xref: /openbmc/linux/drivers/ntb/ntb_transport.c (revision 089a49b6)
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  *
9  *   This program is free software; you can redistribute it and/or modify
10  *   it under the terms of version 2 of the GNU General Public License as
11  *   published by the Free Software Foundation.
12  *
13  *   BSD LICENSE
14  *
15  *   Copyright(c) 2012 Intel Corporation. All rights reserved.
16  *
17  *   Redistribution and use in source and binary forms, with or without
18  *   modification, are permitted provided that the following conditions
19  *   are met:
20  *
21  *     * Redistributions of source code must retain the above copyright
22  *       notice, this list of conditions and the following disclaimer.
23  *     * Redistributions in binary form must reproduce the above copy
24  *       notice, this list of conditions and the following disclaimer in
25  *       the documentation and/or other materials provided with the
26  *       distribution.
27  *     * Neither the name of Intel Corporation nor the names of its
28  *       contributors may be used to endorse or promote products derived
29  *       from this software without specific prior written permission.
30  *
31  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
32  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
33  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
34  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
35  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
36  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
37  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
38  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
39  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
40  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
41  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
42  *
43  * Intel PCIe NTB Linux driver
44  *
45  * Contact Information:
46  * Jon Mason <jon.mason@intel.com>
47  */
48 #include <linux/debugfs.h>
49 #include <linux/delay.h>
50 #include <linux/dmaengine.h>
51 #include <linux/dma-mapping.h>
52 #include <linux/errno.h>
53 #include <linux/export.h>
54 #include <linux/interrupt.h>
55 #include <linux/module.h>
56 #include <linux/pci.h>
57 #include <linux/slab.h>
58 #include <linux/types.h>
59 #include <linux/ntb.h>
60 #include "ntb_hw.h"
61 
62 #define NTB_TRANSPORT_VERSION	3
63 
64 static unsigned int transport_mtu = 0x401E;
65 module_param(transport_mtu, uint, 0644);
66 MODULE_PARM_DESC(transport_mtu, "Maximum size of NTB transport packets");
67 
68 static unsigned char max_num_clients;
69 module_param(max_num_clients, byte, 0644);
70 MODULE_PARM_DESC(max_num_clients, "Maximum number of NTB transport clients");
71 
72 static unsigned int copy_bytes = 1024;
73 module_param(copy_bytes, uint, 0644);
74 MODULE_PARM_DESC(copy_bytes, "Threshold under which NTB will use the CPU to copy instead of DMA");
75 
76 struct ntb_queue_entry {
77 	/* ntb_queue list reference */
78 	struct list_head entry;
79 	/* pointers to data to be transfered */
80 	void *cb_data;
81 	void *buf;
82 	unsigned int len;
83 	unsigned int flags;
84 
85 	struct ntb_transport_qp *qp;
86 	union {
87 		struct ntb_payload_header __iomem *tx_hdr;
88 		struct ntb_payload_header *rx_hdr;
89 	};
90 	unsigned int index;
91 };
92 
93 struct ntb_rx_info {
94 	unsigned int entry;
95 };
96 
97 struct ntb_transport_qp {
98 	struct ntb_transport *transport;
99 	struct ntb_device *ndev;
100 	void *cb_data;
101 	struct dma_chan *dma_chan;
102 
103 	bool client_ready;
104 	bool qp_link;
105 	u8 qp_num;	/* Only 64 QP's are allowed.  0-63 */
106 
107 	struct ntb_rx_info __iomem *rx_info;
108 	struct ntb_rx_info *remote_rx_info;
109 
110 	void (*tx_handler) (struct ntb_transport_qp *qp, void *qp_data,
111 			    void *data, int len);
112 	struct list_head tx_free_q;
113 	spinlock_t ntb_tx_free_q_lock;
114 	void __iomem *tx_mw;
115 	dma_addr_t tx_mw_phys;
116 	unsigned int tx_index;
117 	unsigned int tx_max_entry;
118 	unsigned int tx_max_frame;
119 
120 	void (*rx_handler) (struct ntb_transport_qp *qp, void *qp_data,
121 			    void *data, int len);
122 	struct tasklet_struct rx_work;
123 	struct list_head rx_pend_q;
124 	struct list_head rx_free_q;
125 	spinlock_t ntb_rx_pend_q_lock;
126 	spinlock_t ntb_rx_free_q_lock;
127 	void *rx_buff;
128 	unsigned int rx_index;
129 	unsigned int rx_max_entry;
130 	unsigned int rx_max_frame;
131 	dma_cookie_t last_cookie;
132 
133 	void (*event_handler) (void *data, int status);
134 	struct delayed_work link_work;
135 	struct work_struct link_cleanup;
136 
137 	struct dentry *debugfs_dir;
138 	struct dentry *debugfs_stats;
139 
140 	/* Stats */
141 	u64 rx_bytes;
142 	u64 rx_pkts;
143 	u64 rx_ring_empty;
144 	u64 rx_err_no_buf;
145 	u64 rx_err_oflow;
146 	u64 rx_err_ver;
147 	u64 rx_memcpy;
148 	u64 rx_async;
149 	u64 tx_bytes;
150 	u64 tx_pkts;
151 	u64 tx_ring_full;
152 	u64 tx_err_no_buf;
153 	u64 tx_memcpy;
154 	u64 tx_async;
155 };
156 
157 struct ntb_transport_mw {
158 	size_t size;
159 	void *virt_addr;
160 	dma_addr_t dma_addr;
161 };
162 
163 struct ntb_transport_client_dev {
164 	struct list_head entry;
165 	struct device dev;
166 };
167 
168 struct ntb_transport {
169 	struct list_head entry;
170 	struct list_head client_devs;
171 
172 	struct ntb_device *ndev;
173 	struct ntb_transport_mw *mw;
174 	struct ntb_transport_qp *qps;
175 	unsigned int max_qps;
176 	unsigned long qp_bitmap;
177 	bool transport_link;
178 	struct delayed_work link_work;
179 	struct work_struct link_cleanup;
180 };
181 
182 enum {
183 	DESC_DONE_FLAG = 1 << 0,
184 	LINK_DOWN_FLAG = 1 << 1,
185 };
186 
187 struct ntb_payload_header {
188 	unsigned int ver;
189 	unsigned int len;
190 	unsigned int flags;
191 };
192 
193 enum {
194 	VERSION = 0,
195 	QP_LINKS,
196 	NUM_QPS,
197 	NUM_MWS,
198 	MW0_SZ_HIGH,
199 	MW0_SZ_LOW,
200 	MW1_SZ_HIGH,
201 	MW1_SZ_LOW,
202 	MAX_SPAD,
203 };
204 
205 #define QP_TO_MW(ndev, qp)	((qp) % ntb_max_mw(ndev))
206 #define NTB_QP_DEF_NUM_ENTRIES	100
207 #define NTB_LINK_DOWN_TIMEOUT	10
208 
209 static int ntb_match_bus(struct device *dev, struct device_driver *drv)
210 {
211 	return !strncmp(dev_name(dev), drv->name, strlen(drv->name));
212 }
213 
214 static int ntb_client_probe(struct device *dev)
215 {
216 	const struct ntb_client *drv = container_of(dev->driver,
217 						    struct ntb_client, driver);
218 	struct pci_dev *pdev = container_of(dev->parent, struct pci_dev, dev);
219 	int rc = -EINVAL;
220 
221 	get_device(dev);
222 	if (drv && drv->probe)
223 		rc = drv->probe(pdev);
224 	if (rc)
225 		put_device(dev);
226 
227 	return rc;
228 }
229 
230 static int ntb_client_remove(struct device *dev)
231 {
232 	const struct ntb_client *drv = container_of(dev->driver,
233 						    struct ntb_client, driver);
234 	struct pci_dev *pdev = container_of(dev->parent, struct pci_dev, dev);
235 
236 	if (drv && drv->remove)
237 		drv->remove(pdev);
238 
239 	put_device(dev);
240 
241 	return 0;
242 }
243 
