Merge tag 'pstore-v4.8-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/kees...
[sfrench/cifs-2.6.git] / drivers / ntb / ntb_transport.c
1 /*
2  * This file is provided under a dual BSD/GPLv2 license.  When using or
3  *   redistributing this file, you may do so under either license.
4  *
5  *   GPL LICENSE SUMMARY
6  *
7  *   Copyright(c) 2012 Intel Corporation. All rights reserved.
8  *   Copyright (C) 2015 EMC Corporation. All Rights Reserved.
9  *
10  *   This program is free software; you can redistribute it and/or modify
11  *   it under the terms of version 2 of the GNU General Public License as
12  *   published by the Free Software Foundation.
13  *
14  *   BSD LICENSE
15  *
16  *   Copyright(c) 2012 Intel Corporation. All rights reserved.
17  *   Copyright (C) 2015 EMC Corporation. All Rights Reserved.
18  *
19  *   Redistribution and use in source and binary forms, with or without
20  *   modification, are permitted provided that the following conditions
21  *   are met:
22  *
23  *     * Redistributions of source code must retain the above copyright
24  *       notice, this list of conditions and the following disclaimer.
25  *     * Redistributions in binary form must reproduce the above copy
26  *       notice, this list of conditions and the following disclaimer in
27  *       the documentation and/or other materials provided with the
28  *       distribution.
29  *     * Neither the name of Intel Corporation nor the names of its
30  *       contributors may be used to endorse or promote products derived
31  *       from this software without specific prior written permission.
32  *
33  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
34  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
35  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
36  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
37  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
38  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
39  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
40  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
41  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
42  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
43  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
44  *
45  * PCIe NTB Transport Linux driver
46  *
47  * Contact Information:
48  * Jon Mason <jon.mason@intel.com>
49  */
50 #include <linux/debugfs.h>
51 #include <linux/delay.h>
52 #include <linux/dmaengine.h>
53 #include <linux/dma-mapping.h>
54 #include <linux/errno.h>
55 #include <linux/export.h>
56 #include <linux/interrupt.h>
57 #include <linux/module.h>
58 #include <linux/pci.h>
59 #include <linux/slab.h>
60 #include <linux/types.h>
61 #include <linux/uaccess.h>
62 #include "linux/ntb.h"
63 #include "linux/ntb_transport.h"
64
65 #define NTB_TRANSPORT_VERSION   4
66 #define NTB_TRANSPORT_VER       "4"
67 #define NTB_TRANSPORT_NAME      "ntb_transport"
68 #define NTB_TRANSPORT_DESC      "Software Queue-Pair Transport over NTB"
69
70 MODULE_DESCRIPTION(NTB_TRANSPORT_DESC);
71 MODULE_VERSION(NTB_TRANSPORT_VER);
72 MODULE_LICENSE("Dual BSD/GPL");
73 MODULE_AUTHOR("Intel Corporation");
74
75 static unsigned long max_mw_size;
76 module_param(max_mw_size, ulong, 0644);
77 MODULE_PARM_DESC(max_mw_size, "Limit size of large memory windows");
78
79 static unsigned int transport_mtu = 0x10000;
80 module_param(transport_mtu, uint, 0644);
81 MODULE_PARM_DESC(transport_mtu, "Maximum size of NTB transport packets");
82
83 static unsigned char max_num_clients;
84 module_param(max_num_clients, byte, 0644);
85 MODULE_PARM_DESC(max_num_clients, "Maximum number of NTB transport clients");
86
87 static unsigned int copy_bytes = 1024;
88 module_param(copy_bytes, uint, 0644);
89 MODULE_PARM_DESC(copy_bytes, "Threshold under which NTB will use the CPU to copy instead of DMA");
90
91 static bool use_dma;
92 module_param(use_dma, bool, 0644);
93 MODULE_PARM_DESC(use_dma, "Use DMA engine to perform large data copy");
94
95 static struct dentry *nt_debugfs_dir;
96
97 struct ntb_queue_entry {
98         /* ntb_queue list reference */
99         struct list_head entry;
100         /* pointers to data to be transferred */
101         void *cb_data;
102         void *buf;
103         unsigned int len;
104         unsigned int flags;
105
106         struct ntb_transport_qp *qp;
107         union {
108                 struct ntb_payload_header __iomem *tx_hdr;
109                 struct ntb_payload_header *rx_hdr;
110         };
111         unsigned int index;
112 };
113
114 struct ntb_rx_info {
115         unsigned int entry;
116 };
117
118 struct ntb_transport_qp {
119         struct ntb_transport_ctx *transport;
120         struct ntb_dev *ndev;
121         void *cb_data;
122         struct dma_chan *tx_dma_chan;
123         struct dma_chan *rx_dma_chan;
124
125         bool client_ready;
126         bool link_is_up;
127         bool active;
128
129         u8 qp_num;      /* Only 64 QP's are allowed.  0-63 */
130         u64 qp_bit;
131
132         struct ntb_rx_info __iomem *rx_info;
133         struct ntb_rx_info *remote_rx_info;
134
135         void (*tx_handler)(struct ntb_transport_qp *qp, void *qp_data,
136                            void *data, int len);
137         struct list_head tx_free_q;
138         spinlock_t ntb_tx_free_q_lock;
139         void __iomem *tx_mw;
140         dma_addr_t tx_mw_phys;
141         unsigned int tx_index;
142         unsigned int tx_max_entry;
143         unsigned int tx_max_frame;
144
145         void (*rx_handler)(struct ntb_transport_qp *qp, void *qp_data,
146                            void *data, int len);
147         struct list_head rx_post_q;
148         struct list_head rx_pend_q;
149         struct list_head rx_free_q;
150         /* ntb_rx_q_lock: synchronize access to rx_XXXX_q */
151         spinlock_t ntb_rx_q_lock;
152         void *rx_buff;
153         unsigned int rx_index;
154         unsigned int rx_max_entry;
155         unsigned int rx_max_frame;
156         dma_cookie_t last_cookie;
157         struct tasklet_struct rxc_db_work;
158
159         void (*event_handler)(void *data, int status);
160         struct delayed_work link_work;
161         struct work_struct link_cleanup;
162
163         struct dentry *debugfs_dir;
164         struct dentry *debugfs_stats;
165
166         /* Stats */
167         u64 rx_bytes;
168         u64 rx_pkts;
169         u64 rx_ring_empty;
170         u64 rx_err_no_buf;
171         u64 rx_err_oflow;
172         u64 rx_err_ver;
173         u64 rx_memcpy;
174         u64 rx_async;
175         u64 dma_rx_prep_err;
176         u64 tx_bytes;
177         u64 tx_pkts;
178         u64 tx_ring_full;
179         u64 tx_err_no_buf;
180         u64 tx_memcpy;
181         u64 tx_async;
182         u64 dma_tx_prep_err;
183 };
184
185 struct ntb_transport_mw {
186         phys_addr_t phys_addr;
187         resource_size_t phys_size;
188         resource_size_t xlat_align;
189         resource_size_t xlat_align_size;
190         void __iomem *vbase;
191         size_t xlat_size;
192         size_t buff_size;
193         void *virt_addr;
194         dma_addr_t dma_addr;
195 };
196
197 struct ntb_transport_client_dev {
198         struct list_head entry;
199         struct ntb_transport_ctx *nt;
200         struct device dev;
201 };
202
203 struct ntb_transport_ctx {
204         struct list_head entry;
205         struct list_head client_devs;
206
207         struct ntb_dev *ndev;
208
209         struct ntb_transport_mw *mw_vec;
210         struct ntb_transport_qp *qp_vec;
211         unsigned int mw_count;
212         unsigned int qp_count;
213         u64 qp_bitmap;
214         u64 qp_bitmap_free;
215
216         bool link_is_up;
217         struct delayed_work link_work;
218         struct work_struct link_cleanup;
219
220         struct dentry *debugfs_node_dir;
221 };
222
223 enum {
224         DESC_DONE_FLAG = BIT(0),
225         LINK_DOWN_FLAG = BIT(1),
226 };
227
228 struct ntb_payload_header {
229         unsigned int ver;
230         unsigned int len;
231         unsigned int flags;
232 };
233
234 enum {
235         VERSION = 0,
236         QP_LINKS,
237         NUM_QPS,
238         NUM_MWS,
239         MW0_SZ_HIGH,
240         MW0_SZ_LOW,
241         MW1_SZ_HIGH,
242         MW1_SZ_LOW,
243         MAX_SPAD,
244 };
245
246 #define dev_client_dev(__dev) \
247         container_of((__dev), struct ntb_transport_client_dev, dev)
248
249 #define drv_client(__drv) \
250         container_of((__drv), struct ntb_transport_client, driver)
251
252 #define QP_TO_MW(nt, qp)        ((qp) % nt->mw_count)
253 #define NTB_QP_DEF_NUM_ENTRIES  100
254 #define NTB_LINK_DOWN_TIMEOUT   10
255 #define DMA_RETRIES             20
256 #define DMA_OUT_RESOURCE_TO     50
257
258 static void ntb_transport_rxc_db(unsigned long data);
259 static const struct ntb_ctx_ops ntb_transport_ops;
260 static struct ntb_client ntb_transport_client;
261
262 static int ntb_transport_bus_match(struct device *dev,
263                                    struct device_driver *drv)
264 {
265         return !strncmp(dev_name(dev), drv->name, strlen(drv->name));
266 }
267
268 static int ntb_transport_bus_probe(struct device *dev)
269 {
270         const struct ntb_transport_client *client;
271         int rc = -EINVAL;
272
273         get_device(dev);
274
275         client = drv_client(dev->driver);
276         rc = client->probe(dev);
277         if (rc)
278                 put_device(dev);
279
280         return rc;
281 }
282
283 static int ntb_transport_bus_remove(struct device *dev)
284 {
285         const struct ntb_transport_client *client;
286
287         client = drv_client(dev->driver);
288         client->remove(dev);
289
290         put_device(dev);
291
292         return 0;
293 }
294
295 static struct bus_type ntb_transport_bus = {
296         .name = "ntb_transport",
297         .match = ntb_transport_bus_match,
