projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
xen-pciback: also support disabling of bus-mastering and memory-write-invalidate
[linux-block.git] / drivers / pci / xen-pcifront.c
CommitLineData
956a9202
RW		 1/*
2 * Xen PCI Frontend.
3 *
4 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
5 */
6#include <linux/module.h>
7#include <linux/init.h>
8#include <linux/mm.h>
9#include <xen/xenbus.h>
10#include <xen/events.h>
11#include <xen/grant_table.h>
12#include <xen/page.h>
13#include <linux/spinlock.h>
14#include <linux/pci.h>
15#include <linux/msi.h>
956a9202
RW		 16#include <xen/interface/io/pciif.h>
17#include <asm/xen/pci.h>
18#include <linux/interrupt.h>
60063497 19#include <linux/atomic.h>
956a9202
RW		 20#include <linux/workqueue.h>
21#include <linux/bitops.h>
22#include <linux/time.h>
51c71a3b 23#include <xen/platform_pci.h>
956a9202 24
3d925320 25#include <asm/xen/swiotlb-xen.h>
956a9202
RW		 26#define INVALID_GRANT_REF (0)
27#define INVALID_EVTCHN (-1)
28
29struct pci_bus_entry {
30 struct list_head list;
31 struct pci_bus *bus;
32};
33
34#define _PDEVB_op_active (0)
35#define PDEVB_op_active (1 << (_PDEVB_op_active))
36
37struct pcifront_device {
38 struct xenbus_device *xdev;
39 struct list_head root_buses;
40
41 int evtchn;
42 int gnt_ref;
43
44 int irq;
45
46 /* Lock this when doing any operations in sh_info */
47 spinlock_t sh_info_lock;
48 struct xen_pci_sharedinfo *sh_info;
49 struct work_struct op_work;
50 unsigned long flags;
51
52};
53
54struct pcifront_sd {
55 int domain;
56 struct pcifront_device *pdev;
57};
58
59static inline struct pcifront_device *
60pcifront_get_pdev(struct pcifront_sd *sd)
61{
62 return sd->pdev;
63}
64
65static inline void pcifront_init_sd(struct pcifront_sd *sd,
66 unsigned int domain, unsigned int bus,
67 struct pcifront_device *pdev)
68{
69 sd->domain = domain;
70 sd->pdev = pdev;
71}
72
73static DEFINE_SPINLOCK(pcifront_dev_lock);
74static struct pcifront_device *pcifront_dev;
75
76static int verbose_request;
77module_param(verbose_request, int, 0644);
78
79static int errno_to_pcibios_err(int errno)
80{
81 switch (errno) {
82 case XEN_PCI_ERR_success:
83 return PCIBIOS_SUCCESSFUL;
84
85 case XEN_PCI_ERR_dev_not_found:
86 return PCIBIOS_DEVICE_NOT_FOUND;
87
88 case XEN_PCI_ERR_invalid_offset:
89 case XEN_PCI_ERR_op_failed:
90 return PCIBIOS_BAD_REGISTER_NUMBER;
91
92 case XEN_PCI_ERR_not_implemented:
93 return PCIBIOS_FUNC_NOT_SUPPORTED;
94
95 case XEN_PCI_ERR_access_denied:
96 return PCIBIOS_SET_FAILED;
97 }
98 return errno;
99}
100
101static inline void schedule_pcifront_aer_op(struct pcifront_device *pdev)
102{
103 if (test_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags)
104 && !test_and_set_bit(_PDEVB_op_active, &pdev->flags)) {
105 dev_dbg(&pdev->xdev->dev, "schedule aer frontend job\n");
106 schedule_work(&pdev->op_work);
107 }
108}
109
110static int do_pci_op(struct pcifront_device *pdev, struct xen_pci_op *op)
111{
112 int err = 0;
113 struct xen_pci_op *active_op = &pdev->sh_info->op;
114 unsigned long irq_flags;
115 evtchn_port_t port = pdev->evtchn;
116 unsigned irq = pdev->irq;
117 s64 ns, ns_timeout;
