Commit | Line | Data |
---|---|---|
956a9202 RW |
1 | /* |
2 | * Xen PCI Frontend. | |
3 | * | |
4 | * Author: Ryan Wilson <hap9@epoch.ncsc.mil> | |
5 | */ | |
6 | #include <linux/module.h> | |
7 | #include <linux/init.h> | |
8 | #include <linux/mm.h> | |
9 | #include <xen/xenbus.h> | |
10 | #include <xen/events.h> | |
11 | #include <xen/grant_table.h> | |
12 | #include <xen/page.h> | |
13 | #include <linux/spinlock.h> | |
14 | #include <linux/pci.h> | |
15 | #include <linux/msi.h> | |
16 | #include <xen/xenbus.h> | |
17 | #include <xen/interface/io/pciif.h> | |
18 | #include <asm/xen/pci.h> | |
19 | #include <linux/interrupt.h> | |
20 | #include <asm/atomic.h> | |
21 | #include <linux/workqueue.h> | |
22 | #include <linux/bitops.h> | |
23 | #include <linux/time.h> | |
24 | ||
25 | #define INVALID_GRANT_REF (0) | |
26 | #define INVALID_EVTCHN (-1) | |
27 | ||
28 | struct pci_bus_entry { | |
29 | struct list_head list; | |
30 | struct pci_bus *bus; | |
31 | }; | |
32 | ||
33 | #define _PDEVB_op_active (0) | |
34 | #define PDEVB_op_active (1 << (_PDEVB_op_active)) | |
35 | ||
36 | struct pcifront_device { | |
37 | struct xenbus_device *xdev; | |
38 | struct list_head root_buses; | |
39 | ||
40 | int evtchn; | |
41 | int gnt_ref; | |
42 | ||
43 | int irq; | |
44 | ||
45 | /* Lock this when doing any operations in sh_info */ | |
46 | spinlock_t sh_info_lock; | |
47 | struct xen_pci_sharedinfo *sh_info; | |
48 | struct work_struct op_work; | |
49 | unsigned long flags; | |
50 | ||
51 | }; | |
52 | ||
53 | struct pcifront_sd { | |
54 | int domain; | |
55 | struct pcifront_device *pdev; | |
56 | }; | |
57 | ||
58 | static inline struct pcifront_device * | |
59 | pcifront_get_pdev(struct pcifront_sd *sd) | |
60 | { | |
61 | return sd->pdev; | |
62 | } | |
63 | ||
64 | static inline void pcifront_init_sd(struct pcifront_sd *sd, | |
65 | unsigned int domain, unsigned int bus, | |
66 | struct pcifront_device *pdev) | |
67 | { | |
68 | sd->domain = domain; | |
69 | sd->pdev = pdev; | |
70 | } | |
71 | ||
72 | static DEFINE_SPINLOCK(pcifront_dev_lock); | |
73 | static struct pcifront_device *pcifront_dev; | |
74 | ||
75 | static int verbose_request; | |
76 | module_param(verbose_request, int, 0644); | |
77 | ||
78 | static int errno_to_pcibios_err(int errno) | |
79 | { | |
80 | switch (errno) { | |
81 | case XEN_PCI_ERR_success: | |
82 | return PCIBIOS_SUCCESSFUL; | |
83 | ||
84 | case XEN_PCI_ERR_dev_not_found: | |
85 | return PCIBIOS_DEVICE_NOT_FOUND; | |
86 | ||
87 | case XEN_PCI_ERR_invalid_offset: | |
88 | case XEN_PCI_ERR_op_failed: | |
89 | return PCIBIOS_BAD_REGISTER_NUMBER; | |
90 | ||
91 | case XEN_PCI_ERR_not_implemented: | |
92 | return PCIBIOS_FUNC_NOT_SUPPORTED; | |
93 | ||
94 | case XEN_PCI_ERR_access_denied: | |
95 | return PCIBIOS_SET_FAILED; | |
96 | } | |
97 | return errno; | |
98 | } | |
99 | ||
100 | static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev) | |
101 | { | |
102 | if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags) | |
103 | && !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) { | |
104 | dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n"); | |
105 | schedule_work(&pdev->op_work); | |
106 | } | |
107 | } | |
108 | ||
109 | static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op) | |
110 | { | |
111 | int err = 0; | |
112 | struct xen_pci_op *active_op = &pdev->sh_info->op; | |
113 | unsigned long irq_flags; | |
114 | evtchn_port_t port = pdev->evtchn; | |
115 | unsigned irq = pdev->irq; | |
116 | s64 ns, ns_timeout; | |
117 | struct timeval tv; | |
118 | ||
119 | spin_lock_irqsave(&pdev->sh_info_lock, irq_flags); | |
120 | ||
121 | memcpy(active_op, op, sizeof(struct xen_pci_op)); | |
122 | ||
123 | /* Go */ | |
124 | wmb(); | |
125 | set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags); | |
126 | notify_remote_via_evtchn(port); | |
127 | ||
128 | /* | |
129 | * We set a poll timeout of 3 seconds but give up on return after | |
130 | * 2 seconds. It is better to time out too late rather than too early | |
131 | * (in the latter case we end up continually re-executing poll() with a | |
132 | * timeout in the past). 1s difference gives plenty of slack for error. | |
133 | */ | |
134 | do_gettimeofday(&tv); | |
135 | ns_timeout = timeval_to_ns(&tv) + 2 * (s64)NSEC_PER_SEC; | |
136 | ||
137 | xen_clear_irq_pending(irq); | |
138 | ||
139 | while (test_bit(_XEN_PCIF_active, | |
140 | (unsigned long *)&pdev->sh_info->flags)) { | |
141 | xen_poll_irq_timeout(irq, jiffies + 3*HZ); | |
142 | xen_clear_irq_pending(irq); | |