298         .probe = ntb_transport_bus_probe,
299         .remove = ntb_transport_bus_remove,
300 };
301
302 static LIST_HEAD(ntb_transport_list);
303
304 static int ntb_bus_init(struct ntb_transport_ctx *nt)
305 {
306         list_add_tail(&nt->entry, &ntb_transport_list);
307         return 0;
308 }
309
310 static void ntb_bus_remove(struct ntb_transport_ctx *nt)
311 {
312         struct ntb_transport_client_dev *client_dev, *cd;
313
314         list_for_each_entry_safe(client_dev, cd, &nt->client_devs, entry) {
315                 dev_err(client_dev->dev.parent, "%s still attached to bus, removing\n",
316                         dev_name(&client_dev->dev));
317                 list_del(&client_dev->entry);
318                 device_unregister(&client_dev->dev);
319         }
320
321         list_del(&nt->entry);
322 }
323
324 static void ntb_transport_client_release(struct device *dev)
325 {
326         struct ntb_transport_client_dev *client_dev;
327
328         client_dev = dev_client_dev(dev);
329         kfree(client_dev);
330 }
331
332 /**
333  * ntb_transport_unregister_client_dev - Unregister NTB client device
334  * @device_name: Name of NTB client device
335  *
336  * Unregister an NTB client device with the NTB transport layer
337  */
338 void ntb_transport_unregister_client_dev(char *device_name)
339 {
340         struct ntb_transport_client_dev *client, *cd;
341         struct ntb_transport_ctx *nt;
342
343         list_for_each_entry(nt, &ntb_transport_list, entry)
344                 list_for_each_entry_safe(client, cd, &nt->client_devs, entry)
345                         if (!strncmp(dev_name(&client->dev), device_name,
346                                      strlen(device_name))) {
347                                 list_del(&client->entry);
348                                 device_unregister(&client->dev);
349                         }
350 }
351 EXPORT_SYMBOL_GPL(ntb_transport_unregister_client_dev);
352
353 /**
354  * ntb_transport_register_client_dev - Register NTB client device
355  * @device_name: Name of NTB client device
356  *
357  * Register an NTB client device with the NTB transport layer
358  */
359 int ntb_transport_register_client_dev(char *device_name)
360 {
361         struct ntb_transport_client_dev *client_dev;
362         struct ntb_transport_ctx *nt;
363         int node;
364         int rc, i = 0;
365
366         if (list_empty(&ntb_transport_list))
367                 return -ENODEV;
368
369         list_for_each_entry(nt, &ntb_transport_list, entry) {
370                 struct device *dev;
371
372                 node = dev_to_node(&nt->ndev->dev);
373
374                 client_dev = kzalloc_node(sizeof(*client_dev),
375                                           GFP_KERNEL, node);
376                 if (!client_dev) {
377                         rc = -ENOMEM;
378                         goto err;
379                 }
380
381                 dev = &client_dev->dev;
382
383                 /* setup and register client devices */
384                 dev_set_name(dev, "%s%d", device_name, i);
385                 dev->bus = &ntb_transport_bus;
386                 dev->release = ntb_transport_client_release;
387                 dev->parent = &nt->ndev->dev;
388
389                 rc = device_register(dev);
390                 if (rc) {
391                         kfree(client_dev);
392                         goto err;
393                 }
394
395                 list_add_tail(&client_dev->entry, &nt->client_devs);
396                 i++;
397         }
398
399         return 0;
400
401 err:
402         ntb_transport_unregister_client_dev(device_name);
403
404         return rc;
405 }
406 EXPORT_SYMBOL_GPL(ntb_transport_register_client_dev);
407
408 /**
409  * ntb_transport_register_client - Register NTB client driver
410  * @drv: NTB client driver to be registered
411  *
412  * Register an NTB client driver with the NTB transport layer
413  *
414  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
415  */
416 int ntb_transport_register_client(struct ntb_transport_client *drv)
417 {
418         drv->driver.bus = &ntb_transport_bus;
419
420         if (list_empty(&ntb_transport_list))
421                 return -ENODEV;
422
423         return driver_register(&drv->driver);
424 }
425 EXPORT_SYMBOL_GPL(ntb_transport_register_client);
426
427 /**
428  * ntb_transport_unregister_client - Unregister NTB client driver
429  * @drv: NTB client driver to be unregistered
430  *
431  * Unregister an NTB client driver with the NTB transport layer
432  *
433  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
434  */
435 void ntb_transport_unregister_client(struct ntb_transport_client *drv)
436 {
437         driver_unregister(&drv->driver);
438 }
439 EXPORT_SYMBOL_GPL(ntb_transport_unregister_client);
440
441 static ssize_t debugfs_read(struct file *filp, char __user *ubuf, size_t count,
442                             loff_t *offp)
443 {
444         struct ntb_transport_qp *qp;
445         char *buf;
446         ssize_t ret, out_offset, out_count;
447
448         qp = filp->private_data;
449
450         if (!qp || !qp->link_is_up)
451                 return 0;
452
453         out_count = 1000;
454
455         buf = kmalloc(out_count, GFP_KERNEL);
456         if (!buf)
457                 return -ENOMEM;
458
459         out_offset = 0;
460         out_offset += snprintf(buf + out_offset, out_count - out_offset,
461                                "\nNTB QP stats:\n\n");
462         out_offset += snprintf(buf + out_offset, out_count - out_offset,
463                                "rx_bytes - \t%llu\n", qp->rx_bytes);
464         out_offset += snprintf(buf + out_offset, out_count - out_offset,
465                                "rx_pkts - \t%llu\n", qp->rx_pkts);
466         out_offset += snprintf(buf + out_offset, out_count - out_offset,
467                                "rx_memcpy - \t%llu\n", qp->rx_memcpy);
468         out_offset += snprintf(buf + out_offset, out_count - out_offset,
469                                "rx_async - \t%llu\n", qp->rx_async);
470         out_offset += snprintf(buf + out_offset, out_count - out_offset,
471                                "rx_ring_empty - %llu\n", qp->rx_ring_empty);
472         out_offset += snprintf(buf + out_offset, out_count - out_offset,
473                                "rx_err_no_buf - %llu\n", qp->rx_err_no_buf);
474         out_offset += snprintf(buf + out_offset, out_count - out_offset,
475                                "rx_err_oflow - \t%llu\n", qp->rx_err_oflow);
476         out_offset += snprintf(buf + out_offset, out_count - out_offset,
477                                "rx_err_ver - \t%llu\n", qp->rx_err_ver);
478         out_offset += snprintf(buf + out_offset, out_count - out_offset,
479                                "rx_buff - \t0x%p\n", qp->rx_buff);
480         out_offset += snprintf(buf + out_offset, out_count - out_offset,
481                                "rx_index - \t%u\n", qp->rx_index);
482         out_offset += snprintf(buf + out_offset, out_count - out_offset,
483                                "rx_max_entry - \t%u\n\n", qp->rx_max_entry);
484
485         out_offset += snprintf(buf + out_offset, out_count - out_offset,
486                                "tx_bytes - \t%llu\n", qp->tx_bytes);
487         out_offset += snprintf(buf + out_offset, out_count - out_offset,
488                                "tx_pkts - \t%llu\n", qp->tx_pkts);
489         out_offset += snprintf(buf + out_offset, out_count - out_offset,
490                                "tx_memcpy - \t%llu\n", qp->tx_memcpy);
491         out_offset += snprintf(buf + out_offset, out_count - out_offset,
492                                "tx_async - \t%llu\n", qp->tx_async);
493         out_offset += snprintf(buf + out_offset, out_count - out_offset,
494                                "tx_ring_full - \t%llu\n", qp->tx_ring_full);
495         out_offset += snprintf(buf + out_offset, out_count - out_offset,
496                                "tx_err_no_buf - %llu\n", qp->tx_err_no_buf);
497         out_offset += snprintf(buf + out_offset, out_count - out_offset,
498                                "tx_mw - \t0x%p\n", qp->tx_mw);
499         out_offset += snprintf(buf + out_offset, out_count - out_offset,
500                                "tx_index (H) - \t%u\n", qp->tx_index);
501         out_offset += snprintf(buf + out_offset, out_count - out_offset,
502                                "RRI (T) - \t%u\n",
503                                qp->remote_rx_info->entry);
504         out_offset += snprintf(buf + out_offset, out_count - out_offset,
505                                "tx_max_entry - \t%u\n", qp->tx_max_entry);
506         out_offset += snprintf(buf + out_offset, out_count - out_offset,
507                                "free tx - \t%u\n",
508                                ntb_transport_tx_free_entry(qp));
509         out_offset += snprintf(buf + out_offset, out_count - out_offset,
510                                "DMA tx prep err - \t%llu\n",
511                                qp->dma_tx_prep_err);
512         out_offset += snprintf(buf + out_offset, out_count - out_offset,
513                                "DMA rx prep err - \t%llu\n",
514                                qp->dma_rx_prep_err);
515
516         out_offset += snprintf(buf + out_offset, out_count - out_offset,
517                                "\n");
518         out_offset += snprintf(buf + out_offset, out_count - out_offset,
519                                "Using TX DMA - \t%s\n",
520                                qp->tx_dma_chan ? "Yes" : "No");
521         out_offset += snprintf(buf + out_offset, out_count - out_offset,
522                                "Using RX DMA - \t%s\n",
523                                qp->rx_dma_chan ? "Yes" : "No");
524         out_offset += snprintf(buf + out_offset, out_count - out_offset,
