Add linux-next specific files for 20110801
[linux-2.6/next.git] / drivers / pci / iov.c
blob75e26b511f8c6fc0b150b99bfff5043fa0660900
1 /*
2 * drivers/pci/iov.c
4 * Copyright (C) 2009 Intel Corporation, Yu Zhao <yu.zhao@intel.com>
6 * PCI Express I/O Virtualization (IOV) support.
7 * Single Root IOV 1.0
8 * Address Translation Service 1.0
9 */
11 #include <linux/pci.h>
12 #include <linux/slab.h>
13 #include <linux/mutex.h>
14 #include <linux/export.h>
15 #include <linux/string.h>
16 #include <linux/delay.h>
17 #include <linux/pci-ats.h>
18 #include "pci.h"
20 #define VIRTFN_ID_LEN 16
22 static inline u8 virtfn_bus(struct pci_dev *dev, int id)
24 return dev->bus->number + ((dev->devfn + dev->sriov->offset +
25 dev->sriov->stride * id) >> 8);
28 static inline u8 virtfn_devfn(struct pci_dev *dev, int id)
30 return (dev->devfn + dev->sriov->offset +
31 dev->sriov->stride * id) & 0xff;
34 static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
36 int rc;
37 struct pci_bus *child;
39 if (bus->number == busnr)
40 return bus;
42 child = pci_find_bus(pci_domain_nr(bus), busnr);
43 if (child)
44 return child;
46 child = pci_add_new_bus(bus, NULL, busnr);
47 if (!child)
48 return NULL;
50 child->subordinate = busnr;
51 child->dev.parent = bus->bridge;
52 rc = pci_bus_add_child(child);
53 if (rc) {
54 pci_remove_bus(child);
55 return NULL;
58 return child;
61 static void virtfn_remove_bus(struct pci_bus *bus, int busnr)
63 struct pci_bus *child;
65 if (bus->number == busnr)
66 return;
68 child = pci_find_bus(pci_domain_nr(bus), busnr);
69 BUG_ON(!child);
71 if (list_empty(&child->devices))
72 pci_remove_bus(child);
75 static int virtfn_add(struct pci_dev *dev, int id, int reset)
77 int i;
78 int rc;
79 u64 size;
80 char buf[VIRTFN_ID_LEN];
81 struct pci_dev *virtfn;
82 struct resource *res;
83 struct pci_sriov *iov = dev->sriov;
85 virtfn = alloc_pci_dev();
86 if (!virtfn)
87 return -ENOMEM;
89 mutex_lock(&iov->dev->sriov->lock);
90 virtfn->bus = virtfn_add_bus(dev->bus, virtfn_bus(dev, id));
91 if (!virtfn->bus) {
92 kfree(virtfn);
93 mutex_unlock(&iov->dev->sriov->lock);
94 return -ENOMEM;
96 virtfn->devfn = virtfn_devfn(dev, id);
97 virtfn->vendor = dev->vendor;
98 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
99 pci_setup_device(virtfn);
100 virtfn->dev.parent = dev->dev.parent;
102 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
103 res = dev->resource + PCI_IOV_RESOURCES + i;
104 if (!res->parent)
105 continue;
106 virtfn->resource[i].name = pci_name(virtfn);
107 virtfn->resource[i].flags = res->flags;
108 size = resource_size(res);
109 do_div(size, iov->total);
110 virtfn->resource[i].start = res->start + size * id;
111 virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
112 rc = request_resource(res, &virtfn->resource[i]);
113 BUG_ON(rc);
116 if (reset)
117 __pci_reset_function(virtfn);
119 pci_device_add(virtfn, virtfn->bus);
120 mutex_unlock(&iov->dev->sriov->lock);
122 virtfn->physfn = pci_dev_get(dev);
123 virtfn->is_virtfn = 1;
125 rc = pci_bus_add_device(virtfn);
126 if (rc)
127 goto failed1;
128 sprintf(buf, "virtfn%u", id);
129 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
130 if (rc)
131 goto failed1;
132 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
133 if (rc)
134 goto failed2;
136 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
138 return 0;
140 failed2:
141 sysfs_remove_link(&dev->dev.kobj, buf);
142 failed1:
143 pci_dev_put(dev);
144 mutex_lock(&iov->dev->sriov->lock);
145 pci_remove_bus_device(virtfn);
