Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Xen PCI Frontend
4 *
5 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
6 */
7#include <linux/module.h>
8#include <linux/init.h>
9#include <linux/mm.h>
10#include <xen/xenbus.h>
11#include <xen/events.h>
12#include <xen/grant_table.h>
13#include <xen/page.h>
14#include <linux/spinlock.h>
15#include <linux/pci.h>
16#include <linux/msi.h>
17#include <xen/interface/io/pciif.h>
18#include <asm/xen/pci.h>
19#include <linux/interrupt.h>
20#include <linux/atomic.h>
21#include <linux/workqueue.h>
22#include <linux/bitops.h>
23#include <linux/time.h>
24#include <linux/ktime.h>
25#include <xen/platform_pci.h>
26
27#include <asm/xen/swiotlb-xen.h>
28#define INVALID_GRANT_REF (0)
29#define INVALID_EVTCHN (-1)
30
31struct pci_bus_entry {
32 struct list_head list;
33 struct pci_bus *bus;
34};
35
36#define _PDEVB_op_active (0)
37#define PDEVB_op_active (1 << (_PDEVB_op_active))
38
39struct pcifront_device {
40 struct xenbus_device *xdev;
41 struct list_head root_buses;
42
43 int evtchn;
44 int gnt_ref;
45
46 int irq;
47
48 /* Lock this when doing any operations in sh_info */
49 spinlock_t sh_info_lock;
50 struct xen_pci_sharedinfo *sh_info;
51 struct work_struct op_work;
52 unsigned long flags;
53
54};
55
56struct pcifront_sd {
57 struct pci_sysdata sd;
58 struct pcifront_device *pdev;
59};
60
61static inline struct pcifront_device *
62pcifront_get_pdev(struct pcifront_sd *sd)
63{
64 return sd->pdev;
65}
66
67static inline void pcifront_init_sd(struct pcifront_sd *sd,
68 unsigned int domain, unsigned int bus,
69 struct pcifront_device *pdev)
70{
71 /* Because we do not expose that information via XenBus. */
72 sd->sd.node = first_online_node;
73 sd->sd.domain = domain;
74 sd->pdev = pdev;
75}
76
77static DEFINE_SPINLOCK(pcifront_dev_lock);
78static struct pcifront_device *pcifront_dev;
79
80static int verbose_request;
81module_param(verbose_request, int, 0644);
82
83static int errno_to_pcibios_err(int errno)
84{
85 switch (errno) {
86 case XEN_PCI_ERR_success:
87 return PCIBIOS_SUCCESSFUL;
88
89 case XEN_PCI_ERR_dev_not_found:
90 return PCIBIOS_DEVICE_NOT_FOUND;
91
92 case XEN_PCI_ERR_invalid_offset:
93 case XEN_PCI_ERR_op_failed:
94 return PCIBIOS_BAD_REGISTER_NUMBER;
95
96 case XEN_PCI_ERR_not_implemented:
97 return PCIBIOS_FUNC_NOT_SUPPORTED;
98
99 case XEN_PCI_ERR_access_denied:
100 return PCIBIOS_SET_FAILED;
101 }
102 return errno;
103}
104
105static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev)
106{
107 if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
108 && !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) {
109 dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n");
110 schedule_work(&pdev->op_work);
111 }
112}
113
114static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op)
115{
116 int err = 0;
117 struct xen_pci_op *active_op = &pdev->sh_info->op;
118 unsigned long irq_flags;
119 evtchn_port_t port = pdev->evtchn;
120 unsigned irq = pdev->irq;
121 s64 ns, ns_timeout;
122
123 spin_lock_irqsave(&pdev->sh_info_lock, irq_flags);
124
125 memcpy(active_op, op, sizeof(struct xen_pci_op));
126
127 /* Go */
128 wmb();
129 set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
130 notify_remote_via_evtchn(port);
131
132 /*
133 * We set a poll timeout of 3 seconds but give up on return after
134 * 2 seconds. It is better to time out too late rather than too early
135 * (in the latter case we end up continually re-executing poll() with a
136 * timeout in the past). 1s difference gives plenty of slack for error.
137 */
138 ns_timeout = ktime_get_ns() + 2 * (s64)NSEC_PER_SEC;
139
140 xen_clear_irq_pending(irq);
141
142 while (test_bit(_XEN_PCIF_active,
143 (unsigned long *)&pdev->sh_info->flags)) {
144 xen_poll_irq_timeout(irq, jiffies + 3*HZ);
145 xen_clear_irq_pending(irq);
146 ns = ktime_get_ns();
147 if (ns > ns_timeout) {
148 dev_err(&pdev->xdev->dev,
149 "pciback not responding!!!\n");
150 clear_bit(_XEN_PCIF_active,
151 (unsigned long *)&pdev->sh_info->flags);
152 err = XEN_PCI_ERR_dev_not_found;
153 goto out;
154 }
155 }
156
157 /*
158 * We might lose backend service request since we
159 * reuse same evtchn with pci_conf backend response. So re-schedule
160 * aer pcifront service.
