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)
37 struct pci_bus *child;
39 if (bus->number == busnr)
42 child = pci_find_bus(pci_domain_nr(bus), busnr);
46 child = pci_add_new_bus(bus, NULL, busnr);
50 pci_bus_insert_busn_res(child, busnr, busnr);
51 child->dev.parent = bus->bridge;
52 rc = pci_bus_add_child(child);
54 pci_remove_bus(child);
61 static void virtfn_remove_bus(struct pci_bus *bus, int busnr)
63 struct pci_bus *child;
65 if (bus->number == busnr)
68 child = pci_find_bus(pci_domain_nr(bus), busnr);
71 if (list_empty(&child->devices))
72 pci_remove_bus(child);
75 static int virtfn_add(struct pci_dev *dev, int id, int reset)
80 char buf[VIRTFN_ID_LEN];
81 struct pci_dev *virtfn;
83 struct pci_sriov *iov = dev->sriov;
85 virtfn = alloc_pci_dev();
89 mutex_lock(&iov->dev->sriov->lock);
90 virtfn->bus = virtfn_add_bus(dev->bus, virtfn_bus(dev, id));
93 mutex_unlock(&iov->dev->sriov->lock);
96 virtfn->devfn = virtfn_devfn(dev, id);
97 virtfn->vendor = dev->vendor;
98 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_DID, &virtfn->device);
99 pci_setup_device(virtfn);
100 virtfn->dev.parent = dev->dev.parent;
102 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
103 res = dev->resource + PCI_IOV_RESOURCES + i;
106 virtfn->resource[i].name = pci_name(virtfn);
107 virtfn->resource[i].flags = res->flags;
108 size = resource_size(res);
109 do_div(size, iov->total);
110 virtfn->resource[i].start = res->start + size * id;
111 virtfn->resource[i].end = virtfn->resource[i].start + size - 1;
112 rc = request_resource(res, &virtfn->resource[i]);
117 __pci_reset_function(virtfn);
119 pci_device_add(virtfn, virtfn->bus);
120 mutex_unlock(&iov->dev->sriov->lock);
122 virtfn->physfn = pci_dev_get(dev);
123 virtfn->is_virtfn = 1;
125 rc = pci_bus_add_device(virtfn);
128 sprintf(buf, "virtfn%u", id);
129 rc = sysfs_create_link(&dev->dev.kobj, &virtfn->dev.kobj, buf);
132 rc = sysfs_create_link(&virtfn->dev.kobj, &dev->dev.kobj, "physfn");
136 kobject_uevent(&virtfn->dev.kobj, KOBJ_CHANGE);
141 sysfs_remove_link(&dev->dev.kobj, buf);
144 mutex_lock(&iov->dev->sriov->lock);
145 pci_stop_and_remove_bus_device(virtfn);
146 virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
147 mutex_unlock(&iov->dev->sriov->lock);
152 static void virtfn_remove(struct pci_dev *dev, int id, int reset)
154 char buf[VIRTFN_ID_LEN];
156 struct pci_dev *virtfn;
157 struct pci_sriov *iov = dev->sriov;
159 bus = pci_find_bus(pci_domain_nr(dev->bus), virtfn_bus(dev, id));
163 virtfn = pci_get_slot(bus, virtfn_devfn(dev, id));
170 device_release_driver(&virtfn->dev);
171 __pci_reset_function(virtfn);
174 sprintf(buf, "virtfn%u", id);
175 sysfs_remove_link(&dev->dev.kobj, buf);
177 * pci_stop_dev() could have been called for this virtfn already,
178 * so the directory for the virtfn may have been removed before.
179 * Double check to avoid spurious sysfs warnings.
