Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * PCI Peer 2 Peer DMA support.
4 *
5 * Copyright (c) 2016-2018, Logan Gunthorpe
6 * Copyright (c) 2016-2017, Microsemi Corporation
7 * Copyright (c) 2017, Christoph Hellwig
8 * Copyright (c) 2018, Eideticom Inc.
9 */
10
11#define pr_fmt(fmt) "pci-p2pdma: " fmt
12#include <linux/ctype.h>
13#include <linux/pci-p2pdma.h>
14#include <linux/module.h>
15#include <linux/slab.h>
16#include <linux/genalloc.h>
17#include <linux/memremap.h>
18#include <linux/percpu-refcount.h>
19#include <linux/random.h>
20#include <linux/seq_buf.h>
21#include <linux/xarray.h>
22
23enum pci_p2pdma_map_type {
24 PCI_P2PDMA_MAP_UNKNOWN = 0,
25 PCI_P2PDMA_MAP_NOT_SUPPORTED,
26 PCI_P2PDMA_MAP_BUS_ADDR,
27 PCI_P2PDMA_MAP_THRU_HOST_BRIDGE,
28};
29
30struct pci_p2pdma {
31 struct gen_pool *pool;
32 bool p2pmem_published;
33 struct xarray map_types;
34};
35
36struct pci_p2pdma_pagemap {
37 struct dev_pagemap pgmap;
38 struct pci_dev *provider;
39 u64 bus_offset;
40};
41
42static struct pci_p2pdma_pagemap *to_p2p_pgmap(struct dev_pagemap *pgmap)
43{
44 return container_of(pgmap, struct pci_p2pdma_pagemap, pgmap);
45}
46
47static ssize_t size_show(struct device *dev, struct device_attribute *attr,
48 char *buf)
49{
50 struct pci_dev *pdev = to_pci_dev(dev);
51 size_t size = 0;
52
53 if (pdev->p2pdma->pool)
54 size = gen_pool_size(pdev->p2pdma->pool);
55
56 return snprintf(buf, PAGE_SIZE, "%zd\n", size);
57}
58static DEVICE_ATTR_RO(size);
59
60static ssize_t available_show(struct device *dev, struct device_attribute *attr,
61 char *buf)
62{
63 struct pci_dev *pdev = to_pci_dev(dev);
64 size_t avail = 0;
65
66 if (pdev->p2pdma->pool)
67 avail = gen_pool_avail(pdev->p2pdma->pool);
68
69 return snprintf(buf, PAGE_SIZE, "%zd\n", avail);
70}
71static DEVICE_ATTR_RO(available);
72
73static ssize_t published_show(struct device *dev, struct device_attribute *attr,
74 char *buf)
75{
76 struct pci_dev *pdev = to_pci_dev(dev);
77
78 return snprintf(buf, PAGE_SIZE, "%d\n",
79 pdev->p2pdma->p2pmem_published);
80}
81static DEVICE_ATTR_RO(published);
82
83static struct attribute *p2pmem_attrs[] = {
84 &dev_attr_size.attr,
85 &dev_attr_available.attr,
86 &dev_attr_published.attr,
87 NULL,
88};
89
90static const struct attribute_group p2pmem_group = {
91 .attrs = p2pmem_attrs,
92 .name = "p2pmem",
93};
94
95static void pci_p2pdma_release(void *data)
96{
97 struct pci_dev *pdev = data;
98 struct pci_p2pdma *p2pdma = pdev->p2pdma;
99
100 if (!p2pdma)
101 return;
102
103 /* Flush and disable pci_alloc_p2p_mem() */
104 pdev->p2pdma = NULL;
105 synchronize_rcu();
106
107 gen_pool_destroy(p2pdma->pool);
108 sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group);
109 xa_destroy(&p2pdma->map_types);
110}
111
112static int pci_p2pdma_setup(struct pci_dev *pdev)
113{
114 int error = -ENOMEM;
115 struct pci_p2pdma *p2p;
116
117 p2p = devm_kzalloc(&pdev->dev, sizeof(*p2p), GFP_KERNEL);
118 if (!p2p)
119 return -ENOMEM;
120
121 xa_init(&p2p->map_types);
122
123 p2p->pool = gen_pool_create(PAGE_SHIFT, dev_to_node(&pdev->dev));
124 if (!p2p->pool)
125 goto out;
126
127 error = devm_add_action_or_reset(&pdev->dev, pci_p2pdma_release, pdev);
128 if (error)
129 goto out_pool_destroy;
130
131 pdev->p2pdma = p2p;
132
133 error = sysfs_create_group(&pdev->dev.kobj, &p2pmem_group);
134 if (error)
135 goto out_pool_destroy;
136
137 return 0;
138
139out_pool_destroy:
140 pdev->p2pdma = NULL;
141 gen_pool_destroy(p2p->pool);
142out:
143 devm_kfree(&pdev->dev, p2p);
144 return error;
145}
146
147/**
148 * pci_p2pdma_add_resource - add memory for use as p2p memory
149 * @pdev: the device to add the memory to
150 * @bar: PCI BAR to add
151 * @size: size of the memory to add, may be zero to use the whole BAR
152 * @offset: offset into the PCI BAR
153 *
154 * The memory will be given ZONE_DEVICE struct pages so that it may
155 * be used with any DMA request.
