Loading...
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#include <linux/ktime.h>
24#include <xen/platform_pci.h>
25
26#include <asm/xen/swiotlb-xen.h>
27#define INVALID_GRANT_REF (0)
28#define INVALID_EVTCHN (-1)
29
30struct pci_bus_entry {
31 struct list_head list;
32 struct pci_bus *bus;
33};
34
35#define _PDEVB_op_active (0)
36#define PDEVB_op_active (1 << (_PDEVB_op_active))
37
38struct pcifront_device {
39 struct xenbus_device *xdev;
40 struct list_head root_buses;
41
42 int evtchn;
43 int gnt_ref;
44
45 int irq;
46
47 /* Lock this when doing any operations in sh_info */
48 spinlock_t sh_info_lock;
49 struct xen_pci_sharedinfo *sh_info;
50 struct work_struct op_work;
51 unsigned long flags;
52
53};
54
55struct pcifront_sd {
56 struct pci_sysdata sd;
57 struct pcifront_device *pdev;
58};
59
60static inline struct pcifront_device *
61pcifront_get_pdev(struct pcifront_sd *sd)
62{
63 return sd->pdev;
64}
65
66static inline void pcifront_init_sd(struct pcifront_sd *sd,
67 unsigned int domain, unsigned int bus,
68 struct pcifront_device *pdev)
69{
70 /* Because we do not expose that information via XenBus. */
71 sd->sd.node = first_online_node;
72 sd->sd.domain = domain;
73 sd->pdev = pdev;
74}
75
76static DEFINE_SPINLOCK(pcifront_dev_lock);
77static struct pcifront_device *pcifront_dev;
78
79static int verbose_request;
80module_param(verbose_request, int, 0644);
81
82static int errno_to_pcibios_err(int errno)
83{
84 switch (errno) {
85 case XEN_PCI_ERR_success:
86 return PCIBIOS_SUCCESSFUL;
87
88 case XEN_PCI_ERR_dev_not_found:
89 return PCIBIOS_DEVICE_NOT_FOUND;
90
91 case XEN_PCI_ERR_invalid_offset:
92 case XEN_PCI_ERR_op_failed:
93 return PCIBIOS_BAD_REGISTER_NUMBER;
94
95 case XEN_PCI_ERR_not_implemented:
96 return PCIBIOS_FUNC_NOT_SUPPORTED;
97
98 case XEN_PCI_ERR_access_denied:
99 return PCIBIOS_SET_FAILED;
100 }
101 return errno;
102}
103
104static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev)
105{
106 if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
107 && !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) {
108 dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n");
109 schedule_work(&pdev->op_work);
110 }
111}
112
113static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op)
114{
115 int err = 0;
116 struct xen_pci_op *active_op = &pdev->sh_info->op;
117 unsigned long irq_flags;
118 evtchn_port_t port = pdev->evtchn;
119 unsigned irq = pdev->irq;
120 s64 ns, ns_timeout;
121
122 spin_lock_irqsave(&pdev->sh_info_lock, irq_flags);
123
124 memcpy(active_op, op, sizeof(struct xen_pci_op));
125
126 /* Go */
127 wmb();
128 set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
129 notify_remote_via_evtchn(port);
130
131 /*
132 * We set a poll timeout of 3 seconds but give up on return after
133 * 2 seconds. It is better to time out too late rather than too early
134 * (in the latter case we end up continually re-executing poll() with a
135 * timeout in the past). 1s difference gives plenty of slack for error.
136 */
137 ns_timeout = ktime_get_ns() + 2 * (s64)NSEC_PER_SEC;
138
139 xen_clear_irq_pending(irq);
140
141 while (test_bit(_XEN_PCIF_active,
142 (unsigned long *)&pdev->sh_info->flags)) {
143 xen_poll_irq_timeout(irq, jiffies + 3*HZ);
144 xen_clear_irq_pending(irq);
145 ns = ktime_get_ns();
146 if (ns > ns_timeout) {
147 dev_err(&pdev->xdev->dev,
148 "pciback not responding!!!\n");
149 clear_bit(_XEN_PCIF_active,
150 (unsigned long *)&pdev->sh_info->flags);
151 err = XEN_PCI_ERR_dev_not_found;
152 goto out;
153 }
154 }
155
156 /*
157 * We might lose backend service request since we
158 * reuse same evtchn with pci_conf backend response. So re-schedule
159 * aer pcifront service.