244 static struct bus_type ntb_bus_type = {
245 	.name = "ntb_bus",
246 	.match = ntb_match_bus,
247 	.probe = ntb_client_probe,
248 	.remove = ntb_client_remove,
249 };
250 
251 static LIST_HEAD(ntb_transport_list);
252 
253 static int ntb_bus_init(struct ntb_transport *nt)
254 {
255 	if (list_empty(&ntb_transport_list)) {
256 		int rc = bus_register(&ntb_bus_type);
257 		if (rc)
258 			return rc;
259 	}
260 
261 	list_add(&nt->entry, &ntb_transport_list);
262 
263 	return 0;
264 }
265 
266 static void ntb_bus_remove(struct ntb_transport *nt)
267 {
268 	struct ntb_transport_client_dev *client_dev, *cd;
269 
270 	list_for_each_entry_safe(client_dev, cd, &nt->client_devs, entry) {
271 		dev_err(client_dev->dev.parent, "%s still attached to bus, removing\n",
272 			dev_name(&client_dev->dev));
273 		list_del(&client_dev->entry);
274 		device_unregister(&client_dev->dev);
275 	}
276 
277 	list_del(&nt->entry);
278 
279 	if (list_empty(&ntb_transport_list))
280 		bus_unregister(&ntb_bus_type);
281 }
282 
283 static void ntb_client_release(struct device *dev)
284 {
285 	struct ntb_transport_client_dev *client_dev;
286 	client_dev = container_of(dev, struct ntb_transport_client_dev, dev);
287 
288 	kfree(client_dev);
289 }
290 
291 /**
292  * ntb_unregister_client_dev - Unregister NTB client device
293  * @device_name: Name of NTB client device
294  *
295  * Unregister an NTB client device with the NTB transport layer
296  */
297 void ntb_unregister_client_dev(char *device_name)
298 {
299 	struct ntb_transport_client_dev *client, *cd;
300 	struct ntb_transport *nt;
301 
302 	list_for_each_entry(nt, &ntb_transport_list, entry)
303 		list_for_each_entry_safe(client, cd, &nt->client_devs, entry)
304 			if (!strncmp(dev_name(&client->dev), device_name,
305 				     strlen(device_name))) {
306 				list_del(&client->entry);
307 				device_unregister(&client->dev);
308 			}
309 }
310 EXPORT_SYMBOL_GPL(ntb_unregister_client_dev);
311 
312 /**
313  * ntb_register_client_dev - Register NTB client device
314  * @device_name: Name of NTB client device
315  *
316  * Register an NTB client device with the NTB transport layer
317  */
318 int ntb_register_client_dev(char *device_name)
319 {
320 	struct ntb_transport_client_dev *client_dev;
321 	struct ntb_transport *nt;
322 	int rc, i = 0;
323 
324 	if (list_empty(&ntb_transport_list))
325 		return -ENODEV;
326 
327 	list_for_each_entry(nt, &ntb_transport_list, entry) {
328 		struct device *dev;
329 
330 		client_dev = kzalloc(sizeof(struct ntb_transport_client_dev),
331 				     GFP_KERNEL);
332 		if (!client_dev) {
333 			rc = -ENOMEM;
334 			goto err;
335 		}
336 
337 		dev = &client_dev->dev;
338 
339 		/* setup and register client devices */
340 		dev_set_name(dev, "%s%d", device_name, i);
341 		dev->bus = &ntb_bus_type;
342 		dev->release = ntb_client_release;
343 		dev->parent = &ntb_query_pdev(nt->ndev)->dev;
344 
345 		rc = device_register(dev);
346 		if (rc) {
347 			kfree(client_dev);
348 			goto err;
349 		}
350 
351 		list_add_tail(&client_dev->entry, &nt->client_devs);
352 		i++;
353 	}
354 
355 	return 0;
356 
357 err:
358 	ntb_unregister_client_dev(device_name);
359 
360 	return rc;
361 }
362 EXPORT_SYMBOL_GPL(ntb_register_client_dev);
363 
364 /**
365  * ntb_register_client - Register NTB client driver
366  * @drv: NTB client driver to be registered
367  *
368  * Register an NTB client driver with the NTB transport layer
369  *
370  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
371  */
372 int ntb_register_client(struct ntb_client *drv)
373 {
374 	drv->driver.bus = &ntb_bus_type;
375 
376 	if (list_empty(&ntb_transport_list))
377 		return -ENODEV;
378 
379 	return driver_register(&drv->driver);
380 }
381 EXPORT_SYMBOL_GPL(ntb_register_client);
382 
383 /**
384  * ntb_unregister_client - Unregister NTB client driver
385  * @drv: NTB client driver to be unregistered
386  *
387  * Unregister an NTB client driver with the NTB transport layer
388  *
389  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
390  */
391 void ntb_unregister_client(struct ntb_client *drv)
392 {
393 	driver_unregister(&drv->driver);
394 }
395 EXPORT_SYMBOL_GPL(ntb_unregister_client);
396 
397 static ssize_t debugfs_read(struct file *filp, char __user *ubuf, size_t count,
398 			    loff_t *offp)
399 {
400 	struct ntb_transport_qp *qp;
401 	char *buf;
402 	ssize_t ret, out_offset, out_count;
403 
404 	out_count = 1000;
405 
406 	buf = kmalloc(out_count, GFP_KERNEL);
407 	if (!buf)
408 		return -ENOMEM;
409 
410 	qp = filp->private_data;
411 	out_offset = 0;
412 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
413 			       "NTB QP stats\n");
414 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
415 			       "rx_bytes - \t%llu\n", qp->rx_bytes);
416 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
417 			       "rx_pkts - \t%llu\n", qp->rx_pkts);
418 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
419 			       "rx_memcpy - \t%llu\n", qp->rx_memcpy);
420 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
421 			       "rx_async - \t%llu\n", qp->rx_async);
422 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
423 			       "rx_ring_empty - %llu\n", qp->rx_ring_empty);
424 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
425 			       "rx_err_no_buf - %llu\n", qp->rx_err_no_buf);
426 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
427 			       "rx_err_oflow - \t%llu\n", qp->rx_err_oflow);
428 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
429 			       "rx_err_ver - \t%llu\n", qp->rx_err_ver);
430 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
431 			       "rx_buff - \t%p\n", qp->rx_buff);
432 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
433 			       "rx_index - \t%u\n", qp->rx_index);
434 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
435 			       "rx_max_entry - \t%u\n", qp->rx_max_entry);
436 
437 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
438 			       "tx_bytes - \t%llu\n", qp->tx_bytes);
439 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
440 			       "tx_pkts - \t%llu\n", qp->tx_pkts);
441 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
442 			       "tx_memcpy - \t%llu\n", qp->tx_memcpy);
443 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
444 			       "tx_async - \t%llu\n", qp->tx_async);
445 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
446 			       "tx_ring_full - \t%llu\n", qp->tx_ring_full);
447 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
448 			       "tx_err_no_buf - %llu\n", qp->tx_err_no_buf);
449 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
450 			       "tx_mw - \t%p\n", qp->tx_mw);
451 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
452 			       "tx_index - \t%u\n", qp->tx_index);
453 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
454 			       "tx_max_entry - \t%u\n", qp->tx_max_entry);
455 
456 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
457 			       "\nQP Link %s\n", (qp->qp_link == NTB_LINK_UP) ?
