blob: facad51fbc7ad8cf6e034271d3e0616a4fa8fcae [file] [log] [blame]
Jon Masonfce8a7b2012-11-16 19:27:12 -07001/*
2 * This file is provided under a dual BSD/GPLv2 license. When using or
3 * redistributing this file, you may do so under either license.
4 *
5 * GPL LICENSE SUMMARY
6 *
7 * Copyright(c) 2012 Intel Corporation. All rights reserved.
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of version 2 of the GNU General Public License as
11 * published by the Free Software Foundation.
12 *
13 * BSD LICENSE
14 *
15 * Copyright(c) 2012 Intel Corporation. All rights reserved.
16 *
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions
19 * are met:
20 *
21 * * Redistributions of source code must retain the above copyright
22 * notice, this list of conditions and the following disclaimer.
23 * * Redistributions in binary form must reproduce the above copy
24 * notice, this list of conditions and the following disclaimer in
25 * the documentation and/or other materials provided with the
26 * distribution.
27 * * Neither the name of Intel Corporation nor the names of its
28 * contributors may be used to endorse or promote products derived
29 * from this software without specific prior written permission.
30 *
31 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
32 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
33 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
34 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
35 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
36 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
37 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
38 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
39 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
40 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
41 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
42 *
43 * Intel PCIe NTB Linux driver
44 *
45 * Contact Information:
46 * Jon Mason <jon.mason@intel.com>
47 */
48#include <linux/debugfs.h>
49#include <linux/init.h>
50#include <linux/interrupt.h>
51#include <linux/module.h>
52#include <linux/pci.h>
53#include <linux/slab.h>
54#include "ntb_hw.h"
55#include "ntb_regs.h"
56
57#define NTB_NAME "Intel(R) PCI-E Non-Transparent Bridge Driver"
58#define NTB_VER "0.24"
59
60MODULE_DESCRIPTION(NTB_NAME);
61MODULE_VERSION(NTB_VER);
62MODULE_LICENSE("Dual BSD/GPL");
63MODULE_AUTHOR("Intel Corporation");
64
65enum {
66 NTB_CONN_CLASSIC = 0,
67 NTB_CONN_B2B,
68 NTB_CONN_RP,
69};
70
71enum {
72 NTB_DEV_USD = 0,
73 NTB_DEV_DSD,
74};
75
76enum {
77 SNB_HW = 0,
78 BWD_HW,
79};
80
81/* Translate memory window 0,1 to BAR 2,4 */
82#define MW_TO_BAR(mw) (mw * 2 + 2)
83
84static DEFINE_PCI_DEVICE_TABLE(ntb_pci_tbl) = {
85 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_BWD)},
86 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_JSF)},
87 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_CLASSIC_JSF)},
88 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_RP_JSF)},
89 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_RP_SNB)},
90 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_SNB)},
91 {PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_CLASSIC_SNB)},
92 {0}
93};
94MODULE_DEVICE_TABLE(pci, ntb_pci_tbl);
95
96/**
97 * ntb_register_event_callback() - register event callback
98 * @ndev: pointer to ntb_device instance
99 * @func: callback function to register
100 *
101 * This function registers a callback for any HW driver events such as link
102 * up/down, power management notices and etc.
103 *
104 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
105 */
106int ntb_register_event_callback(struct ntb_device *ndev,
107 void (*func)(void *handle, unsigned int event))
108{
109 if (ndev->event_cb)
110 return -EINVAL;
111
112 ndev->event_cb = func;
113
114 return 0;
115}
116
117/**
118 * ntb_unregister_event_callback() - unregisters the event callback
119 * @ndev: pointer to ntb_device instance
120 *
121 * This function unregisters the existing callback from transport
122 */
123void ntb_unregister_event_callback(struct ntb_device *ndev)
124{
125 ndev->event_cb = NULL;
126}
127
128/**
129 * ntb_register_db_callback() - register a callback for doorbell interrupt
130 * @ndev: pointer to ntb_device instance
131 * @idx: doorbell index to register callback, zero based
132 * @func: callback function to register
133 *
134 * This function registers a callback function for the doorbell interrupt
135 * on the primary side. The function will unmask the doorbell as well to
136 * allow interrupt.
137 *
138 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
139 */
140int ntb_register_db_callback(struct ntb_device *ndev, unsigned int idx,
141 void *data, void (*func)(void *data, int db_num))
142{
143 unsigned long mask;
144
145 if (idx >= ndev->max_cbs || ndev->db_cb[idx].callback) {
146 dev_warn(&ndev->pdev->dev, "Invalid Index.\n");
147 return -EINVAL;
148 }
149
150 ndev->db_cb[idx].callback = func;
151 ndev->db_cb[idx].data = data;
152
153 /* unmask interrupt */
154 mask = readw(ndev->reg_ofs.pdb_mask);
155 clear_bit(idx * ndev->bits_per_vector, &mask);
156 writew(mask, ndev->reg_ofs.pdb_mask);
157
158 return 0;
159}
160
161/**
162 * ntb_unregister_db_callback() - unregister a callback for doorbell interrupt
163 * @ndev: pointer to ntb_device instance
164 * @idx: doorbell index to register callback, zero based
165 *
166 * This function unregisters a callback function for the doorbell interrupt
167 * on the primary side. The function will also mask the said doorbell.