160 */
161 if (test_bit(_XEN_PCIB_active,
162 (unsigned long *)&pdev->sh_info->flags)) {
163 dev_err(&pdev->xdev->dev,
164 "schedule aer pcifront service\n");
165 schedule_pcifront_aer_op(pdev);
166 }
167
168 memcpy(op, active_op, sizeof(struct xen_pci_op));
169
170 err = op->err;
171out:
172 spin_unlock_irqrestore(&pdev->sh_info_lock, irq_flags);
173 return err;
174}
175
176/* Access to this function is spinlocked in drivers/pci/access.c */
177static int pcifront_bus_read(struct pci_bus *bus, unsigned int devfn,
178 int where, int size, u32 *val)
179{
180 int err = 0;
181 struct xen_pci_op op = {
182 .cmd = XEN_PCI_OP_conf_read,
183 .domain = pci_domain_nr(bus),
184 .bus = bus->number,
185 .devfn = devfn,
186 .offset = where,
187 .size = size,
188 };
189 struct pcifront_sd *sd = bus->sysdata;
190 struct pcifront_device *pdev = pcifront_get_pdev(sd);
191
192 if (verbose_request)
193 dev_info(&pdev->xdev->dev,
194 "read dev=%04x:%02x:%02x.%d - offset %x size %d\n",
195 pci_domain_nr(bus), bus->number, PCI_SLOT(devfn),
196 PCI_FUNC(devfn), where, size);
197
198 err = do_pci_op(pdev, &op);
199
200 if (likely(!err)) {
201 if (verbose_request)
202 dev_info(&pdev->xdev->dev, "read got back value %x\n",
203 op.value);
204
205 *val = op.value;
206 } else if (err == -ENODEV) {
207 /* No device here, pretend that it just returned 0 */
208 err = 0;
209 *val = 0;
210 }
211
212 return errno_to_pcibios_err(err);
213}
214
215/* Access to this function is spinlocked in drivers/pci/access.c */
216static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn,
217 int where, int size, u32 val)
218{
219 struct xen_pci_op op = {
220 .cmd = XEN_PCI_OP_conf_write,
221 .domain = pci_domain_nr(bus),
222 .bus = bus->number,
223 .devfn = devfn,
224 .offset = where,
225 .size = size,
226 .value = val,
227 };
228 struct pcifront_sd *sd = bus->sysdata;
229 struct pcifront_device *pdev = pcifront_get_pdev(sd);
230
231 if (verbose_request)
232 dev_info(&pdev->xdev->dev,
233 "write dev=%04x:%02x:%02x.%d - "
234 "offset %x size %d val %x\n",
235 pci_domain_nr(bus), bus->number,
236 PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val);
237
238 return errno_to_pcibios_err(do_pci_op(pdev, &op));
239}
240
241static struct pci_ops pcifront_bus_ops = {
242 .read = pcifront_bus_read,
243 .write = pcifront_bus_write,
244};
245
246#ifdef CONFIG_PCI_MSI
247static int pci_frontend_enable_msix(struct pci_dev *dev,
248 int vector[], int nvec)
249{
250 int err;
251 int i;
252 struct xen_pci_op op = {
253 .cmd = XEN_PCI_OP_enable_msix,
254 .domain = pci_domain_nr(dev->bus),
255 .bus = dev->bus->number,
256 .devfn = dev->devfn,
257 .value = nvec,
258 };
259 struct pcifront_sd *sd = dev->bus->sysdata;
260 struct pcifront_device *pdev = pcifront_get_pdev(sd);
261 struct msi_desc *entry;
262
263 if (nvec > SH_INFO_MAX_VEC) {
264 dev_err(&dev->dev, "too much vector for pci frontend: %x."
265 " Increase SH_INFO_MAX_VEC.\n", nvec);
266 return -EINVAL;
267 }
268
269 i = 0;
270 for_each_pci_msi_entry(entry, dev) {
271 op.msix_entries[i].entry = entry->msi_attrib.entry_nr;
272 /* Vector is useless at this point. */
273 op.msix_entries[i].vector = -1;
274 i++;
275 }
276
277 err = do_pci_op(pdev, &op);
278
279 if (likely(!err)) {
280 if (likely(!op.value)) {
281 /* we get the result */
282 for (i = 0; i < nvec; i++) {
283 if (op.msix_entries[i].vector <= 0) {
284 dev_warn(&dev->dev, "MSI-X entry %d is invalid: %d!\n",
285 i, op.msix_entries[i].vector);
286 err = -EINVAL;
287 vector[i] = -1;
288 continue;
289 }
290 vector[i] = op.msix_entries[i].vector;
291 }
292 } else {
293 printk(KERN_DEBUG "enable msix get value %x\n",
294 op.value);
295 err = op.value;
296 }
297 } else {
298 dev_err(&dev->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 dev_err(&dev->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 dev_warn(&dev->dev, "MSI entry is invalid: %d!\n",
339 op.value);
340 err = -EINVAL;
341 vector[0] = -1;
342 }
343 } else {
344 dev_err(&dev->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 printk(KERN_DEBUG "get no response from backend for disable MSI\n");
367 return;
368 }
369 if (err)
370 /* how can pciback notify us fail? */
371 printk(KERN_DEBUG "get fake response frombackend\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 dev_dbg(&dev->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 struct pci_dev *pcidev;
599 int flag = 0;
600
601 dev_dbg(&pdev->xdev->dev,
602 "pcifront AER process: cmd %x (bus:%x, devfn%x)",
603 cmd, bus, devfn);
604 result = PCI_ERS_RESULT_NONE;
605
606 pcidev = pci_get_bus_and_slot(bus, devfn);
607 if (!pcidev || !pcidev->driver) {
608 dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n");
609 pci_dev_put(pcidev);
610 return result;
611 }
612 pdrv = pcidev->driver;
613
614 if (pdrv) {
615 if (pdrv->err_handler && pdrv->err_handler->error_detected) {
616 dev_dbg(&pcidev->dev,
617 "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 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, "%d",
1042 &state);
1043 if (err != 1)
1044 state = XenbusStateUnknown;
1045
1046 if (state != XenbusStateClosing)
1047 continue;
1048
1049 /* Remove device. */
1050 l = snprintf(str, sizeof(str), "vdev-%d", i);
