1 /*
2 * PCIe AER software error injection support.
3 *
4 * Debuging PCIe AER code is quite difficult because it is hard to
5 * trigger various real hardware errors. Software based error
6 * injection can fake almost all kinds of errors with the help of a
7 * user space helper tool aer-inject, which can be gotten from:
8 * http://www.kernel.org/pub/linux/utils/pci/aer-inject/
9 *
10 * Copyright 2009 Intel Corporation.
11 * Huang Ying <ying.huang@intel.com>
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; version 2
16 * of the License.
17 *
18 */
19
20 #include <linux/module.h>
21 #include <linux/init.h>
22 #include <linux/miscdevice.h>
23 #include <linux/pci.h>
24 #include <linux/slab.h>
25 #include <linux/fs.h>
26 #include <linux/uaccess.h>
27 #include <linux/stddef.h>
28 #include <linux/device.h>
29 #include "aerdrv.h"
30
31 /* Override the existing corrected and uncorrected error masks */
32 static bool aer_mask_override;
33 module_param(aer_mask_override, bool, 0);
34
35 struct aer_error_inj {
36 u8 bus;
37 u8 dev;
38 u8 fn;
39 u32 uncor_status;
40 u32 cor_status;
41 u32 header_log0;
42 u32 header_log1;
43 u32 header_log2;
44 u32 header_log3;
45 u32 domain;
46 };
47
48 struct aer_error {
49 struct list_head list;
50 u32 domain;
51 unsigned int bus;
52 unsigned int devfn;
53 int pos_cap_err;
54
55 u32 uncor_status;
56 u32 cor_status;
57 u32 header_log0;
58 u32 header_log1;
59 u32 header_log2;
60 u32 header_log3;
61 u32 root_status;
62 u32 source_id;
63 };
64
65 struct pci_bus_ops {
66 struct list_head list;
67 struct pci_bus *bus;
68 struct pci_ops *ops;
69 };
70
71 static LIST_HEAD(einjected);
72
73 static LIST_HEAD(pci_bus_ops_list);
74
75 /* Protect einjected and pci_bus_ops_list */
76 static DEFINE_SPINLOCK(inject_lock);
77
aer_error_init(struct aer_error * err,u32 domain,unsigned int bus,unsigned int devfn,int pos_cap_err)78 static void aer_error_init(struct aer_error *err, u32 domain,
79 unsigned int bus, unsigned int devfn,
80 int pos_cap_err)
81 {
82 INIT_LIST_HEAD(&err->list);
83 err->domain = domain;
84 err->bus = bus;
85 err->devfn = devfn;
86 err->pos_cap_err = pos_cap_err;
87 }
88
89 /* inject_lock must be held before calling */
__find_aer_error(u32 domain,unsigned int bus,unsigned int devfn)90 static struct aer_error *__find_aer_error(u32 domain, unsigned int bus,
91 unsigned int devfn)
92 {
93 struct aer_error *err;
94
95 list_for_each_entry(err, &einjected, list) {
96 if (domain == err->domain &&
97 bus == err->bus &&
98 devfn == err->devfn)
99 return err;
100 }
101 return NULL;
102 }
103
104 /* inject_lock must be held before calling */
__find_aer_error_by_dev(struct pci_dev * dev)105 static struct aer_error *__find_aer_error_by_dev(struct pci_dev *dev)
106 {
107 int domain = pci_domain_nr(dev->bus);
108 if (domain < 0)
109 return NULL;
110 return __find_aer_error(domain, dev->bus->number, dev->devfn);
111 }
112
113 /* inject_lock must be held before calling */
__find_pci_bus_ops(struct pci_bus * bus)114 static struct pci_ops *__find_pci_bus_ops(struct pci_bus *bus)
115 {
116 struct pci_bus_ops *bus_ops;
117
118 list_for_each_entry(bus_ops, &pci_bus_ops_list, list) {
119 if (bus_ops->bus == bus)
120 return bus_ops->ops;
121 }
122 return NULL;
123 }
124
pci_bus_ops_pop(void)125 static struct pci_bus_ops *pci_bus_ops_pop(void)
126 {
127 unsigned long flags;
128 struct pci_bus_ops *bus_ops;
129
130 spin_lock_irqsave(&inject_lock, flags);
131 bus_ops = list_first_entry_or_null(&pci_bus_ops_list,
132 struct pci_bus_ops, list);
