?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'for-linus-v3.11-rc3' of git://oss.sgi.com/xfs/xfs
[linux-2.6.git] / drivers / vfio / pci / vfio_pci_config.c
blobaffa34745be92bdfe16e83ebb81ee4073c002729
1 /*
2 * VFIO PCI config space virtualization
3 *
4 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
5 * Author: Alex Williamson <alex.williamson@redhat.com>
6 *
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.
10 *
11 * Derived from original vfio:
12 * Copyright 2010 Cisco Systems, Inc. All rights reserved.
13 * Author: Tom Lyon, pugs@cisco.com
14 */
15
16 /*
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.
24 */
25
26 #include <linux/fs.h>
27 #include <linux/pci.h>
28 #include <linux/uaccess.h>
29 #include <linux/vfio.h>
30 #include <linux/slab.h>
31
32 #include "vfio_pci_private.h"
33
34 #define PCI_CFG_SPACE_SIZE 256
35
36 /* Useful "pseudo" capabilities */
37 #define PCI_CAP_ID_BASIC 0
38 #define PCI_CAP_ID_INVALID 0xFF
39
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))
43
44 /*
45 * Lengths of PCI Config Capabilities
46 * 0: Removed from the user visible capability list
47 * FF: Variable length
48 */
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,
70 };
71
72 /*
73 * Lengths of PCIe/PCI-X Extended Config Capabilities
74 * 0: Removed or masked from the user visible capabilty list
75 * FF: Variable length
76 */
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 */
105 };
106
107 /*
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.
113 */
114 struct perm_bits {
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);
121 };
122
123 #define NO_VIRT 0
124 #define ALL_VIRT 0xFFFFFFFFU
125 #define NO_WRITE 0
126 #define ALL_WRITE 0xFFFFFFFFU
127
128 static int vfio_user_config_read(struct pci_dev *pdev, int offset,
129 __le32 *val, int count)
130 {
131 int ret = -EINVAL;
132 u32 tmp_val = 0;
133
134 switch (count) {
135 case 1:
136 {
137 u8 tmp;
138 ret = pci_user_read_config_byte(pdev, offset, &tmp);
139 tmp_val = tmp;
140 break;
141 }
142 case 2:
143 {
144 u16 tmp;
145 ret = pci_user_read_config_word(pdev, offset, &tmp);
146 tmp_val = tmp;
147 break;
148 }
149 case 4:
150 ret = pci_user_read_config_dword(pdev, offset, &tmp_val);
151 break;
152 }
153
154 *val = cpu_to_le32(tmp_val);
155
156 return pcibios_err_to_errno(ret);
157 }
158
159 static int vfio_user_config_write(struct pci_dev *pdev, int offset,
160 __le32 val, int count)
161 {
162 int ret = -EINVAL;
163 u32 tmp_val = le32_to_cpu(val);
164
165 switch (count) {
166 case 1:
167 ret = pci_user_write_config_byte(pdev, offset, tmp_val);
168 break;
169 case 2:
170 ret = pci_user_write_config_word(pdev, offset, tmp_val);
171 break;
172 case 4:
173 ret = pci_user_write_config_dword(pdev, offset, tmp_val);
174 break;
175 }
176
177 return pcibios_err_to_errno(ret);
178 }
179
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)
183 {
184 __le32 virt = 0;
185
186 memcpy(val, vdev->vconfig + pos, count);
187
188 memcpy(&virt, perm->virt + offset, count);
189
190 /* Any non-virtualized bits? */
191 if (cpu_to_le32(~0U >> (32 - (count * 8))) != virt) {
192 struct pci_dev *pdev = vdev->pdev;
193 __le32 phys_val = 0;
194 int ret;
195
196 ret = vfio_user_config_read(pdev, pos, &phys_val, count);
197 if (ret)
198 return ret;
199
200 *val = (phys_val & ~virt) | (*val & virt);
201 }
202
203 return count;
204 }
205
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)
209 {
210 __le32 virt = 0, write = 0;
211
212 memcpy(&write, perm->write + offset, count);
213
214 if (!write)
215 return count; /* drop, no writable bits */
216
217 memcpy(&virt, perm->virt + offset, count);
218
219 /* Virtualized and writable bits go to vconfig */
220 if (write & virt) {
221 __le32 virt_val = 0;
222
223 memcpy(&virt_val, vdev->vconfig + pos, count);
224
225 virt_val &= ~(write & virt);
226 virt_val |= (val & (write & virt));
227
228 memcpy(vdev->vconfig + pos, &virt_val, count);
229 }
230
231 /* Non-virtualzed and writable bits go to hardware */
232 if (write & ~virt) {
233 struct pci_dev *pdev = vdev->pdev;
234 __le32 phys_val = 0;
235 int ret;
236
237 ret = vfio_user_config_read(pdev, pos, &phys_val, count);
238 if (ret)
239 return ret;
240
241 phys_val &= ~(write & ~virt);
242 phys_val |= (val & (write & ~virt));
243
244 ret = vfio_user_config_write(pdev, pos, phys_val, count);
245 if (ret)
246 return ret;
247 }
248
249 return count;
250 }
251
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)
256 {
257 int ret;
258
259 ret = vfio_user_config_read(vdev->pdev, pos, val, count);
260 if (ret)
261 return pcibios_err_to_errno(ret);
262
263 if (pos >= PCI_CFG_SPACE_SIZE) { /* Extended cap header mangling */
264 if (offset < 4)
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);
272 }
273
274 return count;
275 }
276
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)
281 {
282 int ret;
283
284 ret = vfio_user_config_write(vdev->pdev, pos, val, count);
285 if (ret)
286 return ret;
287
288 return count;
289 }
290
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)
294 {
295 int ret;
296
297 ret = vfio_user_config_read(vdev->pdev, pos, val, count);
298 if (ret)
299 return pcibios_err_to_errno(ret);
300
301 return count;
302 }
303
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 }
307 };
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 }
310 };
311 /*
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.
