2 * VFIO PCI config space virtualization
4 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
5 * Author: Alex Williamson <alex.williamson@redhat.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
11 * Derived from original vfio:
12 * Copyright 2010 Cisco Systems, Inc. All rights reserved.
13 * Author: Tom Lyon, pugs@cisco.com
17 * This code handles reading and writing of PCI configuration registers.
18 * This is hairy because we want to allow a lot of flexibility to the
19 * user driver, but cannot trust it with all of the config fields.
20 * Tables determine which fields can be read and written, as well as
21 * which fields are 'virtualized' - special actions and translations to
22 * make it appear to the user that he has control, when in fact things
23 * must be negotiated with the underlying OS.
27 #include <linux/pci.h>
28 #include <linux/uaccess.h>
29 #include <linux/vfio.h>
31 #include "vfio_pci_private.h"
33 #define PCI_CFG_SPACE_SIZE 256
35 /* Useful "pseudo" capabilities */
36 #define PCI_CAP_ID_BASIC 0
37 #define PCI_CAP_ID_INVALID 0xFF
39 #define is_bar(offset) \
40 ((offset >= PCI_BASE_ADDRESS_0 && offset < PCI_BASE_ADDRESS_5 + 4) || \
41 (offset >= PCI_ROM_ADDRESS && offset < PCI_ROM_ADDRESS + 4))
44 * Lengths of PCI Config Capabilities
45 * 0: Removed from the user visible capability list
48 static u8 pci_cap_length[] = {
49 [PCI_CAP_ID_BASIC] = PCI_STD_HEADER_SIZEOF, /* pci config header */
50 [PCI_CAP_ID_PM] = PCI_PM_SIZEOF,
51 [PCI_CAP_ID_AGP] = PCI_AGP_SIZEOF,
52 [PCI_CAP_ID_VPD] = PCI_CAP_VPD_SIZEOF,
53 [PCI_CAP_ID_SLOTID] = 0, /* bridge - don't care */
54 [PCI_CAP_ID_MSI] = 0xFF, /* 10, 14, 20, or 24 */
55 [PCI_CAP_ID_CHSWP] = 0, /* cpci - not yet */
56 [PCI_CAP_ID_PCIX] = 0xFF, /* 8 or 24 */
57 [PCI_CAP_ID_HT] = 0xFF, /* hypertransport */
58 [PCI_CAP_ID_VNDR] = 0xFF, /* variable */
59 [PCI_CAP_ID_DBG] = 0, /* debug - don't care */
60 [PCI_CAP_ID_CCRC] = 0, /* cpci - not yet */
61 [PCI_CAP_ID_SHPC] = 0, /* hotswap - not yet */
62 [PCI_CAP_ID_SSVID] = 0, /* bridge - don't care */
63 [PCI_CAP_ID_AGP3] = 0, /* AGP8x - not yet */
64 [PCI_CAP_ID_SECDEV] = 0, /* secure device not yet */
65 [PCI_CAP_ID_EXP] = 0xFF, /* 20 or 44 */
66 [PCI_CAP_ID_MSIX] = PCI_CAP_MSIX_SIZEOF,
67 [PCI_CAP_ID_SATA] = 0xFF,
68 [PCI_CAP_ID_AF] = PCI_CAP_AF_SIZEOF,
72 * Lengths of PCIe/PCI-X Extended Config Capabilities
73 * 0: Removed or masked from the user visible capabilty list
76 static u16 pci_ext_cap_length[] = {
77 [PCI_EXT_CAP_ID_ERR] = PCI_ERR_ROOT_COMMAND,
78 [PCI_EXT_CAP_ID_VC] = 0xFF,
79 [PCI_EXT_CAP_ID_DSN] = PCI_EXT_CAP_DSN_SIZEOF,
80 [PCI_EXT_CAP_ID_PWR] = PCI_EXT_CAP_PWR_SIZEOF,
81 [PCI_EXT_CAP_ID_RCLD] = 0, /* root only - don't care */
82 [PCI_EXT_CAP_ID_RCILC] = 0, /* root only - don't care */
83 [PCI_EXT_CAP_ID_RCEC] = 0, /* root only - don't care */
84 [PCI_EXT_CAP_ID_MFVC] = 0xFF,
85 [PCI_EXT_CAP_ID_VC9] = 0xFF, /* same as CAP_ID_VC */
86 [PCI_EXT_CAP_ID_RCRB] = 0, /* root only - don't care */
87 [PCI_EXT_CAP_ID_VNDR] = 0xFF,
88 [PCI_EXT_CAP_ID_CAC] = 0, /* obsolete */
89 [PCI_EXT_CAP_ID_ACS] = 0xFF,
90 [PCI_EXT_CAP_ID_ARI] = PCI_EXT_CAP_ARI_SIZEOF,
91 [PCI_EXT_CAP_ID_ATS] = PCI_EXT_CAP_ATS_SIZEOF,
92 [PCI_EXT_CAP_ID_SRIOV] = PCI_EXT_CAP_SRIOV_SIZEOF,
93 [PCI_EXT_CAP_ID_MRIOV] = 0, /* not yet */
94 [PCI_EXT_CAP_ID_MCAST] = PCI_EXT_CAP_MCAST_ENDPOINT_SIZEOF,
95 [PCI_EXT_CAP_ID_PRI] = PCI_EXT_CAP_PRI_SIZEOF,
96 [PCI_EXT_CAP_ID_AMD_XXX] = 0, /* not yet */
97 [PCI_EXT_CAP_ID_REBAR] = 0xFF,
98 [PCI_EXT_CAP_ID_DPA] = 0xFF,
99 [PCI_EXT_CAP_ID_TPH] = 0xFF,
100 [PCI_EXT_CAP_ID_LTR] = PCI_EXT_CAP_LTR_SIZEOF,
101 [PCI_EXT_CAP_ID_SECPCI] = 0, /* not yet */
102 [PCI_EXT_CAP_ID_PMUX] = 0, /* not yet */
103 [PCI_EXT_CAP_ID_PASID] = 0, /* not yet */
107 * Read/Write Permission Bits - one bit for each bit in capability
108 * Any field can be read if it exists, but what is read depends on
109 * whether the field is 'virtualized', or just pass thru to the
110 * hardware. Any virtualized field is also virtualized for writes.
111 * Writes are only permitted if they have a 1 bit here.
