Commit | Line | Data |
---|---|---|
52916982 LG |
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 | ||
2d7bc010 LG |
11 | #define pr_fmt(fmt) "pci-p2pdma: " fmt |
12 | #include <linux/ctype.h> | |
52916982 LG |
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 | ||
22 | struct pci_p2pdma { | |
23 | struct percpu_ref devmap_ref; | |
24 | struct completion devmap_ref_done; | |
25 | struct gen_pool *pool; | |
26 | bool p2pmem_published; | |
27 | }; | |
28 | ||
cbb8ca69 LG |
29 | static ssize_t size_show(struct device *dev, struct device_attribute *attr, |
30 | char *buf) | |
31 | { | |
32 | struct pci_dev *pdev = to_pci_dev(dev); | |
33 | size_t size = 0; | |
34 | ||
35 | if (pdev->p2pdma->pool) | |
36 | size = gen_pool_size(pdev->p2pdma->pool); | |
37 | ||
38 | return snprintf(buf, PAGE_SIZE, "%zd\n", size); | |
39 | } | |
40 | static DEVICE_ATTR_RO(size); | |
41 | ||
42 | static ssize_t available_show(struct device *dev, struct device_attribute *attr, | |
43 | char *buf) | |
44 | { | |
45 | struct pci_dev *pdev = to_pci_dev(dev); | |
46 | size_t avail = 0; | |
47 | ||
48 | if (pdev->p2pdma->pool) | |
49 | avail = gen_pool_avail(pdev->p2pdma->pool); | |
50 | ||
51 | return snprintf(buf, PAGE_SIZE, "%zd\n", avail); | |
52 | } | |
53 | static DEVICE_ATTR_RO(available); | |
54 | ||
55 | static ssize_t published_show(struct device *dev, struct device_attribute *attr, | |
56 | char *buf) | |
57 | { | |
58 | struct pci_dev *pdev = to_pci_dev(dev); | |
59 | ||
60 | return snprintf(buf, PAGE_SIZE, "%d\n", | |
61 | pdev->p2pdma->p2pmem_published); | |
62 | } | |
63 | static DEVICE_ATTR_RO(published); | |
64 | ||
65 | static struct attribute *p2pmem_attrs[] = { | |
66 | &dev_attr_size.attr, | |
67 | &dev_attr_available.attr, | |
68 | &dev_attr_published.attr, | |
69 | NULL, | |
70 | }; | |
71 | ||
72 | static const struct attribute_group p2pmem_group = { | |
73 | .attrs = p2pmem_attrs, | |
74 | .name = "p2pmem", | |
75 | }; | |
76 | ||
52916982 LG |
77 | static void pci_p2pdma_percpu_release(struct percpu_ref *ref) |
78 | { | |
79 | struct pci_p2pdma *p2p = | |
80 | container_of(ref, struct pci_p2pdma, devmap_ref); | |
81 | ||
82 | complete_all(&p2p->devmap_ref_done); | |
83 | } | |
84 | ||
02917e9f | 85 | static void pci_p2pdma_percpu_kill(struct percpu_ref *ref) |
52916982 | 86 | { |
52916982 LG |
87 | /* |
88 | * pci_p2pdma_add_resource() may be called multiple times | |
89 | * by a driver and may register the percpu_kill devm action multiple | |
90 | * times. We only want the first action to actually kill the | |
91 | * percpu_ref. | |
92 | */ | |
93 | if (percpu_ref_is_dying(ref)) | |
94 | return; | |
95 | ||
96 | percpu_ref_kill(ref); | |
97 | } | |
98 | ||
99 | static void pci_p2pdma_release(void *data) | |
100 | { | |
101 | struct pci_dev *pdev = data; | |
102 | ||
103 | if (!pdev->p2pdma) | |
104 | return; | |
105 | ||
106 | wait_for_completion(&pdev->p2pdma->devmap_ref_done); | |
107 | percpu_ref_exit(&pdev->p2pdma->devmap_ref); | |
108 | ||
109 | gen_pool_destroy(pdev->p2pdma->pool); | |
cbb8ca69 | 110 | sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group); |
52916982 LG |
111 | pdev->p2pdma = NULL; |
112 | } | |
113 | ||
114 | static int pci_p2pdma_setup(struct pci_dev *pdev) | |
115 | { | |
116 | int error = -ENOMEM; | |
117 | struct pci_p2pdma *p2p; | |
118 | ||
119 | p2p = devm_kzalloc(&pdev->dev, sizeof(*p2p), GFP_KERNEL); | |
120 | if (!p2p) | |
121 | return -ENOMEM; | |
122 | ||
123 | p2p->pool = gen_pool_create(PAGE_SHIFT, dev_to_node(&pdev->dev)); | |
124 | if (!p2p->pool) | |
125 | goto out; | |
126 | ||
127 | init_completion(&p2p->devmap_ref_done); | |