1051 if (unlikely(l >= (sizeof(str) - 1))) {
1052 err = -ENOMEM;
1053 goto out;
1054 }
1055 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
1056 "%x:%x:%x.%x", &domain, &bus, &slot, &func);
1057 if (err != 4) {
1058 if (err >= 0)
1059 err = -EINVAL;
1060 xenbus_dev_fatal(pdev->xdev, err,
1061 "Error reading PCI device %d", i);
1062 goto out;
1063 }
1064
1065 pci_dev = pci_get_domain_bus_and_slot(domain, bus,
1066 PCI_DEVFN(slot, func));
1067 if (!pci_dev) {
1068 dev_dbg(&pdev->xdev->dev,
1069 "Cannot get PCI device %04x:%02x:%02x.%d\n",
1070 domain, bus, slot, func);
1071 continue;
1072 }
1073 pci_lock_rescan_remove();
1074 pci_stop_and_remove_bus_device(pci_dev);
1075 pci_dev_put(pci_dev);
1076 pci_unlock_rescan_remove();
1077
1078 dev_dbg(&pdev->xdev->dev,
1079 "PCI device %04x:%02x:%02x.%d removed.\n",
1080 domain, bus, slot, func);
1081 }
1082
1083 err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring);
1084
1085out:
1086 return err;
1087}
1088
1089static void __init_refok pcifront_backend_changed(struct xenbus_device *xdev,
1090 enum xenbus_state be_state)
1091{
1092 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1093
1094 switch (be_state) {
1095 case XenbusStateUnknown:
1096 case XenbusStateInitialising:
1097 case XenbusStateInitWait:
1098 case XenbusStateInitialised:
1099 break;
1100
1101 case XenbusStateConnected:
1102 pcifront_try_connect(pdev);
1103 break;
1104
1105 case XenbusStateClosed:
1106 if (xdev->state == XenbusStateClosed)
1107 break;
1108 /* Missed the backend's CLOSING state -- fallthrough */
1109 case XenbusStateClosing:
1110 dev_warn(&xdev->dev, "backend going away!\n");
1111 pcifront_try_disconnect(pdev);
1112 break;
1113
1114 case XenbusStateReconfiguring:
1115 pcifront_detach_devices(pdev);
1116 break;
1117
1118 case XenbusStateReconfigured:
1119 pcifront_attach_devices(pdev);
1120 break;
1121 }
1122}
1123
1124static int pcifront_xenbus_probe(struct xenbus_device *xdev,
1125 const struct xenbus_device_id *id)
1126{
1127 int err = 0;
1128 struct pcifront_device *pdev = alloc_pdev(xdev);
1129
1130 if (pdev == NULL) {
1131 err = -ENOMEM;
1132 xenbus_dev_fatal(xdev, err,
1133 "Error allocating pcifront_device struct");
1134 goto out;
1135 }
1136
1137 err = pcifront_publish_info(pdev);
1138 if (err)
1139 free_pdev(pdev);
1140
1141out:
1142 return err;
1143}
1144
1145static int pcifront_xenbus_remove(struct xenbus_device *xdev)
1146{
1147 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1148 if (pdev)
1149 free_pdev(pdev);
1150
1151 return 0;
1152}
1153
1154static const struct xenbus_device_id xenpci_ids[] = {
1155 {"pci"},
1156 {""},
1157};
1158
1159static struct xenbus_driver xenpci_driver = {
1160 .name = "pcifront",
1161 .ids = xenpci_ids,
1162 .probe = pcifront_xenbus_probe,
1163 .remove = pcifront_xenbus_remove,
1164 .otherend_changed = pcifront_backend_changed,
1165};
1166
1167static int __init pcifront_init(void)
1168{
1169 if (!xen_pv_domain() || xen_initial_domain())
1170 return -ENODEV;
1171
1172 if (!xen_has_pv_devices())
1173 return -ENODEV;
1174
1175 pci_frontend_registrar(1 /* enable */);
1176
1177 return xenbus_register_frontend(&xenpci_driver);
1178}
1179
1180static void __exit pcifront_cleanup(void)
1181{
1182 xenbus_unregister_driver(&xenpci_driver);
1183 pci_frontend_registrar(0 /* disable */);
1184}
1185module_init(pcifront_init);
1186module_exit(pcifront_cleanup);
1187
1188MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
1189MODULE_LICENSE("GPL");
1190MODULE_ALIAS("xen:pci");
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 errno_to_pcibios_err(int errno)
81{
82 switch (errno) {
83 case XEN_PCI_ERR_success:
84 return PCIBIOS_SUCCESSFUL;
85
86 case XEN_PCI_ERR_dev_not_found:
87 return PCIBIOS_DEVICE_NOT_FOUND;
88
89 case XEN_PCI_ERR_invalid_offset:
90 case XEN_PCI_ERR_op_failed:
91 return PCIBIOS_BAD_REGISTER_NUMBER;
92
93 case XEN_PCI_ERR_not_implemented:
94 return PCIBIOS_FUNC_NOT_SUPPORTED;
95
96 case XEN_PCI_ERR_access_denied:
97 return PCIBIOS_SET_FAILED;
98 }
99 return errno;
100}
101
102static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev)
103{
104 if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
105 && !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) {
106 dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n");
107 schedule_work(&pdev->op_work);
108 }
109}
110
111static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op)
112{
113 int err = 0;
114 struct xen_pci_op *active_op = &pdev->sh_info->op;
115 unsigned long irq_flags;
116 evtchn_port_t port = pdev->evtchn;
117 unsigned irq = pdev->irq;
118 s64 ns, ns_timeout;
119
120 spin_lock_irqsave(&pdev->sh_info_lock, irq_flags);
121
122 memcpy(active_op, op, sizeof(struct xen_pci_op));
123
124 /* Go */
125 wmb();
126 set_bit(_XEN_PCIF_active, (unsigned long *)&pdev->sh_info->flags);
127 notify_remote_via_evtchn(port);
128
129 /*
130 * We set a poll timeout of 3 seconds but give up on return after
131 * 2 seconds. It is better to time out too late rather than too early
132 * (in the latter case we end up continually re-executing poll() with a
133 * timeout in the past). 1s difference gives plenty of slack for error.
