nt->client_devs, entry) 340 if (!strncmp(dev_name(&client->dev), device_name, 341 strlen(device_name))) { 342 list_del(&client->entry); 343 device_unregister(&client->dev); 344 } 345 } 346 EXPORT_SYMBOL_GPL(ntb_transport_unregister_client_dev); 347 348 /** 349 * ntb_transport_register_client_dev - Register NTB client device 350 * @device_name: Name of NTB client device 351 * 352 * Register an NTB client device with the NTB transport layer 353 */ 354 int ntb_transport_register_client_dev(char *device_name) 355 { 356 struct ntb_transport_client_dev *client_dev; 357 struct ntb_transport_ctx *nt; 358 int node; 359 int rc, i = 0; 360 361 if (list_empty(&ntb_transport_list)) 362 return -ENODEV; 363 364 list_for_each_entry(nt, &ntb_transport_list, entry) { 365 struct device *dev; 366 367 node = dev_to_node(&nt->ndev->dev); 368 369 client_dev = kzalloc_node(sizeof(*client_dev), 370 GFP_KERNEL, node); 371 if (!client_dev) { 372 rc = -ENOMEM; 373 goto err; 374 } 375 376 dev = &client_dev->dev; 377 378 /* setup and register client devices */ 379 dev_set_name(dev, "%s%d", device_name, i); 380 dev->bus = &ntb_transport_bus; 381 dev->release = ntb_transport_client_release; 382 dev->parent = &nt->ndev->dev; 383 384 rc = device_register(dev); 385 if (rc) { 386 kfree(client_dev); 387 goto err; 388 } 389 390 list_add_tail(&client_dev->entry, &nt->client_devs); 391 i++; 392 } 393 394 return 0; 395 396 err: 397 ntb_transport_unregister_client_dev(device_name); 398 399 return rc; 400 } 401 EXPORT_SYMBOL_GPL(ntb_transport_register_client_dev); 402 403 /** 404 * ntb_transport_register_client - Register NTB client driver 405 * @drv: NTB client driver to be registered 406 * 407 * Register an NTB client driver with the NTB transport layer 408 * 409 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 410 */ 411 int ntb_transport_register_client(struct ntb_transport_client *drv) 412 { 413 drv->driver.bus = &ntb_transport_bus; 414 415 if (list_empty(&ntb_transport_list)) 416 return -ENODEV; 417 418 return driver_register(&drv->driver); 419 } 420 EXPORT_SYMBOL_GPL(ntb_transport_register_client); 421 422 /** 423 * ntb_transport_unregister_client - Unregister NTB client driver 424 * @drv: NTB client driver to be unregistered 425 * 426 * Unregister an NTB client driver with the NTB transport layer 427 * 428 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 429 */ 430 void ntb_transport_unregister_client(struct ntb_transport_client *drv) 431 { 432 driver_unregister(&drv->driver); 433 } 434 EXPORT_SYMBOL_GPL(ntb_transport_unregister_client); 435 436 static ssize_t debugfs_read(struct file *filp, char __user *ubuf, size_t count, 437 loff_t *offp) 438 { 439 struct ntb_transport_qp *qp; 440 char *buf; 441 ssize_t ret, out_offset, out_count; 442 443 qp = filp->private_data; 444 445 if (!qp || !qp->link_is_up) 446 return 0; 447 448 out_count = 1000; 449 450 buf = kmalloc(out_count, GFP_KERNEL); 451 if (!buf) 452 return -ENOMEM; 453 454 out_offset = 0; 455 out_offset += snprintf(buf + out_offset, out_count - out_offset, 456 "\nNTB QP stats:\n\n"); 457 out_offset += snprintf(buf + out_offset, out_count - out_offset, 458 "rx_bytes - \t%llu\n", qp->rx_bytes); 459 out_offset += snprintf(buf + out_offset, out_count - out_offset, 460 "rx_pkts - \t%llu\n", qp->rx_pkts); 461 out_offset += snprintf(buf + out_offset, out_count - out_offset, 462 "rx_memcpy - \t%llu\n", qp->rx_memcpy); 463 out_offset += snprintf(buf + out_offset, out_count - out_offset, 464 "rx_async - \t%llu\n", qp->rx_async); 465 out_offset += snprintf(buf + out_offset, out_count - out_offset, 466 "rx_ring_empty - %llu\n", qp->rx_ring_empty); 467 out_offset += snprintf(buf + out_offset, out_count - out_offset, 468 "rx_err_no_buf - %llu\n", qp->rx_err_no_buf); 469 out_offset += snprintf(buf + out_offset, out_count - out_offset, 470 "rx_err_oflow - \t%llu\n", qp->rx_err_oflow); 471 out_offset += snprintf(buf + out_offset, out_count - out_offset, 472 "rx_err_ver - \t%llu\n", qp->rx_err_ver); 473 out_offset += snprintf(buf + out_offset, out_count - out_offset, 474 "rx_buff - \t0x%p\n", qp->rx_buff); 475 out_offset += snprintf(buf + out_offset, out_count - out_offset, 476 "rx_index - \t%u\n", qp->rx_index); 477 out_offset += snprintf(buf + out_offset, out_count - out_offset, 478 "rx_max_entry - \t%u\n\n", qp->rx_max_entry); 479 480 out_offset += snprintf(buf + out_offset, out_count - out_offset, 481 "tx_bytes - \t%llu\n", qp->tx_bytes); 482 out_offset += snprintf(buf + out_offset, out_count - out_offset, 483 "tx_pkts - \t%llu\n", qp->tx_pkts); 484 out_offset += snprintf(buf + out_offset, out_count - out_offset, 485 "tx_memcpy - \t%llu\n", qp->tx_memcpy); 486 out_offset += snprintf(buf + out_offset, out_count - out_offset, 487 "tx_async - \t%llu\n", qp->tx_async); 488 out_offset += snprintf(buf + out_offset, out_count - out_offset, 489 "tx_ring_full - \t%llu\n", qp->tx_ring_full); 490 out_offset += snprintf(buf + out_offset, out_count - out_offset, 491 "tx_err_no_buf - %llu\n", qp->tx_err_no_buf); 492 out_offset += snprintf(buf + out_offset, out_count - out_offset, 493 "tx_mw - \t0x%p\n", qp->tx_mw); 494 out_offset += snprintf(buf + out_offset, out_count - out_offset, 495 "tx_index (H) - \t%u\n", qp->tx_index); 496 out_offset += snprintf(buf + out_offset, out_count - out_offset, 497 "RRI (T) - \t%u\n", 498 qp->remote_rx_info->entry); 499 out_offset += snprintf(buf + out_offset, out_count - out_offset, 500 "tx_max_entry - \t%u\n", qp->tx_max_entry); 501 out_offset += snprintf(buf + out_offset, out_count - out_offset, 502 "free tx - \t%u\n", 503 ntb_transport_tx_free_entry(qp)); 504 505 out_offset += snprintf(buf + out_offset, out_count - out_offset, 506 "\n"); 507 out_offset += snprintf(buf + out_offset, out_count - out_offset, 508 "Using TX DMA - \t%s\n", 509 qp->tx_dma_chan ? "Yes" : "No"); 510 out_offset += snprintf(buf + out_offset, out_count - out_offset, 511 "Using RX DMA - \t%s\n", 512 qp->rx_dma_chan ? "Yes" : "No"); 513 out_offset += snprintf(buf + out_offset, out_count - out_offset, 514 "QP Link - \t%s\n", 515 qp->link_is_up ? "Up" : "Down"); 516 out_offset += snprintf(buf + out_offset, out_count - out_offset, 517 "\n"); 518 519 if (out_offset > out_count) 520 out_offset = out_count; 521 522 ret = simple_read_from_buffer(ubuf, count, offp, buf, out_offset); 523 kfree(buf); 524 return ret; 525 } 526 527 static const struct file_operations ntb_qp_debugfs_stats = { 528 .owner = THIS_MODULE, 529 .open = simple_open, 530 .read = debugfs_read, 531 }; 532 533 static void ntb_list_add(spinlock_t *lock, struct list_head *entry, 534 struct list_head *list) 535 { 536 unsigned long flags; 537 538 spin_lock_irqsave(lock, flags); 539 list_add_tail(entry, list); 540 spin_unlock_irqrestore(lock, flags); 541 } 542 543 static struct ntb_queue_entry *ntb_list_rm(spinlock_t *lock, 544 struct list_head *list) 545 { 546 struct ntb_queue_entry *entry; 547 unsigned long flags; 548 549 spin_lock_irqsave(lock, flags); 550 if (list_empty(list)) { 551 entry = NULL; 552 goto out; 553 } 554 entry = list_first_entry(list, struct ntb_queue_entry, entry); 555 list_del(&entry->entry); 556 557 out: 558 spin_unlock_irqrestore(lock, flags); 559 560 return entry; 561 } 562 563 static struct ntb_queue_entry *ntb_list_mv(spinlock_t *lock, 564 struct list_head *list, 565 struct list_head *to_list) 566 { 567 struct ntb_queue_entry *entry; 568 unsigned long flags; 569 570 spin_lock_irqsave(lock, flags); 571 572 if (list_empty(list)) { 573 entry = NULL; 574 } else { 575 entry = list_first_entry(list, struct ntb_queue_entry, entry); 576 list_move_tail(&entry->entry, to_list); 577 } 578 579 spin_unlock_irqrestore(lock, flags); 580 581 return entry; 582 } 583 584 static int ntb_transport_setup_qp_mw(struct ntb_transport_ctx *nt, 585 unsigned int qp_num) 586 { 587 struct ntb_transport_qp *qp = &nt->qp_vec[qp_num]; 588 struct ntb_transport_mw *mw; 589 unsigned int rx_size, num_qps_mw; 590 unsigned int mw_num, mw_count, qp_count; 591 unsigned int i; 592 593 mw_count = nt->mw_count; 594 qp_count = nt->qp_count; 595 596 mw_num = QP_TO_MW(nt, qp_num); 597 mw = &nt->mw_vec[mw_num]; 598 599 if (!mw->virt_addr) 600 return -ENOMEM; 601 602 if (qp_count % mw_count && mw_num + 1 < qp_count / mw_count) 603 num_qps_mw = qp_count / mw_count + 1; 604 else 605 num_qps_mw = qp_count / mw_count; 606 607 rx_size = (unsigned int)mw->xlat_size / num_qps_mw; 608 qp->rx_buff = mw->virt_addr + rx_size * qp_num / mw_count; 609 rx_size -= sizeof(struct ntb_rx_info); 610 611 qp->remote_rx_info = qp->rx_buff + rx_size; 612 613 /* Due to housekeeping, there must be atleast 2 buffs */ 614 qp->rx_max_frame = min(transport_mtu, rx_size / 2); 615 qp->rx_max_entry = rx_size / qp->rx_max_frame; 616 qp->rx_index = 0; 617 618 qp->remote_rx_info->entry = qp->rx_max_entry - 1; 619 620 /* setup the hdr offsets with 0's */ 621 for (i = 0; i < qp->rx_max_entry; i++) { 622 void *offset = (qp->rx_buff + qp->rx_max_frame * (i + 1) - 623 sizeof(struct ntb_payload_header)); 624 memset(offset, 0, sizeof(struct ntb_payload_header)); 625 } 626 627 qp->rx_pkts = 0; 628 qp->tx_pkts = 0; 629 qp->tx_index = 0; 630 631 return 0; 632 } 633 634 static void ntb_free_mw(struct ntb_transport_ctx *nt, int num_mw) 635 { 636 struct ntb_transport_mw *mw = &nt->mw_vec[num_mw]; 637 struct pci_dev *pdev = nt->ndev->pdev; 638 639 if (!mw->virt_addr) 640 return; 641 642 ntb_mw_clear_trans(nt->ndev, num_mw); 643 dma_free_coherent(&pdev->dev, mw->buff_size, 644 mw->virt_addr, mw->dma_addr); 645 mw->xlat_size = 0; 646 mw->buff_size = 0; 647 mw->virt_addr = NULL; 648 } 649 650 static int ntb_set_mw(struct ntb_transport_ctx *nt, int num_mw, 651 resource_size_t size) 652 { 653 struct ntb_transport_mw *mw = &nt->mw_vec[num_mw]; 654 struct pci_dev *pdev = nt->ndev->pdev; 655 size_t xlat_size, buff_size; 656 int rc; 657 658 if (!size) 659 return -EINVAL; 660 661 xlat_size = round_up(size, mw->xlat_align_size); 662 buff_size = round_up(size, mw->xlat_align); 663 664 /* No need to re-setup */ 665 if (mw->xlat_size == xlat_size) 666 return 0; 667 668 if (mw->buff_size) 669 ntb_free_mw(nt, num_mw); 670 671 /* Alloc memory for receiving data. Must be aligned */ 672 mw->xlat_size = xlat_size; 673 mw->buff_size = buff_size; 674 675 mw->virt_addr = dma_alloc_coherent(&pdev->dev, buff_size, 676 &mw->dma_addr, GFP_KERNEL); 677 if (!mw->virt_addr) { 678 mw->xlat_size = 0; 679 mw->buff_size = 0; 680 dev_err(&pdev->dev, "Unable to alloc MW buff of size %zu\n", 681 buff_size); 682 return -ENOMEM; 683 } 684 685 /* 686 * we must ensure that the memory address allocated is BAR size 687 * aligned in order for the XLAT register to take the value. This 688 * is a requirement of the hardware. It is recommended to setup CMA 689 * for BAR sizes equal or greater than 4MB. 690 */ 691 if (!IS_ALIGNED(mw->dma_addr, mw->xlat_align)) { 692 dev_err(&pdev->dev, "DMA memory %pad is not aligned\n", 693 &mw->dma_addr); 694 ntb_free_mw(nt, num_mw); 695 return -ENOMEM; 696 } 697 698 /* Notify HW the memory location of the receive buffer */ 699 rc = ntb_mw_set_trans(nt->ndev, num_mw, mw->dma_addr, mw->xlat_size); 700 if (rc) { 701 dev_err(&pdev->dev, "Unable to set mw%d translation", num_mw); 702 ntb_free_mw(nt, num_mw); 703 return -EIO; 704 } 705 706 return 0; 707 } 708 709 static void ntb_qp_link_down_reset(struct ntb_transport_qp *qp) 710 { 711 qp->link_is_up = false; 712 713 qp->tx_index = 0; 714 qp->rx_index = 0; 715 qp->rx_bytes = 0; 716 qp->rx_pkts = 0; 717 qp->rx_ring_empty = 0; 718 qp->rx_err_no_buf = 0; 719 qp->rx_err_oflow = 0; 720 qp->rx_err_ver = 0; 721 qp->rx_memcpy = 0; 722 qp->rx_async = 0; 723 qp->tx_bytes = 0; 724 qp->tx_pkts = 0; 725 qp->tx_ring_full = 0; 726 qp->tx_err_no_buf = 0; 727 qp->tx_memcpy = 0; 728 qp->tx_async = 0; 729 } 730 731 static void ntb_qp_link_cleanup(struct ntb_transport_qp *qp) 732 { 733 struct ntb_transport_ctx *nt = qp->transport; 734 struct pci_dev *pdev = nt->ndev->pdev; 735 736 dev_info(&pdev->dev, "qp %d: Link Cleanup\n", qp->qp_num); 737 738 cancel_delayed_work_sync(&qp->link_work); 739 ntb_qp_link_down_reset(qp); 740 741 if (qp->event_handler) 742 qp->event_handler(qp->cb_data, qp->link_is_up); 743 } 744 745 static void ntb_qp_link_cleanup_work(struct work_struct *work) 746 { 747 struct ntb_transport_qp *qp = container_of(work, 748 struct ntb_transport_qp, 749 link_cleanup); 750 struct ntb_transport_ctx *nt = qp->transport; 751 752 ntb_qp_link_cleanup(qp); 753 754 if (nt->link_is_up) 755 schedule_delayed_work(&qp->link_work, 756 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 757 } 758 759 static void ntb_qp_link_down(struct ntb_transport_qp *qp) 760 { 761 schedule_work(&qp->link_cleanup); 762 } 763 764 static void ntb_transport_link_cleanup(struct ntb_transport_ctx *nt) 765 { 766 struct ntb_transport_qp *qp; 767 u64 qp_bitmap_alloc; 768 int i; 769 770 qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free; 771 772 /* Pass along the info to any clients */ 773 for (i = 0; i < nt->qp_count; i++) 774 if (qp_bitmap_alloc & BIT_ULL(i)) { 775 qp = &nt->qp_vec[i]; 776 ntb_qp_link_cleanup(qp); 777 cancel_work_sync(&qp->link_cleanup); 778 cancel_delayed_work_sync(&qp->link_work); 779 } 780 781 if (!nt->link_is_up) 782 cancel_delayed_work_sync(&nt->link_work); 783 784 /* The scratchpad registers keep the values if the remote side 785 * goes down, blast them now to give them a sane value the next 786 * time they are accessed 787 */ 788 for (i = 0; i < MAX_SPAD; i++) 789 ntb_spad_write(nt->ndev, i, 0); 790 } 791 792 static void ntb_transport_link_cleanup_work(struct work_struct *work) 793 { 794 struct ntb_transport_ctx *nt = 795 container_of(work, struct ntb_transport_ctx, link_cleanup); 796 797 ntb_transport_link_cleanup(nt); 798 } 799 800 static void ntb_transport_event_callback(void *data) 801 { 802 struct ntb_transport_ctx *nt = data; 803 804 if (ntb_link_is_up(nt->ndev, NULL, NULL) == 1) 805 schedule_delayed_work(&nt->link_work, 0); 806 else 807 schedule_work(&nt->link_cleanup); 808 } 809 810 static void ntb_transport_link_work(struct work_struct *work) 811 { 812 struct ntb_transport_ctx *nt = 813 container_of(work, struct ntb_transport_ctx, link_work.work); 814 struct ntb_dev *ndev = nt->ndev; 815 struct pci_dev *pdev = ndev->pdev; 816 resource_size_t size; 817 u32 val; 818 int rc, i, spad; 819 820 /* send the local info, in the opposite order of the way we read it */ 821 for (i = 0; i < nt->mw_count; i++) { 822 size = nt->mw_vec[i].phys_size; 823 824 if (max_mw_size && size > max_mw_size) 825 size = max_mw_size; 826 827 spad = MW0_SZ_HIGH + (i * 2); 828 ntb_peer_spad_write(ndev, spad, (u32)(size >> 32)); 829 830 spad = MW0_SZ_LOW + (i * 2); 831 ntb_peer_spad_write(ndev, spad, (u32)size); 832 } 833 834 ntb_peer_spad_write(ndev, NUM_MWS, nt->mw_count); 835 836 ntb_peer_spad_write(ndev, NUM_QPS, nt->qp_count); 837 838 ntb_peer_spad_write(ndev, VERSION, NTB_TRANSPORT_VERSION); 839 840 /* Query the remote side for its info */ 841 val = ntb_spad_read(ndev, VERSION); 842 dev_dbg(&pdev->dev, "Remote version = %d\n", val); 843 if (val != NTB_TRANSPORT_VERSION) 844 goto out; 845 846 val = ntb_spad_read(ndev, NUM_QPS); 847 dev_dbg(&pdev->dev, "Remote max number of qps = %d\n", val); 848 if (val != nt->qp_count) 849 goto out; 850 851 val = ntb_spad_read(ndev, NUM_MWS); 852 dev_dbg(&pdev->dev, "Remote number of mws = %d\n", val); 853 if (val != nt->mw_count) 854 goto out; 855 856 for (i = 0; i < nt->mw_count; i++) { 857 u64 val64; 858 859 val = ntb_spad_read(ndev, MW0_SZ_HIGH + (i * 2)); 860 val64 = (u64)val << 32; 861 862 val = ntb_spad_read(ndev, MW0_SZ_LOW + (i * 2)); 863 val64 |= val; 864 865 dev_dbg(&pdev->dev, "Remote MW%d size = %#llx\n", i, val64); 866 867 rc = ntb_set_mw(nt, i, val64); 868 if (rc) 869 goto out1; 870 } 871 872 nt->link_is_up = true; 873 874 for (i = 0; i < nt->qp_count; i++) { 875 struct ntb_transport_qp *qp = &nt->qp_vec[i]; 876 877 ntb_transport_setup_qp_mw(nt, i); 878 879 if (qp->client_ready) 880 schedule_delayed_work(&qp->link_work, 0); 881 } 882 883 return; 884 885 out1: 886 for (i = 0; i < nt->mw_count; i++) 887 ntb_free_mw(nt, i); 888 out: 889 if (ntb_link_is_up(ndev, NULL, NULL) == 1) 890 schedule_delayed_work(&nt->link_work, 891 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 892 } 893 894 static void ntb_qp_link_work(struct work_struct *work) 895 { 896 struct ntb_transport_qp *qp = container_of(work, 897 struct ntb_transport_qp, 898 link_work.work); 899 struct pci_dev *pdev = qp->ndev->pdev; 900 struct ntb_transport_ctx *nt = qp->transport; 901 int val; 902 903 WARN_ON(!nt->link_is_up); 904 905 val = ntb_spad_read(nt->ndev, QP_LINKS); 906 907 ntb_peer_spad_write(nt->ndev, QP_LINKS, val | BIT(qp->qp_num)); 908 909 /* query remote spad for qp ready bits */ 910 ntb_peer_spad_read(nt->ndev, QP_LINKS); 911 dev_dbg_ratelimited(&pdev->dev, "Remote QP link status = %x\n", val); 912 913 /* See if the remote side is up */ 914 if (val & BIT(qp->qp_num)) { 915 dev_info(&pdev->dev, "qp %d: Link Up\n", qp->qp_num); 916 qp->link_is_up = true; 917 918 if (qp->event_handler) 919 qp->event_handler(qp->cb_data, qp->link_is_up); 920 921 tasklet_schedule(&qp->rxc_db_work); 922 } else if (nt->link_is_up) 923 schedule_delayed_work(&qp->link_work, 924 msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT)); 925 } 926 927 static int ntb_transport_init_queue(struct ntb_transport_ctx *nt, 928 unsigned int qp_num) 929 { 930 struct ntb_transport_qp *qp; 931 struct ntb_transport_mw *mw; 932 phys_addr_t mw_base; 933 resource_size_t mw_size; 934 unsigned int num_qps_mw, tx_size; 935 unsigned int mw_num, mw_count, qp_count; 936 u64 qp_offset; 937 938 mw_count = nt->mw_count; 939 qp_count = nt->qp_count; 940 941 mw_num = QP_TO_MW(nt, qp_num); 942 mw = &nt->mw_vec[mw_num]; 943 944 qp = &nt->qp_vec[qp_num]; 945 qp->qp_num = qp_num; 946 qp->transport = nt; 947 qp->ndev = nt->ndev; 948 qp->client_ready = false; 949 qp->event_handler = NULL; 950 ntb_qp_link_down_reset(qp); 951 952 if (qp_count % mw_count && mw_num + 1 < qp_count / mw_count) 953 num_qps_mw = qp_count / mw_count + 1; 954 else 955 num_qps_mw = qp_count / mw_count; 956 957 mw_base = nt->mw_vec[mw_num].phys_addr; 958 mw_size = nt->mw_vec[mw_num].phys_size; 959 960 tx_size = (unsigned int)mw_size / num_qps_mw; 961 qp_offset = tx_size * qp_num / mw_count; 962 963 qp->tx_mw = nt->mw_vec[mw_num].vbase + qp_offset; 964 if (!qp->tx_mw) 965 return -EINVAL; 966 967 qp->tx_mw_phys = mw_base + qp_offset; 968 if (!qp->tx_mw_phys) 969 return -EINVAL; 970 971 tx_size -= sizeof(struct ntb_rx_info); 972 qp->rx_info = qp->tx_mw + tx_size; 973 974 /* Due to housekeeping, there must be atleast 2 buffs */ 975 qp->tx_max_frame = min(transport_mtu, tx_size / 2); 976 qp->tx_max_entry = tx_size / qp->tx_max_frame; 977 978 if (nt->debugfs_node_dir) { 979 char debugfs_name[4]; 980 981 snprintf(debugfs_name, 4, "qp%d", qp_num); 982 qp->debugfs_dir = debugfs_create_dir(debugfs_name, 983 nt->debugfs_node_dir); 984 985 qp->debugfs_stats = debugfs_create_file("stats", S_IRUSR, 986 qp->debugfs_dir, qp, 987 &ntb_qp_debugfs_stats); 988 } else { 989 qp->debugfs_dir = NULL; 990 qp->debugfs_stats = NULL; 991 } 992 993 INIT_DELAYED_WORK(&qp->link_work, ntb_qp_link_work); 994 INIT_WORK(&qp->link_cleanup, ntb_qp_link_cleanup_work); 995 996 spin_lock_init(&qp->ntb_rx_q_lock); 997 spin_lock_init(&qp->ntb_tx_free_q_lock); 998 999 INIT_LIST_HEAD(&qp->rx_post_q); 1000 INIT_LIST_HEAD(&qp->rx_pend_q); 1001 INIT_LIST_HEAD(&qp->rx_free_q); 1002 INIT_LIST_HEAD(&qp->tx_free_q); 1003 1004 tasklet_init(&qp->rxc_db_work, ntb_transport_rxc_db, 1005 (unsigned long)qp); 1006 1007 return 0; 1008 } 1009 1010 static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev) 1011 { 1012 struct ntb_transport_ctx *nt; 1013 struct ntb_transport_mw *mw; 1014 unsigned int mw_count, qp_count; 1015 u64 qp_bitmap; 1016 int node; 1017 int rc, i; 1018 1019 if (ntb_db_is_unsafe(ndev)) 1020 dev_dbg(&ndev->dev, 1021 "doorbell is unsafe, proceed anyway...\n"); 1022 if (ntb_spad_is_unsafe(ndev)) 1023 dev_dbg(&ndev->dev, 1024 "scratchpad is unsafe, proceed anyway...\n"); 1025 1026 node = dev_to_node(&ndev->dev); 1027 1028 nt = kzalloc_node(sizeof(*nt), GFP_KERNEL, node); 1029 if (!nt) 1030 return -ENOMEM; 1031 1032 nt->ndev = ndev; 1033 1034 mw_count = ntb_mw_count(ndev); 1035 1036 nt->mw_count = mw_count; 1037 1038 nt->mw_vec = kzalloc_node(mw_count * sizeof(*nt->mw_vec), 1039 GFP_KERNEL, node); 1040 if (!nt->mw_vec) { 1041 rc = -ENOMEM; 1042 goto err; 1043 } 1044 1045 for (i = 0; i < mw_count; i++) { 1046 mw = &nt->mw_vec[i]; 1047 1048 rc = ntb_mw_get_range(ndev, i, &mw->phys_addr, &mw->phys_size, 1049 &mw->xlat_align, &mw->xlat_align_size); 1050 if (rc) 1051 goto err1; 1052 1053 mw->vbase = ioremap_wc(mw->phys_addr, mw->phys_size); 1054 if (!mw->vbase) { 1055 rc = -ENOMEM; 1056 goto err1; 1057 } 1058 1059 mw->buff_size = 0; 1060 mw->xlat_size = 0; 1061 mw->virt_addr = NULL; 1062 mw->dma_addr = 0; 1063 } 1064 1065 qp_bitmap = ntb_db_valid_mask(ndev); 1066 1067 qp_count = ilog2(qp_bitmap); 1068 if (max_num_clients && max_num_clients < qp_count) 1069 qp_count = max_num_clients; 1070 else if (mw_count < qp_count) 1071 qp_count = mw_count; 1072 1073 qp_bitmap &= BIT_ULL(qp_count) - 1; 1074 1075 nt->qp_count = qp_count; 1076 nt->qp_bitmap = qp_bitmap; 1077 nt->qp_bitmap_free = qp_bitmap; 1078 1079 nt->qp_vec = kzalloc_node(qp_count * sizeof(*nt->qp_vec), 1080 GFP_KERNEL, node); 1081 if (!nt->qp_vec) { 1082 rc = -ENOMEM; 1083 goto err2; 1084 } 1085 1086 if (nt_debugfs_dir) { 1087 nt->debugfs_node_dir = 1088 debugfs_create_dir(pci_name(ndev->pdev), 1089 nt_debugfs_dir); 1090 } 1091 1092 for (i = 0; i < qp_count; i++) { 1093 rc = ntb_transport_init_queue(nt, i); 1094 if (rc) 1095 goto err3; 1096 } 1097 1098 INIT_DELAYED_WORK(&nt->link_work, ntb_transport_link_work); 1099 INIT_WORK(&nt->link_cleanup, ntb_transport_link_cleanup_work); 1100 1101 rc = ntb_set_ctx(ndev, nt, &ntb_transport_ops); 1102 if (rc) 1103 goto err3; 1104 1105 INIT_LIST_HEAD(&nt->client_devs); 1106 rc = ntb_bus_init(nt); 1107 if (rc) 1108 goto err4; 1109 1110 nt->link_is_up = false; 1111 ntb_link_enable(ndev, NTB_SPEED_AUTO, NTB_WIDTH_AUTO); 1112 ntb_link_event(ndev); 1113 1114 return 0; 1115 1116 err4: 1117 ntb_clear_ctx(ndev); 1118 err3: 1119 kfree(nt->qp_vec); 1120 err2: 1121 kfree(nt->mw_vec); 1122 err1: 1123 while (i--) { 1124 mw = &nt->mw_vec[i]; 1125 iounmap(mw->vbase); 1126 } 1127 err: 1128 kfree(nt); 1129 return rc; 1130 } 1131 1132 static void ntb_transport_free(struct ntb_client *self, struct ntb_dev *ndev) 1133 { 1134 struct ntb_transport_ctx *nt = ndev->ctx; 1135 struct ntb_transport_qp *qp; 1136 u64 qp_bitmap_alloc; 1137 int i; 1138 1139 ntb_transport_link_cleanup(nt); 1140 cancel_work_sync(&nt->link_cleanup); 1141 cancel_delayed_work_sync(&nt->link_work); 1142 1143 qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free; 1144 1145 /* verify that all the qp's are freed */ 1146 for (i = 0; i < nt->qp_count; i++) { 1147 qp = &nt->qp_vec[i]; 1148 if (qp_bitmap_alloc & BIT_ULL(i)) 1149 ntb_transport_free_queue(qp); 1150 debugfs_remove_recursive(qp->debugfs_dir); 1151 } 1152 1153 ntb_link_disable(ndev); 1154 ntb_clear_ctx(ndev); 1155 1156 ntb_bus_remove(nt); 1157 1158 for (i = nt->mw_count; i--; ) { 1159 ntb_free_mw(nt, i); 1160 iounmap(nt->mw_vec[i].vbase); 1161 } 1162 1163 kfree(nt->qp_vec); 1164 kfree(nt->mw_vec); 1165 kfree(nt); 1166 } 1167 1168 static void ntb_complete_rxc(struct ntb_transport_qp *qp) 1169 { 1170 struct ntb_queue_entry *entry; 1171 void *cb_data; 1172 unsigned int len; 1173 unsigned long irqflags; 1174 1175 spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags); 1176 1177 while (!list_empty(&qp->rx_post_q)) { 1178 entry = list_first_entry(&qp->rx_post_q, 1179 struct ntb_queue_entry, entry); 1180 if (!(entry->flags & DESC_DONE_FLAG)) 1181 break; 1182 1183 entry->rx_hdr->flags = 0; 1184 iowrite32(entry->index, &qp->rx_info->entry); 1185 1186 cb_data = entry->cb_data; 1187 len = entry->len; 1188 1189 list_move_tail(&entry->entry, &qp->rx_free_q); 1190 1191 spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags); 1192 1193 if (qp->rx_handler && qp->client_ready) 1194 qp->rx_handler(qp, qp->cb_data, cb_data, len); 1195 1196 spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags); 1197 } 1198 1199 spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags); 1200 } 1201 1202 static void ntb_rx_copy_callback(void *data) 1203 { 1204 struct ntb_queue_entry *entry = data; 1205 1206 entry->flags |= DESC_DONE_FLAG; 1207 1208 ntb_complete_rxc(entry->qp); 1209 } 1210 1211 static void ntb_memcpy_rx(struct ntb_queue_entry *entry, void *offset) 1212 { 1213 void *buf = entry->buf; 1214 size_t len = entry->len; 1215 1216 memcpy(buf, offset, len); 1217 1218 /* Ensure