Root/
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/interface/io/pciif.h> |
17 | #include <asm/xen/pci.h> |
18 | #include <linux/interrupt.h> |
19 | #include <linux/atomic.h> |
20 | #include <linux/workqueue.h> |
21 | #include <linux/bitops.h> |
22 | #include <linux/time.h> |
23 | |
24 | #define INVALID_GRANT_REF (0) |
25 | #define INVALID_EVTCHN (-1) |
26 | |
27 | struct pci_bus_entry { |
28 | struct list_head list; |
29 | struct pci_bus *bus; |
30 | }; |
31 | |
32 | #define _PDEVB_op_active (0) |
33 | #define PDEVB_op_active (1 << (_PDEVB_op_active)) |
34 | |
35 | struct pcifront_device { |
36 | struct xenbus_device *xdev; |
37 | struct list_head root_buses; |
38 | |
39 | int evtchn; |
40 | int gnt_ref; |
41 | |
42 | int irq; |
43 | |
44 | /* Lock this when doing any operations in sh_info */ |
45 | spinlock_t sh_info_lock; |
46 | struct xen_pci_sharedinfo *sh_info; |
47 | struct work_struct op_work; |
48 | unsigned long flags; |
49 | |
50 | }; |
51 | |
52 | struct pcifront_sd { |
53 | int domain; |
54 | struct pcifront_device *pdev; |
55 | }; |
56 | |
57 | static inline struct pcifront_device * |
58 | pcifront_get_pdev(struct pcifront_sd *sd) |
59 | { |
60 | return sd->pdev; |
61 | } |
62 | |
63 | static inline void pcifront_init_sd(struct pcifront_sd *sd, |
64 | unsigned int domain, unsigned int bus, |
65 | struct pcifront_device *pdev) |
66 | { |
67 | sd->domain = domain; |
68 | sd->pdev = pdev; |
69 | } |
70 | |
71 | static DEFINE_SPINLOCK(pcifront_dev_lock); |
72 | static struct pcifront_device *pcifront_dev; |
73 | |
74 | static int verbose_request; |
75 | module_param(verbose_request, int, 0644); |
76 | |
77 | static int errno_to_pcibios_err(int errno) |
78 | { |
79 | switch (errno) { |
80 | case XEN_PCI_ERR_success: |
81 | return PCIBIOS_SUCCESSFUL; |
82 | |
83 | case XEN_PCI_ERR_dev_not_found: |
84 | return PCIBIOS_DEVICE_NOT_FOUND; |
85 | |
86 | case XEN_PCI_ERR_invalid_offset: |
87 | case XEN_PCI_ERR_op_failed: |
88 | return PCIBIOS_BAD_REGISTER_NUMBER; |
89 | |
90 | case XEN_PCI_ERR_not_implemented: |
91 | return PCIBIOS_FUNC_NOT_SUPPORTED; |
92 | |
93 | case XEN_PCI_ERR_access_denied: |
94 | return PCIBIOS_SET_FAILED; |
95 | } |
96 | return errno; |
97 | } |
98 | |
99 | static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev) |
100 | { |
101 | if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags) |
102 | && !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) { |
103 | dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n"); |
104 | schedule_work(&pdev->op_work); |
105 | } |
106 | } |
107 | |
108 | static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op) |
109 | { |
110 | int err = 0; |
111 | struct xen_pci_op *active_op = &pdev->sh_info->op; |
112 | unsigned long irq_flags; |
113 | evtchn_port_t port = pdev->evtchn; |
114 | unsigned irq = pdev->irq; |
115 | s64 ns, ns_timeout; |
116 | struct timeval tv; |
117 | |
118 | spin_lock_irqsave(&pdev->sh_info_lock, irq_flags); |
119 | |
120 | memcpy(active_op, op, sizeof(struct xen_pci_op)); |
121 | |
122 | /* Go */ |
123 | wmb(); |
124 | set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags); |
125 | notify_remote_via_evtchn(port); |
126 | |
127 | /* |
128 | * We set a poll timeout of 3 seconds but give up on return after |
129 | * 2 seconds. It is better to time out too late rather than too early |
130 | * (in the latter case we end up continually re-executing poll() with a |
131 | * timeout in the past). 1s difference gives plenty of slack for error. |
132 | */ |
133 | do_gettimeofday(&tv); |
134 | ns_timeout = timeval_to_ns(&tv) + 2 * (s64)NSEC_PER_SEC; |
135 | |
136 | xen_clear_irq_pending(irq); |
137 | |
138 | while (test_bit(_XEN_PCIF_active, |
139 | (unsigned long *)&pdev->sh_info->flags)) { |
140 | xen_poll_irq_timeout(irq, jiffies + 3*HZ); |
141 | xen_clear_irq_pending(irq); |
142 | do_gettimeofday(&tv); |
143 | ns = timeval_to_ns(&tv); |
144 | if (ns > ns_timeout) { |
145 | dev_err(&pdev->xdev->dev, |
146 | "pciback not responding!!!\n"); |
147 | clear_bit(_XEN_PCIF_active, |
148 | (unsigned long *)&pdev->sh_info->flags); |
149 | err = XEN_PCI_ERR_dev_not_found; |
150 | goto out; |
151 | } |
152 | } |
153 | |
154 | /* |
155 | * We might lose backend service request since we |
