2 * VFIO PCI config space virtualization
4 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
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.
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>
30 #include <linux/slab.h>
32 #include "vfio_pci_private.h"
34 #define PCI_CFG_SPACE_SIZE 256
36 /* Useful "pseudo" capabilities */
37 #define PCI_CAP_ID_BASIC 0
38 #define PCI_CAP_ID_INVALID 0xFF
40 #define is_bar(offset) \
41 ((offset >= PCI_BASE_ADDRESS_0 && offset < PCI_BASE_ADDRESS_5 + 4) || \
42 (offset >= PCI_ROM_ADDRESS && offset < PCI_ROM_ADDRESS + 4))
45 * Lengths of PCI Config Capabilities
46 * 0: Removed from the user visible capability list
49 static u8 pci_cap_length[] = {
50 [PCI_CAP_ID_BASIC] = PCI_STD_HEADER_SIZEOF, /* pci config header */
51 [PCI_CAP_ID_PM] = PCI_PM_SIZEOF,
52 [PCI_CAP_ID_AGP] = PCI_AGP_SIZEOF,
53 [PCI_CAP_ID_VPD] = PCI_CAP_VPD_SIZEOF,
54 [PCI_CAP_ID_SLOTID] = 0, /* bridge - don't care */
55 [PCI_CAP_ID_MSI] = 0xFF, /* 10, 14, 20, or 24 */
56 [PCI_CAP_ID_CHSWP] = 0, /* cpci - not yet */
57 [PCI_CAP_ID_PCIX] = 0xFF, /* 8 or 24 */
58 [PCI_CAP_ID_HT] = 0xFF, /* hypertransport */
59 [PCI_CAP_ID_VNDR] = 0xFF, /* variable */
60 [PCI_CAP_ID_DBG] = 0, /* debug - don't care */
61 [PCI_CAP_ID_CCRC] = 0, /* cpci - not yet */
62 [PCI_CAP_ID_SHPC] = 0, /* hotswap - not yet */
63 [PCI_CAP_ID_SSVID] = 0, /* bridge - don't care */
64 [PCI_CAP_ID_AGP3] = 0, /* AGP8x - not yet */
65 [PCI_CAP_ID_SECDEV] = 0, /* secure device not yet */
66 [PCI_CAP_ID_EXP] = 0xFF, /* 20 or 44 */
67 [PCI_CAP_ID_MSIX] = PCI_CAP_MSIX_SIZEOF,
68 [PCI_CAP_ID_SATA] = 0xFF,
69 [PCI_CAP_ID_AF] = PCI_CAP_AF_SIZEOF,
73 * Lengths of PCIe/PCI-X Extended Config Capabilities
74 * 0: Removed or masked from the user visible capabilty list
77 static u16 pci_ext_cap_length[] = {
78 [PCI_EXT_CAP_ID_ERR] = PCI_ERR_ROOT_COMMAND,
79 [PCI_EXT_CAP_ID_VC] = 0xFF,
80 [PCI_EXT_CAP_ID_DSN] = PCI_EXT_CAP_DSN_SIZEOF,
81 [PCI_EXT_CAP_ID_PWR] = PCI_EXT_CAP_PWR_SIZEOF,
82 [PCI_EXT_CAP_ID_RCLD] = 0, /* root only - don't care */
83 [PCI_EXT_CAP_ID_RCILC] = 0, /* root only - don't care */
84 [PCI_EXT_CAP_ID_RCEC] = 0, /* root only - don't care */
85 [PCI_EXT_CAP_ID_MFVC] = 0xFF,
86 [PCI_EXT_CAP_ID_VC9] = 0xFF, /* same as CAP_ID_VC */
87 [PCI_EXT_CAP_ID_RCRB] = 0, /* root only - don't care */
88 [PCI_EXT_CAP_ID_VNDR] = 0xFF,
89 [PCI_EXT_CAP_ID_CAC] = 0, /* obsolete */
90 [PCI_EXT_CAP_ID_ACS] = 0xFF,
91 [PCI_EXT_CAP_ID_ARI] = PCI_EXT_CAP_ARI_SIZEOF,
92 [PCI_EXT_CAP_ID_ATS] = PCI_EXT_CAP_ATS_SIZEOF,
93 [PCI_EXT_CAP_ID_SRIOV] = PCI_EXT_CAP_SRIOV_SIZEOF,
94 [PCI_EXT_CAP_ID_MRIOV] = 0, /* not yet */
95 [PCI_EXT_CAP_ID_MCAST] = PCI_EXT_CAP_MCAST_ENDPOINT_SIZEOF,
96 [PCI_EXT_CAP_ID_PRI] = PCI_EXT_CAP_PRI_SIZEOF,
97 [PCI_EXT_CAP_ID_AMD_XXX] = 0, /* not yet */
98 [PCI_EXT_CAP_ID_REBAR] = 0xFF,
99 [PCI_EXT_CAP_ID_DPA] = 0xFF,
100 [PCI_EXT_CAP_ID_TPH] = 0xFF,
101 [PCI_EXT_CAP_ID_LTR] = PCI_EXT_CAP_LTR_SIZEOF,
102 [PCI_EXT_CAP_ID_SECPCI] = 0, /* not yet */
103 [PCI_EXT_CAP_ID_PMUX] = 0, /* not yet */
104 [PCI_EXT_CAP_ID_PASID] = 0, /* not yet */
108 * Read/Write Permission Bits - one bit for each bit in capability
109 * Any field can be read if it exists, but what is read depends on
110 * whether the field is 'virtualized', or just pass thru to the
111 * hardware. Any virtualized field is also virtualized for writes.
112 * Writes are only permitted if they have a 1 bit here.