that the data is fully copied out before clearing the flag */ 1219 wmb(); 1220 1221 ntb_rx_copy_callback(entry); 1222 } 1223 1224 static void ntb_async_rx(struct ntb_queue_entry *entry, void *offset) 1225 { 1226 struct dma_async_tx_descriptor *txd; 1227 struct ntb_transport_qp *qp = entry->qp; 1228 struct dma_chan *chan = qp->rx_dma_chan; 1229 struct dma_device *device; 1230 size_t pay_off, buff_off, len; 1231 struct dmaengine_unmap_data *unmap; 1232 dma_cookie_t cookie; 1233 void *buf = entry->buf; 1234 1235 len = entry->len; 1236 1237 if (!chan) 1238 goto err; 1239 1240 if (len < copy_bytes) 1241 goto err; 1242 1243 device = chan->device; 1244 pay_off = (size_t)offset & ~PAGE_MASK; 1245 buff_off = (size_t)buf & ~PAGE_MASK; 1246 1247 if (!is_dma_copy_aligned(device, pay_off, buff_off, len)) 1248 goto err; 1249 1250 unmap = dmaengine_get_unmap_data(device->dev, 2, GFP_NOWAIT); 1251 if (!unmap) 1252 goto err; 1253 1254 unmap->len = len; 1255 unmap->addr[0] = dma_map_page(device->dev, virt_to_page(offset), 1256 pay_off, len, DMA_TO_DEVICE); 1257 if (dma_mapping_error(device->dev, unmap->addr[0])) 1258 goto err_get_unmap; 1259 1260 unmap->to_cnt = 1; 1261 1262 unmap->addr[1] = dma_map_page(device->dev, virt_to_page(buf), 1263 buff_off, len, DMA_FROM_DEVICE); 1264 if (dma_mapping_error(device->dev, unmap->addr[1])) 1265 goto err_get_unmap; 1266 1267 unmap->from_cnt = 1; 1268 1269 txd = device->device_prep_dma_memcpy(chan, unmap->addr[1], 1270 unmap->addr[0], len, 1271 DMA_PREP_INTERRUPT); 1272 if (!txd) 1273 goto err_get_unmap; 1274 1275 txd->callback = ntb_rx_copy_callback; 1276 txd->callback_param = entry; 1277 dma_set_unmap(txd, unmap); 1278 1279 cookie = dmaengine_submit(txd); 1280 if (dma_submit_error(cookie)) 1281 goto err_set_unmap; 1282 1283 dmaengine_unmap_put(unmap); 1284 1285 qp->last_cookie = cookie; 1286 1287 qp->rx_async++; 1288 1289 return; 1290 1291 err_set_unmap: 1292 dmaengine_unmap_put(unmap); 1293 err_get_unmap: 1294 dmaengine_unmap_put(unmap); 1295 err: 1296 ntb_memcpy_rx(entry, offset); 1297 qp->rx_memcpy++; 1298 } 1299 1300 static int ntb_process_rxc(struct ntb_transport_qp *qp) 1301 { 1302 struct ntb_payload_header *hdr; 1303 struct ntb_queue_entry *entry; 1304 void *offset; 1305 1306 offset = qp->rx_buff + qp->rx_max_frame * qp->rx_index; 1307 hdr = offset + qp->rx_max_frame - sizeof(struct ntb_payload_header); 1308 1309 dev_dbg(&qp->ndev->pdev->dev, "qp %d: RX ver %u len %d flags %x\n", 1310 qp->qp_num, hdr->ver, hdr->len, hdr->flags); 1311 1312 if (!(hdr->flags & DESC_DONE_FLAG)) { 1313 dev_dbg(&qp->ndev->pdev->dev, "done flag not set\n"); 1314 qp->rx_ring_empty++; 1315 return -EAGAIN; 1316 } 1317 1318 if (hdr->flags & LINK_DOWN_FLAG) { 1319 dev_dbg(&qp->ndev->pdev->dev, "link down flag set\n"); 1320 ntb_qp_link_down(qp); 1321 hdr->flags = 0; 1322 return -EAGAIN; 1323 } 1324 1325 if (hdr->ver != (u32)qp->rx_pkts) { 1326 dev_dbg(&qp->ndev->pdev->dev, 1327 "version mismatch, expected %llu - got %u\n", 1328 qp->rx_pkts, hdr->ver); 1329 qp->rx_err_ver++; 1330 return -EIO; 1331 } 1332 1333 entry = ntb_list_mv(&qp->ntb_rx_q_lock, &qp->rx_pend_q, &qp->rx_post_q); 1334 if (!entry) { 1335 dev_dbg(&qp->ndev->pdev->dev, "no receive buffer\n"); 1336 qp->rx_err_no_buf++; 1337 return -EAGAIN; 1338 } 1339 1340 entry->rx_hdr = hdr; 1341 entry->index = qp->rx_index; 1342 1343 if (hdr->len > entry->len) { 1344 dev_dbg(&qp->ndev->pdev->dev, 1345 "receive buffer overflow! Wanted %d got %d\n", 1346 hdr->len, entry->len); 1347 qp->rx_err_oflow++; 1348 1349 entry->len = -EIO; 1350 entry->flags |= DESC_DONE_FLAG; 1351 1352 ntb_complete_rxc(qp); 1353 } else { 1354 dev_dbg(&qp->ndev->pdev->dev, 1355 "RX OK index %u ver %u size %d into buf size %d\n", 1356 qp->rx_index, hdr->ver, hdr->len, entry->len); 1357 1358 qp->rx_bytes += hdr->len; 1359 qp->rx_pkts++; 1360 1361 entry->len = hdr->len; 1362 1363 ntb_async_rx(entry, offset); 1364 } 1365 1366 qp->rx_index++; 1367 qp->rx_index %= qp->rx_max_entry; 1368 1369 return 0; 1370 } 1371 1372 static void ntb_transport_rxc_db(unsigned long data) 1373 { 1374 struct ntb_transport_qp *qp = (void *)data; 1375 int rc, i; 1376 1377 dev_dbg(&qp->ndev->pdev->dev, "%s: doorbell %d received\n", 1378 __func__, qp->qp_num); 1379 1380 /* Limit the number of packets processed in a single interrupt to 1381 * provide fairness to others 1382 */ 1383 for (i = 0; i < qp->rx_max_entry; i++) { 1384 rc = ntb_process_rxc(qp); 1385 if (rc) 1386 break; 1387 } 1388 1389 if (i && qp->rx_dma_chan) 1390 dma_async_issue_pending(qp->rx_dma_chan); 1391 1392 if (i == qp->rx_max_entry) { 1393 /* there is more work to do */ 1394 tasklet_schedule(&qp->rxc_db_work); 1395 } else if (ntb_db_read(qp->ndev) & BIT_ULL(qp->qp_num)) { 1396 /* the doorbell bit is set: clear it */ 1397 ntb_db_clear(qp->ndev, BIT_ULL(qp->qp_num)); 1398 /* ntb_db_read ensures ntb_db_clear write is committed */ 1399 ntb_db_read(qp->ndev); 1400 1401 /* an interrupt may have arrived between finishing 1402 * ntb_process_rxc and clearing the doorbell bit: 1403 * there might be some more work to do. 1404 */ 1405 tasklet_schedule(&qp->rxc_db_work); 1406 } 1407 } 1408 1409 static void ntb_tx_copy_callback(void *data) 1410 { 1411 struct ntb_queue_entry *entry = data; 1412 struct ntb_transport_qp *qp = entry->qp; 1413 struct ntb_payload_header __iomem *hdr = entry->tx_hdr; 1414 1415 iowrite32(entry->flags | DESC_DONE_FLAG, &hdr->flags); 1416 1417 ntb_peer_db_set(qp->ndev, BIT_ULL(qp->qp_num)); 1418 1419 /* The entry length can only be zero if the packet is intended to be a 1420 * "link down" or similar. Since no payload is being sent in these 1421 * cases, there is nothing to add to the completion queue. 1422 */ 1423 if (entry->len > 0) { 1424 qp->tx_bytes += entry->len; 1425 1426 if (qp->tx_handler) 1427 qp->tx_handler(qp, qp->cb_data, entry->cb_data, 1428 entry->len); 1429 } 1430 1431 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, &qp->tx_free_q); 1432 } 1433 1434 static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset) 1435 { 1436 #ifdef ARCH_HAS_NOCACHE_UACCESS 1437 /* 1438 * Using non-temporal mov to improve performance on non-cached 1439 * writes, even though we aren't actually copying from user space. 