133 if (bus_ops)
134 list_del(&bus_ops->list);
135 spin_unlock_irqrestore(&inject_lock, flags);
136 return bus_ops;
137 }
138
find_pci_config_dword(struct aer_error * err,int where,int * prw1cs)139 static u32 *find_pci_config_dword(struct aer_error *err, int where,
140 int *prw1cs)
141 {
142 int rw1cs = 0;
143 u32 *target = NULL;
144
145 if (err->pos_cap_err == -1)
146 return NULL;
147
148 switch (where - err->pos_cap_err) {
149 case PCI_ERR_UNCOR_STATUS:
150 target = &err->uncor_status;
151 rw1cs = 1;
152 break;
153 case PCI_ERR_COR_STATUS:
154 target = &err->cor_status;
155 rw1cs = 1;
156 break;
157 case PCI_ERR_HEADER_LOG:
158 target = &err->header_log0;
159 break;
160 case PCI_ERR_HEADER_LOG+4:
161 target = &err->header_log1;
162 break;
163 case PCI_ERR_HEADER_LOG+8:
164 target = &err->header_log2;
165 break;
166 case PCI_ERR_HEADER_LOG+12:
167 target = &err->header_log3;
168 break;
169 case PCI_ERR_ROOT_STATUS:
170 target = &err->root_status;
171 rw1cs = 1;
172 break;
173 case PCI_ERR_ROOT_ERR_SRC:
174 target = &err->source_id;
175 break;
176 }
177 if (prw1cs)
178 *prw1cs = rw1cs;
179 return target;
180 }
181
aer_inj_read_config(struct pci_bus * bus,unsigned int devfn,int where,int size,u32 * val)182 static int aer_inj_read_config(struct pci_bus *bus, unsigned int devfn,
183 int where, int size, u32 *val)
184 {
185 u32 *sim;
186 struct aer_error *err;
187 unsigned long flags;
188 struct pci_ops *ops;
189 struct pci_ops *my_ops;
190 int domain;
191 int rv;
192
193 spin_lock_irqsave(&inject_lock, flags);
194 if (size != sizeof(u32))
195 goto out;
196 domain = pci_domain_nr(bus);
197 if (domain < 0)
198 goto out;
199 err = __find_aer_error(domain, bus->number, devfn);
200 if (!err)
201 goto out;
202
203 sim = find_pci_config_dword(err, where, NULL);
204 if (sim) {
205 *val = *sim;
206 spin_unlock_irqrestore(&inject_lock, flags);
207 return 0;
208 }
209 out:
210 ops = __find_pci_bus_ops(bus);
211 /*
212 * pci_lock must already be held, so we can directly
213 * manipulate bus->ops. Many config access functions,
214 * including pci_generic_config_read() require the original
215 * bus->ops be installed to function, so temporarily put them
216 * back.
217 */
218 my_ops = bus->ops;
219 bus->ops = ops;
220 rv = ops->read(bus, devfn, where, size, val);
221 bus->ops = my_ops;
222 spin_unlock_irqrestore(&inject_lock, flags);
223 return rv;
224 }
225
aer_inj_write_config(struct pci_bus * bus,unsigned int devfn,int where,int size,u32 val)226 static int aer_inj_write_config(struct pci_bus *bus, unsigned int devfn,
227 int where, int size, u32 val)
228 {
229 u32 *sim;
230 struct aer_error *err;
231 unsigned long flags;
232 int rw1cs;
233 struct pci_ops *ops;
234 struct pci_ops *my_ops;
235 int domain;
236 int rv;
237
238 spin_lock_irqsave(&inject_lock, flags);
239 if (size != sizeof(u32))
240 goto out;
241 domain = pci_domain_nr(bus);
242 if (domain < 0)
243 goto out;
244 err = __find_aer_error(domain, bus->number, devfn);
245 if (!err)
246 goto out;
247
248 sim = find_pci_config_dword(err, where, &rw1cs);
249 if (sim) {
250 if (rw1cs)
251 *sim ^= val;
252 else
253 *sim = val;
254 spin_unlock_irqrestore(&inject_lock, flags);
255 return 0;
256 }
257 out:
258 ops = __find_pci_bus_ops(bus);
259 /*
260 * pci_lock must already be held, so we can directly
261 * manipulate bus->ops. Many config access functions,
262 * including pci_generic_config_write() require the original
263 * bus->ops be installed to function, so temporarily put them
264 * back.
