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 int pci_iov_add_virtfn(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 rc
= pci_setup_device(virtfn
);
143 virtfn
->dev
.parent
= dev
->dev
.parent
;
144 virtfn
->physfn
= pci_dev_get(dev
);
145 virtfn
->is_virtfn
= 1;
146 virtfn
->multifunction
= 0;
148 for (i
= 0; i
< PCI_SRIOV_NUM_BARS
; i
++) {
149 res
= &dev
->resource
[i
+ PCI_IOV_RESOURCES
];
152 virtfn
->resource
[i
].name
= pci_name(virtfn
);
153 virtfn
->resource
[i
].flags
= res
->flags
;
154 size
= pci_iov_resource_size(dev
, i
+ PCI_IOV_RESOURCES
);
155 virtfn
->resource
[i
].start
= res
->start
+ size
* id
;
156 virtfn
->resource
[i
].end
= virtfn
->resource
[i
].start
+ size
- 1;
157 rc
= request_resource(res
, &virtfn
->resource
[i
]);
162 __pci_reset_function(virtfn
);
164 pci_device_add(virtfn
, virtfn
->bus
);
165 mutex_unlock(&iov
->dev
->sriov
->lock
);
167 pci_bus_add_device(virtfn
);
168 sprintf(buf
, "virtfn%u", id
);
169 rc
= sysfs_create_link(&dev
->dev
.kobj
, &virtfn
->dev
.kobj
, buf
);
172 rc
= sysfs_create_link(&virtfn
->dev
.kobj
, &dev
->dev
.kobj
, "physfn");
176 kobject_uevent(&virtfn
->dev
.kobj
, KOBJ_CHANGE
);
181 sysfs_remove_link(&dev
->dev
.kobj
, buf
);
184 mutex_lock(&iov
->dev
->sriov
->lock
);
185 pci_stop_and_remove_bus_device(virtfn
);
187 virtfn_remove_bus(dev
->bus
, bus
);
189 mutex_unlock(&iov
->dev
->sriov
->lock
);
194 void pci_iov_remove_virtfn(struct pci_dev
*dev
, int id
, int reset
)
196 char buf
[VIRTFN_ID_LEN
];
197 struct pci_dev
*virtfn
;
198 struct pci_sriov
*iov
= dev
->sriov
;
200 virtfn
= pci_get_domain_bus_and_slot(pci_domain_nr(dev
->bus
),
201 pci_iov_virtfn_bus(dev
, id
),
202 pci_iov_virtfn_devfn(dev
, id
));
207 device_release_driver(&virtfn
->dev
);
208 __pci_reset_function(virtfn
);
211 sprintf(buf
, "virtfn%u", id
);
212 sysfs_remove_link(&dev
->dev
.kobj
, buf
);
214 * pci_stop_dev() could have been called for this virtfn already,
215 * so the directory for the virtfn may have been removed before.
216 * Double check to avoid spurious sysfs warnings.
