xref: /openbmc/linux/drivers/ntb/ntb_transport.c (revision 9b9c2cd4)
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, upper_32_bits(size));
829 
830 		spad = MW0_SZ_LOW + (i * 2);
831 		ntb_peer_spad_write(ndev, spad, lower_32_bits(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 	phys_addr_t mw_base;
932 	resource_size_t mw_size;
933 	unsigned int num_qps_mw, tx_size;
934 	unsigned int mw_num, mw_count, qp_count;
935 	u64 qp_offset;
936 
937 	mw_count = nt->mw_count;
938 	qp_count = nt->qp_count;
939 
940 	mw_num = QP_TO_MW(nt, qp_num);
941 
942 	qp = &nt->qp_vec[qp_num];
943 	qp->qp_num = qp_num;
944 	qp->transport = nt;
945 	qp->ndev = nt->ndev;
946 	qp->client_ready = false;
947 	qp->event_handler = NULL;
948 	ntb_qp_link_down_reset(qp);
949 
950 	if (qp_count % mw_count && mw_num + 1 < qp_count / mw_count)
951 		num_qps_mw = qp_count / mw_count + 1;
952 	else
953 		num_qps_mw = qp_count / mw_count;
954 
955 	mw_base = nt->mw_vec[mw_num].phys_addr;
956 	mw_size = nt->mw_vec[mw_num].phys_size;
957 
958 	tx_size = (unsigned int)mw_size / num_qps_mw;
959 	qp_offset = tx_size * (qp_num / mw_count);
960 
961 	qp->tx_mw = nt->mw_vec[mw_num].vbase + qp_offset;
962 	if (!qp->tx_mw)
963 		return -EINVAL;
964 
965 	qp->tx_mw_phys = mw_base + qp_offset;
966 	if (!qp->tx_mw_phys)
967 		return -EINVAL;
968 
969 	tx_size -= sizeof(struct ntb_rx_info);
970 	qp->rx_info = qp->tx_mw + tx_size;
971 
972 	/* Due to housekeeping, there must be atleast 2 buffs */
973 	qp->tx_max_frame = min(transport_mtu, tx_size / 2);
974 	qp->tx_max_entry = tx_size / qp->tx_max_frame;
975 
976 	if (nt->debugfs_node_dir) {
977 		char debugfs_name[4];
978 
979 		snprintf(debugfs_name, 4, "qp%d", qp_num);
980 		qp->debugfs_dir = debugfs_create_dir(debugfs_name,
981 						     nt->debugfs_node_dir);
982 
983 		qp->debugfs_stats = debugfs_create_file("stats", S_IRUSR,
984 							qp->debugfs_dir, qp,
985 							&ntb_qp_debugfs_stats);
986 	} else {
987 		qp->debugfs_dir = NULL;
988 		qp->debugfs_stats = NULL;
989 	}
990 
991 	INIT_DELAYED_WORK(&qp->link_work, ntb_qp_link_work);
992 	INIT_WORK(&qp->link_cleanup, ntb_qp_link_cleanup_work);
993 
994 	spin_lock_init(&qp->ntb_rx_q_lock);
995 	spin_lock_init(&qp->ntb_tx_free_q_lock);
996 
997 	INIT_LIST_HEAD(&qp->rx_post_q);
998 	INIT_LIST_HEAD(&qp->rx_pend_q);
999 	INIT_LIST_HEAD(&qp->rx_free_q);
1000 	INIT_LIST_HEAD(&qp->tx_free_q);
1001 
1002 	tasklet_init(&qp->rxc_db_work, ntb_transport_rxc_db,
1003 		     (unsigned long)qp);
1004 
1005 	return 0;
1006 }
1007 
1008 static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev)
1009 {
1010 	struct ntb_transport_ctx *nt;
1011 	struct ntb_transport_mw *mw;
1012 	unsigned int mw_count, qp_count;
1013 	u64 qp_bitmap;
1014 	int node;
1015 	int rc, i;
1016 
1017 	if (ntb_db_is_unsafe(ndev))
1018 		dev_dbg(&ndev->dev,
1019 			"doorbell is unsafe, proceed anyway...\n");
1020 	if (ntb_spad_is_unsafe(ndev))
1021 		dev_dbg(&ndev->dev,
1022 			"scratchpad is unsafe, proceed anyway...\n");
1023 
1024 	node = dev_to_node(&ndev->dev);
1025 
1026 	nt = kzalloc_node(sizeof(*nt), GFP_KERNEL, node);
1027 	if (!nt)
1028 		return -ENOMEM;
1029 
1030 	nt->ndev = ndev;
1031 
1032 	mw_count = ntb_mw_count(ndev);
1033 
1034 	nt->mw_count = mw_count;
1035 
1036 	nt->mw_vec = kzalloc_node(mw_count * sizeof(*nt->mw_vec),
1037 				  GFP_KERNEL, node);
1038 	if (!nt->mw_vec) {
1039 		rc = -ENOMEM;
1040 		goto err;
1041 	}
1042 
1043 	for (i = 0; i < mw_count; i++) {
1044 		mw = &nt->mw_vec[i];
1045 
1046 		rc = ntb_mw_get_range(ndev, i, &mw->phys_addr, &mw->phys_size,
1047 				      &mw->xlat_align, &mw->xlat_align_size);
1048 		if (rc)
1049 			goto err1;
1050 
1051 		mw->vbase = ioremap_wc(mw->phys_addr, mw->phys_size);
1052 		if (!mw->vbase) {
1053 			rc = -ENOMEM;
1054 			goto err1;
1055 		}
1056 
1057 		mw->buff_size = 0;
1058 		mw->xlat_size = 0;
