2 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
3 * Author: Alex Williamson <alex.williamson@redhat.com>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
9 * Derived from original vfio:
10 * Copyright 2010 Cisco Systems, Inc. All rights reserved.
11 * Author: Tom Lyon, pugs@cisco.com
14 #include <linux/device.h>
15 #include <linux/eventfd.h>
16 #include <linux/interrupt.h>
17 #include <linux/iommu.h>
18 #include <linux/module.h>
19 #include <linux/mutex.h>
20 #include <linux/notifier.h>
21 #include <linux/pci.h>
22 #include <linux/pm_runtime.h>
23 #include <linux/slab.h>
24 #include <linux/types.h>
25 #include <linux/uaccess.h>
26 #include <linux/vfio.h>
28 #include "vfio_pci_private.h"
30 #define DRIVER_VERSION "0.2"
31 #define DRIVER_AUTHOR "Alex Williamson <alex.williamson@redhat.com>"
32 #define DRIVER_DESC "VFIO PCI - User Level meta-driver"
34 static bool nointxmask;
35 module_param_named(nointxmask, nointxmask, bool, S_IRUGO | S_IWUSR);
36 MODULE_PARM_DESC(nointxmask,
37 "Disable support for PCI 2.3 style INTx masking. If this resolves problems for specific devices, report lspci -vvvxxx to linux-pci@vger.kernel.org so the device can be fixed automatically via the broken_intx_masking flag.");
39 static int vfio_pci_enable(struct vfio_pci_device *vdev)
41 struct pci_dev *pdev = vdev->pdev;
46 vdev->reset_works = (pci_reset_function(pdev) == 0);
48 vdev->pci_saved_state = pci_store_saved_state(pdev);
49 if (!vdev->pci_saved_state)
50 pr_debug("%s: Couldn't store %s saved state\n",
51 __func__, dev_name(&pdev->dev));
53 ret = vfio_config_init(vdev);
57 if (likely(!nointxmask))
58 vdev->pci_2_3 = pci_intx_mask_supported(pdev);
60 pci_read_config_word(pdev, PCI_COMMAND, &cmd);
61 if (vdev->pci_2_3 && (cmd & PCI_COMMAND_INTX_DISABLE)) {
62 cmd &= ~PCI_COMMAND_INTX_DISABLE;
63 pci_write_config_word(pdev, PCI_COMMAND, cmd);
66 msix_pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
71 pci_read_config_word(pdev, msix_pos + PCI_MSIX_FLAGS, &flags);
72 pci_read_config_dword(pdev, msix_pos + PCI_MSIX_TABLE, &table);
74 vdev->msix_bar = table & PCI_MSIX_FLAGS_BIRMASK;
75 vdev->msix_offset = table & ~PCI_MSIX_FLAGS_BIRMASK;
76 vdev->msix_size = ((flags & PCI_MSIX_FLAGS_QSIZE) + 1) * 16;
78 vdev->msix_bar = 0xFF;
80 ret = pci_enable_device(pdev);
87 kfree(vdev->pci_saved_state);
88 vdev->pci_saved_state = NULL;
89 vfio_config_free(vdev);
93 static void vfio_pci_disable(struct vfio_pci_device *vdev)
97 pci_disable_device(vdev->pdev);
99 vfio_pci_set_irqs_ioctl(vdev, VFIO_IRQ_SET_DATA_NONE |
100 VFIO_IRQ_SET_ACTION_TRIGGER,
101 vdev->irq_type, 0, 0, NULL);
103 vdev->virq_disabled = false;
105 vfio_config_free(vdev);
107 pci_reset_function(vdev->pdev);
109 if (pci_load_and_free_saved_state(vdev->pdev,
110 &vdev->pci_saved_state) == 0)
111 pci_restore_state(vdev->pdev);
113 pr_info("%s: Couldn't reload %s saved state\n",
114 __func__, dev_name(&vdev->pdev->dev));
116 for (bar = PCI_STD_RESOURCES; bar <= PCI_STD_RESOURCE_END; bar++) {
117 if (!vdev->barmap[bar])
119 pci_iounmap(vdev->pdev, vdev->barmap[bar]);
120 pci_release_selected_regions(vdev->pdev, 1 << bar);
121 vdev->barmap[bar] = NULL;
125 static void vfio_pci_release(void *device_data)
127 struct vfio_pci_device *vdev = device_data;
129 if (atomic_dec_and_test(&vdev->refcnt))
130 vfio_pci_disable(vdev);
132 module_put(THIS_MODULE);
135 static int vfio_pci_open(void *device_data)
137 struct vfio_pci_device *vdev = device_data;