316 */
317 static struct perm_bits unassigned_perms = {
318 .readfn = vfio_raw_config_read,
319 .writefn = vfio_raw_config_write
320 };
321
322 static void free_perm_bits(struct perm_bits *perm)
323 {
324 kfree(perm->virt);
325 kfree(perm->write);
326 perm->virt = NULL;
327 perm->write = NULL;
328 }
329
330 static int alloc_perm_bits(struct perm_bits *perm, int size)
331 {
332 /*
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
339 */
340 size = round_up(size, 4);
341
342 /*
343 * Zero state is
344 * - All Readable, None Writeable, None Virtualized
345 */
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);
350 return -ENOMEM;
351 }
352
353 perm->readfn = vfio_default_config_read;
354 perm->writefn = vfio_default_config_write;
355
356 return 0;
357 }
358
359 /*
360 * Helper functions for filling in permission tables
361 */
362 static inline void p_setb(struct perm_bits *p, int off, u8 virt, u8 write)
363 {
364 p->virt[off] = virt;
365 p->write[off] = write;
366 }
367
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)
370 {
371 *(__le16 *)(&p->virt[off]) = cpu_to_le16(virt);
372 *(__le16 *)(&p->write[off]) = cpu_to_le16(write);
373 }
374
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)
377 {
378 *(__le32 *)(&p->virt[off]) = cpu_to_le32(virt);
379 *(__le32 *)(&p->write[off]) = cpu_to_le32(write);
380 }
381
382 /*
383 * Restore the *real* BARs after we detect a FLR or backdoor reset.
384 * (backdoor = some device specific technique that we didn't catch)
385 */
386 static void vfio_bar_restore(struct vfio_pci_device *vdev)
387 {
388 struct pci_dev *pdev = vdev->pdev;
389 u32 *rbar = vdev->rbar;
390 int i;
391
392 if (pdev->is_virtfn)
393 return;
394
395 pr_info("%s: %s reset recovery - restoring bars\n",
396 __func__, dev_name(&pdev->dev));
397
398 for (i = PCI_BASE_ADDRESS_0; i <= PCI_BASE_ADDRESS_5; i += 4, rbar++)
399 pci_user_write_config_dword(pdev, i, *rbar);
400
401 pci_user_write_config_dword(pdev, PCI_ROM_ADDRESS, *rbar);
402 }
403
404 static __le32 vfio_generate_bar_flags(struct pci_dev *pdev, int bar)
405 {
406 unsigned long flags = pci_resource_flags(pdev, bar);
407 u32 val;
408
409 if (flags & IORESOURCE_IO)
410 return cpu_to_le32(PCI_BASE_ADDRESS_SPACE_IO);
411
412 val = PCI_BASE_ADDRESS_SPACE_MEMORY;
413
414 if (flags & IORESOURCE_PREFETCH)
415 val |= PCI_BASE_ADDRESS_MEM_PREFETCH;
416
417 if (flags & IORESOURCE_MEM_64)
418 val |= PCI_BASE_ADDRESS_MEM_TYPE_64;
419
420 return cpu_to_le32(val);
421 }
422
423 /*
424 * Pretend we're hardware and tweak the values of the *virtual* PCI BARs
425 * to reflect the hardware capabilities. This implements BAR sizing.
426 */
427 static void vfio_bar_fixup(struct vfio_pci_device *vdev)
428 {
429 struct pci_dev *pdev = vdev->pdev;
430 int i;
431 __le32 *bar;
432 u64 mask;
433
434 bar = (__le32 *)&vdev->vconfig[PCI_BASE_ADDRESS_0];
435
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 */
439 continue;
440 }
441
442 mask = ~(pci_resource_len(pdev, i) - 1);
443
444 *bar &= cpu_to_le32((u32)mask);
445 *bar |= vfio_generate_bar_flags(pdev, i);
446
447 if (*bar & cpu_to_le32(PCI_BASE_ADDRESS_MEM_TYPE_64)) {
448 bar++;
449 *bar &= cpu_to_le32((u32)(mask >> 32));
450 i++;
451 }
452 }
453
454 bar = (__le32 *)&vdev->vconfig[PCI_ROM_ADDRESS];
455
456 /*
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.
460 */
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);
465 } else
466 *bar = 0;
467
468 vdev->bardirty = false;
469 }
470
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)
474 {
475 if (is_bar(offset)) /* pos == offset for basic config */
476 vfio_bar_fixup(vdev);
477
478 count = vfio_default_config_read(vdev, pos, count, perm, offset, val);
479
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);
484
485 tmp_val |= cmd & PCI_COMMAND_MEMORY;
486 *val = cpu_to_le32(tmp_val);
487 }
488
489 return count;
490 }
491
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)
495 {
496 struct pci_dev *pdev = vdev->pdev;
497 __le16 *virt_cmd;
498 u16 new_cmd = 0;
499 int ret;
500
501 virt_cmd = (__le16 *)&vdev->vconfig[PCI_COMMAND];
502
503 if (offset == PCI_COMMAND) {
504 bool phys_mem, virt_mem, new_mem, phys_io, virt_io, new_io;
505 u16 phys_cmd;
506
507 ret = pci_user_read_config_word(pdev, PCI_COMMAND, &phys_cmd);
508 if (ret)
509 return ret;
510
511 new_cmd = le32_to_cpu(val);
512
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);
516
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);
520
521 /*
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
528 */
529 if ((new_mem && virt_mem && !phys_mem) ||
530 (new_io && virt_io && !phys_io))
531 vfio_bar_restore(vdev);
532 }
533
534 count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
535 if (count < 0)
536 return count;
537
538 /*
539 * Save current memory/io enable bits in vconfig to allow for
540 * the test above next time.