115 u8 *virt; /* read/write virtual data, not hw */
116 u8 *write; /* writeable bits */
117 int (*readfn)(struct vfio_pci_device *vdev, int pos, int count,
118 struct perm_bits *perm, int offset, __le32 *val);
119 int (*writefn)(struct vfio_pci_device *vdev, int pos, int count,
120 struct perm_bits *perm, int offset, __le32 val);
124 #define ALL_VIRT 0xFFFFFFFFU
126 #define ALL_WRITE 0xFFFFFFFFU
128 static int vfio_user_config_read(struct pci_dev *pdev, int offset,
129 __le32 *val, int count)
138 ret = pci_user_read_config_byte(pdev, offset, &tmp);
145 ret = pci_user_read_config_word(pdev, offset, &tmp);
150 ret = pci_user_read_config_dword(pdev, offset, &tmp_val);
154 *val = cpu_to_le32(tmp_val);
156 return pcibios_err_to_errno(ret);
159 static int vfio_user_config_write(struct pci_dev *pdev, int offset,
160 __le32 val, int count)
163 u32 tmp_val = le32_to_cpu(val);
167 ret = pci_user_write_config_byte(pdev, offset, tmp_val);
170 ret = pci_user_write_config_word(pdev, offset, tmp_val);
173 ret = pci_user_write_config_dword(pdev, offset, tmp_val);
177 return pcibios_err_to_errno(ret);
180 static int vfio_default_config_read(struct vfio_pci_device *vdev, int pos,
181 int count, struct perm_bits *perm,
182 int offset, __le32 *val)
186 memcpy(val, vdev->vconfig + pos, count);
188 memcpy(&virt, perm->virt + offset, count);
190 /* Any non-virtualized bits? */
191 if (cpu_to_le32(~0U >> (32 - (count * 8))) != virt) {
192 struct pci_dev *pdev = vdev->pdev;
196 ret = vfio_user_config_read(pdev, pos, &phys_val, count);
200 *val = (phys_val & ~virt) | (*val & virt);
206 static int vfio_default_config_write(struct vfio_pci_device *vdev, int pos,
207 int count, struct perm_bits *perm,
208 int offset, __le32 val)
210 __le32 virt = 0, write = 0;
212 memcpy(&write, perm->write + offset, count);
215 return count; /* drop, no writable bits */
217 memcpy(&virt, perm->virt + offset, count);
219 /* Virtualized and writable bits go to vconfig */
223 memcpy(&virt_val, vdev->vconfig + pos, count);
225 virt_val &= ~(write & virt);
226 virt_val |= (val & (write & virt));
228 memcpy(vdev->vconfig + pos, &virt_val, count);
231 /* Non-virtualzed and writable bits go to hardware */
233 struct pci_dev *pdev = vdev->pdev;
237 ret = vfio_user_config_read(pdev, pos, &phys_val, count);
241 phys_val &= ~(write & ~virt);
242 phys_val |= (val & (write & ~virt));
244 ret = vfio_user_config_write(pdev, pos, phys_val, count);
252 /* Allow direct read from hardware, except for capability next pointer */
253 static int vfio_direct_config_read(struct vfio_pci_device *vdev, int pos,
254 int count, struct perm_bits *perm,
255 int offset, __le32 *val)
259 ret = vfio_user_config_read(vdev->pdev, pos, val, count);
261 return pcibios_err_to_errno(ret);
263 if (pos >= PCI_CFG_SPACE_SIZE) { /* Extended cap header mangling */
265 memcpy(val, vdev->vconfig + pos, count);
266 } else if (pos >= PCI_STD_HEADER_SIZEOF) { /* Std cap mangling */
267 if (offset == PCI_CAP_LIST_ID && count > 1)
268 memcpy(val, vdev->vconfig + pos,
269 min(PCI_CAP_FLAGS, count));
270 else if (offset == PCI_CAP_LIST_NEXT)
271 memcpy(val, vdev->vconfig + pos, 1);
277 /* Raw access skips any kind of virtualization */
278 static int vfio_raw_config_write(struct vfio_pci_device *vdev, int pos,
279 int count, struct perm_bits *perm,
280 int offset, __le32 val)
284 ret = vfio_user_config_write(vdev->pdev, pos, val, count);
291 static int vfio_raw_config_read(struct vfio_pci_device *vdev, int pos,
292 int count, struct perm_bits *perm,
293 int offset, __le32 *val)
297 ret = vfio_user_config_read(vdev->pdev, pos, val, count);
299 return pcibios_err_to_errno(ret);
304 /* Default capability regions to read-only, no-virtualization */
305 static struct perm_bits cap_perms[PCI_CAP_ID_MAX + 1] = {
306 [0 ... PCI_CAP_ID_MAX] = { .readfn = vfio_direct_config_read }
308 static struct perm_bits ecap_perms[PCI_EXT_CAP_ID_MAX + 1] = {
309 [0 ... PCI_EXT_CAP_ID_MAX] = { .readfn = vfio_direct_config_read }
312 * Default unassigned regions to raw read-write access. Some devices
313 * require this to function as they hide registers between the gaps in
314 * config space (be2net). Like MMIO and I/O port registers, we have
315 * to trust the hardware isolation.
317 static struct perm_bits unassigned_perms = {
318 .readfn = vfio_raw_config_read,
319 .writefn = vfio_raw_config_write
322 static void free_perm_bits(struct perm_bits *perm)
330 static int alloc_perm_bits(struct perm_bits *perm, int size)
333 * Round up all permission bits to the next dword, this lets us
334 * ignore whether a read/write exceeds the defined capability
335 * structure. We can do this because:
336 * - Standard config space is already dword aligned
337 * - Capabilities are all dword alinged (bits 0:1 of next reserved)
338 * - Express capabilities defined as dword aligned
340 size = round_up(size, 4);
344 * - All Readable, None Writeable, None Virtualized
346 perm->virt = kzalloc(size, GFP_KERNEL);
347 perm->write = kzalloc(size, GFP_KERNEL);
348 if (!perm->virt || !perm->write) {
349 free_perm_bits(perm);
353 perm->readfn = vfio_default_config_read;
354 perm->writefn = vfio_default_config_write;
360 * Helper functions for filling in permission tables
362 static inline void p_setb(struct perm_bits *p, int off, u8 virt, u8 write)
365 p->write[off] = write;
368 /* Handle endian-ness - pci and tables are little-endian */
369 static inline void p_setw(struct perm_bits *p, int off, u16 virt, u16 write)
371 *(__le16 *)(&p->virt[off]) = cpu_to_le16(virt);
372 *(__le16 *)(&p->write[off]) = cpu_to_le16(write);
375 /* Handle endian-ness - pci and tables are little-endian */
376 static inline void p_setd(struct perm_bits *p, int off, u32 virt, u32 write)
378 *(__le32 *)(&p->virt[off]) = cpu_to_le32(virt);
379 *(__le32 *)(&p->write[off]) = cpu_to_le32(write);
383 * Restore the *real* BARs after we detect a FLR or backdoor reset.