218 if (virtfn
->dev
.kobj
.sd
)
219 sysfs_remove_link(&virtfn
->dev
.kobj
, "physfn");
221 mutex_lock(&iov
->dev
->sriov
->lock
);
222 pci_stop_and_remove_bus_device(virtfn
);
223 virtfn_remove_bus(dev
->bus
, virtfn
->bus
);
224 mutex_unlock(&iov
->dev
->sriov
->lock
);
226 /* balance pci_get_domain_bus_and_slot() */
231 int __weak
pcibios_sriov_enable(struct pci_dev
*pdev
, u16 num_vfs
)
236 int __weak
pcibios_sriov_disable(struct pci_dev
*pdev
)
241 static int sriov_enable(struct pci_dev
*dev
, int nr_virtfn
)
247 struct resource
*res
;
248 struct pci_dev
*pdev
;
249 struct pci_sriov
*iov
= dev
->sriov
;
259 pci_read_config_word(dev
, iov
->pos
+ PCI_SRIOV_INITIAL_VF
, &initial
);
260 if (initial
> iov
->total_VFs
||
261 (!(iov
->cap
& PCI_SRIOV_CAP_VFM
) && (initial
!= iov
->total_VFs
)))
264 if (nr_virtfn
< 0 || nr_virtfn
> iov
->total_VFs
||
265 (!(iov
->cap
& PCI_SRIOV_CAP_VFM
) && (nr_virtfn
> initial
)))
269 for (i
= 0; i
< PCI_SRIOV_NUM_BARS
; i
++) {
270 bars
|= (1 << (i
+ PCI_IOV_RESOURCES
));
271 res
= &dev
->resource
[i
+ PCI_IOV_RESOURCES
];
275 if (nres
!= iov
->nres
) {
276 dev_err(&dev
->dev
, "not enough MMIO resources for SR-IOV\n");
280 bus
= pci_iov_virtfn_bus(dev
, nr_virtfn
- 1);
281 if (bus
> dev
->bus
->busn_res
.end
) {
282 dev_err(&dev
->dev
, "can't enable %d VFs (bus %02x out of range of %pR)\n",
283 nr_virtfn
, bus
, &dev
->bus
->busn_res
);
287 if (pci_enable_resources(dev
, bars
)) {
288 dev_err(&dev
->dev
, "SR-IOV: IOV BARS not allocated\n");
292 if (iov
->link
!= dev
->devfn
) {
293 pdev
= pci_get_slot(dev
->bus
, iov
->link
);
297 if (!pdev
->is_physfn
) {
302 rc
= sysfs_create_link(&dev
->dev
.kobj
,
303 &pdev
->dev
.kobj
, "dep_link");
309 iov
->initial_VFs
= initial
;
310 if (nr_virtfn
< initial
)
313 rc
= pcibios_sriov_enable(dev
, initial
);
315 dev_err(&dev
->dev
, "failure %d from pcibios_sriov_enable()\n", rc
);
319 pci_iov_set_numvfs(dev
, nr_virtfn
);
320 iov
->ctrl
|= PCI_SRIOV_CTRL_VFE
| PCI_SRIOV_CTRL_MSE
;
321 pci_cfg_access_lock(dev
);
322 pci_write_config_word(dev
, iov
->pos
+ PCI_SRIOV_CTRL
, iov
->ctrl
);
324 pci_cfg_access_unlock(dev
);
326 for (i
= 0; i
< initial
; i
++) {
327 rc
= pci_iov_add_virtfn(dev
, i
, 0);
332 kobject_uevent(&dev
->dev
.kobj
, KOBJ_CHANGE
);
333 iov
->num_VFs
= nr_virtfn
;
339 pci_iov_remove_virtfn(dev
, i
, 0);
341 pcibios_sriov_disable(dev
);
343 iov
->ctrl
&= ~(PCI_SRIOV_CTRL_VFE
| PCI_SRIOV_CTRL_MSE
);
344 pci_cfg_access_lock(dev
);
345 pci_write_config_word(dev
, iov
->pos
+ PCI_SRIOV_CTRL
, iov
->ctrl
);
347 pci_cfg_access_unlock(dev
);
349 if (iov
->link
!= dev
->devfn
)
350 sysfs_remove_link(&dev
->dev
.kobj
, "dep_link");
352 pci_iov_set_numvfs(dev
, 0);
356 static void sriov_disable(struct pci_dev
*dev
)
359 struct pci_sriov
*iov
= dev
->sriov
;
364 for (i
= 0; i
< iov
->num_VFs
; i
++)
365 pci_iov_remove_virtfn(dev
, i
, 0);
367 pcibios_sriov_disable(dev
);
369 iov
->ctrl
&= ~(PCI_SRIOV_CTRL_VFE
| PCI_SRIOV_CTRL_MSE
);
370 pci_cfg_access_lock(dev
);
371 pci_write_config_word(dev
, iov
->pos