134 */
135 ns_timeout = ktime_get_ns() + 2 * (s64)NSEC_PER_SEC;
136
137 xen_clear_irq_pending(irq);
138
139 while (test_bit(_XEN_PCIF_active,
140 (unsigned long *)&pdev->sh_info->flags)) {
141 xen_poll_irq_timeout(irq, jiffies + 3*HZ);
142 xen_clear_irq_pending(irq);
143 ns = ktime_get_ns();
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 dev_dbg(&pdev->xdev->dev,
191 "read dev=%04x:%02x:%02x.%d - offset %x size %d\n",
192 pci_domain_nr(bus), bus->number, PCI_SLOT(devfn),
193 PCI_FUNC(devfn), where, size);
194
195 err = do_pci_op(pdev, &op);
196
197 if (likely(!err)) {
198 dev_dbg(&pdev->xdev->dev, "read got back value %x\n",
199 op.value);
200
201 *val = op.value;
202 } else if (err == -ENODEV) {
203 /* No device here, pretend that it just returned 0 */
204 err = 0;
205 *val = 0;
206 }
207
208 return errno_to_pcibios_err(err);
209}
210
211/* Access to this function is spinlocked in drivers/pci/access.c */
212static int pcifront_bus_write(struct pci_bus *bus, unsigned int devfn,
213 int where, int size, u32 val)
214{
215 struct xen_pci_op op = {
216 .cmd = XEN_PCI_OP_conf_write,
217 .domain = pci_domain_nr(bus),
218 .bus = bus->number,
219 .devfn = devfn,
220 .offset = where,
221 .size = size,
222 .value = val,
223 };
224 struct pcifront_sd *sd = bus->sysdata;
225 struct pcifront_device *pdev = pcifront_get_pdev(sd);
226
227 dev_dbg(&pdev->xdev->dev,
228 "write dev=%04x:%02x:%02x.%d - offset %x size %d val %x\n",
229 pci_domain_nr(bus), bus->number,
230 PCI_SLOT(devfn), PCI_FUNC(devfn), where, size, val);
231
232 return errno_to_pcibios_err(do_pci_op(pdev, &op));
233}
234
235static struct pci_ops pcifront_bus_ops = {
236 .read = pcifront_bus_read,
237 .write = pcifront_bus_write,
238};
239
240#ifdef CONFIG_PCI_MSI
241static int pci_frontend_enable_msix(struct pci_dev *dev,
242 int vector[], int nvec)
243{
244 int err;
245 int i;
246 struct xen_pci_op op = {
247 .cmd = XEN_PCI_OP_enable_msix,
248 .domain = pci_domain_nr(dev->bus),
249 .bus = dev->bus->number,
250 .devfn = dev->devfn,
251 .value = nvec,
252 };
253 struct pcifront_sd *sd = dev->bus->sysdata;
254 struct pcifront_device *pdev = pcifront_get_pdev(sd);
255 struct msi_desc *entry;
256
257 if (nvec > SH_INFO_MAX_VEC) {
258 pci_err(dev, "too many vectors (0x%x) for PCI frontend:"
259 " Increase SH_INFO_MAX_VEC\n", nvec);
260 return -EINVAL;
261 }
262
263 i = 0;
264 for_each_pci_msi_entry(entry, dev) {
265 op.msix_entries[i].entry = entry->msi_attrib.entry_nr;
266 /* Vector is useless at this point. */
267 op.msix_entries[i].vector = -1;
268 i++;
269 }
270
271 err = do_pci_op(pdev, &op);
272
273 if (likely(!err)) {
274 if (likely(!op.value)) {
275 /* we get the result */
276 for (i = 0; i < nvec; i++) {
277 if (op.msix_entries[i].vector <= 0) {
278 pci_warn(dev, "MSI-X entry %d is invalid: %d!\n",
279 i, op.msix_entries[i].vector);
280 err = -EINVAL;
281 vector[i] = -1;
282 continue;
283 }
284 vector[i] = op.msix_entries[i].vector;
285 }
286 } else {
287 pr_info("enable msix get value %x\n", op.value);
288 err = op.value;
289 }
290 } else {
291 pci_err(dev, "enable msix get err %x\n", err);
292 }
293 return err;
294}
295
296static void pci_frontend_disable_msix(struct pci_dev *dev)
297{
298 int err;
299 struct xen_pci_op op = {
300 .cmd = XEN_PCI_OP_disable_msix,
301 .domain = pci_domain_nr(dev->bus),
302 .bus = dev->bus->number,
303 .devfn = dev->devfn,
304 };
305 struct pcifront_sd *sd = dev->bus->sysdata;
306 struct pcifront_device *pdev = pcifront_get_pdev(sd);
307
308 err = do_pci_op(pdev, &op);
309
310 /* What should do for error ? */
311 if (err)
312 pci_err(dev, "pci_disable_msix get err %x\n", err);
313}
314
315static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[])
316{
317 int err;
318 struct xen_pci_op op = {
319 .cmd = XEN_PCI_OP_enable_msi,
320 .domain = pci_domain_nr(dev->bus),
321 .bus = dev->bus->number,
322 .devfn = dev->devfn,
323 };
324 struct pcifront_sd *sd = dev->bus->sysdata;
325 struct pcifront_device *pdev = pcifront_get_pdev(sd);
326
327 err = do_pci_op(pdev, &op);
328 if (likely(!err)) {
329 vector[0] = op.value;
330 if (op.value <= 0) {
331 pci_warn(dev, "MSI entry is invalid: %d!\n",
332 op.value);
333 err = -EINVAL;
334 vector[0] = -1;
335 }
336 } else {
337 pci_err(dev, "pci frontend enable msi failed for dev "
338 "%x:%x\n", op.bus, op.devfn);
339 err = -EINVAL;
340 }
341 return err;
342}
343
344static void pci_frontend_disable_msi(struct pci_dev *dev)
345{
346 int err;
347 struct xen_pci_op op = {
348 .cmd = XEN_PCI_OP_disable_msi,
349 .domain = pci_domain_nr(dev->bus),
350 .bus = dev->bus->number,
351 .devfn = dev->devfn,
352 };
353 struct pcifront_sd *sd = dev->bus->sysdata;
354 struct pcifront_device *pdev = pcifront_get_pdev(sd);
355
356 err = do_pci_op(pdev, &op);
357 if (err == XEN_PCI_ERR_dev_not_found) {
358 /* XXX No response from backend, what shall we do? */
359 pr_info("get no response from backend for disable MSI\n");
360 return;
361 }
362 if (err)
363 /* how can pciback notify us fail? */
364 pr_info("get fake response from backend\n");
365}
366
367static struct xen_pci_frontend_ops pci_frontend_ops = {
368 .enable_msi = pci_frontend_enable_msi,
369 .disable_msi = pci_frontend_disable_msi,
370 .enable_msix = pci_frontend_enable_msix,
371 .disable_msix = pci_frontend_disable_msix,
372};
373
374static void pci_frontend_registrar(int enable)
375{
376 if (enable)
377 xen_pci_frontend = &pci_frontend_ops;
378 else
379 xen_pci_frontend = NULL;
380};
381#else
382static inline void pci_frontend_registrar(int enable) { };
383#endif /* CONFIG_PCI_MSI */
384
385/* Claim resources for the PCI frontend as-is, backend won't allow changes */
386static int pcifront_claim_resource(struct pci_dev *dev, void *data)
387{
388 struct pcifront_device *pdev = data;
389 int i;
390 struct resource *r;
391
392 for (i = 0; i < PCI_NUM_RESOURCES; i++) {
393 r = &dev->resource[i];
394
395 if (!r->parent && r->start && r->flags) {
396 dev_info(&pdev->xdev->dev, "claiming resource %s/%d\n",
397 pci_name(dev), i);
398 if (pci_claim_resource(dev, i)) {
399 dev_err(&pdev->xdev->dev, "Could not claim resource %s/%d! "
400 "Device offline. Try using e820_host=1 in the guest config.\n",
401 pci_name(dev), i);
402 }
403 }
404 }
405
406 return 0;
407}
408
409static int pcifront_scan_bus(struct pcifront_device *pdev,
410 unsigned int domain, unsigned int bus,
411 struct pci_bus *b)
412{
413 struct pci_dev *d;
414 unsigned int devfn;
415
416 /* Scan the bus for functions and add.