525                                "QP Link - \t%s\n",
526                                qp->link_is_up ? "Up" : "Down");
527         out_offset += snprintf(buf + out_offset, out_count - out_offset,
528                                "\n");
529
530         if (out_offset > out_count)
531                 out_offset = out_count;
532
533         ret = simple_read_from_buffer(ubuf, count, offp, buf, out_offset);
534         kfree(buf);
535         return ret;
536 }
537
538 static const struct file_operations ntb_qp_debugfs_stats = {
539         .owner = THIS_MODULE,
540         .open = simple_open,
541         .read = debugfs_read,
542 };
543
544 static void ntb_list_add(spinlock_t *lock, struct list_head *entry,
545                          struct list_head *list)
546 {
547         unsigned long flags;
548
549         spin_lock_irqsave(lock, flags);
550         list_add_tail(entry, list);
551         spin_unlock_irqrestore(lock, flags);
552 }
553
554 static struct ntb_queue_entry *ntb_list_rm(spinlock_t *lock,
555                                            struct list_head *list)
556 {
557         struct ntb_queue_entry *entry;
558         unsigned long flags;
559
560         spin_lock_irqsave(lock, flags);
561         if (list_empty(list)) {
562                 entry = NULL;
563                 goto out;
564         }
565         entry = list_first_entry(list, struct ntb_queue_entry, entry);
566         list_del(&entry->entry);
567
568 out:
569         spin_unlock_irqrestore(lock, flags);
570
571         return entry;
572 }
573
574 static struct ntb_queue_entry *ntb_list_mv(spinlock_t *lock,
575                                            struct list_head *list,
576                                            struct list_head *to_list)
577 {
578         struct ntb_queue_entry *entry;
579         unsigned long flags;
580
581         spin_lock_irqsave(lock, flags);
582
583         if (list_empty(list)) {
584                 entry = NULL;
585         } else {
586                 entry = list_first_entry(list, struct ntb_queue_entry, entry);
587                 list_move_tail(&entry->entry, to_list);
588         }
589
590         spin_unlock_irqrestore(lock, flags);
591
592         return entry;
593 }
594
595 static int ntb_transport_setup_qp_mw(struct ntb_transport_ctx *nt,
596                                      unsigned int qp_num)
597 {
598         struct ntb_transport_qp *qp = &nt->qp_vec[qp_num];
599         struct ntb_transport_mw *mw;
600         unsigned int rx_size, num_qps_mw;
601         unsigned int mw_num, mw_count, qp_count;
602         unsigned int i;
603
604         mw_count = nt->mw_count;
605         qp_count = nt->qp_count;
606
607         mw_num = QP_TO_MW(nt, qp_num);
608         mw = &nt->mw_vec[mw_num];
609
610         if (!mw->virt_addr)
611                 return -ENOMEM;
612
613         if (qp_count % mw_count && mw_num + 1 < qp_count / mw_count)
614                 num_qps_mw = qp_count / mw_count + 1;
615         else
616                 num_qps_mw = qp_count / mw_count;
617
618         rx_size = (unsigned int)mw->xlat_size / num_qps_mw;
619         qp->rx_buff = mw->virt_addr + rx_size * (qp_num / mw_count);
620         rx_size -= sizeof(struct ntb_rx_info);
621
622         qp->remote_rx_info = qp->rx_buff + rx_size;
623
624         /* Due to housekeeping, there must be atleast 2 buffs */
625         qp->rx_max_frame = min(transport_mtu, rx_size / 2);
626         qp->rx_max_entry = rx_size / qp->rx_max_frame;
627         qp->rx_index = 0;
628
629         qp->remote_rx_info->entry = qp->rx_max_entry - 1;
630
631         /* setup the hdr offsets with 0's */
632         for (i = 0; i < qp->rx_max_entry; i++) {
633                 void *offset = (qp->rx_buff + qp->rx_max_frame * (i + 1) -
634                                 sizeof(struct ntb_payload_header));
635                 memset(offset, 0, sizeof(struct ntb_payload_header));
636         }
637
638         qp->rx_pkts = 0;
639         qp->tx_pkts = 0;
640         qp->tx_index = 0;
641
642         return 0;
643 }
644
645 static void ntb_free_mw(struct ntb_transport_ctx *nt, int num_mw)
646 {
647         struct ntb_transport_mw *mw = &nt->mw_vec[num_mw];
648         struct pci_dev *pdev = nt->ndev->pdev;
649
650         if (!mw->virt_addr)
651                 return;
652
653         ntb_mw_clear_trans(nt->ndev, num_mw);
654         dma_free_coherent(&pdev->dev, mw->buff_size,
655                           mw->virt_addr, mw->dma_addr);
656         mw->xlat_size = 0;
657         mw->buff_size = 0;
658         mw->virt_addr = NULL;
659 }
660
661 static int ntb_set_mw(struct ntb_transport_ctx *nt, int num_mw,
662                       resource_size_t size)
663 {
664         struct ntb_transport_mw *mw = &nt->mw_vec[num_mw];
665         struct pci_dev *pdev = nt->ndev->pdev;
666         size_t xlat_size, buff_size;
667         int rc;
668
669         if (!size)
670                 return -EINVAL;
671
672         xlat_size = round_up(size, mw->xlat_align_size);
673         buff_size = round_up(size, mw->xlat_align);
674
675         /* No need to re-setup */
676         if (mw->xlat_size == xlat_size)
677                 return 0;
678
679         if (mw->buff_size)
680                 ntb_free_mw(nt, num_mw);
681
682         /* Alloc memory for receiving data.  Must be aligned */
683         mw->xlat_size = xlat_size;
684         mw->buff_size = buff_size;
685
686         mw->virt_addr = dma_alloc_coherent(&pdev->dev, buff_size,
687                                            &mw->dma_addr, GFP_KERNEL);
688         if (!mw->virt_addr) {
689                 mw->xlat_size = 0;
690                 mw->buff_size = 0;
691                 dev_err(&pdev->dev, "Unable to alloc MW buff of size %zu\n",
692                         buff_size);
693                 return -ENOMEM;
694         }
695
696         /*
697          * we must ensure that the memory address allocated is BAR size
698          * aligned in order for the XLAT register to take the value. This
699          * is a requirement of the hardware. It is recommended to setup CMA
700          * for BAR sizes equal or greater than 4MB.
701          */
702         if (!IS_ALIGNED(mw->dma_addr, mw->xlat_align)) {
703                 dev_err(&pdev->dev, "DMA memory %pad is not aligned\n",
704                         &mw->dma_addr);
705                 ntb_free_mw(nt, num_mw);
706                 return -ENOMEM;
707         }
708
709         /* Notify HW the memory location of the receive buffer */
710         rc = ntb_mw_set_trans(nt->ndev, num_mw, mw->dma_addr, mw->xlat_size);
711         if (rc) {
712                 dev_err(&pdev->dev, "Unable to set mw%d translation", num_mw);
713                 ntb_free_mw(nt, num_mw);
714                 return -EIO;
715         }
716
717         return 0;
718 }
719
720 static void ntb_qp_link_down_reset(struct ntb_transport_qp *qp)
721 {
722         qp->link_is_up = false;
723         qp->active = false;
724
725         qp->tx_index = 0;
726         qp->rx_index = 0;
727         qp->rx_bytes = 0;
728         qp->rx_pkts = 0;
729         qp->rx_ring_empty = 0;
730         qp->rx_err_no_buf = 0;
731         qp->rx_err_oflow = 0;
732         qp->rx_err_ver = 0;
733         qp->rx_memcpy = 0;
734         qp->rx_async = 0;
735         qp->tx_bytes = 0;
736         qp->tx_pkts = 0;
737         qp->tx_ring_full = 0;
738         qp->tx_err_no_buf = 0;
739         qp->tx_memcpy = 0;
740         qp->tx_async = 0;
741         qp->dma_tx_prep_err = 0;
742         qp->dma_rx_prep_err = 0;
743 }
744
745 static void ntb_qp_link_cleanup(struct ntb_transport_qp *qp)
746 {
747         struct ntb_transport_ctx *nt = qp->transport;
748         struct pci_dev *pdev = nt->ndev->pdev;
749
750         dev_info(&pdev->dev, "qp %d: Link Cleanup\n", qp->qp_num);
751
752         cancel_delayed_work_sync(&qp->link_work);
753         ntb_qp_link_down_reset(qp);
754
755         if (qp->event_handler)
756                 qp->event_handler(qp->cb_data, qp->link_is_up);
757 }
758
759 static void ntb_qp_link_cleanup_work(struct work_struct *work)
760 {
761         struct ntb_transport_qp *qp = container_of(work,
762                                                    struct ntb_transport_qp,
763                                                    link_cleanup);
764         struct ntb_transport_ctx *nt = qp->transport;
765
766         ntb_qp_link_cleanup(qp);
767
768         if (nt->link_is_up)
769                 schedule_delayed_work(&qp->link_work,
770                                       msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT));
771 }
772
773 static void ntb_qp_link_down(struct ntb_transport_qp *qp)
774 {
775         schedule_work(&qp->link_cleanup);
776 }
777
778 static void ntb_transport_link_cleanup(struct ntb_transport_ctx *nt)
779 {
780         struct ntb_transport_qp *qp;
781         u64 qp_bitmap_alloc;
782         int i;
783
784         qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free;
785
786         /* Pass along the info to any clients */
787         for (i = 0; i < nt->qp_count; i++)
788                 if (qp_bitmap_alloc & BIT_ULL(i)) {
789                         qp = &nt->qp_vec[i];
790                         ntb_qp_link_cleanup(qp);
791                         cancel_work_sync(&qp->link_cleanup);
792                         cancel_delayed_work_sync(&qp->link_work);
793                 }
794
795         if (!nt->link_is_up)
796                 cancel_delayed_work_sync(&nt->link_work);
797
798         /* The scratchpad registers keep the values if the remote side
799          * goes down, blast them now to give them a sane value the next
800          * time they are accessed
801          */
802         for (i = 0; i < MAX_SPAD; i++)
803                 ntb_spad_write(nt->ndev, i, 0);
804 }
805
806 static void ntb_transport_link_cleanup_work(struct work_struct *work)