118 struct timeval tv;
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 do_gettimeofday(&tv);
136 ns_timeout = timeval_to_ns(&tv) + 2 * (s64)NSEC_PER_SEC;
137
138 xen_clear_irq_pending(irq);
139
140 while (test_bit(_XEN_PCIF_active,
141 (unsigned long *)&pdev->sh_info->flags)) {
142 xen_poll_irq_timeout(irq, jiffies + 3*HZ);
143 xen_clear_irq_pending(irq);
144 do_gettimeofday(&tv);
145 ns = timeval_to_ns(&tv);
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,
e4de866a 194 "read dev=%04x:%02x:%02x.%d - offset %x size %d\n",
956a9202
RW		 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,
e4de866a 233 "write dev=%04x:%02x:%02x.%d - "
956a9202
RW		 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
b8b0f559 241static struct pci_ops pcifront_bus_ops = {
956a9202
RW		 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,
cc0f89c4 248 int vector[], int nvec)
956a9202
RW		 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 list_for_each_entry(entry, &dev->msi_list, list) {
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 */
1d461052
KRW		 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;
cc0f89c4 287 vector[i] = -1;
1d461052
KRW		 288 continue;
289 }
cc0f89c4 290 vector[i] = op.msix_entries[i].vector;
1d461052 291 }
956a9202
RW		 292 } else {
293 printk(KERN_DEBUG "enable msix get value %x\n",
294 op.value);
f09d8432 295 err = op.value;
956a9202
RW		 296 }
297 } else {
298 dev_err(&dev->dev, "enable msix get err %x\n", err);
956a9202 299 }
1d461052 300 return err;
956a9202
RW		 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
cc0f89c4 322static int pci_frontend_enable_msi(struct pci_dev *dev, int vector[])
956a9202
RW		 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)) {
cc0f89c4 336 vector[0] = op.value;
1d461052
KRW		 337 if (op.value <= 0) {
338 dev_warn(&dev->dev, "MSI entry is invalid: %d!\n",
339 op.value);
340 err = -EINVAL;
cc0f89c4 341 vector[0] = -1;
1d461052 342 }
956a9202
RW		 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)) {
917e3e65
KRW		 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",
956a9202
RW		 408 pci_name(dev), i);
409 }
410 }
411 }
412
413 return 0;
414}
415
15856ad5 416static int pcifront_scan_bus(struct pcifront_device *pdev,
956a9202
RW		 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 "
e4de866a 438 "%04x:%02x:%02x.%d found.\n", domain, bus,
956a9202
RW		 439 PCI_SLOT(devfn), PCI_FUNC(devfn));
440 }
441
442 return 0;
443}
444
15856ad5 445static int pcifront_scan_root(struct pcifront_device *pdev,
956a9202
RW		 446 unsigned int domain, unsigned int bus)
447{
448 struct pci_bus *b;
449 struct pcifront_sd *sd = NULL;
450 struct pci_bus_entry *bus_entry = NULL;
451 int err = 0;
452
453#ifndef CONFIG_PCI_DOMAINS
454 if (domain != 0) {
455 dev_err(&pdev->xdev->dev,
456 "PCI Root in non-zero PCI Domain! domain=%d\n", domain);
457 dev_err(&pdev->xdev->dev,
458 "Please compile with CONFIG_PCI_DOMAINS\n");
459 err = -EINVAL;
460 goto err_out;
461 }
462#endif
463
464 dev_info(&pdev->xdev->dev, "Creating PCI Frontend Bus %04x:%02x\n",
465 domain, bus);
466
467 bus_entry = kmalloc(sizeof(*bus_entry), GFP_KERNEL);