417 * We omit handling of PCI bridge attachment because pciback prevents
418 * bridges from being exported.
419 */
420 for (devfn = 0; devfn < 0x100; devfn++) {
421 d = pci_get_slot(b, devfn);
422 if (d) {
423 /* Device is already known. */
424 pci_dev_put(d);
425 continue;
426 }
427
428 d = pci_scan_single_device(b, devfn);
429 if (d)
430 dev_info(&pdev->xdev->dev, "New device on "
431 "%04x:%02x:%02x.%d found.\n", domain, bus,
432 PCI_SLOT(devfn), PCI_FUNC(devfn));
433 }
434
435 return 0;
436}
437
438static int pcifront_scan_root(struct pcifront_device *pdev,
439 unsigned int domain, unsigned int bus)
440{
441 struct pci_bus *b;
442 LIST_HEAD(resources);
443 struct pcifront_sd *sd = NULL;
444 struct pci_bus_entry *bus_entry = NULL;
445 int err = 0;
446 static struct resource busn_res = {
447 .start = 0,
448 .end = 255,
449 .flags = IORESOURCE_BUS,
450 };
451
452#ifndef CONFIG_PCI_DOMAINS
453 if (domain != 0) {
454 dev_err(&pdev->xdev->dev,
455 "PCI Root in non-zero PCI Domain! domain=%d\n", domain);
456 dev_err(&pdev->xdev->dev,
457 "Please compile with CONFIG_PCI_DOMAINS\n");
458 err = -EINVAL;
459 goto err_out;
460 }
461#endif
462
463 dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n",
464 domain, bus);
465
466 bus_entry = kzalloc(sizeof(*bus_entry), GFP_KERNEL);
467 sd = kzalloc(sizeof(*sd), GFP_KERNEL);
468 if (!bus_entry || !sd) {
469 err = -ENOMEM;
470 goto err_out;
471 }
472 pci_add_resource(&resources, &ioport_resource);
473 pci_add_resource(&resources, &iomem_resource);
474 pci_add_resource(&resources, &busn_res);
475 pcifront_init_sd(sd, domain, bus, pdev);
476
477 pci_lock_rescan_remove();
478
479 b = pci_scan_root_bus(&pdev->xdev->dev, bus,
480 &pcifront_bus_ops, sd, &resources);
481 if (!b) {
482 dev_err(&pdev->xdev->dev,
483 "Error creating PCI Frontend Bus!\n");
484 err = -ENOMEM;
485 pci_unlock_rescan_remove();
486 pci_free_resource_list(&resources);
487 goto err_out;
488 }
489
490 bus_entry->bus = b;
491
492 list_add(&bus_entry->list, &pdev->root_buses);
493
494 /* pci_scan_root_bus skips devices which do not have a
495 * devfn==0. The pcifront_scan_bus enumerates all devfn. */
496 err = pcifront_scan_bus(pdev, domain, bus, b);
497
498 /* Claim resources before going "live" with our devices */
499 pci_walk_bus(b, pcifront_claim_resource, pdev);
500
501 /* Create SysFS and notify udev of the devices. Aka: "going live" */
502 pci_bus_add_devices(b);
503
504 pci_unlock_rescan_remove();
505 return err;
506
507err_out:
508 kfree(bus_entry);
509 kfree(sd);
510
511 return err;
512}
513
514static int pcifront_rescan_root(struct pcifront_device *pdev,
515 unsigned int domain, unsigned int bus)
516{
517 int err;
518 struct pci_bus *b;
519
520#ifndef CONFIG_PCI_DOMAINS
521 if (domain != 0) {
522 dev_err(&pdev->xdev->dev,
523 "PCI Root in non-zero PCI Domain! domain=%d\n", domain);
524 dev_err(&pdev->xdev->dev,
525 "Please compile with CONFIG_PCI_DOMAINS\n");
526 return -EINVAL;
527 }
528#endif
529
530 dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n",
531 domain, bus);
532
533 b = pci_find_bus(domain, bus);
534 if (!b)
535 /* If the bus is unknown, create it. */
536 return pcifront_scan_root(pdev, domain, bus);
537
538 err = pcifront_scan_bus(pdev, domain, bus, b);
539
540 /* Claim resources before going "live" with our devices */
541 pci_walk_bus(b, pcifront_claim_resource, pdev);
542
543 /* Create SysFS and notify udev of the devices. Aka: "going live" */
544 pci_bus_add_devices(b);
545
546 return err;
547}
548
549static void free_root_bus_devs(struct pci_bus *bus)
550{
551 struct pci_dev *dev;
552
553 while (!list_empty(&bus->devices)) {
554 dev = container_of(bus->devices.next, struct pci_dev,
555 bus_list);
556 pci_dbg(dev, "removing device\n");
557 pci_stop_and_remove_bus_device(dev);
558 }
559}
560
561static void pcifront_free_roots(struct pcifront_device *pdev)
562{
563 struct pci_bus_entry *bus_entry, *t;
564
565 dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n");
566
567 pci_lock_rescan_remove();
568 list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) {
569 list_del(&bus_entry->list);
570
571 free_root_bus_devs(bus_entry->bus);
572
573 kfree(bus_entry->bus->sysdata);
574
575 device_unregister(bus_entry->bus->bridge);
576 pci_remove_bus(bus_entry->bus);
577
578 kfree(bus_entry);
579 }
580 pci_unlock_rescan_remove();
581}
582
583static pci_ers_result_t pcifront_common_process(int cmd,
584 struct pcifront_device *pdev,
585 pci_channel_state_t state)
586{
587 pci_ers_result_t result;
588 struct pci_driver *pdrv;
589 int bus = pdev->sh_info->aer_op.bus;
590 int devfn = pdev->sh_info->aer_op.devfn;
591 int domain = pdev->sh_info->aer_op.domain;
592 struct pci_dev *pcidev;
593 int flag = 0;
594
595 dev_dbg(&pdev->xdev->dev,
596 "pcifront AER process: cmd %x (bus:%x, devfn%x)",
597 cmd, bus, devfn);
598 result = PCI_ERS_RESULT_NONE;
