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 static int virtfn_add(struct pci_dev *dev, int id, int reset)
65 char buf[VIRTFN_ID_LEN];
66 struct pci_dev *virtfn;
68 struct pci_sriov *iov = dev->sriov;
71 mutex_lock(&iov->dev->sriov->lock);
72 bus = virtfn_add_bus(dev->bus, virtfn_bus(dev, id));
76 virtfn = pci_alloc_dev(bus);
80 virtfn->devfn = virtfn_devfn(dev, id);
81 virtfn->vendor = dev->vendor;
82 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
83 pci_setup_device(virtfn);
84 virtfn->dev.parent = dev->dev.parent;
85 virtfn->physfn = pci_dev_get(dev);
86 virtfn->is_virtfn = 1;
87 virtfn->multifunction = 0;
89 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
90 res = dev->resource + PCI_IOV_RESOURCES + i;
93 virtfn->resource[i].name = pci_name(virtfn);
94 virtfn->resource[i].flags = res->flags;
95 size = resource_size(res);
96 do_div(size, iov->total_VFs);
97 virtfn->resource[i].start = res->start + size * id;
98 virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
99 rc = request_resource(res, &virtfn->resource[i]);
104 __pci_reset_function(virtfn);
106 pci_device_add(virtfn, virtfn->bus);
107 mutex_unlock(&iov->dev->sriov->lock);
109 pci_bus_add_device(virtfn);
110 sprintf(buf, "virtfn%u", id);
111 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
114 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
118 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
123 sysfs_remove_link(&dev->dev.kobj, buf);
126 mutex_lock(&iov->dev->sriov->lock);
127 pci_stop_and_remove_bus_device(virtfn);
129 virtfn_remove_bus(dev->bus, bus);
131 mutex_unlock(&iov->dev->sriov->lock);
136 static void virtfn_remove(struct pci_dev *dev, int id, int reset)
138 char buf[VIRTFN_ID_LEN];
139 struct pci_dev *virtfn;
140 struct pci_sriov *iov = dev->sriov;
142 virtfn = pci_get_domain_bus_and_slot(pci_domain_nr(dev->bus),
144 virtfn_devfn(dev, id));
149 device_release_driver(&virtfn->dev);
150 __pci_reset_function(virtfn);
153 sprintf(buf, "virtfn%u", id);
154 sysfs_remove_link(&dev->dev.kobj, buf);
156 * pci_stop_dev() could have been called for this virtfn already,
157 * so the directory for the virtfn may have been removed before.
158 * Double check to avoid spurious sysfs warnings.
160 if (virtfn->dev.kobj.sd)
161 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
163 mutex_lock(&iov->dev->sriov->lock);
164 pci_stop_and_remove_bus_device(virtfn);
165 virtfn_remove_bus(dev->bus, virtfn->bus);
166 mutex_unlock(&iov->dev->sriov->lock);
168 /* balance pci_get_domain_bus_and_slot() */
173 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
178 u16 offset, stride, initial;
179 struct resource *res;
180 struct pci_dev *pdev;
181 struct pci_sriov *iov = dev->sriov;
191 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
192 if (initial > iov->total_VFs ||
193 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total_VFs)))
196 if (nr_virtfn < 0 || nr_virtfn > iov->total_VFs ||
197 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
200 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
201 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
202 if (!offset || (nr_virtfn > 1 && !stride))
206 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
207 bars |= (1 << (i + PCI_IOV_RESOURCES));
208 res = dev->resource + PCI_IOV_RESOURCES + i;
212 if (nres != iov->nres) {
213 dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
217 iov->offset = offset;
218 iov->stride = stride;
220 bus = virtfn_bus(dev, nr_virtfn - 1);
221 if (bus > dev->bus->busn_res.end) {
222 dev_err(&dev->dev, "can't enable %d VFs (bus %02x out of range of %pR)\n",
223 nr_virtfn, bus, &dev->bus->busn_res);
227 if (pci_enable_resources(dev, bars)) {
228 dev_err(&dev->dev, "SR-IOV: IOV BARS not allocated\n");
232 if (iov->link != dev->devfn) {
233 pdev = pci_get_slot(dev->bus, iov->link);
237 if (!pdev->is_physfn) {
242 rc = sysfs_create_link(&dev->dev.kobj,
243 &pdev->dev.kobj, "dep_link");
249 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
250 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
251 pci_cfg_access_lock(dev);
252 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
254 pci_cfg_access_unlock(dev);
256 iov->initial_VFs = initial;
257 if (nr_virtfn < initial)
260 for (i = 0; i < initial; i++) {
261 rc = virtfn_add(dev, i, 0);
266 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
267 iov->num_VFs = nr_virtfn;
272 for (j = 0; j < i; j++)
273 virtfn_remove(dev, j, 0);
275 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
276 pci_cfg_access_lock(dev);