156 */
157int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size,
158 u64 offset)
159{
160 struct pci_p2pdma_pagemap *p2p_pgmap;
161 struct dev_pagemap *pgmap;
162 void *addr;
163 int error;
164
165 if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM))
166 return -EINVAL;
167
168 if (offset >= pci_resource_len(pdev, bar))
169 return -EINVAL;
170
171 if (!size)
172 size = pci_resource_len(pdev, bar) - offset;
173
174 if (size + offset > pci_resource_len(pdev, bar))
175 return -EINVAL;
176
177 if (!pdev->p2pdma) {
178 error = pci_p2pdma_setup(pdev);
179 if (error)
180 return error;
181 }
182
183 p2p_pgmap = devm_kzalloc(&pdev->dev, sizeof(*p2p_pgmap), GFP_KERNEL);
184 if (!p2p_pgmap)
185 return -ENOMEM;
186
187 pgmap = &p2p_pgmap->pgmap;
188 pgmap->res.start = pci_resource_start(pdev, bar) + offset;
189 pgmap->res.end = pgmap->res.start + size - 1;
190 pgmap->res.flags = pci_resource_flags(pdev, bar);
191 pgmap->type = MEMORY_DEVICE_PCI_P2PDMA;
192
193 p2p_pgmap->provider = pdev;
194 p2p_pgmap->bus_offset = pci_bus_address(pdev, bar) -
195 pci_resource_start(pdev, bar);
196
197 addr = devm_memremap_pages(&pdev->dev, pgmap);
198 if (IS_ERR(addr)) {
199 error = PTR_ERR(addr);
200 goto pgmap_free;
201 }
202
203 error = gen_pool_add_owner(pdev->p2pdma->pool, (unsigned long)addr,
204 pci_bus_address(pdev, bar) + offset,
205 resource_size(&pgmap->res), dev_to_node(&pdev->dev),
206 pgmap->ref);
207 if (error)
208 goto pages_free;
209
210 pci_info(pdev, "added peer-to-peer DMA memory %pR\n",
211 &pgmap->res);
212
213 return 0;
214
215pages_free:
216 devm_memunmap_pages(&pdev->dev, pgmap);
217pgmap_free:
218 devm_kfree(&pdev->dev, pgmap);
219 return error;
220}
221EXPORT_SYMBOL_GPL(pci_p2pdma_add_resource);
222
223/*
224 * Note this function returns the parent PCI device with a
225 * reference taken. It is the caller's responsibility to drop
226 * the reference.
227 */
228static struct pci_dev *find_parent_pci_dev(struct device *dev)
229{
230 struct device *parent;
231
232 dev = get_device(dev);
233
234 while (dev) {
235 if (dev_is_pci(dev))
236 return to_pci_dev(dev);
237
238 parent = get_device(dev->parent);
239 put_device(dev);
240 dev = parent;
241 }
242
243 return NULL;
244}
245
246/*
247 * Check if a PCI bridge has its ACS redirection bits set to redirect P2P
248 * TLPs upstream via ACS. Returns 1 if the packets will be redirected
249 * upstream, 0 otherwise.