384 * (backdoor = some device specific technique that we didn't catch)
386 static void vfio_bar_restore(struct vfio_pci_device *vdev)
388 struct pci_dev *pdev = vdev->pdev;
389 u32 *rbar = vdev->rbar;
395 pr_info("%s: %s reset recovery - restoring bars\n",
396 __func__, dev_name(&pdev->dev));
398 for (i = PCI_BASE_ADDRESS_0; i <= PCI_BASE_ADDRESS_5; i += 4, rbar++)
399 pci_user_write_config_dword(pdev, i, *rbar);
401 pci_user_write_config_dword(pdev, PCI_ROM_ADDRESS, *rbar);
404 static __le32 vfio_generate_bar_flags(struct pci_dev *pdev, int bar)
406 unsigned long flags = pci_resource_flags(pdev, bar);
409 if (flags & IORESOURCE_IO)
410 return cpu_to_le32(PCI_BASE_ADDRESS_SPACE_IO);
412 val = PCI_BASE_ADDRESS_SPACE_MEMORY;
414 if (flags & IORESOURCE_PREFETCH)
415 val |= PCI_BASE_ADDRESS_MEM_PREFETCH;
417 if (flags & IORESOURCE_MEM_64)
418 val |= PCI_BASE_ADDRESS_MEM_TYPE_64;
420 return cpu_to_le32(val);
424 * Pretend we're hardware and tweak the values of the *virtual* PCI BARs
425 * to reflect the hardware capabilities. This implements BAR sizing.
427 static void vfio_bar_fixup(struct vfio_pci_device *vdev)
429 struct pci_dev *pdev = vdev->pdev;
434 bar = (__le32 *)&vdev->vconfig[PCI_BASE_ADDRESS_0];
436 for (i = PCI_STD_RESOURCES; i <= PCI_STD_RESOURCE_END; i++, bar++) {
437 if (!pci_resource_start(pdev, i)) {
438 *bar = 0; /* Unmapped by host = unimplemented to user */
442 mask = ~(pci_resource_len(pdev, i) - 1);
444 *bar &= cpu_to_le32((u32)mask);
445 *bar |= vfio_generate_bar_flags(pdev, i);
447 if (*bar & cpu_to_le32(PCI_BASE_ADDRESS_MEM_TYPE_64)) {
449 *bar &= cpu_to_le32((u32)(mask >> 32));
454 bar = (__le32 *)&vdev->vconfig[PCI_ROM_ADDRESS];
457 * NB. we expose the actual BAR size here, regardless of whether
458 * we can read it. When we report the REGION_INFO for the ROM
459 * we report what PCI tells us is the actual ROM size.
461 if (pci_resource_start(pdev, PCI_ROM_RESOURCE)) {
462 mask = ~(pci_resource_len(pdev, PCI_ROM_RESOURCE) - 1);
463 mask |= PCI_ROM_ADDRESS_ENABLE;
464 *bar &= cpu_to_le32((u32)mask);
468 vdev->bardirty = false;
471 static int vfio_basic_config_read(struct vfio_pci_device *vdev, int pos,
472 int count, struct perm_bits *perm,
473 int offset, __le32 *val)
475 if (is_bar(offset)) /* pos == offset for basic config */
476 vfio_bar_fixup(vdev);
478 count = vfio_default_config_read(vdev, pos, count, perm, offset, val);
480 /* Mask in virtual memory enable for SR-IOV devices */
481 if (offset == PCI_COMMAND && vdev->pdev->is_virtfn) {
482 u16 cmd = le16_to_cpu(*(__le16 *)&vdev->vconfig[PCI_COMMAND]);
483 u32 tmp_val = le32_to_cpu(*val);
485 tmp_val |= cmd & PCI_COMMAND_MEMORY;
486 *val = cpu_to_le32(tmp_val);
492 static int vfio_basic_config_write(struct vfio_pci_device *vdev, int pos,
493 int count, struct perm_bits *perm,
494 int offset, __le32 val)
496 struct pci_dev *pdev = vdev->pdev;
501 virt_cmd = (__le16 *)&vdev->vconfig[PCI_COMMAND];
503 if (offset == PCI_COMMAND) {
504 bool phys_mem, virt_mem, new_mem, phys_io, virt_io, new_io;
507 ret = pci_user_read_config_word(pdev, PCI_COMMAND, &phys_cmd);
511 new_cmd = le32_to_cpu(val);
513 phys_mem = !!(phys_cmd & PCI_COMMAND_MEMORY);
514 virt_mem = !!(le16_to_cpu(*virt_cmd) & PCI_COMMAND_MEMORY);
515 new_mem = !!(new_cmd & PCI_COMMAND_MEMORY);
517 phys_io = !!(phys_cmd & PCI_COMMAND_IO);
518 virt_io = !!(le16_to_cpu(*virt_cmd) & PCI_COMMAND_IO);
519 new_io = !!(new_cmd & PCI_COMMAND_IO);
522 * If the user is writing mem/io enable (new_mem/io) and we
523 * think it's already enabled (virt_mem/io), but the hardware
524 * shows it disabled (phys_mem/io, then the device has
525 * undergone some kind of backdoor reset and needs to be
526 * restored before we allow it to enable the bars.
527 * SR-IOV devices will trigger this, but we catch them later
529 if ((new_mem && virt_mem && !phys_mem) ||
530 (new_io && virt_io && !phys_io))
531 vfio_bar_restore(vdev);
534 count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
539 * Save current memory/io enable bits in vconfig to allow for
540 * the test above next time.
542 if (offset == PCI_COMMAND) {
543 u16 mask = PCI_COMMAND_MEMORY | PCI_COMMAND_IO;
545 *virt_cmd &= cpu_to_le16(~mask);
546 *virt_cmd |= cpu_to_le16(new_cmd & mask);
549 /* Emulate INTx disable */
550 if (offset >= PCI_COMMAND && offset <= PCI_COMMAND + 1) {
551 bool virt_intx_disable;
553 virt_intx_disable = !!(le16_to_cpu(*virt_cmd) &
554 PCI_COMMAND_INTX_DISABLE);
556 if (virt_intx_disable && !vdev->virq_disabled) {
557 vdev->virq_disabled = true;
558 vfio_pci_intx_mask(vdev);
559 } else if (!virt_intx_disable && vdev->virq_disabled) {
560 vdev->virq_disabled = false;
561 vfio_pci_intx_unmask(vdev);
566 vdev->bardirty = true;
571 /* Permissions for the Basic PCI Header */
572 static int __init init_pci_cap_basic_perm(struct perm_bits *perm)
574 if (alloc_perm_bits(perm, PCI_STD_HEADER_SIZEOF))
577 perm->readfn = vfio_basic_config_read;
578 perm->writefn = vfio_basic_config_write;
580 /* Virtualized for SR-IOV functions, which just have FFFF */
581 p_setw(perm, PCI_VENDOR_ID, (u16)ALL_VIRT, NO_WRITE);
582 p_setw(perm, PCI_DEVICE_ID, (u16)ALL_VIRT, NO_WRITE);
585 * Virtualize INTx disable, we use it internally for interrupt
586 * control and can emulate it for non-PCI 2.3 devices.
