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> | |
6dbbd053 | 21 | #include <linux/iommu.h> |
52916982 LG |
22 | |
23 | struct pci_p2pdma { | |
52916982 LG |
24 | struct gen_pool *pool; |
25 | bool p2pmem_published; | |
26 | }; | |
27 | ||
cbb8ca69 LG |
28 | static ssize_t size_show(struct device *dev, struct device_attribute *attr, |
29 | char *buf) | |
30 | { | |
31 | struct pci_dev *pdev = to_pci_dev(dev); | |
32 | size_t size = 0; | |
33 | ||
34 | if (pdev->p2pdma->pool) | |
35 | size = gen_pool_size(pdev->p2pdma->pool); | |
36 | ||
37 | return snprintf(buf, PAGE_SIZE, "%zd\n", size); | |
38 | } | |
39 | static DEVICE_ATTR_RO(size); | |
40 | ||
41 | static ssize_t available_show(struct device *dev, struct device_attribute *attr, | |
42 | char *buf) | |
43 | { | |
44 | struct pci_dev *pdev = to_pci_dev(dev); | |
45 | size_t avail = 0; | |
46 | ||
47 | if (pdev->p2pdma->pool) | |
48 | avail = gen_pool_avail(pdev->p2pdma->pool); | |
49 | ||
50 | return snprintf(buf, PAGE_SIZE, "%zd\n", avail); | |
51 | } | |
52 | static DEVICE_ATTR_RO(available); | |
53 | ||
54 | static ssize_t published_show(struct device *dev, struct device_attribute *attr, | |
55 | char *buf) | |
56 | { | |
57 | struct pci_dev *pdev = to_pci_dev(dev); | |
58 | ||
59 | return snprintf(buf, PAGE_SIZE, "%d\n", | |
60 | pdev->p2pdma->p2pmem_published); | |
61 | } | |
62 | static DEVICE_ATTR_RO(published); | |
63 | ||
64 | static struct attribute *p2pmem_attrs[] = { | |
65 | &dev_attr_size.attr, | |
66 | &dev_attr_available.attr, | |
67 | &dev_attr_published.attr, | |
68 | NULL, | |
69 | }; | |
70 | ||
71 | static const struct attribute_group p2pmem_group = { | |
72 | .attrs = p2pmem_attrs, | |
73 | .name = "p2pmem", | |
74 | }; | |
75 | ||
52916982 LG |
76 | static void pci_p2pdma_release(void *data) |
77 | { | |
78 | struct pci_dev *pdev = data; | |
1570175a | 79 | struct pci_p2pdma *p2pdma = pdev->p2pdma; |
52916982 | 80 | |
1570175a | 81 | if (!p2pdma) |
52916982 LG |
82 | return; |
83 | ||
1570175a DW |
84 | /* Flush and disable pci_alloc_p2p_mem() */ |
85 | pdev->p2pdma = NULL; | |
86 | synchronize_rcu(); | |
52916982 | 87 | |
1570175a | 88 | gen_pool_destroy(p2pdma->pool); |
cbb8ca69 | 89 | sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group); |
52916982 LG |
90 | } |
91 | ||
92 | static int pci_p2pdma_setup(struct pci_dev *pdev) | |
93 | { | |
94 | int error = -ENOMEM; | |
95 | struct pci_p2pdma *p2p; | |
96 | ||
97 | p2p = devm_kzalloc(&pdev->dev, sizeof(*p2p), GFP_KERNEL); | |
98 | if (!p2p) | |
99 | return -ENOMEM; | |
100 | ||
101 | p2p->pool = gen_pool_create(PAGE_SHIFT, dev_to_node(&pdev->dev)); | |
102 | if (!p2p->pool) | |
103 | goto out; | |
104 | ||
52916982 LG |
105 | error = devm_add_action_or_reset(&pdev->dev, pci_p2pdma_release, pdev); |
106 | if (error) | |
107 | goto out_pool_destroy; | |
108 | ||
109 | pdev->p2pdma = p2p; | |
110 | ||
cbb8ca69 LG |
111 | error = sysfs_create_group(&pdev->dev.kobj, &p2pmem_group); |
112 | if (error) | |
113 | goto out_pool_destroy; | |
114 | ||
52916982 LG |
115 | return 0; |
116 | ||
117 | out_pool_destroy: | |
cbb8ca69 | 118 | pdev->p2pdma = NULL; |
52916982 LG |
119 | gen_pool_destroy(p2p->pool); |
120 | out: | |
121 | devm_kfree(&pdev->dev, p2p); | |
122 | return error; | |
123 | } | |
124 | ||
125 | /** | |
126 | * pci_p2pdma_add_resource - add memory for use as p2p memory | |
127 | * @pdev: the device to add the memory to | |
128 | * @bar: PCI BAR to add | |
129 | * @size: size of the memory to add, may be zero to use the whole BAR | |
130 | * @offset: offset into the PCI BAR | |
131 | * | |
132 | * The memory will be given ZONE_DEVICE struct pages so that it may | |