143 | do_gettimeofday(&tv); | |
144 | ns = timeval_to_ns(&tv); | |
145 | if (ns > ns_timeout) { | |
146 | dev_err(&pdev->xdev->dev, | |
147 | "pciback not responding!!!\n"); | |
148 | clear_bit(_XEN_PCIF_active, | |
149 | (unsigned long *)&pdev->sh_info->flags); | |
150 | err = XEN_PCI_ERR_dev_not_found; | |
151 | goto out; | |
152 | } | |
153 | } | |
154 | ||
155 | /* | |
156 | * We might lose backend service request since we | |
157 | * reuse same evtchn with pci_conf backend response. So re-schedule | |
158 | * aer pcifront service. | |
159 | */ | |
160 | if (test_bit(_XEN_PCIB_active, | |
161 | (unsigned long *)&pdev->sh_info->flags)) { | |
162 | dev_err(&pdev->xdev->dev, | |
163 | "schedule aer pcifront service\n"); | |
164 | schedule_pcifront_aer_op(pdev); | |
165 | } | |
166 | ||
167 | memcpy(op, active_op, sizeof(struct xen_pci_op)); | |
168 | ||
169 | err = op->err; | |
170 | out: | |
171 | spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags); | |
172 | return err; | |
173 | } | |
174 | ||
175 | /* Access to this function is spinlocked in drivers/pci/access.c */ | |
176 | static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn, | |
177 | int where, int size, u32 *val) | |
178 | { | |
179 | int err = 0; | |
180 | struct xen_pci_op op = { | |
181 | .cmd = XEN_PCI_OP_conf_read, | |
182 | .domain = pci_domain_nr(bus), | |
183 | .bus = bus->number, | |
184 | .devfn = devfn, | |
185 | .offset = where, | |
186 | .size = size, | |
187 | }; | |
188 | struct pcifront_sd *sd = bus->sysdata; | |
189 | struct pcifront_device *pdev = pcifront_get_pdev(sd); | |
190 | ||
191 | if (verbose_request) | |
192 | dev_info(&pdev->xdev->dev, | |
193 | "read dev=%04x:%02x:%02x.%01x - offset %x size %d\n", | |
194 | pci_domain_nr(bus), bus->number, PCI_SLOT(devfn), | |
195 | PCI_FUNC(devfn), where, size); | |
196 | ||
197 | err = do_pci_op(pdev, &op); | |
198 | ||
199 | if (likely(!err)) { | |
200 | if (verbose_request) | |
201 | dev_info(&pdev->xdev->dev, "read got back value %x\n", | |
202 | op.value); | |
203 | ||
204 | *val = op.value; | |
205 | } else if (err == -ENODEV) { | |
206 | /* No device here, pretend that it just returned 0 */ | |
207 | err = 0; | |
208 | *val = 0; | |
209 | } | |
210 | ||
211 | return errno_to_pcibios_err(err); | |
212 | } | |
213 | ||
214 | /* Access to this function is spinlocked in drivers/pci/access.c */ | |
215 | static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn, | |
216 | int where, int size, u32 val) | |
217 | { | |
218 | struct xen_pci_op op = { | |
219 | .cmd = XEN_PCI_OP_conf_write, | |
220 | .domain = pci_domain_nr(bus), | |
221 | .bus = bus->number, | |
222 | .devfn = devfn, | |
223 | .offset = where, | |
224 | .size = size, | |
225 | .value = val, | |
226 | }; | |
227 | struct pcifront_sd *sd = bus->sysdata; | |
228 | struct pcifront_device *pdev = pcifront_get_pdev(sd); | |
229 | ||
230 | if (verbose_request) | |
231 | dev_info(&pdev->xdev->dev, | |
232 | "write dev=%04x:%02x:%02x.%01x - " | |
233 | "offset %x size %d val %x\n", | |
234 | pci_domain_nr(bus), bus->number, | |
235 | PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val); | |
236 | ||
237 | return errno_to_pcibios_err(do_pci_op(pdev, &op)); | |
238 | } | |
239 | ||
240 | struct pci_ops pcifront_bus_ops = { | |
241 | .read = pcifront_bus_read, | |
242 | .write = pcifront_bus_write, | |
243 | }; | |
244 | ||
245 | #ifdef CONFIG_PCI_MSI | |
246 | static int pci_frontend_enable_msix(struct pci_dev *dev, | |
247 | int **vector, int nvec) | |
248 | { | |
249 | int err; | |
250 | int i; | |
251 | struct xen_pci_op op = { | |
252 | .cmd = XEN_PCI_OP_enable_msix, | |
253 | .domain = pci_domain_nr(dev->bus), | |
254 | .bus = dev->bus->number, | |
255 | .devfn = dev->devfn, | |
256 | .value = nvec, | |
257 | }; | |
258 | struct pcifront_sd *sd = dev->bus->sysdata; | |
259 | struct pcifront_device *pdev = pcifront_get_pdev(sd); | |
260 | struct msi_desc *entry; | |
261 | ||
262 | if (nvec > SH_INFO_MAX_VEC) { | |
263 | dev_err(&dev->dev, "too much vector for pci frontend: %x." | |
264 | " Increase SH_INFO_MAX_VEC.\n", nvec); | |
265 | return -EINVAL; | |
266 | } | |
267 | ||
268 | i = 0; | |
269 | list_for_each_entry(entry, &dev->msi_list, list) { | |
270 | op.msix_entries[i].entry = entry->msi_attrib.entry_nr; | |
271 | /* Vector is useless at this point. */ | |
272 | op.msix_entries[i].vector = -1; | |
273 | i++; | |
274 | } | |
275 | ||
276 | err = do_pci_op(pdev, &op); | |
277 | ||
278 | if (likely(!err)) { | |
279 | if (likely(!op.value)) { | |