277 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
278 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, 0);
280 pci_cfg_access_unlock(dev);
282 if (iov->link != dev->devfn)
283 sysfs_remove_link(&dev->dev.kobj, "dep_link");
288 static void sriov_disable(struct pci_dev *dev)
291 struct pci_sriov *iov = dev->sriov;
296 for (i = 0; i < iov->num_VFs; i++)
297 virtfn_remove(dev, i, 0);
299 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
300 pci_cfg_access_lock(dev);
301 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
303 pci_cfg_access_unlock(dev);
305 if (iov->link != dev->devfn)
306 sysfs_remove_link(&dev->dev.kobj, "dep_link");
309 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, 0);
312 static int sriov_init(struct pci_dev *dev, int pos)
318 u16 ctrl, total, offset, stride;
319 struct pci_sriov *iov;
320 struct resource *res;
321 struct pci_dev *pdev;
323 if (pci_pcie_type(dev) != PCI_EXP_TYPE_RC_END &&
324 pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT)
327 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
328 if (ctrl & PCI_SRIOV_CTRL_VFE) {
329 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
333 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
338 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
343 if (pci_ari_enabled(dev->bus))
344 ctrl |= PCI_SRIOV_CTRL_ARI;
347 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
348 pci_write_config_word(dev, pos + PCI_SRIOV_NUM_VF, 0);
349 pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
350 pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
351 if (!offset || (total > 1 && !stride))
354 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
355 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
356 pgsz &= ~((1 << i) - 1);
361 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
364 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
365 res = dev->resource + PCI_IOV_RESOURCES + i;
366 i += __pci_read_base(dev, pci_bar_unknown, res,
367 pos + PCI_SRIOV_BAR + i * 4);
370 if (resource_size(res) & (PAGE_SIZE - 1)) {
374 res->end = res->start + resource_size(res) * total - 1;
375 dev_info(&dev->dev, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
380 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
389 iov->total_VFs = total;
390 iov->offset = offset;
391 iov->stride = stride;
394 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
395 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
396 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END)
397 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
400 iov->dev = pci_dev_get(pdev);
404 mutex_init(&iov->lock);
412 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
413 res = dev->resource + PCI_IOV_RESOURCES + i;
420 static void sriov_release(struct pci_dev *dev)
422 BUG_ON(dev->sriov->num_VFs);
424 if (dev != dev->sriov->dev)
425 pci_dev_put(dev->sriov->dev);
427 mutex_destroy(&dev->sriov->lock);
433 static void sriov_restore_state(struct pci_dev *dev)
437 struct pci_sriov *iov = dev->sriov;
439 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
440 if (ctrl & PCI_SRIOV_CTRL_VFE)
443 for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
444 pci_update_resource(dev, i);
446 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
447 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->num_VFs);
448 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
449 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
454 * pci_iov_init - initialize the IOV capability
455 * @dev: the PCI device
457 * Returns 0 on success, or negative on failure.
459 int pci_iov_init(struct pci_dev *dev)
463 if (!pci_is_pcie(dev))
466 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
468 return sriov_init(dev, pos);
474 * pci_iov_release - release resources used by the IOV capability
475 * @dev: the PCI device
477 void pci_iov_release(struct pci_dev *dev)
484 * pci_iov_resource_bar - get position of the SR-IOV BAR
485 * @dev: the PCI device
486 * @resno: the resource number
488 * Returns position of the BAR encapsulated in the SR-IOV capability.
490 int pci_iov_resource_bar(struct pci_dev *dev, int resno)
492 if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
495 BUG_ON(!dev->is_physfn);
497 return dev->sriov->pos + PCI_SRIOV_BAR +
498 4 * (resno - PCI_IOV_RESOURCES);
502 * pci_sriov_resource_alignment - get resource alignment for VF BAR
503 * @dev: the PCI device
504 * @resno: the resource number
506 * Returns the alignment of the VF BAR found in the SR-IOV capability.