146 virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
147 mutex_unlock(&iov->dev->sriov->lock);
149 return rc;
152 static void virtfn_remove(struct pci_dev *dev, int id, int reset)
154 char buf[VIRTFN_ID_LEN];
155 struct pci_bus *bus;
156 struct pci_dev *virtfn;
157 struct pci_sriov *iov = dev->sriov;
159 bus = pci_find_bus(pci_domain_nr(dev->bus), virtfn_bus(dev, id));
160 if (!bus)
161 return;
163 virtfn = pci_get_slot(bus, virtfn_devfn(dev, id));
164 if (!virtfn)
165 return;
167 pci_dev_put(virtfn);
169 if (reset) {
170 device_release_driver(&virtfn->dev);
171 __pci_reset_function(virtfn);
174 sprintf(buf, "virtfn%u", id);
175 sysfs_remove_link(&dev->dev.kobj, buf);
176 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
178 mutex_lock(&iov->dev->sriov->lock);
179 pci_remove_bus_device(virtfn);
180 virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
181 mutex_unlock(&iov->dev->sriov->lock);
183 pci_dev_put(dev);
186 static int sriov_migration(struct pci_dev *dev)
188 u16 status;
189 struct pci_sriov *iov = dev->sriov;
191 if (!iov->nr_virtfn)
192 return 0;
194 if (!(iov->cap & PCI_SRIOV_CAP_VFM))
195 return 0;
197 pci_read_config_word(dev, iov->pos + PCI_SRIOV_STATUS, &status);
198 if (!(status & PCI_SRIOV_STATUS_VFM))
199 return 0;
201 schedule_work(&iov->mtask);
203 return 1;
206 static void sriov_migration_task(struct work_struct *work)
208 int i;
209 u8 state;
210 u16 status;
211 struct pci_sriov *iov = container_of(work, struct pci_sriov, mtask);
213 for (i = iov->initial; i < iov->nr_virtfn; i++) {
214 state = readb(iov->mstate + i);
215 if (state == PCI_SRIOV_VFM_MI) {
216 writeb(PCI_SRIOV_VFM_AV, iov->mstate + i);
217 state = readb(iov->mstate + i);
218 if (state == PCI_SRIOV_VFM_AV)
219 virtfn_add(iov->self, i, 1);
220 } else if (state == PCI_SRIOV_VFM_MO) {
221 virtfn_remove(iov->self, i, 1);
222 writeb(PCI_SRIOV_VFM_UA, iov->mstate + i);
223 state = readb(iov->mstate + i);
224 if (state == PCI_SRIOV_VFM_AV)
225 virtfn_add(iov->self, i, 0);
229 pci_read_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, &status);
230 status &= ~PCI_SRIOV_STATUS_VFM;
231 pci_write_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, status);
234 static int sriov_enable_migration(struct pci_dev *dev, int nr_virtfn)
236 int bir;
237 u32 table;
238 resource_size_t pa;
239 struct pci_sriov *iov = dev->sriov;
241 if (nr_virtfn <= iov->initial)
242 return 0;
244 pci_read_config_dword(dev, iov->pos + PCI_SRIOV_VFM, &table);
245 bir = PCI_SRIOV_VFM_BIR(table);
246 if (bir > PCI_STD_RESOURCE_END)
247 return -EIO;
249 table = PCI_SRIOV_VFM_OFFSET(table);
250 if (table + nr_virtfn > pci_resource_len(dev, bir))
251 return -EIO;
253 pa = pci_resource_start(dev, bir) + table;
254 iov->mstate = ioremap(pa, nr_virtfn);
255 if (!iov->mstate)
256 return -ENOMEM;
258 INIT_WORK(&iov->mtask, sriov_migration_task);
260 iov->ctrl |= PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR;
261 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
263 return 0;
266 static void sriov_disable_migration(struct pci_dev *dev)
268 struct pci_sriov *iov = dev->sriov;
270 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR);
271 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
273 cancel_work_sync(&iov->mtask);
274 iounmap(iov->mstate);
277 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
279 int rc;
280 int i, j;
281 int nres;
282 u16 offset, stride, initial;
283 struct resource *res;
284 struct pci_dev *pdev;
285 struct pci_sriov *iov = dev->sriov;
287 if (!nr_virtfn)
288 return 0;
290 if (iov->nr_virtfn)