588 p_setw(perm, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE, (u16)ALL_WRITE);
590 /* Virtualize capability list, we might want to skip/disable */
591 p_setw(perm, PCI_STATUS, PCI_STATUS_CAP_LIST, NO_WRITE);
593 /* No harm to write */
594 p_setb(perm, PCI_CACHE_LINE_SIZE, NO_VIRT, (u8)ALL_WRITE);
595 p_setb(perm, PCI_LATENCY_TIMER, NO_VIRT, (u8)ALL_WRITE);
596 p_setb(perm, PCI_BIST, NO_VIRT, (u8)ALL_WRITE);
598 /* Virtualize all bars, can't touch the real ones */
599 p_setd(perm, PCI_BASE_ADDRESS_0, ALL_VIRT, ALL_WRITE);
600 p_setd(perm, PCI_BASE_ADDRESS_1, ALL_VIRT, ALL_WRITE);
601 p_setd(perm, PCI_BASE_ADDRESS_2, ALL_VIRT, ALL_WRITE);
602 p_setd(perm, PCI_BASE_ADDRESS_3, ALL_VIRT, ALL_WRITE);
603 p_setd(perm, PCI_BASE_ADDRESS_4, ALL_VIRT, ALL_WRITE);
604 p_setd(perm, PCI_BASE_ADDRESS_5, ALL_VIRT, ALL_WRITE);
605 p_setd(perm, PCI_ROM_ADDRESS, ALL_VIRT, ALL_WRITE);
607 /* Allow us to adjust capability chain */
608 p_setb(perm, PCI_CAPABILITY_LIST, (u8)ALL_VIRT, NO_WRITE);
610 /* Sometimes used by sw, just virtualize */
611 p_setb(perm, PCI_INTERRUPT_LINE, (u8)ALL_VIRT, (u8)ALL_WRITE);
615 static int vfio_pm_config_write(struct vfio_pci_device *vdev, int pos,
616 int count, struct perm_bits *perm,
617 int offset, __le32 val)
619 count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
623 if (offset == PCI_PM_CTRL) {
626 switch (le32_to_cpu(val) & PCI_PM_CTRL_STATE_MASK) {
641 pci_set_power_state(vdev->pdev, state);
647 /* Permissions for the Power Management capability */
648 static int __init init_pci_cap_pm_perm(struct perm_bits *perm)
650 if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_PM]))
653 perm->writefn = vfio_pm_config_write;
656 * We always virtualize the next field so we can remove
657 * capabilities from the chain if we want to.
659 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
662 * Power management is defined *per function*, so we can let
663 * the user change power state, but we trap and initiate the
664 * change ourselves, so the state bits are read-only.
666 p_setd(perm, PCI_PM_CTRL, NO_VIRT, ~PCI_PM_CTRL_STATE_MASK);
670 /* Permissions for PCI-X capability */
671 static int __init init_pci_cap_pcix_perm(struct perm_bits *perm)
673 /* Alloc 24, but only 8 are used in v0 */
674 if (alloc_perm_bits(perm, PCI_CAP_PCIX_SIZEOF_V2))
677 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
679 p_setw(perm, PCI_X_CMD, NO_VIRT, (u16)ALL_WRITE);
680 p_setd(perm, PCI_X_ECC_CSR, NO_VIRT, ALL_WRITE);
684 /* Permissions for PCI Express capability */
685 static int __init init_pci_cap_exp_perm(struct perm_bits *perm)
687 /* Alloc larger of two possible sizes */
688 if (alloc_perm_bits(perm, PCI_CAP_EXP_ENDPOINT_SIZEOF_V2))
691 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
694 * Allow writes to device control fields (includes FLR!)
695 * but not to devctl_phantom which could confuse IOMMU
696 * or to the ARI bit in devctl2 which is set at probe time
698 p_setw(perm, PCI_EXP_DEVCTL, NO_VIRT, ~PCI_EXP_DEVCTL_PHANTOM);
699 p_setw(perm, PCI_EXP_DEVCTL2, NO_VIRT, ~PCI_EXP_DEVCTL2_ARI);
703 /* Permissions for Advanced Function capability */
704 static int __init init_pci_cap_af_perm(struct perm_bits *perm)
706 if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_AF]))
709 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
710 p_setb(perm, PCI_AF_CTRL, NO_VIRT, PCI_AF_CTRL_FLR);
714 /* Permissions for Advanced Error Reporting extended capability */
715 static int __init init_pci_ext_cap_err_perm(struct perm_bits *perm)
719 if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_ERR]))
723 * Virtualize the first dword of all express capabilities
724 * because it includes the next pointer. This lets us later
725 * remove capabilities from the chain if we need to.