468 sd = kmalloc(sizeof(*sd), GFP_KERNEL);
469 if (!bus_entry || !sd) {
470 err = -ENOMEM;
471 goto err_out;
472 }
473 pcifront_init_sd(sd, domain, bus, pdev);
474
a83919e0
RW		 475 pci_lock_rescan_remove();
476
956a9202
RW		 477 b = pci_scan_bus_parented(&pdev->xdev->dev, bus,
478 &pcifront_bus_ops, sd);
479 if (!b) {
480 dev_err(&pdev->xdev->dev,
481 "Error creating PCI Frontend Bus!\n");
482 err = -ENOMEM;
a83919e0 483 pci_unlock_rescan_remove();
956a9202
RW		 484 goto err_out;
485 }
486
487 bus_entry->bus = b;
488
489 list_add(&bus_entry->list, &pdev->root_buses);
490
491 /* pci_scan_bus_parented skips devices which do not have a have
492 * devfn==0. The pcifront_scan_bus enumerates all devfn. */
493 err = pcifront_scan_bus(pdev, domain, bus, b);
494
495 /* Claim resources before going "live" with our devices */
496 pci_walk_bus(b, pcifront_claim_resource, pdev);
497
498 /* Create SysFS and notify udev of the devices. Aka: "going live" */
499 pci_bus_add_devices(b);
500
a83919e0 501 pci_unlock_rescan_remove();
956a9202
RW		 502 return err;
503
504err_out:
505 kfree(bus_entry);
506 kfree(sd);
507
508 return err;
509}
510
15856ad5 511static int pcifront_rescan_root(struct pcifront_device *pdev,
956a9202
RW		 512 unsigned int domain, unsigned int bus)
513{
514 int err;
515 struct pci_bus *b;
516
517#ifndef CONFIG_PCI_DOMAINS
518 if (domain != 0) {
519 dev_err(&pdev->xdev->dev,
520 "PCI Root in non-zero PCI Domain! domain=%d\n", domain);
521 dev_err(&pdev->xdev->dev,
522 "Please compile with CONFIG_PCI_DOMAINS\n");
523 return -EINVAL;
524 }
525#endif
526
527 dev_info(&pdev->xdev->dev, "Rescanning PCI Frontend Bus %04x:%02x\n",
528 domain, bus);
529
530 b = pci_find_bus(domain, bus);
531 if (!b)
532 /* If the bus is unknown, create it. */
533 return pcifront_scan_root(pdev, domain, bus);
534
535 err = pcifront_scan_bus(pdev, domain, bus, b);
536
537 /* Claim resources before going "live" with our devices */
538 pci_walk_bus(b, pcifront_claim_resource, pdev);
539
540 /* Create SysFS and notify udev of the devices. Aka: "going live" */
541 pci_bus_add_devices(b);
542
543 return err;
544}
545
546static void free_root_bus_devs(struct pci_bus *bus)
547{
548 struct pci_dev *dev;
549
550 while (!list_empty(&bus->devices)) {
551 dev = container_of(bus->devices.next, struct pci_dev,
552 bus_list);
553 dev_dbg(&dev->dev, "removing device\n");
210647af 554 pci_stop_and_remove_bus_device(dev);
956a9202
RW		 555 }
556}
557
558static void pcifront_free_roots(struct pcifront_device *pdev)
559{
560 struct pci_bus_entry *bus_entry, *t;
561
562 dev_dbg(&pdev->xdev->dev, "cleaning up root buses\n");
563
a83919e0 564 pci_lock_rescan_remove();
956a9202
RW		 565 list_for_each_entry_safe(bus_entry, t, &pdev->root_buses, list) {
566 list_del(&bus_entry->list);
567
568 free_root_bus_devs(bus_entry->bus);
569
570 kfree(bus_entry->bus->sysdata);
571
572 device_unregister(bus_entry->bus->bridge);
573 pci_remove_bus(bus_entry->bus);
574
575 kfree(bus_entry);
576 }
a83919e0 577 pci_unlock_rescan_remove();
956a9202
RW		 578}
579
580static pci_ers_result_t pcifront_common_process(int cmd,
581 struct pcifront_device *pdev,