1059 		mw->virt_addr = NULL;
1060 		mw->dma_addr = 0;
1061 	}
1062 
1063 	qp_bitmap = ntb_db_valid_mask(ndev);
1064 
1065 	qp_count = ilog2(qp_bitmap);
1066 	if (max_num_clients && max_num_clients < qp_count)
1067 		qp_count = max_num_clients;
1068 	else if (mw_count < qp_count)
1069 		qp_count = mw_count;
1070 
1071 	qp_bitmap &= BIT_ULL(qp_count) - 1;
1072 
1073 	nt->qp_count = qp_count;
1074 	nt->qp_bitmap = qp_bitmap;
1075 	nt->qp_bitmap_free = qp_bitmap;
1076 
1077 	nt->qp_vec = kzalloc_node(qp_count * sizeof(*nt->qp_vec),
1078 				  GFP_KERNEL, node);
1079 	if (!nt->qp_vec) {
1080 		rc = -ENOMEM;
1081 		goto err1;
1082 	}
1083 
1084 	if (nt_debugfs_dir) {
1085 		nt->debugfs_node_dir =
1086 			debugfs_create_dir(pci_name(ndev->pdev),
1087 					   nt_debugfs_dir);
1088 	}
1089 
1090 	for (i = 0; i < qp_count; i++) {
1091 		rc = ntb_transport_init_queue(nt, i);
1092 		if (rc)
1093 			goto err2;
1094 	}
1095 
1096 	INIT_DELAYED_WORK(&nt->link_work, ntb_transport_link_work);
1097 	INIT_WORK(&nt->link_cleanup, ntb_transport_link_cleanup_work);
1098 
1099 	rc = ntb_set_ctx(ndev, nt, &ntb_transport_ops);
1100 	if (rc)
1101 		goto err2;
1102 
1103 	INIT_LIST_HEAD(&nt->client_devs);
1104 	rc = ntb_bus_init(nt);
1105 	if (rc)
1106 		goto err3;
1107 
1108 	nt->link_is_up = false;
1109 	ntb_link_enable(ndev, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
1110 	ntb_link_event(ndev);
1111 
1112 	return 0;
1113 
1114 err3:
1115 	ntb_clear_ctx(ndev);
1116 err2:
1117 	kfree(nt->qp_vec);
1118 err1:
1119 	while (i--) {
1120 		mw = &nt->mw_vec[i];
1121 		iounmap(mw->vbase);
1122 	}
1123 	kfree(nt->mw_vec);
1124 err:
1125 	kfree(nt);
1126 	return rc;
1127 }
1128 
1129 static void ntb_transport_free(struct ntb_client *self, struct ntb_dev *ndev)
1130 {
1131 	struct ntb_transport_ctx *nt = ndev->ctx;
1132 	struct ntb_transport_qp *qp;
1133 	u64 qp_bitmap_alloc;
1134 	int i;
1135 
1136 	ntb_transport_link_cleanup(nt);
1137 	cancel_work_sync(&nt->link_cleanup);
1138 	cancel_delayed_work_sync(&nt->link_work);
1139 
1140 	qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free;
1141 
1142 	/* verify that all the qp's are freed */
1143 	for (i = 0; i < nt->qp_count; i++) {
1144 		qp = &nt->qp_vec[i];
1145 		if (qp_bitmap_alloc & BIT_ULL(i))
1146 			ntb_transport_free_queue(qp);
1147 		debugfs_remove_recursive(qp->debugfs_dir);
1148 	}
1149 
1150 	ntb_link_disable(ndev);
1151 	ntb_clear_ctx(ndev);
1152 
1153 	ntb_bus_remove(nt);
1154 
1155 	for (i = nt->mw_count; i--; ) {
1156 		ntb_free_mw(nt, i);
1157 		iounmap(nt->mw_vec[i].vbase);
1158 	}
1159 
1160 	kfree(nt->qp_vec);
1161 	kfree(nt->mw_vec);
1162 	kfree(nt);
1163 }
1164 
1165 static void ntb_complete_rxc(struct ntb_transport_qp *qp)
1166 {
1167 	struct ntb_queue_entry *entry;
1168 	void *cb_data;
1169 	unsigned int len;
1170 	unsigned long irqflags;
1171 
1172 	spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags);
1173 
1174 	while (!list_empty(&qp->rx_post_q)) {
1175 		entry = list_first_entry(&qp->rx_post_q,
1176 					 struct ntb_queue_entry, entry);
1177 		if (!(entry->flags & DESC_DONE_FLAG))
1178 			break;
1179 
1180 		entry->rx_hdr->flags = 0;
1181 		iowrite32(entry->index, &qp->rx_info->entry);
1182 
1183 		cb_data = entry->cb_data;
1184 		len = entry->len;
1185 
1186 		list_move_tail(&entry->entry, &qp->rx_free_q);
1187 
1188 		spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags);
1189 
1190 		if (qp->rx_handler && qp->client_ready)
1191 			qp->rx_handler(qp, qp->cb_data, cb_data, len);
1192 
1193 		spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags);
1194 	}
1195 
1196 	spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags);
1197 }
1198 
1199 static void ntb_rx_copy_callback(void *data)
1200 {
1201 	struct ntb_queue_entry *entry = data;
1202 
1203 	entry->flags |= DESC_DONE_FLAG;
1204 
1205 	ntb_complete_rxc(entry->qp);
1206 }
1207 