181 if (virtfn->dev.kobj.sd)
182 sysfs_remove_link(&virtfn->dev.kobj, "physfn");
184 mutex_lock(&iov->dev->sriov->lock);
185 pci_stop_and_remove_bus_device(virtfn);
186 virtfn_remove_bus(dev->bus, virtfn_bus(dev, id));
187 mutex_unlock(&iov->dev->sriov->lock);
192 static int sriov_migration(struct pci_dev *dev)
195 struct pci_sriov *iov = dev->sriov;
200 if (!(iov->cap & PCI_SRIOV_CAP_VFM))
203 pci_read_config_word(dev, iov->pos + PCI_SRIOV_STATUS, &status);
204 if (!(status & PCI_SRIOV_STATUS_VFM))
207 schedule_work(&iov->mtask);
212 static void sriov_migration_task(struct work_struct *work)
217 struct pci_sriov *iov = container_of(work, struct pci_sriov, mtask);
219 for (i = iov->initial; i < iov->nr_virtfn; i++) {
220 state = readb(iov->mstate + i);
221 if (state == PCI_SRIOV_VFM_MI) {
222 writeb(PCI_SRIOV_VFM_AV, iov->mstate + i);
223 state = readb(iov->mstate + i);
224 if (state == PCI_SRIOV_VFM_AV)
225 virtfn_add(iov->self, i, 1);
226 } else if (state == PCI_SRIOV_VFM_MO) {
227 virtfn_remove(iov->self, i, 1);
228 writeb(PCI_SRIOV_VFM_UA, iov->mstate + i);
229 state = readb(iov->mstate + i);
230 if (state == PCI_SRIOV_VFM_AV)
231 virtfn_add(iov->self, i, 0);
235 pci_read_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, &status);
236 status &= ~PCI_SRIOV_STATUS_VFM;
237 pci_write_config_word(iov->self, iov->pos + PCI_SRIOV_STATUS, status);
240 static int sriov_enable_migration(struct pci_dev *dev, int nr_virtfn)
245 struct pci_sriov *iov = dev->sriov;
247 if (nr_virtfn <= iov->initial)
250 pci_read_config_dword(dev, iov->pos + PCI_SRIOV_VFM, &table);
251 bir = PCI_SRIOV_VFM_BIR(table);
252 if (bir > PCI_STD_RESOURCE_END)
255 table = PCI_SRIOV_VFM_OFFSET(table);
256 if (table + nr_virtfn > pci_resource_len(dev, bir))
259 pa = pci_resource_start(dev, bir) + table;
260 iov->mstate = ioremap(pa, nr_virtfn);
264 INIT_WORK(&iov->mtask, sriov_migration_task);
266 iov->ctrl |= PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR;
267 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
272 static void sriov_disable_migration(struct pci_dev *dev)
274 struct pci_sriov *iov = dev->sriov;
276 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFM | PCI_SRIOV_CTRL_INTR);
277 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
279 cancel_work_sync(&iov->mtask);
280 iounmap(iov->mstate);
283 static int sriov_enable(struct pci_dev *dev, int nr_virtfn)
288 u16 offset, stride, initial;
289 struct resource *res;
290 struct pci_dev *pdev;
291 struct pci_sriov *iov = dev->sriov;
300 pci_read_config_word(dev, iov->pos + PCI_SRIOV_INITIAL_VF, &initial);
301 if (initial > iov->total ||
302 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (initial != iov->total)))
305 if (nr_virtfn < 0 || nr_virtfn > iov->total ||
306 (!(iov->cap & PCI_SRIOV_CAP_VFM) && (nr_virtfn > initial)))
309 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, nr_virtfn);
310 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_OFFSET, &offset);
311 pci_read_config_word(dev, iov->pos + PCI_SRIOV_VF_STRIDE, &stride);
312 if (!offset || (nr_virtfn > 1 && !stride))
316 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
317 bars |= (1 << (i + PCI_IOV_RESOURCES));
318 res = dev->resource + PCI_IOV_RESOURCES + i;
322 if (nres != iov->nres) {
323 dev_err(&dev->dev, "not enough MMIO resources for SR-IOV\n");
327 iov->offset = offset;
328 iov->stride = stride;
330 if (virtfn_bus(dev, nr_virtfn - 1) > dev->bus->busn_res.end) {
331 dev_err(&dev->dev, "SR-IOV: bus number out of range\n");
335 if (pci_enable_resources(dev, bars)) {
336 dev_err(&dev->dev, "SR-IOV: IOV BARS not allocated\n");
340 if (iov->link != dev->devfn) {
341 pdev = pci_get_slot(dev->bus, iov->link);
347 if (!pdev->is_physfn)
350 rc = sysfs_create_link(&dev->dev.kobj,
351 &pdev->dev.kobj, "dep_link");
356 iov->ctrl |= PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE;
357 pci_cfg_access_lock(dev);
