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);
54 static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
56 if (physbus != virtbus && list_empty(&virtbus->devices))
57 pci_remove_bus(virtbus);
60 resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno)
65 return dev->sriov->barsz[resno - PCI_IOV_RESOURCES];
68 static int virtfn_add(struct pci_dev *dev, int id, int reset)
73 char buf[VIRTFN_ID_LEN];
74 struct pci_dev *virtfn;
76 struct pci_sriov *iov = dev->sriov;
79 mutex_lock(&iov->dev->sriov->lock);
80 bus = virtfn_add_bus(dev->bus, virtfn_bus(dev, id));
84 virtfn = pci_alloc_dev(bus);
88 virtfn->devfn = virtfn_devfn(dev, id);
89 virtfn->vendor = dev->vendor;
90 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
91 pci_setup_device(virtfn);
92 virtfn->dev.parent = dev->dev.parent;
93 virtfn->physfn = pci_dev_get(dev);
94 virtfn->is_virtfn = 1;
95 virtfn->multifunction = 0;
97 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
98 res = &dev->resource[i + PCI_IOV_RESOURCES];
101 virtfn->resource[i].name = pci_name(virtfn);
102 virtfn->resource[i].flags = res->flags;
103 size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
104 virtfn->resource[i].start = res->start + size * id;
105 virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
106 rc = request_resource(res, &virtfn->resource[i]);
111 __pci_reset_function(virtfn);
113 pci_device_add(virtfn, virtfn->bus);
114 mutex_unlock(&iov->dev->sriov->lock);
116 pci_bus_add_device(virtfn);
117 sprintf(buf, "virtfn%u", id);
118 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
121 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
125 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
130 sysfs_remove_link(&dev->dev.kobj, buf);
133 mutex_lock(&iov->dev->sriov->lock);
134 pci_stop_and_remove_bus_device(virtfn);
136 virtfn_remove_bus(dev->bus, bus);
138 mutex_unlock(&iov->dev->sriov->lock);
143 static void virtfn_remove(struct pci_dev *dev, int id, int reset)
145 char buf[VIRTFN_ID_LEN];
146 struct pci_dev *virtfn;
147 struct pci_sriov *iov = dev->sriov;
149 virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
151 virtfn_devfn(dev, id));
156 device_release_driver(&virtfn->dev);
157 __pci_reset_function(virtfn);
160 sprintf(buf, "virtfn%u", id);
161 sysfs_remove_link(&dev->dev.kobj, buf);
163 * pci_stop_dev() could have been called for this virtfn already,
164 * so the directory for the virtfn may have been removed before.
165 * Double check to avoid spurious sysfs warnings.
167 if (virtfn->dev.kobj.sd)
168 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
170 mutex_lock(&iov->dev->sriov->lock);
171 pci_stop_and_remove_bus_device(virtfn);
172 virtfn_remove_bus(dev->bus, virtfn->bus);
173 mutex_unlock(&iov->dev->sriov->lock);
175 /* balance pci_get_domain_bus_and_slot() */
180 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
185 u16 offset, stride, initial;
186 struct resource *res;
187 struct pci_dev *pdev;
188 struct pci_sriov *iov = dev->sriov;
198 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
199 if (initial > iov->total_VFs ||
200 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
203 if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
204 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
207 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
208 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
209 if (!offset || (nr_virtfn > 1 && !stride))
213 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
214 bars |= (1 << (i + PCI_IOV_RESOURCES));
215 res = &dev->resource[i + PCI_IOV_RESOURCES];
219 if (nres != iov->nres) {
220 dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
224 iov->offset = offset;
225 iov->stride = stride;
227 bus = virtfn_bus(dev, nr_virtfn - 1);
228 if (bus > dev->bus->busn_res.end) {
229 dev_err(&dev->dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
230 nr_virtfn, bus, &dev->bus->busn_res);
234 if (pci_enable_resources(dev, bars)) {
235 dev_err(&dev->dev, "SR-IOV: IOV BARS not allocated\n");
239 if (iov->link != dev->devfn) {
240 pdev = pci_get_slot(dev->bus, iov->link);
244 if (!pdev->is_physfn) {
249 rc = sysfs_create_link(&dev->dev.kobj,
250 &pdev->dev.kobj, "dep_link");
256 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
257 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
258 pci_cfg_access_lock(dev);
259 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
261 pci_cfg_access_unlock(dev);
263 iov->initial_VFs = initial;