114 u8 *virt; /* read/write virtual data, not hw */
115 u8 *write; /* writeable bits */
116 int (*readfn)(struct vfio_pci_device *vdev, int pos, int count,
117 struct perm_bits *perm, int offset, __le32 *val);
118 int (*writefn)(struct vfio_pci_device *vdev, int pos, int count,
119 struct perm_bits *perm, int offset, __le32 val);
123 #define ALL_VIRT 0xFFFFFFFFU
125 #define ALL_WRITE 0xFFFFFFFFU
127 static int vfio_user_config_read(struct pci_dev *pdev, int offset,
128 __le32 *val, int count)
137 ret = pci_user_read_config_byte(pdev, offset, &tmp);
144 ret = pci_user_read_config_word(pdev, offset, &tmp);
149 ret = pci_user_read_config_dword(pdev, offset, &tmp_val);
153 *val = cpu_to_le32(tmp_val);
155 return pcibios_err_to_errno(ret);
158 static int vfio_user_config_write(struct pci_dev *pdev, int offset,
159 __le32 val, int count)
162 u32 tmp_val = le32_to_cpu(val);
166 ret = pci_user_write_config_byte(pdev, offset, tmp_val);
169 ret = pci_user_write_config_word(pdev, offset, tmp_val);
172 ret = pci_user_write_config_dword(pdev, offset, tmp_val);
176 return pcibios_err_to_errno(ret);
179 static int vfio_default_config_read(struct vfio_pci_device *vdev, int pos,
180 int count, struct perm_bits *perm,
181 int offset, __le32 *val)
185 memcpy(val, vdev->vconfig + pos, count);
187 memcpy(&virt, perm->virt + offset, count);
189 /* Any non-virtualized bits? */
190 if (cpu_to_le32(~0U >> (32 - (count * 8))) != virt) {
191 struct pci_dev *pdev = vdev->pdev;
195 ret = vfio_user_config_read(pdev, pos, &phys_val, count);
199 *val = (phys_val & ~virt) | (*val & virt);
205 static int vfio_default_config_write(struct vfio_pci_device *vdev, int pos,
206 int count, struct perm_bits *perm,
207 int offset, __le32 val)
209 __le32 virt = 0, write = 0;
211 memcpy(&write, perm->write + offset, count);
214 return count; /* drop, no writable bits */
216 memcpy(&virt, perm->virt + offset, count);
218 /* Virtualized and writable bits go to vconfig */
222 memcpy(&virt_val, vdev->vconfig + pos, count);
224 virt_val &= ~(write & virt);
225 virt_val |= (val & (write & virt));
227 memcpy(vdev->vconfig + pos, &virt_val, count);
230 /* Non-virtualzed and writable bits go to hardware */
232 struct pci_dev *pdev = vdev->pdev;
236 ret = vfio_user_config_read(pdev, pos, &phys_val, count);
240 phys_val &= ~(write & ~virt);
241 phys_val |= (val & (write & ~virt));
243 ret = vfio_user_config_write(pdev, pos, phys_val, count);
251 /* Allow direct read from hardware, except for capability next pointer */
252 static int vfio_direct_config_read(struct vfio_pci_device *vdev, int pos,
253 int count, struct perm_bits *perm,
254 int offset, __le32 *val)
258 ret = vfio_user_config_read(vdev->pdev, pos, val, count);
260 return pcibios_err_to_errno(ret);
262 if (pos >= PCI_CFG_SPACE_SIZE) { /* Extended cap header mangling */
264 memcpy(val, vdev->vconfig + pos, count);
265 } else if (pos >= PCI_STD_HEADER_SIZEOF) { /* Std cap mangling */
266 if (offset == PCI_CAP_LIST_ID && count > 1)
267 memcpy(val, vdev->vconfig + pos,
268 min(PCI_CAP_FLAGS, count));
269 else if (offset == PCI_CAP_LIST_NEXT)
270 memcpy(val, vdev->vconfig + pos, 1);
276 static int vfio_direct_config_write(struct vfio_pci_device *vdev, int pos,
277 int count, struct perm_bits *perm,
278 int offset, __le32 val)
282 ret = vfio_user_config_write(vdev->pdev, pos, val, count);
289 /* Default all regions to read-only, no-virtualization */
290 static struct perm_bits cap_perms[PCI_CAP_ID_MAX + 1] = {
291 [0 ... PCI_CAP_ID_MAX] = { .readfn = vfio_direct_config_read }
293 static struct perm_bits ecap_perms[PCI_EXT_CAP_ID_MAX + 1] = {
294 [0 ... PCI_EXT_CAP_ID_MAX] = { .readfn = vfio_direct_config_read }
297 static void free_perm_bits(struct perm_bits *perm)
305 static int alloc_perm_bits(struct perm_bits *perm, int size)
308 * Round up all permission bits to the next dword, this lets us
309 * ignore whether a read/write exceeds the defined capability
310 * structure. We can do this because:
311 * - Standard config space is already dword aligned
312 * - Capabilities are all dword alinged (bits 0:1 of next reserved)
313 * - Express capabilities defined as dword aligned
315 size = round_up(size, 4);
319 * - All Readable, None Writeable, None Virtualized
321 perm->virt = kzalloc(size, GFP_KERNEL);
322 perm->write = kzalloc(size, GFP_KERNEL);
323 if (!perm->virt || !perm->write) {
324 free_perm_bits(perm);
328 perm->readfn = vfio_default_config_read;
329 perm->writefn = vfio_default_config_write;
335 * Helper functions for filling in permission tables
337 static inline void p_setb(struct perm_bits *p, int off, u8 virt, u8 write)
340 p->write[off] = write;
343 /* Handle endian-ness - pci and tables are little-endian */
344 static inline void p_setw(struct perm_bits *p, int off, u16 virt, u16 write)
346 *(__le16 *)(&p->virt[off]) = cpu_to_le16(virt);
347 *(__le16 *)(&p->write[off]) = cpu_to_le16(write);
350 /* Handle endian-ness - pci and tables are little-endian */
351 static inline void p_setd(struct perm_bits *p, int off, u32 virt, u32 write)
353 *(__le32 *)(&p->virt[off]) = cpu_to_le32(virt);
354 *(__le32 *)(&p->write[off]) = cpu_to_le32(write);
358 * Restore the *real* BARs after we detect a FLR or backdoor reset.