358 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
360 pci_cfg_access_unlock(dev);
362 iov->initial = initial;
363 if (nr_virtfn < initial)
366 for (i = 0; i < initial; i++) {
367 rc = virtfn_add(dev, i, 0);
372 if (iov->cap & PCI_SRIOV_CAP_VFM) {
373 rc = sriov_enable_migration(dev, nr_virtfn);
378 kobject_uevent(&dev->dev.kobj, KOBJ_CHANGE);
379 iov->nr_virtfn = nr_virtfn;
384 for (j = 0; j < i; j++)
385 virtfn_remove(dev, j, 0);
387 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
388 pci_cfg_access_lock(dev);
389 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
391 pci_cfg_access_unlock(dev);
393 if (iov->link != dev->devfn)
394 sysfs_remove_link(&dev->dev.kobj, "dep_link");
399 static void sriov_disable(struct pci_dev *dev)
402 struct pci_sriov *iov = dev->sriov;
407 if (iov->cap & PCI_SRIOV_CAP_VFM)
408 sriov_disable_migration(dev);
410 for (i = 0; i < iov->nr_virtfn; i++)
411 virtfn_remove(dev, i, 0);
413 iov->ctrl &= ~(PCI_SRIOV_CTRL_VFE | PCI_SRIOV_CTRL_MSE);
414 pci_cfg_access_lock(dev);
415 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
417 pci_cfg_access_unlock(dev);
419 if (iov->link != dev->devfn)
420 sysfs_remove_link(&dev->dev.kobj, "dep_link");
425 static int sriov_init(struct pci_dev *dev, int pos)
431 u16 ctrl, total, offset, stride;
432 struct pci_sriov *iov;
433 struct resource *res;
434 struct pci_dev *pdev;
436 if (dev->pcie_type != PCI_EXP_TYPE_RC_END &&
437 dev->pcie_type != PCI_EXP_TYPE_ENDPOINT)
440 pci_read_config_word(dev, pos + PCI_SRIOV_CTRL, &ctrl);
441 if (ctrl & PCI_SRIOV_CTRL_VFE) {
442 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, 0);
446 pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
451 list_for_each_entry(pdev, &dev->bus->devices, bus_list)
456 if (pci_ari_enabled(dev->bus))
457 ctrl |= PCI_SRIOV_CTRL_ARI;
460 pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
461 pci_read_config_word(dev, pos + PCI_SRIOV_VF_OFFSET, &offset);
462 pci_read_config_word(dev, pos + PCI_SRIOV_VF_STRIDE, &stride);
463 if (!offset || (total > 1 && !stride))
466 pci_read_config_dword(dev, pos + PCI_SRIOV_SUP_PGSIZE, &pgsz);
467 i = PAGE_SHIFT > 12 ? PAGE_SHIFT - 12 : 0;
468 pgsz &= ~((1 << i) - 1);
473 pci_write_config_dword(dev, pos + PCI_SRIOV_SYS_PGSIZE, pgsz);
476 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
477 res = dev->resource + PCI_IOV_RESOURCES + i;
478 i += __pci_read_base(dev, pci_bar_unknown, res,
479 pos + PCI_SRIOV_BAR + i * 4);
482 if (resource_size(res) & (PAGE_SIZE - 1)) {
486 res->end = res->start + resource_size(res) * total - 1;
490 iov = kzalloc(sizeof(*iov), GFP_KERNEL);
500 iov->offset = offset;
501 iov->stride = stride;
504 pci_read_config_dword(dev, pos + PCI_SRIOV_CAP, &iov->cap);
505 pci_read_config_byte(dev, pos + PCI_SRIOV_FUNC_LINK, &iov->link);
506 if (dev->pcie_type == PCI_EXP_TYPE_RC_END)
507 iov->link = PCI_DEVFN(PCI_SLOT(dev->devfn), iov->link);
510 iov->dev = pci_dev_get(pdev);
514 mutex_init(&iov->lock);
522 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
523 res = dev->resource + PCI_IOV_RESOURCES + i;
530 static void sriov_release(struct pci_dev *dev)
532 BUG_ON(dev->sriov->nr_virtfn);
534 if (dev != dev->sriov->dev)
535 pci_dev_put(dev->sriov->dev);
537 mutex_destroy(&dev->sriov->lock);
543 static void sriov_restore_state(struct pci_dev *dev)
547 struct pci_sriov *iov = dev->sriov;
549 pci_read_config_word(dev, iov->pos + PCI_SRIOV_CTRL, &ctrl);
550 if (ctrl & PCI_SRIOV_CTRL_VFE)
553 for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++)
554 pci_update_resource(dev, i);
556 pci_write_config_dword(dev, iov->pos + PCI_SRIOV_SYS_PGSIZE, iov->pgsz);
557 pci_write_config_word(dev, iov->pos + PCI_SRIOV_NUM_VF, iov->nr_virtfn);
558 pci_write_config_word(dev, iov->pos + PCI_SRIOV_CTRL, iov->ctrl);
559 if (iov->ctrl & PCI_SRIOV_CTRL_VFE)
564 * pci_iov_init - initialize the IOV capability
565 * @dev: the PCI device
567 * Returns 0 on success, or negative on failure.