1208 static void ntb_memcpy_rx(struct ntb_queue_entry *entry, void *offset)
1209 {
1210 	void *buf = entry->buf;
1211 	size_t len = entry->len;
1212 
1213 	memcpy(buf, offset, len);
1214 
1215 	/* Ensure that the data is fully copied out before clearing the flag */
1216 	wmb();
1217 
1218 	ntb_rx_copy_callback(entry);
1219 }
1220 
1221 static void ntb_async_rx(struct ntb_queue_entry *entry, void *offset)
1222 {
1223 	struct dma_async_tx_descriptor *txd;
1224 	struct ntb_transport_qp *qp = entry->qp;
1225 	struct dma_chan *chan = qp->rx_dma_chan;
1226 	struct dma_device *device;
1227 	size_t pay_off, buff_off, len;
1228 	struct dmaengine_unmap_data *unmap;
1229 	dma_cookie_t cookie;
1230 	void *buf = entry->buf;
1231 
1232 	len = entry->len;
1233 
1234 	if (!chan)
1235 		goto err;
1236 
1237 	if (len < copy_bytes)
1238 		goto err;
1239 
1240 	device = chan->device;
1241 	pay_off = (size_t)offset & ~PAGE_MASK;
1242 	buff_off = (size_t)buf & ~PAGE_MASK;
1243 
1244 	if (!is_dma_copy_aligned(device, pay_off, buff_off, len))
1245 		goto err;
1246 
1247 	unmap = dmaengine_get_unmap_data(device->dev, 2, GFP_NOWAIT);
1248 	if (!unmap)
1249 		goto err;
1250 
1251 	unmap->len = len;
1252 	unmap->addr[0] = dma_map_page(device->dev, virt_to_page(offset),
1253 				      pay_off, len, DMA_TO_DEVICE);
1254 	if (dma_mapping_error(device->dev, unmap->addr[0]))
1255 		goto err_get_unmap;
1256 
1257 	unmap->to_cnt = 1;
1258 
1259 	unmap->addr[1] = dma_map_page(device->dev, virt_to_page(buf),
1260 				      buff_off, len, DMA_FROM_DEVICE);
1261 	if (dma_mapping_error(device->dev, unmap->addr[1]))
1262 		goto err_get_unmap;
1263 
1264 	unmap->from_cnt = 1;
1265 
1266 	txd = device->device_prep_dma_memcpy(chan, unmap->addr[1],
1267 					     unmap->addr[0], len,
1268 					     DMA_PREP_INTERRUPT);
1269 	if (!txd)
1270 		goto err_get_unmap;
1271 
1272 	txd->callback = ntb_rx_copy_callback;
1273 	txd->callback_param = entry;
1274 	dma_set_unmap(txd, unmap);
1275 
1276 	cookie = dmaengine_submit(txd);
1277 	if (dma_submit_error(cookie))
1278 		goto err_set_unmap;
1279 
1280 	dmaengine_unmap_put(unmap);
1281 
1282 	qp->last_cookie = cookie;
1283 
1284 	qp->rx_async++;
1285 
1286 	return;
1287 
1288 err_set_unmap:
1289 	dmaengine_unmap_put(unmap);
1290 err_get_unmap:
1291 	dmaengine_unmap_put(unmap);
1292 err:
1293 	ntb_memcpy_rx(entry, offset);
1294 	qp->rx_memcpy++;
1295 }
1296 
1297 static int ntb_process_rxc(struct ntb_transport_qp *qp)
1298 {
1299 	struct ntb_payload_header *hdr;
1300 	struct ntb_queue_entry *entry;
1301 	void *offset;
1302 
1303 	offset = qp->rx_buff + qp->rx_max_frame * qp->rx_index;
1304 	hdr = offset + qp->rx_max_frame - sizeof(struct ntb_payload_header);
1305 
1306 	dev_dbg(&qp->ndev->pdev->dev, "qp %d: RX ver %u len %d flags %x\n",
1307 		qp->qp_num, hdr->ver, hdr->len, hdr->flags);
1308 
1309 	if (!(hdr->flags & DESC_DONE_FLAG)) {
1310 		dev_dbg(&qp->ndev->pdev->dev, "done flag not set\n");
1311 		qp->rx_ring_empty++;
1312 		return -EAGAIN;
1313 	}
1314 
1315 	if (hdr->flags & LINK_DOWN_FLAG) {
1316 		dev_dbg(&qp->ndev->pdev->dev, "link down flag set\n");
1317 		ntb_qp_link_down(qp);
1318 		hdr->flags = 0;
1319 		return -EAGAIN;
1320 	}
1321 
1322 	if (hdr->ver != (u32)qp->rx_pkts) {
1323 		dev_dbg(&qp->ndev->pdev->dev,
1324 			"version mismatch, expected %llu - got %u\n",
1325 			qp->rx_pkts, hdr->ver);
1326 		qp->rx_err_ver++;
1327 		return -EIO;
1328 	}
1329 
1330 	entry = ntb_list_mv(&qp->ntb_rx_q_lock, &qp->rx_pend_q, &qp->rx_post_q);
1331 	if (!entry) {
1332 		dev_dbg(&qp->ndev->pdev->dev, "no receive buffer\n");
1333 		qp->rx_err_no_buf++;
1334 		return -EAGAIN;
1335 	}
1336 
1337 	entry->rx_hdr = hdr;
1338 	entry->index = qp->rx_index;
1339 
1340 	if (hdr->len > entry->len) {
1341 		dev_dbg(&qp->ndev->pdev->dev,
1342 			"receive buffer overflow! Wanted %d got %d\n",
1343 			hdr->len, entry->len);
