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>
18 #define VIRTFN_ID_LEN 16
20 int pci_iov_virtfn_bus(struct pci_dev *dev, int vf_id)
24 return dev->bus->number + ((dev->devfn + dev->sriov->offset +
25 dev->sriov->stride * vf_id) >> 8);
28 int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id)
32 return (dev->devfn + dev->sriov->offset +
33 dev->sriov->stride * vf_id) & 0xff;
37 * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may
38 * change when NumVFs changes.
40 * Update iov->offset and iov->stride when NumVFs is written.
42 static inline void pci_iov_set_numvfs(struct pci_dev *dev, int nr_virtfn)
44 struct pci_sriov *iov = dev->sriov;
46 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
47 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &iov->offset);
48 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &iov->stride);
52 * The PF consumes one bus number. NumVFs, First VF Offset, and VF Stride
53 * determine how many additional bus numbers will be consumed by VFs.
55 * Iterate over all valid NumVFs, validate offset and stride, and calculate
56 * the maximum number of bus numbers that could ever be required.
58 static int compute_max_vf_buses(struct pci_dev *dev)
60 struct pci_sriov *iov = dev->sriov;
61 int nr_virtfn, busnr, rc = 0;
63 for (nr_virtfn = iov->total_VFs; nr_virtfn; nr_virtfn--) {
64 pci_iov_set_numvfs(dev, nr_virtfn);
65 if (!iov->offset || (nr_virtfn > 1 && !iov->stride)) {
70 busnr = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
71 if (busnr > iov->max_VF_buses)
72 iov->max_VF_buses = busnr;
76 pci_iov_set_numvfs(dev, 0);
80 static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
82 struct pci_bus *child;
84 if (bus->number == busnr)
87 child = pci_find_bus(pci_domain_nr(bus), busnr);
91 child = pci_add_new_bus(bus, NULL, busnr);
95 pci_bus_insert_busn_res(child, busnr, busnr);
100 static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
102 if (physbus != virtbus && list_empty(&virtbus->devices))
103 pci_remove_bus(virtbus);
106 resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno)
111 return dev->sriov->barsz[resno - PCI_IOV_RESOURCES];
114 static void pci_read_vf_config_common(struct pci_dev *virtfn)
116 struct pci_dev *physfn = virtfn->physfn;
119 * Some config registers are the same across all associated VFs.
120 * Read them once from VF0 so we can skip reading them from the
123 * PCIe r4.0, sec 9.3.4.1, technically doesn't require all VFs to
124 * have the same Revision ID and Subsystem ID, but we assume they
127 pci_read_config_dword(virtfn, PCI_CLASS_REVISION,
128 &physfn->sriov->class);
129 pci_read_config_byte(virtfn, PCI_HEADER_TYPE,
130 &physfn->sriov->hdr_type);
131 pci_read_config_word(virtfn, PCI_SUBSYSTEM_VENDOR_ID,
132 &physfn->sriov->subsystem_vendor);
133 pci_read_config_word(virtfn, PCI_SUBSYSTEM_ID,
134 &physfn->sriov->subsystem_device);
137 int pci_iov_add_virtfn(struct pci_dev *dev, int id)
142 char buf[VIRTFN_ID_LEN];
143 struct pci_dev *virtfn;
144 struct resource *res;
145 struct pci_sriov *iov = dev->sriov;
148 bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id));
152 virtfn = pci_alloc_dev(bus);
156 virtfn->devfn = pci_iov_virtfn_devfn(dev, id);
157 virtfn->vendor = dev->vendor;
158 virtfn->device = iov->vf_device;
159 virtfn->is_virtfn = 1;
160 virtfn->physfn = pci_dev_get(dev);
163 pci_read_vf_config_common(virtfn);
165 rc = pci_setup_device(virtfn);
169 virtfn->dev.parent = dev->dev.parent;
170 virtfn->multifunction = 0;
172 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
173 res = &dev->resource[i + PCI_IOV_RESOURCES];
176 virtfn->resource[i].name = pci_name(virtfn);
177 virtfn->resource[i].flags = res->flags;
178 size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
179 virtfn->resource[i].start = res->start + size * id;
180 virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
181 rc = request_resource(res, &virtfn->resource[i]);
185 pci_device_add(virtfn, virtfn->bus);
187 sprintf(buf, "virtfn%u", id);
188 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
191 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
195 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
197 pci_bus_add_device(virtfn);
202 sysfs_remove_link(&dev->dev.kobj, buf);
204 pci_stop_and_remove_bus_device(virtfn);
208 virtfn_remove_bus(dev->bus, bus);
214 void pci_iov_remove_virtfn(struct pci_dev *dev, int id)
216 char buf[VIRTFN_ID_LEN];
217 struct pci_dev *virtfn;
219 virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
220 pci_iov_virtfn_bus(dev, id),
221 pci_iov_virtfn_devfn(dev, id));
225 sprintf(buf, "virtfn%u", id);
226 sysfs_remove_link(&dev->dev.kobj, buf);
228 * pci_stop_dev() could have been called for this virtfn already,
229 * so the directory for the virtfn may have been removed before.