128 | error = percpu_ref_init(&p2p->devmap_ref, | |
129 | pci_p2pdma_percpu_release, 0, GFP_KERNEL); | |
130 | if (error) | |
131 | goto out_pool_destroy; | |
132 | ||
133 | error = devm_add_action_or_reset(&pdev->dev, pci_p2pdma_release, pdev); | |
134 | if (error) | |
135 | goto out_pool_destroy; | |
136 | ||
137 | pdev->p2pdma = p2p; | |
138 | ||
cbb8ca69 LG |
139 | error = sysfs_create_group(&pdev->dev.kobj, &p2pmem_group); |
140 | if (error) | |
141 | goto out_pool_destroy; | |
142 | ||
52916982 LG |
143 | return 0; |
144 | ||
145 | out_pool_destroy: | |
cbb8ca69 | 146 | pdev->p2pdma = NULL; |
52916982 LG |
147 | gen_pool_destroy(p2p->pool); |
148 | out: | |
149 | devm_kfree(&pdev->dev, p2p); | |
150 | return error; | |
151 | } | |
152 | ||
153 | /** | |
154 | * pci_p2pdma_add_resource - add memory for use as p2p memory | |
155 | * @pdev: the device to add the memory to | |
156 | * @bar: PCI BAR to add | |
157 | * @size: size of the memory to add, may be zero to use the whole BAR | |
158 | * @offset: offset into the PCI BAR | |
159 | * | |
160 | * The memory will be given ZONE_DEVICE struct pages so that it may | |
161 | * be used with any DMA request. | |
162 | */ | |
163 | int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size, | |
164 | u64 offset) | |
165 | { | |
166 | struct dev_pagemap *pgmap; | |
167 | void *addr; | |
168 | int error; | |
169 | ||
170 | if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) | |
171 | return -EINVAL; | |
172 | ||
173 | if (offset >= pci_resource_len(pdev, bar)) | |
174 | return -EINVAL; | |
175 | ||
176 | if (!size) | |
177 | size = pci_resource_len(pdev, bar) - offset; | |
178 | ||
179 | if (size + offset > pci_resource_len(pdev, bar)) | |
180 | return -EINVAL; | |
181 | ||
182 | if (!pdev->p2pdma) { | |
183 | error = pci_p2pdma_setup(pdev); | |
184 | if (error) | |
185 | return error; | |
186 | } | |
187 | ||
188 | pgmap = devm_kzalloc(&pdev->dev, sizeof(*pgmap), GFP_KERNEL); | |
189 | if (!pgmap) | |
190 | return -ENOMEM; | |
191 | ||
192 | pgmap->res.start = pci_resource_start(pdev, bar) + offset; | |
193 | pgmap->res.end = pgmap->res.start + size - 1; | |
194 | pgmap->res.flags = pci_resource_flags(pdev, bar); | |
195 | pgmap->ref = &pdev->p2pdma->devmap_ref; | |
196 | pgmap->type = MEMORY_DEVICE_PCI_P2PDMA; | |
977196b8 LG |
197 | pgmap->pci_p2pdma_bus_offset = pci_bus_address(pdev, bar) - |
198 | pci_resource_start(pdev, bar); | |
02917e9f | 199 | pgmap->kill = pci_p2pdma_percpu_kill; |
52916982 LG |
200 | |
201 | addr = devm_memremap_pages(&pdev->dev, pgmap); | |
202 | if (IS_ERR(addr)) { | |
203 | error = PTR_ERR(addr); | |
204 | goto pgmap_free; | |
205 | } | |
206 | ||
207 | error = gen_pool_add_virt(pdev->p2pdma->pool, (unsigned long)addr, | |
208 | pci_bus_address(pdev, bar) + offset, | |
209 | resource_size(&pgmap->res), dev_to_node(&pdev->dev)); | |
210 | if (error) | |
211 | goto pgmap_free; | |
212 | ||
52916982 LG |
213 | pci_info(pdev, "added peer-to-peer DMA memory %pR\n", |
214 | &pgmap->res); | |
215 | ||
216 | return 0; | |
217 | ||
218 | pgmap_free: | |
219 | devm_kfree(&pdev->dev, pgmap); | |
220 | return error; | |
221 | } | |
222 | EXPORT_SYMBOL_GPL(pci_p2pdma_add_resource); | |
223 | ||
224 | /* | |
225 | * Note this function returns the parent PCI device with a | |
226 | * reference taken. It is the caller's responsibily to drop | |
227 | * the reference. | |
228 | */ | |
229 | static struct pci_dev *find_parent_pci_dev(struct device *dev) | |
230 | { | |
231 | struct device *parent; | |
232 | ||
233 | dev = get_device(dev); | |
234 | ||
235 | while (dev) { | |
236 | if (dev_is_pci(dev)) | |
237 | return to_pci_dev(dev); | |