458 			       "Up" : "Down");
459 	if (out_offset > out_count)
460 		out_offset = out_count;
461 
462 	ret = simple_read_from_buffer(ubuf, count, offp, buf, out_offset);
463 	kfree(buf);
464 	return ret;
465 }
466 
467 static const struct file_operations ntb_qp_debugfs_stats = {
468 	.owner = THIS_MODULE,
469 	.open = simple_open,
470 	.read = debugfs_read,
471 };
472 
473 static void ntb_list_add(spinlock_t *lock, struct list_head *entry,
474 			 struct list_head *list)
475 {
476 	unsigned long flags;
477 
478 	spin_lock_irqsave(lock, flags);
479 	list_add_tail(entry, list);
480 	spin_unlock_irqrestore(lock, flags);
481 }
482 
483 static struct ntb_queue_entry *ntb_list_rm(spinlock_t *lock,
484 						struct list_head *list)
485 {
486 	struct ntb_queue_entry *entry;
487 	unsigned long flags;
488 
489 	spin_lock_irqsave(lock, flags);
490 	if (list_empty(list)) {
491 		entry = NULL;
492 		goto out;
493 	}
494 	entry = list_first_entry(list, struct ntb_queue_entry, entry);
495 	list_del(&entry->entry);
496 out:
497 	spin_unlock_irqrestore(lock, flags);
498 
499 	return entry;
500 }
501 
502 static void ntb_transport_setup_qp_mw(struct ntb_transport *nt,
503 				      unsigned int qp_num)
504 {
505 	struct ntb_transport_qp *qp = &nt->qps[qp_num];
506 	unsigned int rx_size, num_qps_mw;
507 	u8 mw_num, mw_max;
508 	unsigned int i;
509 
510 	mw_max = ntb_max_mw(nt->ndev);
511 	mw_num = QP_TO_MW(nt->ndev, qp_num);
512 
513 	WARN_ON(nt->mw[mw_num].virt_addr == NULL);
514 
515 	if (nt->max_qps % mw_max && mw_num < nt->max_qps % mw_max)
516 		num_qps_mw = nt->max_qps / mw_max + 1;
517 	else
518 		num_qps_mw = nt->max_qps / mw_max;
519 
520 	rx_size = (unsigned int) nt->mw[mw_num].size / num_qps_mw;
521 	qp->rx_buff = nt->mw[mw_num].virt_addr + qp_num / mw_max * rx_size;
522 	rx_size -= sizeof(struct ntb_rx_info);
523 
524 	qp->remote_rx_info = qp->rx_buff + rx_size;
525 
526 	/* Due to housekeeping, there must be atleast 2 buffs */
527 	qp->rx_max_frame = min(transport_mtu, rx_size / 2);
528 	qp->rx_max_entry = rx_size / qp->rx_max_frame;
529 	qp->rx_index = 0;
530 
531 	qp->remote_rx_info->entry = qp->rx_max_entry - 1;
532 
533 	/* setup the hdr offsets with 0's */
534 	for (i = 0; i < qp->rx_max_entry; i++) {
535 		void *offset = qp->rx_buff + qp->rx_max_frame * (i + 1) -
536 			       sizeof(struct ntb_payload_header);
537 		memset(offset, 0, sizeof(struct ntb_payload_header));
538 	}
539 
540 	qp->rx_pkts = 0;
541 	qp->tx_pkts = 0;
542 	qp->tx_index = 0;
543 }
544 
545 static void ntb_free_mw(struct ntb_transport *nt, int num_mw)
546 {
547 	struct ntb_transport_mw *mw = &nt->mw[num_mw];
548 	struct pci_dev *pdev = ntb_query_pdev(nt->ndev);
549 
550 	if (!mw->virt_addr)
551 		return;
552 
553 	dma_free_coherent(&pdev->dev, mw->size, mw->virt_addr, mw->dma_addr);
554 	mw->virt_addr = NULL;
555 }
556 
557 static int ntb_set_mw(struct ntb_transport *nt, int num_mw, unsigned int size)
558 {
559 	struct ntb_transport_mw *mw = &nt->mw[num_mw];
560 	struct pci_dev *pdev = ntb_query_pdev(nt->ndev);
561 
562 	/* No need to re-setup */
563 	if (mw->size == ALIGN(size, 4096))
564 		return 0;
565 
566 	if (mw->size != 0)
567 		ntb_free_mw(nt, num_mw);
568 
569 	/* Alloc memory for receiving data.  Must be 4k aligned */
570 	mw->size = ALIGN(size, 4096);
571 
572 	mw->virt_addr = dma_alloc_coherent(&pdev->dev, mw->size, &mw->dma_addr,
573 					   GFP_KERNEL);
574 	if (!mw->virt_addr) {
575 		mw->size = 0;
576 		dev_err(&pdev->dev, "Unable to allocate MW buffer of size %d\n",
577 		       (int) mw->size);
578 		return -ENOMEM;
579 	}
580 
581 	/* Notify HW the memory location of the receive buffer */
582 	ntb_set_mw_addr(nt->ndev, num_mw, mw->dma_addr);
583 
584 	return 0;
585 }
586 
587 static void ntb_qp_link_cleanup(struct work_struct *work)
588 {
589 	struct ntb_transport_qp *qp = container_of(work,
590 						   struct ntb_transport_qp,
591 						   link_cleanup);
592 	struct ntb_transport *nt = qp->transport;
593 	struct pci_dev *pdev = ntb_query_pdev(nt->ndev);
594 
595 	if (qp->qp_link == NTB_LINK_DOWN) {
596 		cancel_delayed_work_sync(&qp->link_work);
597 		return;
598 	}
599 
600 	if (qp->event_handler)
601 		qp->event_handler(qp->cb_data, NTB_LINK_DOWN);
602 
603 	dev_info(&pdev->dev, "qp %d: Link Down\n", qp->qp_num);
604 	qp->qp_link = NTB_LINK_DOWN;
605 
606 	if (nt->transport_link == NTB_LINK_UP)
607 		schedule_delayed_work(&qp->link_work,
608 				      msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT));
609 }
610 
611 static void ntb_qp_link_down(struct ntb_transport_qp *qp)
612 {
613 	schedule_work(&qp->link_cleanup);
614 }
615 
616 static void ntb_transport_link_cleanup(struct work_struct *work)
617 {
618 	struct ntb_transport *nt = container_of(work, struct ntb_transport,
619 						link_cleanup);
620 	int i;
621 
622 	if (nt->transport_link == NTB_LINK_DOWN)
623 		cancel_delayed_work_sync(&nt->link_work);
624 	else
625 		nt->transport_link = NTB_LINK_DOWN;
626 
627 	/* Pass along the info to any clients */
628 	for (i = 0; i < nt->max_qps; i++)
629 		if (!test_bit(i, &nt->qp_bitmap))
630 			ntb_qp_link_down(&nt->qps[i]);
631 
632 	/* The scratchpad registers keep the values if the remote side
633 	 * goes down, blast them now to give them a sane value the next
634 	 * time they are accessed
635 	 */
636 	for (i = 0; i < MAX_SPAD; i++)
637 		ntb_write_local_spad(nt->ndev, i, 0);
638 }
639 
640 static void ntb_transport_event_callback(void *data, enum ntb_hw_event event)
641 {
642 	struct ntb_transport *nt = data;
643 
644 	switch (event) {
645 	case NTB_EVENT_HW_LINK_UP:
646 		schedule_delayed_work(&nt->link_work, 0);
647 		break;
648 	case NTB_EVENT_HW_LINK_DOWN:
649 		schedule_work(&nt->link_cleanup);
650 		break;
651 	default:
652 		BUG();
653 	}
654 }
655 
656 static void ntb_transport_link_work(struct work_struct *work)
657 {
658 	struct ntb_transport *nt = container_of(work, struct ntb_transport,
659 						link_work.work);
660 	struct ntb_device *ndev = nt->ndev;
661 	struct pci_dev *pdev = ntb_query_pdev(ndev);
662 	u32 val;
663 	int rc, i;
664 