807 {
808         struct ntb_transport_ctx *nt =
809                 container_of(work, struct ntb_transport_ctx, link_cleanup);
810
811         ntb_transport_link_cleanup(nt);
812 }
813
814 static void ntb_transport_event_callback(void *data)
815 {
816         struct ntb_transport_ctx *nt = data;
817
818         if (ntb_link_is_up(nt->ndev, NULL, NULL) == 1)
819                 schedule_delayed_work(&nt->link_work, 0);
820         else
821                 schedule_work(&nt->link_cleanup);
822 }
823
824 static void ntb_transport_link_work(struct work_struct *work)
825 {
826         struct ntb_transport_ctx *nt =
827                 container_of(work, struct ntb_transport_ctx, link_work.work);
828         struct ntb_dev *ndev = nt->ndev;
829         struct pci_dev *pdev = ndev->pdev;
830         resource_size_t size;
831         u32 val;
832         int rc = 0, i, spad;
833
834         /* send the local info, in the opposite order of the way we read it */
835         for (i = 0; i < nt->mw_count; i++) {
836                 size = nt->mw_vec[i].phys_size;
837
838                 if (max_mw_size && size > max_mw_size)
839                         size = max_mw_size;
840
841                 spad = MW0_SZ_HIGH + (i * 2);
842                 ntb_peer_spad_write(ndev, spad, upper_32_bits(size));
843
844                 spad = MW0_SZ_LOW + (i * 2);
845                 ntb_peer_spad_write(ndev, spad, lower_32_bits(size));
846         }
847
848         ntb_peer_spad_write(ndev, NUM_MWS, nt->mw_count);
849
850         ntb_peer_spad_write(ndev, NUM_QPS, nt->qp_count);
851
852         ntb_peer_spad_write(ndev, VERSION, NTB_TRANSPORT_VERSION);
853
854         /* Query the remote side for its info */
855         val = ntb_spad_read(ndev, VERSION);
856         dev_dbg(&pdev->dev, "Remote version = %d\n", val);
857         if (val != NTB_TRANSPORT_VERSION)
858                 goto out;
859
860         val = ntb_spad_read(ndev, NUM_QPS);
861         dev_dbg(&pdev->dev, "Remote max number of qps = %d\n", val);
862         if (val != nt->qp_count)
863                 goto out;
864
865         val = ntb_spad_read(ndev, NUM_MWS);
866         dev_dbg(&pdev->dev, "Remote number of mws = %d\n", val);
867         if (val != nt->mw_count)
868                 goto out;
869
870         for (i = 0; i < nt->mw_count; i++) {
871                 u64 val64;
872
873                 val = ntb_spad_read(ndev, MW0_SZ_HIGH + (i * 2));
874                 val64 = (u64)val << 32;
875
876                 val = ntb_spad_read(ndev, MW0_SZ_LOW + (i * 2));
877                 val64 |= val;
878
879                 dev_dbg(&pdev->dev, "Remote MW%d size = %#llx\n", i, val64);
880
881                 rc = ntb_set_mw(nt, i, val64);
882                 if (rc)
883                         goto out1;
884         }
885
886         nt->link_is_up = true;
887
888         for (i = 0; i < nt->qp_count; i++) {
889                 struct ntb_transport_qp *qp = &nt->qp_vec[i];
890
891                 ntb_transport_setup_qp_mw(nt, i);
892
893                 if (qp->client_ready)
894                         schedule_delayed_work(&qp->link_work, 0);
895         }
896
897         return;
898
899 out1:
900         for (i = 0; i < nt->mw_count; i++)
901                 ntb_free_mw(nt, i);
902
903         /* if there's an actual failure, we should just bail */
904         if (rc < 0) {
905                 ntb_link_disable(ndev);
906                 return;
907         }
908
909 out:
910         if (ntb_link_is_up(ndev, NULL, NULL) == 1)
911                 schedule_delayed_work(&nt->link_work,
912                                       msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT));
913 }
914
915 static void ntb_qp_link_work(struct work_struct *work)
916 {
917         struct ntb_transport_qp *qp = container_of(work,
918                                                    struct ntb_transport_qp,
919                                                    link_work.work);
920         struct pci_dev *pdev = qp->ndev->pdev;
921         struct ntb_transport_ctx *nt = qp->transport;
922         int val;
923
924         WARN_ON(!nt->link_is_up);
925
926         val = ntb_spad_read(nt->ndev, QP_LINKS);
927
928         ntb_peer_spad_write(nt->ndev, QP_LINKS, val | BIT(qp->qp_num));
929
930         /* query remote spad for qp ready bits */
931         ntb_peer_spad_read(nt->ndev, QP_LINKS);
932         dev_dbg_ratelimited(&pdev->dev, "Remote QP link status = %x\n", val);
933
934         /* See if the remote side is up */
935         if (val & BIT(qp->qp_num)) {
936                 dev_info(&pdev->dev, "qp %d: Link Up\n", qp->qp_num);
937                 qp->link_is_up = true;
938                 qp->active = true;
939
940                 if (qp->event_handler)
941                         qp->event_handler(qp->cb_data, qp->link_is_up);
942
943                 if (qp->active)
944                         tasklet_schedule(&qp->rxc_db_work);
945         } else if (nt->link_is_up)
946                 schedule_delayed_work(&qp->link_work,
947                                       msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT));
948 }
949
950 static int ntb_transport_init_queue(struct ntb_transport_ctx *nt,
951                                     unsigned int qp_num)
952 {
953         struct ntb_transport_qp *qp;
954         phys_addr_t mw_base;
955         resource_size_t mw_size;
956         unsigned int num_qps_mw, tx_size;
957         unsigned int mw_num, mw_count, qp_count;
958         u64 qp_offset;
959
960         mw_count = nt->mw_count;
961         qp_count = nt->qp_count;
962
963         mw_num = QP_TO_MW(nt, qp_num);
964
965         qp = &nt->qp_vec[qp_num];
966         qp->qp_num = qp_num;
967         qp->transport = nt;
968         qp->ndev = nt->ndev;
969         qp->client_ready = false;
970         qp->event_handler = NULL;
971         ntb_qp_link_down_reset(qp);
972
973         if (qp_count % mw_count && mw_num + 1 < qp_count / mw_count)
974                 num_qps_mw = qp_count / mw_count + 1;
975         else
976                 num_qps_mw = qp_count / mw_count;
977
978         mw_base = nt->mw_vec[mw_num].phys_addr;
979         mw_size = nt->mw_vec[mw_num].phys_size;
980
981         tx_size = (unsigned int)mw_size / num_qps_mw;
982         qp_offset = tx_size * (qp_num / mw_count);
983
984         qp->tx_mw = nt->mw_vec[mw_num].vbase + qp_offset;
985         if (!qp->tx_mw)
986                 return -EINVAL;
987
988         qp->tx_mw_phys = mw_base + qp_offset;
989         if (!qp->tx_mw_phys)
990                 return -EINVAL;
991
992         tx_size -= sizeof(struct ntb_rx_info);
993         qp->rx_info = qp->tx_mw + tx_size;
994
995         /* Due to housekeeping, there must be atleast 2 buffs */
996         qp->tx_max_frame = min(transport_mtu, tx_size / 2);
997         qp->tx_max_entry = tx_size / qp->tx_max_frame;
998
999         if (nt->debugfs_node_dir) {
1000                 char debugfs_name[4];
1001
1002                 snprintf(debugfs_name, 4, "qp%d", qp_num);
1003                 qp->debugfs_dir = debugfs_create_dir(debugfs_name,
1004                                                      nt->debugfs_node_dir);
1005
1006                 qp->debugfs_stats = debugfs_create_file("stats", S_IRUSR,
1007                                                         qp->debugfs_dir, qp,
1008                                                         &ntb_qp_debugfs_stats);
1009         } else {
1010                 qp->debugfs_dir = NULL;
1011                 qp->debugfs_stats = NULL;
1012         }
1013
1014         INIT_DELAYED_WORK(&qp->link_work, ntb_qp_link_work);
1015         INIT_WORK(&qp->link_cleanup, ntb_qp_link_cleanup_work);
1016
1017         spin_lock_init(&qp->ntb_rx_q_lock);
1018         spin_lock_init(&qp->ntb_tx_free_q_lock);
1019
1020         INIT_LIST_HEAD(&qp->rx_post_q);
1021         INIT_LIST_HEAD(&qp->rx_pend_q);
1022         INIT_LIST_HEAD(&qp->rx_free_q);
1023         INIT_LIST_HEAD(&qp->tx_free_q);
1024
1025         tasklet_init(&qp->rxc_db_work, ntb_transport_rxc_db,
1026                      (unsigned long)qp);
1027
1028         return 0;
1029 }
1030
1031 static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev)
1032 {
1033         struct ntb_transport_ctx *nt;
1034         struct ntb_transport_mw *mw;
1035         unsigned int mw_count, qp_count;
1036         u64 qp_bitmap;
1037         int node;
1038         int rc, i;
1039
1040         if (ntb_db_is_unsafe(ndev))
1041                 dev_dbg(&ndev->dev,
1042                         "doorbell is unsafe, proceed anyway...\n");
1043         if (ntb_spad_is_unsafe(ndev))
1044                 dev_dbg(&ndev->dev,
1045                         "scratchpad is unsafe, proceed anyway...\n");
1046
1047         node = dev_to_node(&ndev->dev);
1048
1049         nt = kzalloc_node(sizeof(*nt), GFP_KERNEL, node);
1050         if (!nt)
1051                 return -ENOMEM;
1052
1053         nt->ndev = ndev;
1054
1055         mw_count = ntb_mw_count(ndev);
1056
1057         nt->mw_count = mw_count;
1058
1059         nt->mw_vec = kzalloc_node(mw_count * sizeof(*nt->mw_vec),
1060                                   GFP_KERNEL, node);
1061         if (!nt->mw_vec) {
1062                 rc = -ENOMEM;
1063                 goto err;
1064         }
1065
1066         for (i = 0; i < mw_count; i++) {
1067                 mw = &nt->mw_vec[i];
1068
1069                 rc = ntb_mw_get_range(ndev, i, &mw->phys_addr, &mw->phys_size,
1070                                       &mw->xlat_align, &mw->xlat_align_size);
1071                 if (rc)
1072                         goto err1;
1073
1074                 mw->vbase = ioremap_wc(mw->phys_addr, mw->phys_size);
1075                 if (!mw->vbase) {
1076                         rc = -ENOMEM;
1077                         goto err1;
1078                 }
1079
1080                 mw->buff_size = 0;