507 * This is not the same as the resource size which is defined as
508 * the VF BAR size multiplied by the number of VFs. The alignment
509 * is just the VF BAR size.
511 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
514 int reg = pci_iov_resource_bar(dev, resno);
519 __pci_read_base(dev, pci_bar_unknown, &tmp, reg);
520 return resource_alignment(&tmp);
524 * pci_restore_iov_state - restore the state of the IOV capability
525 * @dev: the PCI device
527 void pci_restore_iov_state(struct pci_dev *dev)
530 sriov_restore_state(dev);
534 * pci_iov_bus_range - find bus range used by Virtual Function
537 * Returns max number of buses (exclude current one) used by Virtual
540 int pci_iov_bus_range(struct pci_bus *bus)
546 list_for_each_entry(dev, &bus->devices, bus_list) {
549 busnr = virtfn_bus(dev, dev->sriov->total_VFs - 1);
554 return max ? max - bus->number : 0;
558 * pci_enable_sriov - enable the SR-IOV capability
559 * @dev: the PCI device
560 * @nr_virtfn: number of virtual functions to enable
562 * Returns 0 on success, or negative on failure.
564 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
571 return sriov_enable(dev, nr_virtfn);
573 EXPORT_SYMBOL_GPL(pci_enable_sriov);
576 * pci_disable_sriov - disable the SR-IOV capability
577 * @dev: the PCI device
579 void pci_disable_sriov(struct pci_dev *dev)
588 EXPORT_SYMBOL_GPL(pci_disable_sriov);
591 * pci_num_vf - return number of VFs associated with a PF device_release_driver
592 * @dev: the PCI device
594 * Returns number of VFs, or 0 if SR-IOV is not enabled.
596 int pci_num_vf(struct pci_dev *dev)
601 return dev->sriov->num_VFs;
603 EXPORT_SYMBOL_GPL(pci_num_vf);
606 * pci_vfs_assigned - returns number of VFs are assigned to a guest
607 * @dev: the PCI device
609 * Returns number of VFs belonging to this device that are assigned to a guest.
610 * If device is not a physical function returns 0.
612 int pci_vfs_assigned(struct pci_dev *dev)
614 struct pci_dev *vfdev;
615 unsigned int vfs_assigned = 0;
616 unsigned short dev_id;
618 /* only search if we are a PF */
623 * determine the device ID for the VFs, the vendor ID will be the
624 * same as the PF so there is no need to check for that one
626 pci_read_config_word(dev, dev->sriov->pos + PCI_SRIOV_VF_DID, &dev_id);
628 /* loop through all the VFs to see if we own any that are assigned */
629 vfdev = pci_get_device(dev->vendor, dev_id, NULL);
632 * It is considered assigned if it is a virtual function with
633 * our dev as the physical function and the assigned bit is set
635 if (vfdev->is_virtfn && (vfdev->physfn == dev) &&
636 pci_is_dev_assigned(vfdev))
639 vfdev = pci_get_device(dev->vendor, dev_id, vfdev);
644 EXPORT_SYMBOL_GPL(pci_vfs_assigned);
647 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
648 * @dev: the PCI PF device
649 * @numvfs: number that should be used for TotalVFs supported
651 * Should be called from PF driver's probe routine with
652 * device's mutex held.
654 * Returns 0 if PF is an SRIOV-capable device and
655 * value of numvfs valid. If not a PF return -ENOSYS;
656 * if numvfs is invalid return -EINVAL;
657 * if VFs already enabled, return -EBUSY.
659 int pci_sriov_set_totalvfs(struct pci_dev *dev, u16 numvfs)
663 if (numvfs > dev->sriov->total_VFs)
666 /* Shouldn't change if VFs already enabled */
667 if (dev->sriov->ctrl & PCI_SRIOV_CTRL_VFE)
670 dev->sriov->driver_max_VFs = numvfs;
674 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs);
677 * pci_sriov_get_totalvfs -- get total VFs supported on this device
678 * @dev: the PCI PF device
680 * For a PCIe device with SRIOV support, return the PCIe
681 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
682 * if the driver reduced it. Otherwise 0.
684 int pci_sriov_get_totalvfs(struct pci_dev *dev)
689 if (dev->sriov->driver_max_VFs)
690 return dev->sriov->driver_max_VFs;
692 return dev->sriov->total_VFs;
694 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs);