]>
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> | |
110203be | 21 | #include <linux/xarray.h> |
52916982 | 22 | |
72583084 LG |
23 | enum 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 | ||
52916982 | 30 | struct pci_p2pdma { |
52916982 LG |
31 | struct gen_pool *pool; |
32 | bool p2pmem_published; | |
110203be | 33 | struct xarray map_types; |
52916982 LG |
34 | }; |
35 | ||
a6e6fe65 LG |
36 | struct pci_p2pdma_pagemap { |
37 | struct dev_pagemap pgmap; | |
0afea381 | 38 | struct pci_dev *provider; |
a6e6fe65 LG |
39 | u64 bus_offset; |
40 | }; | |
41 | ||
42 | static struct pci_p2pdma_pagemap *to_p2p_pgmap(struct dev_pagemap *pgmap) | |
43 | { | |
44 | return container_of(pgmap, struct pci_p2pdma_pagemap, pgmap); | |
45 | } | |
46 | ||
cbb8ca69 LG |
47 | static 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 | } | |
58 | static DEVICE_ATTR_RO(size); | |
59 | ||
60 | static 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 | } | |
71 | static DEVICE_ATTR_RO(available); | |
72 | ||
73 | static 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 | } | |
81 | static DEVICE_ATTR_RO(published); | |
82 | ||
83 | static struct attribute *p2pmem_attrs[] = { | |
84 | &dev_attr_size.attr, | |
85 | &dev_attr_available.attr, | |
86 | &dev_attr_published.attr, | |
87 | NULL, | |
88 | }; | |
89 | ||
90 | static const struct attribute_group p2pmem_group = { | |
91 | .attrs = p2pmem_attrs, | |
92 | .name = "p2pmem", | |
93 | }; | |
94 | ||
52916982 LG |
95 | static void pci_p2pdma_release(void *data) |
96 | { | |
97 | struct pci_dev *pdev = data; | |
1570175a | 98 | struct pci_p2pdma *p2pdma = pdev->p2pdma; |
52916982 | 99 | |
1570175a | 100 | if (!p2pdma) |
52916982 LG |
101 | return; |
102 | ||
1570175a DW |
103 | /* Flush and disable pci_alloc_p2p_mem() */ |
104 | pdev->p2pdma = NULL; | |
105 | synchronize_rcu(); | |
52916982 | 106 | |
1570175a | 107 | gen_pool_destroy(p2pdma->pool); |
cbb8ca69 | 108 | sysfs_remove_group(&pdev->dev.kobj, &p2pmem_group); |
110203be | 109 | xa_destroy(&p2pdma->map_types); |
52916982 LG |
110 | } |
111 | ||
112 | static 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 | ||
110203be LG |
121 | xa_init(&p2p->map_types); |
122 | ||
52916982 LG |
123 | p2p->pool = gen_pool_create(PAGE_SHIFT, dev_to_node(&pdev->dev)); |
124 | if (!p2p->pool) | |
125 | goto out; | |
126 | ||
52916982 LG |
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 | ||
cbb8ca69 LG |
133 | error = sysfs_create_group(&pdev->dev.kobj, &p2pmem_group); |
134 | if (error) | |
135 | goto out_pool_destroy; | |
136 | ||
52916982 LG |
137 | return 0; |
138 | ||
139 | out_pool_destroy: | |
cbb8ca69 | 140 | pdev->p2pdma = NULL; |
52916982 LG |
141 | gen_pool_destroy(p2p->pool); |
142 | out: | |
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 | */ | |
157 | int pci_p2pdma_add_resource(struct pci_dev *pdev, int bar, size_t size, | |
158 | u64 offset) | |
159 | { | |
a6e6fe65 | 160 | struct pci_p2pdma_pagemap *p2p_pgmap; |
52916982 LG |
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 | ||
a6e6fe65 LG |
183 | p2p_pgmap = devm_kzalloc(&pdev->dev, sizeof(*p2p_pgmap), GFP_KERNEL); |
184 | if (!p2p_pgmap) | |
52916982 | 185 | return -ENOMEM; |
a6e6fe65 LG |
186 | |
187 | pgmap = &p2p_pgmap->pgmap; | |
52916982 LG |
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); | |
52916982 | 191 | pgmap->type = MEMORY_DEVICE_PCI_P2PDMA; |
a6e6fe65 | 192 | |
0afea381 | 193 | p2p_pgmap->provider = pdev; |
a6e6fe65 | 194 | p2p_pgmap->bus_offset = pci_bus_address(pdev, bar) - |
977196b8 | 195 | pci_resource_start(pdev, bar); |
52916982 LG |
196 | |
197 | addr = devm_memremap_pages(&pdev->dev, pgmap); | |
198 | if (IS_ERR(addr)) { | |
199 | error = PTR_ERR(addr); | |
50f44ee7 | 200 | goto pgmap_free; |
52916982 LG |
201 | } |
202 | ||
1570175a | 203 | error = gen_pool_add_owner(pdev->p2pdma->pool, (unsigned long)addr, |
52916982 | 204 | pci_bus_address(pdev, bar) + offset, |
1570175a | 205 | resource_size(&pgmap->res), dev_to_node(&pdev->dev), |
d0b3517d | 206 | pgmap->ref); |
52916982 | 207 | if (error) |
e615a191 | 208 | goto pages_free; |
52916982 | 209 | |
52916982 LG |