359 * (backdoor = some device specific technique that we didn't catch)
361 static void vfio_bar_restore(struct vfio_pci_device *vdev)
363 struct pci_dev *pdev = vdev->pdev;
364 u32 *rbar = vdev->rbar;
370 pr_info("%s: %s reset recovery - restoring bars\n",
371 __func__, dev_name(&pdev->dev));
373 for (i = PCI_BASE_ADDRESS_0; i <= PCI_BASE_ADDRESS_5; i += 4, rbar++)
374 pci_user_write_config_dword(pdev, i, *rbar);
376 pci_user_write_config_dword(pdev, PCI_ROM_ADDRESS, *rbar);
379 static __le32 vfio_generate_bar_flags(struct pci_dev *pdev, int bar)
381 unsigned long flags = pci_resource_flags(pdev, bar);
384 if (flags & IORESOURCE_IO)
385 return cpu_to_le32(PCI_BASE_ADDRESS_SPACE_IO);
387 val = PCI_BASE_ADDRESS_SPACE_MEMORY;
389 if (flags & IORESOURCE_PREFETCH)
390 val |= PCI_BASE_ADDRESS_MEM_PREFETCH;
392 if (flags & IORESOURCE_MEM_64)
393 val |= PCI_BASE_ADDRESS_MEM_TYPE_64;
395 return cpu_to_le32(val);
399 * Pretend we're hardware and tweak the values of the *virtual* PCI BARs
400 * to reflect the hardware capabilities. This implements BAR sizing.
402 static void vfio_bar_fixup(struct vfio_pci_device *vdev)
404 struct pci_dev *pdev = vdev->pdev;
409 bar = (__le32 *)&vdev->vconfig[PCI_BASE_ADDRESS_0];
411 for (i = PCI_STD_RESOURCES; i <= PCI_STD_RESOURCE_END; i++, bar++) {
412 if (!pci_resource_start(pdev, i)) {
413 *bar = 0; /* Unmapped by host = unimplemented to user */
417 mask = ~(pci_resource_len(pdev, i) - 1);
419 *bar &= cpu_to_le32((u32)mask);
420 *bar |= vfio_generate_bar_flags(pdev, i);
422 if (*bar & cpu_to_le32(PCI_BASE_ADDRESS_MEM_TYPE_64)) {
424 *bar &= cpu_to_le32((u32)(mask >> 32));
429 bar = (__le32 *)&vdev->vconfig[PCI_ROM_ADDRESS];
432 * NB. we expose the actual BAR size here, regardless of whether
433 * we can read it. When we report the REGION_INFO for the ROM
434 * we report what PCI tells us is the actual ROM size.
436 if (pci_resource_start(pdev, PCI_ROM_RESOURCE)) {
437 mask = ~(pci_resource_len(pdev, PCI_ROM_RESOURCE) - 1);
438 mask |= PCI_ROM_ADDRESS_ENABLE;
439 *bar &= cpu_to_le32((u32)mask);
443 vdev->bardirty = false;
446 static int vfio_basic_config_read(struct vfio_pci_device *vdev, int pos,
447 int count, struct perm_bits *perm,
448 int offset, __le32 *val)
450 if (is_bar(offset)) /* pos == offset for basic config */
451 vfio_bar_fixup(vdev);
453 count = vfio_default_config_read(vdev, pos, count, perm, offset, val);
455 /* Mask in virtual memory enable for SR-IOV devices */
456 if (offset == PCI_COMMAND && vdev->pdev->is_virtfn) {
457 u16 cmd = le16_to_cpu(*(__le16 *)&vdev->vconfig[PCI_COMMAND]);
458 u32 tmp_val = le32_to_cpu(*val);
460 tmp_val |= cmd & PCI_COMMAND_MEMORY;
461 *val = cpu_to_le32(tmp_val);
467 static int vfio_basic_config_write(struct vfio_pci_device *vdev, int pos,
468 int count, struct perm_bits *perm,
469 int offset, __le32 val)
471 struct pci_dev *pdev = vdev->pdev;
476 virt_cmd = (__le16 *)&vdev->vconfig[PCI_COMMAND];
478 if (offset == PCI_COMMAND) {
479 bool phys_mem, virt_mem, new_mem, phys_io, virt_io, new_io;
482 ret = pci_user_read_config_word(pdev, PCI_COMMAND, &phys_cmd);
486 new_cmd = le32_to_cpu(val);
488 phys_mem = !!(phys_cmd & PCI_COMMAND_MEMORY);
489 virt_mem = !!(le16_to_cpu(*virt_cmd) & PCI_COMMAND_MEMORY);
490 new_mem = !!(new_cmd & PCI_COMMAND_MEMORY);
492 phys_io = !!(phys_cmd & PCI_COMMAND_IO);
493 virt_io = !!(le16_to_cpu(*virt_cmd) & PCI_COMMAND_IO);
494 new_io = !!(new_cmd & PCI_COMMAND_IO);
497 * If the user is writing mem/io enable (new_mem/io) and we
498 * think it's already enabled (virt_mem/io), but the hardware
499 * shows it disabled (phys_mem/io, then the device has
500 * undergone some kind of backdoor reset and needs to be
501 * restored before we allow it to enable the bars.
502 * SR-IOV devices will trigger this, but we catch them later
504 if ((new_mem && virt_mem && !phys_mem) ||
505 (new_io && virt_io && !phys_io))
506 vfio_bar_restore(vdev);
509 count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
514 * Save current memory/io enable bits in vconfig to allow for
515 * the test above next time.
517 if (offset == PCI_COMMAND) {
518 u16 mask = PCI_COMMAND_MEMORY | PCI_COMMAND_IO;
520 *virt_cmd &= cpu_to_le16(~mask);
521 *virt_cmd |= cpu_to_le16(new_cmd & mask);
524 /* Emulate INTx disable */
525 if (offset >= PCI_COMMAND && offset <= PCI_COMMAND + 1) {
526 bool virt_intx_disable;
528 virt_intx_disable = !!(le16_to_cpu(*virt_cmd) &
529 PCI_COMMAND_INTX_DISABLE);
531 if (virt_intx_disable && !vdev->virq_disabled) {
532 vdev->virq_disabled = true;
533 vfio_pci_intx_mask(vdev);
534 } else if (!virt_intx_disable && vdev->virq_disabled) {
535 vdev->virq_disabled = false;
536 vfio_pci_intx_unmask(vdev);
541 vdev->bardirty = true;
546 /* Permissions for the Basic PCI Header */
547 static int __init init_pci_cap_basic_perm(struct perm_bits *perm)
549 if (alloc_perm_bits(perm, PCI_STD_HEADER_SIZEOF))
552 perm->readfn = vfio_basic_config_read;
553 perm->writefn = vfio_basic_config_write;
555 /* Virtualized for SR-IOV functions, which just have FFFF */
556 p_setw(perm, PCI_VENDOR_ID, (u16)ALL_VIRT, NO_WRITE);
557 p_setw(perm, PCI_DEVICE_ID, (u16)ALL_VIRT, NO_WRITE);
560 * Virtualize INTx disable, we use it internally for interrupt
561 * control and can emulate it for non-PCI 2.3 devices.