161 */
162 if (test_bit(_XEN_PCIB_active,
163 (unsigned long *)&pdev->sh_info->flags)) {
164 dev_err(&pdev->xdev->dev,
165 "schedule aer pcifront service\n");
166 schedule_pcifront_aer_op(pdev);
167 }
168
169 memcpy(op, active_op, sizeof(struct xen_pci_op));
170
171 err = op->err;
172out:
173 spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags);
174 return err;
175}
176
177/* Access to this function is spinlocked in drivers/pci/access.c */
178static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn,
179 int where, int size, u32 *val)
180{
181 int err = 0;
182 struct xen_pci_op op = {
183 .cmd = XEN_PCI_OP_conf_read,
184 .domain = pci_domain_nr(bus),
185 .bus = bus->number,
186 .devfn = devfn,
187 .offset = where,
188 .size = size,
189 };
190 struct pcifront_sd *sd = bus->sysdata;
191 struct pcifront_device *pdev = pcifront_get_pdev(sd);
192
193 if (verbose_request)
194 dev_info(&pdev->xdev->dev,
195 "read dev=%04x:%02x:%02x.%d - offset %x size %d\n",
196 pci_domain_nr(bus), bus->number, PCI_SLOT(devfn),
197 PCI_FUNC(devfn), where, size);
198
199 err = do_pci_op(pdev, &op);
200
201 if (likely(!err)) {
202 if (verbose_request)
203 dev_info(&pdev->xdev->dev, "read got back value %x\n",
204 op.value);
205
206 *val = op.value;
207 } else if (err == -ENODEV) {
208 /* No device here, pretend that it just returned 0 */
209 err = 0;
210 *val = 0;
211 }
212
213 return errno_to_pcibios_err(err);
214}
215
216/* Access to this function is spinlocked in drivers/pci/access.c */
217static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn,
218 int where, int size, u32 val)
219{
220 struct xen_pci_op op = {
221 .cmd = XEN_PCI_OP_conf_write,
222 .domain = pci_domain_nr(bus),
223 .bus = bus->number,
224 .devfn = devfn,
225 .offset = where,
226 .size = size,
227 .value = val,
228 };
229 struct pcifront_sd *sd = bus->sysdata;
230 struct pcifront_device *pdev = pcifront_get_pdev(sd);
231
232 if (verbose_request)
233 dev_info(&pdev->xdev->dev,
234 "write dev=%04x:%02x:%02x.%d - "
235 "offset %x size %d val %x\n",
236 pci_domain_nr(bus), bus->number,
237 PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val);
238
239 return errno_to_pcibios_err(do_pci_op(pdev, &op));
240}
241
242static struct pci_ops pcifront_bus_ops = {
243 .read = pcifront_bus_read,
244 .write = pcifront_bus_write,
245};
246
247#ifdef CONFIG_PCI_MSI
248static int pci_frontend_enable_msix(struct pci_dev *dev,
249 int vector[], int nvec)
250{
251 int err;
252 int i;
253 struct xen_pci_op op = {
254 .cmd = XEN_PCI_OP_enable_msix,
255 .domain = pci_domain_nr(dev->bus),
256 .bus = dev->bus->number,
257 .devfn = dev->devfn,
258 .value = nvec,
259 };
260 struct pcifront_sd *sd = dev->bus->sysdata;
261 struct pcifront_device *pdev = pcifront_get_pdev(sd);
262 struct msi_desc *entry;
263
264 if (nvec > SH_INFO_MAX_VEC) {
265 pci_err(dev, "too many vectors (0x%x) for PCI frontend:"
266 " Increase SH_INFO_MAX_VEC\n", nvec);
267 return -EINVAL;
268 }
269
270 i = 0;
271 for_each_pci_msi_entry(entry, dev) {
272 op.msix_entries[i].entry = entry->msi_attrib.entry_nr;
273 /* Vector is useless at this point. */
274 op.msix_entries[i].vector = -1;
275 i++;
276 }
277
278 err = do_pci_op(pdev, &op);
279
280 if (likely(!err)) {
281 if (likely(!op.value)) {
282 /* we get the result */
283 for (i = 0; i < nvec; i++) {
284 if (op.msix_entries[i].vector <= 0) {
285 pci_warn(dev, "MSI-X entry %d is invalid: %d!\n",
286 i, op.msix_entries[i].vector);
287 err = -EINVAL;
288 vector[i] = -1;
289 continue;
290 }
291 vector[i] = op.msix_entries[i].vector;
292 }
293 } else {