599
600 pcidev = pci_get_domain_bus_and_slot(domain, bus, devfn);
601 if (!pcidev || !pcidev->driver) {
602 dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n");
603 pci_dev_put(pcidev);
604 return result;
605 }
606 pdrv = pcidev->driver;
607
608 if (pdrv) {
609 if (pdrv->err_handler && pdrv->err_handler->error_detected) {
610 pci_dbg(pcidev, "trying to call AER service\n");
611 if (pcidev) {
612 flag = 1;
613 switch (cmd) {
614 case XEN_PCI_OP_aer_detected:
615 result = pdrv->err_handler->
616 error_detected(pcidev, state);
617 break;
618 case XEN_PCI_OP_aer_mmio:
619 result = pdrv->err_handler->
620 mmio_enabled(pcidev);
621 break;
622 case XEN_PCI_OP_aer_slotreset:
623 result = pdrv->err_handler->
624 slot_reset(pcidev);
625 break;
626 case XEN_PCI_OP_aer_resume:
627 pdrv->err_handler->resume(pcidev);
628 break;
629 default:
630 dev_err(&pdev->xdev->dev,
631 "bad request in aer recovery "
632 "operation!\n");
633
634 }
635 }
636 }
637 }
638 if (!flag)
639 result = PCI_ERS_RESULT_NONE;
640
641 return result;
642}
643
644
645static void pcifront_do_aer(struct work_struct *data)
646{
647 struct pcifront_device *pdev =
648 container_of(data, struct pcifront_device, op_work);
649 int cmd = pdev->sh_info->aer_op.cmd;
650 pci_channel_state_t state =
651 (pci_channel_state_t)pdev->sh_info->aer_op.err;
652
653 /*If a pci_conf op is in progress,
654 we have to wait until it is done before service aer op*/
655 dev_dbg(&pdev->xdev->dev,
656 "pcifront service aer bus %x devfn %x\n",
657 pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn);
658
659 pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state);
660
661 /* Post the operation to the guest. */
662 wmb();
663 clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags);
664 notify_remote_via_evtchn(pdev->evtchn);
665
666 /*in case of we lost an aer request in four lines time_window*/
667 smp_mb__before_atomic();
668 clear_bit(_PDEVB_op_active, &pdev->flags);
669 smp_mb__after_atomic();
670
671 schedule_pcifront_aer_op(pdev);
672
673}
674
675static irqreturn_t pcifront_handler_aer(int irq, void *dev)
676{
677 struct pcifront_device *pdev = dev;
678 schedule_pcifront_aer_op(pdev);
679 return IRQ_HANDLED;
680}
681static int pcifront_connect_and_init_dma(struct pcifront_device *pdev)
682{
683 int err = 0;
684
685 spin_lock(&pcifront_dev_lock);
686
687 if (!pcifront_dev) {
688 dev_info(&pdev->xdev->dev, "Installing PCI frontend\n");
689 pcifront_dev = pdev;
690 } else
691 err = -EEXIST;
692
693 spin_unlock(&pcifront_dev_lock);
694
695 if (!err && !swiotlb_nr_tbl()) {
696 err = pci_xen_swiotlb_init_late();
697 if (err)
698 dev_err(&pdev->xdev->dev, "Could not setup SWIOTLB!\n");
699 }
700 return err;
701}
702
703static void pcifront_disconnect(struct pcifront_device *pdev)
704{
705 spin_lock(&pcifront_dev_lock);
706
707 if (pdev == pcifront_dev) {
708 dev_info(&pdev->xdev->dev,
709 "Disconnecting PCI Frontend Buses\n");
710 pcifront_dev = NULL;
711 }
712
713 spin_unlock(&pcifront_dev_lock);
714}
715static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev)
716{
717 struct pcifront_device *pdev;
718
719 pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL);
720 if (pdev == NULL)
721 goto out;
722
723 pdev->sh_info =
724 (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL);
725 if (pdev->sh_info == NULL) {
726 kfree(pdev);
727 pdev = NULL;
728 goto out;
729 }
730 pdev->sh_info->flags = 0;
731
732 /*Flag for registering PV AER handler*/
733 set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags);
734
735 dev_set_drvdata(&xdev->dev, pdev);
736 pdev->xdev = xdev;
737
738 INIT_LIST_HEAD(&pdev->root_buses);
739
740 spin_lock_init(&pdev->sh_info_lock);
741
742 pdev->evtchn = INVALID_EVTCHN;
743 pdev->gnt_ref = INVALID_GRANT_REF;
744 pdev->irq = -1;
745
746 INIT_WORK(&pdev->op_work, pcifront_do_aer);
747
748 dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n",
749 pdev, pdev->sh_info);
750out:
751 return pdev;
752}
753
754static void free_pdev(struct pcifront_device *pdev)
755{
756 dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev);
757
758 pcifront_free_roots(pdev);
759
760 cancel_work_sync(&pdev->op_work);
761
762 if (pdev->irq >= 0)
763 unbind_from_irqhandler(pdev->irq, pdev);
764
765 if (pdev->evtchn != INVALID_EVTCHN)
766 xenbus_free_evtchn(pdev->xdev, pdev->evtchn);
767
768 if (pdev->gnt_ref != INVALID_GRANT_REF)
769 gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */,
770 (unsigned long)pdev->sh_info);
771 else
772 free_page((unsigned long)pdev->sh_info);
773
774 dev_set_drvdata(&pdev->xdev->dev, NULL);
775
776 kfree(pdev);
777}
778
779static int pcifront_publish_info(struct pcifront_device *pdev)
780{
781 int err = 0;
782 struct xenbus_transaction trans;
783 grant_ref_t gref;
784
785 err = xenbus_grant_ring(pdev->xdev, pdev->sh_info, 1, &gref);
786 if (err < 0)