1344 		qp->rx_err_oflow++;
1345 
1346 		entry->len = -EIO;
1347 		entry->flags |= DESC_DONE_FLAG;
1348 
1349 		ntb_complete_rxc(qp);
1350 	} else {
1351 		dev_dbg(&qp->ndev->pdev->dev,
1352 			"RX OK index %u ver %u size %d into buf size %d\n",
1353 			qp->rx_index, hdr->ver, hdr->len, entry->len);
1354 
1355 		qp->rx_bytes += hdr->len;
1356 		qp->rx_pkts++;
1357 
1358 		entry->len = hdr->len;
1359 
1360 		ntb_async_rx(entry, offset);
1361 	}
1362 
1363 	qp->rx_index++;
1364 	qp->rx_index %= qp->rx_max_entry;
1365 
1366 	return 0;
1367 }
1368 
1369 static void ntb_transport_rxc_db(unsigned long data)
1370 {
1371 	struct ntb_transport_qp *qp = (void *)data;
1372 	int rc, i;
1373 
1374 	dev_dbg(&qp->ndev->pdev->dev, "%s: doorbell %d received\n",
1375 		__func__, qp->qp_num);
1376 
1377 	/* Limit the number of packets processed in a single interrupt to
1378 	 * provide fairness to others
1379 	 */
1380 	for (i = 0; i < qp->rx_max_entry; i++) {
1381 		rc = ntb_process_rxc(qp);
1382 		if (rc)
1383 			break;
1384 	}
1385 
1386 	if (i && qp->rx_dma_chan)
1387 		dma_async_issue_pending(qp->rx_dma_chan);
1388 
1389 	if (i == qp->rx_max_entry) {
1390 		/* there is more work to do */
1391 		tasklet_schedule(&qp->rxc_db_work);
1392 	} else if (ntb_db_read(qp->ndev) & BIT_ULL(qp->qp_num)) {
1393 		/* the doorbell bit is set: clear it */
1394 		ntb_db_clear(qp->ndev, BIT_ULL(qp->qp_num));
1395 		/* ntb_db_read ensures ntb_db_clear write is committed */
1396 		ntb_db_read(qp->ndev);
1397 
1398 		/* an interrupt may have arrived between finishing
1399 		 * ntb_process_rxc and clearing the doorbell bit:
1400 		 * there might be some more work to do.
1401 		 */
1402 		tasklet_schedule(&qp->rxc_db_work);
1403 	}
1404 }
1405 
1406 static void ntb_tx_copy_callback(void *data)
1407 {
1408 	struct ntb_queue_entry *entry = data;
1409 	struct ntb_transport_qp *qp = entry->qp;
1410 	struct ntb_payload_header __iomem *hdr = entry->tx_hdr;
1411 
1412 	iowrite32(entry->flags | DESC_DONE_FLAG, &hdr->flags);
1413 
1414 	ntb_peer_db_set(qp->ndev, BIT_ULL(qp->qp_num));
1415 
1416 	/* The entry length can only be zero if the packet is intended to be a
1417 	 * "link down" or similar.  Since no payload is being sent in these
1418 	 * cases, there is nothing to add to the completion queue.
1419 	 */
1420 	if (entry->len > 0) {
1421 		qp->tx_bytes += entry->len;
1422 
1423 		if (qp->tx_handler)
1424 			qp->tx_handler(qp, qp->cb_data, entry->cb_data,
1425 				       entry->len);
1426 	}
1427 
1428 	ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, &qp->tx_free_q);
1429 }
1430 
1431 static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset)
1432 {
1433 #ifdef ARCH_HAS_NOCACHE_UACCESS
1434 	/*
1435 	 * Using non-temporal mov to improve performance on non-cached
1436 	 * writes, even though we aren't actually copying from user space.
1437 	 */
1438 	__copy_from_user_inatomic_nocache(offset, entry->buf, entry->len);
1439 #else
1440 	memcpy_toio(offset, entry->buf, entry->len);
1441 #endif
1442 
1443 	/* Ensure that the data is fully copied out before setting the flags */
1444 	wmb();
1445 
1446 	ntb_tx_copy_callback(entry);
1447 }
1448 
1449 static void ntb_async_tx(struct ntb_transport_qp *qp,
1450 			 struct ntb_queue_entry *entry)
1451 {
1452 	struct ntb_payload_header __iomem *hdr;
1453 	struct dma_async_tx_descriptor *txd;
1454 	struct dma_chan *chan = qp->tx_dma_chan;
1455 	struct dma_device *device;
1456 	size_t dest_off, buff_off;
1457 	struct dmaengine_unmap_data *unmap;
1458 	dma_addr_t dest;
1459 	dma_cookie_t cookie;
1460 	void __iomem *offset;
1461 	size_t len = entry->len;
1462 	void *buf = entry->buf;
1463 
1464 	offset = qp->tx_mw + qp->tx_max_frame * qp->tx_index;
1465 	hdr = offset + qp->tx_max_frame - sizeof(struct ntb_payload_header);
1466 	entry->tx_hdr = hdr;
1467 
1468 	iowrite32(entry->len, &hdr->len);