265 */
266 my_ops = bus->ops;
267 bus->ops = ops;
268 rv = ops->write(bus, devfn, where, size, val);
269 bus->ops = my_ops;
270 spin_unlock_irqrestore(&inject_lock, flags);
271 return rv;
272 }
273
274 static struct pci_ops aer_inj_pci_ops = {
275 .read = aer_inj_read_config,
276 .write = aer_inj_write_config,
277 };
278
pci_bus_ops_init(struct pci_bus_ops * bus_ops,struct pci_bus * bus,struct pci_ops * ops)279 static void pci_bus_ops_init(struct pci_bus_ops *bus_ops,
280 struct pci_bus *bus,
281 struct pci_ops *ops)
282 {
283 INIT_LIST_HEAD(&bus_ops->list);
284 bus_ops->bus = bus;
285 bus_ops->ops = ops;
286 }
287
pci_bus_set_aer_ops(struct pci_bus * bus)288 static int pci_bus_set_aer_ops(struct pci_bus *bus)
289 {
290 struct pci_ops *ops;
291 struct pci_bus_ops *bus_ops;
292 unsigned long flags;
293
294 bus_ops = kmalloc(sizeof(*bus_ops), GFP_KERNEL);
295 if (!bus_ops)
296 return -ENOMEM;
297 ops = pci_bus_set_ops(bus, &aer_inj_pci_ops);
298 spin_lock_irqsave(&inject_lock, flags);
299 if (ops == &aer_inj_pci_ops)
300 goto out;
301 pci_bus_ops_init(bus_ops, bus, ops);
302 list_add(&bus_ops->list, &pci_bus_ops_list);
303 bus_ops = NULL;
304 out:
305 spin_unlock_irqrestore(&inject_lock, flags);
306 kfree(bus_ops);
307 return 0;
308 }
309
find_aer_device_iter(struct device * device,void * data)310 static int find_aer_device_iter(struct device *device, void *data)
311 {
312 struct pcie_device **result = data;
313 struct pcie_device *pcie_dev;
314
315 if (device->bus == &pcie_port_bus_type) {
316 pcie_dev = to_pcie_device(device);
317 if (pcie_dev->service & PCIE_PORT_SERVICE_AER) {
318 *result = pcie_dev;
319 return 1;
320 }
321 }
322 return 0;
323 }
324
find_aer_device(struct pci_dev * dev,struct pcie_device ** result)325 static int find_aer_device(struct pci_dev *dev, struct pcie_device **result)
326 {
327 return device_for_each_child(&dev->dev, result, find_aer_device_iter);
328 }
329
aer_inject(struct aer_error_inj * einj)330 static int aer_inject(struct aer_error_inj *einj)
331 {
332 struct aer_error *err, *rperr;
333 struct aer_error *err_alloc = NULL, *rperr_alloc = NULL;
334 struct pci_dev *dev, *rpdev;
335 struct pcie_device *edev;
336 unsigned long flags;
337 unsigned int devfn = PCI_DEVFN(einj->dev, einj->fn);
338 int pos_cap_err, rp_pos_cap_err;
339 u32 sever, cor_mask, uncor_mask, cor_mask_orig = 0, uncor_mask_orig = 0;
340 int ret = 0;
341
342 dev = pci_get_domain_bus_and_slot(einj->domain, einj->bus, devfn);
343 if (!dev)
344 return -ENODEV;
345 rpdev = pcie_find_root_port(dev);
346 if (!rpdev) {
347 dev_err(&dev->dev, "aer_inject: Root port not found\n");
348 ret = -ENODEV;
349 goto out_put;
350 }
351
352 pos_cap_err = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
353 if (!pos_cap_err) {
354 dev_err(&dev->dev, "aer_inject: Device doesn't support AER\n");
355 ret = -EPROTONOSUPPORT;
356 goto out_put;
357 }
358 pci_read_config_dword(dev, pos_cap_err + PCI_ERR_UNCOR_SEVER, &sever);
359 pci_read_config_dword(dev, pos_cap_err + PCI_ERR_COR_MASK, &cor_mask);
360 pci_read_config_dword(dev, pos_cap_err + PCI_ERR_UNCOR_MASK,
361 &uncor_mask);
362
363 rp_pos_cap_err = pci_find_ext_capability(rpdev, PCI_EXT_CAP_ID_ERR);
364 if (!rp_pos_cap_err) {
365 dev_err(&rpdev->dev,
366 "aer_inject: Root port doesn't support AER\n");
367 ret = -EPROTONOSUPPORT;
368 goto out_put;
369 }
370
371 err_alloc = kzalloc(sizeof(struct aer_error), GFP_KERNEL);
372 if (!err_alloc) {
373 ret = -ENOMEM;
374 goto out_put;
375 }
376 rperr_alloc = kzalloc(sizeof(struct aer_error), GFP_KERNEL);
377 if (!rperr_alloc) {
378 ret = -ENOMEM;
379 goto out_put;
380 }
381
382 if (aer_mask_override) {
383 cor_mask_orig = cor_mask;
384 cor_mask &= !(einj->cor_status);
385 pci_write_config_dword(dev, pos_cap_err + PCI_ERR_COR_MASK,
386 cor_mask);
387
388 uncor_mask_orig = uncor_mask;
389 uncor_mask &= !(einj->uncor_status);
390 pci_write_config_dword(dev, pos_cap_err + PCI_ERR_UNCOR_MASK,
391 uncor_mask);
392 }
393
394 spin_lock_irqsave(&inject_lock, flags);
395
396 err = __find_aer_error_by_dev(dev);
397 if (!err) {
398 err = err_alloc;
399 err_alloc = NULL;
400 aer_error_init(err, einj->domain, einj->bus, devfn,
401 pos_cap_err);
402 list_add(&err->list, &einjected);
403 }
404 err->uncor_status |= einj->uncor_status;
405 err->cor_status |= einj->cor_status;
406 err->header_log0 = einj->header_log0;
407 err->header_log1 = einj->header_log1;
408 err->header_log2 = einj->header_log2;
409 err->header_log3 = einj->header_log3;
410
411 if (!aer_mask_override && einj->cor_status &&
412 !(einj->cor_status & ~cor_mask)) {
413 ret = -EINVAL;
414 dev_warn(&dev->dev,
415 "aer_inject: The correctable error(s) is masked by device\n");
416 spin_unlock_irqrestore(&inject_lock, flags);
417 goto out_put;
418 }
419 if (!aer_mask_override && einj->uncor_status &&
420 !(einj->uncor_status & ~uncor_mask)) {
421 ret = -EINVAL;
422 dev_warn(&dev->dev,
423 "aer_inject: The uncorrectable error(s) is masked by device\n");
424 spin_unlock_irqrestore(&inject_lock, flags);
425 goto out_put;
426 }
427
428 rperr = __find_aer_error_by_dev(rpdev);
429 if (!rperr) {
430 rperr = rperr_alloc;
431 rperr_alloc = NULL;
432 aer_error_init(rperr, pci_domain_nr(rpdev->bus),
433 rpdev->bus->number, rpdev->devfn,
434 rp_pos_cap_err);
435 list_add(&rperr->list, &einjected);
436 }
437 if (einj->cor_status) {
438 if (rperr->root_status & PCI_ERR_ROOT_COR_RCV)
439 rperr->root_status |= PCI_ERR_ROOT_MULTI_COR_RCV;
440 else
441 rperr->root_status |= PCI_ERR_ROOT_COR_RCV;
442 rperr->source_id &= 0xffff0000;
443 rperr->source_id |= (einj->bus << 8) | devfn;
444 }
445 if (einj->uncor_status) {
446 if (rperr->root_status & PCI_ERR_ROOT_UNCOR_RCV)
447 rperr->root_status |= PCI_ERR_ROOT_MULTI_UNCOR_RCV;
448 if (sever & einj->uncor_status) {
449 rperr->root_status |= PCI_ERR_ROOT_FATAL_RCV;
450 if (!(rperr->root_status & PCI_ERR_ROOT_UNCOR_RCV))
451 rperr->root_status |= PCI_ERR_ROOT_FIRST_FATAL;