787 goto out;
788
789 pdev->gnt_ref = gref;
790
791 err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn);
792 if (err)
793 goto out;
794
795 err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer,
796 0, "pcifront", pdev);
797
798 if (err < 0)
799 return err;
800
801 pdev->irq = err;
802
803do_publish:
804 err = xenbus_transaction_start(&trans);
805 if (err) {
806 xenbus_dev_fatal(pdev->xdev, err,
807 "Error writing configuration for backend "
808 "(start transaction)");
809 goto out;
810 }
811
812 err = xenbus_printf(trans, pdev->xdev->nodename,
813 "pci-op-ref", "%u", pdev->gnt_ref);
814 if (!err)
815 err = xenbus_printf(trans, pdev->xdev->nodename,
816 "event-channel", "%u", pdev->evtchn);
817 if (!err)
818 err = xenbus_printf(trans, pdev->xdev->nodename,
819 "magic", XEN_PCI_MAGIC);
820
821 if (err) {
822 xenbus_transaction_end(trans, 1);
823 xenbus_dev_fatal(pdev->xdev, err,
824 "Error writing configuration for backend");
825 goto out;
826 } else {
827 err = xenbus_transaction_end(trans, 0);
828 if (err == -EAGAIN)
829 goto do_publish;
830 else if (err) {
831 xenbus_dev_fatal(pdev->xdev, err,
832 "Error completing transaction "
833 "for backend");
834 goto out;
835 }
836 }
837
838 xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
839
840 dev_dbg(&pdev->xdev->dev, "publishing successful!\n");
841
842out:
843 return err;
844}
845
846static int pcifront_try_connect(struct pcifront_device *pdev)
847{
848 int err = -EFAULT;
849 int i, num_roots, len;
850 char str[64];
851 unsigned int domain, bus;
852
853
854 /* Only connect once */
855 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
856 XenbusStateInitialised)
857 goto out;
858
859 err = pcifront_connect_and_init_dma(pdev);
860 if (err && err != -EEXIST) {
861 xenbus_dev_fatal(pdev->xdev, err,
862 "Error setting up PCI Frontend");
863 goto out;
864 }
865
866 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
867 "root_num", "%d", &num_roots);
868 if (err == -ENOENT) {
869 xenbus_dev_error(pdev->xdev, err,
870 "No PCI Roots found, trying 0000:00");
871 err = pcifront_scan_root(pdev, 0, 0);
872 if (err) {
873 xenbus_dev_fatal(pdev->xdev, err,
874 "Error scanning PCI root 0000:00");
875 goto out;
876 }
877 num_roots = 0;
878 } else if (err != 1) {
879 if (err == 0)
880 err = -EINVAL;
881 xenbus_dev_fatal(pdev->xdev, err,
882 "Error reading number of PCI roots");
883 goto out;
884 }
885
886 for (i = 0; i < num_roots; i++) {
887 len = snprintf(str, sizeof(str), "root-%d", i);
888 if (unlikely(len >= (sizeof(str) - 1))) {
889 err = -ENOMEM;
890 goto out;
891 }
892
893 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
894 "%x:%x", &domain, &bus);
895 if (err != 2) {
896 if (err >= 0)
897 err = -EINVAL;
898 xenbus_dev_fatal(pdev->xdev, err,
899 "Error reading PCI root %d", i);
900 goto out;
901 }
902
903 err = pcifront_scan_root(pdev, domain, bus);
904 if (err) {
905 xenbus_dev_fatal(pdev->xdev, err,
906 "Error scanning PCI root %04x:%02x",
907 domain, bus);
908 goto out;
909 }
910 }
911
912 err = xenbus_switch_state(pdev->xdev, XenbusStateConnected);
913
914out:
915 return err;
916}
917
918static int pcifront_try_disconnect(struct pcifront_device *pdev)
919{
920 int err = 0;
921 enum xenbus_state prev_state;
922
923
924 prev_state = xenbus_read_driver_state(pdev->xdev->nodename);
925
926 if (prev_state >= XenbusStateClosing)
927 goto out;
928
929 if (prev_state == XenbusStateConnected) {
930 pcifront_free_roots(pdev);
931 pcifront_disconnect(pdev);
932 }
933
934 err = xenbus_switch_state(pdev->xdev, XenbusStateClosed);
935
936out:
937
938 return err;
939}
940
941static int pcifront_attach_devices(struct pcifront_device *pdev)
942{
943 int err = -EFAULT;
944 int i, num_roots, len;
945 unsigned int domain, bus;
946 char str[64];
947
948 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
949 XenbusStateReconfiguring)
950 goto out;
951
952 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
953 "root_num", "%d", &num_roots);
954 if (err == -ENOENT) {
955 xenbus_dev_error(pdev->xdev, err,
956 "No PCI Roots found, trying 0000:00");
957 err = pcifront_rescan_root(pdev, 0, 0);
958 if (err) {
959 xenbus_dev_fatal(pdev->xdev, err,
960 "Error scanning PCI root 0000:00");
961 goto out;
962 }
963 num_roots = 0;
964 } else if (err != 1) {
965 if (err == 0)
966 err = -EINVAL;
967 xenbus_dev_fatal(pdev->xdev, err,
968 "Error reading number of PCI roots");
969 goto out;
970 }
971
972 for (i = 0; i < num_roots; i++) {
973 len = snprintf(str, sizeof(str), "root-%d", i);
974 if (unlikely(len >= (sizeof(str) - 1))) {
975 err = -ENOMEM;
976 goto out;
977 }
978
979 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
980 "%x:%x", &domain, &bus);
981 if (err != 2) {
982 if (err >= 0)
983 err = -EINVAL;
984 xenbus_dev_fatal(pdev->xdev, err,
985 "Error reading PCI root %d", i);
986 goto out;
987 }