1469 	iowrite32((u32)qp->tx_pkts, &hdr->ver);
1470 
1471 	if (!chan)
1472 		goto err;
1473 
1474 	if (len < copy_bytes)
1475 		goto err;
1476 
1477 	device = chan->device;
1478 	dest = qp->tx_mw_phys + qp->tx_max_frame * qp->tx_index;
1479 	buff_off = (size_t)buf & ~PAGE_MASK;
1480 	dest_off = (size_t)dest & ~PAGE_MASK;
1481 
1482 	if (!is_dma_copy_aligned(device, buff_off, dest_off, len))
1483 		goto err;
1484 
1485 	unmap = dmaengine_get_unmap_data(device->dev, 1, GFP_NOWAIT);
1486 	if (!unmap)
1487 		goto err;
1488 
1489 	unmap->len = len;
1490 	unmap->addr[0] = dma_map_page(device->dev, virt_to_page(buf),
1491 				      buff_off, len, DMA_TO_DEVICE);
1492 	if (dma_mapping_error(device->dev, unmap->addr[0]))
1493 		goto err_get_unmap;
1494 
1495 	unmap->to_cnt = 1;
1496 
1497 	txd = device->device_prep_dma_memcpy(chan, dest, unmap->addr[0], len,
1498 					     DMA_PREP_INTERRUPT);
1499 	if (!txd)
1500 		goto err_get_unmap;
1501 
1502 	txd->callback = ntb_tx_copy_callback;
1503 	txd->callback_param = entry;
1504 	dma_set_unmap(txd, unmap);
1505 
1506 	cookie = dmaengine_submit(txd);
1507 	if (dma_submit_error(cookie))
1508 		goto err_set_unmap;
1509 
1510 	dmaengine_unmap_put(unmap);
1511 
1512 	dma_async_issue_pending(chan);
1513 	qp->tx_async++;
1514 
1515 	return;
1516 err_set_unmap:
1517 	dmaengine_unmap_put(unmap);
1518 err_get_unmap:
1519 	dmaengine_unmap_put(unmap);
1520 err:
1521 	ntb_memcpy_tx(entry, offset);
1522 	qp->tx_memcpy++;
1523 }
1524 
1525 static int ntb_process_tx(struct ntb_transport_qp *qp,
1526 			  struct ntb_queue_entry *entry)
1527 {
1528 	if (qp->tx_index == qp->remote_rx_info->entry) {
1529 		qp->tx_ring_full++;
1530 		return -EAGAIN;
1531 	}
1532 
1533 	if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) {
1534 		if (qp->tx_handler)
1535 			qp->tx_handler(qp->cb_data, qp, NULL, -EIO);
1536 
1537 		ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry,
1538 			     &qp->tx_free_q);
1539 		return 0;
1540 	}
1541 
1542 	ntb_async_tx(qp, entry);
1543 
1544 	qp->tx_index++;
1545 	qp->tx_index %= qp->tx_max_entry;
1546 
1547 	qp->tx_pkts++;
1548 
1549 	return 0;
1550 }
1551 
1552 static void ntb_send_link_down(struct ntb_transport_qp *qp)
1553 {
1554 	struct pci_dev *pdev = qp->ndev->pdev;
1555 	struct ntb_queue_entry *entry;
1556 	int i, rc;
1557 
1558 	if (!qp->link_is_up)
1559 		return;
1560 
1561 	dev_info(&pdev->dev, "qp %d: Send Link Down\n", qp->qp_num);
1562 
1563 	for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) {
1564 		entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
1565 		if (entry)
1566 			break;
1567 		msleep(100);
1568 	}
1569 
1570 	if (!entry)
1571 		return;
1572 
1573 	entry->cb_data = NULL;
1574 	entry->buf = NULL;
1575 	entry->len = 0;
1576 	entry->flags = LINK_DOWN_FLAG;
1577 
1578 	rc = ntb_process_tx(qp, entry);
1579 	if (rc)
1580 		dev_err(&pdev->dev, "ntb: QP%d unable to send linkdown msg\n",
1581 			qp->qp_num);
1582 
1583 	ntb_qp_link_down_reset(qp);
1584 }
1585 
1586 static bool ntb_dma_filter_fn(struct dma_chan *chan, void *node)
1587 {
1588 	return dev_to_node(&chan->dev->device) == (int)(unsigned long)node;
1589 }
1590 
1591 /**
1592  * ntb_transport_create_queue - Create a new NTB transport layer queue
1593  * @rx_handler: receive callback function
1594  * @tx_handler: transmit callback function
1595  * @event_handler: event callback function
1596  *
1597  * Create a new NTB transport layer queue and provide the queue with a callback
1598  * routine for both transmit and receive.  The receive callback routine will be
1599  * used to pass up data when the transport has received it on the queue.   The
1600  * transmit callback routine will be called when the transport has completed the
1601  * transmission of the data on the queue and the data is ready to be freed.
1602  *
1603  * RETURNS: pointer to newly created ntb_queue, NULL on error.