156 | * reuse same evtchn with pci_conf backend response. So re-schedule |
157 | * aer pcifront service. |
158 | */ |
159 | if (test_bit(_XEN_PCIB_active, |
160 | (unsigned long *)&pdev->sh_info->flags)) { |
161 | dev_err(&pdev->xdev->dev, |
162 | "schedule aer pcifront service\n"); |
163 | schedule_pcifront_aer_op(pdev); |
164 | } |
165 | |
166 | memcpy(op, active_op, sizeof(struct xen_pci_op)); |
167 | |
168 | err = op->err; |
169 | out: |
170 | spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags); |
171 | return err; |
172 | } |
173 | |
174 | /* Access to this function is spinlocked in drivers/pci/access.c */ |
175 | static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn, |
176 | int where, int size, u32 *val) |
177 | { |
178 | int err = 0; |
179 | struct xen_pci_op op = { |
180 | .cmd = XEN_PCI_OP_conf_read, |
181 | .domain = pci_domain_nr(bus), |
182 | .bus = bus->number, |
183 | .devfn = devfn, |
184 | .offset = where, |
185 | .size = size, |
186 | }; |
187 | struct pcifront_sd *sd = bus->sysdata; |
188 | struct pcifront_device *pdev = pcifront_get_pdev(sd); |
189 | |
190 | if (verbose_request) |
191 | dev_info(&pdev->xdev->dev, |
192 | "read dev=%04x:%02x:%02x.%d - offset %x size %d\n", |
193 | pci_domain_nr(bus), bus->number, PCI_SLOT(devfn), |
194 | PCI_FUNC(devfn), where, size); |
195 | |
196 | err = do_pci_op(pdev, &op); |
197 | |
198 | if (likely(!err)) { |
199 | if (verbose_request) |
200 | dev_info(&pdev->xdev->dev, "read got back value %x\n", |
201 | op.value); |
202 | |
203 | *val = op.value; |
204 | } else if (err == -ENODEV) { |
205 | /* No device here, pretend that it just returned 0 */ |
206 | err = 0; |
207 | *val = 0; |
208 | } |
209 | |
210 | return errno_to_pcibios_err(err); |
211 | } |
212 | |
213 | /* Access to this function is spinlocked in drivers/pci/access.c */ |
214 | static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn, |
215 | int where, int size, u32 val) |
216 | { |
217 | struct xen_pci_op op = { |
218 | .cmd = XEN_PCI_OP_conf_write, |
219 | .domain = pci_domain_nr(bus), |
220 | .bus = bus->number, |
221 | .devfn = devfn, |
222 | .offset = where, |
223 | .size = size, |
224 | .value = val, |
225 | }; |
226 | struct pcifront_sd *sd = bus->sysdata; |
227 | struct pcifront_device *pdev = pcifront_get_pdev(sd); |
228 | |
229 | if (verbose_request) |
230 | dev_info(&pdev->xdev->dev, |
231 | "write dev=%04x:%02x:%02x.%d - " |
232 | "offset %x size %d val %x\n", |
233 | pci_domain_nr(bus), bus->number, |
234 | PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val); |
235 | |
236 | return errno_to_pcibios_err(do_pci_op(pdev, &op)); |
237 | } |
238 | |
239 | struct pci_ops pcifront_bus_ops = { |
240 | .read = pcifront_bus_read, |
241 | .write = pcifront_bus_write, |
242 | }; |
243 | |
244 | #ifdef CONFIG_PCI_MSI |
245 | static int pci_frontend_enable_msix(struct pci_dev *dev, |
246 | int vector[], int nvec) |
247 | { |
248 | int err; |
249 | int i; |
250 | struct xen_pci_op op = { |
251 | .cmd = XEN_PCI_OP_enable_msix, |
252 | .domain = pci_domain_nr(dev->bus), |
253 | .bus = dev->bus->number, |
254 | .devfn = dev->devfn, |
255 | .value = nvec, |
256 | }; |
257 | struct pcifront_sd *sd = dev->bus->sysdata; |
258 | struct pcifront_device *pdev = pcifront_get_pdev(sd); |
259 | struct msi_desc *entry; |
260 | |
261 | if (nvec > SH_INFO_MAX_VEC) { |
262 | dev_err(&dev->dev, "too much vector for pci frontend: %x." |
263 | " Increase SH_INFO_MAX_VEC.\n", nvec); |
264 | return -EINVAL; |
265 | } |
266 | |
267 | i = 0; |
268 | list_for_each_entry(entry, &dev->msi_list, list) { |
269 | op.msix_entries[i].entry = entry->msi_attrib.entry_nr; |
270 | /* Vector is useless at this point. */ |
271 | op.msix_entries[i].vector = -1; |
272 | i++; |
273 | } |
274 | |
275 | err = do_pci_op(pdev, &op); |
276 | |
277 | if (likely(!err)) { |
278 | if (likely(!op.value)) { |
279 | /* we get the result */ |
280 | for (i = 0; i < nvec; i++) { |
281 | if (op.msix_entries[i].vector <= 0) { |
282 | dev_warn(&dev->dev, "MSI-X entry %d is invalid: %d!\n", |
283 | i, op.msix_entries[i].vector); |
284 | err = -EINVAL; |
285 | vector[i] = -1; |
286 | continue; |
287 | } |
288 | vector[i] = op.msix_entries[i].vector; |
289 | } |
290 | } else { |
291 | printk(KERN_DEBUG "enable msix get value %x\n", |