988
989 err = pcifront_rescan_root(pdev, domain, bus);
990 if (err) {
991 xenbus_dev_fatal(pdev->xdev, err,
992 "Error scanning PCI root %04x:%02x",
993 domain, bus);
994 goto out;
995 }
996 }
997
998 xenbus_switch_state(pdev->xdev, XenbusStateConnected);
999
1000out:
1001 return err;
1002}
1003
1004static int pcifront_detach_devices(struct pcifront_device *pdev)
1005{
1006 int err = 0;
1007 int i, num_devs;
1008 unsigned int domain, bus, slot, func;
1009 struct pci_dev *pci_dev;
1010 char str[64];
1011
1012 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
1013 XenbusStateConnected)
1014 goto out;
1015
1016 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d",
1017 &num_devs);
1018 if (err != 1) {
1019 if (err >= 0)
1020 err = -EINVAL;
1021 xenbus_dev_fatal(pdev->xdev, err,
1022 "Error reading number of PCI devices");
1023 goto out;
1024 }
1025
1026 /* Find devices being detached and remove them. */
1027 for (i = 0; i < num_devs; i++) {
1028 int l, state;
1029 l = snprintf(str, sizeof(str), "state-%d", i);
1030 if (unlikely(l >= (sizeof(str) - 1))) {
1031 err = -ENOMEM;
1032 goto out;
1033 }
1034 state = xenbus_read_unsigned(pdev->xdev->otherend, str,
1035 XenbusStateUnknown);
1036
1037 if (state != XenbusStateClosing)
1038 continue;
1039
1040 /* Remove device. */
1041 l = snprintf(str, sizeof(str), "vdev-%d", i);
1042 if (unlikely(l >= (sizeof(str) - 1))) {
1043 err = -ENOMEM;
1044 goto out;
1045 }
1046 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
1047 "%x:%x:%x.%x", &domain, &bus, &slot, &func);
1048 if (err != 4) {
1049 if (err >= 0)
1050 err = -EINVAL;
1051 xenbus_dev_fatal(pdev->xdev, err,
1052 "Error reading PCI device %d", i);
1053 goto out;
1054 }
1055
1056 pci_dev = pci_get_domain_bus_and_slot(domain, bus,
1057 PCI_DEVFN(slot, func));
1058 if (!pci_dev) {
1059 dev_dbg(&pdev->xdev->dev,
1060 "Cannot get PCI device %04x:%02x:%02x.%d\n",
1061 domain, bus, slot, func);
1062 continue;
1063 }
1064 pci_lock_rescan_remove();
1065 pci_stop_and_remove_bus_device(pci_dev);
1066 pci_dev_put(pci_dev);
1067 pci_unlock_rescan_remove();
1068
1069 dev_dbg(&pdev->xdev->dev,
1070 "PCI device %04x:%02x:%02x.%d removed.\n",
1071 domain, bus, slot, func);
1072 }
1073
1074 err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring);
1075
1076out:
1077 return err;
1078}
1079
1080static void __ref pcifront_backend_changed(struct xenbus_device *xdev,
1081 enum xenbus_state be_state)
1082{
1083 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1084
1085 switch (be_state) {
1086 case XenbusStateUnknown:
1087 case XenbusStateInitialising:
1088 case XenbusStateInitWait:
1089 case XenbusStateInitialised:
1090 break;
1091
1092 case XenbusStateConnected:
1093 pcifront_try_connect(pdev);
1094 break;
1095
1096 case XenbusStateClosed:
1097 if (xdev->state == XenbusStateClosed)
1098 break;
1099 fallthrough; /* Missed the backend's CLOSING state */
1100 case XenbusStateClosing:
1101 dev_warn(&xdev->dev, "backend going away!\n");
1102 pcifront_try_disconnect(pdev);
1103 break;
1104
1105 case XenbusStateReconfiguring:
1106 pcifront_detach_devices(pdev);
1107 break;
1108
1109 case XenbusStateReconfigured:
1110 pcifront_attach_devices(pdev);
1111 break;
1112 }
1113}
1114
1115static int pcifront_xenbus_probe(struct xenbus_device *xdev,
1116 const struct xenbus_device_id *id)
1117{
1118 int err = 0;
1119 struct pcifront_device *pdev = alloc_pdev(xdev);
1120
1121 if (pdev == NULL) {
1122 err = -ENOMEM;
1123 xenbus_dev_fatal(xdev, err,
1124 "Error allocating pcifront_device struct");
1125 goto out;
1126 }
1127
1128 err = pcifront_publish_info(pdev);
1129 if (err)
1130 free_pdev(pdev);
1131
1132out:
1133 return err;
1134}
1135
1136static int pcifront_xenbus_remove(struct xenbus_device *xdev)
1137{
1138 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1139 if (pdev)
1140 free_pdev(pdev);
1141
1142 return 0;
1143}
1144
1145static const struct xenbus_device_id xenpci_ids[] = {
1146 {"pci"},
1147 {""},
1148};
1149
1150static struct xenbus_driver xenpci_driver = {
1151 .name = "pcifront",
1152 .ids = xenpci_ids,
1153 .probe = pcifront_xenbus_probe,
1154 .remove = pcifront_xenbus_remove,
1155 .otherend_changed = pcifront_backend_changed,
1156};
1157
1158static int __init pcifront_init(void)
1159{
1160 if (!xen_pv_domain() || xen_initial_domain())
1161 return -ENODEV;
1162
1163 if (!xen_has_pv_devices())
1164 return -ENODEV;
1165
1166 pci_frontend_registrar(1 /* enable */);
1167
1168 return xenbus_register_frontend(&xenpci_driver);
1169}
1170
1171static void __exit pcifront_cleanup(void)
1172{
1173 xenbus_unregister_driver(&xenpci_driver);
1174 pci_frontend_registrar(0 /* disable */);
1175}
1176module_init(pcifront_init);
1177module_exit(pcifront_cleanup);
1178
1179MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
1180MODULE_LICENSE("GPL");
1181MODULE_ALIAS("xen:pci");