+ PCI_SRIOV_CTRL
, iov
->ctrl
);
373 pci_cfg_access_unlock(dev
);
375 if (iov
->link
!= dev
->devfn
)
376 sysfs_remove_link(&dev
->dev
.kobj
, "dep_link");
379 pci_iov_set_numvfs(dev
, 0);
382 static int sriov_init(struct pci_dev
*dev
, int pos
)
389 struct pci_sriov
*iov
;
390 struct resource
*res
;
391 struct pci_dev
*pdev
;
393 pci_read_config_word(dev
, pos
+ PCI_SRIOV_CTRL
, &ctrl
);
394 if (ctrl
& PCI_SRIOV_CTRL_VFE
) {
395 pci_write_config_word(dev
, pos
+ PCI_SRIOV_CTRL
, 0);
400 list_for_each_entry(pdev
, &dev
->bus
->devices
, bus_list
)
405 if (pci_ari_enabled(dev
->bus
))
406 ctrl
|= PCI_SRIOV_CTRL_ARI
;
409 pci_write_config_word(dev
, pos
+ PCI_SRIOV_CTRL
, ctrl
);
411 pci_read_config_word(dev
, pos
+ PCI_SRIOV_TOTAL_VF
, &total
);
415 pci_read_config_dword(dev
, pos
+ PCI_SRIOV_SUP_PGSIZE
, &pgsz
);
416 i
= PAGE_SHIFT
> 12 ? PAGE_SHIFT
- 12 : 0;
417 pgsz
&= ~((1 << i
) - 1);
422 pci_write_config_dword(dev
, pos
+ PCI_SRIOV_SYS_PGSIZE
, pgsz
);
424 iov
= kzalloc(sizeof(*iov
), GFP_KERNEL
);
429 for (i
= 0; i
< PCI_SRIOV_NUM_BARS
; i
++) {
430 res
= &dev
->resource
[i
+ PCI_IOV_RESOURCES
];
432 * If it is already FIXED, don't change it, something
433 * (perhaps EA or header fixups) wants it this way.
435 if (res
->flags
& IORESOURCE_PCI_FIXED
)
436 bar64
= (res
->flags
& IORESOURCE_MEM_64
) ? 1 : 0;
438 bar64
= __pci_read_base(dev
, pci_bar_unknown
, res
,
439 pos
+ PCI_SRIOV_BAR
+ i
* 4);
442 if (resource_size(res
) & (PAGE_SIZE
- 1)) {
446 iov
->barsz
[i
] = resource_size(res
);
447 res
->end
= res
->start
+ resource_size(res
) * total
- 1;
448 dev_info(&dev
->dev
, "VF(n) BAR%d space: %pR (contains BAR%d for %d VFs)\n",
457 iov
->total_VFs
= total
;
460 pci_read_config_dword(dev
, pos
+ PCI_SRIOV_CAP
, &iov
->cap
);
461 pci_read_config_byte(dev
, pos
+ PCI_SRIOV_FUNC_LINK
, &iov
->link
);
462 if (pci_pcie_type(dev
) == PCI_EXP_TYPE_RC_END
)
463 iov
->link
= PCI_DEVFN(PCI_SLOT(dev
->devfn
), iov
->link
);
466 iov
->dev
= pci_dev_get(pdev
);
470 mutex_init(&iov
->lock
);
474 rc
= compute_max_vf_buses(dev
);
484 for (i
= 0; i
< PCI_SRIOV_NUM_BARS
; i
++) {
485 res
= &dev
->resource
[i
+ PCI_IOV_RESOURCES
];
493 static void sriov_release(struct pci_dev
*dev
)
495 BUG_ON(dev
->sriov
->num_VFs
);
497 if (dev
!= dev
->sriov
->dev
)
498 pci_dev_put(dev
->sriov
->dev
);
500 mutex_destroy(&dev
->sriov
->lock
);
506 static void sriov_restore_state(struct pci_dev
*dev
)
510 struct pci_sriov
*iov
= dev
->sriov
;
512 pci_read_config_word(dev
, iov
->pos
+ PCI_SRIOV_CTRL
, &ctrl
);
513 if (ctrl
& PCI_SRIOV_CTRL_VFE
)
516 for (i
= PCI_IOV_RESOURCES
; i
<= PCI_IOV_RESOURCE_END
; i
++)
517 pci_update_resource(dev
, i
);
519 pci_write_config_dword(dev
, iov
->pos
+ PCI_SRIOV_SYS_PGSIZE
, iov
->pgsz
);
520 pci_iov_set_numvfs(dev
, iov
->num_VFs
);
521 pci_write_config_word(dev
, iov
->pos
+ PCI_SRIOV_CTRL
, iov
->ctrl
);
522 if (iov
->ctrl
& PCI_SRIOV_CTRL_VFE
)
527 * pci_iov_init - initialize the IOV capability
528 * @dev: the PCI device
530 * Returns 0 on success, or negative on failure.