569 int pci_iov_init(struct pci_dev *dev)
573 if (!pci_is_pcie(dev))
576 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_SRIOV);
578 return sriov_init(dev, pos);
584 * pci_iov_release - release resources used by the IOV capability
585 * @dev: the PCI device
587 void pci_iov_release(struct pci_dev *dev)
594 * pci_iov_resource_bar - get position of the SR-IOV BAR
595 * @dev: the PCI device
596 * @resno: the resource number
597 * @type: the BAR type to be filled in
599 * Returns position of the BAR encapsulated in the SR-IOV capability.
601 int pci_iov_resource_bar(struct pci_dev *dev, int resno,
602 enum pci_bar_type *type)
604 if (resno < PCI_IOV_RESOURCES || resno > PCI_IOV_RESOURCE_END)
607 BUG_ON(!dev->is_physfn);
609 *type = pci_bar_unknown;
611 return dev->sriov->pos + PCI_SRIOV_BAR +
612 4 * (resno - PCI_IOV_RESOURCES);
616 * pci_sriov_resource_alignment - get resource alignment for VF BAR
617 * @dev: the PCI device
618 * @resno: the resource number
620 * Returns the alignment of the VF BAR found in the SR-IOV capability.
621 * This is not the same as the resource size which is defined as
622 * the VF BAR size multiplied by the number of VFs. The alignment
623 * is just the VF BAR size.
625 resource_size_t pci_sriov_resource_alignment(struct pci_dev *dev, int resno)
628 enum pci_bar_type type;
629 int reg = pci_iov_resource_bar(dev, resno, &type);
634 __pci_read_base(dev, type, &tmp, reg);
635 return resource_alignment(&tmp);
639 * pci_restore_iov_state - restore the state of the IOV capability
640 * @dev: the PCI device
642 void pci_restore_iov_state(struct pci_dev *dev)
645 sriov_restore_state(dev);
649 * pci_iov_bus_range - find bus range used by Virtual Function
652 * Returns max number of buses (exclude current one) used by Virtual
655 int pci_iov_bus_range(struct pci_bus *bus)
661 list_for_each_entry(dev, &bus->devices, bus_list) {
664 busnr = virtfn_bus(dev, dev->sriov->total - 1);
669 return max ? max - bus->number : 0;
673 * pci_enable_sriov - enable the SR-IOV capability
674 * @dev: the PCI device
675 * @nr_virtfn: number of virtual functions to enable
677 * Returns 0 on success, or negative on failure.
679 int pci_enable_sriov(struct pci_dev *dev, int nr_virtfn)
686 return sriov_enable(dev, nr_virtfn);
688 EXPORT_SYMBOL_GPL(pci_enable_sriov);
691 * pci_disable_sriov - disable the SR-IOV capability
692 * @dev: the PCI device
694 void pci_disable_sriov(struct pci_dev *dev)
703 EXPORT_SYMBOL_GPL(pci_disable_sriov);
706 * pci_sriov_migration - notify SR-IOV core of Virtual Function Migration
707 * @dev: the PCI device
709 * Returns IRQ_HANDLED if the IRQ is handled, or IRQ_NONE if not.
711 * Physical Function driver is responsible to register IRQ handler using
712 * VF Migration Interrupt Message Number, and call this function when the
713 * interrupt is generated by the hardware.
715 irqreturn_t pci_sriov_migration(struct pci_dev *dev)
720 return sriov_migration(dev) ? IRQ_HANDLED : IRQ_NONE;
722 EXPORT_SYMBOL_GPL(pci_sriov_migration);
725 * pci_num_vf - return number of VFs associated with a PF device_release_driver
726 * @dev: the PCI device
728 * Returns number of VFs, or 0 if SR-IOV is not enabled.
730 int pci_num_vf(struct pci_dev *dev)
732 if (!dev || !dev->is_physfn)
735 return dev->sriov->nr_virtfn;
737 EXPORT_SYMBOL_GPL(pci_num_vf);