541 */
542 if (offset == PCI_COMMAND) {
543 u16 mask = PCI_COMMAND_MEMORY | PCI_COMMAND_IO;
544
545 *virt_cmd &= cpu_to_le16(~mask);
546 *virt_cmd |= cpu_to_le16(new_cmd & mask);
547 }
548
549 /* Emulate INTx disable */
550 if (offset >= PCI_COMMAND && offset <= PCI_COMMAND + 1) {
551 bool virt_intx_disable;
552
553 virt_intx_disable = !!(le16_to_cpu(*virt_cmd) &
554 PCI_COMMAND_INTX_DISABLE);
555
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);
562 }
563 }
564
565 if (is_bar(offset))
566 vdev->bardirty = true;
567
568 return count;
569 }
570
571 /* Permissions for the Basic PCI Header */
572 static int __init init_pci_cap_basic_perm(struct perm_bits *perm)
573 {
574 if (alloc_perm_bits(perm, PCI_STD_HEADER_SIZEOF))
575 return -ENOMEM;
576
577 perm->readfn = vfio_basic_config_read;
578 perm->writefn = vfio_basic_config_write;
579
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);
583
584 /*
585 * Virtualize INTx disable, we use it internally for interrupt
586 * control and can emulate it for non-PCI 2.3 devices.
587 */
588 p_setw(perm, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE, (u16)ALL_WRITE);
589
590 /* Virtualize capability list, we might want to skip/disable */
591 p_setw(perm, PCI_STATUS, PCI_STATUS_CAP_LIST, NO_WRITE);
592
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);
597
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);
606
607 /* Allow us to adjust capability chain */
608 p_setb(perm, PCI_CAPABILITY_LIST, (u8)ALL_VIRT, NO_WRITE);
609
610 /* Sometimes used by sw, just virtualize */
611 p_setb(perm, PCI_INTERRUPT_LINE, (u8)ALL_VIRT, (u8)ALL_WRITE);
612 return 0;
613 }
614
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)
618 {
619 count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
620 if (count < 0)
621 return count;
622
623 if (offset == PCI_PM_CTRL) {
624 pci_power_t state;
625
626 switch (le32_to_cpu(val) & PCI_PM_CTRL_STATE_MASK) {
627 case 0:
628 state = PCI_D0;
629 break;
630 case 1:
631 state = PCI_D1;
632 break;
633 case 2:
634 state = PCI_D2;
635 break;
636 case 3:
637 state = PCI_D3hot;
638 break;
639 }
640
641 pci_set_power_state(vdev->pdev, state);
642 }
643
644 return count;
645 }
646
647 /* Permissions for the Power Management capability */
648 static int __init init_pci_cap_pm_perm(struct perm_bits *perm)
649 {
650 if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_PM]))
651 return -ENOMEM;
652
653 perm->writefn = vfio_pm_config_write;
654
655 /*
656 * We always virtualize the next field so we can remove
657 * capabilities from the chain if we want to.
658 */
659 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
660
661 /*
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.
665 */
666 p_setd(perm, PCI_PM_CTRL, NO_VIRT, ~PCI_PM_CTRL_STATE_MASK);
667 return 0;
668 }
669
670 /* Permissions for PCI-X capability */
671 static int __init init_pci_cap_pcix_perm(struct perm_bits *perm)
672 {
673 /* Alloc 24, but only 8 are used in v0 */
674 if (alloc_perm_bits(perm, PCI_CAP_PCIX_SIZEOF_V2))
675 return -ENOMEM;
676
677 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
678
679 p_setw(perm, PCI_X_CMD, NO_VIRT, (u16)ALL_WRITE);
680 p_setd(perm, PCI_X_ECC_CSR, NO_VIRT, ALL_WRITE);
681 return 0;
682 }
683
684 /* Permissions for PCI Express capability */
685 static int __init init_pci_cap_exp_perm(struct perm_bits *perm)
686 {
687 /* Alloc larger of two possible sizes */
688 if (alloc_perm_bits(perm, PCI_CAP_EXP_ENDPOINT_SIZEOF_V2))
689 return -ENOMEM;
690
691 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
692
693 /*
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
697 */
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);
700 return 0;
701 }
702
703 /* Permissions for Advanced Function capability */
704 static int __init init_pci_cap_af_perm(struct perm_bits *perm)
705 {
706 if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_AF]))
707 return -ENOMEM;
708
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);
711 return 0;
712 }
713
714 /* Permissions for Advanced Error Reporting extended capability */
715 static int __init init_pci_ext_cap_err_perm(struct perm_bits *perm)
716 {
717 u32 mask;
718
719 if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_ERR]))
720 return -ENOMEM;
721
722 /*
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.