230 * Double check to avoid spurious sysfs warnings.
232 if (virtfn->dev.kobj.sd)
233 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
235 pci_stop_and_remove_bus_device(virtfn);
236 virtfn_remove_bus(dev->bus, virtfn->bus);
238 /* balance pci_get_domain_bus_and_slot() */
243 int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
248 int __weak pcibios_sriov_disable(struct pci_dev *pdev)
253 static int sriov_add_vfs(struct pci_dev *dev, u16 num_vfs)
261 for (i = 0; i < num_vfs; i++) {
262 rc = pci_iov_add_virtfn(dev, i);
269 pci_iov_remove_virtfn(dev, i);
274 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
280 struct resource *res;
281 struct pci_dev *pdev;
282 struct pci_sriov *iov = dev->sriov;
292 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
293 if (initial > iov->total_VFs ||
294 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
297 if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
298 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
302 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
303 bars |= (1 << (i + PCI_IOV_RESOURCES));
304 res = &dev->resource[i + PCI_IOV_RESOURCES];
308 if (nres != iov->nres) {
309 pci_err(dev, "not enough MMIO resources for SR-IOV\n");
313 bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
314 if (bus > dev->bus->busn_res.end) {
315 pci_err(dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
316 nr_virtfn, bus, &dev->bus->busn_res);
320 if (pci_enable_resources(dev, bars)) {
321 pci_err(dev, "SR-IOV: IOV BARS not allocated\n");
325 if (iov->link != dev->devfn) {
326 pdev = pci_get_slot(dev->bus, iov->link);
330 if (!pdev->is_physfn) {
335 rc = sysfs_create_link(&dev->dev.kobj,
336 &pdev->dev.kobj, "dep_link");
342 iov->initial_VFs = initial;
343 if (nr_virtfn < initial)
346 rc = pcibios_sriov_enable(dev, initial);
348 pci_err(dev, "failure %d from pcibios_sriov_enable()\n", rc);
352 pci_iov_set_numvfs(dev, nr_virtfn);
353 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
354 pci_cfg_access_lock(dev);
355 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
357 pci_cfg_access_unlock(dev);
359 rc = sriov_add_vfs(dev, initial);
363 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
364 iov->num_VFs = nr_virtfn;
369 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
370 pci_cfg_access_lock(dev);
371 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
373 pci_cfg_access_unlock(dev);
375 pcibios_sriov_disable(dev);
377 if (iov->link != dev->devfn)
378 sysfs_remove_link(&dev->dev.kobj, "dep_link");
380 pci_iov_set_numvfs(dev, 0);
384 static void sriov_del_vfs(struct pci_dev *dev)
386 struct pci_sriov *iov = dev->sriov;
392 for (i = 0; i < iov->num_VFs; i++)
393 pci_iov_remove_virtfn(dev, i);
396 static void sriov_disable(struct pci_dev *dev)
398 struct pci_sriov *iov = dev->sriov;
404 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
405 pci_cfg_access_lock(dev);
406 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
408 pci_cfg_access_unlock(dev);
410 pcibios_sriov_disable(dev);
412 if (iov->link != dev->devfn)
413 sysfs_remove_link(&dev->dev.kobj, "dep_link");
416 pci_iov_set_numvfs(dev, 0);
419 static int sriov_init(struct pci_dev *dev, int pos)
426 struct pci_sriov *iov;
427 struct resource *res;
428 struct pci_dev *pdev;
430 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
431 if (ctrl & PCI_SRIOV_CTRL_VFE) {
432 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
437 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
442 if (pci_ari_enabled(dev->bus))
443 ctrl |= PCI_SRIOV_CTRL_ARI;
446 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
448 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
452 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
453 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
454 pgsz &= ~((1 << i) - 1);
459 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
461 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
466 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
467 res = &dev->resource[i + PCI_IOV_RESOURCES];
469 * If it is already FIXED, don't change it, something
470 * (perhaps EA or header fixups) wants it this way.