294 pr_info("enable msix get value %x\n", op.value);
295 err = op.value;
296 }
297 } else {
298 pci_err(dev, "enable msix get err %x\n", err);
299 }
300 return err;
301}
302
303static void pci_frontend_disable_msix(struct pci_dev *dev)
304{
305 int err;
306 struct xen_pci_op op = {
307 .cmd = XEN_PCI_OP_disable_msix,
308 .domain = pci_domain_nr(dev->bus),
309 .bus = dev->bus->number,
310 .devfn = dev->devfn,
311 };
312 struct pcifront_sd *sd = dev->bus->sysdata;
313 struct pcifront_device *pdev = pcifront_get_pdev(sd);
314
315 err = do_pci_op(pdev, &op);
316
317 /* What should do for error ? */
318 if (err)
319 pci_err(dev, "pci_disable_msix get err %x\n", err);
320}
321
322static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[])
323{
324 int err;
325 struct xen_pci_op op = {
326 .cmd = XEN_PCI_OP_enable_msi,
327 .domain = pci_domain_nr(dev->bus),
328 .bus = dev->bus->number,
329 .devfn = dev->devfn,
330 };
331 struct pcifront_sd *sd = dev->bus->sysdata;
332 struct pcifront_device *pdev = pcifront_get_pdev(sd);
333
334 err = do_pci_op(pdev, &op);
335 if (likely(!err)) {
336 vector[0] = op.value;
337 if (op.value <= 0) {
338 pci_warn(dev, "MSI entry is invalid: %d!\n",
339 op.value);
340 err = -EINVAL;
341 vector[0] = -1;
342 }
343 } else {
344 pci_err(dev, "pci frontend enable msi failed for dev "
345 "%x:%x\n", op.bus, op.devfn);
346 err = -EINVAL;
347 }
348 return err;
349}
350
351static void pci_frontend_disable_msi(struct pci_dev *dev)
352{
353 int err;
354 struct xen_pci_op op = {
355 .cmd = XEN_PCI_OP_disable_msi,
356 .domain = pci_domain_nr(dev->bus),
357 .bus = dev->bus->number,
358 .devfn = dev->devfn,
359 };
360 struct pcifront_sd *sd = dev->bus->sysdata;
361 struct pcifront_device *pdev = pcifront_get_pdev(sd);
362
363 err = do_pci_op(pdev, &op);
364 if (err == XEN_PCI_ERR_dev_not_found) {
365 /* XXX No response from backend, what shall we do? */
366 pr_info("get no response from backend for disable MSI\n");
367 return;
368 }
369 if (err)
370 /* how can pciback notify us fail? */
371 pr_info("get fake response from backend\n");
372}
373
374static struct xen_pci_frontend_ops pci_frontend_ops = {
375 .enable_msi = pci_frontend_enable_msi,
376 .disable_msi = pci_frontend_disable_msi,
377 .enable_msix = pci_frontend_enable_msix,
378 .disable_msix = pci_frontend_disable_msix,
379};
380
381static void pci_frontend_registrar(int enable)
382{
383 if (enable)
384 xen_pci_frontend = &pci_frontend_ops;
385 else
386 xen_pci_frontend = NULL;
387};
388#else
389static inline void pci_frontend_registrar(int enable) { };
390#endif /* CONFIG_PCI_MSI */
391
392/* Claim resources for the PCI frontend as-is, backend won't allow changes */
393static int pcifront_claim_resource(struct pci_dev *dev, void *data)
394{
395 struct pcifront_device *pdev = data;
396 int i;
397 struct resource *r;
398
399 for (i = 0; i < PCI_NUM_RESOURCES; i++) {
400 r = &dev->resource[i];
401
402 if (!r->parent && r->start && r->flags) {
403 dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n",
404 pci_name(dev), i);
405 if (pci_claim_resource(dev, i)) {
406 dev_err(&pdev->xdev->dev, "Could not claim resource %s/%d! "
407 "Device offline. Try using e820_host=1 in the guest config.\n",
408 pci_name(dev), i);
409 }
410 }
411 }
412
413 return 0;
414}
415
416static int pcifront_scan_bus(struct pcifront_device *pdev,
417 unsigned int domain, unsigned int bus,
418 struct pci_bus *b)
419{
420 struct pci_dev *d;
421 unsigned int devfn;
422
423 /* Scan the bus for functions and add.
424 * We omit handling of PCI bridge attachment because pciback prevents
425 * bridges from being exported.