168 */
169void ntb_unregister_db_callback(struct ntb_device *ndev, unsigned int idx)
170{
171 unsigned long mask;
172
173 if (idx >= ndev->max_cbs || !ndev->db_cb[idx].callback)
174 return;
175
176 mask = readw(ndev->reg_ofs.pdb_mask);
177 set_bit(idx * ndev->bits_per_vector, &mask);
178 writew(mask, ndev->reg_ofs.pdb_mask);
179
180 ndev->db_cb[idx].callback = NULL;
181}
182
183/**
184 * ntb_find_transport() - find the transport pointer
185 * @transport: pointer to pci device
186 *
187 * Given the pci device pointer, return the transport pointer passed in when
188 * the transport attached when it was inited.
189 *
190 * RETURNS: pointer to transport.
191 */
192void *ntb_find_transport(struct pci_dev *pdev)
193{
194 struct ntb_device *ndev = pci_get_drvdata(pdev);
195 return ndev->ntb_transport;
196}
197
198/**
199 * ntb_register_transport() - Register NTB transport with NTB HW driver
200 * @transport: transport identifier
201 *
202 * This function allows a transport to reserve the hardware driver for
203 * NTB usage.
204 *
205 * RETURNS: pointer to ntb_device, NULL on error.
206 */
207struct ntb_device *ntb_register_transport(struct pci_dev *pdev, void *transport)
208{
209 struct ntb_device *ndev = pci_get_drvdata(pdev);
210
211 if (ndev->ntb_transport)
212 return NULL;
213
214 ndev->ntb_transport = transport;
215 return ndev;
216}
217
218/**
219 * ntb_unregister_transport() - Unregister the transport with the NTB HW driver
220 * @ndev - ntb_device of the transport to be freed
221 *
222 * This function unregisters the transport from the HW driver and performs any
223 * necessary cleanups.
224 */
225void ntb_unregister_transport(struct ntb_device *ndev)
226{
227 int i;
228
229 if (!ndev->ntb_transport)
230 return;
231
232 for (i = 0; i < ndev->max_cbs; i++)
233 ntb_unregister_db_callback(ndev, i);
234
235 ntb_unregister_event_callback(ndev);
236 ndev->ntb_transport = NULL;
237}
238
239/**
240 * ntb_get_max_spads() - get the total scratch regs usable
241 * @ndev: pointer to ntb_device instance
242 *
243 * This function returns the max 32bit scratchpad registers usable by the
244 * upper layer.
245 *
246 * RETURNS: total number of scratch pad registers available
247 */
248int ntb_get_max_spads(struct ntb_device *ndev)
249{
250 return ndev->limits.max_spads;
251}
252
253/**
254 * ntb_write_local_spad() - write to the secondary scratchpad register
255 * @ndev: pointer to ntb_device instance
256 * @idx: index to the scratchpad register, 0 based
257 * @val: the data value to put into the register
258 *
259 * This function allows writing of a 32bit value to the indexed scratchpad
260 * register. This writes over the data mirrored to the local scratchpad register
261 * by the remote system.
262 *
263 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
264 */
265int ntb_write_local_spad(struct ntb_device *ndev, unsigned int idx, u32 val)
266{
267 if (idx >= ndev->limits.max_spads)
268 return -EINVAL;
269
270 dev_dbg(&ndev->pdev->dev, "Writing %x to local scratch pad index %d\n",
271 val, idx);
272 writel(val, ndev->reg_ofs.spad_read + idx * 4);
273
274 return 0;
275}
276
277/**
278 * ntb_read_local_spad() - read from the primary scratchpad register
279 * @ndev: pointer to ntb_device instance
280 * @idx: index to scratchpad register, 0 based
281 * @val: pointer to 32bit integer for storing the register value
282 *
283 * This function allows reading of the 32bit scratchpad register on
284 * the primary (internal) side. This allows the local system to read data
285 * written and mirrored to the scratchpad register by the remote system.
286 *
287 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
288 */
289int ntb_read_local_spad(struct ntb_device *ndev, unsigned int idx, u32 *val)
290{
291 if (idx >= ndev->limits.max_spads)
292 return -EINVAL;
293
294 *val = readl(ndev->reg_ofs.spad_write + idx * 4);
295 dev_dbg(&ndev->pdev->dev,
296 "Reading %x from local scratch pad index %d\n", *val, idx);
297
298 return 0;
299}
300
301/**
302 * ntb_write_remote_spad() - write to the secondary scratchpad register
303 * @ndev: pointer to ntb_device instance
304 * @idx: index to the scratchpad register, 0 based
305 * @val: the data value to put into the register
306 *
307 * This function allows writing of a 32bit value to the indexed scratchpad
308 * register. The register resides on the secondary (external) side. This allows
309 * the local system to write data to be mirrored to the remote systems
310 * scratchpad register.
