xref: /linux/drivers/ntb/hw/amd/ntb_hw_amd.c (revision 0883c2c06fb5bcf5b9e008270827e63c09a88c1e)
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) 2016 Advanced Micro Devices, Inc. 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) 2016 Advanced Micro Devices, Inc. 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 AMD 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  * AMD PCIe NTB Linux driver
44  *
45  * Contact Information:
46  * Xiangliang Yu <Xiangliang.Yu@amd.com>
47  */
48 
49 #include <linux/debugfs.h>
50 #include <linux/delay.h>
51 #include <linux/init.h>
52 #include <linux/interrupt.h>
53 #include <linux/module.h>
54 #include <linux/acpi.h>
55 #include <linux/pci.h>
56 #include <linux/random.h>
57 #include <linux/slab.h>
58 #include <linux/ntb.h>
59 
60 #include "ntb_hw_amd.h"
61 
62 #define NTB_NAME	"ntb_hw_amd"
63 #define NTB_DESC	"AMD(R) PCI-E Non-Transparent Bridge Driver"
64 #define NTB_VER		"1.0"
65 
66 MODULE_DESCRIPTION(NTB_DESC);
67 MODULE_VERSION(NTB_VER);
68 MODULE_LICENSE("Dual BSD/GPL");
69 MODULE_AUTHOR("AMD Inc.");
70 
71 static const struct file_operations amd_ntb_debugfs_info;
72 static struct dentry *debugfs_dir;
73 
74 static int ndev_mw_to_bar(struct amd_ntb_dev *ndev, int idx)
75 {
76 	if (idx < 0 || idx > ndev->mw_count)
77 		return -EINVAL;
78 
79 	return 1 << idx;
80 }
81 
82 static int amd_ntb_mw_count(struct ntb_dev *ntb)
83 {
84 	return ntb_ndev(ntb)->mw_count;
85 }
86 
87 static int amd_ntb_mw_get_range(struct ntb_dev *ntb, int idx,
88 				phys_addr_t *base,
89 				resource_size_t *size,
90 				resource_size_t *align,
91 				resource_size_t *align_size)
92 {
93 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
94 	int bar;
95 
96 	bar = ndev_mw_to_bar(ndev, idx);
97 	if (bar < 0)
98 		return bar;
99 
100 	if (base)
101 		*base = pci_resource_start(ndev->ntb.pdev, bar);
102 
103 	if (size)
104 		*size = pci_resource_len(ndev->ntb.pdev, bar);
105 
106 	if (align)
107 		*align = SZ_4K;
108 
109 	if (align_size)
110 		*align_size = 1;
111 
112 	return 0;
113 }
114 
115 static int amd_ntb_mw_set_trans(struct ntb_dev *ntb, int idx,
116 				dma_addr_t addr, resource_size_t size)
117 {
118 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
119 	unsigned long xlat_reg, limit_reg = 0;
120 	resource_size_t mw_size;
121 	void __iomem *mmio, *peer_mmio;
122 	u64 base_addr, limit, reg_val;
123 	int bar;
124 
125 	bar = ndev_mw_to_bar(ndev, idx);
126 	if (bar < 0)
127 		return bar;
128 
129 	mw_size = pci_resource_len(ndev->ntb.pdev, bar);
130 
131 	/* make sure the range fits in the usable mw size */
132 	if (size > mw_size)
133 		return -EINVAL;
134 
135 	mmio = ndev->self_mmio;
136 	peer_mmio = ndev->peer_mmio;
137 
138 	base_addr = pci_resource_start(ndev->ntb.pdev, bar);
139 
140 	if (bar != 1) {
141 		xlat_reg = AMD_BAR23XLAT_OFFSET + ((bar - 2) << 3);
142 		limit_reg = AMD_BAR23LMT_OFFSET + ((bar - 2) << 3);
143 
144 		/* Set the limit if supported */
145 		limit = base_addr + size;
146 
147 		/* set and verify setting the translation address */
148 		write64(addr, peer_mmio + xlat_reg);
149 		reg_val = read64(peer_mmio + xlat_reg);
150 		if (reg_val != addr) {
151 			write64(0, peer_mmio + xlat_reg);
152 			return -EIO;
153 		}
154 
155 		/* set and verify setting the limit */
156 		write64(limit, mmio + limit_reg);
157 		reg_val = read64(mmio + limit_reg);
158 		if (reg_val != limit) {
159 			write64(base_addr, mmio + limit_reg);
160 			write64(0, peer_mmio + xlat_reg);
161 			return -EIO;
162 		}
163 	} else {
164 		xlat_reg = AMD_BAR1XLAT_OFFSET;
165 		limit_reg = AMD_BAR1LMT_OFFSET;
166 
167 		/* split bar addr range must all be 32 bit */
168 		if (addr & (~0ull << 32))
169 			return -EINVAL;
170 		if ((addr + size) & (~0ull << 32))
171 			return -EINVAL;
172 
173 		/* Set the limit if supported */
174 		limit = base_addr + size;
175 
176 		/* set and verify setting the translation address */
177 		write64(addr, peer_mmio + xlat_reg);
178 		reg_val = read64(peer_mmio + xlat_reg);
179 		if (reg_val != addr) {
180 			write64(0, peer_mmio + xlat_reg);
181 			return -EIO;
182 		}
183 
184 		/* set and verify setting the limit */
185 		writel(limit, mmio + limit_reg);
186 		reg_val = readl(mmio + limit_reg);
187 		if (reg_val != limit) {
188 			writel(base_addr, mmio + limit_reg);
189 			writel(0, peer_mmio + xlat_reg);
190 			return -EIO;
191 		}
192 	}
193 
194 	return 0;
195 }
196 
197 static int amd_link_is_up(struct amd_ntb_dev *ndev)
198 {
199 	if (!ndev->peer_sta)
200 		return NTB_LNK_STA_ACTIVE(ndev->cntl_sta);
201 
202 	/* If peer_sta is reset or D0 event, the ISR has
203 	 * started a timer to check link status of hardware.