563 p_setw(perm, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE, (u16)ALL_WRITE);
565 /* Virtualize capability list, we might want to skip/disable */
566 p_setw(perm, PCI_STATUS, PCI_STATUS_CAP_LIST, NO_WRITE);
568 /* No harm to write */
569 p_setb(perm, PCI_CACHE_LINE_SIZE, NO_VIRT, (u8)ALL_WRITE);
570 p_setb(perm, PCI_LATENCY_TIMER, NO_VIRT, (u8)ALL_WRITE);
571 p_setb(perm, PCI_BIST, NO_VIRT, (u8)ALL_WRITE);
573 /* Virtualize all bars, can't touch the real ones */
574 p_setd(perm, PCI_BASE_ADDRESS_0, ALL_VIRT, ALL_WRITE);
575 p_setd(perm, PCI_BASE_ADDRESS_1, ALL_VIRT, ALL_WRITE);
576 p_setd(perm, PCI_BASE_ADDRESS_2, ALL_VIRT, ALL_WRITE);
577 p_setd(perm, PCI_BASE_ADDRESS_3, ALL_VIRT, ALL_WRITE);
578 p_setd(perm, PCI_BASE_ADDRESS_4, ALL_VIRT, ALL_WRITE);
579 p_setd(perm, PCI_BASE_ADDRESS_5, ALL_VIRT, ALL_WRITE);
580 p_setd(perm, PCI_ROM_ADDRESS, ALL_VIRT, ALL_WRITE);
582 /* Allow us to adjust capability chain */
583 p_setb(perm, PCI_CAPABILITY_LIST, (u8)ALL_VIRT, NO_WRITE);
585 /* Sometimes used by sw, just virtualize */
586 p_setb(perm, PCI_INTERRUPT_LINE, (u8)ALL_VIRT, (u8)ALL_WRITE);
590 static int vfio_pm_config_write(struct vfio_pci_device *vdev, int pos,
591 int count, struct perm_bits *perm,
592 int offset, __le32 val)
594 count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
598 if (offset == PCI_PM_CTRL) {
601 switch (le32_to_cpu(val) & PCI_PM_CTRL_STATE_MASK) {
616 pci_set_power_state(vdev->pdev, state);
622 /* Permissions for the Power Management capability */
623 static int __init init_pci_cap_pm_perm(struct perm_bits *perm)
625 if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_PM]))
628 perm->writefn = vfio_pm_config_write;
631 * We always virtualize the next field so we can remove
632 * capabilities from the chain if we want to.
634 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
637 * Power management is defined *per function*, so we can let
638 * the user change power state, but we trap and initiate the
639 * change ourselves, so the state bits are read-only.
641 p_setd(perm, PCI_PM_CTRL, NO_VIRT, ~PCI_PM_CTRL_STATE_MASK);
645 /* Permissions for PCI-X capability */
646 static int __init init_pci_cap_pcix_perm(struct perm_bits *perm)
648 /* Alloc 24, but only 8 are used in v0 */
649 if (alloc_perm_bits(perm, PCI_CAP_PCIX_SIZEOF_V2))
652 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
654 p_setw(perm, PCI_X_CMD, NO_VIRT, (u16)ALL_WRITE);
655 p_setd(perm, PCI_X_ECC_CSR, NO_VIRT, ALL_WRITE);
659 /* Permissions for PCI Express capability */
660 static int __init init_pci_cap_exp_perm(struct perm_bits *perm)
662 /* Alloc larger of two possible sizes */
663 if (alloc_perm_bits(perm, PCI_CAP_EXP_ENDPOINT_SIZEOF_V2))
666 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
669 * Allow writes to device control fields (includes FLR!)
670 * but not to devctl_phantom which could confuse IOMMU
671 * or to the ARI bit in devctl2 which is set at probe time
673 p_setw(perm, PCI_EXP_DEVCTL, NO_VIRT, ~PCI_EXP_DEVCTL_PHANTOM);
674 p_setw(perm, PCI_EXP_DEVCTL2, NO_VIRT, ~PCI_EXP_DEVCTL2_ARI);
678 /* Permissions for Advanced Function capability */
679 static int __init init_pci_cap_af_perm(struct perm_bits *perm)
681 if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_AF]))
684 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
685 p_setb(perm, PCI_AF_CTRL, NO_VIRT, PCI_AF_CTRL_FLR);
689 /* Permissions for Advanced Error Reporting extended capability */
690 static int __init init_pci_ext_cap_err_perm(struct perm_bits *perm)
694 if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_ERR]))
698 * Virtualize the first dword of all express capabilities
699 * because it includes the next pointer. This lets us later
700 * remove capabilities from the chain if we need to.