292 | op.value); |
293 | err = op.value; |
294 | } |
295 | } else { |
296 | dev_err(&dev->dev, "enable msix get err %x\n", err); |
297 | } |
298 | return err; |
299 | } |
300 | |
301 | static void pci_frontend_disable_msix(struct pci_dev *dev) |
302 | { |
303 | int err; |
304 | struct xen_pci_op op = { |
305 | .cmd = XEN_PCI_OP_disable_msix, |
306 | .domain = pci_domain_nr(dev->bus), |
307 | .bus = dev->bus->number, |
308 | .devfn = dev->devfn, |
309 | }; |
310 | struct pcifront_sd *sd = dev->bus->sysdata; |
311 | struct pcifront_device *pdev = pcifront_get_pdev(sd); |
312 | |
313 | err = do_pci_op(pdev, &op); |
314 | |
315 | /* What should do for error ? */ |
316 | if (err) |
317 | dev_err(&dev->dev, "pci_disable_msix get err %x\n", err); |
318 | } |
319 | |
320 | static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[]) |
321 | { |
322 | int err; |
323 | struct xen_pci_op op = { |
324 | .cmd = XEN_PCI_OP_enable_msi, |
325 | .domain = pci_domain_nr(dev->bus), |
326 | .bus = dev->bus->number, |
327 | .devfn = dev->devfn, |
328 | }; |
329 | struct pcifront_sd *sd = dev->bus->sysdata; |
330 | struct pcifront_device *pdev = pcifront_get_pdev(sd); |
331 | |
332 | err = do_pci_op(pdev, &op); |
333 | if (likely(!err)) { |
334 | vector[0] = op.value; |
335 | if (op.value <= 0) { |
336 | dev_warn(&dev->dev, "MSI entry is invalid: %d!\n", |
337 | op.value); |
338 | err = -EINVAL; |
339 | vector[0] = -1; |
340 | } |
341 | } else { |
342 | dev_err(&dev->dev, "pci frontend enable msi failed for dev " |
343 | "%x:%x\n", op.bus, op.devfn); |
344 | err = -EINVAL; |
345 | } |
346 | return err; |
347 | } |
348 | |
349 | static void pci_frontend_disable_msi(struct pci_dev *dev) |
350 | { |
351 | int err; |
352 | struct xen_pci_op op = { |
353 | .cmd = XEN_PCI_OP_disable_msi, |
354 | .domain = pci_domain_nr(dev->bus), |
355 | .bus = dev->bus->number, |
356 | .devfn = dev->devfn, |
357 | }; |
358 | struct pcifront_sd *sd = dev->bus->sysdata; |
359 | struct pcifront_device *pdev = pcifront_get_pdev(sd); |
360 | |
361 | err = do_pci_op(pdev, &op); |
362 | if (err == XEN_PCI_ERR_dev_not_found) { |
363 | /* XXX No response from backend, what shall we do? */ |
364 | printk(KERN_DEBUG "get no response from backend for disable MSI\n"); |
365 | return; |
366 | } |
367 | if (err) |
368 | /* how can pciback notify us fail? */ |
369 | printk(KERN_DEBUG "get fake response frombackend\n"); |
370 | } |
371 | |
372 | static struct xen_pci_frontend_ops pci_frontend_ops = { |
373 | .enable_msi = pci_frontend_enable_msi, |
374 | .disable_msi = pci_frontend_disable_msi, |
375 | .enable_msix = pci_frontend_enable_msix, |
376 | .disable_msix = pci_frontend_disable_msix, |
377 | }; |
378 | |
379 | static void pci_frontend_registrar(int enable) |
380 | { |
381 | if (enable) |
382 | xen_pci_frontend = &pci_frontend_ops; |
383 | else |
384 | xen_pci_frontend = NULL; |
385 | }; |
386 | #else |
387 | static inline void pci_frontend_registrar(int enable) { }; |
388 | #endif /* CONFIG_PCI_MSI */ |
389 | |
390 | /* Claim resources for the PCI frontend as-is, backend won't allow changes */ |
391 | static int pcifront_claim_resource(struct pci_dev *dev, void *data) |
392 | { |
393 | struct pcifront_device *pdev = data; |
394 | int i; |
395 | struct resource *r; |
396 | |
397 | for (i = 0; i < PCI_NUM_RESOURCES; i++) { |
398 | r = &dev->resource[i]; |
399 | |
400 | if (!r->parent && r->start && r->flags) { |
401 | dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n", |
402 | pci_name(dev), i); |
403 | if (pci_claim_resource(dev, i)) { |
404 | dev_err(&pdev->xdev->dev, "Could not claim resource %s/%d! " |
405 | "Device offline. Try using e820_host=1 in the guest config.\n", |
406 | pci_name(dev), i); |
407 | } |
408 | } |
409 | } |
410 | |
411 | return 0; |
412 | } |
413 | |
414 | static int __devinit pcifront_scan_bus(struct pcifront_device *pdev, |
415 | unsigned int domain, unsigned int bus, |
416 | struct pci_bus *b) |
417 | { |
418 | struct pci_dev *d; |
419 | unsigned int devfn; |
420 | |
421 | /* Scan the bus for functions and add. |
422 | * We omit handling of PCI bridge attachment because pciback prevents |
423 | * bridges from being exported. |
424 | */ |
425 | for (devfn = 0; devfn < 0x100; devfn++) { |