665 	/* send the local info, in the opposite order of the way we read it */
666 	for (i = 0; i < ntb_max_mw(ndev); i++) {
667 		rc = ntb_write_remote_spad(ndev, MW0_SZ_HIGH + (i * 2),
668 					   ntb_get_mw_size(ndev, i) >> 32);
669 		if (rc) {
670 			dev_err(&pdev->dev, "Error writing %u to remote spad %d\n",
671 				(u32)(ntb_get_mw_size(ndev, i) >> 32),
672 				MW0_SZ_HIGH + (i * 2));
673 			goto out;
674 		}
675 
676 		rc = ntb_write_remote_spad(ndev, MW0_SZ_LOW + (i * 2),
677 					   (u32) ntb_get_mw_size(ndev, i));
678 		if (rc) {
679 			dev_err(&pdev->dev, "Error writing %u to remote spad %d\n",
680 				(u32) ntb_get_mw_size(ndev, i),
681 				MW0_SZ_LOW + (i * 2));
682 			goto out;
683 		}
684 	}
685 
686 	rc = ntb_write_remote_spad(ndev, NUM_MWS, ntb_max_mw(ndev));
687 	if (rc) {
688 		dev_err(&pdev->dev, "Error writing %x to remote spad %d\n",
689 			ntb_max_mw(ndev), NUM_MWS);
690 		goto out;
691 	}
692 
693 	rc = ntb_write_remote_spad(ndev, NUM_QPS, nt->max_qps);
694 	if (rc) {
695 		dev_err(&pdev->dev, "Error writing %x to remote spad %d\n",
696 			nt->max_qps, NUM_QPS);
697 		goto out;
698 	}
699 
700 	rc = ntb_write_remote_spad(ndev, VERSION, NTB_TRANSPORT_VERSION);
701 	if (rc) {
702 		dev_err(&pdev->dev, "Error writing %x to remote spad %d\n",
703 			NTB_TRANSPORT_VERSION, VERSION);
704 		goto out;
705 	}
706 
707 	/* Query the remote side for its info */
708 	rc = ntb_read_remote_spad(ndev, VERSION, &val);
709 	if (rc) {
710 		dev_err(&pdev->dev, "Error reading remote spad %d\n", VERSION);
711 		goto out;
712 	}
713 
714 	if (val != NTB_TRANSPORT_VERSION)
715 		goto out;
716 	dev_dbg(&pdev->dev, "Remote version = %d\n", val);
717 
718 	rc = ntb_read_remote_spad(ndev, NUM_QPS, &val);
719 	if (rc) {
720 		dev_err(&pdev->dev, "Error reading remote spad %d\n", NUM_QPS);
721 		goto out;
722 	}
723 
724 	if (val != nt->max_qps)
725 		goto out;
726 	dev_dbg(&pdev->dev, "Remote max number of qps = %d\n", val);
727 
728 	rc = ntb_read_remote_spad(ndev, NUM_MWS, &val);
729 	if (rc) {
730 		dev_err(&pdev->dev, "Error reading remote spad %d\n", NUM_MWS);
731 		goto out;
732 	}
733 
734 	if (val != ntb_max_mw(ndev))
735 		goto out;
736 	dev_dbg(&pdev->dev, "Remote number of mws = %d\n", val);
737 
738 	for (i = 0; i < ntb_max_mw(ndev); i++) {
739 		u64 val64;
740 
741 		rc = ntb_read_remote_spad(ndev, MW0_SZ_HIGH + (i * 2), &val);
742 		if (rc) {
743 			dev_err(&pdev->dev, "Error reading remote spad %d\n",
744 				MW0_SZ_HIGH + (i * 2));
745 			goto out1;
746 		}
747 
748 		val64 = (u64) val << 32;
749 
750 		rc = ntb_read_remote_spad(ndev, MW0_SZ_LOW + (i * 2), &val);
751 		if (rc) {
752 			dev_err(&pdev->dev, "Error reading remote spad %d\n",
753 				MW0_SZ_LOW + (i * 2));
754 			goto out1;
755 		}
756 
757 		val64 |= val;
758 
759 		dev_dbg(&pdev->dev, "Remote MW%d size = %llu\n", i, val64);
760 
761 		rc = ntb_set_mw(nt, i, val64);
762 		if (rc)
763 			goto out1;
764 	}
765 
766 	nt->transport_link = NTB_LINK_UP;
767 
768 	for (i = 0; i < nt->max_qps; i++) {
769 		struct ntb_transport_qp *qp = &nt->qps[i];
770 
771 		ntb_transport_setup_qp_mw(nt, i);
772 
773 		if (qp->client_ready == NTB_LINK_UP)
774 			schedule_delayed_work(&qp->link_work, 0);
775 	}
776 
777 	return;
778 
779 out1:
780 	for (i = 0; i < ntb_max_mw(ndev); i++)
781 		ntb_free_mw(nt, i);
782 out:
783 	if (ntb_hw_link_status(ndev))
784 		schedule_delayed_work(&nt->link_work,
785 				      msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT));
786 }
787 
788 static void ntb_qp_link_work(struct work_struct *work)
789 {
790 	struct ntb_transport_qp *qp = container_of(work,
791 						   struct ntb_transport_qp,
792 						   link_work.work);
793 	struct pci_dev *pdev = ntb_query_pdev(qp->ndev);
794 	struct ntb_transport *nt = qp->transport;
795 	int rc, val;
796 
797 	WARN_ON(nt->transport_link != NTB_LINK_UP);
798 
799 	rc = ntb_read_local_spad(nt->ndev, QP_LINKS, &val);
800 	if (rc) {
801 		dev_err(&pdev->dev, "Error reading spad %d\n", QP_LINKS);
802 		return;
803 	}
804 
805 	rc = ntb_write_remote_spad(nt->ndev, QP_LINKS, val | 1 << qp->qp_num);
806 	if (rc)
807 		dev_err(&pdev->dev, "Error writing %x to remote spad %d\n",
808 			val | 1 << qp->qp_num, QP_LINKS);
809 
810 	/* query remote spad for qp ready bits */
811 	rc = ntb_read_remote_spad(nt->ndev, QP_LINKS, &val);
812 	if (rc)
813 		dev_err(&pdev->dev, "Error reading remote spad %d\n", QP_LINKS);
814 
815 	dev_dbg(&pdev->dev, "Remote QP link status = %x\n", val);
816 
817 	/* See if the remote side is up */
818 	if (1 << qp->qp_num & val) {
819 		qp->qp_link = NTB_LINK_UP;
820 
821 		dev_info(&pdev->dev, "qp %d: Link Up\n", qp->qp_num);
822 		if (qp->event_handler)
823 			qp->event_handler(qp->cb_data, NTB_LINK_UP);
824 	} else if (nt->transport_link == NTB_LINK_UP)
825 		schedule_delayed_work(&qp->link_work,
826 				      msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT));
827 }
828 
829 static int ntb_transport_init_queue(struct ntb_transport *nt,
830 				     unsigned int qp_num)
831 {
832 	struct ntb_transport_qp *qp;
833 	unsigned int num_qps_mw, tx_size;
834 	u8 mw_num, mw_max;
835 	u64 qp_offset;
836 
837 	mw_max = ntb_max_mw(nt->ndev);
838 	mw_num = QP_TO_MW(nt->ndev, qp_num);
839 
840 	qp = &nt->qps[qp_num];
841 	qp->qp_num = qp_num;
842 	qp->transport = nt;
843 	qp->ndev = nt->ndev;
844 	qp->qp_link = NTB_LINK_DOWN;
845 	qp->client_ready = NTB_LINK_DOWN;
846 	qp->event_handler = NULL;
847 
848 	if (nt->max_qps % mw_max && mw_num < nt->max_qps % mw_max)
849 		num_qps_mw = nt->max_qps / mw_max + 1;
850 	else
851 		num_qps_mw = nt->max_qps / mw_max;
852 
853 	tx_size = (unsigned int) ntb_get_mw_size(qp->ndev, mw_num) / num_qps_mw;
854 	qp_offset = qp_num / mw_max * tx_size;
855 	qp->tx_mw = ntb_get_mw_vbase(nt->ndev, mw_num) + qp_offset;
856 	if (!qp->tx_mw)
857 		return -EINVAL;
858 
859 	qp->tx_mw_phys = ntb_get_mw_base(qp->ndev, mw_num) + qp_offset;
860 	if (!qp->tx_mw_phys)
861 		return -EINVAL;
862 
863 	tx_size -= sizeof(struct ntb_rx_info);
864 	qp->rx_info = qp->tx_mw + tx_size;
865 
866 	/* Due to housekeeping, there must be atleast 2 buffs */
867 	qp->tx_max_frame = min(transport_mtu, tx_size / 2);
868 	qp->tx_max_entry = tx_size / qp->tx_max_frame;
869 
870 	if (ntb_query_debugfs(nt->ndev)) {
871 		char debugfs_name[4];
872 
873 		snprintf(debugfs_name, 4, "qp%d", qp_num);
874 		qp->debugfs_dir = debugfs_create_dir(debugfs_name,
875 						 ntb_query_debugfs(nt->ndev));
876 
877 		qp->debugfs_stats = debugfs_create_file("stats", S_IRUSR,
878 							qp->debugfs_dir, qp,
879 							&ntb_qp_debugfs_stats);
880 	}
881 
882 	INIT_DELAYED_WORK(&qp->link_work, ntb_qp_link_work);
883 	INIT_WORK(&qp->link_cleanup, ntb_qp_link_cleanup);
884 
885 	spin_lock_init(&qp->ntb_rx_pend_q_lock);
886 	spin_lock_init(&qp->ntb_rx_free_q_lock);
887 	spin_lock_init(&qp->ntb_tx_free_q_lock);
888 
889 	INIT_LIST_HEAD(&qp->rx_pend_q);
890 	INIT_LIST_HEAD(&qp->rx_free_q);
891 	INIT_LIST_HEAD(&qp->tx_free_q);
892 
893 	return 0;
894 }
895 
896 int ntb_transport_init(struct pci_dev *pdev)
897 {
898 	struct ntb_transport *nt;
899 	int rc, i;
900 
901 	nt = kzalloc(sizeof(struct ntb_transport), GFP_KERNEL);
902 	if (!nt)
903 		return -ENOMEM;
904 
905 	nt->ndev = ntb_register_transport(pdev, nt);
906 	if (!nt->ndev) {
907 		rc = -EIO;
908 		goto err;
909 	}
910 
911 	nt->mw = kcalloc(ntb_max_mw(nt->ndev), sizeof(struct ntb_transport_mw),
912 			 GFP_KERNEL);
913 	if (!nt->mw) {
914 		rc = -ENOMEM;
915 		goto err1;
916 	}
917 
918 	if (max_num_clients)
919 		nt->max_qps = min(ntb_max_cbs(nt->ndev), max_num_clients);
920 	else
921 		nt->max_qps = min(ntb_max_cbs(nt->ndev), ntb_max_mw(nt->ndev));
922 
923 	nt->qps = kcalloc(nt->max_qps, sizeof(struct ntb_transport_qp),
924 			  GFP_KERNEL);
925 	if (!nt->qps) {
926 		rc = -ENOMEM;
927 		goto err2;
928 	}
929 
930 	nt->qp_bitmap = ((u64) 1 << nt->max_qps) - 1;
931 
932 	for (i = 0; i < nt->max_qps; i++) {
933 		rc = ntb_transport_init_queue(nt, i);
934 		if (rc)
935 			goto err3;
936 	}
937 
938 	INIT_DELAYED_WORK(&nt->link_work, ntb_transport_link_work);
939 	INIT_WORK(&nt->link_cleanup, ntb_transport_link_cleanup);
940 
941 	rc = ntb_register_event_callback(nt->ndev,
942 					 ntb_transport_event_callback);
943 	if (rc)
944 		goto err3;
945 
946 	INIT_LIST_HEAD(&nt->client_devs);
947 	rc = ntb_bus_init(nt);
948 	if (rc)
949 		goto err4;
950 
951 	if (ntb_hw_link_status(nt->ndev))
952 		schedule_delayed_work(&nt->link_work, 0);
953 
954 	return 0;
955 
956 err4:
957 	ntb_unregister_event_callback(nt->ndev);
958 err3:
959 	kfree(nt->qps);
960 err2:
961 	kfree(nt->mw);
962 err1:
963 	ntb_unregister_transport(nt->ndev);
964 err:
965 	kfree(nt);
966 	return rc;
967 }
968 
969 void ntb_transport_free(void *transport)
970 {
971 	struct ntb_transport *nt = transport;
972 	struct ntb_device *ndev = nt->ndev;
973 	int i;
974 
975 	nt->transport_link = NTB_LINK_DOWN;
976 
977 	/* verify that all the qp's are freed */
978 	for (i = 0; i < nt->max_qps; i++) {
979 		if (!test_bit(i, &nt->qp_bitmap))
980 			ntb_transport_free_queue(&nt->qps[i]);
981 		debugfs_remove_recursive(nt->qps[i].debugfs_dir);
982 	}
983 
984 	ntb_bus_remove(nt);
985 
986 	cancel_delayed_work_sync(&nt->link_work);
987 
988 	ntb_unregister_event_callback(ndev);
989 
990 	for (i = 0; i < ntb_max_mw(ndev); i++)
991 		ntb_free_mw(nt, i);
992 
993 	kfree(nt->qps);
994 	kfree(nt->mw);
995 	ntb_unregister_transport(ndev);
996 	kfree(nt);
997 }
998 
999 static void ntb_rx_copy_callback(void *data)
1000 {
1001 	struct ntb_queue_entry *entry = data;
1002 	struct ntb_transport_qp *qp = entry->qp;
1003 	void *cb_data = entry->cb_data;
1004 	unsigned int len = entry->len;
1005 	struct ntb_payload_header *hdr = entry->rx_hdr;
1006 
1007 	/* Ensure that the data is fully copied out before clearing the flag */
1008 	wmb();
1009 	hdr->flags = 0;
1010 
1011 	iowrite32(entry->index, &qp->rx_info->entry);
1012 
1013 	ntb_list_add(&qp->ntb_rx_free_q_lock, &entry->entry, &qp->rx_free_q);
1014 
1015 	if (qp->rx_handler && qp->client_ready == NTB_LINK_UP)
1016 		qp->rx_handler(qp, qp->cb_data, cb_data, len);
1017 }
1018 
1019 static void ntb_memcpy_rx(struct ntb_queue_entry *entry, void *offset)
1020 {
1021 	void *buf = entry->buf;
1022 	size_t len = entry->len;
1023 
1024 	memcpy(buf, offset, len);
1025 
1026 	ntb_rx_copy_callback(entry);
1027 }
1028 
1029 static void ntb_async_rx(struct ntb_queue_entry *entry, void *offset,
1030 			 size_t len)
1031 {
1032 	struct dma_async_tx_descriptor *txd;
1033 	struct ntb_transport_qp *qp = entry->qp;
1034 	struct dma_chan *chan = qp->dma_chan;
1035 	struct dma_device *device;
1036 	size_t pay_off, buff_off;
1037 	dma_addr_t src, dest;
1038 	dma_cookie_t cookie;
1039 	void *buf = entry->buf;
1040 	unsigned long flags;
1041 
1042 	entry->len = len;
1043 
1044 	if (!chan)
1045 		goto err;
1046 
1047 	if (len < copy_bytes)
1048 		goto err1;
1049 
1050 	device = chan->device;
1051 	pay_off = (size_t) offset & ~PAGE_MASK;
1052 	buff_off = (size_t) buf & ~PAGE_MASK;
1053 
1054 	if (!is_dma_copy_aligned(device, pay_off, buff_off, len))
1055 		goto err1;
1056 
1057 	dest = dma_map_single(device->dev, buf, len, DMA_FROM_DEVICE);
1058 	if (dma_mapping_error(device->dev, dest))
1059 		goto err1;
1060 
1061 	src = dma_map_single(device->dev, offset, len, DMA_TO_DEVICE);
1062 	if (dma_mapping_error(device->dev, src))
1063 		goto err2;
1064 
1065 	flags = DMA_COMPL_DEST_UNMAP_SINGLE | DMA_COMPL_SRC_UNMAP_SINGLE |
1066 		DMA_PREP_INTERRUPT;
1067 	txd = device->device_prep_dma_memcpy(chan, dest, src, len, flags);
1068 	if (!txd)
1069 		goto err3;
1070 
1071 	txd->callback = ntb_rx_copy_callback;
1072 	txd->callback_param = entry;
1073 
1074 	cookie = dmaengine_submit(txd);
1075 	if (dma_submit_error(cookie))
1076 		goto err3;
1077 
1078 	qp->last_cookie = cookie;
1079 
1080 	qp->rx_async++;
1081 
1082 	return;
1083 
1084 err3:
1085 	dma_unmap_single(device->dev, src, len, DMA_TO_DEVICE);
1086 err2:
1087 	dma_unmap_single(device->dev, dest, len, DMA_FROM_DEVICE);
1088 err1:
1089 	/* If the callbacks come out of order, the writing of the index to the
1090 	 * last completed will be out of order.  This may result in the
1091 	 * receive stalling forever.