1440 */ 1441 __copy_from_user_inatomic_nocache(offset, entry->buf, entry->len); 1442 #else 1443 memcpy_toio(offset, entry->buf, entry->len); 1444 #endif 1445 1446 /* Ensure that the data is fully copied out before setting the flags */ 1447 wmb(); 1448 1449 ntb_tx_copy_callback(entry); 1450 } 1451 1452 static void ntb_async_tx(struct ntb_transport_qp *qp, 1453 struct ntb_queue_entry *entry) 1454 { 1455 struct ntb_payload_header __iomem *hdr; 1456 struct dma_async_tx_descriptor *txd; 1457 struct dma_chan *chan = qp->tx_dma_chan; 1458 struct dma_device *device; 1459 size_t dest_off, buff_off; 1460 struct dmaengine_unmap_data *unmap; 1461 dma_addr_t dest; 1462 dma_cookie_t cookie; 1463 void __iomem *offset; 1464 size_t len = entry->len; 1465 void *buf = entry->buf; 1466 1467 offset = qp->tx_mw + qp->tx_max_frame * qp->tx_index; 1468 hdr = offset + qp->tx_max_frame - sizeof(struct ntb_payload_header); 1469 entry->tx_hdr = hdr; 1470 1471 iowrite32(entry->len, &hdr->len); 1472 iowrite32((u32)qp->tx_pkts, &hdr->ver); 1473 1474 if (!chan) 1475 goto err; 1476 1477 if (len < copy_bytes) 1478 goto err; 1479 1480 device = chan->device; 1481 dest = qp->tx_mw_phys + qp->tx_max_frame * qp->tx_index; 1482 buff_off = (size_t)buf & ~PAGE_MASK; 1483 dest_off = (size_t)dest & ~PAGE_MASK; 1484 1485 if (!is_dma_copy_aligned(device, buff_off, dest_off, len)) 1486 goto err; 1487 1488 unmap = dmaengine_get_unmap_data(device->dev, 1, GFP_NOWAIT); 1489 if (!unmap) 1490 goto err; 1491 1492 unmap->len = len; 1493 unmap->addr[0] = dma_map_page(device->dev, virt_to_page(buf), 1494 buff_off, len, DMA_TO_DEVICE); 1495 if (dma_mapping_error(device->dev, unmap->addr[0])) 1496 goto err_get_unmap; 1497 1498 unmap->to_cnt = 1; 1499 1500 txd = device->device_prep_dma_memcpy(chan, dest, unmap->addr[0], len, 1501 DMA_PREP_INTERRUPT); 1502 if (!txd) 1503 goto err_get_unmap; 1504 1505 txd->callback = ntb_tx_copy_callback; 1506 txd->callback_param = entry; 1507 dma_set_unmap(txd, unmap); 1508 1509 cookie = dmaengine_submit(txd); 1510 if (dma_submit_error(cookie)) 1511 goto err_set_unmap; 1512 1513 dmaengine_unmap_put(unmap); 1514 1515 dma_async_issue_pending(chan); 1516 qp->tx_async++; 1517 1518 return; 1519 err_set_unmap: 1520 dmaengine_unmap_put(unmap); 1521 err_get_unmap: 1522 dmaengine_unmap_put(unmap); 1523 err: 1524 ntb_memcpy_tx(entry, offset); 1525 qp->tx_memcpy++; 1526 } 1527 1528 static int ntb_process_tx(struct ntb_transport_qp *qp, 1529 struct ntb_queue_entry *entry) 1530 { 1531 if (qp->tx_index == qp->remote_rx_info->entry) { 1532 qp->tx_ring_full++; 1533 return -EAGAIN; 1534 } 1535 1536 if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) { 1537 if (qp->tx_handler) 1538 qp->tx_handler(qp->cb_data, qp, NULL, -EIO); 1539 1540 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1541 &qp->tx_free_q); 1542 return 0; 1543 } 1544 1545 ntb_async_tx(qp, entry); 1546 1547 qp->tx_index++; 1548 qp->tx_index %= qp->tx_max_entry; 1549 1550 qp->tx_pkts++; 1551 1552 return 0; 1553 } 1554 1555 static void ntb_send_link_down(struct ntb_transport_qp *qp) 1556 { 1557 struct pci_dev *pdev = qp->ndev->pdev; 1558 struct ntb_queue_entry *entry; 1559 int i, rc; 1560 1561 if (!qp->link_is_up) 1562 return; 1563 1564 dev_info(&pdev->dev, "qp %d: Send Link Down\n", qp->qp_num); 1565 1566 for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) { 1567 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); 1568 if (entry) 1569 break; 1570 msleep(100); 1571 } 1572 1573 if (!entry) 1574 return; 1575 1576 entry->cb_data = NULL; 1577 entry->buf = NULL; 1578 entry->len = 0; 1579 entry->flags = LINK_DOWN_FLAG; 1580 1581 rc = ntb_process_tx(qp, entry); 1582 if (rc) 1583 dev_err(&pdev->dev, "ntb: QP%d unable to send linkdown msg\n", 1584 qp->qp_num); 1585 1586 ntb_qp_link_down_reset(qp); 1587 } 1588 1589 static bool ntb_dma_filter_fn(struct dma_chan *chan, void *node) 1590 { 1591 return dev_to_node(&chan->dev->device) == (int)(unsigned long)node; 1592 } 1593 1594 /** 1595 * ntb_transport_create_queue - Create a new NTB transport layer queue 1596 * @rx_handler: receive callback function 1597 * @tx_handler: transmit callback function 1598 * @event_handler: event callback function 1599 * 1600 * Create a new NTB transport layer queue and provide the queue with a callback 1601 * routine for both transmit and receive. The receive callback routine will be 1602 * used to pass up data when the transport has received it on the queue. The 1603 * transmit callback routine will be called when the transport has completed the 1604 * transmission of the data on the queue and the data is ready to be freed. 1605 * 1606 * RETURNS: pointer to newly created ntb_queue, NULL on error. 1607 */ 1608 struct ntb_transport_qp * 1609 ntb_transport_create_queue(void *data, struct device *client_dev, 1610 const struct ntb_queue_handlers *handlers) 1611 { 1612 struct ntb_dev *ndev; 1613 struct pci_dev *pdev; 1614 struct ntb_transport_ctx *nt; 1615 struct ntb_queue_entry *entry; 1616 struct ntb_transport_qp *qp; 1617 u64 qp_bit; 1618 unsigned int free_queue; 1619 dma_cap_mask_t dma_mask; 1620 int node; 1621 int i; 1622 1623 ndev = dev_ntb(client_dev->parent); 1624 pdev = ndev->pdev; 1625 nt = ndev->ctx; 1626 1627 node = dev_to_node(&ndev->dev); 1628 1629 free_queue = ffs(nt->qp_bitmap); 1630 if (!free_queue) 1631 goto err; 1632 1633 /* decrement free_queue to make it zero based */ 1634 free_queue--; 1635 1636 qp = &nt->qp_vec[free_queue]; 1637 qp_bit = BIT_ULL(qp->qp_num); 1638 1639 nt->qp_bitmap_free &= ~qp_bit; 1640 1641 qp->cb_data = data; 1642 qp->rx_handler = handlers->rx_handler; 1643 qp->tx_handler = handlers->tx_handler; 1644 qp->event_handler = handlers->event_handler; 1645 1646 dma_cap_zero(dma_mask); 1647 dma_cap_set(DMA_MEMCPY, dma_mask); 1648 1649 if (use_dma) { 1650 qp->tx_dma_chan = 1651 dma_request_channel(dma_mask, ntb_dma_filter_fn, 1652 (void *)(unsigned long)node); 1653 if (!qp->tx_dma_chan) 1654 dev_info(&pdev->dev, "Unable to allocate TX DMA channel\n"); 1655 1656 qp->rx_dma_chan = 1657 dma_request_channel(dma_mask, ntb_dma_filter_fn, 1658 (void *)(unsigned long)node); 1659 if (!qp->rx_dma_chan) 1660 dev_info(&pdev->dev, "Unable