582 pci_channel_state_t state)
583{
584 pci_ers_result_t result;
585 struct pci_driver *pdrv;
586 int bus = pdev->sh_info->aer_op.bus;
587 int devfn = pdev->sh_info->aer_op.devfn;
588 struct pci_dev *pcidev;
589 int flag = 0;
590
591 dev_dbg(&pdev->xdev->dev,
592 "pcifront AER process: cmd %x (bus:%x, devfn%x)",
593 cmd, bus, devfn);
594 result = PCI_ERS_RESULT_NONE;
595
596 pcidev = pci_get_bus_and_slot(bus, devfn);
597 if (!pcidev || !pcidev->driver) {
2a63dd72 598 dev_err(&pdev->xdev->dev, "device or AER driver is NULL\n");
ff0387c3 599 pci_dev_put(pcidev);
956a9202
RW		 600 return result;
601 }
602 pdrv = pcidev->driver;
603
07d25146 604 if (pdrv) {
956a9202
RW		 605 if (pdrv->err_handler && pdrv->err_handler->error_detected) {
606 dev_dbg(&pcidev->dev,
607 "trying to call AER service\n");
608 if (pcidev) {
609 flag = 1;
610 switch (cmd) {
611 case XEN_PCI_OP_aer_detected:
612 result = pdrv->err_handler->
613 error_detected(pcidev, state);
614 break;
615 case XEN_PCI_OP_aer_mmio:
616 result = pdrv->err_handler->
617 mmio_enabled(pcidev);
618 break;
619 case XEN_PCI_OP_aer_slotreset:
620 result = pdrv->err_handler->
621 slot_reset(pcidev);
622 break;
623 case XEN_PCI_OP_aer_resume:
624 pdrv->err_handler->resume(pcidev);
625 break;
626 default:
627 dev_err(&pdev->xdev->dev,
628 "bad request in aer recovery "
629 "operation!\n");
630
631 }
632 }
633 }
956a9202
RW		 634 }
635 if (!flag)
636 result = PCI_ERS_RESULT_NONE;
637
638 return result;
639}
640
641
642static void pcifront_do_aer(struct work_struct *data)
643{
644 struct pcifront_device *pdev =
645 container_of(data, struct pcifront_device, op_work);
646 int cmd = pdev->sh_info->aer_op.cmd;
647 pci_channel_state_t state =
648 (pci_channel_state_t)pdev->sh_info->aer_op.err;
649
650 /*If a pci_conf op is in progress,
651 we have to wait until it is done before service aer op*/
652 dev_dbg(&pdev->xdev->dev,
653 "pcifront service aer bus %x devfn %x\n",
654 pdev->sh_info->aer_op.bus, pdev->sh_info->aer_op.devfn);
655
656 pdev->sh_info->aer_op.err = pcifront_common_process(cmd, pdev, state);
657
658 /* Post the operation to the guest. */
659 wmb();
660 clear_bit(_XEN_PCIB_active, (unsigned long *)&pdev->sh_info->flags);
661 notify_remote_via_evtchn(pdev->evtchn);
662
663 /*in case of we lost an aer request in four lines time_window*/
4e857c58 664 smp_mb__before_atomic();
956a9202 665 clear_bit(_PDEVB_op_active, &pdev->flags);
4e857c58 666 smp_mb__after_atomic();
956a9202
RW		 667
668 schedule_pcifront_aer_op(pdev);
669
670}
671
672static irqreturn_t pcifront_handler_aer(int irq, void *dev)
673{
674 struct pcifront_device *pdev = dev;
675 schedule_pcifront_aer_op(pdev);
676 return IRQ_HANDLED;
677}
3d925320 678static int pcifront_connect_and_init_dma(struct pcifront_device *pdev)
956a9202
RW		 679{
680 int err = 0;
681
682 spin_lock(&pcifront_dev_lock);
683
684 if (!pcifront_dev) {
685 dev_info(&pdev->xdev->dev, "Installing PCI frontend\n");
686 pcifront_dev = pdev;
098b1aea 687 } else
956a9202 688 err = -EEXIST;
098b1aea 689
956a9202
RW		 690 spin_unlock(&pcifront_dev_lock);
691
3d925320
KRW		 692 if (!err && !swiotlb_nr_tbl()) {