426 */
427 for (devfn = 0; devfn < 0x100; devfn++) {
428 d = pci_get_slot(b, devfn);
429 if (d) {
430 /* Device is already known. */
431 pci_dev_put(d);
432 continue;
433 }
434
435 d = pci_scan_single_device(b, devfn);
436 if (d)
437 dev_info(&pdev->xdev->dev, "New device on "
438 "%04x:%02x:%02x.%d found.\n", domain, bus,
439 PCI_SLOT(devfn), PCI_FUNC(devfn));
440 }
441
442 return 0;
443}
444
445static int pcifront_scan_root(struct pcifront_device *pdev,
446 unsigned int domain, unsigned int bus)
447{
448 struct pci_bus *b;
449 LIST_HEAD(resources);
450 struct pcifront_sd *sd = NULL;
451 struct pci_bus_entry *bus_entry = NULL;
452 int err = 0;
453 static struct resource busn_res = {
454 .start = 0,
455 .end = 255,
456 .flags = IORESOURCE_BUS,
457 };
458
459#ifndef CONFIG_PCI_DOMAINS
460 if (domain != 0) {
461 dev_err(&pdev->xdev->dev,
462 "PCI Root in non-zero PCI Domain! domain=%d\n", domain);
463 dev_err(&pdev->xdev->dev,
464 "Please compile with CONFIG_PCI_DOMAINS\n");
465 err = -EINVAL;
466 goto err_out;
467 }
468#endif
469
470 dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n",
471 domain, bus);
472
473 bus_entry = kzalloc(sizeof(*bus_entry), GFP_KERNEL);
474 sd = kzalloc(sizeof(*sd), GFP_KERNEL);
475 if (!bus_entry || !sd) {
476 err = -ENOMEM;
477 goto err_out;
478 }
479 pci_add_resource(&resources, &ioport_resource);
480 pci_add_resource(&resources, &iomem_resource);
481 pci_add_resource(&resources, &busn_res);
482 pcifront_init_sd(sd, domain, bus, pdev);
483
484 pci_lock_rescan_remove();
485
486 b = pci_scan_root_bus(&pdev->xdev->dev, bus,
487 &pcifront_bus_ops, sd, &resources);
488 if (!b) {
489 dev_err(&pdev->xdev->dev,
490 "Error creating PCI Frontend Bus!\n");
491 err = -ENOMEM;
492 pci_unlock_rescan_remove();
493 pci_free_resource_list(&resources);
494 goto err_out;
495 }
496
497 bus_entry->bus = b;
498
499 list_add(&bus_entry->list, &pdev->root_buses);
500
501 /* pci_scan_root_bus skips devices which do not have a
502 * devfn==0. The pcifront_scan_bus enumerates all devfn. */
503 err = pcifront_scan_bus(pdev, domain, bus, b);
504
505 /* Claim resources before going "live" with our devices */
506 pci_walk_bus(b, pcifront_claim_resource, pdev);
507
508 /* Create SysFS and notify udev of the devices. Aka: "going live" */
509 pci_bus_add_devices(b);
510
511 pci_unlock_rescan_remove();
512 return err;
513
514err_out:
515 kfree(bus_entry);
516 kfree(sd);
517
518 return err;
519}
520
521static int pcifront_rescan_root(struct pcifront_device *pdev,
522 unsigned int domain, unsigned int bus)
523{
524 int err;
525 struct pci_bus *b;
526
527#ifndef CONFIG_PCI_DOMAINS
528 if (domain != 0) {
529 dev_err(&pdev->xdev->dev,
530 "PCI Root in non-zero PCI Domain! domain=%d\n", domain);
531 dev_err(&pdev->xdev->dev,
532 "Please compile with CONFIG_PCI_DOMAINS\n");
533 return -EINVAL;
534 }
535#endif
536
537 dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n",
538 domain, bus);
539
540 b = pci_find_bus(domain, bus);
541 if (!b)
542 /* If the bus is unknown, create it. */
543 return pcifront_scan_root(pdev, domain, bus);
544
545 err = pcifront_scan_bus(pdev, domain, bus, b);
546
547 /* Claim resources before going "live" with our devices */
548 pci_walk_bus(b, pcifront_claim_resource, pdev);
549
550 /* Create SysFS and notify udev of the devices. Aka: "going live" */
551 pci_bus_add_devices(b);
552
553 return err;
554}
555
556static void free_root_bus_devs(struct pci_bus *bus)
557{
558 struct pci_dev *dev;
559
560 while (!list_empty(&bus->devices)) {
561 dev = container_of(bus->devices.next, struct pci_dev,
562 bus_list);
563 pci_dbg(dev, "removing device\n");
564 pci_stop_and_remove_bus_device(dev);
565 }
566}
567
568static void pcifront_free_roots(struct pcifront_device *pdev)
569{
570 struct pci_bus_entry *bus_entry, *t;
571
572 dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n");
573
574 pci_lock_rescan_remove();
575 list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) {
576 list_del(&bus_entry->list);
577
578 free_root_bus_devs(bus_entry->bus);
579
580 kfree(bus_entry->bus->sysdata);
581
582 device_unregister(bus_entry->bus->bridge);
583 pci_remove_bus(bus_entry->bus);
584
585 kfree(bus_entry);
586 }
587 pci_unlock_rescan_remove();
588}
589
590static pci_ers_result_t pcifront_common_process(int cmd,
591 struct pcifront_device *pdev,
592 pci_channel_state_t state)
593{
594 pci_ers_result_t result;
595 struct pci_driver *pdrv;
596 int bus = pdev->sh_info->aer_op.bus;
597 int devfn = pdev->sh_info->aer_op.devfn;
598 int domain = pdev->sh_info->aer_op.domain;
599 struct pci_dev *pcidev;
600 int flag = 0;
601
602 dev_dbg(&pdev->xdev->dev,