1604  */
1605 struct ntb_transport_qp *
1606 ntb_transport_create_queue(void *data, struct device *client_dev,
1607 			   const struct ntb_queue_handlers *handlers)
1608 {
1609 	struct ntb_dev *ndev;
1610 	struct pci_dev *pdev;
1611 	struct ntb_transport_ctx *nt;
1612 	struct ntb_queue_entry *entry;
1613 	struct ntb_transport_qp *qp;
1614 	u64 qp_bit;
1615 	unsigned int free_queue;
1616 	dma_cap_mask_t dma_mask;
1617 	int node;
1618 	int i;
1619 
1620 	ndev = dev_ntb(client_dev->parent);
1621 	pdev = ndev->pdev;
1622 	nt = ndev->ctx;
1623 
1624 	node = dev_to_node(&ndev->dev);
1625 
1626 	free_queue = ffs(nt->qp_bitmap);
1627 	if (!free_queue)
1628 		goto err;
1629 
1630 	/* decrement free_queue to make it zero based */
1631 	free_queue--;
1632 
1633 	qp = &nt->qp_vec[free_queue];
1634 	qp_bit = BIT_ULL(qp->qp_num);
1635 
1636 	nt->qp_bitmap_free &= ~qp_bit;
1637 
1638 	qp->cb_data = data;
1639 	qp->rx_handler = handlers->rx_handler;
1640 	qp->tx_handler = handlers->tx_handler;
1641 	qp->event_handler = handlers->event_handler;
1642 
1643 	dma_cap_zero(dma_mask);
1644 	dma_cap_set(DMA_MEMCPY, dma_mask);
1645 
1646 	if (use_dma) {
1647 		qp->tx_dma_chan =
1648 			dma_request_channel(dma_mask, ntb_dma_filter_fn,
1649 					    (void *)(unsigned long)node);
1650 		if (!qp->tx_dma_chan)
1651 			dev_info(&pdev->dev, "Unable to allocate TX DMA channel\n");
1652 
1653 		qp->rx_dma_chan =
1654 			dma_request_channel(dma_mask, ntb_dma_filter_fn,
1655 					    (void *)(unsigned long)node);
1656 		if (!qp->rx_dma_chan)
1657 			dev_info(&pdev->dev, "Unable to allocate RX DMA channel\n");
1658 	} else {
1659 		qp->tx_dma_chan = NULL;
1660 		qp->rx_dma_chan = NULL;
1661 	}
1662 
1663 	dev_dbg(&pdev->dev, "Using %s memcpy for TX\n",
1664 		qp->tx_dma_chan ? "DMA" : "CPU");
1665 
1666 	dev_dbg(&pdev->dev, "Using %s memcpy for RX\n",
1667 		qp->rx_dma_chan ? "DMA" : "CPU");
1668 
1669 	for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
1670 		entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node);
1671 		if (!entry)
1672 			goto err1;
1673 
1674 		entry->qp = qp;
1675 		ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry,
1676 			     &qp->rx_free_q);
1677 	}
1678 
1679 	for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
1680 		entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node);
1681 		if (!entry)
1682 			goto err2;
1683 
1684 		entry->qp = qp;
1685 		ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry,
1686 			     &qp->tx_free_q);
1687 	}
1688 
1689 	ntb_db_clear(qp->ndev, qp_bit);
1690 	ntb_db_clear_mask(qp->ndev, qp_bit);
1691 
1692 	dev_info(&pdev->dev, "NTB Transport QP %d created\n", qp->qp_num);
1693 
1694 	return qp;
1695 
1696 err2:
1697 	while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
1698 		kfree(entry);
1699 err1:
1700 	while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q)))
1701 		kfree(entry);
1702 	if (qp->tx_dma_chan)
1703 		dma_release_channel(qp->tx_dma_chan);
1704 	if (qp->rx_dma_chan)
1705 		dma_release_channel(qp->rx_dma_chan);
1706 	nt->qp_bitmap_free |= qp_bit;
1707 err:
1708 	return NULL;
1709 }
1710 EXPORT_SYMBOL_GPL(ntb_transport_create_queue);
1711 
1712 /**
1713  * ntb_transport_free_queue - Frees NTB transport queue
1714  * @qp: NTB queue to be freed
1715  *
1716  * Frees NTB transport queue
1717  */
1718 void ntb_transport_free_queue(struct ntb_transport_qp *qp)
1719 {
1720 	struct pci_dev *pdev;
1721 	struct ntb_queue_entry *entry;
1722 	u64 qp_bit;
1723 
1724 	if (!qp)
1725 		return;
1726 
1727 	pdev = qp->ndev->pdev;
1728 
1729 	if (qp->tx_dma_chan) {
1730 		struct dma_chan *chan = qp->tx_dma_chan;
1731 		/* Putting the dma_chan to NULL will force any new traffic to be
1732 		 * processed by the CPU instead of the DAM engine
1733 		 */
1734 		qp->tx_dma_chan = NULL;
1735 
1736 		/* Try to be nice and wait for any queued DMA engine
1737 		 * transactions to process before smashing it with a rock
1738 		 */
1739 		dma_sync_wait(chan, qp->last_cookie);
1740 		dmaengine_terminate_all(chan);
1741 		dma_release_channel(chan);
1742 	}
1743 
1744 	if (qp->rx_dma_chan) {
1745 		struct dma_chan *chan = qp->rx_dma_chan;
1746 		/* Putting the dma_chan to NULL will force any new traffic to be
1747 		 * processed by the CPU instead of the DAM engine
1748 		 */
1749 		qp->rx_dma_chan = NULL;
1750 
1751 		/* Try to be nice and wait for any queued DMA engine
1752 		 * transactions to process before smashing it with a rock
1753 		 */
1754 		dma_sync_wait(chan, qp->last_cookie);
1755 		dmaengine_terminate_all(chan);
1756 		dma_release_channel(chan);
1757 	}
1758 
1759 	qp_bit = BIT_ULL(qp->qp_num);
1760 
1761 	ntb_db_set_mask(qp->ndev, qp_bit);
1762 	tasklet_disable(&qp->rxc_db_work);
1763 
1764 	cancel_delayed_work_sync(&qp->link_work);
1765 
1766 	qp->cb_data = NULL;
1767 	qp->rx_handler = NULL;
1768 	qp->tx_handler = NULL;
1769 	qp->event_handler = NULL;
1770 
1771 	while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q)))
1772 		kfree(entry);
1773 
1774 	while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q))) {
1775 		dev_warn(&pdev->dev, "Freeing item from non-empty rx_pend_q\n");
1776 		kfree(entry);
1777 	}
1778 
1779 	while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_post_q))) {
1780 		dev_warn(&pdev->dev, "Freeing item from non-empty rx_post_q\n");
1781 		kfree(entry);
1782 	}
1783 
1784 	while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
1785 		kfree(entry);
1786 
1787 	qp->transport->qp_bitmap_free |= qp_bit;
1788 
1789 	dev_info(&pdev->dev, "NTB Transport QP %d freed\n", qp->qp_num);
1790 }
1791 EXPORT_SYMBOL_GPL(ntb_transport_free_queue);
1792 
1793 /**
1794  * ntb_transport_rx_remove - Dequeues enqueued rx packet
1795  * @qp: NTB queue to be freed
1796  * @len: pointer to variable to write enqueued buffers length
1797  *
1798  * Dequeues unused buffers from receive queue.  Should only be used during
1799  * shutdown of qp.