532 int pci_iov_init(struct pci_dev
*dev
)
536 if (!pci_is_pcie(dev
))
539 pos
= pci_find_ext_capability(dev
, PCI_EXT_CAP_ID_SRIOV
);
541 return sriov_init(dev
, pos
);
547 * pci_iov_release - release resources used by the IOV capability
548 * @dev: the PCI device
550 void pci_iov_release(struct pci_dev
*dev
)
557 * pci_iov_update_resource - update a VF BAR
558 * @dev: the PCI device
559 * @resno: the resource number
561 * Update a VF BAR in the SR-IOV capability of a PF.
563 void pci_iov_update_resource(struct pci_dev
*dev
, int resno
)
565 struct pci_sriov
*iov
= dev
->is_physfn
? dev
->sriov
: NULL
;
566 struct resource
*res
= dev
->resource
+ resno
;
567 int vf_bar
= resno
- PCI_IOV_RESOURCES
;
568 struct pci_bus_region region
;
574 * The generic pci_restore_bars() path calls this for all devices,
575 * including VFs and non-SR-IOV devices. If this is not a PF, we
576 * have nothing to do.
581 pci_read_config_word(dev
, iov
->pos
+ PCI_SRIOV_CTRL
, &cmd
);
582 if ((cmd
& PCI_SRIOV_CTRL_VFE
) && (cmd
& PCI_SRIOV_CTRL_MSE
)) {
583 dev_WARN(&dev
->dev
, "can't update enabled VF BAR%d %pR\n",
589 * Ignore unimplemented BARs, unused resource slots for 64-bit
590 * BARs, and non-movable resources, e.g., those described via
591 * Enhanced Allocation.
596 if (res
->flags
& IORESOURCE_UNSET
)
599 if (res
->flags
& IORESOURCE_PCI_FIXED
)
602 pcibios_resource_to_bus(dev
->bus
, ®ion
, res
);
604 new |= res
->flags
& ~PCI_BASE_ADDRESS_MEM_MASK
;
606 reg
= iov
->pos
+ PCI_SRIOV_BAR
+ 4 * vf_bar
;
607 pci_write_config_dword(dev
, reg
, new);
608 if (res
->flags
& IORESOURCE_MEM_64
) {
609 new = region
.start
>> 16 >> 16;
610 pci_write_config_dword(dev
, reg
+ 4, new);
614 resource_size_t __weak
pcibios_iov_resource_alignment(struct pci_dev
*dev
,
617 return pci_iov_resource_size(dev
, resno
);
621 * pci_sriov_resource_alignment - get resource alignment for VF BAR
622 * @dev: the PCI device
623 * @resno: the resource number
625 * Returns the alignment of the VF BAR found in the SR-IOV capability.
626 * This is not the same as the resource size which is defined as
627 * the VF BAR size multiplied by the number of VFs. The alignment
628 * is just the VF BAR size.
630 resource_size_t
pci_sriov_resource_alignment(struct pci_dev
*dev
, int resno
)
632 return pcibios_iov_resource_alignment(dev
, resno
);
636 * pci_restore_iov_state - restore the state of the IOV capability
637 * @dev: the PCI device
639 void pci_restore_iov_state(struct pci_dev
*dev
)
642 sriov_restore_state(dev
);
646 * pci_iov_bus_range - find bus range used by Virtual Function
649 * Returns max number of buses (exclude current one) used by Virtual
652 int pci_iov_bus_range(struct pci_bus
*bus
)
657 list_for_each_entry(dev
, &bus
->devices
, bus_list
) {
660 if (dev
->sriov
->max_VF_buses
> max
)
661 max
= dev
->sriov
->max_VF_buses
;
664 return max
? max
- bus
->number
: 0;
668 * pci_enable_sriov - enable the SR-IOV capability
669 * @dev: the PCI device
670 * @nr_virtfn: number of virtual functions to enable
672 * Returns 0 on success, or negative on failure.