238 | ||
239 | parent = get_device(dev->parent); | |
240 | put_device(dev); | |
241 | dev = parent; | |
242 | } | |
243 | ||
244 | return NULL; | |
245 | } | |
246 | ||
247 | /* | |
248 | * Check if a PCI bridge has its ACS redirection bits set to redirect P2P | |
249 | * TLPs upstream via ACS. Returns 1 if the packets will be redirected | |
250 | * upstream, 0 otherwise. | |
251 | */ | |
252 | static int pci_bridge_has_acs_redir(struct pci_dev *pdev) | |
253 | { | |
254 | int pos; | |
255 | u16 ctrl; | |
256 | ||
257 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS); | |
258 | if (!pos) | |
259 | return 0; | |
260 | ||
261 | pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl); | |
262 | ||
263 | if (ctrl & (PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC)) | |
264 | return 1; | |
265 | ||
266 | return 0; | |
267 | } | |
268 | ||
269 | static void seq_buf_print_bus_devfn(struct seq_buf *buf, struct pci_dev *pdev) | |
270 | { | |
271 | if (!buf) | |
272 | return; | |
273 | ||
274 | seq_buf_printf(buf, "%s;", pci_name(pdev)); | |
275 | } | |
276 | ||
0f97da83 CK |
277 | /* |
278 | * If we can't find a common upstream bridge take a look at the root | |
279 | * complex and compare it to a whitelist of known good hardware. | |
280 | */ | |
281 | static bool root_complex_whitelist(struct pci_dev *dev) | |
282 | { | |
283 | struct pci_host_bridge *host = pci_find_host_bridge(dev->bus); | |
284 | struct pci_dev *root = pci_get_slot(host->bus, PCI_DEVFN(0, 0)); | |
285 | unsigned short vendor, device; | |
286 | ||
287 | if (!root) | |
288 | return false; | |
289 | ||
290 | vendor = root->vendor; | |
291 | device = root->device; | |
292 | pci_dev_put(root); | |
293 | ||
294 | /* AMD ZEN host bridges can do peer to peer */ | |
295 | if (vendor == PCI_VENDOR_ID_AMD && device == 0x1450) | |
296 | return true; | |
297 | ||
298 | return false; | |
299 | } | |
300 | ||
52916982 LG |
301 | /* |
302 | * Find the distance through the nearest common upstream bridge between | |
303 | * two PCI devices. | |
304 | * | |
305 | * If the two devices are the same device then 0 will be returned. | |
306 | * | |
307 | * If there are two virtual functions of the same device behind the same | |
308 | * bridge port then 2 will be returned (one step down to the PCIe switch, | |
309 | * then one step back to the same device). | |
310 | * | |
311 | * In the case where two devices are connected to the same PCIe switch, the | |
312 | * value 4 will be returned. This corresponds to the following PCI tree: | |
313 | * | |
314 | * -+ Root Port | |
315 | * \+ Switch Upstream Port | |
316 | * +-+ Switch Downstream Port | |
317 | * + \- Device A | |
318 | * \-+ Switch Downstream Port | |
319 | * \- Device B | |
320 | * | |
321 | * The distance is 4 because we traverse from Device A through the downstream | |
322 | * port of the switch, to the common upstream port, back up to the second | |
323 | * downstream port and then to Device B. | |
324 | * | |
325 | * Any two devices that don't have a common upstream bridge will return -1. | |
326 | * In this way devices on separate PCIe root ports will be rejected, which | |
327 | * is what we want for peer-to-peer seeing each PCIe root port defines a | |
328 | * separate hierarchy domain and there's no way to determine whether the root | |
329 | * complex supports forwarding between them. | |
330 | * | |
331 | * In the case where two devices are connected to different PCIe switches, | |
332 | * this function will still return a positive distance as long as both | |
333 | * switches eventually have a common upstream bridge. Note this covers | |