693 err = pci_xen_swiotlb_init_late();
694 if (err)
695 dev_err(&pdev->xdev->dev, "Could not setup SWIOTLB!\n");
696 }
956a9202
RW		 697 return err;
698}
699
700static void pcifront_disconnect(struct pcifront_device *pdev)
701{
702 spin_lock(&pcifront_dev_lock);
703
704 if (pdev == pcifront_dev) {
705 dev_info(&pdev->xdev->dev,
706 "Disconnecting PCI Frontend Buses\n");
707 pcifront_dev = NULL;
708 }
709
710 spin_unlock(&pcifront_dev_lock);
711}
712static struct pcifront_device *alloc_pdev(struct xenbus_device *xdev)
713{
714 struct pcifront_device *pdev;
715
716 pdev = kzalloc(sizeof(struct pcifront_device), GFP_KERNEL);
717 if (pdev == NULL)
718 goto out;
719
720 pdev->sh_info =
721 (struct xen_pci_sharedinfo *)__get_free_page(GFP_KERNEL);
722 if (pdev->sh_info == NULL) {
723 kfree(pdev);
724 pdev = NULL;
725 goto out;
726 }
727 pdev->sh_info->flags = 0;
728
729 /*Flag for registering PV AER handler*/
730 set_bit(_XEN_PCIB_AERHANDLER, (void *)&pdev->sh_info->flags);
731
732 dev_set_drvdata(&xdev->dev, pdev);
733 pdev->xdev = xdev;
734
735 INIT_LIST_HEAD(&pdev->root_buses);
736
737 spin_lock_init(&pdev->sh_info_lock);
738
739 pdev->evtchn = INVALID_EVTCHN;
740 pdev->gnt_ref = INVALID_GRANT_REF;
741 pdev->irq = -1;
742
743 INIT_WORK(&pdev->op_work, pcifront_do_aer);
744
745 dev_dbg(&xdev->dev, "Allocated pdev @ 0x%p pdev->sh_info @ 0x%p\n",
746 pdev, pdev->sh_info);
747out:
748 return pdev;
749}
750
751static void free_pdev(struct pcifront_device *pdev)
752{
753 dev_dbg(&pdev->xdev->dev, "freeing pdev @ 0x%p\n", pdev);
754
755 pcifront_free_roots(pdev);
756
db2e2e6e 757 cancel_work_sync(&pdev->op_work);
956a9202
RW		 758
759 if (pdev->irq >= 0)
760 unbind_from_irqhandler(pdev->irq, pdev);
761
762 if (pdev->evtchn != INVALID_EVTCHN)
763 xenbus_free_evtchn(pdev->xdev, pdev->evtchn);
764
765 if (pdev->gnt_ref != INVALID_GRANT_REF)
766 gnttab_end_foreign_access(pdev->gnt_ref, 0 /* r/w page */,
767 (unsigned long)pdev->sh_info);
768 else
769 free_page((unsigned long)pdev->sh_info);
770
771 dev_set_drvdata(&pdev->xdev->dev, NULL);
772
773 kfree(pdev);
774}
775
776static int pcifront_publish_info(struct pcifront_device *pdev)
777{
778 int err = 0;
779 struct xenbus_transaction trans;
780
781 err = xenbus_grant_ring(pdev->xdev, virt_to_mfn(pdev->sh_info));
782 if (err < 0)
783 goto out;
784
785 pdev->gnt_ref = err;
786
787 err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn);
788 if (err)
789 goto out;
790
791 err = bind_evtchn_to_irqhandler(pdev->evtchn, pcifront_handler_aer,
792 0, "pcifront", pdev);
793
794 if (err < 0)
795 return err;
796
797 pdev->irq = err;
798
799do_publish:
800 err = xenbus_transaction_start(&trans);
801 if (err) {
802 xenbus_dev_fatal(pdev->xdev, err,
803 "Error writing configuration for backend "
804 "(start transaction)");
805 goto out;
806 }
807
808 err = xenbus_printf(trans, pdev->xdev->nodename,
809 "pci-op-ref", "%u", pdev->gnt_ref);
810 if (!err)
811 err = xenbus_printf(trans, pdev->xdev->nodename,
812 "event-channel", "%u", pdev->evtchn);
813 if (!err)
814 err = xenbus_printf(trans, pdev->xdev->nodename,
815 "magic", XEN_PCI_MAGIC);
816
817 if (err) {
