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)
36 struct pci_bus *child;
38 if (bus->number == busnr)
41 child = pci_find_bus(pci_domain_nr(bus), busnr);
45 child = pci_add_new_bus(bus, NULL, busnr);
49 pci_bus_insert_busn_res(child, busnr, busnr);
55 static void virtfn_remove_bus(struct pci_bus *bus, int busnr)
57 struct pci_bus *child;
59 if (bus->number == busnr)
62 child = pci_find_bus(pci_domain_nr(bus), busnr);
65 if (list_empty(&child->devices))
66 pci_remove_bus(child);
69 static int virtfn_add(struct pci_dev *dev, int id, int reset)
74 char buf[VIRTFN_ID_LEN];
75 struct pci_dev *virtfn;
77 struct pci_sriov *iov = dev->sriov;
80 virtfn = pci_alloc_dev(NULL);
84 mutex_lock(&iov->dev->sriov->lock);
85 bus = virtfn_add_bus(dev->bus, virtfn_bus(dev, id));
88 mutex_unlock(&iov->dev->sriov->lock);
91 virtfn->bus = pci_bus_get(bus);
92 virtfn->devfn = virtfn_devfn(dev, id);
93 virtfn->vendor = dev->vendor;
94 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
95 pci_setup_device(virtfn);
96 virtfn->dev.parent = dev->dev.parent;
98 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
99 res = dev->resource + PCI_IOV_RESOURCES + i;
102 virtfn->resource[i].name = pci_name(virtfn);
103 virtfn->resource[i].flags = res->flags;
104 size = resource_size(res);
105 do_div(size, iov->total_VFs);
106 virtfn->resource[i].start = res->start + size * id;
107 virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
108 rc = request_resource(res, &virtfn->resource[i]);
113 __pci_reset_function(virtfn);
115 pci_device_add(virtfn, virtfn->bus);
116 mutex_unlock(&iov->dev->sriov->lock);
118 virtfn->physfn = pci_dev_get(dev);
119 virtfn->is_virtfn = 1;
121 rc = pci_bus_add_device(virtfn);
122 sprintf(buf, "virtfn%u", id);
123 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
126 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
130 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
135 sysfs_remove_link(&dev->dev.kobj, buf);
138 mutex_lock(&iov->dev->sriov->lock);
139 pci_stop_and_remove_bus_device(virtfn);
140 virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
141 mutex_unlock(&iov->dev->sriov->lock);
146 static void virtfn_remove(struct pci_dev *dev, int id, int reset)
148 char buf[VIRTFN_ID_LEN];
150 struct pci_dev *virtfn;
151 struct pci_sriov *iov = dev->sriov;
153 bus = pci_find_bus(pci_domain_nr(dev->bus), virtfn_bus(dev, id));
157 virtfn = pci_get_slot(bus, virtfn_devfn(dev, id));
164 device_release_driver(&virtfn->dev);
165 __pci_reset_function(virtfn);
168 sprintf(buf, "virtfn%u", id);
169 sysfs_remove_link(&dev->dev.kobj, buf);
171 * pci_stop_dev() could have been called for this virtfn already,
172 * so the directory for the virtfn may have been removed before.
173 * Double check to avoid spurious sysfs warnings.
175 if (virtfn->dev.kobj.sd)
176 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
178 mutex_lock(&iov->dev->sriov->lock);
179 pci_stop_and_remove_bus_device(virtfn);
180 virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
181 mutex_unlock(&iov->dev->sriov->lock);
186 static int sriov_migration(struct pci_dev *dev)
189 struct pci_sriov *iov = dev->sriov;
194 if (!(iov->cap & PCI_SRIOV_CAP_VFM))
197 pci_read_config_word(dev, iov->pos + PCI_SRIOV_STATUS, &status);
198 if (!(status & PCI_SRIOV_STATUS_VFM))
201 schedule_work(&iov->mtask);
206 static void sriov_migration_task(struct work_struct *work)
211 struct pci_sriov *iov = container_of(work, struct pci_sriov, mtask);
213 for (i = iov->initial_VFs; i < iov->num_VFs; 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)
239 struct pci_sriov *iov = dev->sriov;
241 if (nr_virtfn <= iov->initial_VFs)
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)
249 table = PCI_SRIOV_VFM_OFFSET(table);
250 if (table + nr_virtfn > pci_resource_len(dev, bir))
253 pa = pci_resource_start(dev, bir) + table;
254 iov->mstate = ioremap(pa, nr_virtfn);
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);
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)
282 u16 offset, stride, initial;
283 struct resource *res;
284 struct pci_dev *pdev;
285 struct pci_sriov *iov = dev->sriov;
294 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
295 if (initial > iov->total_VFs ||
296 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
299 if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
300 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
303 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
304 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
305 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
306 if (!offset || (nr_virtfn > 1 && !stride))
310 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
311 bars |= (1 << (i + PCI_IOV_RESOURCES));
312 res = dev->resource + PCI_IOV_RESOURCES + i;
316 if (nres != iov->nres) {
317 dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
321 iov->offset = offset;
322 iov->stride = stride;
324 if (virtfn_bus(dev, nr_virtfn - 1) > dev->bus->busn_res.end) {