472 if (res->flags & IORESOURCE_PCI_FIXED)
473 bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
475 bar64 = __pci_read_base(dev, pci_bar_unknown, res,
476 pos + PCI_SRIOV_BAR + i * 4);
479 if (resource_size(res) & (PAGE_SIZE - 1)) {
483 iov->barsz[i] = resource_size(res);
484 res->end = res->start + resource_size(res) * total - 1;
485 pci_info(dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
494 iov->total_VFs = total;
495 iov->driver_max_VFs = total;
496 pci_read_config_word(dev, pos + PCI_SRIOV_VF_DID, &iov->vf_device);
499 iov->drivers_autoprobe = true;
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);
512 rc = compute_max_vf_buses(dev);
522 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
523 res = &dev->resource[i + PCI_IOV_RESOURCES];
531 static void sriov_release(struct pci_dev *dev)
533 BUG_ON(dev->sriov->num_VFs);
535 if (dev != dev->sriov->dev)
536 pci_dev_put(dev->sriov->dev);
542 static void sriov_restore_state(struct pci_dev *dev)
546 struct pci_sriov *iov = dev->sriov;
548 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
549 if (ctrl & PCI_SRIOV_CTRL_VFE)
553 * Restore PCI_SRIOV_CTRL_ARI before pci_iov_set_numvfs() because
554 * it reads offset & stride, which depend on PCI_SRIOV_CTRL_ARI.
556 ctrl &= ~PCI_SRIOV_CTRL_ARI;
557 ctrl |= iov->ctrl & PCI_SRIOV_CTRL_ARI;
558 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, ctrl);
560 for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
561 pci_update_resource(dev, i);
563 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
564 pci_iov_set_numvfs(dev, iov->num_VFs);
565 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
566 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
571 * pci_iov_init - initialize the IOV capability
572 * @dev: the PCI device
574 * Returns 0 on success, or negative on failure.
576 int pci_iov_init(struct pci_dev *dev)
580 if (!pci_is_pcie(dev))
583 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
585 return sriov_init(dev, pos);
591 * pci_iov_release - release resources used by the IOV capability
592 * @dev: the PCI device
594 void pci_iov_release(struct pci_dev *dev)
601 * pci_iov_remove - clean up SR-IOV state after PF driver is detached
602 * @dev: the PCI device
604 void pci_iov_remove(struct pci_dev *dev)
606 struct pci_sriov *iov = dev->sriov;
611 iov->driver_max_VFs = iov->total_VFs;
613 pci_warn(dev, "driver left SR-IOV enabled after remove\n");
617 * pci_iov_update_resource - update a VF BAR
618 * @dev: the PCI device
619 * @resno: the resource number
621 * Update a VF BAR in the SR-IOV capability of a PF.
623 void pci_iov_update_resource(struct pci_dev *dev, int resno)
625 struct pci_sriov *iov = dev->is_physfn ? dev->sriov : NULL;
626 struct resource *res = dev->resource + resno;
627 int vf_bar = resno - PCI_IOV_RESOURCES;
628 struct pci_bus_region region;
634 * The generic pci_restore_bars() path calls this for all devices,
635 * including VFs and non-SR-IOV devices. If this is not a PF, we
636 * have nothing to do.
641 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &cmd);
642 if ((cmd & PCI_SRIOV_CTRL_VFE) && (cmd & PCI_SRIOV_CTRL_MSE)) {
643 dev_WARN(&dev->dev, "can't update enabled VF BAR%d %pR\n",
649 * Ignore unimplemented BARs, unused resource slots for 64-bit
650 * BARs, and non-movable resources, e.g., those described via
651 * Enhanced Allocation.
656 if (res->flags & IORESOURCE_UNSET)
659 if (res->flags & IORESOURCE_PCI_FIXED)
662 pcibios_resource_to_bus(dev->bus, ®ion, res);
664 new |= res->flags & ~PCI_BASE_ADDRESS_MEM_MASK;
666 reg = iov->pos + PCI_SRIOV_BAR + 4 * vf_bar;
667 pci_write_config_dword(dev, reg, new);
668 if (res->flags & IORESOURCE_MEM_64) {
669 new = region.start >> 16 >> 16;
670 pci_write_config_dword(dev, reg + 4, new);
674 resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev,
677 return pci_iov_resource_size(dev, resno);
681 * pci_sriov_resource_alignment - get resource alignment for VF BAR
682 * @dev: the PCI device
683 * @resno: the resource number
685 * Returns the alignment of the VF BAR found in the SR-IOV capability.
686 * This is not the same as the resource size which is defined as
687 * the VF BAR size multiplied by the number of VFs. The alignment
688 * is just the VF BAR size.