139 if (!try_module_get(THIS_MODULE))
142 if (atomic_inc_return(&vdev->refcnt) == 1) {
143 int ret = vfio_pci_enable(vdev);
145 module_put(THIS_MODULE);
153 static int vfio_pci_get_irq_count(struct vfio_pci_device *vdev, int irq_type)
155 if (irq_type == VFIO_PCI_INTX_IRQ_INDEX) {
157 pci_read_config_byte(vdev->pdev, PCI_INTERRUPT_PIN, &pin);
161 } else if (irq_type == VFIO_PCI_MSI_IRQ_INDEX) {
165 pos = pci_find_capability(vdev->pdev, PCI_CAP_ID_MSI);
167 pci_read_config_word(vdev->pdev,
168 pos + PCI_MSI_FLAGS, &flags);
170 return 1 << (flags & PCI_MSI_FLAGS_QMASK);
172 } else if (irq_type == VFIO_PCI_MSIX_IRQ_INDEX) {
176 pos = pci_find_capability(vdev->pdev, PCI_CAP_ID_MSIX);
178 pci_read_config_word(vdev->pdev,
179 pos + PCI_MSIX_FLAGS, &flags);
181 return (flags & PCI_MSIX_FLAGS_QSIZE) + 1;
188 static long vfio_pci_ioctl(void *device_data,
189 unsigned int cmd, unsigned long arg)
191 struct vfio_pci_device *vdev = device_data;
194 if (cmd == VFIO_DEVICE_GET_INFO) {
195 struct vfio_device_info info;
197 minsz = offsetofend(struct vfio_device_info, num_irqs);
199 if (copy_from_user(&info, (void __user *)arg, minsz))
202 if (info.argsz < minsz)
205 info.flags = VFIO_DEVICE_FLAGS_PCI;
207 if (vdev->reset_works)
208 info.flags |= VFIO_DEVICE_FLAGS_RESET;
210 info.num_regions = VFIO_PCI_NUM_REGIONS;
211 info.num_irqs = VFIO_PCI_NUM_IRQS;
213 return copy_to_user((void __user *)arg, &info, minsz);
215 } else if (cmd == VFIO_DEVICE_GET_REGION_INFO) {
216 struct pci_dev *pdev = vdev->pdev;
217 struct vfio_region_info info;
219 minsz = offsetofend(struct vfio_region_info, offset);
221 if (copy_from_user(&info, (void __user *)arg, minsz))
224 if (info.argsz < minsz)
227 switch (info.index) {
228 case VFIO_PCI_CONFIG_REGION_INDEX:
229 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
230 info.size = pdev->cfg_size;
231 info.flags = VFIO_REGION_INFO_FLAG_READ |
232 VFIO_REGION_INFO_FLAG_WRITE;
234 case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
235 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
236 info.size = pci_resource_len(pdev, info.index);
242 info.flags = VFIO_REGION_INFO_FLAG_READ |
243 VFIO_REGION_INFO_FLAG_WRITE;
244 if (pci_resource_flags(pdev, info.index) &
245 IORESOURCE_MEM && info.size >= PAGE_SIZE)
246 info.flags |= VFIO_REGION_INFO_FLAG_MMAP;
248 case VFIO_PCI_ROM_REGION_INDEX:
253 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
256 /* Report the BAR size, not the ROM size */
257 info.size = pci_resource_len(pdev, info.index);
261 /* Is it really there? */
262 io = pci_map_rom(pdev, &size);
267 pci_unmap_rom(pdev, io);
269 info.flags = VFIO_REGION_INFO_FLAG_READ;
276 return copy_to_user((void __user *)arg, &info, minsz);
278 } else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) {
279 struct vfio_irq_info info;
281 minsz = offsetofend(struct vfio_irq_info, count);
283 if (copy_from_user(&info, (void __user *)arg, minsz))
286 if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS)
289 info.flags = VFIO_IRQ_INFO_EVENTFD;
291 info.count = vfio_pci_get_irq_count(vdev, info.index);
293 if (info.index == VFIO_PCI_INTX_IRQ_INDEX)
294 info.flags |= (VFIO_IRQ_INFO_MASKABLE |
295 VFIO_IRQ_INFO_AUTOMASKED);
297 info.flags |= VFIO_IRQ_INFO_NORESIZE;
299 return copy_to_user((void __user *)arg, &info, minsz);
301 } else if (cmd == VFIO_DEVICE_SET_IRQS) {
302 struct vfio_irq_set hdr;
306 minsz = offsetofend(struct vfio_irq_set, count);
308 if (copy_from_user(&hdr, (void __user *)arg, minsz))