204 	 * So here just clear status bit. And if peer_sta is
205 	 * D3 or PME_TO, D0/reset event will be happened when
206 	 * system wakeup/poweron, so do nothing here.
207 	 */
208 	if (ndev->peer_sta & AMD_PEER_RESET_EVENT)
209 		ndev->peer_sta &= ~AMD_PEER_RESET_EVENT;
210 	else if (ndev->peer_sta & AMD_PEER_D0_EVENT)
211 		ndev->peer_sta = 0;
212 
213 	return 0;
214 }
215 
216 static int amd_ntb_link_is_up(struct ntb_dev *ntb,
217 			      enum ntb_speed *speed,
218 			      enum ntb_width *width)
219 {
220 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
221 	int ret = 0;
222 
223 	if (amd_link_is_up(ndev)) {
224 		if (speed)
225 			*speed = NTB_LNK_STA_SPEED(ndev->lnk_sta);
226 		if (width)
227 			*width = NTB_LNK_STA_WIDTH(ndev->lnk_sta);
228 
229 		dev_dbg(ndev_dev(ndev), "link is up.\n");
230 
231 		ret = 1;
232 	} else {
233 		if (speed)
234 			*speed = NTB_SPEED_NONE;
235 		if (width)
236 			*width = NTB_WIDTH_NONE;
237 
238 		dev_dbg(ndev_dev(ndev), "link is down.\n");
239 	}
240 
241 	return ret;
242 }
243 
244 static int amd_ntb_link_enable(struct ntb_dev *ntb,
245 			       enum ntb_speed max_speed,
246 			       enum ntb_width max_width)
247 {
248 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
249 	void __iomem *mmio = ndev->self_mmio;
250 	u32 ntb_ctl;
251 
252 	/* Enable event interrupt */
253 	ndev->int_mask &= ~AMD_EVENT_INTMASK;
254 	writel(ndev->int_mask, mmio + AMD_INTMASK_OFFSET);
255 
256 	if (ndev->ntb.topo == NTB_TOPO_SEC)
257 		return -EINVAL;
258 	dev_dbg(ndev_dev(ndev), "Enabling Link.\n");
259 
260 	ntb_ctl = readl(mmio + AMD_CNTL_OFFSET);
261 	ntb_ctl |= (PMM_REG_CTL | SMM_REG_CTL);
262 	writel(ntb_ctl, mmio + AMD_CNTL_OFFSET);
263 
264 	return 0;
265 }
266 
267 static int amd_ntb_link_disable(struct ntb_dev *ntb)
268 {
269 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
270 	void __iomem *mmio = ndev->self_mmio;
271 	u32 ntb_ctl;
272 
273 	/* Disable event interrupt */
274 	ndev->int_mask |= AMD_EVENT_INTMASK;
275 	writel(ndev->int_mask, mmio + AMD_INTMASK_OFFSET);
276 
277 	if (ndev->ntb.topo == NTB_TOPO_SEC)
278 		return -EINVAL;
279 	dev_dbg(ndev_dev(ndev), "Enabling Link.\n");
280 
281 	ntb_ctl = readl(mmio + AMD_CNTL_OFFSET);
282 	ntb_ctl &= ~(PMM_REG_CTL | SMM_REG_CTL);
283 	writel(ntb_ctl, mmio + AMD_CNTL_OFFSET);
284 
285 	return 0;
286 }
287 
288 static u64 amd_ntb_db_valid_mask(struct ntb_dev *ntb)
289 {
290 	return ntb_ndev(ntb)->db_valid_mask;
291 }
292 
293 static int amd_ntb_db_vector_count(struct ntb_dev *ntb)
294 {
295 	return ntb_ndev(ntb)->db_count;
296 }
297 
298 static u64 amd_ntb_db_vector_mask(struct ntb_dev *ntb, int db_vector)
299 {
300 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
301 
302 	if (db_vector < 0 || db_vector > ndev->db_count)
303 		return 0;
304 
305 	return ntb_ndev(ntb)->db_valid_mask & (1 << db_vector);
306 }
307 
308 static u64 amd_ntb_db_read(struct ntb_dev *ntb)
309 {
310 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
311 	void __iomem *mmio = ndev->self_mmio;
312 
313 	return (u64)readw(mmio + AMD_DBSTAT_OFFSET);
314 }
315 
316 static int amd_ntb_db_clear(struct ntb_dev *ntb, u64 db_bits)
317 {
318 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
319 	void __iomem *mmio = ndev->self_mmio;
320 
321 	writew((u16)db_bits, mmio + AMD_DBSTAT_OFFSET);
322 
323 	return 0;
324 }
325 
326 static int amd_ntb_db_set_mask(struct ntb_dev *ntb, u64 db_bits)
327 {
328 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