250 */
251static int pci_bridge_has_acs_redir(struct pci_dev *pdev)
252{
253 int pos;
254 u16 ctrl;
255
256 pos = pdev->acs_cap;
257 if (!pos)
258 return 0;
259
260 pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl);
261
262 if (ctrl & (PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC))
263 return 1;
264
265 return 0;
266}
267
268static void seq_buf_print_bus_devfn(struct seq_buf *buf, struct pci_dev *pdev)
269{
270 if (!buf)
271 return;
272
273 seq_buf_printf(buf, "%s;", pci_name(pdev));
274}
275
276static bool cpu_supports_p2pdma(void)
277{
278#ifdef CONFIG_X86
279 struct cpuinfo_x86 *c = &cpu_data(0);
280
281 /* Any AMD CPU whose family ID is Zen or newer supports p2pdma */
282 if (c->x86_vendor == X86_VENDOR_AMD && c->x86 >= 0x17)
283 return true;
284#endif
285
286 return false;
287}
288
289static const struct pci_p2pdma_whitelist_entry {
290 unsigned short vendor;
291 unsigned short device;
292 enum {
293 REQ_SAME_HOST_BRIDGE = 1 << 0,
294 } flags;
295} pci_p2pdma_whitelist[] = {
296 /* Intel Xeon E5/Core i7 */
297 {PCI_VENDOR_ID_INTEL, 0x3c00, REQ_SAME_HOST_BRIDGE},
298 {PCI_VENDOR_ID_INTEL, 0x3c01, REQ_SAME_HOST_BRIDGE},
299 /* Intel Xeon E7 v3/Xeon E5 v3/Core i7 */
300 {PCI_VENDOR_ID_INTEL, 0x2f00, REQ_SAME_HOST_BRIDGE},
301 {PCI_VENDOR_ID_INTEL, 0x2f01, REQ_SAME_HOST_BRIDGE},
302 /* Intel SkyLake-E */
303 {PCI_VENDOR_ID_INTEL, 0x2030, 0},
304 {PCI_VENDOR_ID_INTEL, 0x2031, 0},
305 {PCI_VENDOR_ID_INTEL, 0x2032, 0},
306 {PCI_VENDOR_ID_INTEL, 0x2033, 0},
307 {PCI_VENDOR_ID_INTEL, 0x2020, 0},
308 {}
309};
310
311static bool __host_bridge_whitelist(struct pci_host_bridge *host,
312 bool same_host_bridge)
313{
314 struct pci_dev *root = pci_get_slot(host->bus, PCI_DEVFN(0, 0));
315 const struct pci_p2pdma_whitelist_entry *entry;
316 unsigned short vendor, device;
317
318 if (!root)
319 return false;
320
321 vendor = root->vendor;
322 device = root->device;
323 pci_dev_put(root);
324
325 for (entry = pci_p2pdma_whitelist; entry->vendor; entry++) {
326 if (vendor != entry->vendor || device != entry->device)
327 continue;
328 if (entry->flags & REQ_SAME_HOST_BRIDGE && !same_host_bridge)
329 return false;
330
331 return true;
332 }
333
334 return false;
335}
336
337/*
338 * If we can't find a common upstream bridge take a look at the root
339 * complex and compare it to a whitelist of known good hardware.
340 */
341static bool host_bridge_whitelist(struct pci_dev *a, struct pci_dev *b)
342{
343 struct pci_host_bridge *host_a = pci_find_host_bridge(a->bus);
344 struct pci_host_bridge *host_b = pci_find_host_bridge(b->bus);
345
346 if (host_a == host_b)
347 return __host_bridge_whitelist(host_a, true);
348
349 if (__host_bridge_whitelist(host_a, false) &&
350 __host_bridge_whitelist(host_b, false))
351 return true;
352
353 return false;
354}
355
356static enum pci_p2pdma_map_type
357__upstream_bridge_distance(struct pci_dev *provider, struct pci_dev *client,
358 int *dist, bool *acs_redirects, struct seq_buf *acs_list)
359{
360 struct pci_dev *a = provider, *b = client, *bb;
361 int dist_a = 0;
362 int dist_b = 0;
363 int acs_cnt = 0;
364
365 if (acs_redirects)
366 *acs_redirects = false;
367
368 /*
369 * Note, we don't need to take references to devices returned by
370 * pci_upstream_bridge() seeing we hold a reference to a child
371 * device which will already hold a reference to the upstream bridge.