726 */
727 p_setd(perm, 0, ALL_VIRT, NO_WRITE);
728
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);
750
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);
761
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);
765 return 0;
766 }
767
768 /* Permissions for Power Budgeting extended capability */
769 static int __init init_pci_ext_cap_pwr_perm(struct perm_bits *perm)
770 {
771 if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_PWR]))
772 return -ENOMEM;
773
774 p_setd(perm, 0, ALL_VIRT, NO_WRITE);
775
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);
778 return 0;
779 }
780
781 /*
782 * Initialize the shared permission tables
783 */
784 void vfio_pci_uninit_perm_bits(void)
785 {
786 free_perm_bits(&cap_perms[PCI_CAP_ID_BASIC]);
787
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]);
792
793 free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_ERR]);
794 free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_PWR]);
795 }
796
797 int __init vfio_pci_init_perm_bits(void)
798 {
799 int ret;
800
801 /* Basic config space */
802 ret = init_pci_cap_basic_perm(&cap_perms[PCI_CAP_ID_BASIC]);
803
804 /* Capabilities */
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]);
811
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;
816
817 if (ret)
818 vfio_pci_uninit_perm_bits();
819
820 return ret;
821 }
822
823 static int vfio_find_cap_start(struct vfio_pci_device *vdev, int pos)
824 {
825 u8 cap;
826 int base = (pos >= PCI_CFG_SPACE_SIZE) ? PCI_CFG_SPACE_SIZE :
827 PCI_STD_HEADER_SIZEOF;
828 cap = vdev->pci_config_map[pos];
829
830 if (cap == PCI_CAP_ID_BASIC)
831 return 0;
832
833 /* XXX Can we have to abutting capabilities of the same type? */
834 while (pos - 1 >= base && vdev->pci_config_map[pos - 1] == cap)
835 pos--;
836
837 return pos;
838 }
839
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)
843 {
844 /* Update max available queue size from msi_qmax */
845 if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
846 __le16 *flags;
847 int start;
848
849 start = vfio_find_cap_start(vdev, pos);
850
851 flags = (__le16 *)&vdev->vconfig[start];
852
853 *flags &= cpu_to_le16(~PCI_MSI_FLAGS_QMASK);
854 *flags |= cpu_to_le16(vdev->msi_qmax << 1);
855 }
856
857 return vfio_default_config_read(vdev, pos, count, perm, offset, val);
858 }
859
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)
863 {
864 count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
865 if (count < 0)
866 return count;
867
868 /* Fixup and write configured queue size and enable to hardware */
869 if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
870 __le16 *pflags;
871 u16 flags;
872 int start, ret;
873
874 start = vfio_find_cap_start(vdev, pos);
875
876 pflags = (__le16 *)&vdev->vconfig[start + PCI_MSI_FLAGS];
877
878 flags = le16_to_cpu(*pflags);
879
880 /* MSI is enabled via ioctl */
881 if (!is_msi(vdev))
882 flags &= ~PCI_MSI_FLAGS_ENABLE;
883
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;
888 }
889
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,
894 flags);
895 if (ret)
896 return pcibios_err_to_errno(ret);
897 }
898
899 return count;
900 }
901
902 /*
903 * MSI determination is per-device, so this routine gets used beyond
904 * initialization time. Don't add __init
905 */
906 static int init_pci_cap_msi_perm(struct perm_bits *perm, int len, u16 flags)
907 {
908 if (alloc_perm_bits(perm, len))
909 return -ENOMEM;
910
911 perm->readfn = vfio_msi_config_read;
912 perm->writefn = vfio_msi_config_write;
913
914 p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
915
916 /*
917 * The upper byte of the control register is reserved,
918 * just setup the lower byte.
919 */
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);
928 }
929 } else {
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);
934 }
935 }
936 return 0;
937 }
938
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)
941 {
942 struct pci_dev *pdev = vdev->pdev;
943 int len, ret;
944 u16 flags;
945
946 ret = pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &flags);
947 if (ret)
948 return pcibios_err_to_errno(ret);
949
950 len = 10; /* Minimum size */
951 if (flags & PCI_MSI_FLAGS_64BIT)
952 len += 4;
953 if (flags & PCI_MSI_FLAGS_MASKBIT)
954 len += 10;
955
956 if (vdev->msi_perm)
957 return len;
958
959 vdev->msi_perm = kmalloc(sizeof(struct perm_bits), GFP_KERNEL);
960 if (!vdev->msi_perm)
961 return -ENOMEM;
962
963 ret = init_pci_cap_msi_perm(vdev->msi_perm, len, flags);
964 if (ret)
965 return ret;
966
967 return len;
968 }
969
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)
972 {
973 struct pci_dev *pdev = vdev->pdev;
974 u32 tmp;
975 int ret, evcc, phases, vc_arb;
976 int len = PCI_CAP_VC_BASE_SIZEOF;
977
978 ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_REG1, &tmp);
979 if (ret)
980 return pcibios_err_to_errno(ret);
981
982 evcc = tmp & PCI_VC_REG1_EVCC; /* extended vc count */
983 ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_REG2, &tmp);
984 if (ret)
985 return pcibios_err_to_errno(ret);
986
987 if (tmp & PCI_VC_REG2_128_PHASE)
988 phases = 128;
989 else if (tmp & PCI_VC_REG2_64_PHASE)
990 phases = 64;
991 else if (tmp & PCI_VC_REG2_32_PHASE)
992 phases = 32;
993 else
994 phases = 0;
995
996 vc_arb = phases * 4;
997
998 /*
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
1003 */
1004 len += (1 + evcc) * PCI_CAP_VC_PER_VC_SIZEOF;
1005 if (vc_arb) {
1006 len = round_up(len, 16);
1007 len += vc_arb / 8;
1008 }
1009 return len;
1010 }
1011
1012 static int vfio_cap_len(struct vfio_pci_device *vdev, u8 cap, u8 pos)
1013 {
1014 struct pci_dev *pdev = vdev->pdev;
1015 u16 word;
1016 u8 byte;
1017 int ret;
1018
1019 switch (cap) {
1020 case PCI_CAP_ID_MSI:
1021 return vfio_msi_cap_len(vdev, pos);
1022 case PCI_CAP_ID_PCIX:
1023 ret = pci_read_config_word(pdev, pos + PCI_X_CMD, &word);
1024 if (ret)
1025 return pcibios_err_to_errno(ret);
1026
1027 if (PCI_X_CMD_VERSION(word)) {
1028 vdev->extended_caps = true;
1029 return PCI_CAP_PCIX_SIZEOF_V2;
1030 } else
1031 return PCI_CAP_PCIX_SIZEOF_V0;
1032 case PCI_CAP_ID_VNDR:
1033 /* length follows next field */
1034 ret = pci_read_config_byte(pdev, pos + PCI_CAP_FLAGS, &byte);
1035 if (ret)
1036 return pcibios_err_to_errno(ret);
1037
1038 return byte;
1039 case PCI_CAP_ID_EXP:
1040 /* length based on version */
1041 vdev->extended_caps = true;
1042
1043 if ((pcie_caps_reg(pdev) & PCI_EXP_FLAGS_VERS) == 1)
1044 return PCI_CAP_EXP_ENDPOINT_SIZEOF_V1;
1045 else
1046 return PCI_CAP_EXP_ENDPOINT_SIZEOF_V2;
1047 case PCI_CAP_ID_HT:
1048 ret = pci_read_config_byte(pdev, pos + 3, &byte);
1049 if (ret)
1050 return pcibios_err_to_errno(ret);
1051
1052 return (byte & HT_3BIT_CAP_MASK) ?