210 | pci_info(pdev, "added peer-to-peer DMA memory %pR\n", |
211 | &pgmap->res); | |
212 | ||
213 | return 0; | |
214 | ||
e615a191 DW |
215 | pages_free: |
216 | devm_memunmap_pages(&pdev->dev, pgmap); | |
52916982 | 217 | pgmap_free: |
d0b3517d | 218 | devm_kfree(&pdev->dev, pgmap); |
52916982 LG |
219 | return error; |
220 | } | |
221 | EXPORT_SYMBOL_GPL(pci_p2pdma_add_resource); | |
222 | ||
223 | /* | |
224 | * Note this function returns the parent PCI device with a | |
f6b6aefe | 225 | * reference taken. It is the caller's responsibility to drop |
52916982 LG |
226 | * the reference. |
227 | */ | |
228 | static 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 | */ | |
251 | static int pci_bridge_has_acs_redir(struct pci_dev *pdev) | |
252 | { | |
253 | int pos; | |
254 | u16 ctrl; | |
255 | ||
256 | pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_ACS); | |
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 | ||
268 | static 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 | ||
494d63b0 LG |
276 | static const struct pci_p2pdma_whitelist_entry { |
277 | unsigned short vendor; | |
278 | unsigned short device; | |
279 | enum { | |
280 | REQ_SAME_HOST_BRIDGE = 1 << 0, | |
281 | } flags; | |
282 | } pci_p2pdma_whitelist[] = { | |
283 | /* AMD ZEN */ | |
284 | {PCI_VENDOR_ID_AMD, 0x1450, 0}, | |
285 | ||
286 | /* Intel Xeon E5/Core i7 */ | |
287 | {PCI_VENDOR_ID_INTEL, 0x3c00, REQ_SAME_HOST_BRIDGE}, | |
288 | {PCI_VENDOR_ID_INTEL, 0x3c01, REQ_SAME_HOST_BRIDGE}, | |
289 | /* Intel Xeon E7 v3/Xeon E5 v3/Core i7 */ | |
290 | {PCI_VENDOR_ID_INTEL, 0x2f00, REQ_SAME_HOST_BRIDGE}, | |
291 | {PCI_VENDOR_ID_INTEL, 0x2f01, REQ_SAME_HOST_BRIDGE}, | |
292 | {} | |
293 | }; | |
294 | ||
b3863578 LG |
295 | /* |
296 | * This lookup function tries to find the PCI device corresponding to a given | |
297 | * host bridge. | |
298 | * | |
299 | * It assumes the host bridge device is the first PCI device in the | |
300 | * bus->devices list and that the devfn is 00.0. These assumptions should hold | |
301 | * for all the devices in the whitelist above. | |
302 | * | |
303 | * This function is equivalent to pci_get_slot(host->bus, 0), however it does | |
304 | * not take the pci_bus_sem lock seeing __host_bridge_whitelist() must not | |
305 | * sleep. | |
306 | * | |
307 | * For this to be safe, the caller should hold a reference to a device on the | |
308 | * bridge, which should ensure the host_bridge device will not be freed | |
309 | * or removed from the head of the devices list. | |
310 | */ | |
311 | static struct pci_dev *pci_host_bridge_dev(struct pci_host_bridge *host) | |
312 | { | |
313 | struct pci_dev *root; | |
314 | ||
315 | root = list_first_entry_or_null(&host->bus->devices, | |
316 | struct pci_dev, bus_list); | |
317 | ||
318 | if (!root) | |
319 | return NULL; | |
320 | if (root->devfn != PCI_DEVFN(0, 0)) | |
321 | return NULL; | |
322 | ||
323 | return root; | |
324 | } | |
325 | ||
494d63b0 LG |
326 | static bool __host_bridge_whitelist(struct pci_host_bridge *host, |
327 | bool same_host_bridge) | |
0f97da83 | 328 | { |
b3863578 | 329 | struct pci_dev *root = pci_host_bridge_dev(host); |
494d63b0 | 330 | const struct pci_p2pdma_whitelist_entry *entry; |
0f97da83 CK |
331 | unsigned short vendor, device; |
332 | ||
333 | if (!root) | |
334 | return false; | |
335 | ||
336 | vendor = root->vendor; | |
337 | device = root->device; | |
0f97da83 | 338 | |
494d63b0 LG |
339 | for (entry = pci_p2pdma_whitelist; entry->vendor; entry++) { |
340 | if (vendor != entry->vendor || device != entry->device) | |
341 | continue; | |
342 | if (entry->flags & REQ_SAME_HOST_BRIDGE && !same_host_bridge) | |
343 | return false; | |
344 | ||
0f97da83 | 345 | return true; |
494d63b0 | 346 | } |
0f97da83 CK |
347 | |
348 | return false; | |
349 | } | |
350 | ||
2c84d818 LG |
351 | /* |
352 | * If we can't find a common upstream bridge take a look at the root | |
353 | * complex and compare it to a whitelist of known good hardware. | |
354 | */ | |
355 | static bool host_bridge_whitelist(struct pci_dev *a, struct pci_dev *b) | |
356 | { | |
357 | struct pci_host_bridge *host_a = pci_find_host_bridge(a->bus); | |
358 | struct pci_host_bridge *host_b = pci_find_host_bridge(b->bus); | |
359 | ||
494d63b0 LG |
360 | if (host_a == host_b) |
361 | return __host_bridge_whitelist(host_a, true); | |
362 | ||
363 | if (__host_bridge_whitelist(host_a, false) && | |
364 | __host_bridge_whitelist(host_b, false)) | |
2c84d818 LG |
365 | return true; |
366 | ||
367 | return false; | |
368 | } | |
369 | ||
72583084 | 370 | static enum pci_p2pdma_map_type |
c6bfaeb5 | 371 | __upstream_bridge_distance(struct pci_dev *provider, struct pci_dev *client, |
72583084 | 372 | int *dist, bool *acs_redirects, struct seq_buf *acs_list) |
52916982 | 373 | { |
0f97da83 | 374 | struct pci_dev *a = provider, *b = client, *bb; |
52916982 LG |
375 | int dist_a = 0; |
376 | int dist_b = 0; | |
52916982 LG |
377 | int acs_cnt = 0; |
378 | ||
72583084 LG |
379 | if (acs_redirects) |
380 | *acs_redirects = false; | |
381 | ||
52916982 LG |
382 | /* |
383 | * Note, we don't need to take references to devices returned by | |
384 | * pci_upstream_bridge() seeing we hold a reference to a child | |
385 | * device which will already hold a reference to the upstream bridge. | |
386 | */ | |
387 | ||
388 | while (a) { | |
389 | dist_b = 0; | |
390 | ||
391 | if (pci_bridge_has_acs_redir(a)) { | |
392 | seq_buf_print_bus_devfn(acs_list, a); | |
393 | acs_cnt++; | |
394 | } | |
395 | ||
396 | bb = b; | |
397 | ||
398 | while (bb) { | |
399 | if (a == bb) | |
400 | goto check_b_path_acs; | |
401 | ||
402 | bb = pci_upstream_bridge(bb); | |
403 | dist_b++; | |
404 | } | |
405 | ||
406 | a = pci_upstream_bridge(a); | |
407 | dist_a++; | |
408 | } | |
409 | ||
72583084 LG |
410 | if (dist) |
411 | *dist = dist_a + dist_b; | |
412 | ||
e2cbfbf7 | 413 | return PCI_P2PDMA_MAP_THRU_HOST_BRIDGE; |
52916982 LG |
414 | |
415 | check_b_path_acs: | |
416 | bb = b; | |
417 | ||
418 | while (bb) { | |
419 | if (a == bb) | |
420 | break; | |
421 | ||
422 | if (pci_bridge_has_acs_redir(bb)) { | |
423 | seq_buf_print_bus_devfn(acs_list, bb); | |
424 | acs_cnt++; | |
425 | } | |
426 | ||
427 | bb = pci_upstream_bridge(bb); | |
428 | } | |
429 | ||
72583084 LG |
430 | if (dist) |
431 | *dist = dist_a + dist_b; | |
52916982 | 432 | |
72583084 LG |
433 | if (acs_cnt) { |
434 | if (acs_redirects) | |
435 | *acs_redirects = true; | |
436 | ||
e2cbfbf7 | 437 | return PCI_P2PDMA_MAP_THRU_HOST_BRIDGE; |
72583084 LG |
438 | } |
439 | ||
440 | return PCI_P2PDMA_MAP_BUS_ADDR; | |
52916982 LG |
441 | } |
442 | ||
110203be LG |
443 | static unsigned long map_types_idx(struct pci_dev *client) |
444 | { | |
445 | return (pci_domain_nr(client->bus) << 16) | | |
446 | (client->bus->number << 8) | client->devfn; | |
447 | } | |
448 | ||
c6bfaeb5 LG |
449 | /* |
450 | * Find the distance through the nearest common upstream bridge between | |
451 | * two PCI devices. | |
452 | * | |
453 | * If the two devices are the same device then 0 will be returned. | |
454 | * | |
455 | * If there are two virtual functions of the same device behind the same | |
456 | * bridge port then 2 will be returned (one step down to the PCIe switch, | |
457 | * then one step back to the same device). | |
458 | * | |
459 | * In the case where two devices are connected to the same PCIe switch, the | |
460 | * value 4 will be returned. This corresponds to the following PCI tree: | |
461 | * | |
462 | * -+ Root Port | |
463 | * \+ Switch Upstream Port | |
464 | * +-+ Switch Downstream Port | |
465 | * + \- Device A | |
466 | * \-+ Switch Downstream Port | |
467 | * \- Device B | |
468 | * | |
469 | * The distance is 4 because we traverse from Device A through the downstream | |
470 | * port of the switch, to the common upstream port, back up to the second | |
471 | * downstream port and then to Device B. | |
472 | * | |
473 | * Any two devices that cannot communicate using p2pdma will return | |
474 | * PCI_P2PDMA_MAP_NOT_SUPPORTED. | |
475 | * | |
476 | * Any two devices that have a data path that goes through the host bridge | |
477 | * will consult a whitelist. If the host bridges are on the whitelist, | |
478 | * this function will return PCI_P2PDMA_MAP_THRU_HOST_BRIDGE. | |
479 | * | |
480 | * If either bridge is not on the whitelist this function returns | |
481 | * PCI_P2PDMA_MAP_NOT_SUPPORTED. | |
482 | * | |
483 | * If a bridge which has any ACS redirection bits set is in the path, | |
484 | * acs_redirects will be set to true. In this case, a list of all infringing | |
485 | * bridge addresses will be populated in acs_list (assuming it's non-null) | |
486 | * for printk purposes. | |
487 | */ | |
488 | static enum pci_p2pdma_map_type | |
489 | upstream_bridge_distance(struct pci_dev *provider, struct pci_dev *client, | |
490 | int *dist, bool *acs_redirects, struct seq_buf *acs_list) | |
491 | { | |
e2cbfbf7 LG |
492 | enum pci_p2pdma_map_type map_type; |
493 | ||
494 | map_type = __upstream_bridge_distance(provider, client, dist, | |
495 | acs_redirects, acs_list); | |
496 | ||
497 | if (map_type == PCI_P2PDMA_MAP_THRU_HOST_BRIDGE) { | |
2c84d818 | 498 | if (!host_bridge_whitelist(provider, client)) |
110203be | 499 | map_type = PCI_P2PDMA_MAP_NOT_SUPPORTED; |
e2cbfbf7 LG |
500 | } |
501 | ||
110203be LG |
502 | if (provider->p2pdma) |
503 | xa_store(&provider->p2pdma->map_types, map_types_idx(client), | |
504 | xa_mk_value(map_type), GFP_KERNEL); | |
505 | ||
e2cbfbf7 | 506 | return map_type; |
c6bfaeb5 LG |
507 | } |
508 | ||
72583084 LG |
509 | static enum pci_p2pdma_map_type |
510 | upstream_bridge_distance_warn(struct pci_dev *provider, struct pci_dev *client, | |
511 | int *dist) | |
52916982 LG |
512 | { |
513 | struct seq_buf acs_list; | |
72583084 | 514 | bool acs_redirects; |
52916982 LG |
515 | int ret; |
516 | ||
517 | seq_buf_init(&acs_list, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); | |
518 | if (!acs_list.buffer) | |
519 | return -ENOMEM; | |
520 | ||
72583084 LG |
521 | ret = upstream_bridge_distance(provider, client, dist, &acs_redirects, |
522 | &acs_list); | |
523 | if (acs_redirects) { | |
524 | pci_warn(client, "ACS redirect is set between the client and provider (%s)\n", | |
52916982 LG |
525 | pci_name(provider)); |
526 | /* Drop final semicolon */ | |
527 | acs_list.buffer[acs_list.len-1] = 0; | |
528 | pci_warn(client, "to disable ACS redirect for this path, add the kernel parameter: pci=disable_acs_redir=%s\n", | |
529 | acs_list.buffer); | |
72583084 | 530 | } |
52916982 | 531 | |
72583084 LG |
532 | if (ret == PCI_P2PDMA_MAP_NOT_SUPPORTED) { |
533 | 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", | |
52916982 LG |
534 | pci_name(provider)); |
535 | } | |
536 | ||
537 | kfree(acs_list.buffer); | |
538 | ||
539 | return ret; | |
540 | } | |
541 | ||
542 | /** | |
f6b6aefe | 543 | * pci_p2pdma_distance_many - Determine the cumulative distance between |
52916982 LG |
544 | * a p2pdma provider and the clients in use. |
545 | * @provider: p2pdma provider to check against the client list | |
546 | * @clients: array of devices to check (NULL-terminated) | |
547 | * @num_clients: number of clients in the array | |
548 | * @verbose: if true, print warnings for devices when we return -1 | |
549 | * | |
27235b15 LG |
550 | * Returns -1 if any of the clients are not compatible, otherwise returns a |
551 | * positive number where a lower number is the preferable choice. (If there's | |
552 | * one client that's the same as the provider it will return 0, which is best | |
553 | * choice). | |
52916982 | 554 | * |
27235b15 LG |
555 | * "compatible" means the provider and the clients are either all behind |
556 | * the same PCI root port or the host bridges connected to each of the devices | |
557 | * are listed in the 'pci_p2pdma_whitelist'. | |
52916982 LG |
558 | */ |
559 | int pci_p2pdma_distance_many(struct pci_dev *provider, struct device **clients, | |
560 | int num_clients, bool verbose) | |
561 | { | |
562 | bool not_supported = false; | |
563 | struct pci_dev *pci_client; | |
72583084 LG |
564 | int total_dist = 0; |
565 | int distance; | |
52916982 LG |
566 | int i, ret; |
567 | ||
568 | if (num_clients == 0) | |
569 | return -1; | |
570 | ||
571 | for (i = 0; i < num_clients; i++) { | |
9c002bb6 LG |
572 | if (IS_ENABLED(CONFIG_DMA_VIRT_OPS) && |
573 | clients[i]->dma_ops == &dma_virt_ops) { | |