334 | * the case of using multiple PCIe switches to achieve a desired level of | |
335 | * fan-out from a root port. The exact distance will be a function of the | |
336 | * number of switches between Device A and Device B. | |
337 | * | |
338 | * If a bridge which has any ACS redirection bits set is in the path | |
339 | * then this functions will return -2. This is so we reject any | |
340 | * cases where the TLPs are forwarded up into the root complex. | |
341 | * In this case, a list of all infringing bridge addresses will be | |
342 | * populated in acs_list (assuming it's non-null) for printk purposes. | |
343 | */ | |
0f97da83 CK |
344 | static int upstream_bridge_distance(struct pci_dev *provider, |
345 | struct pci_dev *client, | |
52916982 LG |
346 | struct seq_buf *acs_list) |
347 | { | |
0f97da83 | 348 | struct pci_dev *a = provider, *b = client, *bb; |
52916982 LG |
349 | int dist_a = 0; |
350 | int dist_b = 0; | |
52916982 LG |
351 | int acs_cnt = 0; |
352 | ||
353 | /* | |
354 | * Note, we don't need to take references to devices returned by | |
355 | * pci_upstream_bridge() seeing we hold a reference to a child | |
356 | * device which will already hold a reference to the upstream bridge. | |
357 | */ | |
358 | ||
359 | while (a) { | |
360 | dist_b = 0; | |
361 | ||
362 | if (pci_bridge_has_acs_redir(a)) { | |
363 | seq_buf_print_bus_devfn(acs_list, a); | |
364 | acs_cnt++; | |
365 | } | |
366 | ||
367 | bb = b; | |
368 | ||
369 | while (bb) { | |
370 | if (a == bb) | |
371 | goto check_b_path_acs; | |
372 | ||
373 | bb = pci_upstream_bridge(bb); | |
374 | dist_b++; | |
375 | } | |
376 | ||
377 | a = pci_upstream_bridge(a); | |
378 | dist_a++; | |
379 | } | |
380 | ||
0f97da83 CK |
381 | /* |
382 | * Allow the connection if both devices are on a whitelisted root | |
383 | * complex, but add an arbitary large value to the distance. | |
384 | */ | |
385 | if (root_complex_whitelist(provider) && | |
386 | root_complex_whitelist(client)) | |
387 | return 0x1000 + dist_a + dist_b; | |
388 | ||
52916982 LG |
389 | return -1; |
390 | ||
391 | check_b_path_acs: | |
392 | bb = b; | |
393 | ||
394 | while (bb) { | |
395 | if (a == bb) | |
396 | break; | |
397 | ||
398 | if (pci_bridge_has_acs_redir(bb)) { | |
399 | seq_buf_print_bus_devfn(acs_list, bb); | |
400 | acs_cnt++; | |
401 | } | |
402 | ||
403 | bb = pci_upstream_bridge(bb); | |
404 | } | |
405 | ||
406 | if (acs_cnt) | |
407 | return -2; | |
408 | ||
409 | return dist_a + dist_b; | |
410 | } | |
411 | ||
412 | static int upstream_bridge_distance_warn(struct pci_dev *provider, | |
413 | struct pci_dev *client) | |
414 | { | |
415 | struct seq_buf acs_list; | |
416 | int ret; | |
417 | ||
418 | seq_buf_init(&acs_list, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); | |
419 | if (!acs_list.buffer) | |
420 | return -ENOMEM; | |
421 | ||
422 | ret = upstream_bridge_distance(provider, client, &acs_list); | |
423 | if (ret == -2) { | |
424 | pci_warn(client, "cannot be used for peer-to-peer DMA as ACS redirect is set between the client and provider (%s)\n", | |
425 | pci_name(provider)); | |
426 | /* Drop final semicolon */ | |
427 | acs_list.buffer[acs_list.len-1] = 0; | |
428 | pci_warn(client, "to disable ACS redirect for this path, add the kernel parameter: pci=disable_acs_redir=%s\n", | |
429 | acs_list.buffer); | |
430 | ||
431 | } else if (ret < 0) { | |
432 | pci_warn(client, "cannot be used for peer-to-peer DMA as the client and provider (%s) do not share an upstream bridge\n", | |
433 | pci_name(provider)); | |
434 | } | |
435 | ||
436 | kfree(acs_list.buffer); | |
437 | ||
438 | return ret; | |
439 | } | |
440 | ||
441 | /** | |