702 p_setd(perm, 0, ALL_VIRT, NO_WRITE);
704 /* Writable bits mask */
705 mask = PCI_ERR_UNC_TRAIN | /* Training */
706 PCI_ERR_UNC_DLP | /* Data Link Protocol */
707 PCI_ERR_UNC_SURPDN | /* Surprise Down */
708 PCI_ERR_UNC_POISON_TLP | /* Poisoned TLP */
709 PCI_ERR_UNC_FCP | /* Flow Control Protocol */
710 PCI_ERR_UNC_COMP_TIME | /* Completion Timeout */
711 PCI_ERR_UNC_COMP_ABORT | /* Completer Abort */
712 PCI_ERR_UNC_UNX_COMP | /* Unexpected Completion */
713 PCI_ERR_UNC_RX_OVER | /* Receiver Overflow */
714 PCI_ERR_UNC_MALF_TLP | /* Malformed TLP */
715 PCI_ERR_UNC_ECRC | /* ECRC Error Status */
716 PCI_ERR_UNC_UNSUP | /* Unsupported Request */
717 PCI_ERR_UNC_ACSV | /* ACS Violation */
718 PCI_ERR_UNC_INTN | /* internal error */
719 PCI_ERR_UNC_MCBTLP | /* MC blocked TLP */
720 PCI_ERR_UNC_ATOMEG | /* Atomic egress blocked */
721 PCI_ERR_UNC_TLPPRE; /* TLP prefix blocked */
722 p_setd(perm, PCI_ERR_UNCOR_STATUS, NO_VIRT, mask);
723 p_setd(perm, PCI_ERR_UNCOR_MASK, NO_VIRT, mask);
724 p_setd(perm, PCI_ERR_UNCOR_SEVER, NO_VIRT, mask);
726 mask = PCI_ERR_COR_RCVR | /* Receiver Error Status */
727 PCI_ERR_COR_BAD_TLP | /* Bad TLP Status */
728 PCI_ERR_COR_BAD_DLLP | /* Bad DLLP Status */
729 PCI_ERR_COR_REP_ROLL | /* REPLAY_NUM Rollover */
730 PCI_ERR_COR_REP_TIMER | /* Replay Timer Timeout */
731 PCI_ERR_COR_ADV_NFAT | /* Advisory Non-Fatal */
732 PCI_ERR_COR_INTERNAL | /* Corrected Internal */
733 PCI_ERR_COR_LOG_OVER; /* Header Log Overflow */
734 p_setd(perm, PCI_ERR_COR_STATUS, NO_VIRT, mask);
735 p_setd(perm, PCI_ERR_COR_MASK, NO_VIRT, mask);
737 mask = PCI_ERR_CAP_ECRC_GENE | /* ECRC Generation Enable */
738 PCI_ERR_CAP_ECRC_CHKE; /* ECRC Check Enable */
739 p_setd(perm, PCI_ERR_CAP, NO_VIRT, mask);
743 /* Permissions for Power Budgeting extended capability */
744 static int __init init_pci_ext_cap_pwr_perm(struct perm_bits *perm)
746 if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_PWR]))
749 p_setd(perm, 0, ALL_VIRT, NO_WRITE);
751 /* Writing the data selector is OK, the info is still read-only */
752 p_setb(perm, PCI_PWR_DATA, NO_VIRT, (u8)ALL_WRITE);
757 * Initialize the shared permission tables
759 void vfio_pci_uninit_perm_bits(void)
761 free_perm_bits(&cap_perms[PCI_CAP_ID_BASIC]);
763 free_perm_bits(&cap_perms[PCI_CAP_ID_PM]);
764 free_perm_bits(&cap_perms[PCI_CAP_ID_PCIX]);
765 free_perm_bits(&cap_perms[PCI_CAP_ID_EXP]);
766 free_perm_bits(&cap_perms[PCI_CAP_ID_AF]);
768 free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_ERR]);
769 free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_PWR]);
772 int __init vfio_pci_init_perm_bits(void)
776 /* Basic config space */
777 ret = init_pci_cap_basic_perm(&cap_perms[PCI_CAP_ID_BASIC]);
780 ret |= init_pci_cap_pm_perm(&cap_perms[PCI_CAP_ID_PM]);
781 cap_perms[PCI_CAP_ID_VPD].writefn = vfio_direct_config_write;
782 ret |= init_pci_cap_pcix_perm(&cap_perms[PCI_CAP_ID_PCIX]);
783 cap_perms[PCI_CAP_ID_VNDR].writefn = vfio_direct_config_write;
784 ret |= init_pci_cap_exp_perm(&cap_perms[PCI_CAP_ID_EXP]);
785 ret |= init_pci_cap_af_perm(&cap_perms[PCI_CAP_ID_AF]);
787 /* Extended capabilities */
788 ret |= init_pci_ext_cap_err_perm(&ecap_perms[PCI_EXT_CAP_ID_ERR]);
789 ret |= init_pci_ext_cap_pwr_perm(&ecap_perms[PCI_EXT_CAP_ID_PWR]);
790 ecap_perms[PCI_EXT_CAP_ID_VNDR].writefn = vfio_direct_config_write;
793 vfio_pci_uninit_perm_bits();
798 static int vfio_find_cap_start(struct vfio_pci_device *vdev, int pos)
801 int base = (pos >= PCI_CFG_SPACE_SIZE) ? PCI_CFG_SPACE_SIZE :
802 PCI_STD_HEADER_SIZEOF;
806 cap = vdev->pci_config_map[pos];
808 if (cap == PCI_CAP_ID_BASIC)
811 /* XXX Can we have to abutting capabilities of the same type? */
812 while (pos - 1 >= base && vdev->pci_config_map[pos - 1] == cap)
818 static int vfio_msi_config_read(struct vfio_pci_device *vdev, int pos,
819 int count, struct perm_bits *perm,
820 int offset, __le32 *val)
822 /* Update max available queue size from msi_qmax */
823 if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
827 start = vfio_find_cap_start(vdev, pos);
829 flags = (__le16 *)&vdev->vconfig[start];
831 *flags &= cpu_to_le16(~PCI_MSI_FLAGS_QMASK);
832 *flags |= cpu_to_le16(vdev->msi_qmax << 1);
835 return vfio_default_config_read(vdev, pos, count, perm, offset, val);
838 static int vfio_msi_config_write(struct vfio_pci_device *vdev, int pos,
839 int count, struct perm_bits *perm,
840 int offset, __le32 val)
842 count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
846 /* Fixup and write configured queue size and enable to hardware */
847 if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
852 start = vfio_find_cap_start(vdev, pos);
854 pflags = (__le16 *)&vdev->vconfig[start + PCI_MSI_FLAGS];
856 flags = le16_to_cpu(*pflags);
858 /* MSI is enabled via ioctl */
860 flags &= ~PCI_MSI_FLAGS_ENABLE;
862 /* Check queue size */
863 if ((flags & PCI_MSI_FLAGS_QSIZE) >> 4 > vdev->msi_qmax) {
864 flags &= ~PCI_MSI_FLAGS_QSIZE;
865 flags |= vdev->msi_qmax << 4;
868 /* Write back to virt and to hardware */
869 *pflags = cpu_to_le16(flags);
870 ret = pci_user_write_config_word(vdev->pdev,
871 start + PCI_MSI_FLAGS,
874 return pcibios_err_to_errno(ret);
881 * MSI determination is per-device, so this routine gets used beyond
882 * initialization time. Don't add __init
884 static int init_pci_cap_msi_perm(struct perm_bits *perm, int len, u16 flags)
886 if (alloc_perm_bits(perm, len))
889 perm->readfn = vfio_msi_config_read;
890 perm->writefn = vfio_msi_config_write;
892 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
895 * The upper byte of the control register is reserved,
896 * just setup the lower byte.