603 "pcifront AER process: cmd %x (bus:%x, devfn%x)",
604 cmd, bus, devfn);
605 result = PCI_ERS_RESULT_NONE;
606
607 pcidev = pci_get_domain_bus_and_slot(domain, bus, devfn);
608 if (!pcidev || !pcidev->driver) {
609 dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n");
610 pci_dev_put(pcidev);
611 return result;
612 }
613 pdrv = pcidev->driver;
614
615 if (pdrv) {
616 if (pdrv->err_handler && pdrv->err_handler->error_detected) {
617 pci_dbg(pcidev, "trying to call AER service\n");
618 if (pcidev) {
619 flag = 1;
620 switch (cmd) {
621 case XEN_PCI_OP_aer_detected:
622 result = pdrv->err_handler->
623 error_detected(pcidev, state);
624 break;
625 case XEN_PCI_OP_aer_mmio:
626 result = pdrv->err_handler->
627 mmio_enabled(pcidev);
628 break;
629 case XEN_PCI_OP_aer_slotreset:
630 result = pdrv->err_handler->
631 slot_reset(pcidev);
632 break;
633 case XEN_PCI_OP_aer_resume:
634 pdrv->err_handler->resume(pcidev);
635 break;
636 default:
637 dev_err(&pdev->xdev->dev,
638 "bad request in aer recovery "
639 "operation!\n");
640
641 }
642 }
643 }
644 }
645 if (!flag)
646 result = PCI_ERS_RESULT_NONE;
647
648 return result;
649}
650
651
652static void pcifront_do_aer(struct work_struct *data)
653{
654 struct pcifront_device *pdev =
655 container_of(data, struct pcifront_device, op_work);
656 int cmd = pdev->sh_info->aer_op.cmd;
657 pci_channel_state_t state =
658 (pci_channel_state_t)pdev->sh_info->aer_op.err;
659
660 /*If a pci_conf op is in progress,
661 we have to wait until it is done before service aer op*/
662 dev_dbg(&pdev->xdev->dev,
663 "pcifront service aer bus %x devfn %x\n",
664 pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn);
665
666 pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state);
667
668 /* Post the operation to the guest. */
669 wmb();
670 clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags);
671 notify_remote_via_evtchn(pdev->evtchn);
672
673 /*in case of we lost an aer request in four lines time_window*/
674 smp_mb__before_atomic();
675 clear_bit(_PDEVB_op_active, &pdev->flags);
676 smp_mb__after_atomic();
677
678 schedule_pcifront_aer_op(pdev);
679
680}
681
682static irqreturn_t pcifront_handler_aer(int irq, void *dev)
683{
684 struct pcifront_device *pdev = dev;
685 schedule_pcifront_aer_op(pdev);
686 return IRQ_HANDLED;
687}
688static int pcifront_connect_and_init_dma(struct pcifront_device *pdev)
689{
690 int err = 0;
691
692 spin_lock(&pcifront_dev_lock);
693
694 if (!pcifront_dev) {
695 dev_info(&pdev->xdev->dev, "Installing PCI frontend\n");
696 pcifront_dev = pdev;
697 } else
698 err = -EEXIST;
699
700 spin_unlock(&pcifront_dev_lock);
701
702 if (!err && !swiotlb_nr_tbl()) {
703 err = pci_xen_swiotlb_init_late();
704 if (err)
705 dev_err(&pdev->xdev->dev, "Could not setup SWIOTLB!\n");
706 }
707 return err;
708}
709
710static void pcifront_disconnect(struct pcifront_device *pdev)
711{
712 spin_lock(&pcifront_dev_lock);
713
714 if (pdev == pcifront_dev) {
715 dev_info(&pdev->xdev->dev,
716 "Disconnecting PCI Frontend Buses\n");
717 pcifront_dev = NULL;
718 }
719
720 spin_unlock(&pcifront_dev_lock);
721}
722static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev)
723{
724 struct pcifront_device *pdev;
725
726 pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL);
727 if (pdev == NULL)
728 goto out;
729
730 pdev->sh_info =
731 (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL);
732 if (pdev->sh_info == NULL) {
733 kfree(pdev);
734 pdev = NULL;
735 goto out;
736 }
737 pdev->sh_info->flags = 0;
738
739 /*Flag for registering PV AER handler*/
740 set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags);
741
742 dev_set_drvdata(&xdev->dev, pdev);
743 pdev->xdev = xdev;
744
745 INIT_LIST_HEAD(&pdev->root_buses);
746
747 spin_lock_init(&pdev->sh_info_lock);
748
749 pdev->evtchn = INVALID_EVTCHN;
750 pdev->gnt_ref = INVALID_GRANT_REF;
751 pdev->irq = -1;
752
753 INIT_WORK(&pdev->op_work, pcifront_do_aer);
754
755 dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n",
756 pdev, pdev->sh_info);
757out:
758 return pdev;
759}
760
761static void free_pdev(struct pcifront_device *pdev)
762{
763 dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev);
764
765 pcifront_free_roots(pdev);
766
767 cancel_work_sync(&pdev->op_work);
768
769 if (pdev->irq >= 0)
770 unbind_from_irqhandler(pdev->irq, pdev);
771
772 if (pdev->evtchn != INVALID_EVTCHN)
773 xenbus_free_evtchn(pdev->xdev, pdev->evtchn);
774
775 if (pdev->gnt_ref != INVALID_GRANT_REF)
776 gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */,
777 (unsigned long)pdev->sh_info);
778 else
779 free_page((unsigned long)pdev->sh_info);
780
781 dev_set_drvdata(&pdev->xdev->dev, NULL);