426 | d = pci_get_slot(b, devfn); |
427 | if (d) { |
428 | /* Device is already known. */ |
429 | pci_dev_put(d); |
430 | continue; |
431 | } |
432 | |
433 | d = pci_scan_single_device(b, devfn); |
434 | if (d) |
435 | dev_info(&pdev->xdev->dev, "New device on " |
436 | "%04x:%02x:%02x.%d found.\n", domain, bus, |
437 | PCI_SLOT(devfn), PCI_FUNC(devfn)); |
438 | } |
439 | |
440 | return 0; |
441 | } |
442 | |
443 | static int __devinit pcifront_scan_root(struct pcifront_device *pdev, |
444 | unsigned int domain, unsigned int bus) |
445 | { |
446 | struct pci_bus *b; |
447 | struct pcifront_sd *sd = NULL; |
448 | struct pci_bus_entry *bus_entry = NULL; |
449 | int err = 0; |
450 | |
451 | #ifndef CONFIG_PCI_DOMAINS |
452 | if (domain != 0) { |
453 | dev_err(&pdev->xdev->dev, |
454 | "PCI Root in non-zero PCI Domain! domain=%d\n", domain); |
455 | dev_err(&pdev->xdev->dev, |
456 | "Please compile with CONFIG_PCI_DOMAINS\n"); |
457 | err = -EINVAL; |
458 | goto err_out; |
459 | } |
460 | #endif |
461 | |
462 | dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n", |
463 | domain, bus); |
464 | |
465 | bus_entry = kmalloc(sizeof(*bus_entry), GFP_KERNEL); |
466 | sd = kmalloc(sizeof(*sd), GFP_KERNEL); |
467 | if (!bus_entry || !sd) { |
468 | err = -ENOMEM; |
469 | goto err_out; |
470 | } |
471 | pcifront_init_sd(sd, domain, bus, pdev); |
472 | |
473 | b = pci_scan_bus_parented(&pdev->xdev->dev, bus, |
474 | &pcifront_bus_ops, sd); |
475 | if (!b) { |
476 | dev_err(&pdev->xdev->dev, |
477 | "Error creating PCI Frontend Bus!\n"); |
478 | err = -ENOMEM; |
479 | goto err_out; |
480 | } |
481 | |
482 | bus_entry->bus = b; |
483 | |
484 | list_add(&bus_entry->list, &pdev->root_buses); |
485 | |
486 | /* pci_scan_bus_parented skips devices which do not have a have |
487 | * devfn==0. The pcifront_scan_bus enumerates all devfn. */ |
488 | err = pcifront_scan_bus(pdev, domain, bus, b); |
489 | |
490 | /* Claim resources before going "live" with our devices */ |
491 | pci_walk_bus(b, pcifront_claim_resource, pdev); |
492 | |
493 | /* Create SysFS and notify udev of the devices. Aka: "going live" */ |
494 | pci_bus_add_devices(b); |
495 | |
496 | return err; |
497 | |
498 | err_out: |
499 | kfree(bus_entry); |
500 | kfree(sd); |
501 | |
502 | return err; |
503 | } |
504 | |
505 | static int __devinit pcifront_rescan_root(struct pcifront_device *pdev, |
506 | unsigned int domain, unsigned int bus) |
507 | { |
508 | int err; |
509 | struct pci_bus *b; |
510 | |
511 | #ifndef CONFIG_PCI_DOMAINS |
512 | if (domain != 0) { |
513 | dev_err(&pdev->xdev->dev, |
514 | "PCI Root in non-zero PCI Domain! domain=%d\n", domain); |
515 | dev_err(&pdev->xdev->dev, |
516 | "Please compile with CONFIG_PCI_DOMAINS\n"); |
517 | return -EINVAL; |
518 | } |
519 | #endif |
520 | |
521 | dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n", |
522 | domain, bus); |
523 | |
524 | b = pci_find_bus(domain, bus); |
525 | if (!b) |
526 | /* If the bus is unknown, create it. */ |
527 | return pcifront_scan_root(pdev, domain, bus); |
528 | |
529 | err = pcifront_scan_bus(pdev, domain, bus, b); |
530 | |
531 | /* Claim resources before going "live" with our devices */ |
532 | pci_walk_bus(b, pcifront_claim_resource, pdev); |
533 | |
534 | /* Create SysFS and notify udev of the devices. Aka: "going live" */ |
535 | pci_bus_add_devices(b); |
536 | |
537 | return err; |
538 | } |
539 | |
540 | static void free_root_bus_devs(struct pci_bus *bus) |
541 | { |
542 | struct pci_dev *dev; |
543 | |
544 | while (!list_empty(&bus->devices)) { |
545 | dev = container_of(bus->devices.next, struct pci_dev, |
546 | bus_list); |
547 | dev_dbg(&dev->dev, "removing device\n"); |
548 | pci_stop_and_remove_bus_device(dev); |
549 | } |
550 | } |
551 | |
552 | static void pcifront_free_roots(struct pcifront_device *pdev) |
553 | { |
554 | struct pci_bus_entry *bus_entry, *t; |
555 | |
556 | dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n"); |
557 | |
558 | list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) { |
559 | list_del(&bus_entry->list); |
560 | |
561 | free_root_bus_devs(bus_entry->bus); |
562 | |
563 | kfree(bus_entry->bus->sysdata); |
564 | |
565 | device_unregister(bus_entry->bus->bridge); |
566 | pci_remove_bus(bus_entry->bus); |
567 | |
