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 int pci_iov_virtfn_bus(struct pci_dev *dev, int vf_id)
26 return dev->bus->number + ((dev->devfn + dev->sriov->offset +
27 dev->sriov->stride * vf_id) >> 8);
30 int pci_iov_virtfn_devfn(struct pci_dev *dev, int vf_id)
34 return (dev->devfn + dev->sriov->offset +
35 dev->sriov->stride * vf_id) & 0xff;
39 * Per SR-IOV spec sec 3.3.10 and 3.3.11, First VF Offset and VF Stride may
40 * change when NumVFs changes.
42 * Update iov->offset and iov->stride when NumVFs is written.
44 static inline void pci_iov_set_numvfs(struct pci_dev *dev, int nr_virtfn)
46 struct pci_sriov *iov = dev->sriov;
48 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
49 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &iov->offset);
50 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &iov->stride);
54 * The PF consumes one bus number. NumVFs, First VF Offset, and VF Stride
55 * determine how many additional bus numbers will be consumed by VFs.
57 * Iterate over all valid NumVFs, validate offset and stride, and calculate
58 * the maximum number of bus numbers that could ever be required.
60 static int compute_max_vf_buses(struct pci_dev *dev)
62 struct pci_sriov *iov = dev->sriov;
63 int nr_virtfn, busnr, rc = 0;
65 for (nr_virtfn = iov->total_VFs; nr_virtfn; nr_virtfn--) {
66 pci_iov_set_numvfs(dev, nr_virtfn);
67 if (!iov->offset || (nr_virtfn > 1 && !iov->stride)) {
72 busnr = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
73 if (busnr > iov->max_VF_buses)
74 iov->max_VF_buses = busnr;
78 pci_iov_set_numvfs(dev, 0);
82 static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
84 struct pci_bus *child;
86 if (bus->number == busnr)
89 child = pci_find_bus(pci_domain_nr(bus), busnr);
93 child = pci_add_new_bus(bus, NULL, busnr);
97 pci_bus_insert_busn_res(child, busnr, busnr);
102 static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
104 if (physbus != virtbus && list_empty(&virtbus->devices))
105 pci_remove_bus(virtbus);
108 resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno)
113 return dev->sriov->barsz[resno - PCI_IOV_RESOURCES];
116 static int virtfn_add(struct pci_dev *dev, int id, int reset)
121 char buf[VIRTFN_ID_LEN];
122 struct pci_dev *virtfn;
123 struct resource *res;
124 struct pci_sriov *iov = dev->sriov;
127 mutex_lock(&iov->dev->sriov->lock);
128 bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id));
132 virtfn = pci_alloc_dev(bus);
136 virtfn->devfn = pci_iov_virtfn_devfn(dev, id);
137 virtfn->vendor = dev->vendor;
138 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
139 pci_setup_device(virtfn);
140 virtfn->dev.parent = dev->dev.parent;
141 virtfn->physfn = pci_dev_get(dev);
142 virtfn->is_virtfn = 1;
143 virtfn->multifunction = 0;
145 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
146 res = &dev->resource[i + PCI_IOV_RESOURCES];
149 virtfn->resource[i].name = pci_name(virtfn);
150 virtfn->resource[i].flags = res->flags;
151 size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
152 virtfn->resource[i].start = res->start + size * id;
153 virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
154 rc = request_resource(res, &virtfn->resource[i]);
159 __pci_reset_function(virtfn);
161 pci_device_add(virtfn, virtfn->bus);
162 mutex_unlock(&iov->dev->sriov->lock);
164 pci_bus_add_device(virtfn);
165 sprintf(buf, "virtfn%u", id);
166 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
169 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
173 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
178 sysfs_remove_link(&dev->dev.kobj, buf);
181 mutex_lock(&iov->dev->sriov->lock);
182 pci_stop_and_remove_bus_device(virtfn);
184 virtfn_remove_bus(dev->bus, bus);
186 mutex_unlock(&iov->dev->sriov->lock);
191 static void virtfn_remove(struct pci_dev *dev, int id, int reset)
193 char buf[VIRTFN_ID_LEN];
194 struct pci_dev *virtfn;
195 struct pci_sriov *iov = dev->sriov;
197 virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
198 pci_iov_virtfn_bus(dev, id),
199 pci_iov_virtfn_devfn(dev, id));
204 device_release_driver(&virtfn->dev);
205 __pci_reset_function(virtfn);
208 sprintf(buf, "virtfn%u", id);
209 sysfs_remove_link(&dev->dev.kobj, buf);
211 * pci_stop_dev() could have been called for this virtfn already,
212 * so the directory for the virtfn may have been removed before.
213 * Double check to avoid spurious sysfs warnings.