782
783 kfree(pdev);
784}
785
786static int pcifront_publish_info(struct pcifront_device *pdev)
787{
788 int err = 0;
789 struct xenbus_transaction trans;
790 grant_ref_t gref;
791
792 err = xenbus_grant_ring(pdev->xdev, pdev->sh_info, 1, &gref);
793 if (err < 0)
794 goto out;
795
796 pdev->gnt_ref = gref;
797
798 err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn);
799 if (err)
800 goto out;
801
802 err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer,
803 0, "pcifront", pdev);
804
805 if (err < 0)
806 return err;
807
808 pdev->irq = err;
809
810do_publish:
811 err = xenbus_transaction_start(&trans);
812 if (err) {
813 xenbus_dev_fatal(pdev->xdev, err,
814 "Error writing configuration for backend "
815 "(start transaction)");
816 goto out;
817 }
818
819 err = xenbus_printf(trans, pdev->xdev->nodename,
820 "pci-op-ref", "%u", pdev->gnt_ref);
821 if (!err)
822 err = xenbus_printf(trans, pdev->xdev->nodename,
823 "event-channel", "%u", pdev->evtchn);
824 if (!err)
825 err = xenbus_printf(trans, pdev->xdev->nodename,
826 "magic", XEN_PCI_MAGIC);
827
828 if (err) {
829 xenbus_transaction_end(trans, 1);
830 xenbus_dev_fatal(pdev->xdev, err,
831 "Error writing configuration for backend");
832 goto out;
833 } else {
834 err = xenbus_transaction_end(trans, 0);
835 if (err == -EAGAIN)
836 goto do_publish;
837 else if (err) {
838 xenbus_dev_fatal(pdev->xdev, err,
839 "Error completing transaction "
840 "for backend");
841 goto out;
842 }
843 }
844
845 xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
846
847 dev_dbg(&pdev->xdev->dev, "publishing successful!\n");
848
849out:
850 return err;
851}
852
853static int pcifront_try_connect(struct pcifront_device *pdev)
854{
855 int err = -EFAULT;
856 int i, num_roots, len;
857 char str[64];
858 unsigned int domain, bus;
859
860
861 /* Only connect once */
862 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
863 XenbusStateInitialised)
864 goto out;
865
866 err = pcifront_connect_and_init_dma(pdev);
867 if (err && err != -EEXIST) {
868 xenbus_dev_fatal(pdev->xdev, err,
869 "Error setting up PCI Frontend");
870 goto out;
871 }
872
873 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
874 "root_num", "%d", &num_roots);
875 if (err == -ENOENT) {
876 xenbus_dev_error(pdev->xdev, err,
877 "No PCI Roots found, trying 0000:00");
878 err = pcifront_scan_root(pdev, 0, 0);
879 if (err) {
880 xenbus_dev_fatal(pdev->xdev, err,
881 "Error scanning PCI root 0000:00");
882 goto out;
883 }
884 num_roots = 0;
885 } else if (err != 1) {
886 if (err == 0)
887 err = -EINVAL;
888 xenbus_dev_fatal(pdev->xdev, err,
889 "Error reading number of PCI roots");
890 goto out;
891 }
892
893 for (i = 0; i < num_roots; i++) {
894 len = snprintf(str, sizeof(str), "root-%d", i);
895 if (unlikely(len >= (sizeof(str) - 1))) {
896 err = -ENOMEM;
897 goto out;
898 }
899
900 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
901 "%x:%x", &domain, &bus);
902 if (err != 2) {
903 if (err >= 0)
904 err = -EINVAL;
905 xenbus_dev_fatal(pdev->xdev, err,
906 "Error reading PCI root %d", i);
907 goto out;
908 }
909
910 err = pcifront_scan_root(pdev, domain, bus);
911 if (err) {
912 xenbus_dev_fatal(pdev->xdev, err,
913 "Error scanning PCI root %04x:%02x",
914 domain, bus);
915 goto out;
916 }
917 }
918
919 err = xenbus_switch_state(pdev->xdev, XenbusStateConnected);
920
921out:
922 return err;
923}
924
925static int pcifront_try_disconnect(struct pcifront_device *pdev)
926{
927 int err = 0;
928 enum xenbus_state prev_state;
929
930
931 prev_state = xenbus_read_driver_state(pdev->xdev->nodename);
932
933 if (prev_state >= XenbusStateClosing)
934 goto out;
935
936 if (prev_state == XenbusStateConnected) {
937 pcifront_free_roots(pdev);
938 pcifront_disconnect(pdev);
939 }
940
941 err = xenbus_switch_state(pdev->xdev, XenbusStateClosed);
942
943out:
944
945 return err;
946}
947
948static int pcifront_attach_devices(struct pcifront_device *pdev)
949{
950 int err = -EFAULT;
951 int i, num_roots, len;
952 unsigned int domain, bus;
953 char str[64];
954
955 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
956 XenbusStateReconfiguring)
957 goto out;
958
959 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
960 "root_num", "%d", &num_roots);
961 if (err == -ENOENT) {
962 xenbus_dev_error(pdev->xdev, err,
963 "No PCI Roots found, trying 0000:00");
964 err = pcifront_rescan_root(pdev, 0, 0);
965 if (err) {
966 xenbus_dev_fatal(pdev->xdev, err,
967 "Error scanning PCI root 0000:00");
968 goto out;
969 }
970 num_roots = 0;
971 } else if (err != 1) {
972 if (err == 0)
973 err = -EINVAL;
974 xenbus_dev_fatal(pdev->xdev, err,
975 "Error reading number of PCI roots");