1053 HT_CAP_SIZEOF_SHORT : HT_CAP_SIZEOF_LONG;
1054 case PCI_CAP_ID_SATA:
1055 ret = pci_read_config_byte(pdev, pos + PCI_SATA_REGS, &byte);
1056 if (ret)
1057 return pcibios_err_to_errno(ret);
1058
1059 byte &= PCI_SATA_REGS_MASK;
1060 if (byte == PCI_SATA_REGS_INLINE)
1061 return PCI_SATA_SIZEOF_LONG;
1062 else
1063 return PCI_SATA_SIZEOF_SHORT;
1064 default:
1065 pr_warn("%s: %s unknown length for pci cap 0x%x@0x%x\n",
1066 dev_name(&pdev->dev), __func__, cap, pos);
1067 }
1068
1069 return 0;
1070 }
1071
1072 static int vfio_ext_cap_len(struct vfio_pci_device *vdev, u16 ecap, u16 epos)
1073 {
1074 struct pci_dev *pdev = vdev->pdev;
1075 u8 byte;
1076 u32 dword;
1077 int ret;
1078
1079 switch (ecap) {
1080 case PCI_EXT_CAP_ID_VNDR:
1081 ret = pci_read_config_dword(pdev, epos + PCI_VSEC_HDR, &dword);
1082 if (ret)
1083 return pcibios_err_to_errno(ret);
1084
1085 return dword >> PCI_VSEC_HDR_LEN_SHIFT;
1086 case PCI_EXT_CAP_ID_VC:
1087 case PCI_EXT_CAP_ID_VC9:
1088 case PCI_EXT_CAP_ID_MFVC:
1089 return vfio_vc_cap_len(vdev, epos);
1090 case PCI_EXT_CAP_ID_ACS:
1091 ret = pci_read_config_byte(pdev, epos + PCI_ACS_CAP, &byte);
1092 if (ret)
1093 return pcibios_err_to_errno(ret);
1094
1095 if (byte & PCI_ACS_EC) {
1096 int bits;
1097
1098 ret = pci_read_config_byte(pdev,
1099 epos + PCI_ACS_EGRESS_BITS,
1100 &byte);
1101 if (ret)
1102 return pcibios_err_to_errno(ret);
1103
1104 bits = byte ? round_up(byte, 32) : 256;
1105 return 8 + (bits / 8);
1106 }
1107 return 8;
1108
1109 case PCI_EXT_CAP_ID_REBAR:
1110 ret = pci_read_config_byte(pdev, epos + PCI_REBAR_CTRL, &byte);
1111 if (ret)
1112 return pcibios_err_to_errno(ret);
1113
1114 byte &= PCI_REBAR_CTRL_NBAR_MASK;
1115 byte >>= PCI_REBAR_CTRL_NBAR_SHIFT;
1116
1117 return 4 + (byte * 8);
1118 case PCI_EXT_CAP_ID_DPA:
1119 ret = pci_read_config_byte(pdev, epos + PCI_DPA_CAP, &byte);
1120 if (ret)
1121 return pcibios_err_to_errno(ret);
1122
1123 byte &= PCI_DPA_CAP_SUBSTATE_MASK;
1124 byte = round_up(byte + 1, 4);
1125 return PCI_DPA_BASE_SIZEOF + byte;
1126 case PCI_EXT_CAP_ID_TPH:
1127 ret = pci_read_config_dword(pdev, epos + PCI_TPH_CAP, &dword);
1128 if (ret)
1129 return pcibios_err_to_errno(ret);
1130
1131 if ((dword & PCI_TPH_CAP_LOC_MASK) == PCI_TPH_LOC_CAP) {
1132 int sts;
1133
1134 sts = byte & PCI_TPH_CAP_ST_MASK;
1135 sts >>= PCI_TPH_CAP_ST_SHIFT;
1136 return PCI_TPH_BASE_SIZEOF + round_up(sts * 2, 4);
1137 }
1138 return PCI_TPH_BASE_SIZEOF;
1139 default:
1140 pr_warn("%s: %s unknown length for pci ecap 0x%x@0x%x\n",
1141 dev_name(&pdev->dev), __func__, ecap, epos);
1142 }
1143
1144 return 0;
1145 }
1146
1147 static int vfio_fill_vconfig_bytes(struct vfio_pci_device *vdev,
1148 int offset, int size)
1149 {
1150 struct pci_dev *pdev = vdev->pdev;
1151 int ret = 0;
1152
1153 /*
1154 * We try to read physical config space in the largest chunks
1155 * we can, assuming that all of the fields support dword access.