372 */
373
374 while (a) {
375 dist_b = 0;
376
377 if (pci_bridge_has_acs_redir(a)) {
378 seq_buf_print_bus_devfn(acs_list, a);
379 acs_cnt++;
380 }
381
382 bb = b;
383
384 while (bb) {
385 if (a == bb)
386 goto check_b_path_acs;
387
388 bb = pci_upstream_bridge(bb);
389 dist_b++;
390 }
391
392 a = pci_upstream_bridge(a);
393 dist_a++;
394 }
395
396 if (dist)
397 *dist = dist_a + dist_b;
398
399 return PCI_P2PDMA_MAP_THRU_HOST_BRIDGE;
400
401check_b_path_acs:
402 bb = b;
403
404 while (bb) {
405 if (a == bb)
406 break;
407
408 if (pci_bridge_has_acs_redir(bb)) {
409 seq_buf_print_bus_devfn(acs_list, bb);
410 acs_cnt++;
411 }
412
413 bb = pci_upstream_bridge(bb);
414 }
415
416 if (dist)
417 *dist = dist_a + dist_b;
418
419 if (acs_cnt) {
420 if (acs_redirects)
421 *acs_redirects = true;
422
423 return PCI_P2PDMA_MAP_THRU_HOST_BRIDGE;
424 }
425
426 return PCI_P2PDMA_MAP_BUS_ADDR;
427}
428
429static unsigned long map_types_idx(struct pci_dev *client)
430{
431 return (pci_domain_nr(client->bus) << 16) |
432 (client->bus->number << 8) | client->devfn;
433}
434
435/*
436 * Find the distance through the nearest common upstream bridge between
437 * two PCI devices.
438 *
439 * If the two devices are the same device then 0 will be returned.
440 *
441 * If there are two virtual functions of the same device behind the same
442 * bridge port then 2 will be returned (one step down to the PCIe switch,
443 * then one step back to the same device).
444 *
445 * In the case where two devices are connected to the same PCIe switch, the
446 * value 4 will be returned. This corresponds to the following PCI tree:
447 *
448 * -+ Root Port
449 * \+ Switch Upstream Port
450 * +-+ Switch Downstream Port
451 * + \- Device A
452 * \-+ Switch Downstream Port
453 * \- Device B
454 *
455 * The distance is 4 because we traverse from Device A through the downstream
456 * port of the switch, to the common upstream port, back up to the second
457 * downstream port and then to Device B.
458 *
459 * Any two devices that cannot communicate using p2pdma will return
460 * PCI_P2PDMA_MAP_NOT_SUPPORTED.
461 *
462 * Any two devices that have a data path that goes through the host bridge
463 * will consult a whitelist. If the host bridges are on the whitelist,
464 * this function will return PCI_P2PDMA_MAP_THRU_HOST_BRIDGE.
465 *
466 * If either bridge is not on the whitelist this function returns
467 * PCI_P2PDMA_MAP_NOT_SUPPORTED.
468 *
469 * If a bridge which has any ACS redirection bits set is in the path,
470 * acs_redirects will be set to true. In this case, a list of all infringing
471 * bridge addresses will be populated in acs_list (assuming it's non-null)
472 * for printk purposes.
473 */
474static enum pci_p2pdma_map_type
475upstream_bridge_distance(struct pci_dev *provider, struct pci_dev *client,
476 int *dist, bool *acs_redirects, struct seq_buf *acs_list)
477{
478 enum pci_p2pdma_map_type map_type;
479
480 map_type = __upstream_bridge_distance(provider, client, dist,
481 acs_redirects, acs_list);
482
483 if (map_type == PCI_P2PDMA_MAP_THRU_HOST_BRIDGE) {
484 if (!cpu_supports_p2pdma() &&
485 !host_bridge_whitelist(provider, client))
486 map_type = PCI_P2PDMA_MAP_NOT_SUPPORTED;
487 }
488
489 if (provider->p2pdma)
490 xa_store(&provider->p2pdma->map_types, map_types_idx(client),
491 xa_mk_value(map_type), GFP_KERNEL);
492
493 return map_type;
494}
495
496static enum pci_p2pdma_map_type
497upstream_bridge_distance_warn(struct pci_dev *provider, struct pci_dev *client,
498 int *dist)
499{
500 struct seq_buf acs_list;
501 bool acs_redirects;
502 int ret;
503
504 seq_buf_init(&acs_list, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE);
505 if (!acs_list.buffer)
506 return -ENOMEM;
507
508 ret = upstream_bridge_distance(provider, client, dist, &acs_redirects,
509 &acs_list);
510 if (acs_redirects) {
511 pci_warn(client, "ACS redirect is set between the client and provider (%s)\n",
512 pci_name(provider));
513 /* Drop final semicolon */
514 acs_list.buffer[acs_list.len-1] = 0;
515 pci_warn(client, "to disable ACS redirect for this path, add the kernel parameter: pci=disable_acs_redir=%s\n",
516 acs_list.buffer);
517 }
518
519 if (ret == PCI_P2PDMA_MAP_NOT_SUPPORTED) {
520 pci_warn(client, "cannot be used for peer-to-peer DMA as the client and provider (%s) do not share an upstream bridge or whitelisted host bridge\n",
521 pci_name(provider));
522 }
523
524 kfree(acs_list.buffer);
525
526 return ret;
527}
528
529/**
530 * pci_p2pdma_distance_many - Determine the cumulative distance between
531 * a p2pdma provider and the clients in use.