291 return -EINVAL;
293 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
294 if (initial > iov->total ||
295 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total)))
296 return -EIO;
298 if (nr_virtfn < 0 || nr_virtfn > iov->total ||
299 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
300 return -EINVAL;
302 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
303 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
304 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
305 if (!offset || (nr_virtfn > 1 && !stride))
306 return -EIO;
308 nres = 0;
309 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
310 res = dev->resource + PCI_IOV_RESOURCES + i;
311 if (res->parent)
312 nres++;
314 if (nres != iov->nres) {
315 dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
316 return -ENOMEM;
319 iov->offset = offset;
320 iov->stride = stride;
322 if (virtfn_bus(dev, nr_virtfn - 1) > dev->bus->subordinate) {
323 dev_err(&dev->dev, "SR-IOV: bus number out of range\n");
324 return -ENOMEM;
327 if (iov->link != dev->devfn) {
328 pdev = pci_get_slot(dev->bus, iov->link);
329 if (!pdev)
330 return -ENODEV;
332 pci_dev_put(pdev);
334 if (!pdev->is_physfn)
335 return -ENODEV;
337 rc = sysfs_create_link(&dev->dev.kobj,
338 &pdev->dev.kobj, "dep_link");
339 if (rc)
340 return rc;
343 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
344 pci_block_user_cfg_access(dev);
345 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
346 msleep(100);
347 pci_unblock_user_cfg_access(dev);
349 iov->initial = initial;
350 if (nr_virtfn < initial)
351 initial = nr_virtfn;
353 for (i = 0; i < initial; i++) {
354 rc = virtfn_add(dev, i, 0);
355 if (rc)
356 goto failed;
359 if (iov->cap & PCI_SRIOV_CAP_VFM) {
360 rc = sriov_enable_migration(dev, nr_virtfn);
361 if (rc)
362 goto failed;
365 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
366 iov->nr_virtfn = nr_virtfn;
368 return 0;
370 failed:
371 for (j = 0; j < i; j++)
372 virtfn_remove(dev, j, 0);
374 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
375 pci_block_user_cfg_access(dev);
376 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
377 ssleep(1);
378 pci_unblock_user_cfg_access(dev);
380 if (iov->link != dev->devfn)
381 sysfs_remove_link(&dev->dev.kobj, "dep_link");
383 return rc;
386 static void sriov_disable(struct pci_dev *dev)
388 int i;
389 struct pci_sriov *iov = dev->sriov;
391 if (!iov->nr_virtfn)
392 return;
394 if (iov->cap & PCI_SRIOV_CAP_VFM)
395 sriov_disable_migration(dev);
397 for (i = 0; i < iov->nr_virtfn; i++)
398 virtfn_remove(dev, i, 0);
400 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
401 pci_block_user_cfg_access(dev);
402 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
403 ssleep(1);
404 pci_unblock_user_cfg_access(dev);
406 if (iov->link != dev->devfn)
407 sysfs_remove_link(&dev->dev.kobj, "dep_link");
409 iov->nr_virtfn = 0;
412 static int sriov_init(struct pci_dev *dev, int pos)
414 int i;
415 int rc;
416 int nres;
417 u32 pgsz;
418 u16 ctrl, total, offset, stride;
419 struct pci_sriov *iov;
420 struct resource *res;
421 struct pci_dev *pdev;
423 if (dev->pcie_type != PCI_EXP_TYPE_RC_END &&
424 dev->pcie_type != PCI_EXP_TYPE_ENDPOINT)
425 return -ENODEV;
427 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
428 if (ctrl & PCI_SRIOV_CTRL_VFE) {
429 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
430 ssleep(1);
433 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
434 if (!total)
435 return 0;
437 ctrl = 0;
438 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