133 | * be used with any DMA request. | |
134 | */ | |
135 | int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size, | |
136 | u64 offset) | |
137 | { | |
138 | struct dev_pagemap *pgmap; | |
139 | void *addr; | |
140 | int error; | |
141 | ||
142 | if (!(pci_resource_flags(pdev, bar) & IORESOURCE_MEM)) | |
143 | return -EINVAL; | |
144 | ||
145 | if (offset >= pci_resource_len(pdev, bar)) | |
146 | return -EINVAL; | |
147 | ||
148 | if (!size) | |
149 | size = pci_resource_len(pdev, bar) - offset; | |
150 | ||
151 | if (size + offset > pci_resource_len(pdev, bar)) | |
152 | return -EINVAL; | |
153 | ||
154 | if (!pdev->p2pdma) { | |
155 | error = pci_p2pdma_setup(pdev); | |
156 | if (error) | |
157 | return error; | |
158 | } | |
159 | ||
d0b3517d CH |
160 | pgmap = devm_kzalloc(&pdev->dev, sizeof(*pgmap), GFP_KERNEL); |
161 | if (!pgmap) | |
52916982 | 162 | return -ENOMEM; |
52916982 LG |
163 | pgmap->res.start = pci_resource_start(pdev, bar) + offset; |
164 | pgmap->res.end = pgmap->res.start + size - 1; | |
165 | pgmap->res.flags = pci_resource_flags(pdev, bar); | |
52916982 | 166 | pgmap->type = MEMORY_DEVICE_PCI_P2PDMA; |
977196b8 LG |
167 | pgmap->pci_p2pdma_bus_offset = pci_bus_address(pdev, bar) - |
168 | pci_resource_start(pdev, bar); | |
52916982 LG |
169 | |
170 | addr = devm_memremap_pages(&pdev->dev, pgmap); | |
171 | if (IS_ERR(addr)) { | |
172 | error = PTR_ERR(addr); | |
50f44ee7 | 173 | goto pgmap_free; |
52916982 LG |
174 | } |
175 | ||
1570175a | 176 | error = gen_pool_add_owner(pdev->p2pdma->pool, (unsigned long)addr, |
52916982 | 177 | pci_bus_address(pdev, bar) + offset, |
1570175a | 178 | resource_size(&pgmap->res), dev_to_node(&pdev->dev), |
d0b3517d | 179 | pgmap->ref); |
52916982 | 180 | if (error) |
e615a191 | 181 | goto pages_free; |
52916982 | 182 | |
52916982 LG |
183 | pci_info(pdev, "added peer-to-peer DMA memory %pR\n", |
184 | &pgmap->res); | |
185 | ||
186 | return 0; | |
187 | ||
e615a191 DW |
188 | pages_free: |
189 | devm_memunmap_pages(&pdev->dev, pgmap); | |
52916982 | 190 | pgmap_free: |
d0b3517d | 191 | devm_kfree(&pdev->dev, pgmap); |
52916982 LG |
192 | return error; |
193 | } | |
194 | EXPORT_SYMBOL_GPL(pci_p2pdma_add_resource); | |
195 | ||
196 | /* | |
197 | * Note this function returns the parent PCI device with a | |
f6b6aefe | 198 | * reference taken. It is the caller's responsibility to drop |
52916982 LG |
199 | * the reference. |
200 | */ | |
201 | static struct pci_dev *find_parent_pci_dev(struct device *dev) | |
202 | { | |
203 | struct device *parent; | |
204 | ||
205 | dev = get_device(dev); | |
206 | ||
207 | while (dev) { | |
208 | if (dev_is_pci(dev)) | |
209 | return to_pci_dev(dev); | |
210 | ||
211 | parent = get_device(dev->parent); | |
212 | put_device(dev); | |
213 | dev = parent; | |
214 | } | |
215 | ||
216 | return NULL; | |
217 | } | |
218 | ||
219 | /* | |
220 | * Check if a PCI bridge has its ACS redirection bits set to redirect P2P | |
221 | * TLPs upstream via ACS. Returns 1 if the packets will be redirected | |
222 | * upstream, 0 otherwise. | |
223 | */ | |
224 | static int pci_bridge_has_acs_redir(struct pci_dev *pdev) | |
225 | { | |
226 | int pos; | |
227 | u16 ctrl; | |
228 | ||
229 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS); | |
230 | if (!pos) | |
231 | return 0; | |
232 | ||
233 | pci_read_config_word(pdev, pos + PCI_ACS_CTRL, &ctrl); | |
234 | ||
235 | if (ctrl & (PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_EC)) | |
236 | return 1; | |
237 | ||
238 | return 0; | |
239 | } | |
240 | ||
241 | static void seq_buf_print_bus_devfn(struct seq_buf *buf, struct pci_dev *pdev) | |