532 * @provider: p2pdma provider to check against the client list
533 * @clients: array of devices to check (NULL-terminated)
534 * @num_clients: number of clients in the array
535 * @verbose: if true, print warnings for devices when we return -1
536 *
537 * Returns -1 if any of the clients are not compatible, otherwise returns a
538 * positive number where a lower number is the preferable choice. (If there's
539 * one client that's the same as the provider it will return 0, which is best
540 * choice).
541 *
542 * "compatible" means the provider and the clients are either all behind
543 * the same PCI root port or the host bridges connected to each of the devices
544 * are listed in the 'pci_p2pdma_whitelist'.
545 */
546int pci_p2pdma_distance_many(struct pci_dev *provider, struct device **clients,
547 int num_clients, bool verbose)
548{
549 bool not_supported = false;
550 struct pci_dev *pci_client;
551 int total_dist = 0;
552 int distance;
553 int i, ret;
554
555 if (num_clients == 0)
556 return -1;
557
558 for (i = 0; i < num_clients; i++) {
559#ifdef CONFIG_DMA_VIRT_OPS
560 if (clients[i]->dma_ops == &dma_virt_ops) {
561 if (verbose)
562 dev_warn(clients[i],
563 "cannot be used for peer-to-peer DMA because the driver makes use of dma_virt_ops\n");
564 return -1;
565 }
566#endif
567
568 pci_client = find_parent_pci_dev(clients[i]);
569 if (!pci_client) {
570 if (verbose)
571 dev_warn(clients[i],
572 "cannot be used for peer-to-peer DMA as it is not a PCI device\n");
573 return -1;
574 }
575
576 if (verbose)
577 ret = upstream_bridge_distance_warn(provider,
578 pci_client, &distance);
579 else
580 ret = upstream_bridge_distance(provider, pci_client,
581 &distance, NULL, NULL);
582
583 pci_dev_put(pci_client);
584
585 if (ret == PCI_P2PDMA_MAP_NOT_SUPPORTED)
586 not_supported = true;
587
588 if (not_supported && !verbose)
589 break;
590
591 total_dist += distance;
592 }
593
594 if (not_supported)
595 return -1;
596
597 return total_dist;
598}
599EXPORT_SYMBOL_GPL(pci_p2pdma_distance_many);
600
601/**
602 * pci_has_p2pmem - check if a given PCI device has published any p2pmem
603 * @pdev: PCI device to check
604 */
605bool pci_has_p2pmem(struct pci_dev *pdev)
606{
607 return pdev->p2pdma && pdev->p2pdma->p2pmem_published;
608}
609EXPORT_SYMBOL_GPL(pci_has_p2pmem);
610
611/**
612 * pci_p2pmem_find - find a peer-to-peer DMA memory device compatible with
613 * the specified list of clients and shortest distance (as determined
614 * by pci_p2pmem_dma())
615 * @clients: array of devices to check (NULL-terminated)
616 * @num_clients: number of client devices in the list
617 *
618 * If multiple devices are behind the same switch, the one "closest" to the
619 * client devices in use will be chosen first. (So if one of the providers is
620 * the same as one of the clients, that provider will be used ahead of any
621 * other providers that are unrelated). If multiple providers are an equal
622 * distance away, one will be chosen at random.
623 *
624 * Returns a pointer to the PCI device with a reference taken (use pci_dev_put
625 * to return the reference) or NULL if no compatible device is found. The
626 * found provider will also be assigned to the client list.