215 if (virtfn->dev.kobj.sd)
216 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
218 mutex_lock(&iov->dev->sriov->lock);
219 pci_stop_and_remove_bus_device(virtfn);
220 virtfn_remove_bus(dev->bus, virtfn->bus);
221 mutex_unlock(&iov->dev->sriov->lock);
223 /* balance pci_get_domain_bus_and_slot() */
228 int __weak pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
233 int __weak pcibios_sriov_disable(struct pci_dev *pdev)
238 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
244 struct resource *res;
245 struct pci_dev *pdev;
246 struct pci_sriov *iov = dev->sriov;
256 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
257 if (initial > iov->total_VFs ||
258 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
261 if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
262 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
266 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
267 bars |= (1 << (i + PCI_IOV_RESOURCES));
268 res = &dev->resource[i + PCI_IOV_RESOURCES];
272 if (nres != iov->nres) {
273 dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
277 bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
278 if (bus > dev->bus->busn_res.end) {
279 dev_err(&dev->dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
280 nr_virtfn, bus, &dev->bus->busn_res);
284 if (pci_enable_resources(dev, bars)) {
285 dev_err(&dev->dev, "SR-IOV: IOV BARS not allocated\n");
289 if (iov->link != dev->devfn) {
290 pdev = pci_get_slot(dev->bus, iov->link);
294 if (!pdev->is_physfn) {
299 rc = sysfs_create_link(&dev->dev.kobj,
300 &pdev->dev.kobj, "dep_link");
306 pci_iov_set_numvfs(dev, nr_virtfn);
307 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
308 pci_cfg_access_lock(dev);
309 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
311 pci_cfg_access_unlock(dev);
313 iov->initial_VFs = initial;
314 if (nr_virtfn < initial)
317 rc = pcibios_sriov_enable(dev, initial);
319 dev_err(&dev->dev, "failure %d from pcibios_sriov_enable()\n", rc);
323 for (i = 0; i < initial; i++) {
324 rc = virtfn_add(dev, i, 0);
329 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
330 iov->num_VFs = nr_virtfn;
336 virtfn_remove(dev, i, 0);
338 pcibios_sriov_disable(dev);
340 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
341 pci_cfg_access_lock(dev);
342 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
344 pci_cfg_access_unlock(dev);
346 if (iov->link != dev->devfn)
347 sysfs_remove_link(&dev->dev.kobj, "dep_link");
349 pci_iov_set_numvfs(dev, 0);
353 static void sriov_disable(struct pci_dev *dev)
356 struct pci_sriov *iov = dev->sriov;
361 for (i = 0; i < iov->num_VFs; i++)
362 virtfn_remove(dev, i, 0);
364 pcibios_sriov_disable(dev);
366 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
367 pci_cfg_access_lock(dev);
368 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
370 pci_cfg_access_unlock(dev);
372 if (iov->link != dev->devfn)
373 sysfs_remove_link(&dev->dev.kobj, "dep_link");
376 pci_iov_set_numvfs(dev, 0);
379 static int sriov_init(struct pci_dev *dev, int pos)
386 struct pci_sriov *iov;
387 struct resource *res;
388 struct pci_dev *pdev;
390 if (pci_pcie_type(dev) != PCI_EXP_TYPE_RC_END &&
391 pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT)
394 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
395 if (ctrl & PCI_SRIOV_CTRL_VFE) {
396 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
401 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
406 if (pci_ari_enabled(dev->bus))
407 ctrl |= PCI_SRIOV_CTRL_ARI;
410 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
412 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
416 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
417 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
418 pgsz &= ~((1 << i) - 1);
423 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
425 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
430 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
431 res = &dev->resource[i + PCI_IOV_RESOURCES];
433 * If it is already FIXED, don't change it, something
434 * (perhaps EA or header fixups) wants it this way.
436 if (res->flags & IORESOURCE_PCI_FIXED)
437 bar64 = (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
439 bar64 = __pci_read_base(dev, pci_bar_unknown, res,
440 pos + PCI_SRIOV_BAR + i * 4);
443 if (resource_size(res) & (PAGE_SIZE - 1)) {
447 iov->barsz[i] = resource_size(res);
448 res->end = res->start + resource_size(res) * total - 1;
449 dev_info(&dev->dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
458 iov->total_VFs = total;
461 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
462 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
463 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
464 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
467 iov->dev = pci_dev_get(pdev);
471 mutex_init(&iov->lock);
475 rc = compute_max_vf_buses(dev);
485 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
486 res = &dev->resource[i + PCI_IOV_RESOURCES];
494 static void sriov_release(struct pci_dev *dev)
496 BUG_ON(dev->sriov->num_VFs);
498 if (dev != dev->sriov->dev)
499 pci_dev_put(dev->sriov->dev);
501 mutex_destroy(&dev->sriov->lock);
507 static void sriov_restore_state(struct pci_dev *dev)
511 struct pci_sriov *iov = dev->sriov;
513 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
514 if (ctrl & PCI_SRIOV_CTRL_VFE)
517 for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
518 pci_update_resource(dev, i);
520 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
521 pci_iov_set_numvfs(dev, iov->num_VFs);
522 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
523 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
528 * pci_iov_init - initialize the IOV capability
529 * @dev: the PCI device
531 * Returns 0 on success, or negative on failure.