976 goto out;
977 }
978
979 for (i = 0; i < num_roots; i++) {
980 len = snprintf(str, sizeof(str), "root-%d", i);
981 if (unlikely(len >= (sizeof(str) - 1))) {
982 err = -ENOMEM;
983 goto out;
984 }
985
986 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
987 "%x:%x", &domain, &bus);
988 if (err != 2) {
989 if (err >= 0)
990 err = -EINVAL;
991 xenbus_dev_fatal(pdev->xdev, err,
992 "Error reading PCI root %d", i);
993 goto out;
994 }
995
996 err = pcifront_rescan_root(pdev, domain, bus);
997 if (err) {
998 xenbus_dev_fatal(pdev->xdev, err,
999 "Error scanning PCI root %04x:%02x",
1000 domain, bus);
1001 goto out;
1002 }
1003 }
1004
1005 xenbus_switch_state(pdev->xdev, XenbusStateConnected);
1006
1007out:
1008 return err;
1009}
1010
1011static int pcifront_detach_devices(struct pcifront_device *pdev)
1012{
1013 int err = 0;
1014 int i, num_devs;
1015 unsigned int domain, bus, slot, func;
1016 struct pci_dev *pci_dev;
1017 char str[64];
1018
1019 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
1020 XenbusStateConnected)
1021 goto out;
1022
1023 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d",
1024 &num_devs);
1025 if (err != 1) {
1026 if (err >= 0)
1027 err = -EINVAL;
1028 xenbus_dev_fatal(pdev->xdev, err,
1029 "Error reading number of PCI devices");
1030 goto out;
1031 }
1032
1033 /* Find devices being detached and remove them. */
1034 for (i = 0; i < num_devs; i++) {
1035 int l, state;
1036 l = snprintf(str, sizeof(str), "state-%d", i);
1037 if (unlikely(l >= (sizeof(str) - 1))) {
1038 err = -ENOMEM;
1039 goto out;
1040 }
1041 state = xenbus_read_unsigned(pdev->xdev->otherend, str,
1042 XenbusStateUnknown);
1043
1044 if (state != XenbusStateClosing)
1045 continue;
1046
1047 /* Remove device. */
1048 l = snprintf(str, sizeof(str), "vdev-%d", i);
1049 if (unlikely(l >= (sizeof(str) - 1))) {
1050 err = -ENOMEM;
1051 goto out;
1052 }
1053 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
1054 "%x:%x:%x.%x", &domain, &bus, &slot, &func);
1055 if (err != 4) {
1056 if (err >= 0)
1057 err = -EINVAL;
1058 xenbus_dev_fatal(pdev->xdev, err,
1059 "Error reading PCI device %d", i);
1060 goto out;
1061 }
1062
1063 pci_dev = pci_get_domain_bus_and_slot(domain, bus,
1064 PCI_DEVFN(slot, func));
1065 if (!pci_dev) {
1066 dev_dbg(&pdev->xdev->dev,
1067 "Cannot get PCI device %04x:%02x:%02x.%d\n",
1068 domain, bus, slot, func);
1069 continue;
1070 }
1071 pci_lock_rescan_remove();
1072 pci_stop_and_remove_bus_device(pci_dev);
1073 pci_dev_put(pci_dev);
1074 pci_unlock_rescan_remove();
1075
1076 dev_dbg(&pdev->xdev->dev,
1077 "PCI device %04x:%02x:%02x.%d removed.\n",
1078 domain, bus, slot, func);
1079 }
1080
1081 err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring);
1082
1083out:
1084 return err;
1085}
1086
1087static void __ref pcifront_backend_changed(struct xenbus_device *xdev,
1088 enum xenbus_state be_state)
1089{
1090 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1091
1092 switch (be_state) {
1093 case XenbusStateUnknown:
1094 case XenbusStateInitialising:
1095 case XenbusStateInitWait:
1096 case XenbusStateInitialised:
1097 break;
1098
1099 case XenbusStateConnected:
1100 pcifront_try_connect(pdev);
1101 break;
1102
1103 case XenbusStateClosed:
1104 if (xdev->state == XenbusStateClosed)
1105 break;
1106 /* fall through - Missed the backend's CLOSING state. */
1107 case XenbusStateClosing:
1108 dev_warn(&xdev->dev, "backend going away!\n");
1109 pcifront_try_disconnect(pdev);
1110 break;
1111
1112 case XenbusStateReconfiguring:
1113 pcifront_detach_devices(pdev);
1114 break;
1115
1116 case XenbusStateReconfigured:
1117 pcifront_attach_devices(pdev);
1118 break;
1119 }
1120}
1121
1122static int pcifront_xenbus_probe(struct xenbus_device *xdev,
1123 const struct xenbus_device_id *id)
1124{
1125 int err = 0;
1126 struct pcifront_device *pdev = alloc_pdev(xdev);
1127
1128 if (pdev == NULL) {
1129 err = -ENOMEM;
1130 xenbus_dev_fatal(xdev, err,
1131 "Error allocating pcifront_device struct");
1132 goto out;
1133 }
1134
1135 err = pcifront_publish_info(pdev);
1136 if (err)
1137 free_pdev(pdev);
1138
1139out:
1140 return err;
1141}
1142
1143static int pcifront_xenbus_remove(struct xenbus_device *xdev)
1144{
1145 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1146 if (pdev)
1147 free_pdev(pdev);
1148
1149 return 0;
1150}
1151
1152static const struct xenbus_device_id xenpci_ids[] = {
1153 {"pci"},
1154 {""},
1155};
1156
1157static struct xenbus_driver xenpci_driver = {
1158 .name = "pcifront",
1159 .ids = xenpci_ids,
1160 .probe = pcifront_xenbus_probe,
1161 .remove = pcifront_xenbus_remove,
1162 .otherend_changed = pcifront_backend_changed,