311 if (hdr.argsz < minsz || hdr.index >= VFIO_PCI_NUM_IRQS ||
312 hdr.flags & ~(VFIO_IRQ_SET_DATA_TYPE_MASK |
313 VFIO_IRQ_SET_ACTION_TYPE_MASK))
316 if (!(hdr.flags & VFIO_IRQ_SET_DATA_NONE)) {
319 if (hdr.flags & VFIO_IRQ_SET_DATA_BOOL)
320 size = sizeof(uint8_t);
321 else if (hdr.flags & VFIO_IRQ_SET_DATA_EVENTFD)
322 size = sizeof(int32_t);
326 if (hdr.argsz - minsz < hdr.count * size ||
327 hdr.count > vfio_pci_get_irq_count(vdev, hdr.index))
330 data = kmalloc(hdr.count * size, GFP_KERNEL);
334 if (copy_from_user(data, (void __user *)(arg + minsz),
341 mutex_lock(&vdev->igate);
343 ret = vfio_pci_set_irqs_ioctl(vdev, hdr.flags, hdr.index,
344 hdr.start, hdr.count, data);
346 mutex_unlock(&vdev->igate);
351 } else if (cmd == VFIO_DEVICE_RESET)
352 return vdev->reset_works ?
353 pci_reset_function(vdev->pdev) : -EINVAL;
358 static ssize_t vfio_pci_read(void *device_data, char __user *buf,
359 size_t count, loff_t *ppos)
361 unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
362 struct vfio_pci_device *vdev = device_data;
363 struct pci_dev *pdev = vdev->pdev;
365 if (index >= VFIO_PCI_NUM_REGIONS)
368 if (index == VFIO_PCI_CONFIG_REGION_INDEX)
369 return vfio_pci_config_readwrite(vdev, buf, count, ppos, false);
370 else if (index == VFIO_PCI_ROM_REGION_INDEX)
371 return vfio_pci_mem_readwrite(vdev, buf, count, ppos, false);
372 else if (pci_resource_flags(pdev, index) & IORESOURCE_IO)
373 return vfio_pci_io_readwrite(vdev, buf, count, ppos, false);
374 else if (pci_resource_flags(pdev, index) & IORESOURCE_MEM)
375 return vfio_pci_mem_readwrite(vdev, buf, count, ppos, false);
380 static ssize_t vfio_pci_write(void *device_data, const char __user *buf,
381 size_t count, loff_t *ppos)
383 unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
384 struct vfio_pci_device *vdev = device_data;
385 struct pci_dev *pdev = vdev->pdev;
387 if (index >= VFIO_PCI_NUM_REGIONS)
390 if (index == VFIO_PCI_CONFIG_REGION_INDEX)
391 return vfio_pci_config_readwrite(vdev, (char __user *)buf,
393 else if (index == VFIO_PCI_ROM_REGION_INDEX)
395 else if (pci_resource_flags(pdev, index) & IORESOURCE_IO)
396 return vfio_pci_io_readwrite(vdev, (char __user *)buf,
398 else if (pci_resource_flags(pdev, index) & IORESOURCE_MEM) {
399 return vfio_pci_mem_readwrite(vdev, (char __user *)buf,
406 static int vfio_pci_mmap(void *device_data, struct vm_area_struct *vma)
408 struct vfio_pci_device *vdev = device_data;
409 struct pci_dev *pdev = vdev->pdev;
411 u64 phys_len, req_len, pgoff, req_start, phys;
414 index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
416 if (vma->vm_end < vma->vm_start)
418 if ((vma->vm_flags & VM_SHARED) == 0)
420 if (index >= VFIO_PCI_ROM_REGION_INDEX)
422 if (!(pci_resource_flags(pdev, index) & IORESOURCE_MEM))
425 phys_len = pci_resource_len(pdev, index);
426 req_len = vma->vm_end - vma->vm_start;
427 pgoff = vma->vm_pgoff &
428 ((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1);
429 req_start = pgoff << PAGE_SHIFT;
431 if (phys_len < PAGE_SIZE || req_start + req_len > phys_len)
434 if (index == vdev->msix_bar) {
436 * Disallow mmaps overlapping the MSI-X table; users don't
437 * get to touch this directly. We could find somewhere
438 * else to map the overlap, but page granularity is only
439 * a recommendation, not a requirement, so the user needs
440 * to know which bits are real. Requiring them to mmap
441 * around the table makes that clear.