1156 * pci_save_state() makes this same assumption and seems to do ok.
1157 */
1158 while (size) {
1159 int filled;
1160
1161 if (size >= 4 && !(offset % 4)) {
1162 __le32 *dwordp = (__le32 *)&vdev->vconfig[offset];
1163 u32 dword;
1164
1165 ret = pci_read_config_dword(pdev, offset, &dword);
1166 if (ret)
1167 return ret;
1168 *dwordp = cpu_to_le32(dword);
1169 filled = 4;
1170 } else if (size >= 2 && !(offset % 2)) {
1171 __le16 *wordp = (__le16 *)&vdev->vconfig[offset];
1172 u16 word;
1173
1174 ret = pci_read_config_word(pdev, offset, &word);
1175 if (ret)
1176 return ret;
1177 *wordp = cpu_to_le16(word);
1178 filled = 2;
1179 } else {
1180 u8 *byte = &vdev->vconfig[offset];
1181 ret = pci_read_config_byte(pdev, offset, byte);
1182 if (ret)
1183 return ret;
1184 filled = 1;
1185 }
1186
1187 offset += filled;
1188 size -= filled;
1189 }
1190
1191 return ret;
1192 }
1193
1194 static int vfio_cap_init(struct vfio_pci_device *vdev)
1195 {
1196 struct pci_dev *pdev = vdev->pdev;
1197 u8 *map = vdev->pci_config_map;
1198 u16 status;
1199 u8 pos, *prev, cap;
1200 int loops, ret, caps = 0;
1201
1202 /* Any capabilities? */
1203 ret = pci_read_config_word(pdev, PCI_STATUS, &status);
1204 if (ret)
1205 return ret;
1206
1207 if (!(status & PCI_STATUS_CAP_LIST))
1208 return 0; /* Done */
1209
1210 ret = pci_read_config_byte(pdev, PCI_CAPABILITY_LIST, &pos);
1211 if (ret)
1212 return ret;
1213
1214 /* Mark the previous position in case we want to skip a capability */
1215 prev = &vdev->vconfig[PCI_CAPABILITY_LIST];
1216
1217 /* We can bound our loop, capabilities are dword aligned */
1218 loops = (PCI_CFG_SPACE_SIZE - PCI_STD_HEADER_SIZEOF) / PCI_CAP_SIZEOF;
1219 while (pos && loops--) {
1220 u8 next;
1221 int i, len = 0;
1222
1223 ret = pci_read_config_byte(pdev, pos, &cap);
1224 if (ret)
1225 return ret;
1226
1227 ret = pci_read_config_byte(pdev,
1228 pos + PCI_CAP_LIST_NEXT, &next);
1229 if (ret)
1230 return ret;
1231
1232 if (cap <= PCI_CAP_ID_MAX) {
1233 len = pci_cap_length[cap];
1234 if (len == 0xFF) { /* Variable length */
1235 len = vfio_cap_len(vdev, cap, pos);
1236 if (len < 0)
1237 return len;
1238 }
1239 }
1240
1241 if (!len) {
1242 pr_info("%s: %s hiding cap 0x%x\n",
1243 __func__, dev_name(&pdev->dev), cap);
1244 *prev = next;
1245 pos = next;
1246 continue;
1247 }
1248
1249 /* Sanity check, do we overlap other capabilities? */
1250 for (i = 0; i < len; i++) {
1251 if (likely(map[pos + i] == PCI_CAP_ID_INVALID))
1252 continue;
1253
1254 pr_warn("%s: %s pci config conflict @0x%x, was cap 0x%x now cap 0x%x\n",
1255 __func__, dev_name(&pdev->dev),
1256 pos + i, map[pos + i], cap);
1257 }
1258
1259 memset(map + pos, cap, len);
1260 ret = vfio_fill_vconfig_bytes(vdev, pos, len);
1261 if (ret)
1262 return ret;
1263
1264 prev = &vdev->vconfig[pos + PCI_CAP_LIST_NEXT];
1265 pos = next;
1266 caps++;
1267 }
1268
1269 /* If we didn't fill any capabilities, clear the status flag */
1270 if (!caps) {
1271 __le16 *vstatus = (__le16 *)&vdev->vconfig[PCI_STATUS];
1272 *vstatus &= ~cpu_to_le16(PCI_STATUS_CAP_LIST);
1273 }
1274
1275 return 0;
1276 }
1277
1278 static int vfio_ecap_init(struct vfio_pci_device *vdev)
1279 {
1280 struct pci_dev *pdev = vdev->pdev;
1281 u8 *map = vdev->pci_config_map;
1282 u16 epos;
1283 __le32 *prev = NULL;
1284 int loops, ret, ecaps = 0;
1285
1286 if (!vdev->extended_caps)
1287 return 0;
1288
1289 epos = PCI_CFG_SPACE_SIZE;
1290
1291 loops = (pdev->cfg_size - PCI_CFG_SPACE_SIZE) / PCI_CAP_SIZEOF;
1292
1293 while (loops-- && epos >= PCI_CFG_SPACE_SIZE) {
1294 u32 header;
1295 u16 ecap;
1296 int i, len = 0;
1297 bool hidden = false;
1298
1299 ret = pci_read_config_dword(pdev, epos, &header);
1300 if (ret)
1301 return ret;
1302
1303 ecap = PCI_EXT_CAP_ID(header);
1304