1163};
1164
1165static int __init pcifront_init(void)
1166{
1167 if (!xen_pv_domain() || xen_initial_domain())
1168 return -ENODEV;
1169
1170 if (!xen_has_pv_devices())
1171 return -ENODEV;
1172
1173 pci_frontend_registrar(1 /* enable */);
1174
1175 return xenbus_register_frontend(&xenpci_driver);
1176}
1177
1178static void __exit pcifront_cleanup(void)
1179{
1180 xenbus_unregister_driver(&xenpci_driver);
1181 pci_frontend_registrar(0 /* disable */);
1182}
1183module_init(pcifront_init);
1184module_exit(pcifront_cleanup);
1185
1186MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
1187MODULE_LICENSE("GPL");
1188MODULE_ALIAS("xen:pci");
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
27struct 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
35struct 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
52struct pcifront_sd {
53 int domain;
54 struct pcifront_device *pdev;
55};
56
57static inline struct pcifront_device *
58pcifront_get_pdev(struct pcifront_sd *sd)
59{
60 return sd->pdev;
61}
62
63static 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
71static DEFINE_SPINLOCK(pcifront_dev_lock);
72static struct pcifront_device *pcifront_dev;
73
74static int verbose_request;
75module_param(verbose_request, int, 0644);
76
77static 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
99static 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
108static 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;
169out:
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 */
175static 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 */
214static 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
239struct pci_ops pcifront_bus_ops = {
240 .read = pcifront_bus_read,
241 .write = pcifront_bus_write,
242};
243
244#ifdef CONFIG_PCI_MSI
245static 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
301static 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
320static 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
349static 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
372static 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
379static 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
387static 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 */
391static 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
414static 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
443static 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
498err_out:
499 kfree(bus_entry);
500 kfree(sd);
501
502 return err;
503}
504
505static 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
540static 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
552static 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
572static 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
635static 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
665static 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}
671static 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
690static 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}
702static 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);
737out:
738 return pdev;
739}
740
741static 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
766static 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
789do_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
828out:
829 return err;
830}
831
832static 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
895out:
896 return err;
897}
898
899static 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
917out:
918
919 return err;
920}
921
922static 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
976out:
977 return err;
978}
979
980static 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
1058out:
1059 return err;
1060}
1061
1062static 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
1094static 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
1111out:
1112 return err;
1113}
1114
1115static 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
1124static const struct xenbus_device_id xenpci_ids[] = {
1125 {"pci"},
1126 {""},
1127};
1128
1129static DEFINE_XENBUS_DRIVER(xenpci, "pcifront",
1130 .probe = pcifront_xenbus_probe,
1131 .remove = pcifront_xenbus_remove,
1132 .otherend_changed = pcifront_backend_changed,
1133);
1134
1135static 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
1145static void __exit pcifront_cleanup(void)
1146{
1147 xenbus_unregister_driver(&xenpci_driver);
1148 pci_frontend_registrar(0 /* disable */);
1149}
1150module_init(pcifront_init);
1151module_exit(pcifront_cleanup);
1152
1153MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
1154MODULE_LICENSE("GPL");
1155MODULE_ALIAS("xen:pci");