280 | /* we get the result */ | |
281 | for (i = 0; i < nvec; i++) | |
282 | *(*vector+i) = op.msix_entries[i].vector; | |
283 | return 0; | |
284 | } else { | |
285 | printk(KERN_DEBUG "enable msix get value %x\n", | |
286 | op.value); | |
287 | return op.value; | |
288 | } | |
289 | } else { | |
290 | dev_err(&dev->dev, "enable msix get err %x\n", err); | |
291 | return err; | |
292 | } | |
293 | } | |
294 | ||
295 | static void pci_frontend_disable_msix(struct pci_dev *dev) | |
296 | { | |
297 | int err; | |
298 | struct xen_pci_op op = { | |
299 | .cmd = XEN_PCI_OP_disable_msix, | |
300 | .domain = pci_domain_nr(dev->bus), | |
301 | .bus = dev->bus->number, | |
302 | .devfn = dev->devfn, | |
303 | }; | |
304 | struct pcifront_sd *sd = dev->bus->sysdata; | |
305 | struct pcifront_device *pdev = pcifront_get_pdev(sd); | |
306 | ||
307 | err = do_pci_op(pdev, &op); | |
308 | ||
309 | /* What should do for error ? */ | |
310 | if (err) | |
311 | dev_err(&dev->dev, "pci_disable_msix get err %x\n", err); | |
312 | } | |
313 | ||
314 | static int pci_frontend_enable_msi(struct pci_dev *dev, int **vector) | |
315 | { | |
316 | int err; | |
317 | struct xen_pci_op op = { | |
318 | .cmd = XEN_PCI_OP_enable_msi, | |
319 | .domain = pci_domain_nr(dev->bus), | |
320 | .bus = dev->bus->number, | |
321 | .devfn = dev->devfn, | |
322 | }; | |
323 | struct pcifront_sd *sd = dev->bus->sysdata; | |
324 | struct pcifront_device *pdev = pcifront_get_pdev(sd); | |
325 | ||
326 | err = do_pci_op(pdev, &op); | |
327 | if (likely(!err)) { | |
328 | *(*vector) = op.value; | |
329 | } else { | |
330 | dev_err(&dev->dev, "pci frontend enable msi failed for dev " | |
331 | "%x:%x\n", op.bus, op.devfn); | |
332 | err = -EINVAL; | |
333 | } | |
334 | return err; | |
335 | } | |
336 | ||
337 | static void pci_frontend_disable_msi(struct pci_dev *dev) | |
338 | { | |
339 | int err; | |
340 | struct xen_pci_op op = { | |
341 | .cmd = XEN_PCI_OP_disable_msi, | |
342 | .domain = pci_domain_nr(dev->bus), | |
343 | .bus = dev->bus->number, | |
344 | .devfn = dev->devfn, | |
345 | }; | |
346 | struct pcifront_sd *sd = dev->bus->sysdata; | |
347 | struct pcifront_device *pdev = pcifront_get_pdev(sd); | |
348 | ||
349 | err = do_pci_op(pdev, &op); | |
350 | if (err == XEN_PCI_ERR_dev_not_found) { | |
351 | /* XXX No response from backend, what shall we do? */ | |
352 | printk(KERN_DEBUG "get no response from backend for disable MSI\n"); | |
353 | return; | |
354 | } | |
355 | if (err) | |
356 | /* how can pciback notify us fail? */ | |
357 | printk(KERN_DEBUG "get fake response frombackend\n"); | |
358 | } | |
359 | ||
360 | static struct xen_pci_frontend_ops pci_frontend_ops = { | |
361 | .enable_msi = pci_frontend_enable_msi, | |
362 | .disable_msi = pci_frontend_disable_msi, | |
363 | .enable_msix = pci_frontend_enable_msix, | |
364 | .disable_msix = pci_frontend_disable_msix, | |
365 | }; | |
366 | ||
367 | static void pci_frontend_registrar(int enable) | |
368 | { | |
369 | if (enable) | |
370 | xen_pci_frontend = &pci_frontend_ops; | |
371 | else | |
372 | xen_pci_frontend = NULL; | |
373 | }; | |
374 | #else | |
375 | static inline void pci_frontend_registrar(int enable) { }; | |
376 | #endif /* CONFIG_PCI_MSI */ | |
377 | ||
378 | /* Claim resources for the PCI frontend as-is, backend won't allow changes */ | |
379 | static int pcifront_claim_resource(struct pci_dev *dev, void *data) | |
380 | { | |
381 | struct pcifront_device *pdev = data; | |
382 | int i; | |
383 | struct resource *r; | |
384 | ||
385 | for (i = 0; i < PCI_NUM_RESOURCES; i++) { | |
386 | r = &dev->resource[i]; | |
387 | ||
388 | if (!r->parent && r->start && r->flags) { | |
389 | dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n", | |
390 | pci_name(dev), i); | |
391 | if (pci_claim_resource(dev, i)) { | |
392 | dev_err(&pdev->xdev->dev, "Could not claim " | |
393 | "resource %s/%d! Device offline. Try " | |
394 | "giving less than 4GB to domain.\n", | |
395 | pci_name(dev), i); | |
396 | } | |
397 | } | |
398 | } | |
399 | ||
400 | return 0; | |
401 | } | |
402 | ||
403 | static int __devinit pcifront_scan_bus(struct pcifront_device *pdev, | |
404 | unsigned int domain, unsigned int bus, | |
405 | struct pci_bus *b) | |
406 | { | |
407 | struct pci_dev *d; | |
408 | unsigned int devfn; | |
409 | ||
410 | /* Scan the bus for functions and add. | |
411 | * We omit handling of PCI bridge attachment because pciback prevents | |
412 | * bridges from being exported. | |
413 | */ | |
414 | for (devfn = 0; devfn < 0x100; devfn++) { | |