311 *
312 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
313 */
314int ntb_write_remote_spad(struct ntb_device *ndev, unsigned int idx, u32 val)
315{
316 if (idx >= ndev->limits.max_spads)
317 return -EINVAL;
318
319 dev_dbg(&ndev->pdev->dev, "Writing %x to remote scratch pad index %d\n",
320 val, idx);
321 writel(val, ndev->reg_ofs.spad_write + idx * 4);
322
323 return 0;
324}
325
326/**
327 * ntb_read_remote_spad() - read from the primary scratchpad register
328 * @ndev: pointer to ntb_device instance
329 * @idx: index to scratchpad register, 0 based
330 * @val: pointer to 32bit integer for storing the register value
331 *
332 * This function allows reading of the 32bit scratchpad register on
333 * the primary (internal) side. This alloows the local system to read the data
334 * it wrote to be mirrored on the remote system.
335 *
336 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
337 */
338int ntb_read_remote_spad(struct ntb_device *ndev, unsigned int idx, u32 *val)
339{
340 if (idx >= ndev->limits.max_spads)
341 return -EINVAL;
342
343 *val = readl(ndev->reg_ofs.spad_read + idx * 4);
344 dev_dbg(&ndev->pdev->dev,
345 "Reading %x from remote scratch pad index %d\n", *val, idx);
346
347 return 0;
348}
349
350/**
351 * ntb_get_mw_vbase() - get virtual addr for the NTB memory window
352 * @ndev: pointer to ntb_device instance
353 * @mw: memory window number
354 *
355 * This function provides the base virtual address of the memory window
356 * specified.
357 *
358 * RETURNS: pointer to virtual address, or NULL on error.
359 */
360void *ntb_get_mw_vbase(struct ntb_device *ndev, unsigned int mw)
361{
362 if (mw > NTB_NUM_MW)
363 return NULL;
364
365 return ndev->mw[mw].vbase;
366}
367
368/**
369 * ntb_get_mw_size() - return size of NTB memory window
370 * @ndev: pointer to ntb_device instance
371 * @mw: memory window number
372 *
373 * This function provides the physical size of the memory window specified
374 *
375 * RETURNS: the size of the memory window or zero on error
376 */
377resource_size_t ntb_get_mw_size(struct ntb_device *ndev, unsigned int mw)
378{
379 if (mw > NTB_NUM_MW)
380 return 0;
381
382 return ndev->mw[mw].bar_sz;
383}
384
385/**
386 * ntb_set_mw_addr - set the memory window address
387 * @ndev: pointer to ntb_device instance
388 * @mw: memory window number
389 * @addr: base address for data
390 *
391 * This function sets the base physical address of the memory window. This
392 * memory address is where data from the remote system will be transfered into
393 * or out of depending on how the transport is configured.
394 */
395void ntb_set_mw_addr(struct ntb_device *ndev, unsigned int mw, u64 addr)
396{
397 if (mw > NTB_NUM_MW)
398 return;
399
400 dev_dbg(&ndev->pdev->dev, "Writing addr %Lx to BAR %d\n", addr,
401 MW_TO_BAR(mw));
402
403 ndev->mw[mw].phys_addr = addr;
404
405 switch (MW_TO_BAR(mw)) {
406 case NTB_BAR_23:
407 writeq(addr, ndev->reg_ofs.sbar2_xlat);
408 break;
409 case NTB_BAR_45:
410 writeq(addr, ndev->reg_ofs.sbar4_xlat);
411 break;
412 }
413}
414
415/**
416 * ntb_ring_sdb() - Set the doorbell on the secondary/external side
417 * @ndev: pointer to ntb_device instance
418 * @db: doorbell to ring
419 *
420 * This function allows triggering of a doorbell on the secondary/external
421 * side that will initiate an interrupt on the remote host
422 *
423 * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
424 */
425void ntb_ring_sdb(struct ntb_device *ndev, unsigned int db)
426{
427 dev_dbg(&ndev->pdev->dev, "%s: ringing doorbell %d\n", __func__, db);
428
429 if (ndev->hw_type == BWD_HW)
430 writeq((u64) 1 << db, ndev->reg_ofs.sdb);
431 else
432 writew(((1 << ndev->bits_per_vector) - 1) <<
433 (db * ndev->bits_per_vector), ndev->reg_ofs.sdb);
434}
435
436static void ntb_link_event(struct ntb_device *ndev, int link_state)
437{
438 unsigned int event;
439
440 if (ndev->link_status == link_state)
441 return;