242 | { | |
243 | if (!buf) | |
244 | return; | |
245 | ||
246 | seq_buf_printf(buf, "%s;", pci_name(pdev)); | |
247 | } | |
248 | ||
0f97da83 CK |
249 | /* |
250 | * If we can't find a common upstream bridge take a look at the root | |
251 | * complex and compare it to a whitelist of known good hardware. | |
252 | */ | |
253 | static bool root_complex_whitelist(struct pci_dev *dev) | |
254 | { | |
255 | struct pci_host_bridge *host = pci_find_host_bridge(dev->bus); | |
256 | struct pci_dev *root = pci_get_slot(host->bus, PCI_DEVFN(0, 0)); | |
257 | unsigned short vendor, device; | |
258 | ||
6dbbd053 LG |
259 | if (iommu_present(dev->dev.bus)) |
260 | return false; | |
261 | ||
0f97da83 CK |
262 | if (!root) |
263 | return false; | |
264 | ||
265 | vendor = root->vendor; | |
266 | device = root->device; | |
267 | pci_dev_put(root); | |
268 | ||
269 | /* AMD ZEN host bridges can do peer to peer */ | |
270 | if (vendor == PCI_VENDOR_ID_AMD && device == 0x1450) | |
271 | return true; | |
272 | ||
273 | return false; | |
274 | } | |
275 | ||
52916982 LG |
276 | /* |
277 | * Find the distance through the nearest common upstream bridge between | |
278 | * two PCI devices. | |
279 | * | |
280 | * If the two devices are the same device then 0 will be returned. | |
281 | * | |
282 | * If there are two virtual functions of the same device behind the same | |
283 | * bridge port then 2 will be returned (one step down to the PCIe switch, | |
284 | * then one step back to the same device). | |
285 | * | |
286 | * In the case where two devices are connected to the same PCIe switch, the | |
287 | * value 4 will be returned. This corresponds to the following PCI tree: | |
288 | * | |
289 | * -+ Root Port | |
290 | * \+ Switch Upstream Port | |
291 | * +-+ Switch Downstream Port | |
292 | * + \- Device A | |
293 | * \-+ Switch Downstream Port | |
294 | * \- Device B | |
295 | * | |
296 | * The distance is 4 because we traverse from Device A through the downstream | |
297 | * port of the switch, to the common upstream port, back up to the second | |
298 | * downstream port and then to Device B. | |
299 | * | |
300 | * Any two devices that don't have a common upstream bridge will return -1. | |
301 | * In this way devices on separate PCIe root ports will be rejected, which | |
302 | * is what we want for peer-to-peer seeing each PCIe root port defines a | |
303 | * separate hierarchy domain and there's no way to determine whether the root | |
304 | * complex supports forwarding between them. | |
305 | * | |
306 | * In the case where two devices are connected to different PCIe switches, | |
307 | * this function will still return a positive distance as long as both | |
308 | * switches eventually have a common upstream bridge. Note this covers | |
309 | * the case of using multiple PCIe switches to achieve a desired level of | |
310 | * fan-out from a root port. The exact distance will be a function of the | |
311 | * number of switches between Device A and Device B. | |
312 | * | |
313 | * If a bridge which has any ACS redirection bits set is in the path | |
314 | * then this functions will return -2. This is so we reject any | |
315 | * cases where the TLPs are forwarded up into the root complex. | |
316 | * In this case, a list of all infringing bridge addresses will be | |
317 | * populated in acs_list (assuming it's non-null) for printk purposes. | |
318 | */ | |
0f97da83 CK |
319 | static int upstream_bridge_distance(struct pci_dev *provider, |
320 | struct pci_dev *client, | |
52916982 LG |
321 | struct seq_buf *acs_list) |
322 | { | |
0f97da83 | 323 | struct pci_dev *a = provider, *b = client, *bb; |
52916982 LG |