329 	void __iomem *mmio = ndev->self_mmio;
330 	unsigned long flags;
331 
332 	if (db_bits & ~ndev->db_valid_mask)
333 		return -EINVAL;
334 
335 	spin_lock_irqsave(&ndev->db_mask_lock, flags);
336 	ndev->db_mask |= db_bits;
337 	writew((u16)ndev->db_mask, mmio + AMD_DBMASK_OFFSET);
338 	spin_unlock_irqrestore(&ndev->db_mask_lock, flags);
339 
340 	return 0;
341 }
342 
343 static int amd_ntb_db_clear_mask(struct ntb_dev *ntb, u64 db_bits)
344 {
345 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
346 	void __iomem *mmio = ndev->self_mmio;
347 	unsigned long flags;
348 
349 	if (db_bits & ~ndev->db_valid_mask)
350 		return -EINVAL;
351 
352 	spin_lock_irqsave(&ndev->db_mask_lock, flags);
353 	ndev->db_mask &= ~db_bits;
354 	writew((u16)ndev->db_mask, mmio + AMD_DBMASK_OFFSET);
355 	spin_unlock_irqrestore(&ndev->db_mask_lock, flags);
356 
357 	return 0;
358 }
359 
360 static int amd_ntb_peer_db_set(struct ntb_dev *ntb, u64 db_bits)
361 {
362 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
363 	void __iomem *mmio = ndev->self_mmio;
364 
365 	writew((u16)db_bits, mmio + AMD_DBREQ_OFFSET);
366 
367 	return 0;
368 }
369 
370 static int amd_ntb_spad_count(struct ntb_dev *ntb)
371 {
372 	return ntb_ndev(ntb)->spad_count;
373 }
374 
375 static u32 amd_ntb_spad_read(struct ntb_dev *ntb, int idx)
376 {
377 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
378 	void __iomem *mmio = ndev->self_mmio;
379 	u32 offset;
380 
381 	if (idx < 0 || idx >= ndev->spad_count)
382 		return 0;
383 
384 	offset = ndev->self_spad + (idx << 2);
385 	return readl(mmio + AMD_SPAD_OFFSET + offset);
386 }
387 
388 static int amd_ntb_spad_write(struct ntb_dev *ntb,
389 			      int idx, u32 val)
390 {
391 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
392 	void __iomem *mmio = ndev->self_mmio;
393 	u32 offset;
394 
395 	if (idx < 0 || idx >= ndev->spad_count)
396 		return -EINVAL;
397 
398 	offset = ndev->self_spad + (idx << 2);
399 	writel(val, mmio + AMD_SPAD_OFFSET + offset);
400 
401 	return 0;
402 }
403 
404 static u32 amd_ntb_peer_spad_read(struct ntb_dev *ntb, int idx)
405 {
406 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
407 	void __iomem *mmio = ndev->self_mmio;
408 	u32 offset;
409 
410 	if (idx < 0 || idx >= ndev->spad_count)
411 		return -EINVAL;
412 
413 	offset = ndev->peer_spad + (idx << 2);
414 	return readl(mmio + AMD_SPAD_OFFSET + offset);
415 }
416 
417 static int amd_ntb_peer_spad_write(struct ntb_dev *ntb,
418 				   int idx, u32 val)
419 {
420 	struct amd_ntb_dev *ndev = ntb_ndev(ntb);
421 	void __iomem *mmio = ndev->self_mmio;
422 	u32 offset;
423 
424 	if (idx < 0 || idx >= ndev->spad_count)
425 		return -EINVAL;
426 
427 	offset = ndev->peer_spad + (idx << 2);
428 	writel(val, mmio + AMD_SPAD_OFFSET + offset);
429 
430 	return 0;
431 }
432 
433 static const struct ntb_dev_ops amd_ntb_ops = {
434 	.mw_count		= amd_ntb_mw_count,
435 	.mw_get_range		= amd_ntb_mw_get_range,
436 	.mw_set_trans		= amd_ntb_mw_set_trans,
437 	.link_is_up		= amd_ntb_link_is_up,
438 	.link_enable		= amd_ntb_link_enable,
439 	.link_disable		= amd_ntb_link_disable,
440 	.db_valid_mask		= amd_ntb_db_valid_mask,
441 	.db_vector_count	= amd_ntb_db_vector_count,
442 	.db_vector_mask		= amd_ntb_db_vector_mask,
443 	.db_read		= amd_ntb_db_read,
444 	.db_clear		= amd_ntb_db_clear,
445 	.db_set_mask		= amd_ntb_db_set_mask,
446 	.db_clear_mask		= amd_ntb_db_clear_mask,
447 	.peer_db_set		= amd_ntb_peer_db_set,
448 	.spad_count		= amd_ntb_spad_count,
449 	.spad_read		= amd_ntb_spad_read,