898 p_setb(perm, PCI_MSI_FLAGS, (u8)ALL_VIRT, (u8)ALL_WRITE);
899 p_setd(perm, PCI_MSI_ADDRESS_LO, ALL_VIRT, ALL_WRITE);
900 if (flags & PCI_MSI_FLAGS_64BIT) {
901 p_setd(perm, PCI_MSI_ADDRESS_HI, ALL_VIRT, ALL_WRITE);
902 p_setw(perm, PCI_MSI_DATA_64, (u16)ALL_VIRT, (u16)ALL_WRITE);
903 if (flags & PCI_MSI_FLAGS_MASKBIT) {
904 p_setd(perm, PCI_MSI_MASK_64, NO_VIRT, ALL_WRITE);
905 p_setd(perm, PCI_MSI_PENDING_64, NO_VIRT, ALL_WRITE);
908 p_setw(perm, PCI_MSI_DATA_32, (u16)ALL_VIRT, (u16)ALL_WRITE);
909 if (flags & PCI_MSI_FLAGS_MASKBIT) {
910 p_setd(perm, PCI_MSI_MASK_32, NO_VIRT, ALL_WRITE);
911 p_setd(perm, PCI_MSI_PENDING_32, NO_VIRT, ALL_WRITE);
917 /* Determine MSI CAP field length; initialize msi_perms on 1st call per vdev */
918 static int vfio_msi_cap_len(struct vfio_pci_device *vdev, u8 pos)
920 struct pci_dev *pdev = vdev->pdev;
924 ret = pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &flags);
926 return pcibios_err_to_errno(ret);
928 len = 10; /* Minimum size */
929 if (flags & PCI_MSI_FLAGS_64BIT)
931 if (flags & PCI_MSI_FLAGS_MASKBIT)
937 vdev->msi_perm = kmalloc(sizeof(struct perm_bits), GFP_KERNEL);
941 ret = init_pci_cap_msi_perm(vdev->msi_perm, len, flags);
948 /* Determine extended capability length for VC (2 & 9) and MFVC */
949 static int vfio_vc_cap_len(struct vfio_pci_device *vdev, u16 pos)
951 struct pci_dev *pdev = vdev->pdev;
953 int ret, evcc, phases, vc_arb;
954 int len = PCI_CAP_VC_BASE_SIZEOF;
956 ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_REG1, &tmp);
958 return pcibios_err_to_errno(ret);
960 evcc = tmp & PCI_VC_REG1_EVCC; /* extended vc count */
961 ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_REG2, &tmp);
963 return pcibios_err_to_errno(ret);
965 if (tmp & PCI_VC_REG2_128_PHASE)
967 else if (tmp & PCI_VC_REG2_64_PHASE)
969 else if (tmp & PCI_VC_REG2_32_PHASE)
977 * Port arbitration tables are root & switch only;
978 * function arbitration tables are function 0 only.
979 * In either case, we'll never let user write them so
980 * we don't care how big they are
982 len += (1 + evcc) * PCI_CAP_VC_PER_VC_SIZEOF;
984 len = round_up(len, 16);
990 static int vfio_cap_len(struct vfio_pci_device *vdev, u8 cap, u8 pos)
992 struct pci_dev *pdev = vdev->pdev;
999 return vfio_msi_cap_len(vdev, pos);
1000 case PCI_CAP_ID_PCIX:
1001 ret = pci_read_config_word(pdev, pos + PCI_X_CMD, &word);
1003 return pcibios_err_to_errno(ret);
1005 if (PCI_X_CMD_VERSION(word)) {
1006 vdev->extended_caps = true;
1007 return PCI_CAP_PCIX_SIZEOF_V2;
1009 return PCI_CAP_PCIX_SIZEOF_V0;
1010 case PCI_CAP_ID_VNDR:
1011 /* length follows next field */
1012 ret = pci_read_config_byte(pdev, pos + PCI_CAP_FLAGS, &byte);
1014 return pcibios_err_to_errno(ret);
1017 case PCI_CAP_ID_EXP:
1018 /* length based on version */
1019 ret = pci_read_config_word(pdev, pos + PCI_EXP_FLAGS, &word);
1021 return pcibios_err_to_errno(ret);
1023 vdev->extended_caps = true;
1025 if ((word & PCI_EXP_FLAGS_VERS) == 1)
1026 return PCI_CAP_EXP_ENDPOINT_SIZEOF_V1;
1028 return PCI_CAP_EXP_ENDPOINT_SIZEOF_V2;
1030 ret = pci_read_config_byte(pdev, pos + 3, &byte);
1032 return pcibios_err_to_errno(ret);
1034 return (byte & HT_3BIT_CAP_MASK) ?
1035 HT_CAP_SIZEOF_SHORT : HT_CAP_SIZEOF_LONG;
1036 case PCI_CAP_ID_SATA:
1037 ret = pci_read_config_byte(pdev, pos + PCI_SATA_REGS, &byte);
1039 return pcibios_err_to_errno(ret);
1041 byte &= PCI_SATA_REGS_MASK;
1042 if (byte == PCI_SATA_REGS_INLINE)
1043 return PCI_SATA_SIZEOF_LONG;
1045 return PCI_SATA_SIZEOF_SHORT;
1047 pr_warn("%s: %s unknown length for pci cap 0x%x@0x%x\n",
1048 dev_name(&pdev->dev), __func__, cap, pos);
1054 static int vfio_ext_cap_len(struct vfio_pci_device *vdev, u16 ecap, u16 epos)
1056 struct pci_dev *pdev = vdev->pdev;
1062 case PCI_EXT_CAP_ID_VNDR:
1063 ret = pci_read_config_dword(pdev, epos + PCI_VSEC_HDR, &dword);
1065 return pcibios_err_to_errno(ret);
1067 return dword >> PCI_VSEC_HDR_LEN_SHIFT;
1068 case PCI_EXT_CAP_ID_VC:
1069 case PCI_EXT_CAP_ID_VC9:
1070 case PCI_EXT_CAP_ID_MFVC:
1071 return vfio_vc_cap_len(vdev, epos);
1072 case PCI_EXT_CAP_ID_ACS:
1073 ret = pci_read_config_byte(pdev, epos + PCI_ACS_CAP, &byte);
1075 return pcibios_err_to_errno(ret);
1077 if (byte & PCI_ACS_EC) {
1080 ret = pci_read_config_byte(pdev,
1081 epos + PCI_ACS_EGRESS_BITS,
1084 return pcibios_err_to_errno(ret);
1086 bits = byte ? round_up(byte, 32) : 256;
1087 return 8 + (bits / 8);
1091 case PCI_EXT_CAP_ID_REBAR:
1092 ret = pci_read_config_byte(pdev, epos + PCI_REBAR_CTRL, &byte);
1094 return pcibios_err_to_errno(ret);
1096 byte &= PCI_REBAR_CTRL_NBAR_MASK;
1097 byte >>= PCI_REBAR_CTRL_NBAR_SHIFT;
1099 return 4 + (byte * 8);
1100 case PCI_EXT_CAP_ID_DPA:
1101 ret = pci_read_config_byte(pdev, epos + PCI_DPA_CAP, &byte);
1103 return pcibios_err_to_errno(ret);
1105 byte &= PCI_DPA_CAP_SUBSTATE_MASK;
1106 byte = round_up(byte + 1, 4);
1107 return PCI_DPA_BASE_SIZEOF + byte;
1108 case PCI_EXT_CAP_ID_TPH:
1109 ret = pci_read_config_dword(pdev, epos + PCI_TPH_CAP, &dword);
1111 return pcibios_err_to_errno(ret);
1113 if ((dword & PCI_TPH_CAP_LOC_MASK) == PCI_TPH_LOC_CAP) {
1116 sts = byte & PCI_TPH_CAP_ST_MASK;
1117 sts >>= PCI_TPH_CAP_ST_SHIFT;
1118 return PCI_TPH_BASE_SIZEOF + round_up(sts * 2, 4);
1120 return PCI_TPH_BASE_SIZEOF;
1122 pr_warn("%s: %s unknown length for pci ecap 0x%x@0x%x\n",
1123 dev_name(&pdev->dev), __func__, ecap, epos);
1129 static int vfio_fill_vconfig_bytes(struct vfio_pci_device *vdev,
1130 int offset, int size)
1132 struct pci_dev *pdev = vdev->pdev;
1136 * We try to read physical config space in the largest chunks
1137 * we can, assuming that all of the fields support dword access.