324 | int dist_a = 0; |
325 | int dist_b = 0; | |
52916982 LG |
326 | int acs_cnt = 0; |
327 | ||
328 | /* | |
329 | * Note, we don't need to take references to devices returned by | |
330 | * pci_upstream_bridge() seeing we hold a reference to a child | |
331 | * device which will already hold a reference to the upstream bridge. | |
332 | */ | |
333 | ||
334 | while (a) { | |
335 | dist_b = 0; | |
336 | ||
337 | if (pci_bridge_has_acs_redir(a)) { | |
338 | seq_buf_print_bus_devfn(acs_list, a); | |
339 | acs_cnt++; | |
340 | } | |
341 | ||
342 | bb = b; | |
343 | ||
344 | while (bb) { | |
345 | if (a == bb) | |
346 | goto check_b_path_acs; | |
347 | ||
348 | bb = pci_upstream_bridge(bb); | |
349 | dist_b++; | |
350 | } | |
351 | ||
352 | a = pci_upstream_bridge(a); | |
353 | dist_a++; | |
354 | } | |
355 | ||
0f97da83 CK |
356 | /* |
357 | * Allow the connection if both devices are on a whitelisted root | |
f6b6aefe | 358 | * complex, but add an arbitrary large value to the distance. |
0f97da83 CK |
359 | */ |
360 | if (root_complex_whitelist(provider) && | |
361 | root_complex_whitelist(client)) | |
362 | return 0x1000 + dist_a + dist_b; | |
363 | ||
52916982 LG |
364 | return -1; |
365 | ||
366 | check_b_path_acs: | |
367 | bb = b; | |
368 | ||
369 | while (bb) { | |
370 | if (a == bb) | |
371 | break; | |
372 | ||
373 | if (pci_bridge_has_acs_redir(bb)) { | |
374 | seq_buf_print_bus_devfn(acs_list, bb); | |
375 | acs_cnt++; | |
376 | } | |
377 | ||
378 | bb = pci_upstream_bridge(bb); | |
379 | } | |
380 | ||
381 | if (acs_cnt) | |
382 | return -2; | |
383 | ||
384 | return dist_a + dist_b; | |
385 | } | |
386 | ||
387 | static int upstream_bridge_distance_warn(struct pci_dev *provider, | |
388 | struct pci_dev *client) | |
389 | { | |
390 | struct seq_buf acs_list; | |
391 | int ret; | |
392 | ||
393 | seq_buf_init(&acs_list, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); | |
394 | if (!acs_list.buffer) | |
395 | return -ENOMEM; | |
396 | ||
397 | ret = upstream_bridge_distance(provider, client, &acs_list); | |
398 | if (ret == -2) { | |
399 | pci_warn(client, "cannot be used for peer-to-peer DMA as ACS redirect is set between the client and provider (%s)\n", | |
400 | pci_name(provider)); | |
401 | /* Drop final semicolon */ | |
402 | acs_list.buffer[acs_list.len-1] = 0; | |
403 | pci_warn(client, "to disable ACS redirect for this path, add the kernel parameter: pci=disable_acs_redir=%s\n", | |
404 | acs_list.buffer); | |
405 | ||
406 | } else if (ret < 0) { | |
407 | pci_warn(client, "cannot be used for peer-to-peer DMA as the client and provider (%s) do not share an upstream bridge\n", | |
408 | pci_name(provider)); | |
409 | } | |
410 | ||
411 | kfree(acs_list.buffer); | |
412 | ||
413 | return ret; | |
414 | } | |
415 | ||
416 | /** | |
f6b6aefe | 417 | * pci_p2pdma_distance_many - Determine the cumulative distance between |
52916982 LG |
418 | * a p2pdma provider and the clients in use. |
419 | * @provider: p2pdma provider to check against the client list | |
420 | * @clients: array of devices to check (NULL-terminated) | |
421 | * @num_clients: number of clients in the array | |
422 | * @verbose: if true, print warnings for devices when we return -1 | |
423 | * | |
424 | * Returns -1 if any of the clients are not compatible (behind the same | |
425 | * root port as the provider), otherwise returns a positive number where | |
fcf9ab35 | 426 | * a lower number is the preferable choice. (If there's one client |
52916982 LG |
427 | * that's the same as the provider it will return 0, which is best choice). |
428 | * | |
429 | * For now, "compatible" means the provider and the clients are all behind | |