674 int pci_enable_sriov(struct pci_dev
*dev
, int nr_virtfn
)
681 return sriov_enable(dev
, nr_virtfn
);
683 EXPORT_SYMBOL_GPL(pci_enable_sriov
);
686 * pci_disable_sriov - disable the SR-IOV capability
687 * @dev: the PCI device
689 void pci_disable_sriov(struct pci_dev
*dev
)
698 EXPORT_SYMBOL_GPL(pci_disable_sriov
);
701 * pci_num_vf - return number of VFs associated with a PF device_release_driver
702 * @dev: the PCI device
704 * Returns number of VFs, or 0 if SR-IOV is not enabled.
706 int pci_num_vf(struct pci_dev
*dev
)
711 return dev
->sriov
->num_VFs
;
713 EXPORT_SYMBOL_GPL(pci_num_vf
);
716 * pci_vfs_assigned - returns number of VFs are assigned to a guest
717 * @dev: the PCI device
719 * Returns number of VFs belonging to this device that are assigned to a guest.
720 * If device is not a physical function returns 0.
722 int pci_vfs_assigned(struct pci_dev
*dev
)
724 struct pci_dev
*vfdev
;
725 unsigned int vfs_assigned
= 0;
726 unsigned short dev_id
;
728 /* only search if we are a PF */
733 * determine the device ID for the VFs, the vendor ID will be the
734 * same as the PF so there is no need to check for that one
736 pci_read_config_word(dev
, dev
->sriov
->pos
+ PCI_SRIOV_VF_DID
, &dev_id
);
738 /* loop through all the VFs to see if we own any that are assigned */
739 vfdev
= pci_get_device(dev
->vendor
, dev_id
, NULL
);
742 * It is considered assigned if it is a virtual function with
743 * our dev as the physical function and the assigned bit is set
745 if (vfdev
->is_virtfn
&& (vfdev
->physfn
== dev
) &&
746 pci_is_dev_assigned(vfdev
))
749 vfdev
= pci_get_device(dev
->vendor
, dev_id
, vfdev
);
754 EXPORT_SYMBOL_GPL(pci_vfs_assigned
);
757 * pci_sriov_set_totalvfs -- reduce the TotalVFs available
758 * @dev: the PCI PF device
759 * @numvfs: number that should be used for TotalVFs supported
761 * Should be called from PF driver's probe routine with
762 * device's mutex held.
764 * Returns 0 if PF is an SRIOV-capable device and
765 * value of numvfs valid. If not a PF return -ENOSYS;
766 * if numvfs is invalid return -EINVAL;
767 * if VFs already enabled, return -EBUSY.
769 int pci_sriov_set_totalvfs(struct pci_dev
*dev
, u16 numvfs
)
773 if (numvfs
> dev
->sriov
->total_VFs
)
776 /* Shouldn't change if VFs already enabled */
777 if (dev
->sriov
->ctrl
& PCI_SRIOV_CTRL_VFE
)
780 dev
->sriov
->driver_max_VFs
= numvfs
;
784 EXPORT_SYMBOL_GPL(pci_sriov_set_totalvfs
);
787 * pci_sriov_get_totalvfs -- get total VFs supported on this device
788 * @dev: the PCI PF device
790 * For a PCIe device with SRIOV support, return the PCIe
791 * SRIOV capability value of TotalVFs or the value of driver_max_VFs
792 * if the driver reduced it. Otherwise 0.
794 int pci_sriov_get_totalvfs(struct pci_dev
*dev
)
799 if (dev
->sriov
->driver_max_VFs
)
800 return dev
->sriov
->driver_max_VFs
;
802 return dev
->sriov
->total_VFs
;
804 EXPORT_SYMBOL_GPL(pci_sriov_get_totalvfs
);