442 | * pci_p2pdma_distance_many - Determive the cumulative distance between | |
443 | * a p2pdma provider and the clients in use. | |
444 | * @provider: p2pdma provider to check against the client list | |
445 | * @clients: array of devices to check (NULL-terminated) | |
446 | * @num_clients: number of clients in the array | |
447 | * @verbose: if true, print warnings for devices when we return -1 | |
448 | * | |
449 | * Returns -1 if any of the clients are not compatible (behind the same | |
450 | * root port as the provider), otherwise returns a positive number where | |
fcf9ab35 | 451 | * a lower number is the preferable choice. (If there's one client |
52916982 LG |
452 | * that's the same as the provider it will return 0, which is best choice). |
453 | * | |
454 | * For now, "compatible" means the provider and the clients are all behind | |
455 | * the same PCI root port. This cuts out cases that may work but is safest | |
456 | * for the user. Future work can expand this to white-list root complexes that | |
457 | * can safely forward between each ports. | |
458 | */ | |
459 | int pci_p2pdma_distance_many(struct pci_dev *provider, struct device **clients, | |
460 | int num_clients, bool verbose) | |
461 | { | |
462 | bool not_supported = false; | |
463 | struct pci_dev *pci_client; | |
464 | int distance = 0; | |
465 | int i, ret; | |
466 | ||
467 | if (num_clients == 0) | |
468 | return -1; | |
469 | ||
470 | for (i = 0; i < num_clients; i++) { | |
471 | pci_client = find_parent_pci_dev(clients[i]); | |
472 | if (!pci_client) { | |
473 | if (verbose) | |
474 | dev_warn(clients[i], | |
475 | "cannot be used for peer-to-peer DMA as it is not a PCI device\n"); | |
476 | return -1; | |
477 | } | |
478 | ||
479 | if (verbose) | |
480 | ret = upstream_bridge_distance_warn(provider, | |
481 | pci_client); | |
482 | else | |
483 | ret = upstream_bridge_distance(provider, pci_client, | |
484 | NULL); | |
485 | ||
486 | pci_dev_put(pci_client); | |
487 | ||
488 | if (ret < 0) | |
489 | not_supported = true; | |
490 | ||
491 | if (not_supported && !verbose) | |
492 | break; | |
493 | ||
494 | distance += ret; | |
495 | } | |
496 | ||
497 | if (not_supported) | |
498 | return -1; | |
499 | ||
500 | return distance; | |
501 | } | |
502 | EXPORT_SYMBOL_GPL(pci_p2pdma_distance_many); | |
503 | ||
504 | /** | |
505 | * pci_has_p2pmem - check if a given PCI device has published any p2pmem | |
506 | * @pdev: PCI device to check | |
507 | */ | |
508 | bool pci_has_p2pmem(struct pci_dev *pdev) | |
509 | { | |
510 | return pdev->p2pdma && pdev->p2pdma->p2pmem_published; | |
511 | } | |
512 | EXPORT_SYMBOL_GPL(pci_has_p2pmem); | |
513 | ||
514 | /** | |
515 | * pci_p2pmem_find - find a peer-to-peer DMA memory device compatible with | |
516 | * the specified list of clients and shortest distance (as determined | |
517 | * by pci_p2pmem_dma()) | |
518 | * @clients: array of devices to check (NULL-terminated) | |
519 | * @num_clients: number of client devices in the list | |
520 | * | |
521 | * If multiple devices are behind the same switch, the one "closest" to the | |
fcf9ab35 | 522 | * client devices in use will be chosen first. (So if one of the providers is |
52916982 LG |
523 | * the same as one of the clients, that provider will be used ahead of any |
524 | * other providers that are unrelated). If multiple providers are an equal | |
525 | * distance away, one will be chosen at random. | |
526 | * | |
527 | * Returns a pointer to the PCI device with a reference taken (use pci_dev_put | |
528 | * to return the reference) or NULL if no compatible device is found. The | |
529 | * found provider will also be assigned to the client list. | |
