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 /* Fake capability ID for standard config space */
35 #define PCI_CAP_ID_BASIC	0
36 
37 #define is_bar(offset)	\
38 	((offset >= PCI_BASE_ADDRESS_0 && offset < PCI_BASE_ADDRESS_5 + 4) || \
39 	 (offset >= PCI_ROM_ADDRESS && offset < PCI_ROM_ADDRESS + 4))
40 
41 /*
42  * Lengths of PCI Config Capabilities
43  *   0: Removed from the user visible capability list
44  *   FF: Variable length
45  */
46 static const u8 pci_cap_length[PCI_CAP_ID_MAX + 1] = {
47 	[PCI_CAP_ID_BASIC]	= PCI_STD_HEADER_SIZEOF, /* pci config header */
48 	[PCI_CAP_ID_PM]		= PCI_PM_SIZEOF,
49 	[PCI_CAP_ID_AGP]	= PCI_AGP_SIZEOF,
50 	[PCI_CAP_ID_VPD]	= PCI_CAP_VPD_SIZEOF,
51 	[PCI_CAP_ID_SLOTID]	= 0,		/* bridge - don't care */
52 	[PCI_CAP_ID_MSI]	= 0xFF,		/* 10, 14, 20, or 24 */
53 	[PCI_CAP_ID_CHSWP]	= 0,		/* cpci - not yet */
54 	[PCI_CAP_ID_PCIX]	= 0xFF,		/* 8 or 24 */
55 	[PCI_CAP_ID_HT]		= 0xFF,		/* hypertransport */
56 	[PCI_CAP_ID_VNDR]	= 0xFF,		/* variable */
57 	[PCI_CAP_ID_DBG]	= 0,		/* debug - don't care */
58 	[PCI_CAP_ID_CCRC]	= 0,		/* cpci - not yet */
59 	[PCI_CAP_ID_SHPC]	= 0,		/* hotswap - not yet */
60 	[PCI_CAP_ID_SSVID]	= 0,		/* bridge - don't care */
61 	[PCI_CAP_ID_AGP3]	= 0,		/* AGP8x - not yet */
62 	[PCI_CAP_ID_SECDEV]	= 0,		/* secure device not yet */
63 	[PCI_CAP_ID_EXP]	= 0xFF,		/* 20 or 44 */
64 	[PCI_CAP_ID_MSIX]	= PCI_CAP_MSIX_SIZEOF,
65 	[PCI_CAP_ID_SATA]	= 0xFF,
66 	[PCI_CAP_ID_AF]		= PCI_CAP_AF_SIZEOF,
67 };
68 
69 /*
70  * Lengths of PCIe/PCI-X Extended Config Capabilities
71  *   0: Removed or masked from the user visible capability list
72  *   FF: Variable length
73  */
74 static const u16 pci_ext_cap_length[PCI_EXT_CAP_ID_MAX + 1] = {
75 	[PCI_EXT_CAP_ID_ERR]	=	PCI_ERR_ROOT_COMMAND,
76 	[PCI_EXT_CAP_ID_VC]	=	0xFF,
77 	[PCI_EXT_CAP_ID_DSN]	=	PCI_EXT_CAP_DSN_SIZEOF,
78 	[PCI_EXT_CAP_ID_PWR]	=	PCI_EXT_CAP_PWR_SIZEOF,
79 	[PCI_EXT_CAP_ID_RCLD]	=	0,	/* root only - don't care */
80 	[PCI_EXT_CAP_ID_RCILC]	=	0,	/* root only - don't care */
81 	[PCI_EXT_CAP_ID_RCEC]	=	0,	/* root only - don't care */
82 	[PCI_EXT_CAP_ID_MFVC]	=	0xFF,
83 	[PCI_EXT_CAP_ID_VC9]	=	0xFF,	/* same as CAP_ID_VC */
84 	[PCI_EXT_CAP_ID_RCRB]	=	0,	/* root only - don't care */
85 	[PCI_EXT_CAP_ID_VNDR]	=	0xFF,
86 	[PCI_EXT_CAP_ID_CAC]	=	0,	/* obsolete */
87 	[PCI_EXT_CAP_ID_ACS]	=	0xFF,
88 	[PCI_EXT_CAP_ID_ARI]	=	PCI_EXT_CAP_ARI_SIZEOF,
89 	[PCI_EXT_CAP_ID_ATS]	=	PCI_EXT_CAP_ATS_SIZEOF,