627 */
628struct pci_dev *pci_p2pmem_find_many(struct device **clients, int num_clients)
629{
630 struct pci_dev *pdev = NULL;
631 int distance;
632 int closest_distance = INT_MAX;
633 struct pci_dev **closest_pdevs;
634 int dev_cnt = 0;
635 const int max_devs = PAGE_SIZE / sizeof(*closest_pdevs);
636 int i;
637
638 closest_pdevs = kmalloc(PAGE_SIZE, GFP_KERNEL);
639 if (!closest_pdevs)
640 return NULL;
641
642 while ((pdev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, pdev))) {
643 if (!pci_has_p2pmem(pdev))
644 continue;
645
646 distance = pci_p2pdma_distance_many(pdev, clients,
647 num_clients, false);
648 if (distance < 0 || distance > closest_distance)
649 continue;
650
651 if (distance == closest_distance && dev_cnt >= max_devs)
652 continue;
653
654 if (distance < closest_distance) {
655 for (i = 0; i < dev_cnt; i++)
656 pci_dev_put(closest_pdevs[i]);
657
658 dev_cnt = 0;
659 closest_distance = distance;
660 }
661
662 closest_pdevs[dev_cnt++] = pci_dev_get(pdev);
663 }
664
665 if (dev_cnt)
666 pdev = pci_dev_get(closest_pdevs[prandom_u32_max(dev_cnt)]);
667
668 for (i = 0; i < dev_cnt; i++)
669 pci_dev_put(closest_pdevs[i]);
670
671 kfree(closest_pdevs);
672 return pdev;
673}
674EXPORT_SYMBOL_GPL(pci_p2pmem_find_many);
675
676/**
677 * pci_alloc_p2p_mem - allocate peer-to-peer DMA memory
678 * @pdev: the device to allocate memory from
679 * @size: number of bytes to allocate
680 *
681 * Returns the allocated memory or NULL on error.
682 */
683void *pci_alloc_p2pmem(struct pci_dev *pdev, size_t size)
684{
685 void *ret = NULL;
686 struct percpu_ref *ref;
687
688 /*
689 * Pairs with synchronize_rcu() in pci_p2pdma_release() to
690 * ensure pdev->p2pdma is non-NULL for the duration of the
691 * read-lock.
692 */
693 rcu_read_lock();
694 if (unlikely(!pdev->p2pdma))
695 goto out;
696
697 ret = (void *)gen_pool_alloc_owner(pdev->p2pdma->pool, size,
698 (void **) &ref);
699 if (!ret)
700 goto out;
701
702 if (unlikely(!percpu_ref_tryget_live(ref))) {
703 gen_pool_free(pdev->p2pdma->pool, (unsigned long) ret, size);
704 ret = NULL;
705 goto out;
706 }
707out:
708 rcu_read_unlock();
709 return ret;
710}
711EXPORT_SYMBOL_GPL(pci_alloc_p2pmem);
712
713/**
714 * pci_free_p2pmem - free peer-to-peer DMA memory
715 * @pdev: the device the memory was allocated from
716 * @addr: address of the memory that was allocated
717 * @size: number of bytes that were allocated
718 */
719void pci_free_p2pmem(struct pci_dev *pdev, void *addr, size_t size)
720{
721 struct percpu_ref *ref;
722
723 gen_pool_free_owner(pdev->p2pdma->pool, (uintptr_t)addr, size,
724 (void **) &ref);
725 percpu_ref_put(ref);
726}
727EXPORT_SYMBOL_GPL(pci_free_p2pmem);
728
729/**
730 * pci_virt_to_bus - return the PCI bus address for a given virtual
731 * address obtained with pci_alloc_p2pmem()
732 * @pdev: the device the memory was allocated from
733 * @addr: address of the memory that was allocated
734 */
735pci_bus_addr_t pci_p2pmem_virt_to_bus(struct pci_dev *pdev, void *addr)
736{
737 if (!addr)
738 return 0;
739 if (!pdev->p2pdma)
740 return 0;
741
742 /*
743 * Note: when we added the memory to the pool we used the PCI
744 * bus address as the physical address. So gen_pool_virt_to_phys()
745 * actually returns the bus address despite the misleading name.