1081                 mw->xlat_size = 0;
1082                 mw->virt_addr = NULL;
1083                 mw->dma_addr = 0;
1084         }
1085
1086         qp_bitmap = ntb_db_valid_mask(ndev);
1087
1088         qp_count = ilog2(qp_bitmap);
1089         if (max_num_clients && max_num_clients < qp_count)
1090                 qp_count = max_num_clients;
1091         else if (mw_count < qp_count)
1092                 qp_count = mw_count;
1093
1094         qp_bitmap &= BIT_ULL(qp_count) - 1;
1095
1096         nt->qp_count = qp_count;
1097         nt->qp_bitmap = qp_bitmap;
1098         nt->qp_bitmap_free = qp_bitmap;
1099
1100         nt->qp_vec = kzalloc_node(qp_count * sizeof(*nt->qp_vec),
1101                                   GFP_KERNEL, node);
1102         if (!nt->qp_vec) {
1103                 rc = -ENOMEM;
1104                 goto err1;
1105         }
1106
1107         if (nt_debugfs_dir) {
1108                 nt->debugfs_node_dir =
1109                         debugfs_create_dir(pci_name(ndev->pdev),
1110                                            nt_debugfs_dir);
1111         }
1112
1113         for (i = 0; i < qp_count; i++) {
1114                 rc = ntb_transport_init_queue(nt, i);
1115                 if (rc)
1116                         goto err2;
1117         }
1118
1119         INIT_DELAYED_WORK(&nt->link_work, ntb_transport_link_work);
1120         INIT_WORK(&nt->link_cleanup, ntb_transport_link_cleanup_work);
1121
1122         rc = ntb_set_ctx(ndev, nt, &ntb_transport_ops);
1123         if (rc)
1124                 goto err2;
1125
1126         INIT_LIST_HEAD(&nt->client_devs);
1127         rc = ntb_bus_init(nt);
1128         if (rc)
1129                 goto err3;
1130
1131         nt->link_is_up = false;
1132         ntb_link_enable(ndev, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
1133         ntb_link_event(ndev);
1134
1135         return 0;
1136
1137 err3:
1138         ntb_clear_ctx(ndev);
1139 err2:
1140         kfree(nt->qp_vec);
1141 err1:
1142         while (i--) {
1143                 mw = &nt->mw_vec[i];
1144                 iounmap(mw->vbase);
1145         }
1146         kfree(nt->mw_vec);
1147 err:
1148         kfree(nt);
1149         return rc;
1150 }
1151
1152 static void ntb_transport_free(struct ntb_client *self, struct ntb_dev *ndev)
1153 {
1154         struct ntb_transport_ctx *nt = ndev->ctx;
1155         struct ntb_transport_qp *qp;
1156         u64 qp_bitmap_alloc;
1157         int i;
1158
1159         ntb_transport_link_cleanup(nt);
1160         cancel_work_sync(&nt->link_cleanup);
1161         cancel_delayed_work_sync(&nt->link_work);
1162
1163         qp_bitmap_alloc = nt->qp_bitmap & ~nt->qp_bitmap_free;
1164
1165         /* verify that all the qp's are freed */
1166         for (i = 0; i < nt->qp_count; i++) {
1167                 qp = &nt->qp_vec[i];
1168                 if (qp_bitmap_alloc & BIT_ULL(i))
1169                         ntb_transport_free_queue(qp);
1170                 debugfs_remove_recursive(qp->debugfs_dir);
1171         }
1172
1173         ntb_link_disable(ndev);
1174         ntb_clear_ctx(ndev);
1175
1176         ntb_bus_remove(nt);
1177
1178         for (i = nt->mw_count; i--; ) {
1179                 ntb_free_mw(nt, i);
1180                 iounmap(nt->mw_vec[i].vbase);
1181         }
1182
1183         kfree(nt->qp_vec);
1184         kfree(nt->mw_vec);
1185         kfree(nt);
1186 }
1187
1188 static void ntb_complete_rxc(struct ntb_transport_qp *qp)
1189 {
1190         struct ntb_queue_entry *entry;
1191         void *cb_data;
1192         unsigned int len;
1193         unsigned long irqflags;
1194
1195         spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags);
1196
1197         while (!list_empty(&qp->rx_post_q)) {
1198                 entry = list_first_entry(&qp->rx_post_q,
1199                                          struct ntb_queue_entry, entry);
1200                 if (!(entry->flags & DESC_DONE_FLAG))
1201                         break;
1202
1203                 entry->rx_hdr->flags = 0;
1204                 iowrite32(entry->index, &qp->rx_info->entry);
1205
1206                 cb_data = entry->cb_data;
1207                 len = entry->len;
1208
1209                 list_move_tail(&entry->entry, &qp->rx_free_q);
1210
1211                 spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags);
1212
1213                 if (qp->rx_handler && qp->client_ready)
1214                         qp->rx_handler(qp, qp->cb_data, cb_data, len);
1215
1216                 spin_lock_irqsave(&qp->ntb_rx_q_lock, irqflags);
1217         }
1218
1219         spin_unlock_irqrestore(&qp->ntb_rx_q_lock, irqflags);
1220 }
1221
1222 static void ntb_rx_copy_callback(void *data)
1223 {
1224         struct ntb_queue_entry *entry = data;
1225
1226         entry->flags |= DESC_DONE_FLAG;
1227
1228         ntb_complete_rxc(entry->qp);
1229 }
1230
1231 static void ntb_memcpy_rx(struct ntb_queue_entry *entry, void *offset)
1232 {
1233         void *buf = entry->buf;
1234         size_t len = entry->len;
1235
1236         memcpy(buf, offset, len);
1237
1238         /* Ensure that the data is fully copied out before clearing the flag */
1239         wmb();
1240
1241         ntb_rx_copy_callback(entry);
1242 }
1243
1244 static void ntb_async_rx(struct ntb_queue_entry *entry, void *offset)
1245 {
1246         struct dma_async_tx_descriptor *txd;
1247         struct ntb_transport_qp *qp = entry->qp;
1248         struct dma_chan *chan = qp->rx_dma_chan;
1249         struct dma_device *device;
1250         size_t pay_off, buff_off, len;
1251         struct dmaengine_unmap_data *unmap;
1252         dma_cookie_t cookie;
1253         void *buf = entry->buf;
1254         int retries = 0;
1255
1256         len = entry->len;
1257
1258         if (!chan)
1259                 goto err;
1260
1261         if (len < copy_bytes)
1262                 goto err;
1263
1264         device = chan->device;
1265         pay_off = (size_t)offset & ~PAGE_MASK;
1266         buff_off = (size_t)buf & ~PAGE_MASK;
1267
1268         if (!is_dma_copy_aligned(device, pay_off, buff_off, len))
1269                 goto err;
1270
1271         unmap = dmaengine_get_unmap_data(device->dev, 2, GFP_NOWAIT);
1272         if (!unmap)
1273                 goto err;
1274
1275         unmap->len = len;
1276         unmap->addr[0] = dma_map_page(device->dev, virt_to_page(offset),
1277                                       pay_off, len, DMA_TO_DEVICE);
1278         if (dma_mapping_error(device->dev, unmap->addr[0]))
1279                 goto err_get_unmap;
1280
1281         unmap->to_cnt = 1;
1282
1283         unmap->addr[1] = dma_map_page(device->dev, virt_to_page(buf),
1284                                       buff_off, len, DMA_FROM_DEVICE);
1285         if (dma_mapping_error(device->dev, unmap->addr[1]))
1286                 goto err_get_unmap;
1287
1288         unmap->from_cnt = 1;
1289
1290         for (retries = 0; retries < DMA_RETRIES; retries++) {
1291                 txd = device->device_prep_dma_memcpy(chan, unmap->addr[1],
1292                                                      unmap->addr[0], len,
1293                                                      DMA_PREP_INTERRUPT);
1294                 if (txd)
1295                         break;
1296
1297                 set_current_state(TASK_INTERRUPTIBLE);
1298                 schedule_timeout(DMA_OUT_RESOURCE_TO);
1299         }
1300
1301         if (!txd) {
1302                 qp->dma_rx_prep_err++;
1303                 goto err_get_unmap;
1304         }
1305
1306         txd->callback = ntb_rx_copy_callback;
1307         txd->callback_param = entry;
1308         dma_set_unmap(txd, unmap);
1309
1310         cookie = dmaengine_submit(txd);
1311         if (dma_submit_error(cookie))
1312                 goto err_set_unmap;
1313
1314         dmaengine_unmap_put(unmap);
1315
1316         qp->last_cookie = cookie;
1317
1318         qp->rx_async++;
1319
1320         return;
1321
1322 err_set_unmap:
1323         dmaengine_unmap_put(unmap);
1324 err_get_unmap:
1325         dmaengine_unmap_put(unmap);
1326 err:
1327         ntb_memcpy_rx(entry, offset);
1328         qp->rx_memcpy++;
1329 }
1330
1331 static int ntb_process_rxc(struct ntb_transport_qp *qp)
1332 {
1333         struct ntb_payload_header *hdr;
1334         struct ntb_queue_entry *entry;
1335         void *offset;
1336
1337         offset = qp->rx_buff + qp->rx_max_frame * qp->rx_index;
1338         hdr = offset + qp->rx_max_frame - sizeof(struct ntb_payload_header);
1339
1340         dev_dbg(&qp->ndev->pdev->dev, "qp %d: RX ver %u len %d flags %x\n",
1341                 qp->qp_num, hdr->ver, hdr->len, hdr->flags);
1342
1343         if (!(hdr->flags & DESC_DONE_FLAG)) {
1344                 dev_dbg(&qp->ndev->pdev->dev, "done flag not set\n");
1345                 qp->rx_ring_empty++;
1346                 return -EAGAIN;
1347         }
1348
1349         if (hdr->flags & LINK_DOWN_FLAG) {
1350                 dev_dbg(&qp->ndev->pdev->dev, "link down flag set\n");
1351                 ntb_qp_link_down(qp);
1352                 hdr->flags = 0;
1353                 return -EAGAIN;
1354         }
1355
1356         if (hdr->ver != (u32)qp->rx_pkts) {
1357                 dev_dbg(&qp->ndev->pdev->dev,
1358                         "version mismatch, expected %llu - got %u\n",
1359                         qp->rx_pkts, hdr->ver);
1360                 qp->rx_err_ver++;
1361                 return -EIO;
1362         }
1363
1364         entry = ntb_list_mv(&qp->ntb_rx_q_lock, &qp->rx_pend_q, &qp->rx_post_q);
1365         if (!entry) {
1366                 dev_dbg(&qp->ndev->pdev->dev, "no receive buffer\n");
1367                 qp->rx_err_no_buf++;
1368                 return -EAGAIN;
1369         }
1370
1371         entry->rx_hdr = hdr;
1372         entry->index = qp->rx_index;
1373
1374         if (hdr->len > entry->len) {