1138 * pci_save_state() makes this same assumption and seems to do ok.
1143 if (size >= 4 && !(offset % 4)) {
1144 __le32 *dwordp = (__le32 *)&vdev->vconfig[offset];
1147 ret = pci_read_config_dword(pdev, offset, &dword);
1150 *dwordp = cpu_to_le32(dword);
1152 } else if (size >= 2 && !(offset % 2)) {
1153 __le16 *wordp = (__le16 *)&vdev->vconfig[offset];
1156 ret = pci_read_config_word(pdev, offset, &word);
1159 *wordp = cpu_to_le16(word);
1162 u8 *byte = &vdev->vconfig[offset];
1163 ret = pci_read_config_byte(pdev, offset, byte);
1176 static int vfio_cap_init(struct vfio_pci_device *vdev)
1178 struct pci_dev *pdev = vdev->pdev;
1179 u8 *map = vdev->pci_config_map;
1182 int loops, ret, caps = 0;
1184 /* Any capabilities? */
1185 ret = pci_read_config_word(pdev, PCI_STATUS, &status);
1189 if (!(status & PCI_STATUS_CAP_LIST))
1190 return 0; /* Done */
1192 ret = pci_read_config_byte(pdev, PCI_CAPABILITY_LIST, &pos);
1196 /* Mark the previous position in case we want to skip a capability */
1197 prev = &vdev->vconfig[PCI_CAPABILITY_LIST];
1199 /* We can bound our loop, capabilities are dword aligned */
1200 loops = (PCI_CFG_SPACE_SIZE - PCI_STD_HEADER_SIZEOF) / PCI_CAP_SIZEOF;
1201 while (pos && loops--) {
1205 ret = pci_read_config_byte(pdev, pos, &cap);
1209 ret = pci_read_config_byte(pdev,
1210 pos + PCI_CAP_LIST_NEXT, &next);
1214 if (cap <= PCI_CAP_ID_MAX) {
1215 len = pci_cap_length[cap];
1216 if (len == 0xFF) { /* Variable length */
1217 len = vfio_cap_len(vdev, cap, pos);
1224 pr_info("%s: %s hiding cap 0x%x\n",
1225 __func__, dev_name(&pdev->dev), cap);
1231 /* Sanity check, do we overlap other capabilities? */
1232 for (i = 0; i < len; i += 4) {
1233 if (likely(map[(pos + i) / 4] == PCI_CAP_ID_INVALID))
1236 pr_warn("%s: %s pci config conflict @0x%x, was cap 0x%x now cap 0x%x\n",
1237 __func__, dev_name(&pdev->dev),
1238 pos + i, map[pos + i], cap);
1241 memset(map + (pos / 4), cap, len / 4);
1242 ret = vfio_fill_vconfig_bytes(vdev, pos, len);
1246 prev = &vdev->vconfig[pos + PCI_CAP_LIST_NEXT];
1251 /* If we didn't fill any capabilities, clear the status flag */
1253 __le16 *vstatus = (__le16 *)&vdev->vconfig[PCI_STATUS];
1254 *vstatus &= ~cpu_to_le16(PCI_STATUS_CAP_LIST);
1260 static int vfio_ecap_init(struct vfio_pci_device *vdev)
1262 struct pci_dev *pdev = vdev->pdev;
1263 u8 *map = vdev->pci_config_map;
1265 __le32 *prev = NULL;
1266 int loops, ret, ecaps = 0;
1268 if (!vdev->extended_caps)
1271 epos = PCI_CFG_SPACE_SIZE;
1273 loops = (pdev->cfg_size - PCI_CFG_SPACE_SIZE) / PCI_CAP_SIZEOF;
1275 while (loops-- && epos >= PCI_CFG_SPACE_SIZE) {
1279 bool hidden = false;
1281 ret = pci_read_config_dword(pdev, epos, &header);
1285 ecap = PCI_EXT_CAP_ID(header);
1287 if (ecap <= PCI_EXT_CAP_ID_MAX) {
1288 len = pci_ext_cap_length[ecap];
1290 len = vfio_ext_cap_len(vdev, ecap, epos);
1297 pr_info("%s: %s hiding ecap 0x%x@0x%x\n",
1298 __func__, dev_name(&pdev->dev), ecap, epos);
1300 /* If not the first in the chain, we can skip over it */
1302 u32 val = epos = PCI_EXT_CAP_NEXT(header);
1303 *prev &= cpu_to_le32(~(0xffcU << 20));
1304 *prev |= cpu_to_le32(val << 20);
1309 * Otherwise, fill in a placeholder, the direct
1310 * readfn will virtualize this automatically
1312 len = PCI_CAP_SIZEOF;
1316 for (i = 0; i < len; i += 4) {
1317 if (likely(map[(epos + i) / 4] == PCI_CAP_ID_INVALID))
1320 pr_warn("%s: %s pci config conflict @0x%x, was ecap 0x%x now ecap 0x%x\n",
1321 __func__, dev_name(&pdev->dev),
1322 epos + i, map[epos + i], ecap);
1326 * Even though ecap is 2 bytes, we're currently a long way
1327 * from exceeding 1 byte capabilities. If we ever make it
1328 * up to 0xFF we'll need to up this to a two-byte, byte map.
1330 BUILD_BUG_ON(PCI_EXT_CAP_ID_MAX >= PCI_CAP_ID_INVALID);
1332 memset(map + (epos / 4), ecap, len / 4);
1333 ret = vfio_fill_vconfig_bytes(vdev, epos, len);
1338 * If we're just using this capability to anchor the list,
1339 * hide the real ID. Only count real ecaps. XXX PCI spec
1340 * indicates to use cap id = 0, version = 0, next = 0 if
1341 * ecaps are absent, hope users check all the way to next.