444 /* If neither entirely above nor below, then it overlaps */
445 if (!(req_start >= vdev->msix_offset + vdev->msix_size ||
446 req_start + req_len <= vdev->msix_offset))
451 * Even though we don't make use of the barmap for the mmap,
452 * we need to request the region and the barmap tracks that.
454 if (!vdev->barmap[index]) {
455 ret = pci_request_selected_regions(pdev,
456 1 << index, "vfio-pci");
460 vdev->barmap[index] = pci_iomap(pdev, index, 0);
463 vma->vm_private_data = vdev;
464 vma->vm_flags |= VM_IO;
465 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
467 phys = (pci_resource_start(pdev, index) >> PAGE_SHIFT) + pgoff;
469 return remap_pfn_range(vma, vma->vm_start, phys,
470 req_len, vma->vm_page_prot);
473 static const struct vfio_device_ops vfio_pci_ops = {
475 .open = vfio_pci_open,
476 .release = vfio_pci_release,
477 .ioctl = vfio_pci_ioctl,
478 .read = vfio_pci_read,
479 .write = vfio_pci_write,
480 .mmap = vfio_pci_mmap,
483 static int vfio_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
486 struct vfio_pci_device *vdev;
487 struct iommu_group *group;
490 pci_read_config_byte(pdev, PCI_HEADER_TYPE, &type);
491 if ((type & PCI_HEADER_TYPE) != PCI_HEADER_TYPE_NORMAL)
494 group = iommu_group_get(&pdev->dev);
498 vdev = kzalloc(sizeof(*vdev), GFP_KERNEL);
500 iommu_group_put(group);
505 vdev->irq_type = VFIO_PCI_NUM_IRQS;
506 mutex_init(&vdev->igate);
507 spin_lock_init(&vdev->irqlock);
508 atomic_set(&vdev->refcnt, 0);
510 ret = vfio_add_group_dev(&pdev->dev, &vfio_pci_ops, vdev);
512 iommu_group_put(group);
519 static void vfio_pci_remove(struct pci_dev *pdev)
521 struct vfio_pci_device *vdev;
523 vdev = vfio_del_group_dev(&pdev->dev);
527 iommu_group_put(pdev->dev.iommu_group);
531 static struct pci_driver vfio_pci_driver = {
533 .id_table = NULL, /* only dynamic ids */
534 .probe = vfio_pci_probe,
535 .remove = vfio_pci_remove,
538 static void __exit vfio_pci_cleanup(void)
540 pci_unregister_driver(&vfio_pci_driver);
541 vfio_pci_virqfd_exit();
542 vfio_pci_uninit_perm_bits();
545 static int __init vfio_pci_init(void)
549 /* Allocate shared config space permision data used by all devices */
550 ret = vfio_pci_init_perm_bits();
554 /* Start the virqfd cleanup handler */
555 ret = vfio_pci_virqfd_init();
559 /* Register and scan for devices */
560 ret = pci_register_driver(&vfio_pci_driver);
567 vfio_pci_virqfd_exit();
569 vfio_pci_uninit_perm_bits();
573 module_init(vfio_pci_init);
574 module_exit(vfio_pci_cleanup);
576 MODULE_VERSION(DRIVER_VERSION);
577 MODULE_LICENSE("GPL v2");
578 MODULE_AUTHOR(DRIVER_AUTHOR);
579 MODULE_DESCRIPTION(DRIVER_DESC);