430 | * the same PCI root port. This cuts out cases that may work but is safest | |
431 | * for the user. Future work can expand this to white-list root complexes that | |
432 | * can safely forward between each ports. | |
433 | */ | |
434 | int pci_p2pdma_distance_many(struct pci_dev *provider, struct device **clients, | |
435 | int num_clients, bool verbose) | |
436 | { | |
437 | bool not_supported = false; | |
438 | struct pci_dev *pci_client; | |
439 | int distance = 0; | |
440 | int i, ret; | |
441 | ||
442 | if (num_clients == 0) | |
443 | return -1; | |
444 | ||
445 | for (i = 0; i < num_clients; i++) { | |
9c002bb6 LG |
446 | if (IS_ENABLED(CONFIG_DMA_VIRT_OPS) && |
447 | clients[i]->dma_ops == &dma_virt_ops) { | |
448 | if (verbose) | |
449 | dev_warn(clients[i], | |
450 | "cannot be used for peer-to-peer DMA because the driver makes use of dma_virt_ops\n"); | |
451 | return -1; | |
452 | } | |
453 | ||
52916982 LG |
454 | pci_client = find_parent_pci_dev(clients[i]); |
455 | if (!pci_client) { | |
456 | if (verbose) | |
457 | dev_warn(clients[i], | |
458 | "cannot be used for peer-to-peer DMA as it is not a PCI device\n"); | |
459 | return -1; | |
460 | } | |
461 | ||
462 | if (verbose) | |
463 | ret = upstream_bridge_distance_warn(provider, | |
464 | pci_client); | |
465 | else | |
466 | ret = upstream_bridge_distance(provider, pci_client, | |
467 | NULL); | |
468 | ||
469 | pci_dev_put(pci_client); | |
470 | ||
471 | if (ret < 0) | |
472 | not_supported = true; | |
473 | ||
474 | if (not_supported && !verbose) | |
475 | break; | |
476 | ||
477 | distance += ret; | |
478 | } | |
479 | ||
480 | if (not_supported) | |
481 | return -1; | |
482 | ||
483 | return distance; | |
484 | } | |
485 | EXPORT_SYMBOL_GPL(pci_p2pdma_distance_many); | |
486 | ||
487 | /** | |
488 | * pci_has_p2pmem - check if a given PCI device has published any p2pmem | |
489 | * @pdev: PCI device to check | |
490 | */ | |
491 | bool pci_has_p2pmem(struct pci_dev *pdev) | |
492 | { | |
493 | return pdev->p2pdma && pdev->p2pdma->p2pmem_published; | |
494 | } | |
495 | EXPORT_SYMBOL_GPL(pci_has_p2pmem); | |
496 | ||
497 | /** | |
498 | * pci_p2pmem_find - find a peer-to-peer DMA memory device compatible with | |
499 | * the specified list of clients and shortest distance (as determined | |
500 | * by pci_p2pmem_dma()) | |
501 | * @clients: array of devices to check (NULL-terminated) | |
502 | * @num_clients: number of client devices in the list | |
503 | * | |
504 | * If multiple devices are behind the same switch, the one "closest" to the | |
fcf9ab35 | 505 | * client devices in use will be chosen first. (So if one of the providers is |
52916982 LG |
506 | * the same as one of the clients, that provider will be used ahead of any |
507 | * other providers that are unrelated). If multiple providers are an equal | |
508 | * distance away, one will be chosen at random. | |
509 | * | |
510 | * Returns a pointer to the PCI device with a reference taken (use pci_dev_put | |
511 | * to return the reference) or NULL if no compatible device is found. The | |
512 | * found provider will also be assigned to the client list. | |
513 | */ | |
514 | struct pci_dev *pci_p2pmem_find_many(struct device **clients, int num_clients) | |
515 | { | |
516 | struct pci_dev *pdev = NULL; | |
517 | int distance; | |
518 | int closest_distance = INT_MAX; | |
519 | struct pci_dev **closest_pdevs; | |
520 | int dev_cnt = 0; | |
521 | const int max_devs = PAGE_SIZE / sizeof(*closest_pdevs); | |
522 | int i; | |
523 | ||
524 | closest_pdevs = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
525 | if (!closest_pdevs) | |