568 | kfree(bus_entry); |
569 | } |
570 | } |
571 | |
572 | static pci_ers_result_t pcifront_common_process(int cmd, |
573 | struct pcifront_device *pdev, |
574 | pci_channel_state_t state) |
575 | { |
576 | pci_ers_result_t result; |
577 | struct pci_driver *pdrv; |
578 | int bus = pdev->sh_info->aer_op.bus; |
579 | int devfn = pdev->sh_info->aer_op.devfn; |
580 | struct pci_dev *pcidev; |
581 | int flag = 0; |
582 | |
583 | dev_dbg(&pdev->xdev->dev, |
584 | "pcifront AER process: cmd %x (bus:%x, devfn%x)", |
585 | cmd, bus, devfn); |
586 | result = PCI_ERS_RESULT_NONE; |
587 | |
588 | pcidev = pci_get_bus_and_slot(bus, devfn); |
589 | if (!pcidev || !pcidev->driver) { |
590 | dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n"); |
591 | if (pcidev) |
592 | pci_dev_put(pcidev); |
593 | return result; |
594 | } |
595 | pdrv = pcidev->driver; |
596 | |
597 | if (pdrv) { |
598 | if (pdrv->err_handler && pdrv->err_handler->error_detected) { |
599 | dev_dbg(&pcidev->dev, |
600 | "trying to call AER service\n"); |
601 | if (pcidev) { |
602 | flag = 1; |
603 | switch (cmd) { |
604 | case XEN_PCI_OP_aer_detected: |
605 | result = pdrv->err_handler-> |
606 | error_detected(pcidev, state); |
607 | break; |
608 | case XEN_PCI_OP_aer_mmio: |
609 | result = pdrv->err_handler-> |
610 | mmio_enabled(pcidev); |
611 | break; |
612 | case XEN_PCI_OP_aer_slotreset: |
613 | result = pdrv->err_handler-> |
614 | slot_reset(pcidev); |
615 | break; |
616 | case XEN_PCI_OP_aer_resume: |
617 | pdrv->err_handler->resume(pcidev); |
618 | break; |
619 | default: |
620 | dev_err(&pdev->xdev->dev, |
621 | "bad request in aer recovery " |
622 | "operation!\n"); |
623 | |
624 | } |
625 | } |
626 | } |
627 | } |
628 | if (!flag) |
629 | result = PCI_ERS_RESULT_NONE; |
630 | |
631 | return result; |
632 | } |
633 | |
634 | |
635 | static void pcifront_do_aer(struct work_struct *data) |
636 | { |
637 | struct pcifront_device *pdev = |
638 | container_of(data, struct pcifront_device, op_work); |
639 | int cmd = pdev->sh_info->aer_op.cmd; |
640 | pci_channel_state_t state = |
641 | (pci_channel_state_t)pdev->sh_info->aer_op.err; |
642 | |
643 | /*If a pci_conf op is in progress, |
644 | we have to wait until it is done before service aer op*/ |
645 | dev_dbg(&pdev->xdev->dev, |
646 | "pcifront service aer bus %x devfn %x\n", |
647 | pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn); |
648 | |
649 | pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state); |
650 | |
651 | /* Post the operation to the guest. */ |
652 | wmb(); |
653 | clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags); |
654 | notify_remote_via_evtchn(pdev->evtchn); |
655 | |
656 | /*in case of we lost an aer request in four lines time_window*/ |
657 | smp_mb__before_clear_bit(); |
658 | clear_bit(_PDEVB_op_active, &pdev->flags); |
659 | smp_mb__after_clear_bit(); |
660 | |
661 | schedule_pcifront_aer_op(pdev); |
662 | |
663 | } |
664 | |
665 | static irqreturn_t pcifront_handler_aer(int irq, void *dev) |
666 | { |
667 | struct pcifront_device *pdev = dev; |
668 | schedule_pcifront_aer_op(pdev); |
669 | return IRQ_HANDLED; |
670 | } |
671 | static int pcifront_connect(struct pcifront_device *pdev) |
672 | { |
673 | int err = 0; |
674 | |
675 | spin_lock(&pcifront_dev_lock); |
676 | |
677 | if (!pcifront_dev) { |
678 | dev_info(&pdev->xdev->dev, "Installing PCI frontend\n"); |
679 | pcifront_dev = pdev; |
680 | } else { |
681 | dev_err(&pdev->xdev->dev, "PCI frontend already installed!\n"); |
682 | err = -EEXIST; |
683 | } |
684 | |
685 | spin_unlock(&pcifront_dev_lock); |
686 | |
687 | return err; |
688 | } |
689 | |
690 | static void pcifront_disconnect(struct pcifront_device *pdev) |
691 | { |
692 | spin_lock(&pcifront_dev_lock); |
693 | |
694 | if (pdev == pcifront_dev) { |
695 | dev_info(&pdev->xdev->dev, |
696 | "Disconnecting PCI Frontend Buses\n"); |
697 | pcifront_dev = NULL; |
698 | } |
699 | |
700 | spin_unlock(&pcifront_dev_lock); |
701 | } |
702 | static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev) |
703 | { |
704 | struct pcifront_device *pdev; |
705 | |
706 | pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL); |
707 | if (pdev == NULL) |
708 | goto out; |
709 | |
710 | pdev->sh_info = |