727 p_setd(perm, 0, ALL_VIRT, NO_WRITE);
729 /* Writable bits mask */
730 mask = PCI_ERR_UNC_TRAIN | /* Training */
731 PCI_ERR_UNC_DLP | /* Data Link Protocol */
732 PCI_ERR_UNC_SURPDN | /* Surprise Down */
733 PCI_ERR_UNC_POISON_TLP | /* Poisoned TLP */
734 PCI_ERR_UNC_FCP | /* Flow Control Protocol */
735 PCI_ERR_UNC_COMP_TIME | /* Completion Timeout */
736 PCI_ERR_UNC_COMP_ABORT | /* Completer Abort */
737 PCI_ERR_UNC_UNX_COMP | /* Unexpected Completion */
738 PCI_ERR_UNC_RX_OVER | /* Receiver Overflow */
739 PCI_ERR_UNC_MALF_TLP | /* Malformed TLP */
740 PCI_ERR_UNC_ECRC | /* ECRC Error Status */
741 PCI_ERR_UNC_UNSUP | /* Unsupported Request */
742 PCI_ERR_UNC_ACSV | /* ACS Violation */
743 PCI_ERR_UNC_INTN | /* internal error */
744 PCI_ERR_UNC_MCBTLP | /* MC blocked TLP */
745 PCI_ERR_UNC_ATOMEG | /* Atomic egress blocked */
746 PCI_ERR_UNC_TLPPRE; /* TLP prefix blocked */
747 p_setd(perm, PCI_ERR_UNCOR_STATUS, NO_VIRT, mask);
748 p_setd(perm, PCI_ERR_UNCOR_MASK, NO_VIRT, mask);
749 p_setd(perm, PCI_ERR_UNCOR_SEVER, NO_VIRT, mask);
751 mask = PCI_ERR_COR_RCVR | /* Receiver Error Status */
752 PCI_ERR_COR_BAD_TLP | /* Bad TLP Status */
753 PCI_ERR_COR_BAD_DLLP | /* Bad DLLP Status */
754 PCI_ERR_COR_REP_ROLL | /* REPLAY_NUM Rollover */
755 PCI_ERR_COR_REP_TIMER | /* Replay Timer Timeout */
756 PCI_ERR_COR_ADV_NFAT | /* Advisory Non-Fatal */
757 PCI_ERR_COR_INTERNAL | /* Corrected Internal */
758 PCI_ERR_COR_LOG_OVER; /* Header Log Overflow */
759 p_setd(perm, PCI_ERR_COR_STATUS, NO_VIRT, mask);
760 p_setd(perm, PCI_ERR_COR_MASK, NO_VIRT, mask);
762 mask = PCI_ERR_CAP_ECRC_GENE | /* ECRC Generation Enable */
763 PCI_ERR_CAP_ECRC_CHKE; /* ECRC Check Enable */
764 p_setd(perm, PCI_ERR_CAP, NO_VIRT, mask);
768 /* Permissions for Power Budgeting extended capability */
769 static int __init init_pci_ext_cap_pwr_perm(struct perm_bits *perm)
771 if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_PWR]))
774 p_setd(perm, 0, ALL_VIRT, NO_WRITE);
776 /* Writing the data selector is OK, the info is still read-only */
777 p_setb(perm, PCI_PWR_DATA, NO_VIRT, (u8)ALL_WRITE);
782 * Initialize the shared permission tables
784 void vfio_pci_uninit_perm_bits(void)
786 free_perm_bits(&cap_perms[PCI_CAP_ID_BASIC]);
788 free_perm_bits(&cap_perms[PCI_CAP_ID_PM]);
789 free_perm_bits(&cap_perms[PCI_CAP_ID_PCIX]);
790 free_perm_bits(&cap_perms[PCI_CAP_ID_EXP]);
791 free_perm_bits(&cap_perms[PCI_CAP_ID_AF]);
793 free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_ERR]);
794 free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_PWR]);
797 int __init vfio_pci_init_perm_bits(void)
801 /* Basic config space */
802 ret = init_pci_cap_basic_perm(&cap_perms[PCI_CAP_ID_BASIC]);
805 ret |= init_pci_cap_pm_perm(&cap_perms[PCI_CAP_ID_PM]);
806 cap_perms[PCI_CAP_ID_VPD].writefn = vfio_raw_config_write;
807 ret |= init_pci_cap_pcix_perm(&cap_perms[PCI_CAP_ID_PCIX]);
808 cap_perms[PCI_CAP_ID_VNDR].writefn = vfio_raw_config_write;
809 ret |= init_pci_cap_exp_perm(&cap_perms[PCI_CAP_ID_EXP]);
810 ret |= init_pci_cap_af_perm(&cap_perms[PCI_CAP_ID_AF]);
812 /* Extended capabilities */
813 ret |= init_pci_ext_cap_err_perm(&ecap_perms[PCI_EXT_CAP_ID_ERR]);
814 ret |= init_pci_ext_cap_pwr_perm(&ecap_perms[PCI_EXT_CAP_ID_PWR]);
815 ecap_perms[PCI_EXT_CAP_ID_VNDR].writefn = vfio_raw_config_write;
818 vfio_pci_uninit_perm_bits();
823 static int vfio_find_cap_start(struct vfio_pci_device *vdev, int pos)
826 int base = (pos >= PCI_CFG_SPACE_SIZE) ? PCI_CFG_SPACE_SIZE :
827 PCI_STD_HEADER_SIZEOF;
828 cap = vdev->pci_config_map[pos];
830 if (cap == PCI_CAP_ID_BASIC)
833 /* XXX Can we have to abutting capabilities of the same type? */
834 while (pos - 1 >= base && vdev->pci_config_map[pos - 1] == cap)
840 static int vfio_msi_config_read(struct vfio_pci_device *vdev, int pos,
841 int count, struct perm_bits *perm,
842 int offset, __le32 *val)
844 /* Update max available queue size from msi_qmax */
845 if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
849 start = vfio_find_cap_start(vdev, pos);
851 flags = (__le16 *)&vdev->vconfig[start];
853 *flags &= cpu_to_le16(~PCI_MSI_FLAGS_QMASK);
854 *flags |= cpu_to_le16(vdev->msi_qmax << 1);
857 return vfio_default_config_read(vdev, pos, count, perm, offset, val);
860 static int vfio_msi_config_write(struct vfio_pci_device *vdev, int pos,
861 int count, struct perm_bits *perm,
862 int offset, __le32 val)
864 count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
868 /* Fixup and write configured queue size and enable to hardware */
869 if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
874 start = vfio_find_cap_start(vdev, pos);
876 pflags = (__le16 *)&vdev->vconfig[start + PCI_MSI_FLAGS];
878 flags = le16_to_cpu(*pflags);
880 /* MSI is enabled via ioctl */
882 flags &= ~PCI_MSI_FLAGS_ENABLE;
884 /* Check queue size */
885 if ((flags & PCI_MSI_FLAGS_QSIZE) >> 4 > vdev->msi_qmax) {
886 flags &= ~PCI_MSI_FLAGS_QSIZE;
887 flags |= vdev->msi_qmax << 4;
890 /* Write back to virt and to hardware */
891 *pflags = cpu_to_le16(flags);
892 ret = pci_user_write_config_word(vdev->pdev,
893 start + PCI_MSI_FLAGS,
896 return pcibios_err_to_errno(ret);
903 * MSI determination is per-device, so this routine gets used beyond
904 * initialization time. Don't add __init
906 static int init_pci_cap_msi_perm(struct perm_bits *perm, int len, u16 flags)
908 if (alloc_perm_bits(perm, len))
911 perm->readfn = vfio_msi_config_read;
912 perm->writefn = vfio_msi_config_write;
914 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
917 * The upper byte of the control register is reserved,
918 * just setup the lower byte.