530 | */ | |
531 | struct pci_dev *pci_p2pmem_find_many(struct device **clients, int num_clients) | |
532 | { | |
533 | struct pci_dev *pdev = NULL; | |
534 | int distance; | |
535 | int closest_distance = INT_MAX; | |
536 | struct pci_dev **closest_pdevs; | |
537 | int dev_cnt = 0; | |
538 | const int max_devs = PAGE_SIZE / sizeof(*closest_pdevs); | |
539 | int i; | |
540 | ||
541 | closest_pdevs = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
542 | if (!closest_pdevs) | |
543 | return NULL; | |
544 | ||
545 | while ((pdev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, pdev))) { | |
546 | if (!pci_has_p2pmem(pdev)) | |
547 | continue; | |
548 | ||
549 | distance = pci_p2pdma_distance_many(pdev, clients, | |
550 | num_clients, false); | |
551 | if (distance < 0 || distance > closest_distance) | |
552 | continue; | |
553 | ||
554 | if (distance == closest_distance && dev_cnt >= max_devs) | |
555 | continue; | |
556 | ||
557 | if (distance < closest_distance) { | |
558 | for (i = 0; i < dev_cnt; i++) | |
559 | pci_dev_put(closest_pdevs[i]); | |
560 | ||
561 | dev_cnt = 0; | |
562 | closest_distance = distance; | |
563 | } | |
564 | ||
565 | closest_pdevs[dev_cnt++] = pci_dev_get(pdev); | |
566 | } | |
567 | ||
568 | if (dev_cnt) | |
569 | pdev = pci_dev_get(closest_pdevs[prandom_u32_max(dev_cnt)]); | |
570 | ||
571 | for (i = 0; i < dev_cnt; i++) | |
572 | pci_dev_put(closest_pdevs[i]); | |
573 | ||
574 | kfree(closest_pdevs); | |
575 | return pdev; | |
576 | } | |
577 | EXPORT_SYMBOL_GPL(pci_p2pmem_find_many); | |
578 | ||
579 | /** | |
580 | * pci_alloc_p2p_mem - allocate peer-to-peer DMA memory | |
581 | * @pdev: the device to allocate memory from | |
582 | * @size: number of bytes to allocate | |
583 | * | |
584 | * Returns the allocated memory or NULL on error. | |
585 | */ | |
586 | void *pci_alloc_p2pmem(struct pci_dev *pdev, size_t size) | |
587 | { | |
588 | void *ret; | |
589 | ||
590 | if (unlikely(!pdev->p2pdma)) | |
591 | return NULL; | |
592 | ||
593 | if (unlikely(!percpu_ref_tryget_live(&pdev->p2pdma->devmap_ref))) | |
594 | return NULL; | |
595 | ||
596 | ret = (void *)gen_pool_alloc(pdev->p2pdma->pool, size); | |
597 | ||
598 | if (unlikely(!ret)) | |
599 | percpu_ref_put(&pdev->p2pdma->devmap_ref); | |
600 | ||
601 | return ret; | |
602 | } | |
603 | EXPORT_SYMBOL_GPL(pci_alloc_p2pmem); | |
604 | ||
605 | /** | |
606 | * pci_free_p2pmem - free peer-to-peer DMA memory | |
607 | * @pdev: the device the memory was allocated from | |
608 | * @addr: address of the memory that was allocated | |
fcf9ab35 | 609 | * @size: number of bytes that were allocated |
52916982 LG |
610 | */ |
611 | void pci_free_p2pmem(struct pci_dev *pdev, void *addr, size_t size) | |
612 | { | |
613 | gen_pool_free(pdev->p2pdma->pool, (uintptr_t)addr, size); | |
614 | percpu_ref_put(&pdev->p2pdma->devmap_ref); | |
615 | } | |
616 | EXPORT_SYMBOL_GPL(pci_free_p2pmem); | |
617 | ||
618 | /** | |
619 | * pci_virt_to_bus - return the PCI bus address for a given virtual | |
620 | * address obtained with pci_alloc_p2pmem() | |
621 | * @pdev: the device the memory was allocated from | |
622 | * @addr: address of the memory that was allocated | |
623 | */ | |
624 | pci_bus_addr_t pci_p2pmem_virt_to_bus(struct pci_dev *pdev, void *addr) | |
625 | { | |
626 | if (!addr) | |
627 | return 0; | |
628 | if (!pdev->p2pdma) | |
629 | return 0; | |
630 | ||
631 | /* | |
632 | * Note: when we added the memory to the pool we used the PCI | |
633 | * bus address as the physical address. So gen_pool_virt_to_phys() | |
634 | * actually returns the bus address despite the misleading name. | |