690 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
692 return pcibios_iov_resource_alignment(dev, resno);
696 * pci_restore_iov_state - restore the state of the IOV capability
697 * @dev: the PCI device
699 void pci_restore_iov_state(struct pci_dev *dev)
702 sriov_restore_state(dev);
706 * pci_vf_drivers_autoprobe - set PF property drivers_autoprobe for VFs
707 * @dev: the PCI device
708 * @auto_probe: set VF drivers auto probe flag
710 void pci_vf_drivers_autoprobe(struct pci_dev *dev, bool auto_probe)
713 dev->sriov->drivers_autoprobe = auto_probe;
717 * pci_iov_bus_range - find bus range used by Virtual Function
720 * Returns max number of buses (exclude current one) used by Virtual
723 int pci_iov_bus_range(struct pci_bus *bus)
728 list_for_each_entry(dev, &bus->devices, bus_list) {
731 if (dev->sriov->max_VF_buses > max)
732 max = dev->sriov->max_VF_buses;
735 return max ? max - bus->number : 0;
739 * pci_enable_sriov - enable the SR-IOV capability
740 * @dev: the PCI device
741 * @nr_virtfn: number of virtual functions to enable
743 * Returns 0 on success, or negative on failure.
745 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
752 return sriov_enable(dev, nr_virtfn);
754 EXPORT_SYMBOL_GPL(pci_enable_sriov);
757 * pci_disable_sriov - disable the SR-IOV capability
758 * @dev: the PCI device
760 void pci_disable_sriov(struct pci_dev *dev)
769 EXPORT_SYMBOL_GPL(pci_disable_sriov);
772 * pci_num_vf - return number of VFs associated with a PF device_release_driver
773 * @dev: the PCI device
775 * Returns number of VFs, or 0 if SR-IOV is not enabled.
777 int pci_num_vf(struct pci_dev *dev)
782 return dev->sriov->num_VFs;
784 EXPORT_SYMBOL_GPL(pci_num_vf);
787 * pci_vfs_assigned - returns number of VFs are assigned to a guest
788 * @dev: the PCI device
790 * Returns number of VFs belonging to this device that are assigned to a guest.
791 * If device is not a physical function returns 0.
793 int pci_vfs_assigned(struct pci_dev *dev)
795 struct pci_dev *vfdev;
796 unsigned int vfs_assigned = 0;
797 unsigned short dev_id;
799 /* only search if we are a PF */
804 * determine the device ID for the VFs, the vendor ID will be the
805 * same as the PF so there is no need to check for that one
807 dev_id = dev->sriov->vf_device;
809 /* loop through all the VFs to see if we own any that are assigned */
810 vfdev = pci_get_device(dev->vendor, dev_id, NULL);
813 * It is considered assigned if it is a virtual function with
814 * our dev as the physical function and the assigned bit is set
816 if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
817 pci_is_dev_assigned(vfdev))
820 vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
825 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
828 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
829 * @dev: the PCI PF device
830 * @numvfs: number that should be used for TotalVFs supported
832 * Should be called from PF driver's probe routine with
833 * device's mutex held.
835 * Returns 0 if PF is an SRIOV-capable device and
836 * value of numvfs valid. If not a PF return -ENOSYS;
837 * if numvfs is invalid return -EINVAL;
838 * if VFs already enabled, return -EBUSY.
840 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
845 if (numvfs > dev->sriov->total_VFs)
848 /* Shouldn't change if VFs already enabled */
849 if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
852 dev->sriov->driver_max_VFs = numvfs;
855 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
858 * pci_sriov_get_totalvfs -- get total VFs supported on this device
859 * @dev: the PCI PF device
861 * For a PCIe device with SRIOV support, return the PCIe
862 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
863 * if the driver reduced it. Otherwise 0.
865 int pci_sriov_get_totalvfs(struct pci_dev *dev)
870 return dev->sriov->driver_max_VFs;
872 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);
875 * pci_sriov_configure_simple - helper to configure SR-IOV
876 * @dev: the PCI device
877 * @nr_virtfn: number of virtual functions to enable, 0 to disable
879 * Enable or disable SR-IOV for devices that don't require any PF setup
880 * before enabling SR-IOV. Return value is negative on error, or number of
881 * VFs allocated on success.
883 int pci_sriov_configure_simple(struct pci_dev *dev, int nr_virtfn)
892 if (pci_vfs_assigned(dev)) {
893 pci_warn(dev, "Cannot modify SR-IOV while VFs are assigned\n");
897 if (nr_virtfn == 0) {
902 rc = sriov_enable(dev, nr_virtfn);
908 EXPORT_SYMBOL_GPL(pci_sriov_configure_simple);