920 p_setb(perm, PCI_MSI_FLAGS, (u8)ALL_VIRT, (u8)ALL_WRITE);
921 p_setd(perm, PCI_MSI_ADDRESS_LO, ALL_VIRT, ALL_WRITE);
922 if (flags & PCI_MSI_FLAGS_64BIT) {
923 p_setd(perm, PCI_MSI_ADDRESS_HI, ALL_VIRT, ALL_WRITE);
924 p_setw(perm, PCI_MSI_DATA_64, (u16)ALL_VIRT, (u16)ALL_WRITE);
925 if (flags & PCI_MSI_FLAGS_MASKBIT) {
926 p_setd(perm, PCI_MSI_MASK_64, NO_VIRT, ALL_WRITE);
927 p_setd(perm, PCI_MSI_PENDING_64, NO_VIRT, ALL_WRITE);
930 p_setw(perm, PCI_MSI_DATA_32, (u16)ALL_VIRT, (u16)ALL_WRITE);
931 if (flags & PCI_MSI_FLAGS_MASKBIT) {
932 p_setd(perm, PCI_MSI_MASK_32, NO_VIRT, ALL_WRITE);
933 p_setd(perm, PCI_MSI_PENDING_32, NO_VIRT, ALL_WRITE);
939 /* Determine MSI CAP field length; initialize msi_perms on 1st call per vdev */
940 static int vfio_msi_cap_len(struct vfio_pci_device *vdev, u8 pos)
942 struct pci_dev *pdev = vdev->pdev;
946 ret = pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &flags);
948 return pcibios_err_to_errno(ret);
950 len = 10; /* Minimum size */
951 if (flags & PCI_MSI_FLAGS_64BIT)
953 if (flags & PCI_MSI_FLAGS_MASKBIT)
959 vdev->msi_perm = kmalloc(sizeof(struct perm_bits), GFP_KERNEL);
963 ret = init_pci_cap_msi_perm(vdev->msi_perm, len, flags);
970 /* Determine extended capability length for VC (2 & 9) and MFVC */
971 static int vfio_vc_cap_len(struct vfio_pci_device *vdev, u16 pos)
973 struct pci_dev *pdev = vdev->pdev;
975 int ret, evcc, phases, vc_arb;
976 int len = PCI_CAP_VC_BASE_SIZEOF;
978 ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_CAP1, &tmp);
980 return pcibios_err_to_errno(ret);
982 evcc = tmp & PCI_VC_CAP1_EVCC; /* extended vc count */
983 ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_CAP2, &tmp);
985 return pcibios_err_to_errno(ret);
987 if (tmp & PCI_VC_CAP2_128_PHASE)
989 else if (tmp & PCI_VC_CAP2_64_PHASE)
991 else if (tmp & PCI_VC_CAP2_32_PHASE)
999 * Port arbitration tables are root & switch only;
1000 * function arbitration tables are function 0 only.
1001 * In either case, we'll never let user write them so
1002 * we don't care how big they are
1004 len += (1 + evcc) * PCI_CAP_VC_PER_VC_SIZEOF;
1006 len = round_up(len, 16);
1012 static int vfio_cap_len(struct vfio_pci_device *vdev, u8 cap, u8 pos)
1014 struct pci_dev *pdev = vdev->pdev;
1021 case PCI_CAP_ID_MSI:
1022 return vfio_msi_cap_len(vdev, pos);
1023 case PCI_CAP_ID_PCIX:
1024 ret = pci_read_config_word(pdev, pos + PCI_X_CMD, &word);
1026 return pcibios_err_to_errno(ret);
1028 if (PCI_X_CMD_VERSION(word)) {
1029 /* Test for extended capabilities */
1030 pci_read_config_dword(pdev, PCI_CFG_SPACE_SIZE, &dword);
1031 vdev->extended_caps = (dword != 0);
1032 return PCI_CAP_PCIX_SIZEOF_V2;
1034 return PCI_CAP_PCIX_SIZEOF_V0;
1035 case PCI_CAP_ID_VNDR:
1036 /* length follows next field */
1037 ret = pci_read_config_byte(pdev, pos + PCI_CAP_FLAGS, &byte);
1039 return pcibios_err_to_errno(ret);
1042 case PCI_CAP_ID_EXP:
1043 /* Test for extended capabilities */
1044 pci_read_config_dword(pdev, PCI_CFG_SPACE_SIZE, &dword);
1045 vdev->extended_caps = (dword != 0);
1047 /* length based on version */
1048 if ((pcie_caps_reg(pdev) & PCI_EXP_FLAGS_VERS) == 1)
1049 return PCI_CAP_EXP_ENDPOINT_SIZEOF_V1;
1051 return PCI_CAP_EXP_ENDPOINT_SIZEOF_V2;
1053 ret = pci_read_config_byte(pdev, pos + 3, &byte);
1055 return pcibios_err_to_errno(ret);
1057 return (byte & HT_3BIT_CAP_MASK) ?
1058 HT_CAP_SIZEOF_SHORT : HT_CAP_SIZEOF_LONG;
1059 case PCI_CAP_ID_SATA:
1060 ret = pci_read_config_byte(pdev, pos + PCI_SATA_REGS, &byte);
1062 return pcibios_err_to_errno(ret);
1064 byte &= PCI_SATA_REGS_MASK;
1065 if (byte == PCI_SATA_REGS_INLINE)
1066 return PCI_SATA_SIZEOF_LONG;
1068 return PCI_SATA_SIZEOF_SHORT;
1070 pr_warn("%s: %s unknown length for pci cap 0x%x@0x%x\n",
1071 dev_name(&pdev->dev), __func__, cap, pos);
1077 static int vfio_ext_cap_len(struct vfio_pci_device *vdev, u16 ecap, u16 epos)
1079 struct pci_dev *pdev = vdev->pdev;
1085 case PCI_EXT_CAP_ID_VNDR:
1086 ret = pci_read_config_dword(pdev, epos + PCI_VSEC_HDR, &dword);
1088 return pcibios_err_to_errno(ret);
1090 return dword >> PCI_VSEC_HDR_LEN_SHIFT;
1091 case PCI_EXT_CAP_ID_VC:
1092 case PCI_EXT_CAP_ID_VC9:
1093 case PCI_EXT_CAP_ID_MFVC:
1094 return vfio_vc_cap_len(vdev, epos);
1095 case PCI_EXT_CAP_ID_ACS:
1096 ret = pci_read_config_byte(pdev, epos + PCI_ACS_CAP, &byte);
1098 return pcibios_err_to_errno(ret);
1100 if (byte & PCI_ACS_EC) {
1103 ret = pci_read_config_byte(pdev,
1104 epos + PCI_ACS_EGRESS_BITS,
1107 return pcibios_err_to_errno(ret);
1109 bits = byte ? round_up(byte, 32) : 256;
1110 return 8 + (bits / 8);
1114 case PCI_EXT_CAP_ID_REBAR:
1115 ret = pci_read_config_byte(pdev, epos + PCI_REBAR_CTRL, &byte);
1117 return pcibios_err_to_errno(ret);
1119 byte &= PCI_REBAR_CTRL_NBAR_MASK;
1120 byte >>= PCI_REBAR_CTRL_NBAR_SHIFT;
1122 return 4 + (byte * 8);
1123 case PCI_EXT_CAP_ID_DPA:
1124 ret = pci_read_config_byte(pdev, epos + PCI_DPA_CAP, &byte);
1126 return pcibios_err_to_errno(ret);
1128 byte &= PCI_DPA_CAP_SUBSTATE_MASK;
1129 byte = round_up(byte + 1, 4);
1130 return PCI_DPA_BASE_SIZEOF + byte;
1131 case PCI_EXT_CAP_ID_TPH:
1132 ret = pci_read_config_dword(pdev, epos + PCI_TPH_CAP, &dword);
1134 return pcibios_err_to_errno(ret);
1136 if ((dword & PCI_TPH_CAP_LOC_MASK) == PCI_TPH_LOC_CAP) {
1139 sts = byte & PCI_TPH_CAP_ST_MASK;
1140 sts >>= PCI_TPH_CAP_ST_SHIFT;
1141 return PCI_TPH_BASE_SIZEOF + round_up(sts * 2, 4);
1143 return PCI_TPH_BASE_SIZEOF;
1145 pr_warn("%s: %s unknown length for pci ecap 0x%x@0x%x\n",
1146 dev_name(&pdev->dev), __func__, ecap, epos);
1152 static int vfio_fill_vconfig_bytes(struct vfio_pci_device *vdev,
1153 int offset, int size)
1155 struct pci_dev *pdev = vdev->pdev;
1159 * We try to read physical config space in the largest chunks
1160 * we can, assuming that all of the fields support dword access.