1092 	 */
1093 	dma_sync_wait(chan, qp->last_cookie);
1094 err:
1095 	ntb_memcpy_rx(entry, offset);
1096 	qp->rx_memcpy++;
1097 }
1098 
1099 static int ntb_process_rxc(struct ntb_transport_qp *qp)
1100 {
1101 	struct ntb_payload_header *hdr;
1102 	struct ntb_queue_entry *entry;
1103 	void *offset;
1104 
1105 	offset = qp->rx_buff + qp->rx_max_frame * qp->rx_index;
1106 	hdr = offset + qp->rx_max_frame - sizeof(struct ntb_payload_header);
1107 
1108 	entry = ntb_list_rm(&qp->ntb_rx_pend_q_lock, &qp->rx_pend_q);
1109 	if (!entry) {
1110 		dev_dbg(&ntb_query_pdev(qp->ndev)->dev,
1111 			"no buffer - HDR ver %u, len %d, flags %x\n",
1112 			hdr->ver, hdr->len, hdr->flags);
1113 		qp->rx_err_no_buf++;
1114 		return -ENOMEM;
1115 	}
1116 
1117 	if (!(hdr->flags & DESC_DONE_FLAG)) {
1118 		ntb_list_add(&qp->ntb_rx_pend_q_lock, &entry->entry,
1119 			     &qp->rx_pend_q);
1120 		qp->rx_ring_empty++;
1121 		return -EAGAIN;
1122 	}
1123 
1124 	if (hdr->ver != (u32) qp->rx_pkts) {
1125 		dev_dbg(&ntb_query_pdev(qp->ndev)->dev,
1126 			"qp %d: version mismatch, expected %llu - got %u\n",
1127 			qp->qp_num, qp->rx_pkts, hdr->ver);
1128 		ntb_list_add(&qp->ntb_rx_pend_q_lock, &entry->entry,
1129 			     &qp->rx_pend_q);
1130 		qp->rx_err_ver++;
1131 		return -EIO;
1132 	}
1133 
1134 	if (hdr->flags & LINK_DOWN_FLAG) {
1135 		ntb_qp_link_down(qp);
1136 
1137 		goto err;
1138 	}
1139 
1140 	dev_dbg(&ntb_query_pdev(qp->ndev)->dev,
1141 		"rx offset %u, ver %u - %d payload received, buf size %d\n",
1142 		qp->rx_index, hdr->ver, hdr->len, entry->len);
1143 
1144 	qp->rx_bytes += hdr->len;
1145 	qp->rx_pkts++;
1146 
1147 	if (hdr->len > entry->len) {
1148 		qp->rx_err_oflow++;
1149 		dev_dbg(&ntb_query_pdev(qp->ndev)->dev,
1150 			"RX overflow! Wanted %d got %d\n",
1151 			hdr->len, entry->len);
1152 
1153 		goto err;
1154 	}
1155 
1156 	entry->index = qp->rx_index;
1157 	entry->rx_hdr = hdr;
1158 
1159 	ntb_async_rx(entry, offset, hdr->len);
1160 
1161 out:
1162 	qp->rx_index++;
1163 	qp->rx_index %= qp->rx_max_entry;
1164 
1165 	return 0;
1166 
1167 err:
1168 	ntb_list_add(&qp->ntb_rx_pend_q_lock, &entry->entry,
1169 		     &qp->rx_pend_q);
1170 	/* Ensure that the data is fully copied out before clearing the flag */
1171 	wmb();
1172 	hdr->flags = 0;
1173 	iowrite32(qp->rx_index, &qp->rx_info->entry);
1174 
1175 	goto out;
1176 }
1177 
1178 static void ntb_transport_rx(unsigned long data)
1179 {
1180 	struct ntb_transport_qp *qp = (struct ntb_transport_qp *)data;
1181 	int rc, i;
1182 
1183 	/* Limit the number of packets processed in a single interrupt to
1184 	 * provide fairness to others
1185 	 */
1186 	for (i = 0; i < qp->rx_max_entry; i++) {
1187 		rc = ntb_process_rxc(qp);
1188 		if (rc)
1189 			break;
1190 	}
1191 
1192 	if (qp->dma_chan)
1193 		dma_async_issue_pending(qp->dma_chan);
1194 }
1195 
1196 static void ntb_transport_rxc_db(void *data, int db_num)
1197 {
1198 	struct ntb_transport_qp *qp = data;
1199 
1200 	dev_dbg(&ntb_query_pdev(qp->ndev)->dev, "%s: doorbell %d received\n",
1201 		__func__, db_num);
1202 
1203 	tasklet_schedule(&qp->rx_work);
1204 }
1205 
1206 static void ntb_tx_copy_callback(void *data)
1207 {
1208 	struct ntb_queue_entry *entry = data;
1209 	struct ntb_transport_qp *qp = entry->qp;
1210 	struct ntb_payload_header __iomem *hdr = entry->tx_hdr;
1211 
1212 	/* Ensure that the data is fully copied out before setting the flags */
1213 	wmb();
1214 	iowrite32(entry->flags | DESC_DONE_FLAG, &hdr->flags);
1215 
1216 	ntb_ring_doorbell(qp->ndev, qp->qp_num);
1217 
1218 	/* The entry length can only be zero if the packet is intended to be a
1219 	 * "link down" or similar.  Since no payload is being sent in these
1220 	 * cases, there is nothing to add to the completion queue.
1221 	 */
1222 	if (entry->len > 0) {
1223 		qp->tx_bytes += entry->len;
1224 
1225 		if (qp->tx_handler)
1226 			qp->tx_handler(qp, qp->cb_data, entry->cb_data,
1227 				       entry->len);
1228 	}
1229 
1230 	ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, &qp->tx_free_q);
1231 }
1232 
1233 static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset)
1234 {
1235 	memcpy_toio(offset, entry->buf, entry->len);
1236 
1237 	ntb_tx_copy_callback(entry);
1238 }
1239 
1240 static void ntb_async_tx(struct ntb_transport_qp *qp,
1241 			 struct ntb_queue_entry *entry)
1242 {
1243 	struct ntb_payload_header __iomem *hdr;
1244 	struct dma_async_tx_descriptor *txd;
1245 	struct dma_chan *chan = qp->dma_chan;
1246 	struct dma_device *device;
1247 	size_t dest_off, buff_off;
1248 	dma_addr_t src, dest;
1249 	dma_cookie_t cookie;
1250 	void __iomem *offset;
1251 	size_t len = entry->len;
1252 	void *buf = entry->buf;
1253 	unsigned long flags;
1254 
1255 	offset = qp->tx_mw + qp->tx_max_frame * qp->tx_index;
1256 	hdr = offset + qp->tx_max_frame - sizeof(struct ntb_payload_header);
1257 	entry->tx_hdr = hdr;
1258 
1259 	iowrite32(entry->len, &hdr->len);
1260 	iowrite32((u32) qp->tx_pkts, &hdr->ver);
1261 
1262 	if (!chan)
1263 		goto err;
1264 
1265 	if (len < copy_bytes)
1266 		goto err;
1267 
1268 	device = chan->device;
1269 	dest = qp->tx_mw_phys + qp->tx_max_frame * qp->tx_index;
1270 	buff_off = (size_t) buf & ~PAGE_MASK;
1271 	dest_off = (size_t) dest & ~PAGE_MASK;
1272 
1273 	if (!is_dma_copy_aligned(device, buff_off, dest_off, len))
1274 		goto err;
1275 
1276 	src = dma_map_single(device->dev, buf, len, DMA_TO_DEVICE);
1277 	if (dma_mapping_error(device->dev, src))
1278 		goto err;
1279 
1280 	flags = DMA_COMPL_SRC_UNMAP_SINGLE | DMA_PREP_INTERRUPT;
1281 	txd = device->device_prep_dma_memcpy(chan, dest, src, len, flags);
1282 	if (!txd)
1283 		goto err1;
1284 
1285 	txd->callback = ntb_tx_copy_callback;
1286 	txd->callback_param = entry;
1287 
1288 	cookie = dmaengine_submit(txd);
1289 	if (dma_submit_error(cookie))
1290 		goto err1;
1291 
1292 	dma_async_issue_pending(chan);
1293 	qp->tx_async++;
1294 
1295 	return;
1296 err1:
1297 	dma_unmap_single(device->dev, src, len, DMA_TO_DEVICE);
1298 err:
1299 	ntb_memcpy_tx(entry, offset);
1300 	qp->tx_memcpy++;
1301 }
1302 
1303 static int ntb_process_tx(struct ntb_transport_qp *qp,
1304 			  struct ntb_queue_entry *entry)
1305 {
1306 	dev_dbg(&ntb_query_pdev(qp->ndev)->dev, "%lld - tx %u, entry len %d flags %x buff %p\n",
1307 		qp->tx_pkts, qp->tx_index, entry->len, entry->flags,
1308 		entry->buf);