450 	.spad_write		= amd_ntb_spad_write,
451 	.peer_spad_read		= amd_ntb_peer_spad_read,
452 	.peer_spad_write	= amd_ntb_peer_spad_write,
453 };
454 
455 static void amd_ack_smu(struct amd_ntb_dev *ndev, u32 bit)
456 {
457 	void __iomem *mmio = ndev->self_mmio;
458 	int reg;
459 
460 	reg = readl(mmio + AMD_SMUACK_OFFSET);
461 	reg |= bit;
462 	writel(reg, mmio + AMD_SMUACK_OFFSET);
463 
464 	ndev->peer_sta |= bit;
465 }
466 
467 static void amd_handle_event(struct amd_ntb_dev *ndev, int vec)
468 {
469 	void __iomem *mmio = ndev->self_mmio;
470 	u32 status;
471 
472 	status = readl(mmio + AMD_INTSTAT_OFFSET);
473 	if (!(status & AMD_EVENT_INTMASK))
474 		return;
475 
476 	dev_dbg(ndev_dev(ndev), "status = 0x%x and vec = %d\n", status, vec);
477 
478 	status &= AMD_EVENT_INTMASK;
479 	switch (status) {
480 	case AMD_PEER_FLUSH_EVENT:
481 		dev_info(ndev_dev(ndev), "Flush is done.\n");
482 		break;
483 	case AMD_PEER_RESET_EVENT:
484 		amd_ack_smu(ndev, AMD_PEER_RESET_EVENT);
485 
486 		/* link down first */
487 		ntb_link_event(&ndev->ntb);
488 		/* polling peer status */
489 		schedule_delayed_work(&ndev->hb_timer, AMD_LINK_HB_TIMEOUT);
490 
491 		break;
492 	case AMD_PEER_D3_EVENT:
493 	case AMD_PEER_PMETO_EVENT:
494 		amd_ack_smu(ndev, status);
495 
496 		/* link down */
497 		ntb_link_event(&ndev->ntb);
498 
499 		break;
500 	case AMD_PEER_D0_EVENT:
501 		mmio = ndev->peer_mmio;
502 		status = readl(mmio + AMD_PMESTAT_OFFSET);
503 		/* check if this is WAKEUP event */
504 		if (status & 0x1)
505 			dev_info(ndev_dev(ndev), "Wakeup is done.\n");
506 
507 		amd_ack_smu(ndev, AMD_PEER_D0_EVENT);
508 
509 		/* start a timer to poll link status */
510 		schedule_delayed_work(&ndev->hb_timer,
511 				      AMD_LINK_HB_TIMEOUT);
512 		break;
513 	default:
514 		dev_info(ndev_dev(ndev), "event status = 0x%x.\n", status);
515 		break;
516 	}
517 }
518 
519 static irqreturn_t ndev_interrupt(struct amd_ntb_dev *ndev, int vec)
520 {
521 	dev_dbg(ndev_dev(ndev), "vec %d\n", vec);
522 
523 	if (vec > (AMD_DB_CNT - 1) || (ndev->msix_vec_count == 1))
524 		amd_handle_event(ndev, vec);
525 
526 	if (vec < AMD_DB_CNT)
527 		ntb_db_event(&ndev->ntb, vec);
528 
529 	return IRQ_HANDLED;
530 }
531 
532 static irqreturn_t ndev_vec_isr(int irq, void *dev)
533 {
534 	struct amd_ntb_vec *nvec = dev;
535 
536 	return ndev_interrupt(nvec->ndev, nvec->num);
537 }
538 
539 static irqreturn_t ndev_irq_isr(int irq, void *dev)
540 {
541 	struct amd_ntb_dev *ndev = dev;
542 
543 	return ndev_interrupt(ndev, irq - ndev_pdev(ndev)->irq);
544 }
545 
546 static int ndev_init_isr(struct amd_ntb_dev *ndev,
547 			 int msix_min, int msix_max)
548 {
549 	struct pci_dev *pdev;
550 	int rc, i, msix_count, node;
551 
552 	pdev = ndev_pdev(ndev);
553 
554 	node = dev_to_node(&pdev->dev);
555 
556 	ndev->db_mask = ndev->db_valid_mask;
557 
558 	/* Try to set up msix irq */
559 	ndev->vec = kzalloc_node(msix_max * sizeof(*ndev->vec),
560 				 GFP_KERNEL, node);
561 	if (!ndev->vec)
562 		goto err_msix_vec_alloc;
563 
564 	ndev->msix = kzalloc_node(msix_max * sizeof(*ndev->msix),
565 				  GFP_KERNEL, node);
566 	if (!ndev->msix)
567 		goto err_msix_alloc;
568 
569 	for (i = 0; i < msix_max; ++i)
570 		ndev->msix[i].entry = i;
571 
572 	msix_count = pci_enable_msix_range(pdev, ndev->msix,
573 					   msix_min, msix_max);
574 	if (msix_count < 0)
575 		goto err_msix_enable;
576 
577 	/* NOTE: Disable MSIX if msix count is less than 16 because of
578 	 * hardware limitation.
