1 // SPDX-License-Identifier: GPL-2.0
3 * PCI Express I/O Virtualization (IOV) support
5 * Address Translation Service 1.0
7 * Copyright (C) 2009 Intel Corporation, Yu Zhao <yu.zhao@intel.com>
10 #include <linux/pci.h>
11 #include <linux/slab.h>
12 #include <linux/mutex.h>
13 #include <linux/export.h>
14 #include <linux/string.h>
15 #include <linux/delay.h>
16 #include <linux/pci-ats.h>
19 #define VIRTFN_ID_LEN 16
21 int pci_iov_virtfn_bus(struct pci_dev *dev, int vf_id)
25 return dev->bus->number + ((dev->devfn + dev->sriov->offset +
26 dev->sriov->stride * vf_id) >> 8);
29 int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id)
33 return (dev->devfn + dev->sriov->offset +
34 dev->sriov->stride * vf_id) & 0xff;
38 * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may
39 * change when NumVFs changes.
41 * Update iov->offset and iov->stride when NumVFs is written.
43 static inline void pci_iov_set_numvfs(struct pci_dev *dev, int nr_virtfn)
45 struct pci_sriov *iov = dev->sriov;
47 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
48 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &iov->offset);
49 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &iov->stride);
53 * The PF consumes one bus number. NumVFs, First VF Offset, and VF Stride
54 * determine how many additional bus numbers will be consumed by VFs.
56 * Iterate over all valid NumVFs, validate offset and stride, and calculate
57 * the maximum number of bus numbers that could ever be required.
59 static int compute_max_vf_buses(struct pci_dev *dev)
61 struct pci_sriov *iov = dev->sriov;
62 int nr_virtfn, busnr, rc = 0;
64 for (nr_virtfn = iov->total_VFs; nr_virtfn; nr_virtfn--) {
65 pci_iov_set_numvfs(dev, nr_virtfn);
66 if (!iov->offset || (nr_virtfn > 1 && !iov->stride)) {
71 busnr = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
72 if (busnr > iov->max_VF_buses)
73 iov->max_VF_buses = busnr;
77 pci_iov_set_numvfs(dev, 0);
81 static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
83 struct pci_bus *child;
85 if (bus->number == busnr)
88 child = pci_find_bus(pci_domain_nr(bus), busnr);
92 child = pci_add_new_bus(bus, NULL, busnr);
96 pci_bus_insert_busn_res(child, busnr, busnr);
101 static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
103 if (physbus != virtbus && list_empty(&virtbus->devices))
104 pci_remove_bus(virtbus);
107 resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno)
112 return dev->sriov->barsz[resno - PCI_IOV_RESOURCES];
115 static void pci_read_vf_config_common(struct pci_dev *virtfn)
117 struct pci_dev *physfn = virtfn->physfn;
120 * Some config registers are the same across all associated VFs.
121 * Read them once from VF0 so we can skip reading them from the
124 * PCIe r4.0, sec 9.3.4.1, technically doesn't require all VFs to
125 * have the same Revision ID and Subsystem ID, but we assume they
128 pci_read_config_dword(virtfn, PCI_CLASS_REVISION,
129 &physfn->sriov->class);
130 pci_read_config_byte(virtfn, PCI_HEADER_TYPE,
131 &physfn->sriov->hdr_type);
132 pci_read_config_word(virtfn, PCI_SUBSYSTEM_VENDOR_ID,
133 &physfn->sriov->subsystem_vendor);
134 pci_read_config_word(virtfn, PCI_SUBSYSTEM_ID,
135 &physfn->sriov->subsystem_device);
138 int pci_iov_add_virtfn(struct pci_dev *dev, int id)
143 char buf[VIRTFN_ID_LEN];
144 struct pci_dev *virtfn;
145 struct resource *res;
146 struct pci_sriov *iov = dev->sriov;
149 bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id));
153 virtfn = pci_alloc_dev(bus);
157 virtfn->devfn = pci_iov_virtfn_devfn(dev, id);
158 virtfn->vendor = dev->vendor;
159 virtfn->device = iov->vf_device;
160 virtfn->is_virtfn = 1;
161 virtfn->physfn = pci_dev_get(dev);
164 pci_read_vf_config_common(virtfn);
166 rc = pci_setup_device(virtfn);
170 virtfn->dev.parent = dev->dev.parent;
171 virtfn->multifunction = 0;
173 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
174 res = &dev->resource[i + PCI_IOV_RESOURCES];
177 virtfn->resource[i].name = pci_name(virtfn);
178 virtfn->resource[i].flags = res->flags;
179 size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
180 virtfn->resource[i].start = res->start + size * id;
181 virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
182 rc = request_resource(res, &virtfn->resource[i]);
186 pci_device_add(virtfn, virtfn->bus);
188 sprintf(buf, "virtfn%u", id);
189 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
192 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
196 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
198 pci_bus_add_device(virtfn);
203 sysfs_remove_link(&dev->dev.kobj, buf);
205 pci_stop_and_remove_bus_device(virtfn);
209 virtfn_remove_bus(dev->bus, bus);
215 void pci_iov_remove_virtfn(struct pci_dev *dev, int id)
217 char buf[VIRTFN_ID_LEN];
218 struct pci_dev *virtfn;
220 virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
221 pci_iov_virtfn_bus(dev, id),
222 pci_iov_virtfn_devfn(dev, id));
226 sprintf(buf, "virtfn%u", id);
227 sysfs_remove_link(&dev->dev.kobj, buf);
229 * pci_stop_dev() could have been called for this virtfn already,
230 * so the directory for the virtfn may have been removed before.