1309 	if (qp->tx_index == qp->remote_rx_info->entry) {
1310 		qp->tx_ring_full++;
1311 		return -EAGAIN;
1312 	}
1313 
1314 	if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) {
1315 		if (qp->tx_handler)
1316 			qp->tx_handler(qp->cb_data, qp, NULL, -EIO);
1317 
1318 		ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry,
1319 			     &qp->tx_free_q);
1320 		return 0;
1321 	}
1322 
1323 	ntb_async_tx(qp, entry);
1324 
1325 	qp->tx_index++;
1326 	qp->tx_index %= qp->tx_max_entry;
1327 
1328 	qp->tx_pkts++;
1329 
1330 	return 0;
1331 }
1332 
1333 static void ntb_send_link_down(struct ntb_transport_qp *qp)
1334 {
1335 	struct pci_dev *pdev = ntb_query_pdev(qp->ndev);
1336 	struct ntb_queue_entry *entry;
1337 	int i, rc;
1338 
1339 	if (qp->qp_link == NTB_LINK_DOWN)
1340 		return;
1341 
1342 	qp->qp_link = NTB_LINK_DOWN;
1343 	dev_info(&pdev->dev, "qp %d: Link Down\n", qp->qp_num);
1344 
1345 	for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) {
1346 		entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
1347 		if (entry)
1348 			break;
1349 		msleep(100);
1350 	}
1351 
1352 	if (!entry)
1353 		return;
1354 
1355 	entry->cb_data = NULL;
1356 	entry->buf = NULL;
1357 	entry->len = 0;
1358 	entry->flags = LINK_DOWN_FLAG;
1359 
1360 	rc = ntb_process_tx(qp, entry);
1361 	if (rc)
1362 		dev_err(&pdev->dev, "ntb: QP%d unable to send linkdown msg\n",
1363 			qp->qp_num);
1364 }
1365 
1366 /**
1367  * ntb_transport_create_queue - Create a new NTB transport layer queue
1368  * @rx_handler: receive callback function
1369  * @tx_handler: transmit callback function
1370  * @event_handler: event callback function
1371  *
1372  * Create a new NTB transport layer queue and provide the queue with a callback
1373  * routine for both transmit and receive.  The receive callback routine will be
1374  * used to pass up data when the transport has received it on the queue.   The
1375  * transmit callback routine will be called when the transport has completed the
1376  * transmission of the data on the queue and the data is ready to be freed.
1377  *
1378  * RETURNS: pointer to newly created ntb_queue, NULL on error.
1379  */
1380 struct ntb_transport_qp *
1381 ntb_transport_create_queue(void *data, struct pci_dev *pdev,
1382 			   const struct ntb_queue_handlers *handlers)
1383 {
1384 	struct ntb_queue_entry *entry;
1385 	struct ntb_transport_qp *qp;
1386 	struct ntb_transport *nt;
1387 	unsigned int free_queue;
1388 	int rc, i;
1389 
1390 	nt = ntb_find_transport(pdev);
1391 	if (!nt)
1392 		goto err;
1393 
1394 	free_queue = ffs(nt->qp_bitmap);
1395 	if (!free_queue)
1396 		goto err;
1397 
1398 	/* decrement free_queue to make it zero based */
1399 	free_queue--;
1400 
1401 	clear_bit(free_queue, &nt->qp_bitmap);
1402 
1403 	qp = &nt->qps[free_queue];
1404 	qp->cb_data = data;
1405 	qp->rx_handler = handlers->rx_handler;
1406 	qp->tx_handler = handlers->tx_handler;
1407 	qp->event_handler = handlers->event_handler;
1408 
1409 	qp->dma_chan = dma_find_channel(DMA_MEMCPY);
1410 	if (!qp->dma_chan)
1411 		dev_info(&pdev->dev, "Unable to allocate DMA channel, using CPU instead\n");
1412 	else
1413 		dmaengine_get();
1414 
1415 	for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
1416 		entry = kzalloc(sizeof(struct ntb_queue_entry), GFP_ATOMIC);
1417 		if (!entry)
1418 			goto err1;
1419 
1420 		entry->qp = qp;
1421 		ntb_list_add(&qp->ntb_rx_free_q_lock, &entry->entry,
1422 			     &qp->rx_free_q);
1423 	}
1424 
1425 	for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
1426 		entry = kzalloc(sizeof(struct ntb_queue_entry), GFP_ATOMIC);
1427 		if (!entry)
1428 			goto err2;
1429 
1430 		entry->qp = qp;
1431 		ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry,
1432 			     &qp->tx_free_q);
1433 	}
1434 
1435 	tasklet_init(&qp->rx_work, ntb_transport_rx, (unsigned long) qp);
1436 
1437 	rc = ntb_register_db_callback(qp->ndev, free_queue, qp,
1438 				      ntb_transport_rxc_db);
1439 	if (rc)
1440 		goto err3;
1441 
1442 	dev_info(&pdev->dev, "NTB Transport QP %d created\n", qp->qp_num);
1443 
1444 	return qp;
1445 
1446 err3:
1447 	tasklet_disable(&qp->rx_work);
1448 err2:
1449 	while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
1450 		kfree(entry);
1451 err1:
1452 	while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q)))
1453 		kfree(entry);
1454 	set_bit(free_queue, &nt->qp_bitmap);
1455 err:
1456 	return NULL;
1457 }
1458 EXPORT_SYMBOL_GPL(ntb_transport_create_queue);
1459 
1460 /**
1461  * ntb_transport_free_queue - Frees NTB transport queue
1462  * @qp: NTB queue to be freed
1463  *
1464  * Frees NTB transport queue
1465  */
1466 void ntb_transport_free_queue(struct ntb_transport_qp *qp)
1467 {
1468 	struct pci_dev *pdev;
1469 	struct ntb_queue_entry *entry;
1470 
1471 	if (!qp)
1472 		return;
1473 
1474 	pdev = ntb_query_pdev(qp->ndev);
1475 
1476 	if (qp->dma_chan) {
1477 		struct dma_chan *chan = qp->dma_chan;
1478 		/* Putting the dma_chan to NULL will force any new traffic to be
1479 		 * processed by the CPU instead of the DAM engine
1480 		 */
1481 		qp->dma_chan = NULL;
1482 
1483 		/* Try to be nice and wait for any queued DMA engine
1484 		 * transactions to process before smashing it with a rock
1485 		 */
1486 		dma_sync_wait(chan, qp->last_cookie);
1487 		dmaengine_terminate_all(chan);
1488 		dmaengine_put();
1489 	}
1490 
1491 	ntb_unregister_db_callback(qp->ndev, qp->qp_num);
1492 	tasklet_disable(&qp->rx_work);
1493 
1494 	cancel_delayed_work_sync(&qp->link_work);
1495 
1496 	while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q)))
1497 		kfree(entry);
1498 
1499 	while ((entry = ntb_list_rm(&qp->ntb_rx_pend_q_lock, &qp->rx_pend_q))) {
1500 		dev_warn(&pdev->dev, "Freeing item from a non-empty queue\n");
1501 		kfree(entry);
1502 	}
1503 
1504 	while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
1505 		kfree(entry);
1506 
1507 	set_bit(qp->qp_num, &qp->transport->qp_bitmap);
1508 
1509 	dev_info(&pdev->dev, "NTB Transport QP %d freed\n", qp->qp_num);
1510 }
1511 EXPORT_SYMBOL_GPL(ntb_transport_free_queue);
1512 
1513 /**
1514  * ntb_transport_rx_remove - Dequeues enqueued rx packet
1515  * @qp: NTB queue to be freed
1516  * @len: pointer to variable to write enqueued buffers length
1517  *
1518  * Dequeues unused buffers from receive queue.  Should only be used during
1519  * shutdown of qp.