415 | d = pci_get_slot(b, devfn); | |
416 | if (d) { | |
417 | /* Device is already known. */ | |
418 | pci_dev_put(d); | |
419 | continue; | |
420 | } | |
421 | ||
422 | d = pci_scan_single_device(b, devfn); | |
423 | if (d) | |
424 | dev_info(&pdev->xdev->dev, "New device on " | |
425 | "%04x:%02x:%02x.%02x found.\n", domain, bus, | |
426 | PCI_SLOT(devfn), PCI_FUNC(devfn)); | |
427 | } | |
428 | ||
429 | return 0; | |
430 | } | |
431 | ||
432 | static int __devinit pcifront_scan_root(struct pcifront_device *pdev, | |
433 | unsigned int domain, unsigned int bus) | |
434 | { | |
435 | struct pci_bus *b; | |
436 | struct pcifront_sd *sd = NULL; | |
437 | struct pci_bus_entry *bus_entry = NULL; | |
438 | int err = 0; | |
439 | ||
440 | #ifndef CONFIG_PCI_DOMAINS | |
441 | if (domain != 0) { | |
442 | dev_err(&pdev->xdev->dev, | |
443 | "PCI Root in non-zero PCI Domain! domain=%d\n", domain); | |
444 | dev_err(&pdev->xdev->dev, | |
445 | "Please compile with CONFIG_PCI_DOMAINS\n"); | |
446 | err = -EINVAL; | |
447 | goto err_out; | |
448 | } | |
449 | #endif | |
450 | ||
451 | dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n", | |
452 | domain, bus); | |
453 | ||
454 | bus_entry = kmalloc(sizeof(*bus_entry), GFP_KERNEL); | |
455 | sd = kmalloc(sizeof(*sd), GFP_KERNEL); | |
456 | if (!bus_entry || !sd) { | |
457 | err = -ENOMEM; | |
458 | goto err_out; | |
459 | } | |
460 | pcifront_init_sd(sd, domain, bus, pdev); | |
461 | ||
462 | b = pci_scan_bus_parented(&pdev->xdev->dev, bus, | |
463 | &pcifront_bus_ops, sd); | |
464 | if (!b) { | |
465 | dev_err(&pdev->xdev->dev, | |
466 | "Error creating PCI Frontend Bus!\n"); | |
467 | err = -ENOMEM; | |
468 | goto err_out; | |
469 | } | |
470 | ||
471 | bus_entry->bus = b; | |
472 | ||
473 | list_add(&bus_entry->list, &pdev->root_buses); | |
474 | ||
475 | /* pci_scan_bus_parented skips devices which do not have a have | |
476 | * devfn==0. The pcifront_scan_bus enumerates all devfn. */ | |
477 | err = pcifront_scan_bus(pdev, domain, bus, b); | |
478 | ||
479 | /* Claim resources before going "live" with our devices */ | |
480 | pci_walk_bus(b, pcifront_claim_resource, pdev); | |
481 | ||
482 | /* Create SysFS and notify udev of the devices. Aka: "going live" */ | |
483 | pci_bus_add_devices(b); | |
484 | ||
485 | return err; | |
486 | ||
487 | err_out: | |
488 | kfree(bus_entry); | |
489 | kfree(sd); | |
490 | ||
491 | return err; | |
492 | } | |
493 | ||
494 | static int __devinit pcifront_rescan_root(struct pcifront_device *pdev, | |
495 | unsigned int domain, unsigned int bus) | |
496 | { | |
497 | int err; | |
498 | struct pci_bus *b; | |
499 | ||
500 | #ifndef CONFIG_PCI_DOMAINS | |
501 | if (domain != 0) { | |
502 | dev_err(&pdev->xdev->dev, | |
503 | "PCI Root in non-zero PCI Domain! domain=%d\n", domain); | |
504 | dev_err(&pdev->xdev->dev, | |
505 | "Please compile with CONFIG_PCI_DOMAINS\n"); | |
506 | return -EINVAL; | |
507 | } | |
508 | #endif | |
509 | ||
510 | dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n", | |
511 | domain, bus); | |
512 | ||
513 | b = pci_find_bus(domain, bus); | |
514 | if (!b) | |
515 | /* If the bus is unknown, create it. */ | |
516 | return pcifront_scan_root(pdev, domain, bus); | |
517 | ||
518 | err = pcifront_scan_bus(pdev, domain, bus, b); | |
519 | ||
520 | /* Claim resources before going "live" with our devices */ | |
521 | pci_walk_bus(b, pcifront_claim_resource, pdev); | |
522 | ||
523 | /* Create SysFS and notify udev of the devices. Aka: "going live" */ | |
524 | pci_bus_add_devices(b); | |
525 | ||
526 | return err; | |
527 | } | |
528 | ||
529 | static void free_root_bus_devs(struct pci_bus *bus) | |
530 | { | |
531 | struct pci_dev *dev; | |
532 | ||
533 | while (!list_empty(&bus->devices)) { | |
534 | dev = container_of(bus->devices.next, struct pci_dev, | |
535 | bus_list); | |
536 | dev_dbg(&dev->dev, "removing device\n"); | |
537 | pci_remove_bus_device(dev); | |
538 | } | |
539 | } | |
540 | ||
541 | static void pcifront_free_roots(struct pcifront_device *pdev) | |
542 | { | |
543 | struct pci_bus_entry *bus_entry, *t; | |
544 | ||
545 | dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n"); | |
546 | ||
547 | list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) { | |
548 | list_del(&bus_entry->list); | |
549 | ||
550 | free_root_bus_devs(bus_entry->bus); | |
551 | ||
552 | kfree(bus_entry->bus->sysdata); | |
553 | ||
554 | device_unregister(bus_entry->bus->bridge); | |
555 | pci_remove_bus(bus_entry->bus); | |
556 | ||
557 | kfree(bus_entry); | |