579 	 */
580 	if (msix_count < msix_min) {
581 		pci_disable_msix(pdev);
582 		goto err_msix_enable;
583 	}
584 
585 	for (i = 0; i < msix_count; ++i) {
586 		ndev->vec[i].ndev = ndev;
587 		ndev->vec[i].num = i;
588 		rc = request_irq(ndev->msix[i].vector, ndev_vec_isr, 0,
589 				 "ndev_vec_isr", &ndev->vec[i]);
590 		if (rc)
591 			goto err_msix_request;
592 	}
593 
594 	dev_dbg(ndev_dev(ndev), "Using msix interrupts\n");
595 	ndev->db_count = msix_min;
596 	ndev->msix_vec_count = msix_max;
597 	return 0;
598 
599 err_msix_request:
600 	while (i-- > 0)
601 		free_irq(ndev->msix[i].vector, ndev);
602 	pci_disable_msix(pdev);
603 err_msix_enable:
604 	kfree(ndev->msix);
605 err_msix_alloc:
606 	kfree(ndev->vec);
607 err_msix_vec_alloc:
608 	ndev->msix = NULL;
609 	ndev->vec = NULL;
610 
611 	/* Try to set up msi irq */
612 	rc = pci_enable_msi(pdev);
613 	if (rc)
614 		goto err_msi_enable;
615 
616 	rc = request_irq(pdev->irq, ndev_irq_isr, 0,
617 			 "ndev_irq_isr", ndev);
618 	if (rc)
619 		goto err_msi_request;
620 
621 	dev_dbg(ndev_dev(ndev), "Using msi interrupts\n");
622 	ndev->db_count = 1;
623 	ndev->msix_vec_count = 1;
624 	return 0;
625 
626 err_msi_request:
627 	pci_disable_msi(pdev);
628 err_msi_enable:
629 
630 	/* Try to set up intx irq */
631 	pci_intx(pdev, 1);
632 
633 	rc = request_irq(pdev->irq, ndev_irq_isr, IRQF_SHARED,
634 			 "ndev_irq_isr", ndev);
635 	if (rc)
636 		goto err_intx_request;
637 
638 	dev_dbg(ndev_dev(ndev), "Using intx interrupts\n");
639 	ndev->db_count = 1;
640 	ndev->msix_vec_count = 1;
641 	return 0;
642 
643 err_intx_request:
644 	return rc;
645 }
646 
647 static void ndev_deinit_isr(struct amd_ntb_dev *ndev)
648 {
649 	struct pci_dev *pdev;
650 	void __iomem *mmio = ndev->self_mmio;
651 	int i;
652 
653 	pdev = ndev_pdev(ndev);
654 
655 	/* Mask all doorbell interrupts */
656 	ndev->db_mask = ndev->db_valid_mask;
657 	writel(ndev->db_mask, mmio + AMD_DBMASK_OFFSET);
658 
659 	if (ndev->msix) {
660 		i = ndev->msix_vec_count;
661 		while (i--)
662 			free_irq(ndev->msix[i].vector, &ndev->vec[i]);
663 		pci_disable_msix(pdev);
664 		kfree(ndev->msix);
665 		kfree(ndev->vec);
666 	} else {
667 		free_irq(pdev->irq, ndev);
668 		if (pci_dev_msi_enabled(pdev))
669 			pci_disable_msi(pdev);
670 		else
671 			pci_intx(pdev, 0);
672 	}
673 }
674 
675 static ssize_t ndev_debugfs_read(struct file *filp, char __user *ubuf,
676 				 size_t count, loff_t *offp)
677 {
678 	struct amd_ntb_dev *ndev;
679 	void __iomem *mmio;
680 	char *buf;
681 	size_t buf_size;
682 	ssize_t ret, off;
683 	union { u64 v64; u32 v32; u16 v16; } u;
684 
685 	ndev = filp->private_data;
686 	mmio = ndev->self_mmio;
687 
688 	buf_size = min(count, 0x800ul);
689 
690 	buf = kmalloc(buf_size, GFP_KERNEL);
691 	if (!buf)
692 		return -ENOMEM;
693 
694 	off = 0;
695 
696 	off += scnprintf(buf + off, buf_size - off,
697 			 "NTB Device Information:\n");
698 
699 	off += scnprintf(buf + off, buf_size - off,
700 			 "Connection Topology -\t%s\n",
701 			 ntb_topo_string(ndev->ntb.topo));
702 
703 	off += scnprintf(buf + off, buf_size - off,
704 			 "LNK STA -\t\t%#06x\n", ndev->lnk_sta);
705 
706 	if (!amd_link_is_up(ndev)) {
707 		off += scnprintf(buf + off, buf_size - off,
708 				 "Link Status -\t\tDown\n");
709 	} else {
710 		off += scnprintf(buf + off, buf_size - off,
711 				 "Link Status -\t\tUp\n");
712 		off += scnprintf(buf + off, buf_size - off,
713 				 "Link Speed -\t\tPCI-E Gen %u\n",
714 				 NTB_LNK_STA_SPEED(ndev->lnk_sta));
715 		off += scnprintf(buf + off, buf_size - off,