1520  *
1521  * RETURNS: NULL error value on error, or void* for success.
1522  */
1523 void *ntb_transport_rx_remove(struct ntb_transport_qp *qp, unsigned int *len)
1524 {
1525 	struct ntb_queue_entry *entry;
1526 	void *buf;
1527 
1528 	if (!qp || qp->client_ready == NTB_LINK_UP)
1529 		return NULL;
1530 
1531 	entry = ntb_list_rm(&qp->ntb_rx_pend_q_lock, &qp->rx_pend_q);
1532 	if (!entry)
1533 		return NULL;
1534 
1535 	buf = entry->cb_data;
1536 	*len = entry->len;
1537 
1538 	ntb_list_add(&qp->ntb_rx_free_q_lock, &entry->entry, &qp->rx_free_q);
1539 
1540 	return buf;
1541 }
1542 EXPORT_SYMBOL_GPL(ntb_transport_rx_remove);
1543 
1544 /**
1545  * ntb_transport_rx_enqueue - Enqueue a new NTB queue entry
1546  * @qp: NTB transport layer queue the entry is to be enqueued on
1547  * @cb: per buffer pointer for callback function to use
1548  * @data: pointer to data buffer that incoming packets will be copied into
1549  * @len: length of the data buffer
1550  *
1551  * Enqueue a new receive buffer onto the transport queue into which a NTB
1552  * payload can be received into.
1553  *
1554  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
1555  */
1556 int ntb_transport_rx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data,
1557 			     unsigned int len)
1558 {
1559 	struct ntb_queue_entry *entry;
1560 
1561 	if (!qp)
1562 		return -EINVAL;
1563 
1564 	entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q);
1565 	if (!entry)
1566 		return -ENOMEM;
1567 
1568 	entry->cb_data = cb;
1569 	entry->buf = data;
1570 	entry->len = len;
1571 
1572 	ntb_list_add(&qp->ntb_rx_pend_q_lock, &entry->entry, &qp->rx_pend_q);
1573 
1574 	return 0;
1575 }
1576 EXPORT_SYMBOL_GPL(ntb_transport_rx_enqueue);
1577 
1578 /**
1579  * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry
1580  * @qp: NTB transport layer queue the entry is to be enqueued on
1581  * @cb: per buffer pointer for callback function to use
1582  * @data: pointer to data buffer that will be sent
1583  * @len: length of the data buffer
1584  *
1585  * Enqueue a new transmit buffer onto the transport queue from which a NTB
1586  * payload will be transmitted.  This assumes that a lock is being held to
1587  * serialize access to the qp.
1588  *
1589  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
1590  */
1591 int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data,
1592 			     unsigned int len)
1593 {
1594 	struct ntb_queue_entry *entry;
1595 	int rc;
1596 
1597 	if (!qp || qp->qp_link != NTB_LINK_UP || !len)
1598 		return -EINVAL;
1599 
1600 	entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
1601 	if (!entry) {
1602 		qp->tx_err_no_buf++;
1603 		return -ENOMEM;
1604 	}
1605 
1606 	entry->cb_data = cb;
1607 	entry->buf = data;
1608 	entry->len = len;
1609 	entry->flags = 0;
1610 
1611 	rc = ntb_process_tx(qp, entry);
1612 	if (rc)
1613 		ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry,
1614 			     &qp->tx_free_q);
1615 
1616 	return rc;
1617 }
1618 EXPORT_SYMBOL_GPL(ntb_transport_tx_enqueue);
1619 
1620 /**
1621  * ntb_transport_link_up - Notify NTB transport of client readiness to use queue
1622  * @qp: NTB transport layer queue to be enabled
1623  *
1624  * Notify NTB transport layer of client readiness to use queue
1625  */
1626 void ntb_transport_link_up(struct ntb_transport_qp *qp)
1627 {
1628 	if (!qp)
1629 		return;
1630 
1631 	qp->client_ready = NTB_LINK_UP;
1632 
1633 	if (qp->transport->transport_link == NTB_LINK_UP)
1634 		schedule_delayed_work(&qp->link_work, 0);
1635 }
1636 EXPORT_SYMBOL_GPL(ntb_transport_link_up);
1637 
1638 /**
1639  * ntb_transport_link_down - Notify NTB transport to no longer enqueue data
1640  * @qp: NTB transport layer queue to be disabled
1641  *
1642  * Notify NTB transport layer of client's desire to no longer receive data on
1643  * transport queue specified.  It is the client's responsibility to ensure all
1644  * entries on queue are purged or otherwise handled appropriately.
1645  */
1646 void ntb_transport_link_down(struct ntb_transport_qp *qp)
1647 {
1648 	struct pci_dev *pdev;
1649 	int rc, val;
1650 
1651 	if (!qp)
1652 		return;
1653 
1654 	pdev = ntb_query_pdev(qp->ndev);
1655 	qp->client_ready = NTB_LINK_DOWN;
1656 
1657 	rc = ntb_read_local_spad(qp->ndev, QP_LINKS, &val);
1658 	if (rc) {
1659 		dev_err(&pdev->dev, "Error reading spad %d\n", QP_LINKS);
1660 		return;
1661 	}
1662 
1663 	rc = ntb_write_remote_spad(qp->ndev, QP_LINKS,
1664 				   val & ~(1 << qp->qp_num));
1665 	if (rc)
1666 		dev_err(&pdev->dev, "Error writing %x to remote spad %d\n",
1667 			val & ~(1 << qp->qp_num), QP_LINKS);
1668 
1669 	if (qp->qp_link == NTB_LINK_UP)
1670 		ntb_send_link_down(qp);
1671 	else
1672 		cancel_delayed_work_sync(&qp->link_work);
1673 }
1674 EXPORT_SYMBOL_GPL(ntb_transport_link_down);
1675 
1676 /**
1677  * ntb_transport_link_query - Query transport link state
1678  * @qp: NTB transport layer queue to be queried
1679  *
1680  * Query connectivity to the remote system of the NTB transport queue
1681  *
1682  * RETURNS: true for link up or false for link down
1683  */
1684 bool ntb_transport_link_query(struct ntb_transport_qp *qp)
1685 {
1686 	if (!qp)
1687 		return false;
1688 
1689 	return qp->qp_link == NTB_LINK_UP;
1690 }
1691 EXPORT_SYMBOL_GPL(ntb_transport_link_query);
1692 
1693 /**
1694  * ntb_transport_qp_num - Query the qp number
1695  * @qp: NTB transport layer queue to be queried
1696  *
1697  * Query qp number of the NTB transport queue
1698  *
1699  * RETURNS: a zero based number specifying the qp number
1700  */
1701 unsigned char ntb_transport_qp_num(struct ntb_transport_qp *qp)
1702 {
1703 	if (!qp)
1704 		return 0;
1705 
1706 	return qp->qp_num;
1707 }
1708 EXPORT_SYMBOL_GPL(ntb_transport_qp_num);
1709 
1710 /**
1711  * ntb_transport_max_size - Query the max payload size of a qp
1712  * @qp: NTB transport layer queue to be queried
1713  *
1714  * Query the maximum payload size permissible on the given qp
1715  *
1716  * RETURNS: the max payload size of a qp
1717  */
1718 unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp)
1719 {
1720 	unsigned int max;
1721 
1722 	if (!qp)
1723 		return 0;
1724 
1725 	if (!qp->dma_chan)
1726 		return qp->tx_max_frame - sizeof(struct ntb_payload_header);
1727 
1728 	/* If DMA engine usage is possible, try to find the max size for that */
1729 	max = qp->tx_max_frame - sizeof(struct ntb_payload_header);
1730 	max -= max % (1 << qp->dma_chan->device->copy_align);
1731 
1732 	return max;
1733 }
1734 EXPORT_SYMBOL_GPL(ntb_transport_max_size);
1735