526 | return NULL; | |
527 | ||
528 | while ((pdev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, pdev))) { | |
529 | if (!pci_has_p2pmem(pdev)) | |
530 | continue; | |
531 | ||
532 | distance = pci_p2pdma_distance_many(pdev, clients, | |
533 | num_clients, false); | |
534 | if (distance < 0 || distance > closest_distance) | |
535 | continue; | |
536 | ||
537 | if (distance == closest_distance && dev_cnt >= max_devs) | |
538 | continue; | |
539 | ||
540 | if (distance < closest_distance) { | |
541 | for (i = 0; i < dev_cnt; i++) | |
542 | pci_dev_put(closest_pdevs[i]); | |
543 | ||
544 | dev_cnt = 0; | |
545 | closest_distance = distance; | |
546 | } | |
547 | ||
548 | closest_pdevs[dev_cnt++] = pci_dev_get(pdev); | |
549 | } | |
550 | ||
551 | if (dev_cnt) | |
552 | pdev = pci_dev_get(closest_pdevs[prandom_u32_max(dev_cnt)]); | |
553 | ||
554 | for (i = 0; i < dev_cnt; i++) | |
555 | pci_dev_put(closest_pdevs[i]); | |
556 | ||
557 | kfree(closest_pdevs); | |
558 | return pdev; | |
559 | } | |
560 | EXPORT_SYMBOL_GPL(pci_p2pmem_find_many); | |
561 | ||
562 | /** | |
563 | * pci_alloc_p2p_mem - allocate peer-to-peer DMA memory | |
564 | * @pdev: the device to allocate memory from | |
565 | * @size: number of bytes to allocate | |
566 | * | |
567 | * Returns the allocated memory or NULL on error. | |
568 | */ | |
569 | void *pci_alloc_p2pmem(struct pci_dev *pdev, size_t size) | |
570 | { | |
1570175a DW |
571 | void *ret = NULL; |
572 | struct percpu_ref *ref; | |
52916982 | 573 | |
1570175a DW |
574 | /* |
575 | * Pairs with synchronize_rcu() in pci_p2pdma_release() to | |
576 | * ensure pdev->p2pdma is non-NULL for the duration of the | |
577 | * read-lock. | |
578 | */ | |
579 | rcu_read_lock(); | |
52916982 | 580 | if (unlikely(!pdev->p2pdma)) |
1570175a | 581 | goto out; |
52916982 | 582 | |
1570175a DW |
583 | ret = (void *)gen_pool_alloc_owner(pdev->p2pdma->pool, size, |
584 | (void **) &ref); | |
585 | if (!ret) | |
586 | goto out; | |
52916982 | 587 | |
1570175a DW |
588 | if (unlikely(!percpu_ref_tryget_live(ref))) { |
589 | gen_pool_free(pdev->p2pdma->pool, (unsigned long) ret, size); | |
590 | ret = NULL; | |
591 | goto out; | |
592 | } | |
593 | out: | |
594 | rcu_read_unlock(); | |
52916982 LG |
595 | return ret; |
596 | } | |
597 | EXPORT_SYMBOL_GPL(pci_alloc_p2pmem); | |
598 | ||
599 | /** | |
600 | * pci_free_p2pmem - free peer-to-peer DMA memory | |
601 | * @pdev: the device the memory was allocated from | |
602 | * @addr: address of the memory that was allocated | |
fcf9ab35 | 603 | * @size: number of bytes that were allocated |
52916982 LG |
604 | */ |
605 | void pci_free_p2pmem(struct pci_dev *pdev, void *addr, size_t size) | |
606 | { | |
1570175a DW |
607 | struct percpu_ref *ref; |
608 | ||
609 | gen_pool_free_owner(pdev->p2pdma->pool, (uintptr_t)addr, size, | |
610 | (void **) &ref); | |
611 | percpu_ref_put(ref); | |
52916982 LG |
612 | } |
613 | EXPORT_SYMBOL_GPL(pci_free_p2pmem); | |
614 | ||
615 | /** | |
616 | * pci_virt_to_bus - return the PCI bus address for a given virtual | |
617 | * address obtained with pci_alloc_p2pmem() | |
618 | * @pdev: the device the memory was allocated from | |
619 | * @addr: address of the memory that was allocated | |
620 | */ | |
621 | pci_bus_addr_t pci_p2pmem_virt_to_bus(struct pci_dev *pdev, void *addr) | |
622 | { | |
623 | if (!addr) | |
624 | return 0; | |
625 | if (!pdev->p2pdma) | |
626 | return 0; | |
627 | ||
628 | /* | |
629 | * Note: when we added the memory to the pool we used the PCI | |
630 | * bus address as the physical address. So gen_pool_virt_to_phys() | |
631 | * actually returns the bus address despite the misleading name. | |