1800  *
1801  * RETURNS: NULL error value on error, or void* for success.
1802  */
1803 void *ntb_transport_rx_remove(struct ntb_transport_qp *qp, unsigned int *len)
1804 {
1805 	struct ntb_queue_entry *entry;
1806 	void *buf;
1807 
1808 	if (!qp || qp->client_ready)
1809 		return NULL;
1810 
1811 	entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q);
1812 	if (!entry)
1813 		return NULL;
1814 
1815 	buf = entry->cb_data;
1816 	*len = entry->len;
1817 
1818 	ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_free_q);
1819 
1820 	return buf;
1821 }
1822 EXPORT_SYMBOL_GPL(ntb_transport_rx_remove);
1823 
1824 /**
1825  * ntb_transport_rx_enqueue - Enqueue a new NTB queue entry
1826  * @qp: NTB transport layer queue the entry is to be enqueued on
1827  * @cb: per buffer pointer for callback function to use
1828  * @data: pointer to data buffer that incoming packets will be copied into
1829  * @len: length of the data buffer
1830  *
1831  * Enqueue a new receive buffer onto the transport queue into which a NTB
1832  * payload can be received into.
1833  *
1834  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
1835  */
1836 int ntb_transport_rx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data,
1837 			     unsigned int len)
1838 {
1839 	struct ntb_queue_entry *entry;
1840 
1841 	if (!qp)
1842 		return -EINVAL;
1843 
1844 	entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q);
1845 	if (!entry)
1846 		return -ENOMEM;
1847 
1848 	entry->cb_data = cb;
1849 	entry->buf = data;
1850 	entry->len = len;
1851 	entry->flags = 0;
1852 
1853 	ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_pend_q);
1854 
1855 	tasklet_schedule(&qp->rxc_db_work);
1856 
1857 	return 0;
1858 }
1859 EXPORT_SYMBOL_GPL(ntb_transport_rx_enqueue);
1860 
1861 /**
1862  * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry
1863  * @qp: NTB transport layer queue the entry is to be enqueued on
1864  * @cb: per buffer pointer for callback function to use
1865  * @data: pointer to data buffer that will be sent
1866  * @len: length of the data buffer
1867  *
1868  * Enqueue a new transmit buffer onto the transport queue from which a NTB
1869  * payload will be transmitted.  This assumes that a lock is being held to
1870  * serialize access to the qp.
1871  *
1872  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
1873  */
1874 int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data,
1875 			     unsigned int len)
1876 {
1877 	struct ntb_queue_entry *entry;
1878 	int rc;
1879 
1880 	if (!qp || !qp->link_is_up || !len)
1881 		return -EINVAL;
1882 
1883 	entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
1884 	if (!entry) {
1885 		qp->tx_err_no_buf++;
1886 		return -EBUSY;
1887 	}
1888 
1889 	entry->cb_data = cb;
1890 	entry->buf = data;
1891 	entry->len = len;
1892 	entry->flags = 0;
1893 
1894 	rc = ntb_process_tx(qp, entry);
1895 	if (rc)
1896 		ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry,
1897 			     &qp->tx_free_q);
1898 
1899 	return rc;
1900 }
1901 EXPORT_SYMBOL_GPL(ntb_transport_tx_enqueue);
1902 
1903 /**
1904  * ntb_transport_link_up - Notify NTB transport of client readiness to use queue
1905  * @qp: NTB transport layer queue to be enabled
1906  *
1907  * Notify NTB transport layer of client readiness to use queue
1908  */
1909 void ntb_transport_link_up(struct ntb_transport_qp *qp)
1910 {
1911 	if (!qp)
1912 		return;
1913 
1914 	qp->client_ready = true;
1915 
1916 	if (qp->transport->link_is_up)
1917 		schedule_delayed_work(&qp->link_work, 0);
1918 }
1919 EXPORT_SYMBOL_GPL(ntb_transport_link_up);
1920 
1921 /**
1922  * ntb_transport_link_down - Notify NTB transport to no longer enqueue data
1923  * @qp: NTB transport layer queue to be disabled
1924  *
1925  * Notify NTB transport layer of client's desire to no longer receive data on
1926  * transport queue specified.  It is the client's responsibility to ensure all
1927  * entries on queue are purged or otherwise handled appropriately.