1344 *(__le32 *)&vdev->vconfig[epos] &=
1345 cpu_to_le32((0xffcU << 20));
1349 prev = (__le32 *)&vdev->vconfig[epos];
1350 epos = PCI_EXT_CAP_NEXT(header);
1354 *(u32 *)&vdev->vconfig[PCI_CFG_SPACE_SIZE] = 0;
1360 * For each device we allocate a pci_config_map that indicates the
1361 * capability occupying each dword and thus the struct perm_bits we
1362 * use for read and write. We also allocate a virtualized config
1363 * space which tracks reads and writes to bits that we emulate for
1364 * the user. Initial values filled from device.
1366 * Using shared stuct perm_bits between all vfio-pci devices saves
1367 * us from allocating cfg_size buffers for virt and write for every
1368 * device. We could remove vconfig and allocate individual buffers
1369 * for each area requring emulated bits, but the array of pointers
1370 * would be comparable in size (at least for standard config space).
1372 int vfio_config_init(struct vfio_pci_device *vdev)
1374 struct pci_dev *pdev = vdev->pdev;
1379 * Config space, caps and ecaps are all dword aligned, so we can
1380 * use one byte per dword to record the type.
1382 map = kmalloc(pdev->cfg_size / 4, GFP_KERNEL);
1386 vconfig = kmalloc(pdev->cfg_size, GFP_KERNEL);
1392 vdev->pci_config_map = map;
1393 vdev->vconfig = vconfig;
1395 memset(map, PCI_CAP_ID_BASIC, PCI_STD_HEADER_SIZEOF / 4);
1396 memset(map + (PCI_STD_HEADER_SIZEOF / 4), PCI_CAP_ID_INVALID,
1397 (pdev->cfg_size - PCI_STD_HEADER_SIZEOF) / 4);
1399 ret = vfio_fill_vconfig_bytes(vdev, 0, PCI_STD_HEADER_SIZEOF);
1403 vdev->bardirty = true;
1406 * XXX can we just pci_load_saved_state/pci_restore_state?
1407 * may need to rebuild vconfig after that
1410 /* For restore after reset */
1411 vdev->rbar[0] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_0]);
1412 vdev->rbar[1] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_1]);
1413 vdev->rbar[2] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_2]);
1414 vdev->rbar[3] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_3]);
1415 vdev->rbar[4] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_4]);
1416 vdev->rbar[5] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_5]);
1417 vdev->rbar[6] = le32_to_cpu(*(__le32 *)&vconfig[PCI_ROM_ADDRESS]);
1419 if (pdev->is_virtfn) {
1420 *(__le16 *)&vconfig[PCI_VENDOR_ID] = cpu_to_le16(pdev->vendor);
1421 *(__le16 *)&vconfig[PCI_DEVICE_ID] = cpu_to_le16(pdev->device);
1424 ret = vfio_cap_init(vdev);
1428 ret = vfio_ecap_init(vdev);
1436 vdev->pci_config_map = NULL;
1438 vdev->vconfig = NULL;
1439 return pcibios_err_to_errno(ret);
1442 void vfio_config_free(struct vfio_pci_device *vdev)
1444 kfree(vdev->vconfig);
1445 vdev->vconfig = NULL;
1446 kfree(vdev->pci_config_map);
1447 vdev->pci_config_map = NULL;
1448 kfree(vdev->msi_perm);
1449 vdev->msi_perm = NULL;
1452 static ssize_t vfio_config_do_rw(struct vfio_pci_device *vdev, char __user *buf,
1453 size_t count, loff_t *ppos, bool iswrite)
1455 struct pci_dev *pdev = vdev->pdev;
1456 struct perm_bits *perm;
1458 int cap_start = 0, offset;
1460 ssize_t ret = count;
1462 if (*ppos < 0 || *ppos + count > pdev->cfg_size)
1466 * gcc can't seem to figure out we're a static function, only called
1467 * with count of 1/2/4 and hits copy_from_user_overflow without this.
1469 if (count > sizeof(val))
1472 cap_id = vdev->pci_config_map[*ppos / 4];
1474 if (cap_id == PCI_CAP_ID_INVALID) {
1476 return ret; /* drop */
1479 * Per PCI spec 3.0, section 6.1, reads from reserved and
1480 * unimplemented registers return 0
1482 if (copy_to_user(buf, &val, count))
1489 * All capabilities are minimum 4 bytes and aligned on dword
1490 * boundaries. Since we don't support unaligned accesses, we're
1491 * only ever accessing a single capability.
1493 if (*ppos >= PCI_CFG_SPACE_SIZE) {
1494 WARN_ON(cap_id > PCI_EXT_CAP_ID_MAX);
1496 perm = &ecap_perms[cap_id];
1497 cap_start = vfio_find_cap_start(vdev, *ppos);
1500 WARN_ON(cap_id > PCI_CAP_ID_MAX);
1502 perm = &cap_perms[cap_id];
1504 if (cap_id == PCI_CAP_ID_MSI)
1505 perm = vdev->msi_perm;
1507 if (cap_id > PCI_CAP_ID_BASIC)
1508 cap_start = vfio_find_cap_start(vdev, *ppos);
1511 WARN_ON(!cap_start && cap_id != PCI_CAP_ID_BASIC);
1512 WARN_ON(cap_start > *ppos);
1514 offset = *ppos - cap_start;
1520 if (copy_from_user(&val, buf, count))
1523 ret = perm->writefn(vdev, *ppos, count, perm, offset, val);
1526 ret = perm->readfn(vdev, *ppos, count,
1527 perm, offset, &val);
1532 if (copy_to_user(buf, &val, count))
1539 ssize_t vfio_pci_config_rw(struct vfio_pci_device *vdev, char __user *buf,
1540 size_t count, loff_t *ppos, bool iswrite)
1546 pos &= VFIO_PCI_OFFSET_MASK;
1549 * We want to both keep the access size the caller users as well as
1550 * support reading large chunks of config space in a single call.
1551 * PCI doesn't support unaligned accesses, so we can safely break
1555 if (count >= 4 && !(pos % 4))
1556 ret = vfio_config_do_rw(vdev, buf, 4, &pos, iswrite);
1557 else if (count >= 2 && !(pos % 2))
1558 ret = vfio_config_do_rw(vdev, buf, 2, &pos, iswrite);
1560 ret = vfio_config_do_rw(vdev, buf, 1, &pos, iswrite);