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 and calculate the maximum number of bus
58 * numbers that could ever be required.
60 static inline u8 virtfn_max_buses(struct pci_dev *dev)
62 struct pci_sriov *iov = dev->sriov;
67 for (nr_virtfn = 1; nr_virtfn <= iov->total_VFs; nr_virtfn++) {
68 pci_iov_set_numvfs(dev, nr_virtfn);
69 busnr = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
77 static struct pci_bus *virtfn_add_bus(struct pci_bus *bus, int busnr)
79 struct pci_bus *child;
81 if (bus->number == busnr)
84 child = pci_find_bus(pci_domain_nr(bus), busnr);
88 child = pci_add_new_bus(bus, NULL, busnr);
92 pci_bus_insert_busn_res(child, busnr, busnr);
97 static void virtfn_remove_bus(struct pci_bus *physbus, struct pci_bus *virtbus)
99 if (physbus != virtbus && list_empty(&virtbus->devices))
100 pci_remove_bus(virtbus);
103 resource_size_t pci_iov_resource_size(struct pci_dev *dev, int resno)
108 return dev->sriov->barsz[resno - PCI_IOV_RESOURCES];
111 static int virtfn_add(struct pci_dev *dev, int id, int reset)
116 char buf[VIRTFN_ID_LEN];
117 struct pci_dev *virtfn;
118 struct resource *res;
119 struct pci_sriov *iov = dev->sriov;
122 mutex_lock(&iov->dev->sriov->lock);
123 bus = virtfn_add_bus(dev->bus, pci_iov_virtfn_bus(dev, id));
127 virtfn = pci_alloc_dev(bus);
131 virtfn->devfn = pci_iov_virtfn_devfn(dev, id);
132 virtfn->vendor = dev->vendor;
133 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
134 pci_setup_device(virtfn);
135 virtfn->dev.parent = dev->dev.parent;
136 virtfn->physfn = pci_dev_get(dev);
137 virtfn->is_virtfn = 1;
138 virtfn->multifunction = 0;
140 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
141 res = &dev->resource[i + PCI_IOV_RESOURCES];
144 virtfn->resource[i].name = pci_name(virtfn);
145 virtfn->resource[i].flags = res->flags;
146 size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
147 virtfn->resource[i].start = res->start + size * id;
148 virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
149 rc = request_resource(res, &virtfn->resource[i]);
154 __pci_reset_function(virtfn);
156 pci_device_add(virtfn, virtfn->bus);
157 mutex_unlock(&iov->dev->sriov->lock);
159 pci_bus_add_device(virtfn);
160 sprintf(buf, "virtfn%u", id);
161 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
164 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
168 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
173 sysfs_remove_link(&dev->dev.kobj, buf);
176 mutex_lock(&iov->dev->sriov->lock);
177 pci_stop_and_remove_bus_device(virtfn);
179 virtfn_remove_bus(dev->bus, bus);
181 mutex_unlock(&iov->dev->sriov->lock);
186 static void virtfn_remove(struct pci_dev *dev, int id, int reset)
188 char buf[VIRTFN_ID_LEN];
189 struct pci_dev *virtfn;
190 struct pci_sriov *iov = dev->sriov;
192 virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
193 pci_iov_virtfn_bus(dev, id),
194 pci_iov_virtfn_devfn(dev, id));
199 device_release_driver(&virtfn->dev);
200 __pci_reset_function(virtfn);
203 sprintf(buf, "virtfn%u", id);
204 sysfs_remove_link(&dev->dev.kobj, buf);
206 * pci_stop_dev() could have been called for this virtfn already,
207 * so the directory for the virtfn may have been removed before.
208 * Double check to avoid spurious sysfs warnings.