1928  */
1929 void ntb_transport_link_down(struct ntb_transport_qp *qp)
1930 {
1931 	int val;
1932 
1933 	if (!qp)
1934 		return;
1935 
1936 	qp->client_ready = false;
1937 
1938 	val = ntb_spad_read(qp->ndev, QP_LINKS);
1939 
1940 	ntb_peer_spad_write(qp->ndev, QP_LINKS,
1941 			    val & ~BIT(qp->qp_num));
1942 
1943 	if (qp->link_is_up)
1944 		ntb_send_link_down(qp);
1945 	else
1946 		cancel_delayed_work_sync(&qp->link_work);
1947 }
1948 EXPORT_SYMBOL_GPL(ntb_transport_link_down);
1949 
1950 /**
1951  * ntb_transport_link_query - Query transport link state
1952  * @qp: NTB transport layer queue to be queried
1953  *
1954  * Query connectivity to the remote system of the NTB transport queue
1955  *
1956  * RETURNS: true for link up or false for link down
1957  */
1958 bool ntb_transport_link_query(struct ntb_transport_qp *qp)
1959 {
1960 	if (!qp)
1961 		return false;
1962 
1963 	return qp->link_is_up;
1964 }
1965 EXPORT_SYMBOL_GPL(ntb_transport_link_query);
1966 
1967 /**
1968  * ntb_transport_qp_num - Query the qp number
1969  * @qp: NTB transport layer queue to be queried
1970  *
1971  * Query qp number of the NTB transport queue
1972  *
1973  * RETURNS: a zero based number specifying the qp number
1974  */
1975 unsigned char ntb_transport_qp_num(struct ntb_transport_qp *qp)
1976 {
1977 	if (!qp)
1978 		return 0;
1979 
1980 	return qp->qp_num;
1981 }
1982 EXPORT_SYMBOL_GPL(ntb_transport_qp_num);
1983 
1984 /**
1985  * ntb_transport_max_size - Query the max payload size of a qp
1986  * @qp: NTB transport layer queue to be queried
1987  *
1988  * Query the maximum payload size permissible on the given qp
1989  *
1990  * RETURNS: the max payload size of a qp
1991  */
1992 unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp)
1993 {
1994 	unsigned int max_size;
1995 	unsigned int copy_align;
1996 	struct dma_chan *rx_chan, *tx_chan;
1997 
1998 	if (!qp)
1999 		return 0;
2000 
2001 	rx_chan = qp->rx_dma_chan;
2002 	tx_chan = qp->tx_dma_chan;
2003 
2004 	copy_align = max(rx_chan ? rx_chan->device->copy_align : 0,
2005 			 tx_chan ? tx_chan->device->copy_align : 0);
2006 
2007 	/* If DMA engine usage is possible, try to find the max size for that */
2008 	max_size = qp->tx_max_frame - sizeof(struct ntb_payload_header);
2009 	max_size = round_down(max_size, 1 << copy_align);
2010 
2011 	return max_size;
2012 }
2013 EXPORT_SYMBOL_GPL(ntb_transport_max_size);
2014 
2015 unsigned int ntb_transport_tx_free_entry(struct ntb_transport_qp *qp)
2016 {
2017 	unsigned int head = qp->tx_index;
2018 	unsigned int tail = qp->remote_rx_info->entry;
2019 
2020 	return tail > head ? tail - head : qp->tx_max_entry + tail - head;
2021 }
2022 EXPORT_SYMBOL_GPL(ntb_transport_tx_free_entry);
2023 
2024 static void ntb_transport_doorbell_callback(void *data, int vector)
2025 {
2026 	struct ntb_transport_ctx *nt = data;
2027 	struct ntb_transport_qp *qp;
2028 	u64 db_bits;
2029 	unsigned int qp_num;
2030 
2031 	db_bits = (nt->qp_bitmap & ~nt->qp_bitmap_free &
2032 		   ntb_db_vector_mask(nt->ndev, vector));
2033 
2034 	while (db_bits) {
2035 		qp_num = __ffs(db_bits);
2036 		qp = &nt->qp_vec[qp_num];
2037 
2038 		tasklet_schedule(&qp->rxc_db_work);
2039 
2040 		db_bits &= ~BIT_ULL(qp_num);
2041 	}
2042 }
2043 
2044 static const struct ntb_ctx_ops ntb_transport_ops = {
2045 	.link_event = ntb_transport_event_callback,
2046 	.db_event = ntb_transport_doorbell_callback,
2047 };
2048 
2049 static struct ntb_client ntb_transport_client = {
2050 	.ops = {
2051 		.probe = ntb_transport_probe,
2052 		.remove = ntb_transport_free,
2053 	},
2054 };
2055 
2056 static int __init ntb_transport_init(void)
2057 {
2058 	int rc;
2059 
2060 	pr_info("%s, version %s\n", NTB_TRANSPORT_DESC, NTB_TRANSPORT_VER);
2061 
2062 	if (debugfs_initialized())
2063 		nt_debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
2064 
2065 	rc = bus_register(&ntb_transport_bus);
2066 	if (rc)
2067 		goto err_bus;
2068 
2069 	rc = ntb_register_client(&ntb_transport_client);
2070 	if (rc)
2071 		goto err_client;
2072 
2073 	return 0;
2074 
2075 err_client:
2076 	bus_unregister(&ntb_transport_bus);
2077 err_bus:
2078 	debugfs_remove_recursive(nt_debugfs_dir);
2079 	return rc;
2080 }
2081 module_init(ntb_transport_init);
2082 
2083 static void __exit ntb_transport_exit(void)
2084 {
2085 	debugfs_remove_recursive(nt_debugfs_dir);
2086 
2087 	ntb_unregister_client(&ntb_transport_client);
2088 	bus_unregister(&ntb_transport_bus);
2089 }
2090 module_exit(ntb_transport_exit);
2091