442
443 if (link_state == NTB_LINK_UP) {
444 u16 status;
445
446 dev_info(&ndev->pdev->dev, "Link Up\n");
447 ndev->link_status = NTB_LINK_UP;
448 event = NTB_EVENT_HW_LINK_UP;
449
450 if (ndev->hw_type == BWD_HW)
451 status = readw(ndev->reg_ofs.lnk_stat);
452 else {
453 int rc = pci_read_config_word(ndev->pdev,
454 SNB_LINK_STATUS_OFFSET,
455 &status);
456 if (rc)
457 return;
458 }
459 dev_info(&ndev->pdev->dev, "Link Width %d, Link Speed %d\n",
460 (status & NTB_LINK_WIDTH_MASK) >> 4,
461 (status & NTB_LINK_SPEED_MASK));
462 } else {
463 dev_info(&ndev->pdev->dev, "Link Down\n");
464 ndev->link_status = NTB_LINK_DOWN;
465 event = NTB_EVENT_HW_LINK_DOWN;
466 }
467
468 /* notify the upper layer if we have an event change */
469 if (ndev->event_cb)
470 ndev->event_cb(ndev->ntb_transport, event);
471}
472
473static int ntb_link_status(struct ntb_device *ndev)
474{
475 int link_state;
476
477 if (ndev->hw_type == BWD_HW) {
478 u32 ntb_cntl;
479
480 ntb_cntl = readl(ndev->reg_ofs.lnk_cntl);
481 if (ntb_cntl & BWD_CNTL_LINK_DOWN)
482 link_state = NTB_LINK_DOWN;
483 else
484 link_state = NTB_LINK_UP;
485 } else {
486 u16 status;
487 int rc;
488
489 rc = pci_read_config_word(ndev->pdev, SNB_LINK_STATUS_OFFSET,
490 &status);
491 if (rc)
492 return rc;
493
494 if (status & NTB_LINK_STATUS_ACTIVE)
495 link_state = NTB_LINK_UP;
496 else
497 link_state = NTB_LINK_DOWN;
498 }
499
500 ntb_link_event(ndev, link_state);
501
502 return 0;
503}
504
505/* BWD doesn't have link status interrupt, poll on that platform */
506static void bwd_link_poll(struct work_struct *work)
507{
508 struct ntb_device *ndev = container_of(work, struct ntb_device,
509 hb_timer.work);
510 unsigned long ts = jiffies;
511
512 /* If we haven't gotten an interrupt in a while, check the BWD link
513 * status bit
514 */
515 if (ts > ndev->last_ts + NTB_HB_TIMEOUT) {
516 int rc = ntb_link_status(ndev);
517 if (rc)
518 dev_err(&ndev->pdev->dev,
519 "Error determining link status\n");
520 }
521
522 schedule_delayed_work(&ndev->hb_timer, NTB_HB_TIMEOUT);
523}
524
525static int ntb_xeon_setup(struct ntb_device *ndev)
526{
527 int rc;
528 u8 val;
529
530 ndev->hw_type = SNB_HW;
531
532 rc = pci_read_config_byte(ndev->pdev, NTB_PPD_OFFSET, &val);
533 if (rc)
534 return rc;
535
536 switch (val & SNB_PPD_CONN_TYPE) {
537 case NTB_CONN_B2B:
538 ndev->conn_type = NTB_CONN_B2B;
539 break;
540 case NTB_CONN_CLASSIC:
541 case NTB_CONN_RP:
542 default:
543 dev_err(&ndev->pdev->dev, "Only B2B supported at this time\n");
544 return -EINVAL;
545 }
546
547 if (val & SNB_PPD_DEV_TYPE)
548 ndev->dev_type = NTB_DEV_DSD;
549 else
550 ndev->dev_type = NTB_DEV_USD;
551
552 ndev->reg_ofs.pdb = ndev->reg_base + SNB_PDOORBELL_OFFSET;
553 ndev->reg_ofs.pdb_mask = ndev->reg_base + SNB_PDBMSK_OFFSET;
554 ndev->reg_ofs.sbar2_xlat = ndev->reg_base + SNB_SBAR2XLAT_OFFSET;
555 ndev->reg_ofs.sbar4_xlat = ndev->reg_base + SNB_SBAR4XLAT_OFFSET;
556 ndev->reg_ofs.lnk_cntl = ndev->reg_base + SNB_NTBCNTL_OFFSET;
557 ndev->reg_ofs.lnk_stat = ndev->reg_base + SNB_LINK_STATUS_OFFSET;
558 ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET;
559 ndev->reg_ofs.spci_cmd = ndev->reg_base + SNB_PCICMD_OFFSET;
560
561 if (ndev->conn_type == NTB_CONN_B2B) {
562 ndev->reg_ofs.sdb = ndev->reg_base + SNB_B2B_DOORBELL_OFFSET;
563 ndev->reg_ofs.spad_write = ndev->reg_base + SNB_B2B_SPAD_OFFSET;
564 ndev->limits.max_spads = SNB_MAX_SPADS;
565 } else {
566 ndev->reg_ofs.sdb = ndev->reg_base + SNB_SDOORBELL_OFFSET;
567 ndev->reg_ofs.spad_write = ndev->reg_base + SNB_SPAD_OFFSET;
568 ndev->limits.max_spads = SNB_MAX_COMPAT_SPADS;
569 }
570
571 ndev->limits.max_db_bits = SNB_MAX_DB_BITS;
572 ndev->limits.msix_cnt = SNB_MSIX_CNT;
573 ndev->bits_per_vector = SNB_DB_BITS_PER_VEC;
574
575 return 0;
576}
577
578static int ntb_bwd_setup(struct ntb_device *ndev)
579{
580 int rc;
581 u32 val;
582
583 ndev->hw_type = BWD_HW;
584