818 xenbus_transaction_end(trans, 1);
819 xenbus_dev_fatal(pdev->xdev, err,
820 "Error writing configuration for backend");
821 goto out;
822 } else {
823 err = xenbus_transaction_end(trans, 0);
824 if (err == -EAGAIN)
825 goto do_publish;
826 else if (err) {
827 xenbus_dev_fatal(pdev->xdev, err,
828 "Error completing transaction "
829 "for backend");
830 goto out;
831 }
832 }
833
834 xenbus_switch_state(pdev->xdev, XenbusStateInitialised);
835
836 dev_dbg(&pdev->xdev->dev, "publishing successful!\n");
837
838out:
839 return err;
840}
841
15856ad5 842static int pcifront_try_connect(struct pcifront_device *pdev)
956a9202
RW		 843{
844 int err = -EFAULT;
845 int i, num_roots, len;
846 char str[64];
847 unsigned int domain, bus;
848
849
850 /* Only connect once */
851 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
852 XenbusStateInitialised)
853 goto out;
854
3d925320 855 err = pcifront_connect_and_init_dma(pdev);
098b1aea 856 if (err && err != -EEXIST) {
956a9202 857 xenbus_dev_fatal(pdev->xdev, err,
3d925320 858 "Error setting up PCI Frontend");
956a9202
RW		 859 goto out;
860 }
861
862 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
863 "root_num", "%d", &num_roots);
864 if (err == -ENOENT) {
865 xenbus_dev_error(pdev->xdev, err,
866 "No PCI Roots found, trying 0000:00");
867 err = pcifront_scan_root(pdev, 0, 0);
23cf1d00
CG		 868 if (err) {
869 xenbus_dev_fatal(pdev->xdev, err,
870 "Error scanning PCI root 0000:00");
871 goto out;
872 }
956a9202
RW		 873 num_roots = 0;
874 } else if (err != 1) {
875 if (err == 0)
876 err = -EINVAL;
877 xenbus_dev_fatal(pdev->xdev, err,
878 "Error reading number of PCI roots");
879 goto out;
880 }
881
882 for (i = 0; i < num_roots; i++) {
883 len = snprintf(str, sizeof(str), "root-%d", i);
884 if (unlikely(len >= (sizeof(str) - 1))) {
885 err = -ENOMEM;
886 goto out;
887 }
888
889 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
890 "%x:%x", &domain, &bus);
891 if (err != 2) {
892 if (err >= 0)
893 err = -EINVAL;
894 xenbus_dev_fatal(pdev->xdev, err,
895 "Error reading PCI root %d", i);
896 goto out;
897 }
898
899 err = pcifront_scan_root(pdev, domain, bus);
900 if (err) {
901 xenbus_dev_fatal(pdev->xdev, err,
902 "Error scanning PCI root %04x:%02x",
903 domain, bus);
904 goto out;
905 }
906 }
907
908 err = xenbus_switch_state(pdev->xdev, XenbusStateConnected);
909
910out:
911 return err;
912}
913
914static int pcifront_try_disconnect(struct pcifront_device *pdev)
915{
916 int err = 0;
917 enum xenbus_state prev_state;
918
919
920 prev_state = xenbus_read_driver_state(pdev->xdev->nodename);
921
922 if (prev_state >= XenbusStateClosing)
923 goto out;
924
925 if (prev_state == XenbusStateConnected) {
926 pcifront_free_roots(pdev);
927 pcifront_disconnect(pdev);
928 }
929
930 err = xenbus_switch_state(pdev->xdev, XenbusStateClosed);
931
932out:
933
934 return err;
935}
936
15856ad5 937static int pcifront_attach_devices(struct pcifront_device *pdev)
956a9202
RW		 938{
939 int err = -EFAULT;
940 int i, num_roots, len;
941 unsigned int domain, bus;
942 char str[64];
943
944 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
945 XenbusStateReconfiguring)