711 | (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL); |
712 | if (pdev->sh_info == NULL) { |
713 | kfree(pdev); |
714 | pdev = NULL; |
715 | goto out; |
716 | } |
717 | pdev->sh_info->flags = 0; |
718 | |
719 | /*Flag for registering PV AER handler*/ |
720 | set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags); |
721 | |
722 | dev_set_drvdata(&xdev->dev, pdev); |
723 | pdev->xdev = xdev; |
724 | |
725 | INIT_LIST_HEAD(&pdev->root_buses); |
726 | |
727 | spin_lock_init(&pdev->sh_info_lock); |
728 | |
729 | pdev->evtchn = INVALID_EVTCHN; |
730 | pdev->gnt_ref = INVALID_GRANT_REF; |
731 | pdev->irq = -1; |
732 | |
733 | INIT_WORK(&pdev->op_work, pcifront_do_aer); |
734 | |
735 | dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n", |
736 | pdev, pdev->sh_info); |
737 | out: |
738 | return pdev; |
739 | } |
740 | |
741 | static void free_pdev(struct pcifront_device *pdev) |
742 | { |
743 | dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev); |
744 | |
745 | pcifront_free_roots(pdev); |
746 | |
747 | cancel_work_sync(&pdev->op_work); |
748 | |
749 | if (pdev->irq >= 0) |
750 | unbind_from_irqhandler(pdev->irq, pdev); |
751 | |
752 | if (pdev->evtchn != INVALID_EVTCHN) |
753 | xenbus_free_evtchn(pdev->xdev, pdev->evtchn); |
754 | |
755 | if (pdev->gnt_ref != INVALID_GRANT_REF) |
756 | gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */, |
757 | (unsigned long)pdev->sh_info); |
758 | else |
759 | free_page((unsigned long)pdev->sh_info); |
760 | |
761 | dev_set_drvdata(&pdev->xdev->dev, NULL); |
762 | |
763 | kfree(pdev); |
764 | } |
765 | |
766 | static int pcifront_publish_info(struct pcifront_device *pdev) |
767 | { |
768 | int err = 0; |
769 | struct xenbus_transaction trans; |
770 | |
771 | err = xenbus_grant_ring(pdev->xdev, virt_to_mfn(pdev->sh_info)); |
772 | if (err < 0) |
773 | goto out; |
774 | |
775 | pdev->gnt_ref = err; |
776 | |
777 | err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn); |
778 | if (err) |
779 | goto out; |
780 | |
781 | err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer, |
782 | 0, "pcifront", pdev); |
783 | |
784 | if (err < 0) |
785 | return err; |
786 | |
787 | pdev->irq = err; |
788 | |
789 | do_publish: |
790 | err = xenbus_transaction_start(&trans); |
791 | if (err) { |
792 | xenbus_dev_fatal(pdev->xdev, err, |
793 | "Error writing configuration for backend " |
794 | "(start transaction)"); |
795 | goto out; |
796 | } |
797 | |
798 | err = xenbus_printf(trans, pdev->xdev->nodename, |
799 | "pci-op-ref", "%u", pdev->gnt_ref); |
800 | if (!err) |
801 | err = xenbus_printf(trans, pdev->xdev->nodename, |
802 | "event-channel", "%u", pdev->evtchn); |
803 | if (!err) |
804 | err = xenbus_printf(trans, pdev->xdev->nodename, |
805 | "magic", XEN_PCI_MAGIC); |
806 | |
807 | if (err) { |
808 | xenbus_transaction_end(trans, 1); |
809 | xenbus_dev_fatal(pdev->xdev, err, |
810 | "Error writing configuration for backend"); |
811 | goto out; |
812 | } else { |
813 | err = xenbus_transaction_end(trans, 0); |
814 | if (err == -EAGAIN) |
815 | goto do_publish; |
816 | else if (err) { |
817 | xenbus_dev_fatal(pdev->xdev, err, |
818 | "Error completing transaction " |
819 | "for backend"); |
820 | goto out; |
821 | } |
822 | } |
823 | |
824 | xenbus_switch_state(pdev->xdev, XenbusStateInitialised); |
825 | |
826 | dev_dbg(&pdev->xdev->dev, "publishing successful!\n"); |
827 | |
828 | out: |
829 | return err; |
830 | } |
831 | |
832 | static int __devinit pcifront_try_connect(struct pcifront_device *pdev) |
833 | { |
834 | int err = -EFAULT; |
835 | int i, num_roots, len; |
836 | char str[64]; |
837 | unsigned int domain, bus; |
838 | |
839 | |
840 | /* Only connect once */ |
841 | if (xenbus_read_driver_state(pdev->xdev->nodename) != |
842 | XenbusStateInitialised) |
843 | goto out; |
844 | |
845 | err = pcifront_connect(pdev); |
846 | if (err) { |
847 | xenbus_dev_fatal(pdev->xdev, err, |
848 | "Error connecting PCI Frontend"); |
849 | goto out; |
850 | } |
851 | |
852 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, |
853 | "root_num", "%d", &num_roots); |
854 | if (err == -ENOENT) { |
855 | xenbus_dev_error(pdev->xdev, err, |
856 | "No PCI Roots found, trying 0000:00"); |
857 | err = pcifront_scan_root(pdev, 0, 0); |
858 | num_roots = 0; |
859 | } else if (err != 1) { |