1375                 dev_dbg(&qp->ndev->pdev->dev,
1376                         "receive buffer overflow! Wanted %d got %d\n",
1377                         hdr->len, entry->len);
1378                 qp->rx_err_oflow++;
1379
1380                 entry->len = -EIO;
1381                 entry->flags |= DESC_DONE_FLAG;
1382
1383                 ntb_complete_rxc(qp);
1384         } else {
1385                 dev_dbg(&qp->ndev->pdev->dev,
1386                         "RX OK index %u ver %u size %d into buf size %d\n",
1387                         qp->rx_index, hdr->ver, hdr->len, entry->len);
1388
1389                 qp->rx_bytes += hdr->len;
1390                 qp->rx_pkts++;
1391
1392                 entry->len = hdr->len;
1393
1394                 ntb_async_rx(entry, offset);
1395         }
1396
1397         qp->rx_index++;
1398         qp->rx_index %= qp->rx_max_entry;
1399
1400         return 0;
1401 }
1402
1403 static void ntb_transport_rxc_db(unsigned long data)
1404 {
1405         struct ntb_transport_qp *qp = (void *)data;
1406         int rc, i;
1407
1408         dev_dbg(&qp->ndev->pdev->dev, "%s: doorbell %d received\n",
1409                 __func__, qp->qp_num);
1410
1411         /* Limit the number of packets processed in a single interrupt to
1412          * provide fairness to others
1413          */
1414         for (i = 0; i < qp->rx_max_entry; i++) {
1415                 rc = ntb_process_rxc(qp);
1416                 if (rc)
1417                         break;
1418         }
1419
1420         if (i && qp->rx_dma_chan)
1421                 dma_async_issue_pending(qp->rx_dma_chan);
1422
1423         if (i == qp->rx_max_entry) {
1424                 /* there is more work to do */
1425                 if (qp->active)
1426                         tasklet_schedule(&qp->rxc_db_work);
1427         } else if (ntb_db_read(qp->ndev) & BIT_ULL(qp->qp_num)) {
1428                 /* the doorbell bit is set: clear it */
1429                 ntb_db_clear(qp->ndev, BIT_ULL(qp->qp_num));
1430                 /* ntb_db_read ensures ntb_db_clear write is committed */
1431                 ntb_db_read(qp->ndev);
1432
1433                 /* an interrupt may have arrived between finishing
1434                  * ntb_process_rxc and clearing the doorbell bit:
1435                  * there might be some more work to do.
1436                  */
1437                 if (qp->active)
1438                         tasklet_schedule(&qp->rxc_db_work);
1439         }
1440 }
1441
1442 static void ntb_tx_copy_callback(void *data)
1443 {
1444         struct ntb_queue_entry *entry = data;
1445         struct ntb_transport_qp *qp = entry->qp;
1446         struct ntb_payload_header __iomem *hdr = entry->tx_hdr;
1447
1448         iowrite32(entry->flags | DESC_DONE_FLAG, &hdr->flags);
1449
1450         ntb_peer_db_set(qp->ndev, BIT_ULL(qp->qp_num));
1451
1452         /* The entry length can only be zero if the packet is intended to be a
1453          * "link down" or similar.  Since no payload is being sent in these
1454          * cases, there is nothing to add to the completion queue.
1455          */
1456         if (entry->len > 0) {
1457                 qp->tx_bytes += entry->len;
1458
1459                 if (qp->tx_handler)
1460                         qp->tx_handler(qp, qp->cb_data, entry->cb_data,
1461                                        entry->len);
1462         }
1463
1464         ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, &qp->tx_free_q);
1465 }
1466
1467 static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset)
1468 {
1469 #ifdef ARCH_HAS_NOCACHE_UACCESS
1470         /*
1471          * Using non-temporal mov to improve performance on non-cached
1472          * writes, even though we aren't actually copying from user space.
1473          */
1474         __copy_from_user_inatomic_nocache(offset, entry->buf, entry->len);
1475 #else
1476         memcpy_toio(offset, entry->buf, entry->len);
1477 #endif
1478
1479         /* Ensure that the data is fully copied out before setting the flags */
1480         wmb();
1481
1482         ntb_tx_copy_callback(entry);
1483 }
1484
1485 static void ntb_async_tx(struct ntb_transport_qp *qp,
1486                          struct ntb_queue_entry *entry)
1487 {
1488         struct ntb_payload_header __iomem *hdr;
1489         struct dma_async_tx_descriptor *txd;
1490         struct dma_chan *chan = qp->tx_dma_chan;
1491         struct dma_device *device;
1492         size_t dest_off, buff_off;
1493         struct dmaengine_unmap_data *unmap;
1494         dma_addr_t dest;
1495         dma_cookie_t cookie;
1496         void __iomem *offset;
1497         size_t len = entry->len;
1498         void *buf = entry->buf;
1499         int retries = 0;
1500
1501         offset = qp->tx_mw + qp->tx_max_frame * qp->tx_index;
1502         hdr = offset + qp->tx_max_frame - sizeof(struct ntb_payload_header);
1503         entry->tx_hdr = hdr;
1504
1505         iowrite32(entry->len, &hdr->len);
1506         iowrite32((u32)qp->tx_pkts, &hdr->ver);
1507
1508         if (!chan)
1509                 goto err;
1510
1511         if (len < copy_bytes)
1512                 goto err;
1513
1514         device = chan->device;
1515         dest = qp->tx_mw_phys + qp->tx_max_frame * qp->tx_index;
1516         buff_off = (size_t)buf & ~PAGE_MASK;
1517         dest_off = (size_t)dest & ~PAGE_MASK;
1518
1519         if (!is_dma_copy_aligned(device, buff_off, dest_off, len))
1520                 goto err;
1521
1522         unmap = dmaengine_get_unmap_data(device->dev, 1, GFP_NOWAIT);
1523         if (!unmap)
1524                 goto err;
1525
1526         unmap->len = len;
1527         unmap->addr[0] = dma_map_page(device->dev, virt_to_page(buf),
1528                                       buff_off, len, DMA_TO_DEVICE);
1529         if (dma_mapping_error(device->dev, unmap->addr[0]))
1530                 goto err_get_unmap;
1531
1532         unmap->to_cnt = 1;
1533
1534         for (retries = 0; retries < DMA_RETRIES; retries++) {
1535                 txd = device->device_prep_dma_memcpy(chan, dest, unmap->addr[0],
1536                                                      len, DMA_PREP_INTERRUPT);
1537                 if (txd)
1538                         break;
1539
1540                 set_current_state(TASK_INTERRUPTIBLE);
1541                 schedule_timeout(DMA_OUT_RESOURCE_TO);
1542         }
1543
1544         if (!txd) {
1545                 qp->dma_tx_prep_err++;
1546                 goto err_get_unmap;
1547         }
1548
1549         txd->callback = ntb_tx_copy_callback;
1550         txd->callback_param = entry;
1551         dma_set_unmap(txd, unmap);
1552
1553         cookie = dmaengine_submit(txd);
1554         if (dma_submit_error(cookie))
1555                 goto err_set_unmap;
1556
1557         dmaengine_unmap_put(unmap);
1558
1559         dma_async_issue_pending(chan);
1560         qp->tx_async++;
1561
1562         return;
1563 err_set_unmap:
1564         dmaengine_unmap_put(unmap);
1565 err_get_unmap:
1566         dmaengine_unmap_put(unmap);
1567 err:
1568         ntb_memcpy_tx(entry, offset);
1569         qp->tx_memcpy++;
1570 }
1571
1572 static int ntb_process_tx(struct ntb_transport_qp *qp,
1573                           struct ntb_queue_entry *entry)
1574 {
1575         if (qp->tx_index == qp->remote_rx_info->entry) {
1576                 qp->tx_ring_full++;
1577                 return -EAGAIN;
1578         }
1579
1580         if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) {
1581                 if (qp->tx_handler)
1582                         qp->tx_handler(qp, qp->cb_data, NULL, -EIO);
1583
1584                 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry,
1585                              &qp->tx_free_q);
1586                 return 0;
1587         }
1588
1589         ntb_async_tx(qp, entry);
1590
1591         qp->tx_index++;
1592         qp->tx_index %= qp->tx_max_entry;
1593
1594         qp->tx_pkts++;
1595
1596         return 0;
1597 }
1598
1599 static void ntb_send_link_down(struct ntb_transport_qp *qp)
1600 {
1601         struct pci_dev *pdev = qp->ndev->pdev;
1602         struct ntb_queue_entry *entry;
1603         int i, rc;
1604
1605         if (!qp->link_is_up)
1606                 return;
1607
1608         dev_info(&pdev->dev, "qp %d: Send Link Down\n", qp->qp_num);
1609
1610         for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) {
1611                 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
1612                 if (entry)
1613                         break;
1614                 msleep(100);
1615         }
1616
1617         if (!entry)
1618                 return;
1619
1620         entry->cb_data = NULL;
1621         entry->buf = NULL;
1622         entry->len = 0;
1623         entry->flags = LINK_DOWN_FLAG;
1624
1625         rc = ntb_process_tx(qp, entry);
1626         if (rc)
1627                 dev_err(&pdev->dev, "ntb: QP%d unable to send linkdown msg\n",
1628                         qp->qp_num);
1629
1630         ntb_qp_link_down_reset(qp);
1631 }
1632
1633 static bool ntb_dma_filter_fn(struct dma_chan *chan, void *node)
1634 {
1635         return dev_to_node(&chan->dev->device) == (int)(unsigned long)node;
1636 }
1637
1638 /**
1639  * ntb_transport_create_queue - Create a new NTB transport layer queue
1640  * @rx_handler: receive callback function
1641  * @tx_handler: transmit callback function
1642  * @event_handler: event callback function
1643  *
1644  * Create a new NTB transport layer queue and provide the queue with a callback
1645  * routine for both transmit and receive.  The receive callback routine will be
1646  * used to pass up data when the transport has received it on the queue.   The
1647  * transmit callback routine will be called when the transport has completed the
1648  * transmission of the data on the queue and the data is ready to be freed.