585 rc = pci_read_config_dword(ndev->pdev, NTB_PPD_OFFSET, &val);
586 if (rc)
587 return rc;
588
589 switch ((val & BWD_PPD_CONN_TYPE) >> 8) {
590 case NTB_CONN_B2B:
591 ndev->conn_type = NTB_CONN_B2B;
592 break;
593 case NTB_CONN_RP:
594 default:
595 dev_err(&ndev->pdev->dev, "Only B2B supported at this time\n");
596 return -EINVAL;
597 }
598
599 if (val & BWD_PPD_DEV_TYPE)
600 ndev->dev_type = NTB_DEV_DSD;
601 else
602 ndev->dev_type = NTB_DEV_USD;
603
604 /* Initiate PCI-E link training */
605 rc = pci_write_config_dword(ndev->pdev, NTB_PPD_OFFSET,
606 val | BWD_PPD_INIT_LINK);
607 if (rc)
608 return rc;
609
610 ndev->reg_ofs.pdb = ndev->reg_base + BWD_PDOORBELL_OFFSET;
611 ndev->reg_ofs.pdb_mask = ndev->reg_base + BWD_PDBMSK_OFFSET;
612 ndev->reg_ofs.sbar2_xlat = ndev->reg_base + BWD_SBAR2XLAT_OFFSET;
613 ndev->reg_ofs.sbar4_xlat = ndev->reg_base + BWD_SBAR4XLAT_OFFSET;
614 ndev->reg_ofs.lnk_cntl = ndev->reg_base + BWD_NTBCNTL_OFFSET;
615 ndev->reg_ofs.lnk_stat = ndev->reg_base + BWD_LINK_STATUS_OFFSET;
616 ndev->reg_ofs.spad_read = ndev->reg_base + BWD_SPAD_OFFSET;
617 ndev->reg_ofs.spci_cmd = ndev->reg_base + BWD_PCICMD_OFFSET;
618
619 if (ndev->conn_type == NTB_CONN_B2B) {
620 ndev->reg_ofs.sdb = ndev->reg_base + BWD_B2B_DOORBELL_OFFSET;
621 ndev->reg_ofs.spad_write = ndev->reg_base + BWD_B2B_SPAD_OFFSET;
622 ndev->limits.max_spads = BWD_MAX_SPADS;
623 } else {
624 ndev->reg_ofs.sdb = ndev->reg_base + BWD_PDOORBELL_OFFSET;
625 ndev->reg_ofs.spad_write = ndev->reg_base + BWD_SPAD_OFFSET;
626 ndev->limits.max_spads = BWD_MAX_COMPAT_SPADS;
627 }
628
629 ndev->limits.max_db_bits = BWD_MAX_DB_BITS;
630 ndev->limits.msix_cnt = BWD_MSIX_CNT;
631 ndev->bits_per_vector = BWD_DB_BITS_PER_VEC;
632
633 /* Since bwd doesn't have a link interrupt, setup a poll timer */
634 INIT_DELAYED_WORK(&ndev->hb_timer, bwd_link_poll);
635 schedule_delayed_work(&ndev->hb_timer, NTB_HB_TIMEOUT);
636
637 return 0;
638}
639
640static int __devinit ntb_device_setup(struct ntb_device *ndev)
641{
642 int rc;
643
644 switch (ndev->pdev->device) {
645 case PCI_DEVICE_ID_INTEL_NTB_2ND_SNB:
646 case PCI_DEVICE_ID_INTEL_NTB_RP_JSF:
647 case PCI_DEVICE_ID_INTEL_NTB_RP_SNB:
648 case PCI_DEVICE_ID_INTEL_NTB_CLASSIC_JSF:
649 case PCI_DEVICE_ID_INTEL_NTB_CLASSIC_SNB:
650 case PCI_DEVICE_ID_INTEL_NTB_B2B_JSF:
651 case PCI_DEVICE_ID_INTEL_NTB_B2B_SNB:
652 rc = ntb_xeon_setup(ndev);
653 break;
654 case PCI_DEVICE_ID_INTEL_NTB_B2B_BWD:
655 rc = ntb_bwd_setup(ndev);
656 break;
657 default:
658 rc = -ENODEV;
659 }
660
661 /* Enable Bus Master and Memory Space on the secondary side */
662 writew(PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER, ndev->reg_ofs.spci_cmd);
663
664 return rc;
665}
666
667static void ntb_device_free(struct ntb_device *ndev)
668{
669 if (ndev->hw_type == BWD_HW)
670 cancel_delayed_work_sync(&ndev->hb_timer);
671}
672
673static irqreturn_t bwd_callback_msix_irq(int irq, void *data)
674{
675 struct ntb_db_cb *db_cb = data;
676 struct ntb_device *ndev = db_cb->ndev;
677
678 dev_dbg(&ndev->pdev->dev, "MSI-X irq %d received for DB %d\n", irq,
679 db_cb->db_num);
680
681 if (db_cb->callback)
682 db_cb->callback(db_cb->data, db_cb->db_num);
683
684 /* No need to check for the specific HB irq, any interrupt means
685 * we're connected.
686 */
687 ndev->last_ts = jiffies;
688
689 writeq((u64) 1 << db_cb->db_num, ndev->reg_ofs.pdb);
690
691 return IRQ_HANDLED;
692}
693
694static irqreturn_t xeon_callback_msix_irq(int irq, void *data)
695{
696 struct ntb_db_cb *db_cb = data;
697 struct ntb_device *ndev = db_cb->ndev;
698
699 dev_dbg(&ndev->pdev->dev, "MSI-X irq %d received for DB %d\n", irq,
700 db_cb->db_num);
701
702 if (db_cb->callback)
703 db_cb->callback(db_cb->data, db_cb->db_num);
704
705 /* On Sandybridge, there are 16 bits in the interrupt register
706 * but only 4 vectors. So, 5 bits are assigned to the first 3
707 * vectors, with the 4th having a single bit for link
708 * interrupts.