231 * Double check to avoid spurious sysfs warnings.
233 if (virtfn->dev.kobj.sd)
234 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
236 pci_stop_and_remove_bus_device(virtfn);
237 virtfn_remove_bus(dev->bus, virtfn->bus);
239 /* balance pci_get_domain_bus_and_slot() */
244 int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
249 int __weak pcibios_sriov_disable(struct pci_dev *pdev)
254 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
260 struct resource *res;
261 struct pci_dev *pdev;
262 struct pci_sriov *iov = dev->sriov;
272 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
273 if (initial > iov->total_VFs ||
274 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
277 if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
278 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
282 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
283 bars |= (1 << (i + PCI_IOV_RESOURCES));
284 res = &dev->resource[i + PCI_IOV_RESOURCES];
288 if (nres != iov->nres) {
289 pci_err(dev, "not enough MMIO resources for SR-IOV\n");
293 bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
294 if (bus > dev->bus->busn_res.end) {
295 pci_err(dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
296 nr_virtfn, bus, &dev->bus->busn_res);
300 if (pci_enable_resources(dev, bars)) {
301 pci_err(dev, "SR-IOV: IOV BARS not allocated\n");
305 if (iov->link != dev->devfn) {
306 pdev = pci_get_slot(dev->bus, iov->link);
310 if (!pdev->is_physfn) {
315 rc = sysfs_create_link(&dev->dev.kobj,
316 &pdev->dev.kobj, "dep_link");
322 iov->initial_VFs = initial;
323 if (nr_virtfn < initial)
326 rc = pcibios_sriov_enable(dev, initial);
328 pci_err(dev, "failure %d from pcibios_sriov_enable()\n", rc);
332 pci_iov_set_numvfs(dev, nr_virtfn);
333 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
334 pci_cfg_access_lock(dev);
335 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
337 pci_cfg_access_unlock(dev);
339 for (i = 0; i < initial; i++) {
340 rc = pci_iov_add_virtfn(dev, i);
345 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
346 iov->num_VFs = nr_virtfn;
352 pci_iov_remove_virtfn(dev, i);
355 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
356 pci_cfg_access_lock(dev);
357 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
359 pci_cfg_access_unlock(dev);
361 pcibios_sriov_disable(dev);
363 if (iov->link != dev->devfn)
364 sysfs_remove_link(&dev->dev.kobj, "dep_link");
366 pci_iov_set_numvfs(dev, 0);
370 static void sriov_disable(struct pci_dev *dev)
373 struct pci_sriov *iov = dev->sriov;
378 for (i = 0; i < iov->num_VFs; i++)
379 pci_iov_remove_virtfn(dev, i);
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 pcibios_sriov_disable(dev);
389 if (iov->link != dev->devfn)
390 sysfs_remove_link(&dev->dev.kobj, "dep_link");
393 pci_iov_set_numvfs(dev, 0);
396 static int sriov_init(struct pci_dev *dev, int pos)
403 struct pci_sriov *iov;
404 struct resource *res;
405 struct pci_dev *pdev;
407 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
408 if (ctrl & PCI_SRIOV_CTRL_VFE) {
409 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
414 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
419 if (pci_ari_enabled(dev->bus))
420 ctrl |= PCI_SRIOV_CTRL_ARI;
423 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
425 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
429 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
430 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
431 pgsz &= ~((1 << i) - 1);
436 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
438 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
443 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
444 res = &dev->resource[i + PCI_IOV_RESOURCES];
446 * If it is already FIXED, don't change it, something
447 * (perhaps EA or header fixups) wants it this way.