946 goto out;
947
948 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend,
949 "root_num", "%d", &num_roots);
950 if (err == -ENOENT) {
951 xenbus_dev_error(pdev->xdev, err,
952 "No PCI Roots found, trying 0000:00");
953 err = pcifront_rescan_root(pdev, 0, 0);
23cf1d00
CG		 954 if (err) {
955 xenbus_dev_fatal(pdev->xdev, err,
956 "Error scanning PCI root 0000:00");
957 goto out;
958 }
956a9202
RW		 959 num_roots = 0;
960 } else if (err != 1) {
961 if (err == 0)
962 err = -EINVAL;
963 xenbus_dev_fatal(pdev->xdev, err,
964 "Error reading number of PCI roots");
965 goto out;
966 }
967
968 for (i = 0; i < num_roots; i++) {
969 len = snprintf(str, sizeof(str), "root-%d", i);
970 if (unlikely(len >= (sizeof(str) - 1))) {
971 err = -ENOMEM;
972 goto out;
973 }
974
975 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
976 "%x:%x", &domain, &bus);
977 if (err != 2) {
978 if (err >= 0)
979 err = -EINVAL;
980 xenbus_dev_fatal(pdev->xdev, err,
981 "Error reading PCI root %d", i);
982 goto out;
983 }
984
985 err = pcifront_rescan_root(pdev, domain, bus);
986 if (err) {
987 xenbus_dev_fatal(pdev->xdev, err,
988 "Error scanning PCI root %04x:%02x",
989 domain, bus);
990 goto out;
991 }
992 }
993
994 xenbus_switch_state(pdev->xdev, XenbusStateConnected);
995
996out:
997 return err;
998}
999
1000static int pcifront_detach_devices(struct pcifront_device *pdev)
1001{
1002 int err = 0;
1003 int i, num_devs;
1004 unsigned int domain, bus, slot, func;
956a9202
RW		 1005 struct pci_dev *pci_dev;
1006 char str[64];
1007
1008 if (xenbus_read_driver_state(pdev->xdev->nodename) !=
1009 XenbusStateConnected)
1010 goto out;
1011
1012 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, "num_devs", "%d",
1013 &num_devs);
1014 if (err != 1) {
1015 if (err >= 0)
1016 err = -EINVAL;
1017 xenbus_dev_fatal(pdev->xdev, err,
1018 "Error reading number of PCI devices");
1019 goto out;
1020 }
1021
1022 /* Find devices being detached and remove them. */
1023 for (i = 0; i < num_devs; i++) {
1024 int l, state;
1025 l = snprintf(str, sizeof(str), "state-%d", i);
1026 if (unlikely(l >= (sizeof(str) - 1))) {
1027 err = -ENOMEM;
1028 goto out;
1029 }
1030 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str, "%d",
1031 &state);
1032 if (err != 1)
1033 state = XenbusStateUnknown;
1034
1035 if (state != XenbusStateClosing)
1036 continue;
1037
1038 /* Remove device. */
1039 l = snprintf(str, sizeof(str), "vdev-%d", i);
1040 if (unlikely(l >= (sizeof(str) - 1))) {
1041 err = -ENOMEM;
1042 goto out;
1043 }
1044 err = xenbus_scanf(XBT_NIL, pdev->xdev->otherend, str,
1045 "%x:%x:%x.%x", &domain, &bus, &slot, &func);
1046 if (err != 4) {
1047 if (err >= 0)
1048 err = -EINVAL;
1049 xenbus_dev_fatal(pdev->xdev, err,
1050 "Error reading PCI device %d", i);
1051 goto out;
1052 }
1053
2ccc246d
JL		 1054 pci_dev = pci_get_domain_bus_and_slot(domain, bus,
1055 PCI_DEVFN(slot, func));
956a9202
RW		 1056 if (!pci_dev) {
1057 dev_dbg(&pdev->xdev->dev,
e4de866a 1058 "Cannot get PCI device %04x:%02x:%02x.%d\n",
956a9202
RW		 1059 domain, bus, slot, func);
1060 continue;
1061 }
a83919e0 1062 pci_lock_rescan_remove();