709 */
710 writew(((1 << ndev->bits_per_vector) - 1) <<
711 (db_cb->db_num * ndev->bits_per_vector), ndev->reg_ofs.pdb);
712
713 return IRQ_HANDLED;
714}
715
716/* Since we do not have a HW doorbell in BWD, this is only used in JF/JT */
717static irqreturn_t xeon_event_msix_irq(int irq, void *dev)
718{
719 struct ntb_device *ndev = dev;
720 int rc;
721
722 dev_dbg(&ndev->pdev->dev, "MSI-X irq %d received for Events\n", irq);
723
724 rc = ntb_link_status(ndev);
725 if (rc)
726 dev_err(&ndev->pdev->dev, "Error determining link status\n");
727
728 /* bit 15 is always the link bit */
729 writew(1 << ndev->limits.max_db_bits, ndev->reg_ofs.pdb);
730
731 return IRQ_HANDLED;
732}
733
734static irqreturn_t ntb_interrupt(int irq, void *dev)
735{
736 struct ntb_device *ndev = dev;
737 unsigned int i = 0;
738
739 if (ndev->hw_type == BWD_HW) {
740 u64 pdb = readq(ndev->reg_ofs.pdb);
741
742 dev_dbg(&ndev->pdev->dev, "irq %d - pdb = %Lx\n", irq, pdb);
743
744 while (pdb) {
745 i = __ffs(pdb);
746 pdb &= pdb - 1;
747 bwd_callback_msix_irq(irq, &ndev->db_cb[i]);
748 }
749 } else {
750 u16 pdb = readw(ndev->reg_ofs.pdb);
751
752 dev_dbg(&ndev->pdev->dev, "irq %d - pdb = %x sdb %x\n", irq,
753 pdb, readw(ndev->reg_ofs.sdb));
754
755 if (pdb & SNB_DB_HW_LINK) {
756 xeon_event_msix_irq(irq, dev);
757 pdb &= ~SNB_DB_HW_LINK;
758 }
759
760 while (pdb) {
761 i = __ffs(pdb);
762 pdb &= pdb - 1;
763 xeon_callback_msix_irq(irq, &ndev->db_cb[i]);
764 }
765 }
766
767 return IRQ_HANDLED;
768}
769
770static int ntb_setup_msix(struct ntb_device *ndev)
771{
772 struct pci_dev *pdev = ndev->pdev;
773 struct msix_entry *msix;
774 int msix_entries;
775 int rc, i, pos;
776 u16 val;
777
778 pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
779 if (!pos) {
780 rc = -EIO;
781 goto err;
782 }
783
784 rc = pci_read_config_word(pdev, pos + PCI_MSIX_FLAGS, &val);
785 if (rc)
786 goto err;
787
788 msix_entries = msix_table_size(val);
789 if (msix_entries > ndev->limits.msix_cnt) {
790 rc = -EINVAL;
791 goto err;
792 }
793
794 ndev->msix_entries = kmalloc(sizeof(struct msix_entry) * msix_entries,
795 GFP_KERNEL);
796 if (!ndev->msix_entries) {
797 rc = -ENOMEM;
798 goto err;
799 }
800
801 for (i = 0; i < msix_entries; i++)
802 ndev->msix_entries[i].entry = i;
803
804 rc = pci_enable_msix(pdev, ndev->msix_entries, msix_entries);
805 if (rc < 0)
806 goto err1;
807 if (rc > 0) {
808 /* On SNB, the link interrupt is always tied to 4th vector. If
809 * we can't get all 4, then we can't use MSI-X.