452 } else
453 rperr->root_status |= PCI_ERR_ROOT_NONFATAL_RCV;
454 rperr->root_status |= PCI_ERR_ROOT_UNCOR_RCV;
455 rperr->source_id &= 0x0000ffff;
456 rperr->source_id |= ((einj->bus << 8) | devfn) << 16;
457 }
458 spin_unlock_irqrestore(&inject_lock, flags);
459
460 if (aer_mask_override) {
461 pci_write_config_dword(dev, pos_cap_err + PCI_ERR_COR_MASK,
462 cor_mask_orig);
463 pci_write_config_dword(dev, pos_cap_err + PCI_ERR_UNCOR_MASK,
464 uncor_mask_orig);
465 }
466
467 ret = pci_bus_set_aer_ops(dev->bus);
468 if (ret)
469 goto out_put;
470 ret = pci_bus_set_aer_ops(rpdev->bus);
471 if (ret)
472 goto out_put;
473
474 if (find_aer_device(rpdev, &edev)) {
475 if (!get_service_data(edev)) {
476 dev_warn(&edev->device,
477 "aer_inject: AER service is not initialized\n");
478 ret = -EPROTONOSUPPORT;
479 goto out_put;
480 }
481 dev_info(&edev->device,
482 "aer_inject: Injecting errors %08x/%08x into device %s\n",
483 einj->cor_status, einj->uncor_status, pci_name(dev));
484 aer_irq(-1, edev);
485 } else {
486 dev_err(&rpdev->dev, "aer_inject: AER device not found\n");
487 ret = -ENODEV;
488 }
489 out_put:
490 kfree(err_alloc);
491 kfree(rperr_alloc);
492 pci_dev_put(dev);
493 return ret;
494 }
495
aer_inject_write(struct file * filp,const char __user * ubuf,size_t usize,loff_t * off)496 static ssize_t aer_inject_write(struct file *filp, const char __user *ubuf,
497 size_t usize, loff_t *off)
498 {
499 struct aer_error_inj einj;
500 int ret;
501
502 if (!capable(CAP_SYS_ADMIN))
503 return -EPERM;
504 if (usize < offsetof(struct aer_error_inj, domain) ||
505 usize > sizeof(einj))
506 return -EINVAL;
507
508 memset(&einj, 0, sizeof(einj));
509 if (copy_from_user(&einj, ubuf, usize))
510 return -EFAULT;
511
512 ret = aer_inject(&einj);
513 return ret ? ret : usize;
514 }
515
516 static const struct file_operations aer_inject_fops = {
517 .write = aer_inject_write,
518 .owner = THIS_MODULE,
519 .llseek = noop_llseek,
520 };
521
522 static struct miscdevice aer_inject_device = {
523 .minor = MISC_DYNAMIC_MINOR,
524 .name = "aer_inject",
525 .fops = &aer_inject_fops,
526 };
527
aer_inject_init(void)528 static int __init aer_inject_init(void)
529 {
530 return misc_register(&aer_inject_device);
531 }
532
aer_inject_exit(void)533 static void __exit aer_inject_exit(void)
534 {
535 struct aer_error *err, *err_next;
536 unsigned long flags;
537 struct pci_bus_ops *bus_ops;
538
539 misc_deregister(&aer_inject_device);
540
541 while ((bus_ops = pci_bus_ops_pop())) {
542 pci_bus_set_ops(bus_ops->bus, bus_ops->ops);
543 kfree(bus_ops);
544 }
545
546 spin_lock_irqsave(&inject_lock, flags);
547 list_for_each_entry_safe(err, err_next, &einjected, list) {
548 list_del(&err->list);
549 kfree(err);
550 }
551 spin_unlock_irqrestore(&inject_lock, flags);
552 }
553
554 module_init(aer_inject_init);
555 module_exit(aer_inject_exit);
556
557 MODULE_DESCRIPTION("PCIe AER software error injector");
558 MODULE_LICENSE("GPL");
559