210647af 1063 pci_stop_and_remove_bus_device(pci_dev);
956a9202 1064 pci_dev_put(pci_dev);
a83919e0 1065 pci_unlock_rescan_remove();
956a9202
RW		 1066
1067 dev_dbg(&pdev->xdev->dev,
e4de866a 1068 "PCI device %04x:%02x:%02x.%d removed.\n",
956a9202
RW		 1069 domain, bus, slot, func);
1070 }
1071
1072 err = xenbus_switch_state(pdev->xdev, XenbusStateReconfiguring);
1073
1074out:
1075 return err;
1076}
1077
1078static void __init_refok pcifront_backend_changed(struct xenbus_device *xdev,
1079 enum xenbus_state be_state)
1080{
1081 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1082
1083 switch (be_state) {
1084 case XenbusStateUnknown:
1085 case XenbusStateInitialising:
1086 case XenbusStateInitWait:
1087 case XenbusStateInitialised:
956a9202
RW		 1088 break;
1089
1090 case XenbusStateConnected:
1091 pcifront_try_connect(pdev);
1092 break;
1093
d5af64de
DV		 1094 case XenbusStateClosed:
1095 if (xdev->state == XenbusStateClosed)
1096 break;
1097 /* Missed the backend's CLOSING state -- fallthrough */
956a9202
RW		 1098 case XenbusStateClosing:
1099 dev_warn(&xdev->dev, "backend going away!\n");
1100 pcifront_try_disconnect(pdev);
1101 break;
1102
1103 case XenbusStateReconfiguring:
1104 pcifront_detach_devices(pdev);
1105 break;
1106
1107 case XenbusStateReconfigured:
1108 pcifront_attach_devices(pdev);
1109 break;
1110 }
1111}
1112
1113static int pcifront_xenbus_probe(struct xenbus_device *xdev,
1114 const struct xenbus_device_id *id)
1115{
1116 int err = 0;
1117 struct pcifront_device *pdev = alloc_pdev(xdev);
1118
1119 if (pdev == NULL) {
1120 err = -ENOMEM;
1121 xenbus_dev_fatal(xdev, err,
1122 "Error allocating pcifront_device struct");
1123 goto out;
1124 }
1125
1126 err = pcifront_publish_info(pdev);
1127 if (err)
1128 free_pdev(pdev);
1129
1130out:
1131 return err;
1132}
1133
1134static int pcifront_xenbus_remove(struct xenbus_device *xdev)
1135{
1136 struct pcifront_device *pdev = dev_get_drvdata(&xdev->dev);
1137 if (pdev)
1138 free_pdev(pdev);
1139
1140 return 0;
1141}
1142
1143static const struct xenbus_device_id xenpci_ids[] = {
1144 {"pci"},
1145 {""},
1146};
1147
95afae48
DV		 1148static struct xenbus_driver xenpci_driver = {
1149 .name = "pcifront",
1150 .ids = xenpci_ids,
956a9202
RW		 1151 .probe = pcifront_xenbus_probe,
1152 .remove = pcifront_xenbus_remove,
1153 .otherend_changed = pcifront_backend_changed,
95afae48 1154};
956a9202
RW		 1155
1156static int __init pcifront_init(void)
1157{
1158 if (!xen_pv_domain() || xen_initial_domain())
1159 return -ENODEV;
1160
51c71a3b
KRW		 1161 if (!xen_has_pv_devices())
1162 return -ENODEV;
1163
956a9202
RW		 1164 pci_frontend_registrar(1 /* enable */);
1165
73db144b 1166 return xenbus_register_frontend(&xenpci_driver);
956a9202
RW		 1167}
1168
1169static void __exit pcifront_cleanup(void)
1170{
73db144b 1171 xenbus_unregister_driver(&xenpci_driver);
956a9202
RW		 1172 pci_frontend_registrar(0 /* disable */);
1173}
1174module_init(pcifront_init);
1175module_exit(pcifront_cleanup);
1176
1177MODULE_DESCRIPTION("Xen PCI passthrough frontend.");
1178MODULE_LICENSE("GPL");
1179MODULE_ALIAS("xen:pci");
Linux 6.x block layer and io_uring tree(s)