810 */
811 if (ndev->hw_type != BWD_HW) {
812 rc = -EIO;
813 goto err1;
814 }
815
816 dev_warn(&pdev->dev,
817 "Only %d MSI-X vectors. Limiting the number of queues to that number.\n",
818 rc);
819 msix_entries = rc;
820 }
821
822 for (i = 0; i < msix_entries; i++) {
823 msix = &ndev->msix_entries[i];
824 WARN_ON(!msix->vector);
825
826 /* Use the last MSI-X vector for Link status */
827 if (ndev->hw_type == BWD_HW) {
828 rc = request_irq(msix->vector, bwd_callback_msix_irq, 0,
829 "ntb-callback-msix", &ndev->db_cb[i]);
830 if (rc)
831 goto err2;
832 } else {
833 if (i == msix_entries - 1) {
834 rc = request_irq(msix->vector,
835 xeon_event_msix_irq, 0,
836 "ntb-event-msix", ndev);
837 if (rc)
838 goto err2;
839 } else {
840 rc = request_irq(msix->vector,
841 xeon_callback_msix_irq, 0,
842 "ntb-callback-msix",
843 &ndev->db_cb[i]);
844 if (rc)
845 goto err2;
846 }
847 }
848 }
849
850 ndev->num_msix = msix_entries;
851 if (ndev->hw_type == BWD_HW)
852 ndev->max_cbs = msix_entries;
853 else
854 ndev->max_cbs = msix_entries - 1;
855
856 return 0;
857
858err2:
859 while (--i >= 0) {
860 msix = &ndev->msix_entries[i];
861 if (ndev->hw_type != BWD_HW && i == ndev->num_msix - 1)
862 free_irq(msix->vector, ndev);
863 else
864 free_irq(msix->vector, &ndev->db_cb[i]);
865 }
866 pci_disable_msix(pdev);
867err1:
868 kfree(ndev->msix_entries);
869 dev_err(&pdev->dev, "Error allocating MSI-X interrupt\n");
870err:
871 ndev->num_msix = 0;
872 return rc;
873}
874
875static int ntb_setup_msi(struct ntb_device *ndev)
876{
877 struct pci_dev *pdev = ndev->pdev;
878 int rc;
879
880 rc = pci_enable_msi(pdev);
881 if (rc)
882 return rc;
883
884 rc = request_irq(pdev->irq, ntb_interrupt, 0, "ntb-msi", ndev);
885 if (rc) {
886 pci_disable_msi(pdev);
887 dev_err(&pdev->dev, "Error allocating MSI interrupt\n");
888 return rc;
889 }
890
891 return 0;
892}
893
894static int ntb_setup_intx(struct ntb_device *ndev)
895{
896 struct pci_dev *pdev = ndev->pdev;
897 int rc;
898
899 pci_msi_off(pdev);
900
901 /* Verify intx is enabled */
902 pci_intx(pdev, 1);
903
904 rc = request_irq(pdev->irq, ntb_interrupt, IRQF_SHARED, "ntb-intx",
905 ndev);
906 if (rc)
907 return rc;
908
909 return 0;
910}
911
912static int __devinit ntb_setup_interrupts(struct ntb_device *ndev)
913{
914 int rc;
915
916 /* On BWD, disable all interrupts. On SNB, disable all but Link
917 * Interrupt. The rest will be unmasked as callbacks are registered.
918 */
919 if (ndev->hw_type == BWD_HW)
920 writeq(~0, ndev->reg_ofs.pdb_mask);
921 else
922 writew(~(1 << ndev->limits.max_db_bits),
923 ndev->reg_ofs.pdb_mask);
924
925 rc = ntb_setup_msix(ndev);
926 if (!rc)
927 goto done;
928
929 ndev->bits_per_vector = 1;
930 ndev->max_cbs = ndev->limits.max_db_bits;
931
932 rc = ntb_setup_msi(ndev);
933 if (!rc)
934 goto done;
935
936 rc = ntb_setup_intx(ndev);
937 if (rc) {
938 dev_err(&ndev->pdev->dev, "no usable interrupts\n");
939 return rc;
940 }
941
942done:
943 return 0;
944}
945
946static void __devexit ntb_free_interrupts(struct ntb_device *ndev)
947{
948 struct pci_dev *pdev = ndev->pdev;
949
950 /* mask interrupts */
951 if (ndev->hw_type == BWD_HW)
952 writeq(~0, ndev->reg_ofs.pdb_mask);
953 else
954 writew(~0, ndev->reg_ofs.pdb_mask);
955
956 if (ndev->num_msix) {
957 struct msix_entry *msix;
958 u32 i;
959
960 for (i = 0; i < ndev->num_msix; i++) {
961 msix = &ndev->msix_entries[i];
962 if (ndev->hw_type != BWD_HW && i == ndev->num_msix - 1)
963 free_irq(msix->vector, ndev);
964 else
965 free_irq(msix->vector, &ndev->db_cb[i]);
966 }
967 pci_disable_msix(pdev);
968 } else {
969 free_irq(pdev->irq, ndev);
970
971 if (pci_dev_msi_enabled(pdev))
972 pci_disable_msi(pdev);
973 }
974}
975
976static int __devinit ntb_create_callbacks(struct ntb_device *ndev)
977{
978 int i;
979
980 /* Checken-egg issue. We won't know how many callbacks are necessary
981 * until we see how many MSI-X vectors we get, but these pointers need
982 * to be passed into the MSI-X register fucntion. So, we allocate the
983 * max, knowing that they might not all be used, to work around this.