264 if (nr_virtfn < initial)
267 for (i = 0; i < initial; i++) {
268 rc = virtfn_add(dev, i, 0);
273 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
274 iov->num_VFs = nr_virtfn;
279 for (j = 0; j < i; j++)
280 virtfn_remove(dev, j, 0);
282 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
283 pci_cfg_access_lock(dev);
284 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
285 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, 0);
287 pci_cfg_access_unlock(dev);
289 if (iov->link != dev->devfn)
290 sysfs_remove_link(&dev->dev.kobj, "dep_link");
295 static void sriov_disable(struct pci_dev *dev)
298 struct pci_sriov *iov = dev->sriov;
303 for (i = 0; i < iov->num_VFs; i++)
304 virtfn_remove(dev, i, 0);
306 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
307 pci_cfg_access_lock(dev);
308 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
310 pci_cfg_access_unlock(dev);
312 if (iov->link != dev->devfn)
313 sysfs_remove_link(&dev->dev.kobj, "dep_link");
316 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, 0);
319 static int sriov_init(struct pci_dev *dev, int pos)
325 u16 ctrl, total, offset, stride;
326 struct pci_sriov *iov;
327 struct resource *res;
328 struct pci_dev *pdev;
330 if (pci_pcie_type(dev) != PCI_EXP_TYPE_RC_END &&
331 pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT)
334 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
335 if (ctrl & PCI_SRIOV_CTRL_VFE) {
336 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
340 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
345 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
350 if (pci_ari_enabled(dev->bus))
351 ctrl |= PCI_SRIOV_CTRL_ARI;
354 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
355 pci_write_config_word(dev, pos + PCI_SRIOV_NUM_VF, 0);
356 pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
357 pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
358 if (!offset || (total > 1 && !stride))
361 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
362 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
363 pgsz &= ~((1 << i) - 1);
368 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
370 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
375 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
376 res = &dev->resource[i + PCI_IOV_RESOURCES];
377 bar64 = __pci_read_base(dev, pci_bar_unknown, res,
378 pos + PCI_SRIOV_BAR + i * 4);
381 if (resource_size(res) & (PAGE_SIZE - 1)) {
385 iov->barsz[i] = resource_size(res);
386 res->end = res->start + resource_size(res) * total - 1;
387 dev_info(&dev->dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
396 iov->total_VFs = total;
397 iov->offset = offset;
398 iov->stride = stride;
401 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
402 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
403 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
404 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
407 iov->dev = pci_dev_get(pdev);
411 mutex_init(&iov->lock);
419 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
420 res = &dev->resource[i + PCI_IOV_RESOURCES];
428 static void sriov_release(struct pci_dev *dev)
430 BUG_ON(dev->sriov->num_VFs);
432 if (dev != dev->sriov->dev)
433 pci_dev_put(dev->sriov->dev);
435 mutex_destroy(&dev->sriov->lock);
441 static void sriov_restore_state(struct pci_dev *dev)
445 struct pci_sriov *iov = dev->sriov;
447 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
448 if (ctrl & PCI_SRIOV_CTRL_VFE)
451 for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
452 pci_update_resource(dev, i);
454 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
455 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->num_VFs);
456 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
457 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
462 * pci_iov_init - initialize the IOV capability
463 * @dev: the PCI device
465 * Returns 0 on success, or negative on failure.
467 int pci_iov_init(struct pci_dev *dev)
471 if (!pci_is_pcie(dev))
474 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
476 return sriov_init(dev, pos);
482 * pci_iov_release - release resources used by the IOV capability
483 * @dev: the PCI device
485 void pci_iov_release(struct pci_dev *dev)
492 * pci_iov_resource_bar - get position of the SR-IOV BAR
493 * @dev: the PCI device
494 * @resno: the resource number
496 * Returns position of the BAR encapsulated in the SR-IOV capability.