860 | if (err == 0) |
861 | err = -EINVAL; |
862 | xenbus_dev_fatal(pdev->xdev, err, |
863 | "Error reading number of PCI roots"); |
864 | goto out; |
865 | } |
866 | |
867 | for (i = 0; i < num_roots; i++) { |
868 | len = snprintf(str, sizeof(str), "root-%d", i); |
869 | if (unlikely(len >= (sizeof(str) - 1))) { |
870 | err = -ENOMEM; |
871 | goto out; |
872 | } |
873 | |
874 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, |
875 | "%x:%x", &domain, &bus); |
876 | if (err != 2) { |
877 | if (err >= 0) |
878 | err = -EINVAL; |
879 | xenbus_dev_fatal(pdev->xdev, err, |
880 | "Error reading PCI root %d", i); |
881 | goto out; |
882 | } |
883 | |
884 | err = pcifront_scan_root(pdev, domain, bus); |
885 | if (err) { |
886 | xenbus_dev_fatal(pdev->xdev, err, |
887 | "Error scanning PCI root %04x:%02x", |
888 | domain, bus); |
889 | goto out; |
890 | } |
891 | } |
892 | |
893 | err = xenbus_switch_state(pdev->xdev, XenbusStateConnected); |
894 | |
895 | out: |
896 | return err; |
897 | } |
898 | |
899 | static int pcifront_try_disconnect(struct pcifront_device *pdev) |
900 | { |
901 | int err = 0; |
902 | enum xenbus_state prev_state; |
903 | |
904 | |
905 | prev_state = xenbus_read_driver_state(pdev->xdev->nodename); |
906 | |
907 | if (prev_state >= XenbusStateClosing) |
908 | goto out; |
909 | |
910 | if (prev_state == XenbusStateConnected) { |
911 | pcifront_free_roots(pdev); |
912 | pcifront_disconnect(pdev); |
913 | } |
914 | |
915 | err = xenbus_switch_state(pdev->xdev, XenbusStateClosed); |
916 | |
917 | out: |
918 | |
919 | return err; |
920 | } |
921 | |
922 | static int __devinit pcifront_attach_devices(struct pcifront_device *pdev) |
923 | { |
924 | int err = -EFAULT; |
925 | int i, num_roots, len; |
926 | unsigned int domain, bus; |
927 | char str[64]; |
928 | |
929 | if (xenbus_read_driver_state(pdev->xdev->nodename) != |
930 | XenbusStateReconfiguring) |
931 | goto out; |
932 | |
933 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, |
934 | "root_num", "%d", &num_roots); |
935 | if (err == -ENOENT) { |
936 | xenbus_dev_error(pdev->xdev, err, |
937 | "No PCI Roots found, trying 0000:00"); |
938 | err = pcifront_rescan_root(pdev, 0, 0); |
939 | num_roots = 0; |
940 | } else if (err != 1) { |
941 | if (err == 0) |
942 | err = -EINVAL; |
943 | xenbus_dev_fatal(pdev->xdev, err, |
944 | "Error reading number of PCI roots"); |
945 | goto out; |
946 | } |
947 | |
948 | for (i = 0; i < num_roots; i++) { |
949 | len = snprintf(str, sizeof(str), "root-%d", i); |
950 | if (unlikely(len >= (sizeof(str) - 1))) { |
951 | err = -ENOMEM; |
952 | goto out; |
953 | } |
954 | |
955 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, |
956 | "%x:%x", &domain, &bus); |
957 | if (err != 2) { |
958 | if (err >= 0) |
959 | err = -EINVAL; |
960 | xenbus_dev_fatal(pdev->xdev, err, |
961 | "Error reading PCI root %d", i); |
962 | goto out; |
963 | } |
964 | |
965 | err = pcifront_rescan_root(pdev, domain, bus); |
966 | if (err) { |
967 | xenbus_dev_fatal(pdev->xdev, err, |
968 | "Error scanning PCI root %04x:%02x", |
969 | domain, bus); |
970 | goto out; |
971 | } |
972 | } |
973 | |
974 | xenbus_switch_state(pdev->xdev, XenbusStateConnected); |
975 | |
976 | out: |
977 | return err; |
978 | } |
979 | |
980 | static int pcifront_detach_devices(struct pcifront_device *pdev) |
981 | { |
982 | int err = 0; |
983 | int i, num_devs; |
984 | unsigned int domain, bus, slot, func; |
985 | struct pci_bus *pci_bus; |
986 | struct pci_dev *pci_dev; |
987 | char str[64]; |
988 | |
989 | if (xenbus_read_driver_state(pdev->xdev->nodename) != |
990 | XenbusStateConnected) |
991 | goto out; |
992 | |
993 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d", |
994 | &num_devs); |
995 | if (err != 1) { |
996 | if (err >= 0) |
997 | err = -EINVAL; |
998 | xenbus_dev_fatal(pdev->xdev, err, |
999 | "Error reading number of PCI devices"); |
1000 | goto out; |
1001 | } |
1002 | |
1003 | /* Find devices being detached and remove them. */ |
1004 | for (i = 0; i < num_devs; i++) { |
1005 | int l, state; |
1006 | l = snprintf(str, sizeof(str), "state-%d", i); |
1007 | if (unlikely(l >= (sizeof(str) - 1))) { |
1008 | err = -ENOMEM; |
1009 | goto out; |
1010 | } |
1011 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, "%d", |