439 if (pdev->is_physfn)
440 goto found;
442 pdev = NULL;
443 if (pci_ari_enabled(dev->bus))
444 ctrl |= PCI_SRIOV_CTRL_ARI;
446 found:
447 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
448 pci_write_config_word(dev, pos + PCI_SRIOV_NUM_VF, total);
449 pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
450 pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
451 if (!offset || (total > 1 && !stride))
452 return -EIO;
454 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
455 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
456 pgsz &= ~((1 << i) - 1);
457 if (!pgsz)
458 return -EIO;
460 pgsz &= ~(pgsz - 1);
461 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
463 nres = 0;
464 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
465 res = dev->resource + PCI_IOV_RESOURCES + i;
466 i += __pci_read_base(dev, pci_bar_unknown, res,
467 pos + PCI_SRIOV_BAR + i * 4);
468 if (!res->flags)
469 continue;
470 if (resource_size(res) & (PAGE_SIZE - 1)) {
471 rc = -EIO;
472 goto failed;
474 res->end = res->start + resource_size(res) * total - 1;
475 nres++;
478 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
479 if (!iov) {
480 rc = -ENOMEM;
481 goto failed;
484 iov->pos = pos;
485 iov->nres = nres;
486 iov->ctrl = ctrl;
487 iov->total = total;
488 iov->offset = offset;
489 iov->stride = stride;
490 iov->pgsz = pgsz;
491 iov->self = dev;
492 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
493 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
494 if (dev->pcie_type == PCI_EXP_TYPE_RC_END)
495 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
497 if (pdev)
498 iov->dev = pci_dev_get(pdev);
499 else
500 iov->dev = dev;
502 mutex_init(&iov->lock);
504 dev->sriov = iov;
505 dev->is_physfn = 1;
507 return 0;
509 failed:
510 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
511 res = dev->resource + PCI_IOV_RESOURCES + i;
512 res->flags = 0;
515 return rc;
518 static void sriov_release(struct pci_dev *dev)
520 BUG_ON(dev->sriov->nr_virtfn);
522 if (dev != dev->sriov->dev)
523 pci_dev_put(dev->sriov->dev);
525 mutex_destroy(&dev->sriov->lock);
527 kfree(dev->sriov);
528 dev->sriov = NULL;
531 static void sriov_restore_state(struct pci_dev *dev)
533 int i;
534 u16 ctrl;
535 struct pci_sriov *iov = dev->sriov;
537 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
538 if (ctrl & PCI_SRIOV_CTRL_VFE)
539 return;
541 for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
542 pci_update_resource(dev, i);
544 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
545 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->nr_virtfn);
546 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
547 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
548 msleep(100);
552 * pci_iov_init - initialize the IOV capability
553 * @dev: the PCI device
555 * Returns 0 on success, or negative on failure.
557 int pci_iov_init(struct pci_dev *dev)
559 int pos;
561 if (!pci_is_pcie(dev))
562 return -ENODEV;
564 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
565 if (pos)
566 return sriov_init(dev, pos);
568 return -ENODEV;
572 * pci_iov_release - release resources used by the IOV capability
573 * @dev: the PCI device
575 void pci_iov_release(struct pci_dev *dev)
577 if (dev->is_physfn)
578 sriov_release(dev);
582 * pci_iov_resource_bar - get position of the SR-IOV BAR
583 * @dev: the PCI device
584 * @resno: the resource number
585 * @type: the BAR type to be filled in
587 * Returns position of the BAR encapsulated in the SR-IOV capability.