498 int pci_iov_resource_bar(struct pci_dev *dev, int resno)
500 if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
503 BUG_ON(!dev->is_physfn);
505 return dev->sriov->pos + PCI_SRIOV_BAR +
506 4 * (resno - PCI_IOV_RESOURCES);
510 * pci_sriov_resource_alignment - get resource alignment for VF BAR
511 * @dev: the PCI device
512 * @resno: the resource number
514 * Returns the alignment of the VF BAR found in the SR-IOV capability.
515 * This is not the same as the resource size which is defined as
516 * the VF BAR size multiplied by the number of VFs. The alignment
517 * is just the VF BAR size.
519 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
521 return pci_iov_resource_size(dev, resno);
525 * pci_restore_iov_state - restore the state of the IOV capability
526 * @dev: the PCI device
528 void pci_restore_iov_state(struct pci_dev *dev)
531 sriov_restore_state(dev);
535 * pci_iov_bus_range - find bus range used by Virtual Function
538 * Returns max number of buses (exclude current one) used by Virtual
541 int pci_iov_bus_range(struct pci_bus *bus)
547 list_for_each_entry(dev, &bus->devices, bus_list) {
550 busnr = virtfn_bus(dev, dev->sriov->total_VFs - 1);
555 return max ? max - bus->number : 0;
559 * pci_enable_sriov - enable the SR-IOV capability
560 * @dev: the PCI device
561 * @nr_virtfn: number of virtual functions to enable
563 * Returns 0 on success, or negative on failure.
565 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
572 return sriov_enable(dev, nr_virtfn);
574 EXPORT_SYMBOL_GPL(pci_enable_sriov);
577 * pci_disable_sriov - disable the SR-IOV capability
578 * @dev: the PCI device
580 void pci_disable_sriov(struct pci_dev *dev)
589 EXPORT_SYMBOL_GPL(pci_disable_sriov);
592 * pci_num_vf - return number of VFs associated with a PF device_release_driver
593 * @dev: the PCI device
595 * Returns number of VFs, or 0 if SR-IOV is not enabled.
597 int pci_num_vf(struct pci_dev *dev)
602 return dev->sriov->num_VFs;
604 EXPORT_SYMBOL_GPL(pci_num_vf);
607 * pci_vfs_assigned - returns number of VFs are assigned to a guest
608 * @dev: the PCI device
610 * Returns number of VFs belonging to this device that are assigned to a guest.
611 * If device is not a physical function returns 0.
613 int pci_vfs_assigned(struct pci_dev *dev)
615 struct pci_dev *vfdev;
616 unsigned int vfs_assigned = 0;
617 unsigned short dev_id;
619 /* only search if we are a PF */
624 * determine the device ID for the VFs, the vendor ID will be the
625 * same as the PF so there is no need to check for that one
627 pci_read_config_word(dev, dev->sriov->pos + PCI_SRIOV_VF_DID, &dev_id);
629 /* loop through all the VFs to see if we own any that are assigned */
630 vfdev = pci_get_device(dev->vendor, dev_id, NULL);
633 * It is considered assigned if it is a virtual function with
634 * our dev as the physical function and the assigned bit is set
636 if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
637 pci_is_dev_assigned(vfdev))
640 vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
645 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
648 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
649 * @dev: the PCI PF device
650 * @numvfs: number that should be used for TotalVFs supported
652 * Should be called from PF driver's probe routine with
653 * device's mutex held.
655 * Returns 0 if PF is an SRIOV-capable device and
656 * value of numvfs valid. If not a PF return -ENOSYS;
657 * if numvfs is invalid return -EINVAL;
658 * if VFs already enabled, return -EBUSY.
660 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
664 if (numvfs > dev->sriov->total_VFs)
667 /* Shouldn't change if VFs already enabled */
668 if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
671 dev->sriov->driver_max_VFs = numvfs;
675 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
678 * pci_sriov_get_totalvfs -- get total VFs supported on this device
679 * @dev: the PCI PF device
681 * For a PCIe device with SRIOV support, return the PCIe
682 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
683 * if the driver reduced it. Otherwise 0.
685 int pci_sriov_get_totalvfs(struct pci_dev *dev)
690 if (dev->sriov->driver_max_VFs)
691 return dev->sriov->driver_max_VFs;
693 return dev->sriov->total_VFs;
695 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);