1305 if (ecap <= PCI_EXT_CAP_ID_MAX) {
1306 len = pci_ext_cap_length[ecap];
1307 if (len == 0xFF) {
1308 len = vfio_ext_cap_len(vdev, ecap, epos);
1309 if (len < 0)
1310 return ret;
1311 }
1312 }
1313
1314 if (!len) {
1315 pr_info("%s: %s hiding ecap 0x%x@0x%x\n",
1316 __func__, dev_name(&pdev->dev), ecap, epos);
1317
1318 /* If not the first in the chain, we can skip over it */
1319 if (prev) {
1320 u32 val = epos = PCI_EXT_CAP_NEXT(header);
1321 *prev &= cpu_to_le32(~(0xffcU << 20));
1322 *prev |= cpu_to_le32(val << 20);
1323 continue;
1324 }
1325
1326 /*
1327 * Otherwise, fill in a placeholder, the direct
1328 * readfn will virtualize this automatically
1329 */
1330 len = PCI_CAP_SIZEOF;
1331 hidden = true;
1332 }
1333
1334 for (i = 0; i < len; i++) {
1335 if (likely(map[epos + i] == PCI_CAP_ID_INVALID))
1336 continue;
1337
1338 pr_warn("%s: %s pci config conflict @0x%x, was ecap 0x%x now ecap 0x%x\n",
1339 __func__, dev_name(&pdev->dev),
1340 epos + i, map[epos + i], ecap);
1341 }
1342
1343 /*
1344 * Even though ecap is 2 bytes, we're currently a long way
1345 * from exceeding 1 byte capabilities. If we ever make it
1346 * up to 0xFF we'll need to up this to a two-byte, byte map.
1347 */
1348 BUILD_BUG_ON(PCI_EXT_CAP_ID_MAX >= PCI_CAP_ID_INVALID);
1349
1350 memset(map + epos, ecap, len);
1351 ret = vfio_fill_vconfig_bytes(vdev, epos, len);
1352 if (ret)
1353 return ret;
1354
1355 /*
1356 * If we're just using this capability to anchor the list,
1357 * hide the real ID. Only count real ecaps. XXX PCI spec
1358 * indicates to use cap id = 0, version = 0, next = 0 if
1359 * ecaps are absent, hope users check all the way to next.
1360 */
1361 if (hidden)
1362 *(__le32 *)&vdev->vconfig[epos] &=
1363 cpu_to_le32((0xffcU << 20));
1364 else
1365 ecaps++;
1366
1367 prev = (__le32 *)&vdev->vconfig[epos];
1368 epos = PCI_EXT_CAP_NEXT(header);
1369 }
1370
1371 if (!ecaps)
1372 *(u32 *)&vdev->vconfig[PCI_CFG_SPACE_SIZE] = 0;
1373
1374 return 0;
1375 }
1376
1377 /*
1378 * For each device we allocate a pci_config_map that indicates the
1379 * capability occupying each dword and thus the struct perm_bits we
1380 * use for read and write. We also allocate a virtualized config
1381 * space which tracks reads and writes to bits that we emulate for
1382 * the user. Initial values filled from device.
1383 *
1384 * Using shared stuct perm_bits between all vfio-pci devices saves
1385 * us from allocating cfg_size buffers for virt and write for every
1386 * device. We could remove vconfig and allocate individual buffers
1387 * for each area requring emulated bits, but the array of pointers
1388 * would be comparable in size (at least for standard config space).
1389 */
1390 int vfio_config_init(struct vfio_pci_device *vdev)
1391 {
1392 struct pci_dev *pdev = vdev->pdev;
1393 u8 *map, *vconfig;
1394 int ret;
1395
1396 /*
1397 * Config space, caps and ecaps are all dword aligned, so we could
1398 * use one byte per dword to record the type. However, there are
1399 * no requiremenst on the length of a capability, so the gap between
1400 * capabilities needs byte granularity.
1401 */
1402 map = kmalloc(pdev->cfg_size, GFP_KERNEL);
1403 if (!map)
1404 return -ENOMEM;
1405
1406 vconfig = kmalloc(pdev->cfg_size, GFP_KERNEL);
1407 if (!vconfig) {
1408 kfree(map);
1409 return -ENOMEM;
1410 }
1411
1412 vdev->pci_config_map = map;
1413 vdev->vconfig = vconfig;
1414
1415 memset(map, PCI_CAP_ID_BASIC, PCI_STD_HEADER_SIZEOF);
1416 memset(map + PCI_STD_HEADER_SIZEOF, PCI_CAP_ID_INVALID,
1417 pdev->cfg_size - PCI_STD_HEADER_SIZEOF);
1418
1419 ret = vfio_fill_vconfig_bytes(vdev, 0, PCI_STD_HEADER_SIZEOF);
1420 if (ret)
1421 goto out;
1422
1423 vdev->bardirty = true;
1424
1425 /*
1426 * XXX can we just pci_load_saved_state/pci_restore_state?