558 | } | |
559 | } | |
560 | ||
561 | static pci_ers_result_t pcifront_common_process(int cmd, | |
562 | struct pcifront_device *pdev, | |
563 | pci_channel_state_t state) | |
564 | { | |
565 | pci_ers_result_t result; | |
566 | struct pci_driver *pdrv; | |
567 | int bus = pdev->sh_info->aer_op.bus; | |
568 | int devfn = pdev->sh_info->aer_op.devfn; | |
569 | struct pci_dev *pcidev; | |
570 | int flag = 0; | |
571 | ||
572 | dev_dbg(&pdev->xdev->dev, | |
573 | "pcifront AER process: cmd %x (bus:%x, devfn%x)", | |
574 | cmd, bus, devfn); | |
575 | result = PCI_ERS_RESULT_NONE; | |
576 | ||
577 | pcidev = pci_get_bus_and_slot(bus, devfn); | |
578 | if (!pcidev || !pcidev->driver) { | |
579 | dev_err(&pcidev->dev, | |
580 | "device or driver is NULL\n"); | |
581 | return result; | |
582 | } | |
583 | pdrv = pcidev->driver; | |
584 | ||
585 | if (get_driver(&pdrv->driver)) { | |
586 | if (pdrv->err_handler && pdrv->err_handler->error_detected) { | |
587 | dev_dbg(&pcidev->dev, | |
588 | "trying to call AER service\n"); | |
589 | if (pcidev) { | |
590 | flag = 1; | |
591 | switch (cmd) { | |
592 | case XEN_PCI_OP_aer_detected: | |
593 | result = pdrv->err_handler-> | |
594 | error_detected(pcidev, state); | |
595 | break; | |
596 | case XEN_PCI_OP_aer_mmio: | |
597 | result = pdrv->err_handler-> | |
598 | mmio_enabled(pcidev); | |
599 | break; | |
600 | case XEN_PCI_OP_aer_slotreset: | |
601 | result = pdrv->err_handler-> | |
602 | slot_reset(pcidev); | |
603 | break; | |
604 | case XEN_PCI_OP_aer_resume: | |
605 | pdrv->err_handler->resume(pcidev); | |
606 | break; | |
607 | default: | |
608 | dev_err(&pdev->xdev->dev, | |
609 | "bad request in aer recovery " | |
610 | "operation!\n"); | |
611 | ||
612 | } | |
613 | } | |
614 | } | |
615 | put_driver(&pdrv->driver); | |
616 | } | |
617 | if (!flag) | |
618 | result = PCI_ERS_RESULT_NONE; | |
619 | ||
620 | return result; | |
621 | } | |
622 | ||
623 | ||
624 | static void pcifront_do_aer(struct work_struct *data) | |
625 | { | |
626 | struct pcifront_device *pdev = | |
627 | container_of(data, struct pcifront_device, op_work); | |
628 | int cmd = pdev->sh_info->aer_op.cmd; | |
629 | pci_channel_state_t state = | |
630 | (pci_channel_state_t)pdev->sh_info->aer_op.err; | |
631 | ||
632 | /*If a pci_conf op is in progress, | |
633 | we have to wait until it is done before service aer op*/ | |
634 | dev_dbg(&pdev->xdev->dev, | |
635 | "pcifront service aer bus %x devfn %x\n", | |
636 | pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn); | |
637 | ||
638 | pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state); | |
639 | ||
640 | /* Post the operation to the guest. */ | |
641 | wmb(); | |
642 | clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags); | |
643 | notify_remote_via_evtchn(pdev->evtchn); | |
644 | ||
645 | /*in case of we lost an aer request in four lines time_window*/ | |
646 | smp_mb__before_clear_bit(); | |
647 | clear_bit(_PDEVB_op_active, &pdev->flags); | |
648 | smp_mb__after_clear_bit(); | |
649 | ||
650 | schedule_pcifront_aer_op(pdev); | |
651 | ||
652 | } | |
653 | ||
654 | static irqreturn_t pcifront_handler_aer(int irq, void *dev) | |
655 | { | |
656 | struct pcifront_device *pdev = dev; | |
657 | schedule_pcifront_aer_op(pdev); | |
658 | return IRQ_HANDLED; | |
659 | } | |
660 | static int pcifront_connect(struct pcifront_device *pdev) | |
661 | { | |
662 | int err = 0; | |
663 | ||
664 | spin_lock(&pcifront_dev_lock); | |
665 | ||
666 | if (!pcifront_dev) { | |
667 | dev_info(&pdev->xdev->dev, "Installing PCI frontend\n"); | |
668 | pcifront_dev = pdev; | |
669 | } else { | |
670 | dev_err(&pdev->xdev->dev, "PCI frontend already installed!\n"); | |
671 | err = -EEXIST; | |
672 | } | |
673 | ||
674 | spin_unlock(&pcifront_dev_lock); | |
675 | ||
676 | return err; | |
677 | } | |
678 | ||
679 | static void pcifront_disconnect(struct pcifront_device *pdev) | |
680 | { | |
681 | spin_lock(&pcifront_dev_lock); | |
682 | ||
683 | if (pdev == pcifront_dev) { | |
684 | dev_info(&pdev->xdev->dev, | |
685 | "Disconnecting PCI Frontend Buses\n"); | |
686 | pcifront_dev = NULL; | |
687 | } | |
688 | ||
689 | spin_unlock(&pcifront_dev_lock); | |
690 | } | |
691 | static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev) | |
692 | { | |
693 | struct pcifront_device *pdev; | |
694 | ||
695 | pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL); | |
696 | if (pdev == NULL) | |
697 | goto out; | |
698 | ||
699 | pdev->sh_info = | |
700 | (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL); | |