746 */
747 return gen_pool_virt_to_phys(pdev->p2pdma->pool, (unsigned long)addr);
748}
749EXPORT_SYMBOL_GPL(pci_p2pmem_virt_to_bus);
750
751/**
752 * pci_p2pmem_alloc_sgl - allocate peer-to-peer DMA memory in a scatterlist
753 * @pdev: the device to allocate memory from
754 * @nents: the number of SG entries in the list
755 * @length: number of bytes to allocate
756 *
757 * Return: %NULL on error or &struct scatterlist pointer and @nents on success
758 */
759struct scatterlist *pci_p2pmem_alloc_sgl(struct pci_dev *pdev,
760 unsigned int *nents, u32 length)
761{
762 struct scatterlist *sg;
763 void *addr;
764
765 sg = kzalloc(sizeof(*sg), GFP_KERNEL);
766 if (!sg)
767 return NULL;
768
769 sg_init_table(sg, 1);
770
771 addr = pci_alloc_p2pmem(pdev, length);
772 if (!addr)
773 goto out_free_sg;
774
775 sg_set_buf(sg, addr, length);
776 *nents = 1;
777 return sg;
778
779out_free_sg:
780 kfree(sg);
781 return NULL;
782}
783EXPORT_SYMBOL_GPL(pci_p2pmem_alloc_sgl);
784
785/**
786 * pci_p2pmem_free_sgl - free a scatterlist allocated by pci_p2pmem_alloc_sgl()
787 * @pdev: the device to allocate memory from
788 * @sgl: the allocated scatterlist
789 */
790void pci_p2pmem_free_sgl(struct pci_dev *pdev, struct scatterlist *sgl)
791{
792 struct scatterlist *sg;
793 int count;
794
795 for_each_sg(sgl, sg, INT_MAX, count) {
796 if (!sg)
797 break;
798
799 pci_free_p2pmem(pdev, sg_virt(sg), sg->length);
800 }
801 kfree(sgl);
802}
803EXPORT_SYMBOL_GPL(pci_p2pmem_free_sgl);
804
805/**
806 * pci_p2pmem_publish - publish the peer-to-peer DMA memory for use by
807 * other devices with pci_p2pmem_find()
808 * @pdev: the device with peer-to-peer DMA memory to publish
809 * @publish: set to true to publish the memory, false to unpublish it
810 *
811 * Published memory can be used by other PCI device drivers for
812 * peer-2-peer DMA operations. Non-published memory is reserved for
813 * exclusive use of the device driver that registers the peer-to-peer
814 * memory.
815 */
816void pci_p2pmem_publish(struct pci_dev *pdev, bool publish)
817{
818 if (pdev->p2pdma)
819 pdev->p2pdma->p2pmem_published = publish;
820}
821EXPORT_SYMBOL_GPL(pci_p2pmem_publish);
822
823static enum pci_p2pdma_map_type pci_p2pdma_map_type(struct pci_dev *provider,
824 struct pci_dev *client)
825{
826 if (!provider->p2pdma)
827 return PCI_P2PDMA_MAP_NOT_SUPPORTED;
828
829 return xa_to_value(xa_load(&provider->p2pdma->map_types,
830 map_types_idx(client)));
831}
832
833static int __pci_p2pdma_map_sg(struct pci_p2pdma_pagemap *p2p_pgmap,
834 struct device *dev, struct scatterlist *sg, int nents)
835{
836 struct scatterlist *s;
837 phys_addr_t paddr;
838 int i;
839
840 /*
841 * p2pdma mappings are not compatible with devices that use
842 * dma_virt_ops. If the upper layers do the right thing
843 * this should never happen because it will be prevented
844 * by the check in pci_p2pdma_distance_many()
845 */
846#ifdef CONFIG_DMA_VIRT_OPS
847 if (WARN_ON_ONCE(dev->dma_ops == &dma_virt_ops))
848 return 0;
849#endif
850
851 for_each_sg(sg, s, nents, i) {
852 paddr = sg_phys(s);
853
854 s->dma_address = paddr - p2p_pgmap->bus_offset;
855 sg_dma_len(s) = s->length;
856 }
857
858 return nents;
859}
860
861/**
862 * pci_p2pdma_map_sg - map a PCI peer-to-peer scatterlist for DMA
863 * @dev: device doing the DMA request
864 * @sg: scatter list to map
865 * @nents: elements in the scatterlist
866 * @dir: DMA direction
867 * @attrs: DMA attributes passed to dma_map_sg() (if called)
868 *
869 * Scatterlists mapped with this function should be unmapped using
870 * pci_p2pdma_unmap_sg_attrs().
871 *
872 * Returns the number of SG entries mapped or 0 on error.