635 | */ | |
636 | return gen_pool_virt_to_phys(pdev->p2pdma->pool, (unsigned long)addr); | |
637 | } | |
638 | EXPORT_SYMBOL_GPL(pci_p2pmem_virt_to_bus); | |
639 | ||
640 | /** | |
641 | * pci_p2pmem_alloc_sgl - allocate peer-to-peer DMA memory in a scatterlist | |
642 | * @pdev: the device to allocate memory from | |
643 | * @nents: the number of SG entries in the list | |
644 | * @length: number of bytes to allocate | |
645 | * | |
fcf9ab35 | 646 | * Return: %NULL on error or &struct scatterlist pointer and @nents on success |
52916982 LG |
647 | */ |
648 | struct scatterlist *pci_p2pmem_alloc_sgl(struct pci_dev *pdev, | |
649 | unsigned int *nents, u32 length) | |
650 | { | |
651 | struct scatterlist *sg; | |
652 | void *addr; | |
653 | ||
654 | sg = kzalloc(sizeof(*sg), GFP_KERNEL); | |
655 | if (!sg) | |
656 | return NULL; | |
657 | ||
658 | sg_init_table(sg, 1); | |
659 | ||
660 | addr = pci_alloc_p2pmem(pdev, length); | |
661 | if (!addr) | |
662 | goto out_free_sg; | |
663 | ||
664 | sg_set_buf(sg, addr, length); | |
665 | *nents = 1; | |
666 | return sg; | |
667 | ||
668 | out_free_sg: | |
669 | kfree(sg); | |
670 | return NULL; | |
671 | } | |
672 | EXPORT_SYMBOL_GPL(pci_p2pmem_alloc_sgl); | |
673 | ||
674 | /** | |
675 | * pci_p2pmem_free_sgl - free a scatterlist allocated by pci_p2pmem_alloc_sgl() | |
676 | * @pdev: the device to allocate memory from | |
677 | * @sgl: the allocated scatterlist | |
678 | */ | |
679 | void pci_p2pmem_free_sgl(struct pci_dev *pdev, struct scatterlist *sgl) | |
680 | { | |
681 | struct scatterlist *sg; | |
682 | int count; | |
683 | ||
684 | for_each_sg(sgl, sg, INT_MAX, count) { | |
685 | if (!sg) | |
686 | break; | |
687 | ||
688 | pci_free_p2pmem(pdev, sg_virt(sg), sg->length); | |
689 | } | |
690 | kfree(sgl); | |
691 | } | |
692 | EXPORT_SYMBOL_GPL(pci_p2pmem_free_sgl); | |
693 | ||
694 | /** | |
695 | * pci_p2pmem_publish - publish the peer-to-peer DMA memory for use by | |
696 | * other devices with pci_p2pmem_find() | |
697 | * @pdev: the device with peer-to-peer DMA memory to publish | |
698 | * @publish: set to true to publish the memory, false to unpublish it | |
699 | * | |
700 | * Published memory can be used by other PCI device drivers for | |
701 | * peer-2-peer DMA operations. Non-published memory is reserved for | |
fcf9ab35 | 702 | * exclusive use of the device driver that registers the peer-to-peer |
52916982 LG |
703 | * memory. |
704 | */ | |
705 | void pci_p2pmem_publish(struct pci_dev *pdev, bool publish) | |
706 | { | |
707 | if (pdev->p2pdma) | |
708 | pdev->p2pdma->p2pmem_published = publish; | |
709 | } | |
710 | EXPORT_SYMBOL_GPL(pci_p2pmem_publish); | |
977196b8 LG |
711 | |
712 | /** | |
713 | * pci_p2pdma_map_sg - map a PCI peer-to-peer scatterlist for DMA | |
714 | * @dev: device doing the DMA request | |
715 | * @sg: scatter list to map | |
716 | * @nents: elements in the scatterlist | |
717 | * @dir: DMA direction | |
718 | * | |
719 | * Scatterlists mapped with this function should not be unmapped in any way. | |
720 | * | |
721 | * Returns the number of SG entries mapped or 0 on error. | |
722 | */ | |
723 | int pci_p2pdma_map_sg(struct device *dev, struct scatterlist *sg, int nents, | |
724 | enum dma_data_direction dir) | |
725 | { | |
726 | struct dev_pagemap *pgmap; | |
727 | struct scatterlist *s; | |
728 | phys_addr_t paddr; | |
729 | int i; | |
730 | ||
731 | /* | |
732 | * p2pdma mappings are not compatible with devices that use | |
733 | * dma_virt_ops. If the upper layers do the right thing | |
734 | * this should never happen because it will be prevented | |
735 | * by the check in pci_p2pdma_add_client() | |