325 dev_err(&dev->dev, "SR-IOV: bus number out of range\n");
329 if (pci_enable_resources(dev, bars)) {
330 dev_err(&dev->dev, "SR-IOV: IOV BARS not allocated\n");
334 if (iov->link != dev->devfn) {
335 pdev = pci_get_slot(dev->bus, iov->link);
341 if (!pdev->is_physfn)
344 rc = sysfs_create_link(&dev->dev.kobj,
345 &pdev->dev.kobj, "dep_link");
350 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
351 pci_cfg_access_lock(dev);
352 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
354 pci_cfg_access_unlock(dev);
356 iov->initial_VFs = initial;
357 if (nr_virtfn < initial)
360 for (i = 0; i < initial; i++) {
361 rc = virtfn_add(dev, i, 0);
366 if (iov->cap & PCI_SRIOV_CAP_VFM) {
367 rc = sriov_enable_migration(dev, nr_virtfn);
372 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
373 iov->num_VFs = nr_virtfn;
378 for (j = 0; j < i; j++)
379 virtfn_remove(dev, j, 0);
381 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
382 pci_cfg_access_lock(dev);
383 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
385 pci_cfg_access_unlock(dev);
387 if (iov->link != dev->devfn)
388 sysfs_remove_link(&dev->dev.kobj, "dep_link");
393 static void sriov_disable(struct pci_dev *dev)
396 struct pci_sriov *iov = dev->sriov;
401 if (iov->cap & PCI_SRIOV_CAP_VFM)
402 sriov_disable_migration(dev);
404 for (i = 0; i < iov->num_VFs; i++)
405 virtfn_remove(dev, i, 0);
407 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
408 pci_cfg_access_lock(dev);
409 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
411 pci_cfg_access_unlock(dev);
413 if (iov->link != dev->devfn)
414 sysfs_remove_link(&dev->dev.kobj, "dep_link");
419 static int sriov_init(struct pci_dev *dev, int pos)
425 u16 ctrl, total, offset, stride;
426 struct pci_sriov *iov;
427 struct resource *res;
428 struct pci_dev *pdev;
430 if (pci_pcie_type(dev) != PCI_EXP_TYPE_RC_END &&
431 pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT)
434 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
435 if (ctrl & PCI_SRIOV_CTRL_VFE) {
436 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
440 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
445 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
450 if (pci_ari_enabled(dev->bus))
451 ctrl |= PCI_SRIOV_CTRL_ARI;
454 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
455 pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
456 pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
457 if (!offset || (total > 1 && !stride))
460 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
461 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
462 pgsz &= ~((1 << i) - 1);
467 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
470 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
471 res = dev->resource + PCI_IOV_RESOURCES + i;
472 i += __pci_read_base(dev, pci_bar_unknown, res,
473 pos + PCI_SRIOV_BAR + i * 4);
476 if (resource_size(res) & (PAGE_SIZE - 1)) {
480 res->end = res->start + resource_size(res) * total - 1;
484 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
493 iov->total_VFs = total;
494 iov->offset = offset;
495 iov->stride = stride;
498 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
499 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
500 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
501 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
504 iov->dev = pci_dev_get(pdev);
508 mutex_init(&iov->lock);
516 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
517 res = dev->resource + PCI_IOV_RESOURCES + i;
524 static void sriov_release(struct pci_dev *dev)
526 BUG_ON(dev->sriov->num_VFs);
528 if (dev != dev->sriov->dev)
529 pci_dev_put(dev->sriov->dev);
531 mutex_destroy(&dev->sriov->lock);
537 static void sriov_restore_state(struct pci_dev *dev)
541 struct pci_sriov *iov = dev->sriov;
543 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
544 if (ctrl & PCI_SRIOV_CTRL_VFE)
547 for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
548 pci_update_resource(dev, i);
550 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
551 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->num_VFs);
552 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
553 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
558 * pci_iov_init - initialize the IOV capability
559 * @dev: the PCI device
561 * Returns 0 on success, or negative on failure.
563 int pci_iov_init(struct pci_dev *dev)
567 if (!pci_is_pcie(dev))
570 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
572 return sriov_init(dev, pos);
578 * pci_iov_release - release resources used by the IOV capability
579 * @dev: the PCI device
581 void pci_iov_release(struct pci_dev *dev)
588 * pci_iov_resource_bar - get position of the SR-IOV BAR
589 * @dev: the PCI device
590 * @resno: the resource number
591 * @type: the BAR type to be filled in
593 * Returns position of the BAR encapsulated in the SR-IOV capability.