632 | */ | |
633 | return gen_pool_virt_to_phys(pdev->p2pdma->pool, (unsigned long)addr); | |
634 | } | |
635 | EXPORT_SYMBOL_GPL(pci_p2pmem_virt_to_bus); | |
636 | ||
637 | /** | |
638 | * pci_p2pmem_alloc_sgl - allocate peer-to-peer DMA memory in a scatterlist | |
639 | * @pdev: the device to allocate memory from | |
640 | * @nents: the number of SG entries in the list | |
641 | * @length: number of bytes to allocate | |
642 | * | |
fcf9ab35 | 643 | * Return: %NULL on error or &struct scatterlist pointer and @nents on success |
52916982 LG |
644 | */ |
645 | struct scatterlist *pci_p2pmem_alloc_sgl(struct pci_dev *pdev, | |
646 | unsigned int *nents, u32 length) | |
647 | { | |
648 | struct scatterlist *sg; | |
649 | void *addr; | |
650 | ||
651 | sg = kzalloc(sizeof(*sg), GFP_KERNEL); | |
652 | if (!sg) | |
653 | return NULL; | |
654 | ||
655 | sg_init_table(sg, 1); | |
656 | ||
657 | addr = pci_alloc_p2pmem(pdev, length); | |
658 | if (!addr) | |
659 | goto out_free_sg; | |
660 | ||
661 | sg_set_buf(sg, addr, length); | |
662 | *nents = 1; | |
663 | return sg; | |
664 | ||
665 | out_free_sg: | |
666 | kfree(sg); | |
667 | return NULL; | |
668 | } | |
669 | EXPORT_SYMBOL_GPL(pci_p2pmem_alloc_sgl); | |
670 | ||
671 | /** | |
672 | * pci_p2pmem_free_sgl - free a scatterlist allocated by pci_p2pmem_alloc_sgl() | |
673 | * @pdev: the device to allocate memory from | |
674 | * @sgl: the allocated scatterlist | |
675 | */ | |
676 | void pci_p2pmem_free_sgl(struct pci_dev *pdev, struct scatterlist *sgl) | |
677 | { | |
678 | struct scatterlist *sg; | |
679 | int count; | |
680 | ||
681 | for_each_sg(sgl, sg, INT_MAX, count) { | |
682 | if (!sg) | |
683 | break; | |
684 | ||
685 | pci_free_p2pmem(pdev, sg_virt(sg), sg->length); | |
686 | } | |
687 | kfree(sgl); | |
688 | } | |
689 | EXPORT_SYMBOL_GPL(pci_p2pmem_free_sgl); | |
690 | ||
691 | /** | |
692 | * pci_p2pmem_publish - publish the peer-to-peer DMA memory for use by | |
693 | * other devices with pci_p2pmem_find() | |
694 | * @pdev: the device with peer-to-peer DMA memory to publish | |
695 | * @publish: set to true to publish the memory, false to unpublish it | |
696 | * | |
697 | * Published memory can be used by other PCI device drivers for | |
698 | * peer-2-peer DMA operations. Non-published memory is reserved for | |
fcf9ab35 | 699 | * exclusive use of the device driver that registers the peer-to-peer |
52916982 LG |
700 | * memory. |
701 | */ | |
702 | void pci_p2pmem_publish(struct pci_dev *pdev, bool publish) | |
703 | { | |
704 | if (pdev->p2pdma) | |
705 | pdev->p2pdma->p2pmem_published = publish; | |
706 | } | |
707 | EXPORT_SYMBOL_GPL(pci_p2pmem_publish); | |
977196b8 LG |
708 | |
709 | /** | |
710 | * pci_p2pdma_map_sg - map a PCI peer-to-peer scatterlist for DMA | |
711 | * @dev: device doing the DMA request | |
712 | * @sg: scatter list to map | |
713 | * @nents: elements in the scatterlist | |
714 | * @dir: DMA direction | |
715 | * | |
716 | * Scatterlists mapped with this function should not be unmapped in any way. | |
717 | * | |
718 | * Returns the number of SG entries mapped or 0 on error. | |
719 | */ | |
720 | int pci_p2pdma_map_sg(struct device *dev, struct scatterlist *sg, int nents, | |
721 | enum dma_data_direction dir) | |
722 | { | |
723 | struct dev_pagemap *pgmap; | |
724 | struct scatterlist *s; | |
725 | phys_addr_t paddr; | |
726 | int i; | |
727 | ||
728 | /* | |
729 | * p2pdma mappings are not compatible with devices that use | |
730 | * dma_virt_ops. If the upper layers do the right thing | |
731 | * this should never happen because it will be prevented | |
9c002bb6 | 732 | * by the check in pci_p2pdma_distance_many() |
