4 * Copyright (C) 2009 Intel Corporation, Yu Zhao <yu.zhao@intel.com>
6 * PCI Express I/O Virtualization (IOV) support.
8 * Address Translation Service 1.0
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>
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)
37 struct pci_bus *child;
39 if (bus->number == busnr)
42 child = pci_find_bus(pci_domain_nr(bus), busnr);
46 child = pci_add_new_bus(bus, NULL, busnr);
50 pci_bus_insert_busn_res(child, busnr, busnr);
51 child->dev.parent = bus->bridge;
52 rc = pci_bus_add_child(child);
54 pci_remove_bus(child);
61 static void virtfn_remove_bus(struct pci_bus *bus, int busnr)
63 struct pci_bus *child;
65 if (bus->number == busnr)
68 child = pci_find_bus(pci_domain_nr(bus), busnr);
71 if (list_empty(&child->devices))
72 pci_remove_bus(child);
75 static int virtfn_add(struct pci_dev *dev, int id, int reset)
80 char buf[VIRTFN_ID_LEN];
81 struct pci_dev *virtfn;
83 struct pci_sriov *iov = dev->sriov;
85 virtfn = alloc_pci_dev();
89 mutex_lock(&iov->dev->sriov->lock);
90 virtfn->bus = virtfn_add_bus(dev->bus, virtfn_bus(dev, id));
93 mutex_unlock(&iov->dev->sriov->lock);
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;
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]);
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);
128 sprintf(buf, "virtfn%u", id);
129 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
132 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
136 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
141 sysfs_remove_link(&dev->dev.kobj, buf);
144 mutex_lock(&iov->dev->sriov->lock);
145 pci_stop_and_remove_bus_device(virtfn);
146 virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
147 mutex_unlock(&iov->dev->sriov->lock);
152 static void virtfn_remove(struct pci_dev *dev, int id, int reset)
154 char buf[VIRTFN_ID_LEN];
155 struct pci_dev *virtfn;
156 struct pci_sriov *iov = dev->sriov;
158 virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
159 virtfn_bus(dev, id), virtfn_devfn(dev, id));
166 device_release_driver(&virtfn->dev);
167 __pci_reset_function(virtfn);
170 sprintf(buf, "virtfn%u", id);
171 sysfs_remove_link(&dev->dev.kobj, buf);
173 * pci_stop_dev() could have been called for this virtfn already,
174 * so the directory for the virtfn may have been removed before.
175 * Double check to avoid spurious sysfs warnings.
177 if (virtfn->dev.kobj.sd)
178 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
180 mutex_lock(&iov->dev->sriov->lock);
181 pci_stop_and_remove_bus_device(virtfn);
182 virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
183 mutex_unlock(&iov->dev->sriov->lock);
188 static int sriov_migration(struct pci_dev *dev)
191 struct pci_sriov *iov = dev->sriov;
196 if (!(iov->cap & PCI_SRIOV_CAP_VFM))
199 pci_read_config_word(dev, iov->pos + PCI_SRIOV_STATUS, &status);
200 if (!(status & PCI_SRIOV_STATUS_VFM))
203 schedule_work(&iov->mtask);
208 static void sriov_migration_task(struct work_struct *work)
213 struct pci_sriov *iov = container_of(work, struct pci_sriov, mtask);
215 for (i = iov->initial; i < iov->nr_virtfn; i++) {
216 state = readb(iov->mstate + i);
217 if (state == PCI_SRIOV_VFM_MI) {
218 writeb(PCI_SRIOV_VFM_AV, iov->mstate + i);
219 state = readb(iov->mstate + i);
220 if (state == PCI_SRIOV_VFM_AV)
221 virtfn_add(iov->self, i, 1);
222 } else if (state == PCI_SRIOV_VFM_MO) {
223 virtfn_remove(iov->self, i, 1);
224 writeb(PCI_SRIOV_VFM_UA, iov->mstate + i);
225 state = readb(iov->mstate + i);
226 if (state == PCI_SRIOV_VFM_AV)
227 virtfn_add(iov->self, i, 0);
231 pci_read_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, &status);
232 status &= ~PCI_SRIOV_STATUS_VFM;
233 pci_write_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, status);
236 static int sriov_enable_migration(struct pci_dev *dev, int nr_virtfn)
241 struct pci_sriov *iov = dev->sriov;
243 if (nr_virtfn <= iov->initial)
246 pci_read_config_dword(dev, iov->pos + PCI_SRIOV_VFM, &table);
247 bir = PCI_SRIOV_VFM_BIR(table);
248 if (bir > PCI_STD_RESOURCE_END)
251 table = PCI_SRIOV_VFM_OFFSET(table);
252 if (table + nr_virtfn > pci_resource_len(dev, bir))
255 pa = pci_resource_start(dev, bir) + table;
256 iov->mstate = ioremap(pa, nr_virtfn);
260 INIT_WORK(&iov->mtask, sriov_migration_task);
262 iov->ctrl |= PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR;
263 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
268 static void sriov_disable_migration(struct pci_dev *dev)
270 struct pci_sriov *iov = dev->sriov;
272 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR);
273 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
275 cancel_work_sync(&iov->mtask);
276 iounmap(iov->mstate);
279 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
284 u16 offset, stride, initial;
285 struct resource *res;
286 struct pci_dev *pdev;
287 struct pci_sriov *iov = dev->sriov;
296 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
297 if (initial > iov->total ||
298 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total)))