574 | if (verbose) | |
575 | dev_warn(clients[i], | |
576 | "cannot be used for peer-to-peer DMA because the driver makes use of dma_virt_ops\n"); | |
577 | return -1; | |
578 | } | |
579 | ||
52916982 LG |
580 | pci_client = find_parent_pci_dev(clients[i]); |
581 | if (!pci_client) { | |
582 | if (verbose) | |
583 | dev_warn(clients[i], | |
584 | "cannot be used for peer-to-peer DMA as it is not a PCI device\n"); | |
585 | return -1; | |
586 | } | |
587 | ||
588 | if (verbose) | |
589 | ret = upstream_bridge_distance_warn(provider, | |
72583084 | 590 | pci_client, &distance); |
52916982 LG |
591 | else |
592 | ret = upstream_bridge_distance(provider, pci_client, | |
72583084 | 593 | &distance, NULL, NULL); |
52916982 LG |
594 | |
595 | pci_dev_put(pci_client); | |
596 | ||
72583084 | 597 | if (ret == PCI_P2PDMA_MAP_NOT_SUPPORTED) |
52916982 LG |
598 | not_supported = true; |
599 | ||
600 | if (not_supported && !verbose) | |
601 | break; | |
602 | ||
72583084 | 603 | total_dist += distance; |
52916982 LG |
604 | } |
605 | ||
606 | if (not_supported) | |
607 | return -1; | |
608 | ||
72583084 | 609 | return total_dist; |
52916982 LG |
610 | } |
611 | EXPORT_SYMBOL_GPL(pci_p2pdma_distance_many); | |
612 | ||
613 | /** | |
614 | * pci_has_p2pmem - check if a given PCI device has published any p2pmem | |
615 | * @pdev: PCI device to check | |
616 | */ | |
617 | bool pci_has_p2pmem(struct pci_dev *pdev) | |
618 | { | |
619 | return pdev->p2pdma && pdev->p2pdma->p2pmem_published; | |
620 | } | |
621 | EXPORT_SYMBOL_GPL(pci_has_p2pmem); | |
622 | ||
623 | /** | |
624 | * pci_p2pmem_find - find a peer-to-peer DMA memory device compatible with | |
625 | * the specified list of clients and shortest distance (as determined | |
626 | * by pci_p2pmem_dma()) | |
627 | * @clients: array of devices to check (NULL-terminated) | |
628 | * @num_clients: number of client devices in the list | |
629 | * | |
630 | * If multiple devices are behind the same switch, the one "closest" to the | |
fcf9ab35 | 631 | * client devices in use will be chosen first. (So if one of the providers is |
52916982 LG |
632 | * the same as one of the clients, that provider will be used ahead of any |
633 | * other providers that are unrelated). If multiple providers are an equal | |
634 | * distance away, one will be chosen at random. | |
635 | * | |
636 | * Returns a pointer to the PCI device with a reference taken (use pci_dev_put | |
637 | * to return the reference) or NULL if no compatible device is found. The | |
638 | * found provider will also be assigned to the client list. | |
639 | */ | |
640 | struct pci_dev *pci_p2pmem_find_many(struct device **clients, int num_clients) | |
641 | { | |
642 | struct pci_dev *pdev = NULL; | |
643 | int distance; | |
644 | int closest_distance = INT_MAX; | |
645 | struct pci_dev **closest_pdevs; | |
646 | int dev_cnt = 0; | |
647 | const int max_devs = PAGE_SIZE / sizeof(*closest_pdevs); | |
648 | int i; | |
649 | ||
650 | closest_pdevs = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
651 | if (!closest_pdevs) | |
652 | return NULL; | |
653 | ||
654 | while ((pdev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, pdev))) { | |
655 | if (!pci_has_p2pmem(pdev)) | |
656 | continue; | |
657 | ||
658 | distance = pci_p2pdma_distance_many(pdev, clients, | |
659 | num_clients, false); | |
660 | if (distance < 0 || distance > closest_distance) | |
661 | continue; | |
662 | ||
663 | if (distance == closest_distance && dev_cnt >= max_devs) | |
664 | continue; | |
665 | ||
666 | if (distance < closest_distance) { | |
667 | for (i = 0; i < dev_cnt; i++) | |
668 | pci_dev_put(closest_pdevs[i]); | |
669 | ||
670 | dev_cnt = 0; | |
671 | closest_distance = distance; | |
672 | } | |
673 | ||
674 | closest_pdevs[dev_cnt++] = pci_dev_get(pdev); | |
675 | } | |
676 | ||
677 | if (dev_cnt) | |
678 | pdev = pci_dev_get(closest_pdevs[prandom_u32_max(dev_cnt)]); | |
679 | ||
680 | for (i = 0; i < dev_cnt; i++) | |
681 | pci_dev_put(closest_pdevs[i]); | |
682 | ||
683 | kfree(closest_pdevs); | |
684 | return pdev; | |
685 | } | |
686 | EXPORT_SYMBOL_GPL(pci_p2pmem_find_many); | |
687 | ||
688 | /** | |