589 int pci_iov_resource_bar(struct pci_dev *dev, int resno,
590 enum pci_bar_type *type)
592 if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
593 return 0;
595 BUG_ON(!dev->is_physfn);
597 *type = pci_bar_unknown;
599 return dev->sriov->pos + PCI_SRIOV_BAR +
600 4 * (resno - PCI_IOV_RESOURCES);
604 * pci_sriov_resource_alignment - get resource alignment for VF BAR
605 * @dev: the PCI device
606 * @resno: the resource number
608 * Returns the alignment of the VF BAR found in the SR-IOV capability.
609 * This is not the same as the resource size which is defined as
610 * the VF BAR size multiplied by the number of VFs. The alignment
611 * is just the VF BAR size.
613 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
615 struct resource tmp;
616 enum pci_bar_type type;
617 int reg = pci_iov_resource_bar(dev, resno, &type);
619 if (!reg)
620 return 0;
622 __pci_read_base(dev, type, &tmp, reg);
623 return resource_alignment(&tmp);
627 * pci_restore_iov_state - restore the state of the IOV capability
628 * @dev: the PCI device
630 void pci_restore_iov_state(struct pci_dev *dev)
632 if (dev->is_physfn)
633 sriov_restore_state(dev);
637 * pci_iov_bus_range - find bus range used by Virtual Function
638 * @bus: the PCI bus
640 * Returns max number of buses (exclude current one) used by Virtual
641 * Functions.
643 int pci_iov_bus_range(struct pci_bus *bus)
645 int max = 0;
646 u8 busnr;
647 struct pci_dev *dev;
649 list_for_each_entry(dev, &bus->devices, bus_list) {
650 if (!dev->is_physfn)
651 continue;
652 busnr = virtfn_bus(dev, dev->sriov->total - 1);
653 if (busnr > max)
654 max = busnr;
657 return max ? max - bus->number : 0;
661 * pci_enable_sriov - enable the SR-IOV capability
662 * @dev: the PCI device
663 * @nr_virtfn: number of virtual functions to enable
665 * Returns 0 on success, or negative on failure.
667 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
669 might_sleep();
671 if (!dev->is_physfn)
672 return -ENODEV;
674 return sriov_enable(dev, nr_virtfn);
676 EXPORT_SYMBOL_GPL(pci_enable_sriov);
679 * pci_disable_sriov - disable the SR-IOV capability
680 * @dev: the PCI device
682 void pci_disable_sriov(struct pci_dev *dev)
684 might_sleep();
686 if (!dev->is_physfn)
687 return;
689 sriov_disable(dev);
691 EXPORT_SYMBOL_GPL(pci_disable_sriov);
694 * pci_sriov_migration - notify SR-IOV core of Virtual Function Migration
695 * @dev: the PCI device
697 * Returns IRQ_HANDLED if the IRQ is handled, or IRQ_NONE if not.
699 * Physical Function driver is responsible to register IRQ handler using
700 * VF Migration Interrupt Message Number, and call this function when the
701 * interrupt is generated by the hardware.
703 irqreturn_t pci_sriov_migration(struct pci_dev *dev)
705 if (!dev->is_physfn)
706 return IRQ_NONE;
708 return sriov_migration(dev) ? IRQ_HANDLED : IRQ_NONE;
710 EXPORT_SYMBOL_GPL(pci_sriov_migration);
713 * pci_num_vf - return number of VFs associated with a PF device_release_driver
714 * @dev: the PCI device
716 * Returns number of VFs, or 0 if SR-IOV is not enabled.
718 int pci_num_vf(struct pci_dev *dev)
720 if (!dev || !dev->is_physfn)
721 return 0;
722 else
723 return dev->sriov->nr_virtfn;
725 EXPORT_SYMBOL_GPL(pci_num_vf);
727 static int ats_alloc_one(struct pci_dev *dev, int ps)
729 int pos;
730 u16 cap;
731 struct pci_ats *ats;
733 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ATS);
734 if (!pos)
735 return -ENODEV;
737 ats = kzalloc(sizeof(*ats), GFP_KERNEL);
738 if (!ats)
739 return -ENOMEM;
741 ats->pos = pos;
742 ats->stu = ps;
743 pci_read_config_word(dev, pos + PCI_ATS_CAP, &cap);
744 ats->qdep = PCI_ATS_CAP_QDEP(cap) ? PCI_ATS_CAP_QDEP(cap) :
745 PCI_ATS_MAX_QDEP;
746 dev->ats = ats;
748 return 0;
751 static void ats_free_one(struct pci_dev *dev)
753 kfree(dev->ats);
754 dev->ats = NULL;
758 * pci_enable_ats - enable the ATS capability
759 * @dev: the PCI device
760 * @ps: the IOMMU page shift
762 * Returns 0 on success, or negative on failure.