716 				 "Link Width -\t\tx%u\n",
717 				 NTB_LNK_STA_WIDTH(ndev->lnk_sta));
718 	}
719 
720 	off += scnprintf(buf + off, buf_size - off,
721 			 "Memory Window Count -\t%u\n", ndev->mw_count);
722 	off += scnprintf(buf + off, buf_size - off,
723 			 "Scratchpad Count -\t%u\n", ndev->spad_count);
724 	off += scnprintf(buf + off, buf_size - off,
725 			 "Doorbell Count -\t%u\n", ndev->db_count);
726 	off += scnprintf(buf + off, buf_size - off,
727 			 "MSIX Vector Count -\t%u\n", ndev->msix_vec_count);
728 
729 	off += scnprintf(buf + off, buf_size - off,
730 			 "Doorbell Valid Mask -\t%#llx\n", ndev->db_valid_mask);
731 
732 	u.v32 = readl(ndev->self_mmio + AMD_DBMASK_OFFSET);
733 	off += scnprintf(buf + off, buf_size - off,
734 			 "Doorbell Mask -\t\t\t%#06x\n", u.v32);
735 
736 	u.v32 = readl(mmio + AMD_DBSTAT_OFFSET);
737 	off += scnprintf(buf + off, buf_size - off,
738 			 "Doorbell Bell -\t\t\t%#06x\n", u.v32);
739 
740 	off += scnprintf(buf + off, buf_size - off,
741 			 "\nNTB Incoming XLAT:\n");
742 
743 	u.v64 = read64(mmio + AMD_BAR1XLAT_OFFSET);
744 	off += scnprintf(buf + off, buf_size - off,
745 			 "XLAT1 -\t\t%#018llx\n", u.v64);
746 
747 	u.v64 = read64(ndev->self_mmio + AMD_BAR23XLAT_OFFSET);
748 	off += scnprintf(buf + off, buf_size - off,
749 			 "XLAT23 -\t\t%#018llx\n", u.v64);
750 
751 	u.v64 = read64(ndev->self_mmio + AMD_BAR45XLAT_OFFSET);
752 	off += scnprintf(buf + off, buf_size - off,
753 			 "XLAT45 -\t\t%#018llx\n", u.v64);
754 
755 	u.v32 = readl(mmio + AMD_BAR1LMT_OFFSET);
756 	off += scnprintf(buf + off, buf_size - off,
757 			 "LMT1 -\t\t\t%#06x\n", u.v32);
758 
759 	u.v64 = read64(ndev->self_mmio + AMD_BAR23LMT_OFFSET);
760 	off += scnprintf(buf + off, buf_size - off,
761 			 "LMT23 -\t\t\t%#018llx\n", u.v64);
762 
763 	u.v64 = read64(ndev->self_mmio + AMD_BAR45LMT_OFFSET);
764 	off += scnprintf(buf + off, buf_size - off,
765 			 "LMT45 -\t\t\t%#018llx\n", u.v64);
766 
767 	ret = simple_read_from_buffer(ubuf, count, offp, buf, off);
768 	kfree(buf);
769 	return ret;
770 }
771 
772 static void ndev_init_debugfs(struct amd_ntb_dev *ndev)
773 {
774 	if (!debugfs_dir) {
775 		ndev->debugfs_dir = NULL;
776 		ndev->debugfs_info = NULL;
777 	} else {
778 		ndev->debugfs_dir =
779 			debugfs_create_dir(ndev_name(ndev), debugfs_dir);
780 		if (!ndev->debugfs_dir)
781 			ndev->debugfs_info = NULL;
782 		else
783 			ndev->debugfs_info =
784 				debugfs_create_file("info", S_IRUSR,
785 						    ndev->debugfs_dir, ndev,
786 						    &amd_ntb_debugfs_info);
787 	}
788 }
789 
790 static void ndev_deinit_debugfs(struct amd_ntb_dev *ndev)
791 {
792 	debugfs_remove_recursive(ndev->debugfs_dir);
793 }
794 
795 static inline void ndev_init_struct(struct amd_ntb_dev *ndev,
796 				    struct pci_dev *pdev)
797 {
798 	ndev->ntb.pdev = pdev;
799 	ndev->ntb.topo = NTB_TOPO_NONE;
800 	ndev->ntb.ops = &amd_ntb_ops;
801 	ndev->int_mask = AMD_EVENT_INTMASK;
802 	spin_lock_init(&ndev->db_mask_lock);
803 }
804 
805 static int amd_poll_link(struct amd_ntb_dev *ndev)
806 {
807 	void __iomem *mmio = ndev->peer_mmio;
808 	u32 reg, stat;
809 	int rc;
810 
811 	reg = readl(mmio + AMD_SIDEINFO_OFFSET);
812 	reg &= NTB_LIN_STA_ACTIVE_BIT;
813 
814 	dev_dbg(ndev_dev(ndev), "%s: reg_val = 0x%x.\n", __func__, reg);
815 
816 	if (reg == ndev->cntl_sta)
817 		return 0;
818 
819 	ndev->cntl_sta = reg;
820 
821 	rc = pci_read_config_dword(ndev->ntb.pdev,
822 				   AMD_LINK_STATUS_OFFSET, &stat);
823 	if (rc)
824 		return 0;
825 	ndev->lnk_sta = stat;
826 
827 	return 1;
828 }
829 