689 | * pci_alloc_p2p_mem - allocate peer-to-peer DMA memory | |
690 | * @pdev: the device to allocate memory from | |
691 | * @size: number of bytes to allocate | |
692 | * | |
693 | * Returns the allocated memory or NULL on error. | |
694 | */ | |
695 | void *pci_alloc_p2pmem(struct pci_dev *pdev, size_t size) | |
696 | { | |
1570175a DW |
697 | void *ret = NULL; |
698 | struct percpu_ref *ref; | |
52916982 | 699 | |
1570175a DW |
700 | /* |
701 | * Pairs with synchronize_rcu() in pci_p2pdma_release() to | |
702 | * ensure pdev->p2pdma is non-NULL for the duration of the | |
703 | * read-lock. | |
704 | */ | |
705 | rcu_read_lock(); | |
52916982 | 706 | if (unlikely(!pdev->p2pdma)) |
1570175a | 707 | goto out; |
52916982 | 708 | |
1570175a DW |
709 | ret = (void *)gen_pool_alloc_owner(pdev->p2pdma->pool, size, |
710 | (void **) &ref); | |
711 | if (!ret) | |
712 | goto out; | |
52916982 | 713 | |
1570175a DW |
714 | if (unlikely(!percpu_ref_tryget_live(ref))) { |
715 | gen_pool_free(pdev->p2pdma->pool, (unsigned long) ret, size); | |
716 | ret = NULL; | |
717 | goto out; | |
718 | } | |
719 | out: | |
720 | rcu_read_unlock(); | |
52916982 LG |
721 | return ret; |
722 | } | |
723 | EXPORT_SYMBOL_GPL(pci_alloc_p2pmem); | |
724 | ||
725 | /** | |
726 | * pci_free_p2pmem - free peer-to-peer DMA memory | |
727 | * @pdev: the device the memory was allocated from | |
728 | * @addr: address of the memory that was allocated | |
fcf9ab35 | 729 | * @size: number of bytes that were allocated |
52916982 LG |
730 | */ |
731 | void pci_free_p2pmem(struct pci_dev *pdev, void *addr, size_t size) | |
732 | { | |
1570175a DW |
733 | struct percpu_ref *ref; |
734 | ||
735 | gen_pool_free_owner(pdev->p2pdma->pool, (uintptr_t)addr, size, | |
736 | (void **) &ref); | |
737 | percpu_ref_put(ref); | |
52916982 LG |
738 | } |
739 | EXPORT_SYMBOL_GPL(pci_free_p2pmem); | |
740 | ||
741 | /** | |
742 | * pci_virt_to_bus - return the PCI bus address for a given virtual | |
743 | * address obtained with pci_alloc_p2pmem() | |
744 | * @pdev: the device the memory was allocated from | |
745 | * @addr: address of the memory that was allocated | |
746 | */ | |
747 | pci_bus_addr_t pci_p2pmem_virt_to_bus(struct pci_dev *pdev, void *addr) | |
748 | { | |
749 | if (!addr) | |
750 | return 0; | |
751 | if (!pdev->p2pdma) | |
752 | return 0; | |
753 | ||
754 | /* | |
755 | * Note: when we added the memory to the pool we used the PCI | |
756 | * bus address as the physical address. So gen_pool_virt_to_phys() | |
757 | * actually returns the bus address despite the misleading name. | |
758 | */ | |
759 | return gen_pool_virt_to_phys(pdev->p2pdma->pool, (unsigned long)addr); | |
760 | } | |
761 | EXPORT_SYMBOL_GPL(pci_p2pmem_virt_to_bus); | |
762 | ||
763 | /** | |
764 | * pci_p2pmem_alloc_sgl - allocate peer-to-peer DMA memory in a scatterlist | |
765 | * @pdev: the device to allocate memory from | |
766 | * @nents: the number of SG entries in the list | |
767 | * @length: number of bytes to allocate | |
768 | * | |
fcf9ab35 | 769 | * Return: %NULL on error or &struct scatterlist pointer and @nents on success |
52916982 LG |
770 | */ |
771 | struct scatterlist *pci_p2pmem_alloc_sgl(struct pci_dev *pdev, | |
772 | unsigned int *nents, u32 length) | |
773 | { | |
774 | struct scatterlist *sg; | |
775 | void *addr; | |
776 | ||
777 | sg = kzalloc(sizeof(*sg), GFP_KERNEL); | |
778 | if (!sg) | |
779 | return NULL; | |
780 | ||
781 | sg_init_table(sg, 1); | |
782 | ||
783 | addr = pci_alloc_p2pmem(pdev, length); | |
784 | if (!addr) | |
785 | goto out_free_sg; | |
786 | ||
787 | sg_set_buf(sg, addr, length); | |
788 | *nents = 1; | |
789 | return sg; | |
790 | ||
791 | out_free_sg: | |
792 | kfree(sg); | |
793 | return NULL; | |
794 | } | |
795 | EXPORT_SYMBOL_GPL(pci_p2pmem_alloc_sgl); | |
796 | ||
797 | /** | |
798 | * pci_p2pmem_free_sgl - free a scatterlist allocated by pci_p2pmem_alloc_sgl() | |
799 | * @pdev: the device to allocate memory from | |
800 | * @sgl: the allocated scatterlist | |
801 | */ | |
802 | void pci_p2pmem_free_sgl(struct pci_dev *pdev, struct scatterlist *sgl) | |
803 | { | |
804 | struct scatterlist *sg; | |