701 | if (pdev->sh_info == NULL) { | |
702 | kfree(pdev); | |
703 | pdev = NULL; | |
704 | goto out; | |
705 | } | |
706 | pdev->sh_info->flags = 0; | |
707 | ||
708 | /*Flag for registering PV AER handler*/ | |
709 | set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags); | |
710 | ||
711 | dev_set_drvdata(&xdev->dev, pdev); | |
712 | pdev->xdev = xdev; | |
713 | ||
714 | INIT_LIST_HEAD(&pdev->root_buses); | |
715 | ||
716 | spin_lock_init(&pdev->sh_info_lock); | |
717 | ||
718 | pdev->evtchn = INVALID_EVTCHN; | |
719 | pdev->gnt_ref = INVALID_GRANT_REF; | |
720 | pdev->irq = -1; | |
721 | ||
722 | INIT_WORK(&pdev->op_work, pcifront_do_aer); | |
723 | ||
724 | dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n", | |
725 | pdev, pdev->sh_info); | |
726 | out: | |
727 | return pdev; | |
728 | } | |
729 | ||
730 | static void free_pdev(struct pcifront_device *pdev) | |
731 | { | |
732 | dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev); | |
733 | ||
734 | pcifront_free_roots(pdev); | |
735 | ||
736 | /*For PCIE_AER error handling job*/ | |
737 | flush_scheduled_work(); | |
738 | ||
739 | if (pdev->irq >= 0) | |
740 | unbind_from_irqhandler(pdev->irq, pdev); | |
741 | ||
742 | if (pdev->evtchn != INVALID_EVTCHN) | |
743 | xenbus_free_evtchn(pdev->xdev, pdev->evtchn); | |
744 | ||
745 | if (pdev->gnt_ref != INVALID_GRANT_REF) | |
746 | gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */, | |
747 | (unsigned long)pdev->sh_info); | |
748 | else | |
749 | free_page((unsigned long)pdev->sh_info); | |
750 | ||
751 | dev_set_drvdata(&pdev->xdev->dev, NULL); | |
752 | ||
753 | kfree(pdev); | |
754 | } | |
755 | ||
756 | static int pcifront_publish_info(struct pcifront_device *pdev) | |
757 | { | |
758 | int err = 0; | |
759 | struct xenbus_transaction trans; | |
760 | ||
761 | err = xenbus_grant_ring(pdev->xdev, virt_to_mfn(pdev->sh_info)); | |
762 | if (err < 0) | |
763 | goto out; | |
764 | ||
765 | pdev->gnt_ref = err; | |
766 | ||
767 | err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn); | |
768 | if (err) | |
769 | goto out; | |
770 | ||
771 | err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer, | |
772 | 0, "pcifront", pdev); | |
773 | ||
774 | if (err < 0) | |
775 | return err; | |
776 | ||
777 | pdev->irq = err; | |
778 | ||
779 | do_publish: | |
780 | err = xenbus_transaction_start(&trans); | |
781 | if (err) { | |
782 | xenbus_dev_fatal(pdev->xdev, err, | |
783 | "Error writing configuration for backend " | |
784 | "(start transaction)"); | |
785 | goto out; | |
786 | } | |
787 | ||
788 | err = xenbus_printf(trans, pdev->xdev->nodename, | |
789 | "pci-op-ref", "%u", pdev->gnt_ref); | |
790 | if (!err) | |
791 | err = xenbus_printf(trans, pdev->xdev->nodename, | |
792 | "event-channel", "%u", pdev->evtchn); | |
793 | if (!err) | |
794 | err = xenbus_printf(trans, pdev->xdev->nodename, | |
795 | "magic", XEN_PCI_MAGIC); | |
796 | ||
797 | if (err) { | |
798 | xenbus_transaction_end(trans, 1); | |
799 | xenbus_dev_fatal(pdev->xdev, err, | |
800 | "Error writing configuration for backend"); | |
801 | goto out; | |
802 | } else { | |
803 | err = xenbus_transaction_end(trans, 0); | |
804 | if (err == -EAGAIN) | |
805 | goto do_publish; | |
806 | else if (err) { | |
807 | xenbus_dev_fatal(pdev->xdev, err, | |
808 | "Error completing transaction " | |
809 | "for backend"); | |
810 | goto out; | |
811 | } | |
812 | } | |
813 | ||
814 | xenbus_switch_state(pdev->xdev, XenbusStateInitialised); | |
815 | ||
816 | dev_dbg(&pdev->xdev->dev, "publishing successful!\n"); | |
817 | ||
818 | out: | |
819 | return err; | |
820 | } | |
821 | ||
822 | static int __devinit pcifront_try_connect(struct pcifront_device *pdev) | |
823 | { | |
824 | int err = -EFAULT; | |
825 | int i, num_roots, len; | |
826 | char str[64]; | |
827 | unsigned int domain, bus; | |
828 | ||
829 | ||
830 | /* Only connect once */ | |
831 | if (xenbus_read_driver_state(pdev->xdev->nodename) != | |
832 | XenbusStateInitialised) | |
833 | goto out; | |
834 | ||
835 | err = pcifront_connect(pdev); | |
836 | if (err) { | |
837 | xenbus_dev_fatal(pdev->xdev, err, | |
838 | "Error connecting PCI Frontend"); | |
839 | goto out; | |
840 | } | |
841 | ||
842 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, | |
843 | "root_num", "%d", &num_roots); | |
844 | if (err == -ENOENT) { | |
845 | xenbus_dev_error(pdev->xdev, err, | |
846 | "No PCI Roots found, trying 0000:00"); | |
847 | err = pcifront_scan_root(pdev, 0, 0); | |
848 | num_roots = 0; | |
849 | } else if (err != 1) { | |