830 static void amd_link_hb(struct work_struct *work)
831 {
832 	struct amd_ntb_dev *ndev = hb_ndev(work);
833 
834 	if (amd_poll_link(ndev))
835 		ntb_link_event(&ndev->ntb);
836 
837 	if (!amd_link_is_up(ndev))
838 		schedule_delayed_work(&ndev->hb_timer, AMD_LINK_HB_TIMEOUT);
839 }
840 
841 static int amd_init_isr(struct amd_ntb_dev *ndev)
842 {
843 	return ndev_init_isr(ndev, AMD_DB_CNT, AMD_MSIX_VECTOR_CNT);
844 }
845 
846 static void amd_init_side_info(struct amd_ntb_dev *ndev)
847 {
848 	void __iomem *mmio = ndev->self_mmio;
849 	unsigned int reg;
850 
851 	reg = readl(mmio + AMD_SIDEINFO_OFFSET);
852 	if (!(reg & AMD_SIDE_READY)) {
853 		reg |= AMD_SIDE_READY;
854 		writel(reg, mmio + AMD_SIDEINFO_OFFSET);
855 	}
856 }
857 
858 static void amd_deinit_side_info(struct amd_ntb_dev *ndev)
859 {
860 	void __iomem *mmio = ndev->self_mmio;
861 	unsigned int reg;
862 
863 	reg = readl(mmio + AMD_SIDEINFO_OFFSET);
864 	if (reg & AMD_SIDE_READY) {
865 		reg &= ~AMD_SIDE_READY;
866 		writel(reg, mmio + AMD_SIDEINFO_OFFSET);
867 		readl(mmio + AMD_SIDEINFO_OFFSET);
868 	}
869 }
870 
871 static int amd_init_ntb(struct amd_ntb_dev *ndev)
872 {
873 	void __iomem *mmio = ndev->self_mmio;
874 
875 	ndev->mw_count = AMD_MW_CNT;
876 	ndev->spad_count = AMD_SPADS_CNT;
877 	ndev->db_count = AMD_DB_CNT;
878 
879 	switch (ndev->ntb.topo) {
880 	case NTB_TOPO_PRI:
881 	case NTB_TOPO_SEC:
882 		ndev->spad_count >>= 1;
883 		if (ndev->ntb.topo == NTB_TOPO_PRI) {
884 			ndev->self_spad = 0;
885 			ndev->peer_spad = 0x20;
886 		} else {
887 			ndev->self_spad = 0x20;
888 			ndev->peer_spad = 0;
889 		}
890 
891 		INIT_DELAYED_WORK(&ndev->hb_timer, amd_link_hb);
892 		schedule_delayed_work(&ndev->hb_timer, AMD_LINK_HB_TIMEOUT);
893 
894 		break;
895 	default:
896 		dev_err(ndev_dev(ndev), "AMD NTB does not support B2B mode.\n");
897 		return -EINVAL;
898 	}
899 
900 	ndev->db_valid_mask = BIT_ULL(ndev->db_count) - 1;
901 
902 	/* Mask event interrupts */
903 	writel(ndev->int_mask, mmio + AMD_INTMASK_OFFSET);
904 
905 	return 0;
906 }
907 
908 static enum ntb_topo amd_get_topo(struct amd_ntb_dev *ndev)
909 {
910 	void __iomem *mmio = ndev->self_mmio;
911 	u32 info;
912 
913 	info = readl(mmio + AMD_SIDEINFO_OFFSET);
914 	if (info & AMD_SIDE_MASK)
915 		return NTB_TOPO_SEC;
916 	else
917 		return NTB_TOPO_PRI;
918 }
919 
920 static int amd_init_dev(struct amd_ntb_dev *ndev)
921 {
922 	struct pci_dev *pdev;
923 	int rc = 0;
924 
925 	pdev = ndev_pdev(ndev);
926 
927 	ndev->ntb.topo = amd_get_topo(ndev);
928 	dev_dbg(ndev_dev(ndev), "AMD NTB topo is %s\n",
929 		ntb_topo_string(ndev->ntb.topo));
930 
931 	rc = amd_init_ntb(ndev);
932 	if (rc)
933 		return rc;
934 
935 	rc = amd_init_isr(ndev);
936 	if (rc) {
937 		dev_err(ndev_dev(ndev), "fail to init isr.\n");
938 		return rc;
939 	}
940 
941 	ndev->db_valid_mask = BIT_ULL(ndev->db_count) - 1;
942 
943 	return 0;
944 }
945 
946 static void amd_deinit_dev(struct amd_ntb_dev *ndev)
947 {
948 	cancel_delayed_work_sync(&ndev->hb_timer);
949 
950 	ndev_deinit_isr(ndev);
951 }
952 
953 static int amd_ntb_init_pci(struct amd_ntb_dev *ndev,
954 			    struct pci_dev *pdev)
955 {
956 	int rc;
957 
958 	pci_set_drvdata(pdev, ndev);
959 
960 	rc = pci_enable_device(pdev);
961 	if (rc)
962 		goto err_pci_enable;
963 
964 	rc = pci_request_regions(pdev, NTB_NAME);
965 	if (rc)
966 		goto err_pci_regions;
967 
968 	pci_set_master(pdev);
969 
970 	rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
971 	if (rc) {
972 		rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