1649  *
1650  * RETURNS: pointer to newly created ntb_queue, NULL on error.
1651  */
1652 struct ntb_transport_qp *
1653 ntb_transport_create_queue(void *data, struct device *client_dev,
1654                            const struct ntb_queue_handlers *handlers)
1655 {
1656         struct ntb_dev *ndev;
1657         struct pci_dev *pdev;
1658         struct ntb_transport_ctx *nt;
1659         struct ntb_queue_entry *entry;
1660         struct ntb_transport_qp *qp;
1661         u64 qp_bit;
1662         unsigned int free_queue;
1663         dma_cap_mask_t dma_mask;
1664         int node;
1665         int i;
1666
1667         ndev = dev_ntb(client_dev->parent);
1668         pdev = ndev->pdev;
1669         nt = ndev->ctx;
1670
1671         node = dev_to_node(&ndev->dev);
1672
1673         free_queue = ffs(nt->qp_bitmap);
1674         if (!free_queue)
1675                 goto err;
1676
1677         /* decrement free_queue to make it zero based */
1678         free_queue--;
1679
1680         qp = &nt->qp_vec[free_queue];
1681         qp_bit = BIT_ULL(qp->qp_num);
1682
1683         nt->qp_bitmap_free &= ~qp_bit;
1684
1685         qp->cb_data = data;
1686         qp->rx_handler = handlers->rx_handler;
1687         qp->tx_handler = handlers->tx_handler;
1688         qp->event_handler = handlers->event_handler;
1689
1690         dma_cap_zero(dma_mask);
1691         dma_cap_set(DMA_MEMCPY, dma_mask);
1692
1693         if (use_dma) {
1694                 qp->tx_dma_chan =
1695                         dma_request_channel(dma_mask, ntb_dma_filter_fn,
1696                                             (void *)(unsigned long)node);
1697                 if (!qp->tx_dma_chan)
1698                         dev_info(&pdev->dev, "Unable to allocate TX DMA channel\n");
1699
1700                 qp->rx_dma_chan =
1701                         dma_request_channel(dma_mask, ntb_dma_filter_fn,
1702                                             (void *)(unsigned long)node);
1703                 if (!qp->rx_dma_chan)
1704                         dev_info(&pdev->dev, "Unable to allocate RX DMA channel\n");
1705         } else {
1706                 qp->tx_dma_chan = NULL;
1707                 qp->rx_dma_chan = NULL;
1708         }
1709
1710         dev_dbg(&pdev->dev, "Using %s memcpy for TX\n",
1711                 qp->tx_dma_chan ? "DMA" : "CPU");
1712
1713         dev_dbg(&pdev->dev, "Using %s memcpy for RX\n",
1714                 qp->rx_dma_chan ? "DMA" : "CPU");
1715
1716         for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
1717                 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node);
1718                 if (!entry)
1719                         goto err1;
1720
1721                 entry->qp = qp;
1722                 ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry,
1723                              &qp->rx_free_q);
1724         }
1725
1726         for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
1727                 entry = kzalloc_node(sizeof(*entry), GFP_ATOMIC, node);
1728                 if (!entry)
1729                         goto err2;
1730
1731                 entry->qp = qp;
1732                 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry,
1733                              &qp->tx_free_q);
1734         }
1735
1736         ntb_db_clear(qp->ndev, qp_bit);
1737         ntb_db_clear_mask(qp->ndev, qp_bit);
1738
1739         dev_info(&pdev->dev, "NTB Transport QP %d created\n", qp->qp_num);
1740
1741         return qp;
1742
1743 err2:
1744         while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
1745                 kfree(entry);
1746 err1:
1747         while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q)))
1748                 kfree(entry);
1749         if (qp->tx_dma_chan)
1750                 dma_release_channel(qp->tx_dma_chan);
1751         if (qp->rx_dma_chan)
1752                 dma_release_channel(qp->rx_dma_chan);
1753         nt->qp_bitmap_free |= qp_bit;
1754 err:
1755         return NULL;
1756 }
1757 EXPORT_SYMBOL_GPL(ntb_transport_create_queue);
1758
1759 /**
1760  * ntb_transport_free_queue - Frees NTB transport queue
1761  * @qp: NTB queue to be freed
1762  *
1763  * Frees NTB transport queue
1764  */
1765 void ntb_transport_free_queue(struct ntb_transport_qp *qp)
1766 {
1767         struct pci_dev *pdev;
1768         struct ntb_queue_entry *entry;
1769         u64 qp_bit;
1770
1771         if (!qp)
1772                 return;
1773
1774         pdev = qp->ndev->pdev;
1775
1776         qp->active = false;
1777
1778         if (qp->tx_dma_chan) {
1779                 struct dma_chan *chan = qp->tx_dma_chan;
1780                 /* Putting the dma_chan to NULL will force any new traffic to be
1781                  * processed by the CPU instead of the DAM engine
1782                  */
1783                 qp->tx_dma_chan = NULL;
1784
1785                 /* Try to be nice and wait for any queued DMA engine
1786                  * transactions to process before smashing it with a rock
1787                  */
1788                 dma_sync_wait(chan, qp->last_cookie);
1789                 dmaengine_terminate_all(chan);
1790                 dma_release_channel(chan);
1791         }
1792
1793         if (qp->rx_dma_chan) {
1794                 struct dma_chan *chan = qp->rx_dma_chan;
1795                 /* Putting the dma_chan to NULL will force any new traffic to be
1796                  * processed by the CPU instead of the DAM engine
1797                  */
1798                 qp->rx_dma_chan = NULL;
1799
1800                 /* Try to be nice and wait for any queued DMA engine
1801                  * transactions to process before smashing it with a rock
1802                  */
1803                 dma_sync_wait(chan, qp->last_cookie);
1804                 dmaengine_terminate_all(chan);
1805                 dma_release_channel(chan);
1806         }
1807
1808         qp_bit = BIT_ULL(qp->qp_num);
1809
1810         ntb_db_set_mask(qp->ndev, qp_bit);
1811         tasklet_kill(&qp->rxc_db_work);
1812
1813         cancel_delayed_work_sync(&qp->link_work);
1814
1815         qp->cb_data = NULL;
1816         qp->rx_handler = NULL;
1817         qp->tx_handler = NULL;
1818         qp->event_handler = NULL;
1819
1820         while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q)))
1821                 kfree(entry);
1822
1823         while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q))) {
1824                 dev_warn(&pdev->dev, "Freeing item from non-empty rx_pend_q\n");
1825                 kfree(entry);
1826         }
1827
1828         while ((entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_post_q))) {
1829                 dev_warn(&pdev->dev, "Freeing item from non-empty rx_post_q\n");
1830                 kfree(entry);
1831         }
1832
1833         while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
1834                 kfree(entry);
1835
1836         qp->transport->qp_bitmap_free |= qp_bit;
1837
1838         dev_info(&pdev->dev, "NTB Transport QP %d freed\n", qp->qp_num);
1839 }
1840 EXPORT_SYMBOL_GPL(ntb_transport_free_queue);
1841
1842 /**
1843  * ntb_transport_rx_remove - Dequeues enqueued rx packet
1844  * @qp: NTB queue to be freed
1845  * @len: pointer to variable to write enqueued buffers length
1846  *
1847  * Dequeues unused buffers from receive queue.  Should only be used during
1848  * shutdown of qp.
1849  *
1850  * RETURNS: NULL error value on error, or void* for success.