449 if (res->flags & IORESOURCE_PCI_FIXED)
450 bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
452 bar64 = __pci_read_base(dev, pci_bar_unknown, res,
453 pos + PCI_SRIOV_BAR + i * 4);
456 if (resource_size(res) & (PAGE_SIZE - 1)) {
460 iov->barsz[i] = resource_size(res);
461 res->end = res->start + resource_size(res) * total - 1;
462 pci_info(dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
471 iov->total_VFs = total;
472 pci_read_config_word(dev, pos + PCI_SRIOV_VF_DID, &iov->vf_device);
475 iov->drivers_autoprobe = true;
476 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
477 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
478 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
479 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
482 iov->dev = pci_dev_get(pdev);
488 rc = compute_max_vf_buses(dev);
498 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
499 res = &dev->resource[i + PCI_IOV_RESOURCES];
507 static void sriov_release(struct pci_dev *dev)
509 BUG_ON(dev->sriov->num_VFs);
511 if (dev != dev->sriov->dev)
512 pci_dev_put(dev->sriov->dev);
518 static void sriov_restore_state(struct pci_dev *dev)
522 struct pci_sriov *iov = dev->sriov;
524 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
525 if (ctrl & PCI_SRIOV_CTRL_VFE)
529 * Restore PCI_SRIOV_CTRL_ARI before pci_iov_set_numvfs() because
530 * it reads offset & stride, which depend on PCI_SRIOV_CTRL_ARI.
532 ctrl &= ~PCI_SRIOV_CTRL_ARI;
533 ctrl |= iov->ctrl & PCI_SRIOV_CTRL_ARI;
534 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, ctrl);
536 for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
537 pci_update_resource(dev, i);
539 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
540 pci_iov_set_numvfs(dev, iov->num_VFs);
541 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
542 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
547 * pci_iov_init - initialize the IOV capability
548 * @dev: the PCI device
550 * Returns 0 on success, or negative on failure.
552 int pci_iov_init(struct pci_dev *dev)
556 if (!pci_is_pcie(dev))
559 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
561 return sriov_init(dev, pos);
567 * pci_iov_release - release resources used by the IOV capability
568 * @dev: the PCI device
570 void pci_iov_release(struct pci_dev *dev)
577 * pci_iov_update_resource - update a VF BAR
578 * @dev: the PCI device
579 * @resno: the resource number
581 * Update a VF BAR in the SR-IOV capability of a PF.
583 void pci_iov_update_resource(struct pci_dev *dev, int resno)
585 struct pci_sriov *iov = dev->is_physfn ? dev->sriov : NULL;
586 struct resource *res = dev->resource + resno;
587 int vf_bar = resno - PCI_IOV_RESOURCES;
588 struct pci_bus_region region;
594 * The generic pci_restore_bars() path calls this for all devices,
595 * including VFs and non-SR-IOV devices. If this is not a PF, we
596 * have nothing to do.
601 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &cmd);
602 if ((cmd & PCI_SRIOV_CTRL_VFE) && (cmd & PCI_SRIOV_CTRL_MSE)) {
603 dev_WARN(&dev->dev, "can't update enabled VF BAR%d %pR\n",
609 * Ignore unimplemented BARs, unused resource slots for 64-bit
610 * BARs, and non-movable resources, e.g., those described via
611 * Enhanced Allocation.
616 if (res->flags & IORESOURCE_UNSET)
619 if (res->flags & IORESOURCE_PCI_FIXED)
622 pcibios_resource_to_bus(dev->bus, ®ion, res);
624 new |= res->flags & ~PCI_BASE_ADDRESS_MEM_MASK;
626 reg = iov->pos + PCI_SRIOV_BAR + 4 * vf_bar;
627 pci_write_config_dword(dev, reg, new);
628 if (res->flags & IORESOURCE_MEM_64) {
629 new = region.start >> 16 >> 16;
630 pci_write_config_dword(dev, reg + 4, new);
634 resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev,
637 return pci_iov_resource_size(dev, resno);
641 * pci_sriov_resource_alignment - get resource alignment for VF BAR
642 * @dev: the PCI device
643 * @resno: the resource number
645 * Returns the alignment of the VF BAR found in the SR-IOV capability.