1012 | &state); |
1013 | if (err != 1) |
1014 | state = XenbusStateUnknown; |
1015 | |
1016 | if (state != XenbusStateClosing) |
1017 | continue; |
1018 | |
1019 | /* Remove device. */ |
1020 | l = snprintf(str, sizeof(str), "vdev-%d", i); |
1021 | if (unlikely(l >= (sizeof(str) - 1))) { |
1022 | err = -ENOMEM; |
1023 | goto out; |
1024 | } |
1025 | err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, |
1026 | "%x:%x:%x.%x", &domain, &bus, &slot, &func); |
1027 | if (err != 4) { |
1028 | if (err >= 0) |
1029 | err = -EINVAL; |
1030 | xenbus_dev_fatal(pdev->xdev, err, |
1031 | "Error reading PCI device %d", i); |
1032 | goto out; |
1033 | } |
1034 | |
1035 | pci_bus = pci_find_bus(domain, bus); |
1036 | if (!pci_bus) { |
1037 | dev_dbg(&pdev->xdev->dev, "Cannot get bus %04x:%02x\n", |
1038 | domain, bus); |
1039 | continue; |
1040 | } |
1041 | pci_dev = pci_get_slot(pci_bus, PCI_DEVFN(slot, func)); |
1042 | if (!pci_dev) { |
1043 | dev_dbg(&pdev->xdev->dev, |
1044 | "Cannot get PCI device %04x:%02x:%02x.%d\n", |
1045 | domain, bus, slot, func); |
1046 | continue; |
1047 | } |
1048 | pci_stop_and_remove_bus_device(pci_dev); |
1049 | pci_dev_put(pci_dev); |
1050 | |
1051 | dev_dbg(&pdev->xdev->dev, |
1052 | "PCI device %04x:%02x:%02x.%d removed.\n", |
1053 | domain, bus, slot, func); |
1054 | } |
1055 | |
1056 | err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring); |
1057 | |
1058 | out: |
1059 | return err; |
1060 | } |
1061 | |
1062 | static void __init_refok pcifront_backend_changed(struct xenbus_device *xdev, |
1063 | enum xenbus_state be_state) |
1064 | { |
1065 | struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev); |
1066 | |
1067 | switch (be_state) { |
1068 | case XenbusStateUnknown: |
1069 | case XenbusStateInitialising: |
1070 | case XenbusStateInitWait: |
1071 | case XenbusStateInitialised: |
1072 | case XenbusStateClosed: |
1073 | break; |
1074 | |
1075 | case XenbusStateConnected: |
1076 | pcifront_try_connect(pdev); |
1077 | break; |
1078 | |
1079 | case XenbusStateClosing: |
1080 | dev_warn(&xdev->dev, "backend going away!\n"); |
1081 | pcifront_try_disconnect(pdev); |
1082 | break; |
1083 | |
1084 | case XenbusStateReconfiguring: |
1085 | pcifront_detach_devices(pdev); |
1086 | break; |
1087 | |
1088 | case XenbusStateReconfigured: |
1089 | pcifront_attach_devices(pdev); |
1090 | break; |
1091 | } |
1092 | } |
1093 | |
1094 | static int pcifront_xenbus_probe(struct xenbus_device *xdev, |
1095 | const struct xenbus_device_id *id) |
1096 | { |
1097 | int err = 0; |
1098 | struct pcifront_device *pdev = alloc_pdev(xdev); |
1099 | |
1100 | if (pdev == NULL) { |
1101 | err = -ENOMEM; |
1102 | xenbus_dev_fatal(xdev, err, |
1103 | "Error allocating pcifront_device struct"); |
1104 | goto out; |
1105 | } |
1106 | |
1107 | err = pcifront_publish_info(pdev); |
1108 | if (err) |
1109 | free_pdev(pdev); |
1110 | |
1111 | out: |
1112 | return err; |
1113 | } |
1114 | |
1115 | static int pcifront_xenbus_remove(struct xenbus_device *xdev) |
1116 | { |
1117 | struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev); |
1118 | if (pdev) |
1119 | free_pdev(pdev); |
1120 | |
1121 | return 0; |
1122 | } |
1123 | |
1124 | static const struct xenbus_device_id xenpci_ids[] = { |
1125 | {"pci"}, |
1126 | {""}, |
1127 | }; |
1128 | |
1129 | static DEFINE_XENBUS_DRIVER(xenpci, "pcifront", |
1130 | .probe = pcifront_xenbus_probe, |
1131 | .remove = pcifront_xenbus_remove, |
1132 | .otherend_changed = pcifront_backend_changed, |
1133 | ); |
1134 | |
1135 | static int __init pcifront_init(void) |
1136 | { |
1137 | if (!xen_pv_domain() || xen_initial_domain()) |
1138 | return -ENODEV; |
1139 | |
1140 | pci_frontend_registrar(1 /* enable */); |
1141 | |
1142 | return xenbus_register_frontend(&xenpci_driver); |
1143 | } |
1144 | |
1145 | static void __exit pcifront_cleanup(void) |
1146 | { |
1147 | xenbus_unregister_driver(&xenpci_driver); |
1148 | pci_frontend_registrar(0 /* disable */); |
1149 | } |
1150 | module_init(pcifront_init); |
1151 | module_exit(pcifront_cleanup); |
1152 | |
1153 | MODULE_DESCRIPTION("Xen PCI passthrough frontend."); |
1154 | MODULE_LICENSE("GPL"); |
1155 | MODULE_ALIAS("xen:pci"); |
1156 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9