1161 * pci_save_state() makes this same assumption and seems to do ok.
1166 if (size >= 4 && !(offset % 4)) {
1167 __le32 *dwordp = (__le32 *)&vdev->vconfig[offset];
1170 ret = pci_read_config_dword(pdev, offset, &dword);
1173 *dwordp = cpu_to_le32(dword);
1175 } else if (size >= 2 && !(offset % 2)) {
1176 __le16 *wordp = (__le16 *)&vdev->vconfig[offset];
1179 ret = pci_read_config_word(pdev, offset, &word);
1182 *wordp = cpu_to_le16(word);
1185 u8 *byte = &vdev->vconfig[offset];
1186 ret = pci_read_config_byte(pdev, offset, byte);
1199 static int vfio_cap_init(struct vfio_pci_device *vdev)
1201 struct pci_dev *pdev = vdev->pdev;
1202 u8 *map = vdev->pci_config_map;
1205 int loops, ret, caps = 0;
1207 /* Any capabilities? */
1208 ret = pci_read_config_word(pdev, PCI_STATUS, &status);
1212 if (!(status & PCI_STATUS_CAP_LIST))
1213 return 0; /* Done */
1215 ret = pci_read_config_byte(pdev, PCI_CAPABILITY_LIST, &pos);
1219 /* Mark the previous position in case we want to skip a capability */
1220 prev = &vdev->vconfig[PCI_CAPABILITY_LIST];
1222 /* We can bound our loop, capabilities are dword aligned */
1223 loops = (PCI_CFG_SPACE_SIZE - PCI_STD_HEADER_SIZEOF) / PCI_CAP_SIZEOF;
1224 while (pos && loops--) {
1228 ret = pci_read_config_byte(pdev, pos, &cap);
1232 ret = pci_read_config_byte(pdev,
1233 pos + PCI_CAP_LIST_NEXT, &next);
1237 if (cap <= PCI_CAP_ID_MAX) {
1238 len = pci_cap_length[cap];
1239 if (len == 0xFF) { /* Variable length */
1240 len = vfio_cap_len(vdev, cap, pos);
1247 pr_info("%s: %s hiding cap 0x%x\n",
1248 __func__, dev_name(&pdev->dev), cap);
1254 /* Sanity check, do we overlap other capabilities? */
1255 for (i = 0; i < len; i++) {
1256 if (likely(map[pos + i] == PCI_CAP_ID_INVALID))
1259 pr_warn("%s: %s pci config conflict @0x%x, was cap 0x%x now cap 0x%x\n",
1260 __func__, dev_name(&pdev->dev),
1261 pos + i, map[pos + i], cap);
1264 memset(map + pos, cap, len);
1265 ret = vfio_fill_vconfig_bytes(vdev, pos, len);
1269 prev = &vdev->vconfig[pos + PCI_CAP_LIST_NEXT];
1274 /* If we didn't fill any capabilities, clear the status flag */
1276 __le16 *vstatus = (__le16 *)&vdev->vconfig[PCI_STATUS];
1277 *vstatus &= ~cpu_to_le16(PCI_STATUS_CAP_LIST);
1283 static int vfio_ecap_init(struct vfio_pci_device *vdev)
1285 struct pci_dev *pdev = vdev->pdev;
1286 u8 *map = vdev->pci_config_map;
1288 __le32 *prev = NULL;
1289 int loops, ret, ecaps = 0;
1291 if (!vdev->extended_caps)
1294 epos = PCI_CFG_SPACE_SIZE;
1296 loops = (pdev->cfg_size - PCI_CFG_SPACE_SIZE) / PCI_CAP_SIZEOF;
1298 while (loops-- && epos >= PCI_CFG_SPACE_SIZE) {
1302 bool hidden = false;
1304 ret = pci_read_config_dword(pdev, epos, &header);
1308 ecap = PCI_EXT_CAP_ID(header);
1310 if (ecap <= PCI_EXT_CAP_ID_MAX) {
1311 len = pci_ext_cap_length[ecap];
1313 len = vfio_ext_cap_len(vdev, ecap, epos);
1320 pr_info("%s: %s hiding ecap 0x%x@0x%x\n",
1321 __func__, dev_name(&pdev->dev), ecap, epos);
1323 /* If not the first in the chain, we can skip over it */
1325 u32 val = epos = PCI_EXT_CAP_NEXT(header);
1326 *prev &= cpu_to_le32(~(0xffcU << 20));
1327 *prev |= cpu_to_le32(val << 20);
1332 * Otherwise, fill in a placeholder, the direct
1333 * readfn will virtualize this automatically
1335 len = PCI_CAP_SIZEOF;
1339 for (i = 0; i < len; i++) {
1340 if (likely(map[epos + i] == PCI_CAP_ID_INVALID))
1343 pr_warn("%s: %s pci config conflict @0x%x, was ecap 0x%x now ecap 0x%x\n",
1344 __func__, dev_name(&pdev->dev),
1345 epos + i, map[epos + i], ecap);
1349 * Even though ecap is 2 bytes, we're currently a long way
1350 * from exceeding 1 byte capabilities. If we ever make it
1351 * up to 0xFF we'll need to up this to a two-byte, byte map.