533 int pci_iov_init(struct pci_dev *dev)
537 if (!pci_is_pcie(dev))
540 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
542 return sriov_init(dev, pos);
548 * pci_iov_release - release resources used by the IOV capability
549 * @dev: the PCI device
551 void pci_iov_release(struct pci_dev *dev)
558 * pci_iov_resource_bar - get position of the SR-IOV BAR
559 * @dev: the PCI device
560 * @resno: the resource number
562 * Returns position of the BAR encapsulated in the SR-IOV capability.
564 int pci_iov_resource_bar(struct pci_dev *dev, int resno)
566 if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
569 BUG_ON(!dev->is_physfn);
571 return dev->sriov->pos + PCI_SRIOV_BAR +
572 4 * (resno - PCI_IOV_RESOURCES);
575 resource_size_t __weak pcibios_iov_resource_alignment(struct pci_dev *dev,
578 return pci_iov_resource_size(dev, resno);
582 * pci_sriov_resource_alignment - get resource alignment for VF BAR
583 * @dev: the PCI device
584 * @resno: the resource number
586 * Returns the alignment of the VF BAR found in the SR-IOV capability.
587 * This is not the same as the resource size which is defined as
588 * the VF BAR size multiplied by the number of VFs. The alignment
589 * is just the VF BAR size.
591 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
593 return pcibios_iov_resource_alignment(dev, resno);
597 * pci_restore_iov_state - restore the state of the IOV capability
598 * @dev: the PCI device
600 void pci_restore_iov_state(struct pci_dev *dev)
603 sriov_restore_state(dev);
607 * pci_iov_bus_range - find bus range used by Virtual Function
610 * Returns max number of buses (exclude current one) used by Virtual
613 int pci_iov_bus_range(struct pci_bus *bus)
618 list_for_each_entry(dev, &bus->devices, bus_list) {
621 if (dev->sriov->max_VF_buses > max)
622 max = dev->sriov->max_VF_buses;
625 return max ? max - bus->number : 0;
629 * pci_enable_sriov - enable the SR-IOV capability
630 * @dev: the PCI device
631 * @nr_virtfn: number of virtual functions to enable
633 * Returns 0 on success, or negative on failure.
635 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
642 return sriov_enable(dev, nr_virtfn);
644 EXPORT_SYMBOL_GPL(pci_enable_sriov);
647 * pci_disable_sriov - disable the SR-IOV capability
648 * @dev: the PCI device
650 void pci_disable_sriov(struct pci_dev *dev)
659 EXPORT_SYMBOL_GPL(pci_disable_sriov);
662 * pci_num_vf - return number of VFs associated with a PF device_release_driver
663 * @dev: the PCI device
665 * Returns number of VFs, or 0 if SR-IOV is not enabled.
667 int pci_num_vf(struct pci_dev *dev)
672 return dev->sriov->num_VFs;
674 EXPORT_SYMBOL_GPL(pci_num_vf);
677 * pci_vfs_assigned - returns number of VFs are assigned to a guest
678 * @dev: the PCI device
680 * Returns number of VFs belonging to this device that are assigned to a guest.
681 * If device is not a physical function returns 0.
683 int pci_vfs_assigned(struct pci_dev *dev)
685 struct pci_dev *vfdev;
686 unsigned int vfs_assigned = 0;
687 unsigned short dev_id;
689 /* only search if we are a PF */
694 * determine the device ID for the VFs, the vendor ID will be the
695 * same as the PF so there is no need to check for that one
697 pci_read_config_word(dev, dev->sriov->pos + PCI_SRIOV_VF_DID, &dev_id);
699 /* loop through all the VFs to see if we own any that are assigned */
700 vfdev = pci_get_device(dev->vendor, dev_id, NULL);
703 * It is considered assigned if it is a virtual function with
704 * our dev as the physical function and the assigned bit is set
706 if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
707 pci_is_dev_assigned(vfdev))
710 vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
715 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
718 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
719 * @dev: the PCI PF device
720 * @numvfs: number that should be used for TotalVFs supported
722 * Should be called from PF driver's probe routine with
723 * device's mutex held.
725 * Returns 0 if PF is an SRIOV-capable device and
726 * value of numvfs valid. If not a PF return -ENOSYS;
727 * if numvfs is invalid return -EINVAL;
728 * if VFs already enabled, return -EBUSY.
730 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
734 if (numvfs > dev->sriov->total_VFs)
737 /* Shouldn't change if VFs already enabled */
738 if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
741 dev->sriov->driver_max_VFs = numvfs;
745 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
748 * pci_sriov_get_totalvfs -- get total VFs supported on this device
749 * @dev: the PCI PF device
751 * For a PCIe device with SRIOV support, return the PCIe
752 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
753 * if the driver reduced it. Otherwise 0.
755 int pci_sriov_get_totalvfs(struct pci_dev *dev)
760 if (dev->sriov->driver_max_VFs)
761 return dev->sriov->driver_max_VFs;
763 return dev->sriov->total_VFs;
765 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);