646 * This is not the same as the resource size which is defined as
647 * the VF BAR size multiplied by the number of VFs. The alignment
648 * is just the VF BAR size.
650 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
652 return pcibios_iov_resource_alignment(dev, resno);
656 * pci_restore_iov_state - restore the state of the IOV capability
657 * @dev: the PCI device
659 void pci_restore_iov_state(struct pci_dev *dev)
662 sriov_restore_state(dev);
666 * pci_vf_drivers_autoprobe - set PF property drivers_autoprobe for VFs
667 * @dev: the PCI device
668 * @auto_probe: set VF drivers auto probe flag
670 void pci_vf_drivers_autoprobe(struct pci_dev *dev, bool auto_probe)
673 dev->sriov->drivers_autoprobe = auto_probe;
677 * pci_iov_bus_range - find bus range used by Virtual Function
680 * Returns max number of buses (exclude current one) used by Virtual
683 int pci_iov_bus_range(struct pci_bus *bus)
688 list_for_each_entry(dev, &bus->devices, bus_list) {
691 if (dev->sriov->max_VF_buses > max)
692 max = dev->sriov->max_VF_buses;
695 return max ? max - bus->number : 0;
699 * pci_enable_sriov - enable the SR-IOV capability
700 * @dev: the PCI device
701 * @nr_virtfn: number of virtual functions to enable
703 * Returns 0 on success, or negative on failure.
705 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
712 return sriov_enable(dev, nr_virtfn);
714 EXPORT_SYMBOL_GPL(pci_enable_sriov);
717 * pci_disable_sriov - disable the SR-IOV capability
718 * @dev: the PCI device
720 void pci_disable_sriov(struct pci_dev *dev)
729 EXPORT_SYMBOL_GPL(pci_disable_sriov);
732 * pci_num_vf - return number of VFs associated with a PF device_release_driver
733 * @dev: the PCI device
735 * Returns number of VFs, or 0 if SR-IOV is not enabled.
737 int pci_num_vf(struct pci_dev *dev)
742 return dev->sriov->num_VFs;
744 EXPORT_SYMBOL_GPL(pci_num_vf);
747 * pci_vfs_assigned - returns number of VFs are assigned to a guest
748 * @dev: the PCI device
750 * Returns number of VFs belonging to this device that are assigned to a guest.
751 * If device is not a physical function returns 0.
753 int pci_vfs_assigned(struct pci_dev *dev)
755 struct pci_dev *vfdev;
756 unsigned int vfs_assigned = 0;
757 unsigned short dev_id;
759 /* only search if we are a PF */
764 * determine the device ID for the VFs, the vendor ID will be the
765 * same as the PF so there is no need to check for that one
767 dev_id = dev->sriov->vf_device;
769 /* loop through all the VFs to see if we own any that are assigned */
770 vfdev = pci_get_device(dev->vendor, dev_id, NULL);
773 * It is considered assigned if it is a virtual function with
774 * our dev as the physical function and the assigned bit is set
776 if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
777 pci_is_dev_assigned(vfdev))
780 vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
785 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
788 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
789 * @dev: the PCI PF device
790 * @numvfs: number that should be used for TotalVFs supported
792 * Should be called from PF driver's probe routine with
793 * device's mutex held.
795 * Returns 0 if PF is an SRIOV-capable device and
796 * value of numvfs valid. If not a PF return -ENOSYS;
797 * if numvfs is invalid return -EINVAL;
798 * if VFs already enabled, return -EBUSY.
800 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
804 if (numvfs > dev->sriov->total_VFs)
807 /* Shouldn't change if VFs already enabled */
808 if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
811 dev->sriov->driver_max_VFs = numvfs;
815 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
818 * pci_sriov_get_totalvfs -- get total VFs supported on this device
819 * @dev: the PCI PF device
821 * For a PCIe device with SRIOV support, return the PCIe
822 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
823 * if the driver reduced it. Otherwise 0.
825 int pci_sriov_get_totalvfs(struct pci_dev *dev)
830 if (dev->sriov->driver_max_VFs)
831 return dev->sriov->driver_max_VFs;
833 return dev->sriov->total_VFs;
835 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);