to allocate RX DMA channel\n"); 1661 } else { 1662 qp->tx_dma_chan = NULL; 1663 qp->rx_dma_chan = NULL; 1664 } 1665 1666 dev_dbg(&pdev->dev, "Using %s memcpy for TX\n", 1667 qp->tx_dma_chan ? "DMA" : "CPU"); 1668 1669 dev_dbg(&pdev->dev, "Using %s memcpy for RX\n", 1670 qp->rx_dma_chan ? "DMA" : "CPU"); 1671 1672 for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) { 1673 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 1674 if (!entry) 1675 goto err1; 1676 1677 entry->qp = qp; 1678 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, 1679 &qp->rx_free_q); 1680 } 1681 1682 for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) { 1683 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node); 1684 if (!entry) 1685 goto err2; 1686 1687 entry->qp = qp; 1688 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1689 &qp->tx_free_q); 1690 } 1691 1692 ntb_db_clear(qp->ndev, qp_bit); 1693 ntb_db_clear_mask(qp->ndev, qp_bit); 1694 1695 dev_info(&pdev->dev, "NTB Transport QP %d created\n", qp->qp_num); 1696 1697 return qp; 1698 1699 err2: 1700 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) 1701 kfree(entry); 1702 err1: 1703 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q))) 1704 kfree(entry); 1705 if (qp->tx_dma_chan) 1706 dma_release_channel(qp->tx_dma_chan); 1707 if (qp->rx_dma_chan) 1708 dma_release_channel(qp->rx_dma_chan); 1709 nt->qp_bitmap_free |= qp_bit; 1710 err: 1711 return NULL; 1712 } 1713 EXPORT_SYMBOL_GPL(ntb_transport_create_queue); 1714 1715 /** 1716 * ntb_transport_free_queue - Frees NTB transport queue 1717 * @qp: NTB queue to be freed 1718 * 1719 * Frees NTB transport queue 1720 */ 1721 void ntb_transport_free_queue(struct ntb_transport_qp *qp) 1722 { 1723 struct pci_dev *pdev; 1724 struct ntb_queue_entry *entry; 1725 u64 qp_bit; 1726 1727 if (!qp) 1728 return; 1729 1730 pdev = qp->ndev->pdev; 1731 1732 if (qp->tx_dma_chan) { 1733 struct dma_chan *chan = qp->tx_dma_chan; 1734 /* Putting the dma_chan to NULL will force any new traffic to be 1735 * processed by the CPU instead of the DAM engine 1736 */ 1737 qp->tx_dma_chan = NULL; 1738 1739 /* Try to be nice and wait for any queued DMA engine 1740 * transactions to process before smashing it with a rock 1741 */ 1742 dma_sync_wait(chan, qp->last_cookie); 1743 dmaengine_terminate_all(chan); 1744 dma_release_channel(chan); 1745 } 1746 1747 if (qp->rx_dma_chan) { 1748 struct dma_chan *chan = qp->rx_dma_chan; 1749 /* Putting the dma_chan to NULL will force any new traffic to be 1750 * processed by the CPU instead of the DAM engine 1751 */ 1752 qp->rx_dma_chan = NULL; 1753 1754 /* Try to be nice and wait for any queued DMA engine 1755 * transactions to process before smashing it with a rock 1756 */ 1757 dma_sync_wait(chan, qp->last_cookie); 1758 dmaengine_terminate_all(chan); 1759 dma_release_channel(chan); 1760 } 1761 1762 qp_bit = BIT_ULL(qp->qp_num); 1763 1764 ntb_db_set_mask(qp->ndev, qp_bit); 1765 tasklet_disable(&qp->rxc_db_work); 1766 1767 cancel_delayed_work_sync(&qp->link_work); 1768 1769 qp->cb_data = NULL; 1770 qp->rx_handler = NULL; 1771 qp->tx_handler = NULL; 1772 qp->event_handler = NULL; 1773 1774 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q))) 1775 kfree(entry); 1776 1777 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q))) { 1778 dev_warn(&pdev->dev, "Freeing item from non-empty rx_pend_q\n"); 1779 kfree(entry); 1780 } 1781 1782 while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_post_q))) { 1783 dev_warn(&pdev->dev, "Freeing item from non-empty rx_post_q\n"); 1784 kfree(entry); 1785 } 1786 1787 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q))) 1788 kfree(entry); 1789 1790 qp->transport->qp_bitmap_free |= qp_bit; 1791 1792 dev_info(&pdev->dev, "NTB Transport QP %d freed\n", qp->qp_num); 1793 } 1794 EXPORT_SYMBOL_GPL(ntb_transport_free_queue); 1795 1796 /** 1797 * ntb_transport_rx_remove - Dequeues enqueued rx packet 1798 * @qp: NTB queue to be freed 1799 * @len: pointer to variable to write enqueued buffers length 1800 * 1801 * Dequeues unused buffers from receive queue. Should only be used during 1802 * shutdown of qp. 1803 * 1804 * RETURNS: NULL error value on error, or void* for success. 1805 */ 1806 void *ntb_transport_rx_remove(struct ntb_transport_qp *qp, unsigned int *len) 1807 { 1808 struct ntb_queue_entry *entry; 1809 void *buf; 1810 1811 if (!qp || qp->client_ready) 1812 return NULL; 1813 1814 entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q); 1815 if (!entry) 1816 return NULL; 1817 1818 buf = entry->cb_data; 1819 *len = entry->len; 1820 1821 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_free_q); 1822 1823 return buf; 1824 } 1825 EXPORT_SYMBOL_GPL(ntb_transport_rx_remove); 1826 1827 /** 1828 * ntb_transport_rx_enqueue - Enqueue a new NTB queue entry 1829 * @qp: NTB transport layer queue the entry is to be enqueued on 1830 * @cb: per buffer pointer for callback function to use 1831 * @data: pointer to data buffer that incoming packets will be copied into 1832 * @len: length of the data buffer 1833 * 1834 * Enqueue a new receive buffer onto the transport queue into which a NTB 1835 * payload can be received into. 1836 * 1837 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 1838 */ 1839 int ntb_transport_rx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, 1840 unsigned int len) 1841 { 1842 struct ntb_queue_entry *entry; 1843 1844 if (!qp) 1845 return -EINVAL; 1846 1847 entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q); 1848 if (!entry) 1849 return -ENOMEM; 1850 1851 entry->cb_data = cb; 1852 entry->buf = data; 1853 entry->len = len; 1854 entry->flags = 0; 1855 1856 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_pend_q); 1857 1858 tasklet_schedule(&qp->rxc_db_work); 1859 1860 return 0; 1861 } 1862 EXPORT_SYMBOL_GPL(ntb_transport_rx_enqueue); 1863 1864 /** 1865 * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry 1866 * @qp: NTB transport layer queue the entry is to be enqueued on 1867 * @cb: per buffer pointer for callback function to use 1868 * @data: pointer to data buffer that will be sent 1869 * @len: length of the data buffer 1870 * 1871 * Enqueue a new transmit buffer onto the transport queue from which a NTB 1872 * payload will be transmitted. This assumes that a lock is being held to 1873 * serialize access to the qp. 1874 * 1875 * RETURNS: An appropriate -ERRNO error value on error, or zero for success. 