850 | if (err == 0) | |
851 | err = -EINVAL; | |
852 | xenbus_dev_fatal(pdev->xdev, err, | |
853 | "Error reading number of PCI roots"); | |
854 | goto out; | |
855 | } | |
856 | ||
857 | for (i = 0; i < num_roots; i++) { | |
858 | len = snprintf(str, sizeof(str), "root-%d", i); | |
859 | if (unlikely(len >= (sizeof(str) - 1))) { | |
860 | err = -ENOMEM; | |
861 | goto out; | |
862 | } | |
863 | ||
864 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, | |
865 | "%x:%x", &domain, &bus); | |
866 | if (err != 2) { | |
867 | if (err >= 0) | |
868 | err = -EINVAL; | |
869 | xenbus_dev_fatal(pdev->xdev, err, | |
870 | "Error reading PCI root %d", i); | |
871 | goto out; | |
872 | } | |
873 | ||
874 | err = pcifront_scan_root(pdev, domain, bus); | |
875 | if (err) { | |
876 | xenbus_dev_fatal(pdev->xdev, err, | |
877 | "Error scanning PCI root %04x:%02x", | |
878 | domain, bus); | |
879 | goto out; | |
880 | } | |
881 | } | |
882 | ||
883 | err = xenbus_switch_state(pdev->xdev, XenbusStateConnected); | |
884 | ||
885 | out: | |
886 | return err; | |
887 | } | |
888 | ||
889 | static int pcifront_try_disconnect(struct pcifront_device *pdev) | |
890 | { | |
891 | int err = 0; | |
892 | enum xenbus_state prev_state; | |
893 | ||
894 | ||
895 | prev_state = xenbus_read_driver_state(pdev->xdev->nodename); | |
896 | ||
897 | if (prev_state >= XenbusStateClosing) | |
898 | goto out; | |
899 | ||
900 | if (prev_state == XenbusStateConnected) { | |
901 | pcifront_free_roots(pdev); | |
902 | pcifront_disconnect(pdev); | |
903 | } | |
904 | ||
905 | err = xenbus_switch_state(pdev->xdev, XenbusStateClosed); | |
906 | ||
907 | out: | |
908 | ||
909 | return err; | |
910 | } | |
911 | ||
912 | static int __devinit pcifront_attach_devices(struct pcifront_device *pdev) | |
913 | { | |
914 | int err = -EFAULT; | |
915 | int i, num_roots, len; | |
916 | unsigned int domain, bus; | |
917 | char str[64]; | |
918 | ||
919 | if (xenbus_read_driver_state(pdev->xdev->nodename) != | |
920 | XenbusStateReconfiguring) | |
921 | goto out; | |
922 | ||
923 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, | |
924 | "root_num", "%d", &num_roots); | |
925 | if (err == -ENOENT) { | |
926 | xenbus_dev_error(pdev->xdev, err, | |
927 | "No PCI Roots found, trying 0000:00"); | |
928 | err = pcifront_rescan_root(pdev, 0, 0); | |
929 | num_roots = 0; | |
930 | } else if (err != 1) { | |
931 | if (err == 0) | |
932 | err = -EINVAL; | |
933 | xenbus_dev_fatal(pdev->xdev, err, | |
934 | "Error reading number of PCI roots"); | |
935 | goto out; | |
936 | } | |
937 | ||
938 | for (i = 0; i < num_roots; i++) { | |
939 | len = snprintf(str, sizeof(str), "root-%d", i); | |
940 | if (unlikely(len >= (sizeof(str) - 1))) { | |
941 | err = -ENOMEM; | |
942 | goto out; | |
943 | } | |
944 | ||
945 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, | |
946 | "%x:%x", &domain, &bus); | |
947 | if (err != 2) { | |
948 | if (err >= 0) | |
949 | err = -EINVAL; | |
950 | xenbus_dev_fatal(pdev->xdev, err, | |
951 | "Error reading PCI root %d", i); | |
952 | goto out; | |
953 | } | |
954 | ||
955 | err = pcifront_rescan_root(pdev, domain, bus); | |
956 | if (err) { | |
957 | xenbus_dev_fatal(pdev->xdev, err, | |
958 | "Error scanning PCI root %04x:%02x", | |
959 | domain, bus); | |
960 | goto out; | |
961 | } | |
962 | } | |
963 | ||
964 | xenbus_switch_state(pdev->xdev, XenbusStateConnected); | |
965 | ||
966 | out: | |
967 | return err; | |
968 | } | |
969 | ||
970 | static int pcifront_detach_devices(struct pcifront_device *pdev) | |
971 | { | |
972 | int err = 0; | |
973 | int i, num_devs; | |
974 | unsigned int domain, bus, slot, func; | |
975 | struct pci_bus *pci_bus; | |
976 | struct pci_dev *pci_dev; | |
977 | char str[64]; | |
978 | ||
979 | if (xenbus_read_driver_state(pdev->xdev->nodename) != | |
980 | XenbusStateConnected) | |
981 | goto out; | |
982 | ||
983 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d", | |
984 | &num_devs); | |
985 | if (err != 1) { | |
986 | if (err >= 0) | |
987 | err = -EINVAL; | |
988 | xenbus_dev_fatal(pdev->xdev, err, | |
989 | "Error reading number of PCI devices"); | |
990 | goto out; | |
991 | } | |
992 | ||
993 | /* Find devices being detached and remove them. */ | |
994 | for (i = 0; i < num_devs; i++) { | |
995 | int l, state; | |
996 | l = snprintf(str, sizeof(str), "state-%d", i); | |
997 | if (unlikely(l >= (sizeof(str) - 1))) { | |
998 | err = -ENOMEM; | |
999 | goto out; | |
1000 | } | |
1001 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, "%d", | |
1002 | &state); | |