210 if (virtfn->dev.kobj.sd)
211 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
213 mutex_lock(&iov->dev->sriov->lock);
214 pci_stop_and_remove_bus_device(virtfn);
215 virtfn_remove_bus(dev->bus, virtfn->bus);
216 mutex_unlock(&iov->dev->sriov->lock);
218 /* balance pci_get_domain_bus_and_slot() */
223 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
228 u16 offset, stride, initial;
229 struct resource *res;
230 struct pci_dev *pdev;
231 struct pci_sriov *iov = dev->sriov;
241 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
242 if (initial > iov->total_VFs ||
243 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
246 if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
247 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
250 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
251 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
252 if (!offset || (nr_virtfn > 1 && !stride))
256 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
257 bars |= (1 << (i + PCI_IOV_RESOURCES));
258 res = &dev->resource[i + PCI_IOV_RESOURCES];
262 if (nres != iov->nres) {
263 dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
267 iov->offset = offset;
268 iov->stride = stride;
270 bus = pci_iov_virtfn_bus(dev, nr_virtfn - 1);
271 if (bus > dev->bus->busn_res.end) {
272 dev_err(&dev->dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
273 nr_virtfn, bus, &dev->bus->busn_res);
277 if (pci_enable_resources(dev, bars)) {
278 dev_err(&dev->dev, "SR-IOV: IOV BARS not allocated\n");
282 if (iov->link != dev->devfn) {
283 pdev = pci_get_slot(dev->bus, iov->link);
287 if (!pdev->is_physfn) {
292 rc = sysfs_create_link(&dev->dev.kobj,
293 &pdev->dev.kobj, "dep_link");
299 pci_iov_set_numvfs(dev, nr_virtfn);
300 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
301 pci_cfg_access_lock(dev);
302 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
304 pci_cfg_access_unlock(dev);
306 iov->initial_VFs = initial;
307 if (nr_virtfn < initial)
310 for (i = 0; i < initial; i++) {
311 rc = virtfn_add(dev, i, 0);
316 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
317 iov->num_VFs = nr_virtfn;
322 for (j = 0; j < i; j++)
323 virtfn_remove(dev, j, 0);
325 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
326 pci_cfg_access_lock(dev);
327 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
328 pci_iov_set_numvfs(dev, 0);
330 pci_cfg_access_unlock(dev);
332 if (iov->link != dev->devfn)
333 sysfs_remove_link(&dev->dev.kobj, "dep_link");
338 static void sriov_disable(struct pci_dev *dev)
341 struct pci_sriov *iov = dev->sriov;
346 for (i = 0; i < iov->num_VFs; i++)
347 virtfn_remove(dev, i, 0);
349 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
350 pci_cfg_access_lock(dev);
351 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
353 pci_cfg_access_unlock(dev);
355 if (iov->link != dev->devfn)
356 sysfs_remove_link(&dev->dev.kobj, "dep_link");
359 pci_iov_set_numvfs(dev, 0);
362 static int sriov_init(struct pci_dev *dev, int pos)
368 u16 ctrl, total, offset, stride;
369 struct pci_sriov *iov;
370 struct resource *res;
371 struct pci_dev *pdev;
373 if (pci_pcie_type(dev) != PCI_EXP_TYPE_RC_END &&
374 pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT)
377 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
378 if (ctrl & PCI_SRIOV_CTRL_VFE) {
379 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
383 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
388 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
393 if (pci_ari_enabled(dev->bus))
394 ctrl |= PCI_SRIOV_CTRL_ARI;
397 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
398 pci_write_config_word(dev, pos + PCI_SRIOV_NUM_VF, 0);
399 pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
400 pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
401 if (!offset || (total > 1 && !stride))
404 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
405 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
406 pgsz &= ~((1 << i) - 1);
411 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
413 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
418 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
419 res = &dev->resource[i + PCI_IOV_RESOURCES];
420 bar64 = __pci_read_base(dev, pci_bar_unknown, res,
421 pos + PCI_SRIOV_BAR + i * 4);
424 if (resource_size(res) & (PAGE_SIZE - 1)) {
428 iov->barsz[i] = resource_size(res);
429 res->end = res->start + resource_size(res) * total - 1;
430 dev_info(&dev->dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
439 iov->total_VFs = total;
440 iov->offset = offset;
441 iov->stride = stride;
444 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
445 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
446 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
447 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
450 iov->dev = pci_dev_get(pdev);
454 mutex_init(&iov->lock);
458 iov->max_VF_buses = virtfn_max_buses(dev);
463 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
464 res = &dev->resource[i + PCI_IOV_RESOURCES];
472 static void sriov_release(struct pci_dev *dev)
474 BUG_ON(dev->sriov->num_VFs);
476 if (dev != dev->sriov->dev)
477 pci_dev_put(dev->sriov->dev);
479 mutex_destroy(&dev->sriov->lock);
485 static void sriov_restore_state(struct pci_dev *dev)
489 struct pci_sriov *iov = dev->sriov;
491 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
492 if (ctrl & PCI_SRIOV_CTRL_VFE)
495 for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
496 pci_update_resource(dev, i);
498 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
499 pci_iov_set_numvfs(dev, iov->num_VFs);
500 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
501 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
506 * pci_iov_init - initialize the IOV capability
507 * @dev: the PCI device
509 * Returns 0 on success, or negative on failure.