1427 * may need to rebuild vconfig after that
1428 */
1429
1430 /* For restore after reset */
1431 vdev->rbar[0] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_0]);
1432 vdev->rbar[1] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_1]);
1433 vdev->rbar[2] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_2]);
1434 vdev->rbar[3] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_3]);
1435 vdev->rbar[4] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_4]);
1436 vdev->rbar[5] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_5]);
1437 vdev->rbar[6] = le32_to_cpu(*(__le32 *)&vconfig[PCI_ROM_ADDRESS]);
1438
1439 if (pdev->is_virtfn) {
1440 *(__le16 *)&vconfig[PCI_VENDOR_ID] = cpu_to_le16(pdev->vendor);
1441 *(__le16 *)&vconfig[PCI_DEVICE_ID] = cpu_to_le16(pdev->device);
1442 }
1443
1444 ret = vfio_cap_init(vdev);
1445 if (ret)
1446 goto out;
1447
1448 ret = vfio_ecap_init(vdev);
1449 if (ret)
1450 goto out;
1451
1452 return 0;
1453
1454 out:
1455 kfree(map);
1456 vdev->pci_config_map = NULL;
1457 kfree(vconfig);
1458 vdev->vconfig = NULL;
1459 return pcibios_err_to_errno(ret);
1460 }
1461
1462 void vfio_config_free(struct vfio_pci_device *vdev)
1463 {
1464 kfree(vdev->vconfig);
1465 vdev->vconfig = NULL;
1466 kfree(vdev->pci_config_map);
1467 vdev->pci_config_map = NULL;
1468 kfree(vdev->msi_perm);
1469 vdev->msi_perm = NULL;
1470 }
1471
1472 /*
1473 * Find the remaining number of bytes in a dword that match the given
1474 * position. Stop at either the end of the capability or the dword boundary.
1475 */
1476 static size_t vfio_pci_cap_remaining_dword(struct vfio_pci_device *vdev,
1477 loff_t pos)
1478 {
1479 u8 cap = vdev->pci_config_map[pos];
1480 size_t i;
1481
1482 for (i = 1; (pos + i) % 4 && vdev->pci_config_map[pos + i] == cap; i++)
1483 /* nop */;
1484
1485 return i;
1486 }
1487
1488 static ssize_t vfio_config_do_rw(struct vfio_pci_device *vdev, char __user *buf,
1489 size_t count, loff_t *ppos, bool iswrite)
1490 {
1491 struct pci_dev *pdev = vdev->pdev;
1492 struct perm_bits *perm;
1493 __le32 val = 0;
1494 int cap_start = 0, offset;
1495 u8 cap_id;
1496 ssize_t ret;
1497
1498 if (*ppos < 0 || *ppos >= pdev->cfg_size ||
1499 *ppos + count > pdev->cfg_size)
1500 return -EFAULT;
1501
1502 /*
1503 * Chop accesses into aligned chunks containing no more than a
1504 * single capability. Caller increments to the next chunk.
1505 */
1506 count = min(count, vfio_pci_cap_remaining_dword(vdev, *ppos));
1507 if (count >= 4 && !(*ppos % 4))
1508 count = 4;
1509 else if (count >= 2 && !(*ppos % 2))
1510 count = 2;
1511 else
1512 count = 1;
1513
1514 ret = count;
1515
1516 cap_id = vdev->pci_config_map[*ppos];
1517
1518 if (cap_id == PCI_CAP_ID_INVALID) {
1519 perm = &unassigned_perms;
1520 cap_start = *ppos;
1521 } else {
1522 if (*ppos >= PCI_CFG_SPACE_SIZE) {
1523 WARN_ON(cap_id > PCI_EXT_CAP_ID_MAX);
1524
1525 perm = &ecap_perms[cap_id];
1526 cap_start = vfio_find_cap_start(vdev, *ppos);
1527 } else {
1528 WARN_ON(cap_id > PCI_CAP_ID_MAX);
1529
1530 perm = &cap_perms[cap_id];
1531
1532 if (cap_id == PCI_CAP_ID_MSI)
1533 perm = vdev->msi_perm;
1534
1535 if (cap_id > PCI_CAP_ID_BASIC)
1536 cap_start = vfio_find_cap_start(vdev, *ppos);
1537 }
1538 }
1539
1540 WARN_ON(!cap_start && cap_id != PCI_CAP_ID_BASIC);
1541 WARN_ON(cap_start > *ppos);
1542
1543 offset = *ppos - cap_start;
1544
1545 if (iswrite) {
1546 if (!perm->writefn)
1547 return ret;
1548
1549 if (copy_from_user(&val, buf, count))
1550 return -EFAULT;
1551
1552 ret = perm->writefn(vdev, *ppos, count, perm, offset, val);
1553 } else {
1554 if (perm->readfn) {
1555 ret = perm->readfn(vdev, *ppos, count,
1556 perm, offset, &val);
1557 if (ret < 0)
1558 return ret;
1559 }
1560
1561 if (copy_to_user(buf, &val, count))
1562 return -EFAULT;
1563 }
1564
1565 return ret;
1566 }
1567
1568 ssize_t vfio_pci_config_rw(struct vfio_pci_device *vdev, char __user *buf,
1569 size_t count, loff_t *ppos, bool iswrite)
1570 {
1571 size_t done = 0;
1572 int ret = 0;
1573 loff_t pos = *ppos;
1574
1575 pos &= VFIO_PCI_OFFSET_MASK;
1576
1577 while (count) {
1578 ret = vfio_config_do_rw(vdev, buf, count, &pos, iswrite);
1579 if (ret < 0)
1580 return ret;
1581
1582 count -= ret;
1583 done += ret;
1584 buf += ret;
1585 pos += ret;
1586 }
1587
1588 *ppos += done;
1589
1590 return done;
1591 }
Linus' kernel tree