973 		if (rc)
974 			goto err_dma_mask;
975 		dev_warn(ndev_dev(ndev), "Cannot DMA highmem\n");
976 	}
977 
978 	rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
979 	if (rc) {
980 		rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
981 		if (rc)
982 			goto err_dma_mask;
983 		dev_warn(ndev_dev(ndev), "Cannot DMA consistent highmem\n");
984 	}
985 
986 	ndev->self_mmio = pci_iomap(pdev, 0, 0);
987 	if (!ndev->self_mmio) {
988 		rc = -EIO;
989 		goto err_dma_mask;
990 	}
991 	ndev->peer_mmio = ndev->self_mmio + AMD_PEER_OFFSET;
992 
993 	return 0;
994 
995 err_dma_mask:
996 	pci_clear_master(pdev);
997 err_pci_regions:
998 	pci_disable_device(pdev);
999 err_pci_enable:
1000 	pci_set_drvdata(pdev, NULL);
1001 	return rc;
1002 }
1003 
1004 static void amd_ntb_deinit_pci(struct amd_ntb_dev *ndev)
1005 {
1006 	struct pci_dev *pdev = ndev_pdev(ndev);
1007 
1008 	pci_iounmap(pdev, ndev->self_mmio);
1009 
1010 	pci_clear_master(pdev);
1011 	pci_release_regions(pdev);
1012 	pci_disable_device(pdev);
1013 	pci_set_drvdata(pdev, NULL);
1014 }
1015 
1016 static int amd_ntb_pci_probe(struct pci_dev *pdev,
1017 			     const struct pci_device_id *id)
1018 {
1019 	struct amd_ntb_dev *ndev;
1020 	int rc, node;
1021 
1022 	node = dev_to_node(&pdev->dev);
1023 
1024 	ndev = kzalloc_node(sizeof(*ndev), GFP_KERNEL, node);
1025 	if (!ndev) {
1026 		rc = -ENOMEM;
1027 		goto err_ndev;
1028 	}
1029 
1030 	ndev_init_struct(ndev, pdev);
1031 
1032 	rc = amd_ntb_init_pci(ndev, pdev);
1033 	if (rc)
1034 		goto err_init_pci;
1035 
1036 	rc = amd_init_dev(ndev);
1037 	if (rc)
1038 		goto err_init_dev;
1039 
1040 	/* write side info */
1041 	amd_init_side_info(ndev);
1042 
1043 	amd_poll_link(ndev);
1044 
1045 	ndev_init_debugfs(ndev);
1046 
1047 	rc = ntb_register_device(&ndev->ntb);
1048 	if (rc)
1049 		goto err_register;
1050 
1051 	dev_info(&pdev->dev, "NTB device registered.\n");
1052 
1053 	return 0;
1054 
1055 err_register:
1056 	ndev_deinit_debugfs(ndev);
1057 	amd_deinit_dev(ndev);
1058 err_init_dev:
1059 	amd_ntb_deinit_pci(ndev);
1060 err_init_pci:
1061 	kfree(ndev);
1062 err_ndev:
1063 	return rc;
1064 }
1065 
1066 static void amd_ntb_pci_remove(struct pci_dev *pdev)
1067 {
1068 	struct amd_ntb_dev *ndev = pci_get_drvdata(pdev);
1069 
1070 	ntb_unregister_device(&ndev->ntb);
1071 	ndev_deinit_debugfs(ndev);
1072 	amd_deinit_side_info(ndev);
1073 	amd_deinit_dev(ndev);
1074 	amd_ntb_deinit_pci(ndev);
1075 	kfree(ndev);
1076 }
1077 
1078 static const struct file_operations amd_ntb_debugfs_info = {
1079 	.owner = THIS_MODULE,
1080 	.open = simple_open,
1081 	.read = ndev_debugfs_read,
1082 };
1083 
1084 static const struct pci_device_id amd_ntb_pci_tbl[] = {
1085 	{PCI_VDEVICE(AMD, PCI_DEVICE_ID_AMD_NTB)},
1086 	{0}
1087 };
1088 MODULE_DEVICE_TABLE(pci, amd_ntb_pci_tbl);
1089 
1090 static struct pci_driver amd_ntb_pci_driver = {
1091 	.name		= KBUILD_MODNAME,
1092 	.id_table	= amd_ntb_pci_tbl,
1093 	.probe		= amd_ntb_pci_probe,
1094 	.remove		= amd_ntb_pci_remove,
1095 };
1096 
1097 static int __init amd_ntb_pci_driver_init(void)
1098 {
1099 	pr_info("%s %s\n", NTB_DESC, NTB_VER);
1100 
1101 	if (debugfs_initialized())
1102 		debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
1103 
1104 	return pci_register_driver(&amd_ntb_pci_driver);
1105 }
1106 module_init(amd_ntb_pci_driver_init);
1107 
1108 static void __exit amd_ntb_pci_driver_exit(void)
1109 {
1110 	pci_unregister_driver(&amd_ntb_pci_driver);
1111 	debugfs_remove_recursive(debugfs_dir);
1112 }
1113 module_exit(amd_ntb_pci_driver_exit);
1114