511 int pci_iov_init(struct pci_dev *dev)
515 if (!pci_is_pcie(dev))
518 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
520 return sriov_init(dev, pos);
526 * pci_iov_release - release resources used by the IOV capability
527 * @dev: the PCI device
529 void pci_iov_release(struct pci_dev *dev)
536 * pci_iov_resource_bar - get position of the SR-IOV BAR
537 * @dev: the PCI device
538 * @resno: the resource number
540 * Returns position of the BAR encapsulated in the SR-IOV capability.
542 int pci_iov_resource_bar(struct pci_dev *dev, int resno)
544 if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
547 BUG_ON(!dev->is_physfn);
549 return dev->sriov->pos + PCI_SRIOV_BAR +
550 4 * (resno - PCI_IOV_RESOURCES);
554 * pci_sriov_resource_alignment - get resource alignment for VF BAR
555 * @dev: the PCI device
556 * @resno: the resource number
558 * Returns the alignment of the VF BAR found in the SR-IOV capability.
559 * This is not the same as the resource size which is defined as
560 * the VF BAR size multiplied by the number of VFs. The alignment
561 * is just the VF BAR size.
563 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
565 return pci_iov_resource_size(dev, resno);
569 * pci_restore_iov_state - restore the state of the IOV capability
570 * @dev: the PCI device
572 void pci_restore_iov_state(struct pci_dev *dev)
575 sriov_restore_state(dev);
579 * pci_iov_bus_range - find bus range used by Virtual Function
582 * Returns max number of buses (exclude current one) used by Virtual
585 int pci_iov_bus_range(struct pci_bus *bus)
590 list_for_each_entry(dev, &bus->devices, bus_list) {
593 if (dev->sriov->max_VF_buses > max)
594 max = dev->sriov->max_VF_buses;
597 return max ? max - bus->number : 0;
601 * pci_enable_sriov - enable the SR-IOV capability
602 * @dev: the PCI device
603 * @nr_virtfn: number of virtual functions to enable
605 * Returns 0 on success, or negative on failure.
607 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
614 return sriov_enable(dev, nr_virtfn);
616 EXPORT_SYMBOL_GPL(pci_enable_sriov);
619 * pci_disable_sriov - disable the SR-IOV capability
620 * @dev: the PCI device
622 void pci_disable_sriov(struct pci_dev *dev)
631 EXPORT_SYMBOL_GPL(pci_disable_sriov);
634 * pci_num_vf - return number of VFs associated with a PF device_release_driver
635 * @dev: the PCI device
637 * Returns number of VFs, or 0 if SR-IOV is not enabled.
639 int pci_num_vf(struct pci_dev *dev)
644 return dev->sriov->num_VFs;
646 EXPORT_SYMBOL_GPL(pci_num_vf);
649 * pci_vfs_assigned - returns number of VFs are assigned to a guest
650 * @dev: the PCI device
652 * Returns number of VFs belonging to this device that are assigned to a guest.
653 * If device is not a physical function returns 0.
655 int pci_vfs_assigned(struct pci_dev *dev)
657 struct pci_dev *vfdev;
658 unsigned int vfs_assigned = 0;
659 unsigned short dev_id;
661 /* only search if we are a PF */
666 * determine the device ID for the VFs, the vendor ID will be the
667 * same as the PF so there is no need to check for that one
669 pci_read_config_word(dev, dev->sriov->pos + PCI_SRIOV_VF_DID, &dev_id);
671 /* loop through all the VFs to see if we own any that are assigned */
672 vfdev = pci_get_device(dev->vendor, dev_id, NULL);
675 * It is considered assigned if it is a virtual function with
676 * our dev as the physical function and the assigned bit is set
678 if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
679 pci_is_dev_assigned(vfdev))
682 vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
687 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
690 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
691 * @dev: the PCI PF device
692 * @numvfs: number that should be used for TotalVFs supported
694 * Should be called from PF driver's probe routine with
695 * device's mutex held.
697 * Returns 0 if PF is an SRIOV-capable device and
698 * value of numvfs valid. If not a PF return -ENOSYS;
699 * if numvfs is invalid return -EINVAL;
700 * if VFs already enabled, return -EBUSY.
702 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
706 if (numvfs > dev->sriov->total_VFs)
709 /* Shouldn't change if VFs already enabled */
710 if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
713 dev->sriov->driver_max_VFs = numvfs;
717 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
720 * pci_sriov_get_totalvfs -- get total VFs supported on this device
721 * @dev: the PCI PF device
723 * For a PCIe device with SRIOV support, return the PCIe
724 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
725 * if the driver reduced it. Otherwise 0.
727 int pci_sriov_get_totalvfs(struct pci_dev *dev)
732 if (dev->sriov->driver_max_VFs)
733 return dev->sriov->driver_max_VFs;
735 return dev->sriov->total_VFs;
737 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);