301 if (nr_virtfn < 0 || nr_virtfn > iov->total ||
302 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
305 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
306 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
307 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
308 if (!offset || (nr_virtfn > 1 && !stride))
312 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
313 bars |= (1 << (i + PCI_IOV_RESOURCES));
314 res = dev->resource + PCI_IOV_RESOURCES + i;
318 if (nres != iov->nres) {
319 dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
323 iov->offset = offset;
324 iov->stride = stride;
326 if (virtfn_bus(dev, nr_virtfn - 1) > dev->bus->busn_res.end) {
327 dev_err(&dev->dev, "SR-IOV: bus number out of range\n");
331 if (pci_enable_resources(dev, bars)) {
332 dev_err(&dev->dev, "SR-IOV: IOV BARS not allocated\n");
336 if (iov->link != dev->devfn) {
337 pdev = pci_get_slot(dev->bus, iov->link);
343 if (!pdev->is_physfn)
346 rc = sysfs_create_link(&dev->dev.kobj,
347 &pdev->dev.kobj, "dep_link");
352 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
353 pci_cfg_access_lock(dev);
354 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
356 pci_cfg_access_unlock(dev);
358 iov->initial = initial;
359 if (nr_virtfn < initial)
362 for (i = 0; i < initial; i++) {
363 rc = virtfn_add(dev, i, 0);
368 if (iov->cap & PCI_SRIOV_CAP_VFM) {
369 rc = sriov_enable_migration(dev, nr_virtfn);
374 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
375 iov->nr_virtfn = nr_virtfn;
380 for (j = 0; j < i; j++)
381 virtfn_remove(dev, j, 0);
383 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
384 pci_cfg_access_lock(dev);
385 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
387 pci_cfg_access_unlock(dev);
389 if (iov->link != dev->devfn)
390 sysfs_remove_link(&dev->dev.kobj, "dep_link");
395 static void sriov_disable(struct pci_dev *dev)
398 struct pci_sriov *iov = dev->sriov;
403 if (iov->cap & PCI_SRIOV_CAP_VFM)
404 sriov_disable_migration(dev);
406 for (i = 0; i < iov->nr_virtfn; i++)
407 virtfn_remove(dev, i, 0);
409 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
410 pci_cfg_access_lock(dev);
411 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
413 pci_cfg_access_unlock(dev);
415 if (iov->link != dev->devfn)
416 sysfs_remove_link(&dev->dev.kobj, "dep_link");
421 static int sriov_init(struct pci_dev *dev, int pos)
427 u16 ctrl, total, offset, stride;
428 struct pci_sriov *iov;
429 struct resource *res;
430 struct pci_dev *pdev;
432 if (pci_pcie_type(dev) != PCI_EXP_TYPE_RC_END &&
433 pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT)
436 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
437 if (ctrl & PCI_SRIOV_CTRL_VFE) {
438 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
442 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
447 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
452 if (pci_ari_enabled(dev->bus))
453 ctrl |= PCI_SRIOV_CTRL_ARI;
456 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
457 pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
458 pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
459 if (!offset || (total > 1 && !stride))
462 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
463 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
464 pgsz &= ~((1 << i) - 1);
469 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
472 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
473 res = dev->resource + PCI_IOV_RESOURCES + i;
474 i += __pci_read_base(dev, pci_bar_unknown, res,
475 pos + PCI_SRIOV_BAR + i * 4);
478 if (resource_size(res) & (PAGE_SIZE - 1)) {
482 res->end = res->start + resource_size(res) * total - 1;
486 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
496 iov->offset = offset;
497 iov->stride = stride;
500 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
501 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
502 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
503 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
506 iov->dev = pci_dev_get(pdev);
510 mutex_init(&iov->lock);
518 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
519 res = dev->resource + PCI_IOV_RESOURCES + i;
526 static void sriov_release(struct pci_dev *dev)
528 BUG_ON(dev->sriov->nr_virtfn);
530 if (dev != dev->sriov->dev)
531 pci_dev_put(dev->sriov->dev);
533 mutex_destroy(&dev->sriov->lock);
539 static void sriov_restore_state(struct pci_dev *dev)
543 struct pci_sriov *iov = dev->sriov;
545 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
546 if (ctrl & PCI_SRIOV_CTRL_VFE)
549 for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
550 pci_update_resource(dev, i);
552 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
553 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->nr_virtfn);
554 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
555 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
560 * pci_iov_init - initialize the IOV capability
561 * @dev: the PCI device
563 * Returns 0 on success, or negative on failure.