1353 BUILD_BUG_ON(PCI_EXT_CAP_ID_MAX >= PCI_CAP_ID_INVALID);
1355 memset(map + epos, ecap, len);
1356 ret = vfio_fill_vconfig_bytes(vdev, epos, len);
1361 * If we're just using this capability to anchor the list,
1362 * hide the real ID. Only count real ecaps. XXX PCI spec
1363 * indicates to use cap id = 0, version = 0, next = 0 if
1364 * ecaps are absent, hope users check all the way to next.
1367 *(__le32 *)&vdev->vconfig[epos] &=
1368 cpu_to_le32((0xffcU << 20));
1372 prev = (__le32 *)&vdev->vconfig[epos];
1373 epos = PCI_EXT_CAP_NEXT(header);
1377 *(u32 *)&vdev->vconfig[PCI_CFG_SPACE_SIZE] = 0;
1383 * For each device we allocate a pci_config_map that indicates the
1384 * capability occupying each dword and thus the struct perm_bits we
1385 * use for read and write. We also allocate a virtualized config
1386 * space which tracks reads and writes to bits that we emulate for
1387 * the user. Initial values filled from device.
1389 * Using shared stuct perm_bits between all vfio-pci devices saves
1390 * us from allocating cfg_size buffers for virt and write for every
1391 * device. We could remove vconfig and allocate individual buffers
1392 * for each area requring emulated bits, but the array of pointers
1393 * would be comparable in size (at least for standard config space).
1395 int vfio_config_init(struct vfio_pci_device *vdev)
1397 struct pci_dev *pdev = vdev->pdev;
1402 * Config space, caps and ecaps are all dword aligned, so we could
1403 * use one byte per dword to record the type. However, there are
1404 * no requiremenst on the length of a capability, so the gap between
1405 * capabilities needs byte granularity.
1407 map = kmalloc(pdev->cfg_size, GFP_KERNEL);
1411 vconfig = kmalloc(pdev->cfg_size, GFP_KERNEL);
1417 vdev->pci_config_map = map;
1418 vdev->vconfig = vconfig;
1420 memset(map, PCI_CAP_ID_BASIC, PCI_STD_HEADER_SIZEOF);
1421 memset(map + PCI_STD_HEADER_SIZEOF, PCI_CAP_ID_INVALID,
1422 pdev->cfg_size - PCI_STD_HEADER_SIZEOF);
1424 ret = vfio_fill_vconfig_bytes(vdev, 0, PCI_STD_HEADER_SIZEOF);
1428 vdev->bardirty = true;
1431 * XXX can we just pci_load_saved_state/pci_restore_state?
1432 * may need to rebuild vconfig after that
1435 /* For restore after reset */
1436 vdev->rbar[0] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_0]);
1437 vdev->rbar[1] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_1]);
1438 vdev->rbar[2] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_2]);
1439 vdev->rbar[3] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_3]);
1440 vdev->rbar[4] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_4]);
1441 vdev->rbar[5] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_5]);
1442 vdev->rbar[6] = le32_to_cpu(*(__le32 *)&vconfig[PCI_ROM_ADDRESS]);
1444 if (pdev->is_virtfn) {
1445 *(__le16 *)&vconfig[PCI_VENDOR_ID] = cpu_to_le16(pdev->vendor);
1446 *(__le16 *)&vconfig[PCI_DEVICE_ID] = cpu_to_le16(pdev->device);
1449 ret = vfio_cap_init(vdev);
1453 ret = vfio_ecap_init(vdev);
1461 vdev->pci_config_map = NULL;
1463 vdev->vconfig = NULL;
1464 return pcibios_err_to_errno(ret);
1467 void vfio_config_free(struct vfio_pci_device *vdev)
1469 kfree(vdev->vconfig);
1470 vdev->vconfig = NULL;
1471 kfree(vdev->pci_config_map);
1472 vdev->pci_config_map = NULL;
1473 kfree(vdev->msi_perm);
1474 vdev->msi_perm = NULL;
1478 * Find the remaining number of bytes in a dword that match the given
1479 * position. Stop at either the end of the capability or the dword boundary.
1481 static size_t vfio_pci_cap_remaining_dword(struct vfio_pci_device *vdev,
1484 u8 cap = vdev->pci_config_map[pos];
1487 for (i = 1; (pos + i) % 4 && vdev->pci_config_map[pos + i] == cap; i++)
1493 static ssize_t vfio_config_do_rw(struct vfio_pci_device *vdev, char __user *buf,
1494 size_t count, loff_t *ppos, bool iswrite)
1496 struct pci_dev *pdev = vdev->pdev;
1497 struct perm_bits *perm;
1499 int cap_start = 0, offset;
1503 if (*ppos < 0 || *ppos >= pdev->cfg_size ||
1504 *ppos + count > pdev->cfg_size)
1508 * Chop accesses into aligned chunks containing no more than a
1509 * single capability. Caller increments to the next chunk.
1511 count = min(count, vfio_pci_cap_remaining_dword(vdev, *ppos));
1512 if (count >= 4 && !(*ppos % 4))
1514 else if (count >= 2 && !(*ppos % 2))
1521 cap_id = vdev->pci_config_map[*ppos];
1523 if (cap_id == PCI_CAP_ID_INVALID) {
1524 perm = &unassigned_perms;
1527 if (*ppos >= PCI_CFG_SPACE_SIZE) {
1528 WARN_ON(cap_id > PCI_EXT_CAP_ID_MAX);
1530 perm = &ecap_perms[cap_id];
1531 cap_start = vfio_find_cap_start(vdev, *ppos);
1533 WARN_ON(cap_id > PCI_CAP_ID_MAX);
1535 perm = &cap_perms[cap_id];
1537 if (cap_id == PCI_CAP_ID_MSI)
1538 perm = vdev->msi_perm;
1540 if (cap_id > PCI_CAP_ID_BASIC)
1541 cap_start = vfio_find_cap_start(vdev, *ppos);
1545 WARN_ON(!cap_start && cap_id != PCI_CAP_ID_BASIC);
1546 WARN_ON(cap_start > *ppos);
1548 offset = *ppos - cap_start;
1554 if (copy_from_user(&val, buf, count))
1557 ret = perm->writefn(vdev, *ppos, count, perm, offset, val);
1560 ret = perm->readfn(vdev, *ppos, count,
1561 perm, offset, &val);
1566 if (copy_to_user(buf, &val, count))
1573 ssize_t vfio_pci_config_rw(struct vfio_pci_device *vdev, char __user *buf,
1574 size_t count, loff_t *ppos, bool iswrite)
1580 pos &= VFIO_PCI_OFFSET_MASK;
1583 ret = vfio_config_do_rw(vdev, buf, count, &pos, iswrite);