805 | int count; | |
806 | ||
807 | for_each_sg(sgl, sg, INT_MAX, count) { | |
808 | if (!sg) | |
809 | break; | |
810 | ||
811 | pci_free_p2pmem(pdev, sg_virt(sg), sg->length); | |
812 | } | |
813 | kfree(sgl); | |
814 | } | |
815 | EXPORT_SYMBOL_GPL(pci_p2pmem_free_sgl); | |
816 | ||
817 | /** | |
818 | * pci_p2pmem_publish - publish the peer-to-peer DMA memory for use by | |
819 | * other devices with pci_p2pmem_find() | |
820 | * @pdev: the device with peer-to-peer DMA memory to publish | |
821 | * @publish: set to true to publish the memory, false to unpublish it | |
822 | * | |
823 | * Published memory can be used by other PCI device drivers for | |
824 | * peer-2-peer DMA operations. Non-published memory is reserved for | |
fcf9ab35 | 825 | * exclusive use of the device driver that registers the peer-to-peer |
52916982 LG |
826 | * memory. |
827 | */ | |
828 | void pci_p2pmem_publish(struct pci_dev *pdev, bool publish) | |
829 | { | |
830 | if (pdev->p2pdma) | |
831 | pdev->p2pdma->p2pmem_published = publish; | |
832 | } | |
833 | EXPORT_SYMBOL_GPL(pci_p2pmem_publish); | |
977196b8 | 834 | |
5d52e1ab LG |
835 | static enum pci_p2pdma_map_type pci_p2pdma_map_type(struct pci_dev *provider, |
836 | struct pci_dev *client) | |
837 | { | |
838 | if (!provider->p2pdma) | |
839 | return PCI_P2PDMA_MAP_NOT_SUPPORTED; | |
840 | ||
841 | return xa_to_value(xa_load(&provider->p2pdma->map_types, | |
842 | map_types_idx(client))); | |
843 | } | |
844 | ||
142c242e LG |
845 | static int __pci_p2pdma_map_sg(struct pci_p2pdma_pagemap *p2p_pgmap, |
846 | struct device *dev, struct scatterlist *sg, int nents) | |
977196b8 | 847 | { |
977196b8 LG |
848 | struct scatterlist *s; |
849 | phys_addr_t paddr; | |
850 | int i; | |
851 | ||
852 | /* | |
853 | * p2pdma mappings are not compatible with devices that use | |
854 | * dma_virt_ops. If the upper layers do the right thing | |
855 | * this should never happen because it will be prevented | |
9c002bb6 | 856 | * by the check in pci_p2pdma_distance_many() |
977196b8 LG |
857 | */ |
858 | if (WARN_ON_ONCE(IS_ENABLED(CONFIG_DMA_VIRT_OPS) && | |
859 | dev->dma_ops == &dma_virt_ops)) | |
860 | return 0; | |
861 | ||
862 | for_each_sg(sg, s, nents, i) { | |
977196b8 LG |
863 | paddr = sg_phys(s); |
864 | ||
a6e6fe65 | 865 | s->dma_address = paddr - p2p_pgmap->bus_offset; |
977196b8 LG |
866 | sg_dma_len(s) = s->length; |
867 | } | |
868 | ||
869 | return nents; | |
870 | } | |
142c242e LG |
871 | |
872 | /** | |
873 | * pci_p2pdma_map_sg - map a PCI peer-to-peer scatterlist for DMA | |
874 | * @dev: device doing the DMA request | |
875 | * @sg: scatter list to map | |
876 | * @nents: elements in the scatterlist | |
877 | * @dir: DMA direction | |
878 | * @attrs: DMA attributes passed to dma_map_sg() (if called) | |
879 | * | |
880 | * Scatterlists mapped with this function should be unmapped using | |
881 | * pci_p2pdma_unmap_sg_attrs(). | |
882 | * | |
883 | * Returns the number of SG entries mapped or 0 on error. | |
884 | */ | |
885 | int pci_p2pdma_map_sg_attrs(struct device *dev, struct scatterlist *sg, | |
886 | int nents, enum dma_data_direction dir, unsigned long attrs) | |
887 | { | |
888 | struct pci_p2pdma_pagemap *p2p_pgmap = | |
889 | to_p2p_pgmap(sg_page(sg)->pgmap); | |
5d52e1ab | 890 | struct pci_dev *client; |
142c242e | 891 | |
5d52e1ab LG |
892 | if (WARN_ON_ONCE(!dev_is_pci(dev))) |
893 | return 0; | |
894 | ||
895 | client = to_pci_dev(dev); | |
896 | ||
897 | switch (pci_p2pdma_map_type(p2p_pgmap->provider, client)) { | |
898 | case PCI_P2PDMA_MAP_THRU_HOST_BRIDGE: | |
899 | return dma_map_sg_attrs(dev, sg, nents, dir, attrs); | |
900 | case PCI_P2PDMA_MAP_BUS_ADDR: | |
901 | return __pci_p2pdma_map_sg(p2p_pgmap, dev, sg, nents); | |
902 | default: | |
903 | WARN_ON_ONCE(1); | |
904 | return 0; | |
905 | } | |
142c242e | 906 | } |
2b9f4bb2 | 907 | EXPORT_SYMBOL_GPL(pci_p2pdma_map_sg_attrs); |
2d7bc010 | 908 | |
7f73eac3 LG |
909 | /** |
910 | * pci_p2pdma_unmap_sg - unmap a PCI peer-to-peer scatterlist that was | |
911 | * mapped with pci_p2pdma_map_sg() | |
912 | * @dev: device doing the DMA request | |