595 int pci_iov_resource_bar(struct pci_dev *dev, int resno,
596 enum pci_bar_type *type)
598 if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
601 BUG_ON(!dev->is_physfn);
603 *type = pci_bar_unknown;
605 return dev->sriov->pos + PCI_SRIOV_BAR +
606 4 * (resno - PCI_IOV_RESOURCES);
610 * pci_sriov_resource_alignment - get resource alignment for VF BAR
611 * @dev: the PCI device
612 * @resno: the resource number
614 * Returns the alignment of the VF BAR found in the SR-IOV capability.
615 * This is not the same as the resource size which is defined as
616 * the VF BAR size multiplied by the number of VFs. The alignment
617 * is just the VF BAR size.
619 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
622 enum pci_bar_type type;
623 int reg = pci_iov_resource_bar(dev, resno, &type);
628 __pci_read_base(dev, type, &tmp, reg);
629 return resource_alignment(&tmp);
633 * pci_restore_iov_state - restore the state of the IOV capability
634 * @dev: the PCI device
636 void pci_restore_iov_state(struct pci_dev *dev)
639 sriov_restore_state(dev);
643 * pci_iov_bus_range - find bus range used by Virtual Function
646 * Returns max number of buses (exclude current one) used by Virtual
649 int pci_iov_bus_range(struct pci_bus *bus)
655 list_for_each_entry(dev, &bus->devices, bus_list) {
658 busnr = virtfn_bus(dev, dev->sriov->total_VFs - 1);
663 return max ? max - bus->number : 0;
667 * pci_enable_sriov - enable the SR-IOV capability
668 * @dev: the PCI device
669 * @nr_virtfn: number of virtual functions to enable
671 * Returns 0 on success, or negative on failure.
673 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
680 return sriov_enable(dev, nr_virtfn);
682 EXPORT_SYMBOL_GPL(pci_enable_sriov);
685 * pci_disable_sriov - disable the SR-IOV capability
686 * @dev: the PCI device
688 void pci_disable_sriov(struct pci_dev *dev)
697 EXPORT_SYMBOL_GPL(pci_disable_sriov);
700 * pci_sriov_migration - notify SR-IOV core of Virtual Function Migration
701 * @dev: the PCI device
703 * Returns IRQ_HANDLED if the IRQ is handled, or IRQ_NONE if not.
705 * Physical Function driver is responsible to register IRQ handler using
706 * VF Migration Interrupt Message Number, and call this function when the
707 * interrupt is generated by the hardware.
709 irqreturn_t pci_sriov_migration(struct pci_dev *dev)
714 return sriov_migration(dev) ? IRQ_HANDLED : IRQ_NONE;
716 EXPORT_SYMBOL_GPL(pci_sriov_migration);
719 * pci_num_vf - return number of VFs associated with a PF device_release_driver
720 * @dev: the PCI device
722 * Returns number of VFs, or 0 if SR-IOV is not enabled.
724 int pci_num_vf(struct pci_dev *dev)
729 return dev->sriov->num_VFs;
731 EXPORT_SYMBOL_GPL(pci_num_vf);
734 * pci_vfs_assigned - returns number of VFs are assigned to a guest
735 * @dev: the PCI device
737 * Returns number of VFs belonging to this device that are assigned to a guest.
738 * If device is not a physical function returns -ENODEV.
740 int pci_vfs_assigned(struct pci_dev *dev)
742 struct pci_dev *vfdev;
743 unsigned int vfs_assigned = 0;
744 unsigned short dev_id;
746 /* only search if we are a PF */
751 * determine the device ID for the VFs, the vendor ID will be the
752 * same as the PF so there is no need to check for that one
754 pci_read_config_word(dev, dev->sriov->pos + PCI_SRIOV_VF_DID, &dev_id);
756 /* loop through all the VFs to see if we own any that are assigned */
757 vfdev = pci_get_device(dev->vendor, dev_id, NULL);
760 * It is considered assigned if it is a virtual function with
761 * our dev as the physical function and the assigned bit is set
763 if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
764 (vfdev->dev_flags & PCI_DEV_FLAGS_ASSIGNED))
767 vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
772 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
775 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
776 * @dev: the PCI PF device
777 * @numvfs: number that should be used for TotalVFs supported
779 * Should be called from PF driver's probe routine with
780 * device's mutex held.
782 * Returns 0 if PF is an SRIOV-capable device and
783 * value of numvfs valid. If not a PF with VFS, return -EINVAL;
784 * if VFs already enabled, return -EBUSY.
786 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
788 if (!dev->is_physfn || (numvfs > dev->sriov->total_VFs))
791 /* Shouldn't change if VFs already enabled */
792 if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
795 dev->sriov->driver_max_VFs = numvfs;
799 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
802 * pci_sriov_get_totalvfs -- get total VFs supported on this devic3
803 * @dev: the PCI PF device
805 * For a PCIe device with SRIOV support, return the PCIe
806 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
807 * if the driver reduced it. Otherwise, -EINVAL.
809 int pci_sriov_get_totalvfs(struct pci_dev *dev)
814 if (dev->sriov->driver_max_VFs)
815 return dev->sriov->driver_max_VFs;
817 return dev->sriov->total_VFs;
819 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);