984 */
985 ndev->db_cb = kcalloc(ndev->limits.max_db_bits,
986 sizeof(struct ntb_db_cb),
987 GFP_KERNEL);
988 if (!ndev->db_cb)
989 return -ENOMEM;
990
991 for (i = 0; i < ndev->limits.max_db_bits; i++) {
992 ndev->db_cb[i].db_num = i;
993 ndev->db_cb[i].ndev = ndev;
994 }
995
996 return 0;
997}
998
999static void ntb_free_callbacks(struct ntb_device *ndev)
1000{
1001 int i;
1002
1003 for (i = 0; i < ndev->limits.max_db_bits; i++)
1004 ntb_unregister_db_callback(ndev, i);
1005
1006 kfree(ndev->db_cb);
1007}
1008
1009static int __devinit
1010ntb_pci_probe(struct pci_dev *pdev,
1011 __attribute__((unused)) const struct pci_device_id *id)
1012{
1013 struct ntb_device *ndev;
1014 int rc, i;
1015
1016 ndev = kzalloc(sizeof(struct ntb_device), GFP_KERNEL);
1017 if (!ndev)
1018 return -ENOMEM;
1019
1020 ndev->pdev = pdev;
1021 ndev->link_status = NTB_LINK_DOWN;
1022 pci_set_drvdata(pdev, ndev);
1023
1024 rc = pci_enable_device(pdev);
1025 if (rc)
1026 goto err;
1027
1028 pci_set_master(ndev->pdev);
1029
1030 rc = pci_request_selected_regions(pdev, NTB_BAR_MASK, KBUILD_MODNAME);
1031 if (rc)
1032 goto err1;
1033
1034 ndev->reg_base = pci_ioremap_bar(pdev, NTB_BAR_MMIO);
1035 if (!ndev->reg_base) {
1036 dev_warn(&pdev->dev, "Cannot remap BAR 0\n");
1037 rc = -EIO;
1038 goto err2;
1039 }
1040
1041 for (i = 0; i < NTB_NUM_MW; i++) {
1042 ndev->mw[i].bar_sz = pci_resource_len(pdev, MW_TO_BAR(i));
1043 ndev->mw[i].vbase =
1044 ioremap_wc(pci_resource_start(pdev, MW_TO_BAR(i)),
1045 ndev->mw[i].bar_sz);
1046 dev_info(&pdev->dev, "MW %d size %d\n", i,
1047 (u32) pci_resource_len(pdev, MW_TO_BAR(i)));
1048 if (!ndev->mw[i].vbase) {
1049 dev_warn(&pdev->dev, "Cannot remap BAR %d\n",
1050 MW_TO_BAR(i));
1051 rc = -EIO;
1052 goto err3;
1053 }
1054 }
1055
1056 rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
1057 if (rc) {
1058 rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1059 if (rc)
1060 goto err3;
1061
1062 dev_warn(&pdev->dev, "Cannot DMA highmem\n");
1063 }
1064
1065 rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
1066 if (rc) {
1067 rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1068 if (rc)
1069 goto err3;
1070
1071 dev_warn(&pdev->dev, "Cannot DMA consistent highmem\n");
1072 }
1073
1074 rc = ntb_device_setup(ndev);
1075 if (rc)
1076 goto err3;
1077
1078 rc = ntb_create_callbacks(ndev);
1079 if (rc)
1080 goto err4;
1081
1082 rc = ntb_setup_interrupts(ndev);
1083 if (rc)
1084 goto err5;
1085
1086 /* The scratchpad registers keep the values between rmmod/insmod,
1087 * blast them now
1088 */
1089 for (i = 0; i < ndev->limits.max_spads; i++) {
1090 ntb_write_local_spad(ndev, i, 0);
1091 ntb_write_remote_spad(ndev, i, 0);
1092 }
1093
1094 rc = ntb_transport_init(pdev);
1095 if (rc)
1096 goto err6;
1097
1098 /* Let's bring the NTB link up */
1099 writel(NTB_CNTL_BAR23_SNOOP | NTB_CNTL_BAR45_SNOOP,
1100 ndev->reg_ofs.lnk_cntl);
1101
1102 return 0;
1103
1104err6:
1105 ntb_free_interrupts(ndev);
1106err5:
1107 ntb_free_callbacks(ndev);
1108err4:
1109 ntb_device_free(ndev);
1110err3:
1111 for (i--; i >= 0; i--)
1112 iounmap(ndev->mw[i].vbase);
1113 iounmap(ndev->reg_base);
1114err2:
1115 pci_release_selected_regions(pdev, NTB_BAR_MASK);
1116err1:
1117 pci_disable_device(pdev);
1118err:
1119 kfree(ndev);
1120
1121 dev_err(&pdev->dev, "Error loading %s module\n", KBUILD_MODNAME);
1122 return rc;
1123}
1124
1125static void __devexit ntb_pci_remove(struct pci_dev *pdev)
1126{
1127 struct ntb_device *ndev = pci_get_drvdata(pdev);
1128 int i;
1129 u32 ntb_cntl;
1130
1131 /* Bring NTB link down */
1132 ntb_cntl = readl(ndev->reg_ofs.lnk_cntl);
1133 ntb_cntl |= NTB_LINK_DISABLE;
1134 writel(ntb_cntl, ndev->reg_ofs.lnk_cntl);
1135
1136 ntb_transport_free(ndev->ntb_transport);
1137
1138 ntb_free_interrupts(ndev);
1139 ntb_free_callbacks(ndev);
1140 ntb_device_free(ndev);
1141
1142 for (i = 0; i < NTB_NUM_MW; i++)
1143 iounmap(ndev->mw[i].vbase);
1144
1145 iounmap(ndev->reg_base);
1146 pci_release_selected_regions(pdev, NTB_BAR_MASK);
1147 pci_disable_device(pdev);
1148 kfree(ndev);
1149}
1150
1151static struct pci_driver ntb_pci_driver = {
1152 .name = KBUILD_MODNAME,
1153 .id_table = ntb_pci_tbl,
1154 .probe = ntb_pci_probe,
1155 .remove = __devexit_p(ntb_pci_remove),
1156};
1157module_pci_driver(ntb_pci_driver);