913 | * @sg: scatter list to map | |
914 | * @nents: number of elements returned by pci_p2pdma_map_sg() | |
915 | * @dir: DMA direction | |
916 | * @attrs: DMA attributes passed to dma_unmap_sg() (if called) | |
917 | */ | |
918 | void pci_p2pdma_unmap_sg_attrs(struct device *dev, struct scatterlist *sg, | |
919 | int nents, enum dma_data_direction dir, unsigned long attrs) | |
920 | { | |
5d52e1ab LG |
921 | struct pci_p2pdma_pagemap *p2p_pgmap = |
922 | to_p2p_pgmap(sg_page(sg)->pgmap); | |
923 | enum pci_p2pdma_map_type map_type; | |
924 | struct pci_dev *client; | |
925 | ||
926 | if (WARN_ON_ONCE(!dev_is_pci(dev))) | |
927 | return; | |
928 | ||
929 | client = to_pci_dev(dev); | |
930 | ||
931 | map_type = pci_p2pdma_map_type(p2p_pgmap->provider, client); | |
932 | ||
933 | if (map_type == PCI_P2PDMA_MAP_THRU_HOST_BRIDGE) | |
934 | dma_unmap_sg_attrs(dev, sg, nents, dir, attrs); | |
7f73eac3 LG |
935 | } |
936 | EXPORT_SYMBOL_GPL(pci_p2pdma_unmap_sg_attrs); | |
937 | ||
2d7bc010 LG |
938 | /** |
939 | * pci_p2pdma_enable_store - parse a configfs/sysfs attribute store | |
940 | * to enable p2pdma | |
941 | * @page: contents of the value to be stored | |
942 | * @p2p_dev: returns the PCI device that was selected to be used | |
943 | * (if one was specified in the stored value) | |
944 | * @use_p2pdma: returns whether to enable p2pdma or not | |
945 | * | |
946 | * Parses an attribute value to decide whether to enable p2pdma. | |
fcf9ab35 | 947 | * The value can select a PCI device (using its full BDF device |
2d7bc010 LG |
948 | * name) or a boolean (in any format strtobool() accepts). A false |
949 | * value disables p2pdma, a true value expects the caller | |
950 | * to automatically find a compatible device and specifying a PCI device | |
951 | * expects the caller to use the specific provider. | |
952 | * | |
953 | * pci_p2pdma_enable_show() should be used as the show operation for | |
954 | * the attribute. | |
955 | * | |
956 | * Returns 0 on success | |
957 | */ | |
958 | int pci_p2pdma_enable_store(const char *page, struct pci_dev **p2p_dev, | |
959 | bool *use_p2pdma) | |
960 | { | |
961 | struct device *dev; | |
962 | ||
963 | dev = bus_find_device_by_name(&pci_bus_type, NULL, page); | |
964 | if (dev) { | |
965 | *use_p2pdma = true; | |
966 | *p2p_dev = to_pci_dev(dev); | |
967 | ||
968 | if (!pci_has_p2pmem(*p2p_dev)) { | |
969 | pci_err(*p2p_dev, | |
970 | "PCI device has no peer-to-peer memory: %s\n", | |
971 | page); | |
972 | pci_dev_put(*p2p_dev); | |
973 | return -ENODEV; | |
974 | } | |
975 | ||
976 | return 0; | |
977 | } else if ((page[0] == '0' || page[0] == '1') && !iscntrl(page[1])) { | |
978 | /* | |
979 | * If the user enters a PCI device that doesn't exist | |
980 | * like "0000:01:00.1", we don't want strtobool to think | |
981 | * it's a '0' when it's clearly not what the user wanted. | |
982 | * So we require 0's and 1's to be exactly one character. | |
983 | */ | |
984 | } else if (!strtobool(page, use_p2pdma)) { | |
985 | return 0; | |
986 | } | |
987 | ||
988 | pr_err("No such PCI device: %.*s\n", (int)strcspn(page, "\n"), page); | |
989 | return -ENODEV; | |
990 | } | |
991 | EXPORT_SYMBOL_GPL(pci_p2pdma_enable_store); | |
992 | ||
993 | /** | |
994 | * pci_p2pdma_enable_show - show a configfs/sysfs attribute indicating | |
995 | * whether p2pdma is enabled | |
996 | * @page: contents of the stored value | |
997 | * @p2p_dev: the selected p2p device (NULL if no device is selected) | |
fcf9ab35 | 998 | * @use_p2pdma: whether p2pdma has been enabled |
2d7bc010 LG |
999 | * |
1000 | * Attributes that use pci_p2pdma_enable_store() should use this function | |
1001 | * to show the value of the attribute. | |
1002 | * | |
1003 | * Returns 0 on success | |
1004 | */ | |
1005 | ssize_t pci_p2pdma_enable_show(char *page, struct pci_dev *p2p_dev, | |
1006 | bool use_p2pdma) | |
1007 | { | |
1008 | if (!use_p2pdma) | |
1009 | return sprintf(page, "0\n"); | |
1010 | ||
1011 | if (!p2p_dev) | |
1012 | return sprintf(page, "1\n"); | |
1013 | ||
1014 | return sprintf(page, "%s\n", pci_name(p2p_dev)); | |
1015 | } | |
1016 | EXPORT_SYMBOL_GPL(pci_p2pdma_enable_show); |