873 */
874int pci_p2pdma_map_sg_attrs(struct device *dev, struct scatterlist *sg,
875 int nents, enum dma_data_direction dir, unsigned long attrs)
876{
877 struct pci_p2pdma_pagemap *p2p_pgmap =
878 to_p2p_pgmap(sg_page(sg)->pgmap);
879 struct pci_dev *client;
880
881 if (WARN_ON_ONCE(!dev_is_pci(dev)))
882 return 0;
883
884 client = to_pci_dev(dev);
885
886 switch (pci_p2pdma_map_type(p2p_pgmap->provider, client)) {
887 case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE:
888 return dma_map_sg_attrs(dev, sg, nents, dir, attrs);
889 case PCI_P2PDMA_MAP_BUS_ADDR:
890 return __pci_p2pdma_map_sg(p2p_pgmap, dev, sg, nents);
891 default:
892 WARN_ON_ONCE(1);
893 return 0;
894 }
895}
896EXPORT_SYMBOL_GPL(pci_p2pdma_map_sg_attrs);
897
898/**
899 * pci_p2pdma_unmap_sg - unmap a PCI peer-to-peer scatterlist that was
900 * mapped with pci_p2pdma_map_sg()
901 * @dev: device doing the DMA request
902 * @sg: scatter list to map
903 * @nents: number of elements returned by pci_p2pdma_map_sg()
904 * @dir: DMA direction
905 * @attrs: DMA attributes passed to dma_unmap_sg() (if called)
906 */
907void pci_p2pdma_unmap_sg_attrs(struct device *dev, struct scatterlist *sg,
908 int nents, enum dma_data_direction dir, unsigned long attrs)
909{
910 struct pci_p2pdma_pagemap *p2p_pgmap =
911 to_p2p_pgmap(sg_page(sg)->pgmap);
912 enum pci_p2pdma_map_type map_type;
913 struct pci_dev *client;
914
915 if (WARN_ON_ONCE(!dev_is_pci(dev)))
916 return;
917
918 client = to_pci_dev(dev);
919
920 map_type = pci_p2pdma_map_type(p2p_pgmap->provider, client);
921
922 if (map_type == PCI_P2PDMA_MAP_THRU_HOST_BRIDGE)
923 dma_unmap_sg_attrs(dev, sg, nents, dir, attrs);
924}
925EXPORT_SYMBOL_GPL(pci_p2pdma_unmap_sg_attrs);
926
927/**
928 * pci_p2pdma_enable_store - parse a configfs/sysfs attribute store
929 * to enable p2pdma
930 * @page: contents of the value to be stored
931 * @p2p_dev: returns the PCI device that was selected to be used
932 * (if one was specified in the stored value)
933 * @use_p2pdma: returns whether to enable p2pdma or not
934 *
935 * Parses an attribute value to decide whether to enable p2pdma.
936 * The value can select a PCI device (using its full BDF device
937 * name) or a boolean (in any format strtobool() accepts). A false
938 * value disables p2pdma, a true value expects the caller
939 * to automatically find a compatible device and specifying a PCI device
940 * expects the caller to use the specific provider.
941 *
942 * pci_p2pdma_enable_show() should be used as the show operation for
943 * the attribute.
944 *
945 * Returns 0 on success
946 */
947int pci_p2pdma_enable_store(const char *page, struct pci_dev **p2p_dev,
948 bool *use_p2pdma)
949{
950 struct device *dev;
951
952 dev = bus_find_device_by_name(&pci_bus_type, NULL, page);
953 if (dev) {
954 *use_p2pdma = true;
955 *p2p_dev = to_pci_dev(dev);
956
957 if (!pci_has_p2pmem(*p2p_dev)) {
958 pci_err(*p2p_dev,
959 "PCI device has no peer-to-peer memory: %s\n",
960 page);
961 pci_dev_put(*p2p_dev);
962 return -ENODEV;
963 }
964
965 return 0;
966 } else if ((page[0] == '0' || page[0] == '1') && !iscntrl(page[1])) {
967 /*
968 * If the user enters a PCI device that doesn't exist
969 * like "0000:01:00.1", we don't want strtobool to think
970 * it's a '0' when it's clearly not what the user wanted.
971 * So we require 0's and 1's to be exactly one character.
972 */
973 } else if (!strtobool(page, use_p2pdma)) {
974 return 0;
975 }
976
977 pr_err("No such PCI device: %.*s\n", (int)strcspn(page, "\n"), page);
978 return -ENODEV;
979}
980EXPORT_SYMBOL_GPL(pci_p2pdma_enable_store);
981
982/**
983 * pci_p2pdma_enable_show - show a configfs/sysfs attribute indicating
984 * whether p2pdma is enabled
985 * @page: contents of the stored value
986 * @p2p_dev: the selected p2p device (NULL if no device is selected)
987 * @use_p2pdma: whether p2pdma has been enabled
988 *
989 * Attributes that use pci_p2pdma_enable_store() should use this function
990 * to show the value of the attribute.
991 *
992 * Returns 0 on success
993 */
994ssize_t pci_p2pdma_enable_show(char *page, struct pci_dev *p2p_dev,
995 bool use_p2pdma)
996{
997 if (!use_p2pdma)
998 return sprintf(page, "0\n");
999
1000 if (!p2p_dev)
1001 return sprintf(page, "1\n");
1002
1003 return sprintf(page, "%s\n", pci_name(p2p_dev));
1004}
1005EXPORT_SYMBOL_GPL(pci_p2pdma_enable_show);