736 | */ | |
737 | if (WARN_ON_ONCE(IS_ENABLED(CONFIG_DMA_VIRT_OPS) && | |
738 | dev->dma_ops == &dma_virt_ops)) | |
739 | return 0; | |
740 | ||
741 | for_each_sg(sg, s, nents, i) { | |
742 | pgmap = sg_page(s)->pgmap; | |
743 | paddr = sg_phys(s); | |
744 | ||
745 | s->dma_address = paddr - pgmap->pci_p2pdma_bus_offset; | |
746 | sg_dma_len(s) = s->length; | |
747 | } | |
748 | ||
749 | return nents; | |
750 | } | |
751 | EXPORT_SYMBOL_GPL(pci_p2pdma_map_sg); | |
2d7bc010 LG |
752 | |
753 | /** | |
754 | * pci_p2pdma_enable_store - parse a configfs/sysfs attribute store | |
755 | * to enable p2pdma | |
756 | * @page: contents of the value to be stored | |
757 | * @p2p_dev: returns the PCI device that was selected to be used | |
758 | * (if one was specified in the stored value) | |
759 | * @use_p2pdma: returns whether to enable p2pdma or not | |
760 | * | |
761 | * Parses an attribute value to decide whether to enable p2pdma. | |
fcf9ab35 | 762 | * The value can select a PCI device (using its full BDF device |
2d7bc010 LG |
763 | * name) or a boolean (in any format strtobool() accepts). A false |
764 | * value disables p2pdma, a true value expects the caller | |
765 | * to automatically find a compatible device and specifying a PCI device | |
766 | * expects the caller to use the specific provider. | |
767 | * | |
768 | * pci_p2pdma_enable_show() should be used as the show operation for | |
769 | * the attribute. | |
770 | * | |
771 | * Returns 0 on success | |
772 | */ | |
773 | int pci_p2pdma_enable_store(const char *page, struct pci_dev **p2p_dev, | |
774 | bool *use_p2pdma) | |
775 | { | |
776 | struct device *dev; | |
777 | ||
778 | dev = bus_find_device_by_name(&pci_bus_type, NULL, page); | |
779 | if (dev) { | |
780 | *use_p2pdma = true; | |
781 | *p2p_dev = to_pci_dev(dev); | |
782 | ||
783 | if (!pci_has_p2pmem(*p2p_dev)) { | |
784 | pci_err(*p2p_dev, | |
785 | "PCI device has no peer-to-peer memory: %s\n", | |
786 | page); | |
787 | pci_dev_put(*p2p_dev); | |
788 | return -ENODEV; | |
789 | } | |
790 | ||
791 | return 0; | |
792 | } else if ((page[0] == '0' || page[0] == '1') && !iscntrl(page[1])) { | |
793 | /* | |
794 | * If the user enters a PCI device that doesn't exist | |
795 | * like "0000:01:00.1", we don't want strtobool to think | |
796 | * it's a '0' when it's clearly not what the user wanted. | |
797 | * So we require 0's and 1's to be exactly one character. | |
798 | */ | |
799 | } else if (!strtobool(page, use_p2pdma)) { | |
800 | return 0; | |
801 | } | |
802 | ||
803 | pr_err("No such PCI device: %.*s\n", (int)strcspn(page, "\n"), page); | |
804 | return -ENODEV; | |
805 | } | |
806 | EXPORT_SYMBOL_GPL(pci_p2pdma_enable_store); | |
807 | ||
808 | /** | |
809 | * pci_p2pdma_enable_show - show a configfs/sysfs attribute indicating | |
810 | * whether p2pdma is enabled | |
811 | * @page: contents of the stored value | |
812 | * @p2p_dev: the selected p2p device (NULL if no device is selected) | |
fcf9ab35 | 813 | * @use_p2pdma: whether p2pdma has been enabled |
2d7bc010 LG |
814 | * |
815 | * Attributes that use pci_p2pdma_enable_store() should use this function | |
816 | * to show the value of the attribute. | |
817 | * | |
818 | * Returns 0 on success | |
819 | */ | |
820 | ssize_t pci_p2pdma_enable_show(char *page, struct pci_dev *p2p_dev, | |
821 | bool use_p2pdma) | |
822 | { | |
823 | if (!use_p2pdma) | |
824 | return sprintf(page, "0\n"); | |
825 | ||
826 | if (!p2p_dev) | |
827 | return sprintf(page, "1\n"); | |
828 | ||
829 | return sprintf(page, "%s\n", pci_name(p2p_dev)); | |
830 | } | |
831 | EXPORT_SYMBOL_GPL(pci_p2pdma_enable_show); |