1003 | if (err != 1) | |
1004 | state = XenbusStateUnknown; | |
1005 | ||
1006 | if (state != XenbusStateClosing) | |
1007 | continue; | |
1008 | ||
1009 | /* Remove device. */ | |
1010 | l = snprintf(str, sizeof(str), "vdev-%d", i); | |
1011 | if (unlikely(l >= (sizeof(str) - 1))) { | |
1012 | err = -ENOMEM; | |
1013 | goto out; | |
1014 | } | |
1015 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, | |
1016 | "%x:%x:%x.%x", &domain, &bus, &slot, &func); | |
1017 | if (err != 4) { | |
1018 | if (err >= 0) | |
1019 | err = -EINVAL; | |
1020 | xenbus_dev_fatal(pdev->xdev, err, | |
1021 | "Error reading PCI device %d", i); | |
1022 | goto out; | |
1023 | } | |
1024 | ||
1025 | pci_bus = pci_find_bus(domain, bus); | |
1026 | if (!pci_bus) { | |
1027 | dev_dbg(&pdev->xdev->dev, "Cannot get bus %04x:%02x\n", | |
1028 | domain, bus); | |
1029 | continue; | |
1030 | } | |
1031 | pci_dev = pci_get_slot(pci_bus, PCI_DEVFN(slot, func)); | |
1032 | if (!pci_dev) { | |
1033 | dev_dbg(&pdev->xdev->dev, | |
1034 | "Cannot get PCI device %04x:%02x:%02x.%02x\n", | |
1035 | domain, bus, slot, func); | |
1036 | continue; | |
1037 | } | |
1038 | pci_remove_bus_device(pci_dev); | |
1039 | pci_dev_put(pci_dev); | |
1040 | ||
1041 | dev_dbg(&pdev->xdev->dev, | |
1042 | "PCI device %04x:%02x:%02x.%02x removed.\n", | |
1043 | domain, bus, slot, func); | |
1044 | } | |
1045 | ||
1046 | err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring); | |
1047 | ||
1048 | out: | |
1049 | return err; | |
1050 | } | |
1051 | ||
1052 | static void __init_refok pcifront_backend_changed(struct xenbus_device *xdev, | |
1053 | enum xenbus_state be_state) | |
1054 | { | |
1055 | struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev); | |
1056 | ||
1057 | switch (be_state) { | |
1058 | case XenbusStateUnknown: | |
1059 | case XenbusStateInitialising: | |
1060 | case XenbusStateInitWait: | |
1061 | case XenbusStateInitialised: | |
1062 | case XenbusStateClosed: | |
1063 | break; | |
1064 | ||
1065 | case XenbusStateConnected: | |
1066 | pcifront_try_connect(pdev); | |
1067 | break; | |
1068 | ||
1069 | case XenbusStateClosing: | |
1070 | dev_warn(&xdev->dev, "backend going away!\n"); | |
1071 | pcifront_try_disconnect(pdev); | |
1072 | break; | |
1073 | ||
1074 | case XenbusStateReconfiguring: | |
1075 | pcifront_detach_devices(pdev); | |
1076 | break; | |
1077 | ||
1078 | case XenbusStateReconfigured: | |
1079 | pcifront_attach_devices(pdev); | |
1080 | break; | |
1081 | } | |
1082 | } | |
1083 | ||
1084 | static int pcifront_xenbus_probe(struct xenbus_device *xdev, | |
1085 | const struct xenbus_device_id *id) | |
1086 | { | |
1087 | int err = 0; | |
1088 | struct pcifront_device *pdev = alloc_pdev(xdev); | |
1089 | ||
1090 | if (pdev == NULL) { | |
1091 | err = -ENOMEM; | |
1092 | xenbus_dev_fatal(xdev, err, | |
1093 | "Error allocating pcifront_device struct"); | |
1094 | goto out; | |
1095 | } | |
1096 | ||
1097 | err = pcifront_publish_info(pdev); | |
1098 | if (err) | |
1099 | free_pdev(pdev); | |
1100 | ||
1101 | out: | |
1102 | return err; | |
1103 | } | |
1104 | ||
1105 | static int pcifront_xenbus_remove(struct xenbus_device *xdev) | |
1106 | { | |
1107 | struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev); | |
1108 | if (pdev) | |
1109 | free_pdev(pdev); | |
1110 | ||
1111 | return 0; | |
1112 | } | |
1113 | ||
1114 | static const struct xenbus_device_id xenpci_ids[] = { | |
1115 | {"pci"}, | |
1116 | {""}, | |
1117 | }; | |
1118 | ||
1119 | static struct xenbus_driver xenbus_pcifront_driver = { | |
1120 | .name = "pcifront", | |
1121 | .owner = THIS_MODULE, | |
1122 | .ids = xenpci_ids, | |
1123 | .probe = pcifront_xenbus_probe, | |
1124 | .remove = pcifront_xenbus_remove, | |
1125 | .otherend_changed = pcifront_backend_changed, | |
1126 | }; | |
1127 | ||
1128 | static int __init pcifront_init(void) | |
1129 | { | |
1130 | if (!xen_pv_domain() || xen_initial_domain()) | |
1131 | return -ENODEV; | |
1132 | ||
1133 | pci_frontend_registrar(1 /* enable */); | |
1134 | ||
1135 | return xenbus_register_frontend(&xenbus_pcifront_driver); | |
1136 | } | |
1137 | ||
1138 | static void __exit pcifront_cleanup(void) | |
1139 | { | |
1140 | xenbus_unregister_driver(&xenbus_pcifront_driver); | |
1141 | pci_frontend_registrar(0 /* disable */); | |
1142 | } | |
1143 | module_init(pcifront_init); | |
1144 | module_exit(pcifront_cleanup); | |
1145 | ||
1146 | MODULE_DESCRIPTION("Xen PCI passthrough frontend."); | |
1147 | MODULE_LICENSE("GPL"); | |
1148 | MODULE_ALIAS("xen:pci"); |