565 int pci_iov_init(struct pci_dev *dev)
569 if (!pci_is_pcie(dev))
572 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
574 return sriov_init(dev, pos);
580 * pci_iov_release - release resources used by the IOV capability
581 * @dev: the PCI device
583 void pci_iov_release(struct pci_dev *dev)
590 * pci_iov_resource_bar - get position of the SR-IOV BAR
591 * @dev: the PCI device
592 * @resno: the resource number
593 * @type: the BAR type to be filled in
595 * Returns position of the BAR encapsulated in the SR-IOV capability.
597 int pci_iov_resource_bar(struct pci_dev *dev, int resno,
598 enum pci_bar_type *type)
600 if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
603 BUG_ON(!dev->is_physfn);
605 *type = pci_bar_unknown;
607 return dev->sriov->pos + PCI_SRIOV_BAR +
608 4 * (resno - PCI_IOV_RESOURCES);
612 * pci_sriov_resource_alignment - get resource alignment for VF BAR
613 * @dev: the PCI device
614 * @resno: the resource number
616 * Returns the alignment of the VF BAR found in the SR-IOV capability.
617 * This is not the same as the resource size which is defined as
618 * the VF BAR size multiplied by the number of VFs. The alignment
619 * is just the VF BAR size.
621 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
624 enum pci_bar_type type;
625 int reg = pci_iov_resource_bar(dev, resno, &type);
630 __pci_read_base(dev, type, &tmp, reg);
631 return resource_alignment(&tmp);
635 * pci_restore_iov_state - restore the state of the IOV capability
636 * @dev: the PCI device
638 void pci_restore_iov_state(struct pci_dev *dev)
641 sriov_restore_state(dev);
645 * pci_iov_bus_range - find bus range used by Virtual Function
648 * Returns max number of buses (exclude current one) used by Virtual
651 int pci_iov_bus_range(struct pci_bus *bus)
657 list_for_each_entry(dev, &bus->devices, bus_list) {
660 busnr = virtfn_bus(dev, dev->sriov->total - 1);
665 return max ? max - bus->number : 0;
669 * pci_enable_sriov - enable the SR-IOV capability
670 * @dev: the PCI device
671 * @nr_virtfn: number of virtual functions to enable
673 * Returns 0 on success, or negative on failure.
675 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
682 return sriov_enable(dev, nr_virtfn);
684 EXPORT_SYMBOL_GPL(pci_enable_sriov);
687 * pci_disable_sriov - disable the SR-IOV capability
688 * @dev: the PCI device
690 void pci_disable_sriov(struct pci_dev *dev)
699 EXPORT_SYMBOL_GPL(pci_disable_sriov);
702 * pci_sriov_migration - notify SR-IOV core of Virtual Function Migration
703 * @dev: the PCI device
705 * Returns IRQ_HANDLED if the IRQ is handled, or IRQ_NONE if not.
707 * Physical Function driver is responsible to register IRQ handler using
708 * VF Migration Interrupt Message Number, and call this function when the
709 * interrupt is generated by the hardware.
711 irqreturn_t pci_sriov_migration(struct pci_dev *dev)
716 return sriov_migration(dev) ? IRQ_HANDLED : IRQ_NONE;
718 EXPORT_SYMBOL_GPL(pci_sriov_migration);
721 * pci_num_vf - return number of VFs associated with a PF device_release_driver
722 * @dev: the PCI device
724 * Returns number of VFs, or 0 if SR-IOV is not enabled.
726 int pci_num_vf(struct pci_dev *dev)
728 if (!dev || !dev->is_physfn)
731 return dev->sriov->nr_virtfn;
733 EXPORT_SYMBOL_GPL(pci_num_vf);