764 int pci_enable_ats(struct pci_dev *dev, int ps)
766 int rc;
767 u16 ctrl;
769 BUG_ON(dev->ats && dev->ats->is_enabled);
771 if (ps < PCI_ATS_MIN_STU)
772 return -EINVAL;
774 if (dev->is_physfn || dev->is_virtfn) {
775 struct pci_dev *pdev = dev->is_physfn ? dev : dev->physfn;
777 mutex_lock(&pdev->sriov->lock);
778 if (pdev->ats)
779 rc = pdev->ats->stu == ps ? 0 : -EINVAL;
780 else
781 rc = ats_alloc_one(pdev, ps);
783 if (!rc)
784 pdev->ats->ref_cnt++;
785 mutex_unlock(&pdev->sriov->lock);
786 if (rc)
787 return rc;
790 if (!dev->is_physfn) {
791 rc = ats_alloc_one(dev, ps);
792 if (rc)
793 return rc;
796 ctrl = PCI_ATS_CTRL_ENABLE;
797 if (!dev->is_virtfn)
798 ctrl |= PCI_ATS_CTRL_STU(ps - PCI_ATS_MIN_STU);
799 pci_write_config_word(dev, dev->ats->pos + PCI_ATS_CTRL, ctrl);
801 dev->ats->is_enabled = 1;
803 return 0;
807 * pci_disable_ats - disable the ATS capability
808 * @dev: the PCI device
810 void pci_disable_ats(struct pci_dev *dev)
812 u16 ctrl;
814 BUG_ON(!dev->ats || !dev->ats->is_enabled);
816 pci_read_config_word(dev, dev->ats->pos + PCI_ATS_CTRL, &ctrl);
817 ctrl &= ~PCI_ATS_CTRL_ENABLE;
818 pci_write_config_word(dev, dev->ats->pos + PCI_ATS_CTRL, ctrl);
820 dev->ats->is_enabled = 0;
822 if (dev->is_physfn || dev->is_virtfn) {
823 struct pci_dev *pdev = dev->is_physfn ? dev : dev->physfn;
825 mutex_lock(&pdev->sriov->lock);
826 pdev->ats->ref_cnt--;
827 if (!pdev->ats->ref_cnt)
828 ats_free_one(pdev);
829 mutex_unlock(&pdev->sriov->lock);
832 if (!dev->is_physfn)
833 ats_free_one(dev);
837 * pci_ats_queue_depth - query the ATS Invalidate Queue Depth
838 * @dev: the PCI device
840 * Returns the queue depth on success, or negative on failure.
842 * The ATS spec uses 0 in the Invalidate Queue Depth field to
843 * indicate that the function can accept 32 Invalidate Request.
844 * But here we use the `real' values (i.e. 1~32) for the Queue
845 * Depth; and 0 indicates the function shares the Queue with
846 * other functions (doesn't exclusively own a Queue).
848 int pci_ats_queue_depth(struct pci_dev *dev)
850 int pos;
851 u16 cap;
853 if (dev->is_virtfn)
854 return 0;
856 if (dev->ats)
857 return dev->ats->qdep;
859 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ATS);
860 if (!pos)
861 return -ENODEV;
863 pci_read_config_word(dev, pos + PCI_ATS_CAP, &cap);
865 return PCI_ATS_CAP_QDEP(cap) ? PCI_ATS_CAP_QDEP(cap) :
866 PCI_ATS_MAX_QDEP;