977196b8 LG |
733 | */ |
734 | if (WARN_ON_ONCE(IS_ENABLED(CONFIG_DMA_VIRT_OPS) && | |
735 | dev->dma_ops == &dma_virt_ops)) | |
736 | return 0; | |
737 | ||
738 | for_each_sg(sg, s, nents, i) { | |
739 | pgmap = sg_page(s)->pgmap; | |
740 | paddr = sg_phys(s); | |
741 | ||
742 | s->dma_address = paddr - pgmap->pci_p2pdma_bus_offset; | |
743 | sg_dma_len(s) = s->length; | |
744 | } | |
745 | ||
746 | return nents; | |
747 | } | |
748 | EXPORT_SYMBOL_GPL(pci_p2pdma_map_sg); | |
2d7bc010 LG |
749 | |
750 | /** | |
751 | * pci_p2pdma_enable_store - parse a configfs/sysfs attribute store | |
752 | * to enable p2pdma | |
753 | * @page: contents of the value to be stored | |
754 | * @p2p_dev: returns the PCI device that was selected to be used | |
755 | * (if one was specified in the stored value) | |
756 | * @use_p2pdma: returns whether to enable p2pdma or not | |
757 | * | |
758 | * Parses an attribute value to decide whether to enable p2pdma. | |
fcf9ab35 | 759 | * The value can select a PCI device (using its full BDF device |
2d7bc010 LG |
760 | * name) or a boolean (in any format strtobool() accepts). A false |
761 | * value disables p2pdma, a true value expects the caller | |
762 | * to automatically find a compatible device and specifying a PCI device | |
763 | * expects the caller to use the specific provider. | |
764 | * | |
765 | * pci_p2pdma_enable_show() should be used as the show operation for | |
766 | * the attribute. | |
767 | * | |
768 | * Returns 0 on success | |
769 | */ | |
770 | int pci_p2pdma_enable_store(const char *page, struct pci_dev **p2p_dev, | |
771 | bool *use_p2pdma) | |
772 | { | |
773 | struct device *dev; | |
774 | ||
775 | dev = bus_find_device_by_name(&pci_bus_type, NULL, page); | |
776 | if (dev) { | |
777 | *use_p2pdma = true; | |
778 | *p2p_dev = to_pci_dev(dev); | |
779 | ||
780 | if (!pci_has_p2pmem(*p2p_dev)) { | |
781 | pci_err(*p2p_dev, | |
782 | "PCI device has no peer-to-peer memory: %s\n", | |
783 | page); | |
784 | pci_dev_put(*p2p_dev); | |
785 | return -ENODEV; | |
786 | } | |
787 | ||
788 | return 0; | |
789 | } else if ((page[0] == '0' || page[0] == '1') && !iscntrl(page[1])) { | |
790 | /* | |
791 | * If the user enters a PCI device that doesn't exist | |
792 | * like "0000:01:00.1", we don't want strtobool to think | |
793 | * it's a '0' when it's clearly not what the user wanted. | |
794 | * So we require 0's and 1's to be exactly one character. | |
795 | */ | |
796 | } else if (!strtobool(page, use_p2pdma)) { | |
797 | return 0; | |
798 | } | |
799 | ||
800 | pr_err("No such PCI device: %.*s\n", (int)strcspn(page, "\n"), page); | |
801 | return -ENODEV; | |
802 | } | |
803 | EXPORT_SYMBOL_GPL(pci_p2pdma_enable_store); | |
804 | ||
805 | /** | |
806 | * pci_p2pdma_enable_show - show a configfs/sysfs attribute indicating | |
807 | * whether p2pdma is enabled | |
808 | * @page: contents of the stored value | |
809 | * @p2p_dev: the selected p2p device (NULL if no device is selected) | |
fcf9ab35 | 810 | * @use_p2pdma: whether p2pdma has been enabled |
2d7bc010 LG |
811 | * |
812 | * Attributes that use pci_p2pdma_enable_store() should use this function | |
813 | * to show the value of the attribute. | |
814 | * | |
815 | * Returns 0 on success | |
816 | */ | |
817 | ssize_t pci_p2pdma_enable_show(char *page, struct pci_dev *p2p_dev, | |
818 | bool use_p2pdma) | |
819 | { | |
820 | if (!use_p2pdma) | |
821 | return sprintf(page, "0\n"); | |
822 | ||
823 | if (!p2p_dev) | |
824 | return sprintf(page, "1\n"); | |
825 | ||
826 | return sprintf(page, "%s\n", pci_name(p2p_dev)); | |
827 | } | |
828 | EXPORT_SYMBOL_GPL(pci_p2pdma_enable_show); |