1876 */ 1877 int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data, 1878 unsigned int len) 1879 { 1880 struct ntb_queue_entry *entry; 1881 int rc; 1882 1883 if (!qp || !qp->link_is_up || !len) 1884 return -EINVAL; 1885 1886 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q); 1887 if (!entry) { 1888 qp->tx_err_no_buf++; 1889 return -EBUSY; 1890 } 1891 1892 entry->cb_data = cb; 1893 entry->buf = data; 1894 entry->len = len; 1895 entry->flags = 0; 1896 1897 rc = ntb_process_tx(qp, entry); 1898 if (rc) 1899 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, 1900 &qp->tx_free_q); 1901 1902 return rc; 1903 } 1904 EXPORT_SYMBOL_GPL(ntb_transport_tx_enqueue); 1905 1906 /** 1907 * ntb_transport_link_up - Notify NTB transport of client readiness to use queue 1908 * @qp: NTB transport layer queue to be enabled 1909 * 1910 * Notify NTB transport layer of client readiness to use queue 1911 */ 1912 void ntb_transport_link_up(struct ntb_transport_qp *qp) 1913 { 1914 if (!qp) 1915 return; 1916 1917 qp->client_ready = true; 1918 1919 if (qp->transport->link_is_up) 1920 schedule_delayed_work(&qp->link_work, 0); 1921 } 1922 EXPORT_SYMBOL_GPL(ntb_transport_link_up); 1923 1924 /** 1925 * ntb_transport_link_down - Notify NTB transport to no longer enqueue data 1926 * @qp: NTB transport layer queue to be disabled 1927 * 1928 * Notify NTB transport layer of client's desire to no longer receive data on 1929 * transport queue specified. It is the client's responsibility to ensure all 1930 * entries on queue are purged or otherwise handled appropriately. 1931 */ 1932 void ntb_transport_link_down(struct ntb_transport_qp *qp) 1933 { 1934 struct pci_dev *pdev; 1935 int val; 1936 1937 if (!qp) 1938 return; 1939 1940 pdev = qp->ndev->pdev; 1941 qp->client_ready = false; 1942 1943 val = ntb_spad_read(qp->ndev, QP_LINKS); 1944 1945 ntb_peer_spad_write(qp->ndev, QP_LINKS, 1946 val & ~BIT(qp->qp_num)); 1947 1948 if (qp->link_is_up) 1949 ntb_send_link_down(qp); 1950 else 1951 cancel_delayed_work_sync(&qp->link_work); 1952 } 1953 EXPORT_SYMBOL_GPL(ntb_transport_link_down); 1954 1955 /** 1956 * ntb_transport_link_query - Query transport link state 1957 * @qp: NTB transport layer queue to be queried 1958 * 1959 * Query connectivity to the remote system of the NTB transport queue 1960 * 1961 * RETURNS: true for link up or false for link down 1962 */ 1963 bool ntb_transport_link_query(struct ntb_transport_qp *qp) 1964 { 1965 if (!qp) 1966 return false; 1967 1968 return qp->link_is_up; 1969 } 1970 EXPORT_SYMBOL_GPL(ntb_transport_link_query); 1971 1972 /** 1973 * ntb_transport_qp_num - Query the qp number 1974 * @qp: NTB transport layer queue to be queried 1975 * 1976 * Query qp number of the NTB transport queue 1977 * 1978 * RETURNS: a zero based number specifying the qp number 1979 */ 1980 unsigned char ntb_transport_qp_num(struct ntb_transport_qp *qp) 1981 { 1982 if (!qp) 1983 return 0; 1984 1985 return qp->qp_num; 1986 } 1987 EXPORT_SYMBOL_GPL(ntb_transport_qp_num); 1988 1989 /** 1990 * ntb_transport_max_size - Query the max payload size of a qp 1991 * @qp: NTB transport layer queue to be queried 1992 * 1993 * Query the maximum payload size permissible on the given qp 1994 * 1995 * RETURNS: the max payload size of a qp 1996 */ 1997 unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp) 1998 { 1999 unsigned int max; 2000 unsigned int copy_align; 2001 2002 if (!qp) 2003 return 0; 2004 2005 if (!qp->tx_dma_chan && !qp->rx_dma_chan) 2006 return qp->tx_max_frame - sizeof(struct ntb_payload_header); 2007 2008 copy_align = max(qp->tx_dma_chan->device->copy_align, 2009 qp->rx_dma_chan->device->copy_align); 2010 2011 /* If DMA engine usage is possible, try to find the max size for that */ 2012 max = qp->tx_max_frame - sizeof(struct ntb_payload_header); 2013 max -= max % (1 << copy_align); 2014 2015 return max; 2016 } 2017 EXPORT_SYMBOL_GPL(ntb_transport_max_size); 2018 2019 unsigned int ntb_transport_tx_free_entry(struct ntb_transport_qp *qp) 2020 { 2021 unsigned int head = qp->tx_index; 2022 unsigned int tail = qp->remote_rx_info->entry; 2023 2024 return tail > head ? tail - head : qp->tx_max_entry + tail - head; 2025 } 2026 EXPORT_SYMBOL_GPL(ntb_transport_tx_free_entry); 2027 2028 static void ntb_transport_doorbell_callback(void *data, int vector) 2029 { 2030 struct ntb_transport_ctx *nt = data; 2031 struct ntb_transport_qp *qp; 2032 u64 db_bits; 2033 unsigned int qp_num; 2034 2035 db_bits = (nt->qp_bitmap & ~nt->qp_bitmap_free & 2036 ntb_db_vector_mask(nt->ndev, vector)); 2037 2038 while (db_bits) { 2039 qp_num = __ffs(db_bits); 2040 qp = &nt->qp_vec[qp_num]; 2041 2042 tasklet_schedule(&qp->rxc_db_work); 2043 2044 db_bits &= ~BIT_ULL(qp_num); 2045 } 2046 } 2047 2048 static const struct ntb_ctx_ops ntb_transport_ops = { 2049 .link_event = ntb_transport_event_callback, 2050 .db_event = ntb_transport_doorbell_callback, 2051 }; 2052 2053 static struct ntb_client ntb_transport_client = { 2054 .ops = { 2055 .probe = ntb_transport_probe, 2056 .remove = ntb_transport_free, 2057 }, 2058 }; 2059 2060 static int __init ntb_transport_init(void) 2061 { 2062 int rc; 2063 2064 pr_info("%s, version %s\n", NTB_TRANSPORT_DESC, NTB_TRANSPORT_VER); 2065 2066 if (debugfs_initialized()) 2067 nt_debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL); 2068 2069 rc = bus_register(&ntb_transport_bus); 2070 if (rc) 2071 goto err_bus; 2072 2073 rc = ntb_register_client(&ntb_transport_client); 2074 if (rc) 2075 goto err_client; 2076 2077 return 0; 2078 2079 err_client: 2080 bus_unregister(&ntb_transport_bus); 2081 err_bus: 2082 debugfs_remove_recursive(nt_debugfs_dir); 2083 return rc; 2084 } 2085 module_init(ntb_transport_init); 2086 2087 static void __exit ntb_transport_exit(void) 2088 { 2089 debugfs_remove_recursive(nt_debugfs_dir); 2090 2091 ntb_unregister_client(&ntb_transport_client); 2092 bus_unregister(&ntb_transport_bus); 2093 } 2094 module_exit(ntb_transport_exit); 2095