1851  */
1852 void *ntb_transport_rx_remove(struct ntb_transport_qp *qp, unsigned int *len)
1853 {
1854         struct ntb_queue_entry *entry;
1855         void *buf;
1856
1857         if (!qp || qp->client_ready)
1858                 return NULL;
1859
1860         entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_pend_q);
1861         if (!entry)
1862                 return NULL;
1863
1864         buf = entry->cb_data;
1865         *len = entry->len;
1866
1867         ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_free_q);
1868
1869         return buf;
1870 }
1871 EXPORT_SYMBOL_GPL(ntb_transport_rx_remove);
1872
1873 /**
1874  * ntb_transport_rx_enqueue - Enqueue a new NTB queue entry
1875  * @qp: NTB transport layer queue the entry is to be enqueued on
1876  * @cb: per buffer pointer for callback function to use
1877  * @data: pointer to data buffer that incoming packets will be copied into
1878  * @len: length of the data buffer
1879  *
1880  * Enqueue a new receive buffer onto the transport queue into which a NTB
1881  * payload can be received into.
1882  *
1883  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
1884  */
1885 int ntb_transport_rx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data,
1886                              unsigned int len)
1887 {
1888         struct ntb_queue_entry *entry;
1889
1890         if (!qp)
1891                 return -EINVAL;
1892
1893         entry = ntb_list_rm(&qp->ntb_rx_q_lock, &qp->rx_free_q);
1894         if (!entry)
1895                 return -ENOMEM;
1896
1897         entry->cb_data = cb;
1898         entry->buf = data;
1899         entry->len = len;
1900         entry->flags = 0;
1901
1902         ntb_list_add(&qp->ntb_rx_q_lock, &entry->entry, &qp->rx_pend_q);
1903
1904         if (qp->active)
1905                 tasklet_schedule(&qp->rxc_db_work);
1906
1907         return 0;
1908 }
1909 EXPORT_SYMBOL_GPL(ntb_transport_rx_enqueue);
1910
1911 /**
1912  * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry
1913  * @qp: NTB transport layer queue the entry is to be enqueued on
1914  * @cb: per buffer pointer for callback function to use
1915  * @data: pointer to data buffer that will be sent
1916  * @len: length of the data buffer
1917  *
1918  * Enqueue a new transmit buffer onto the transport queue from which a NTB
1919  * payload will be transmitted.  This assumes that a lock is being held to
1920  * serialize access to the qp.
1921  *
1922  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
1923  */
1924 int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data,
1925                              unsigned int len)
1926 {
1927         struct ntb_queue_entry *entry;
1928         int rc;
1929
1930         if (!qp || !qp->link_is_up || !len)
1931                 return -EINVAL;
1932
1933         entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
1934         if (!entry) {
1935                 qp->tx_err_no_buf++;
1936                 return -EBUSY;
1937         }
1938
1939         entry->cb_data = cb;
1940         entry->buf = data;
1941         entry->len = len;
1942         entry->flags = 0;
1943
1944         rc = ntb_process_tx(qp, entry);
1945         if (rc)
1946                 ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry,
1947                              &qp->tx_free_q);
1948
1949         return rc;
1950 }
1951 EXPORT_SYMBOL_GPL(ntb_transport_tx_enqueue);
1952
1953 /**
1954  * ntb_transport_link_up - Notify NTB transport of client readiness to use queue
1955  * @qp: NTB transport layer queue to be enabled
1956  *
1957  * Notify NTB transport layer of client readiness to use queue
1958  */
1959 void ntb_transport_link_up(struct ntb_transport_qp *qp)
1960 {
1961         if (!qp)
1962                 return;
1963
1964         qp->client_ready = true;
1965
1966         if (qp->transport->link_is_up)
1967                 schedule_delayed_work(&qp->link_work, 0);
1968 }
1969 EXPORT_SYMBOL_GPL(ntb_transport_link_up);
1970
1971 /**
1972  * ntb_transport_link_down - Notify NTB transport to no longer enqueue data
1973  * @qp: NTB transport layer queue to be disabled
1974  *
1975  * Notify NTB transport layer of client's desire to no longer receive data on
1976  * transport queue specified.  It is the client's responsibility to ensure all
1977  * entries on queue are purged or otherwise handled appropriately.
1978  */
1979 void ntb_transport_link_down(struct ntb_transport_qp *qp)
1980 {
1981         int val;
1982
1983         if (!qp)
1984                 return;
1985
1986         qp->client_ready = false;
1987
1988         val = ntb_spad_read(qp->ndev, QP_LINKS);
1989
1990         ntb_peer_spad_write(qp->ndev, QP_LINKS,
1991                             val & ~BIT(qp->qp_num));
1992
1993         if (qp->link_is_up)
1994                 ntb_send_link_down(qp);
1995         else
1996                 cancel_delayed_work_sync(&qp->link_work);
1997 }
1998 EXPORT_SYMBOL_GPL(ntb_transport_link_down);
1999
2000 /**
2001  * ntb_transport_link_query - Query transport link state
2002  * @qp: NTB transport layer queue to be queried
2003  *
2004  * Query connectivity to the remote system of the NTB transport queue
2005  *
2006  * RETURNS: true for link up or false for link down
2007  */
2008 bool ntb_transport_link_query(struct ntb_transport_qp *qp)
2009 {
2010         if (!qp)
2011                 return false;
2012
2013         return qp->link_is_up;
2014 }
2015 EXPORT_SYMBOL_GPL(ntb_transport_link_query);
2016
2017 /**
2018  * ntb_transport_qp_num - Query the qp number
2019  * @qp: NTB transport layer queue to be queried
2020  *
2021  * Query qp number of the NTB transport queue
2022  *
2023  * RETURNS: a zero based number specifying the qp number
2024  */
2025 unsigned char ntb_transport_qp_num(struct ntb_transport_qp *qp)
2026 {
2027         if (!qp)
2028                 return 0;
2029
2030         return qp->qp_num;
2031 }
2032 EXPORT_SYMBOL_GPL(ntb_transport_qp_num);
2033
2034 /**
2035  * ntb_transport_max_size - Query the max payload size of a qp
2036  * @qp: NTB transport layer queue to be queried
2037  *
2038  * Query the maximum payload size permissible on the given qp
2039  *
2040  * RETURNS: the max payload size of a qp
2041  */
2042 unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp)
2043 {
2044         unsigned int max_size;
2045         unsigned int copy_align;
2046         struct dma_chan *rx_chan, *tx_chan;
2047
2048         if (!qp)
2049                 return 0;
2050
2051         rx_chan = qp->rx_dma_chan;
2052         tx_chan = qp->tx_dma_chan;
2053
2054         copy_align = max(rx_chan ? rx_chan->device->copy_align : 0,
2055                          tx_chan ? tx_chan->device->copy_align : 0);
2056
2057         /* If DMA engine usage is possible, try to find the max size for that */
2058         max_size = qp->tx_max_frame - sizeof(struct ntb_payload_header);
2059         max_size = round_down(max_size, 1 << copy_align);
2060
2061         return max_size;
2062 }
2063 EXPORT_SYMBOL_GPL(ntb_transport_max_size);
2064
2065 unsigned int ntb_transport_tx_free_entry(struct ntb_transport_qp *qp)
2066 {
2067         unsigned int head = qp->tx_index;
2068         unsigned int tail = qp->remote_rx_info->entry;
2069
2070         return tail > head ? tail - head : qp->tx_max_entry + tail - head;
2071 }
2072 EXPORT_SYMBOL_GPL(ntb_transport_tx_free_entry);
2073
2074 static void ntb_transport_doorbell_callback(void *data, int vector)
2075 {
2076         struct ntb_transport_ctx *nt = data;
2077         struct ntb_transport_qp *qp;
2078         u64 db_bits;
2079         unsigned int qp_num;
2080
2081         db_bits = (nt->qp_bitmap & ~nt->qp_bitmap_free &
2082                    ntb_db_vector_mask(nt->ndev, vector));
2083
2084         while (db_bits) {
2085                 qp_num = __ffs(db_bits);
2086                 qp = &nt->qp_vec[qp_num];
2087
2088                 if (qp->active)
2089                         tasklet_schedule(&qp->rxc_db_work);
2090
2091                 db_bits &= ~BIT_ULL(qp_num);
2092         }
2093 }
2094
2095 static const struct ntb_ctx_ops ntb_transport_ops = {
2096         .link_event = ntb_transport_event_callback,
2097         .db_event = ntb_transport_doorbell_callback,
2098 };
2099
2100 static struct ntb_client ntb_transport_client = {
2101         .ops = {
2102                 .probe = ntb_transport_probe,
2103                 .remove = ntb_transport_free,
2104         },
2105 };
2106
2107 static int __init ntb_transport_init(void)
2108 {
2109         int rc;
2110
2111         pr_info("%s, version %s\n", NTB_TRANSPORT_DESC, NTB_TRANSPORT_VER);
2112
2113         if (debugfs_initialized())
2114                 nt_debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
2115
2116         rc = bus_register(&ntb_transport_bus);
2117         if (rc)
2118                 goto err_bus;
2119
2120         rc = ntb_register_client(&ntb_transport_client);
2121         if (rc)
2122                 goto err_client;
2123
2124         return 0;
2125
2126 err_client:
2127         bus_unregister(&ntb_transport_bus);
2128 err_bus:
2129         debugfs_remove_recursive(nt_debugfs_dir);
2130         return rc;
2131 }
2132 module_init(ntb_transport_init);
2133
2134 static void __exit ntb_transport_exit(void)
2135 {
2136         debugfs_remove_recursive(nt_debugfs_dir);
2137
2138         ntb_unregister_client(&ntb_transport_client);
2139         bus_unregister(&ntb_transport_bus);
2140 }
2141 module_exit(ntb_transport_exit);