90 	[PCI_EXT_CAP_ID_SRIOV]	=	PCI_EXT_CAP_SRIOV_SIZEOF,
91 	[PCI_EXT_CAP_ID_MRIOV]	=	0,	/* not yet */
92 	[PCI_EXT_CAP_ID_MCAST]	=	PCI_EXT_CAP_MCAST_ENDPOINT_SIZEOF,
93 	[PCI_EXT_CAP_ID_PRI]	=	PCI_EXT_CAP_PRI_SIZEOF,
94 	[PCI_EXT_CAP_ID_AMD_XXX] =	0,	/* not yet */
95 	[PCI_EXT_CAP_ID_REBAR]	=	0xFF,
96 	[PCI_EXT_CAP_ID_DPA]	=	0xFF,
97 	[PCI_EXT_CAP_ID_TPH]	=	0xFF,
98 	[PCI_EXT_CAP_ID_LTR]	=	PCI_EXT_CAP_LTR_SIZEOF,
99 	[PCI_EXT_CAP_ID_SECPCI]	=	0,	/* not yet */
100 	[PCI_EXT_CAP_ID_PMUX]	=	0,	/* not yet */
101 	[PCI_EXT_CAP_ID_PASID]	=	0,	/* not yet */
102 };
103 
104 /*
105  * Read/Write Permission Bits - one bit for each bit in capability
106  * Any field can be read if it exists, but what is read depends on
107  * whether the field is 'virtualized', or just pass thru to the
108  * hardware.  Any virtualized field is also virtualized for writes.
109  * Writes are only permitted if they have a 1 bit here.
110  */
111 struct perm_bits {
112 	u8	*virt;		/* read/write virtual data, not hw */
113 	u8	*write;		/* writeable bits */
114 	int	(*readfn)(struct vfio_pci_device *vdev, int pos, int count,
115 			  struct perm_bits *perm, int offset, __le32 *val);
116 	int	(*writefn)(struct vfio_pci_device *vdev, int pos, int count,
117 			   struct perm_bits *perm, int offset, __le32 val);
118 };
119 
120 #define	NO_VIRT		0
121 #define	ALL_VIRT	0xFFFFFFFFU
122 #define	NO_WRITE	0
123 #define	ALL_WRITE	0xFFFFFFFFU
124 
125 static int vfio_user_config_read(struct pci_dev *pdev, int offset,
126 				 __le32 *val, int count)
127 {
128 	int ret = -EINVAL;
129 	u32 tmp_val = 0;
130 
131 	switch (count) {
132 	case 1:
133 	{
134 		u8 tmp;
135 		ret = pci_user_read_config_byte(pdev, offset, &tmp);
136 		tmp_val = tmp;
137 		break;
138 	}
139 	case 2:
140 	{
141 		u16 tmp;
142 		ret = pci_user_read_config_word(pdev, offset, &tmp);
143 		tmp_val = tmp;
144 		break;
145 	}
146 	case 4:
147 		ret = pci_user_read_config_dword(pdev, offset, &tmp_val);
148 		break;
149 	}
150 
151 	*val = cpu_to_le32(tmp_val);
152 
153 	return ret;
154 }
155 
156 static int vfio_user_config_write(struct pci_dev *pdev, int offset,
157 				  __le32 val, int count)
158 {
159 	int ret = -EINVAL;
160 	u32 tmp_val = le32_to_cpu(val);
161 
162 	switch (count) {
163 	case 1:
164 		ret = pci_user_write_config_byte(pdev, offset, tmp_val);
165 		break;
166 	case 2:
167 		ret = pci_user_write_config_word(pdev, offset, tmp_val);
168 		break;
169 	case 4:
170 		ret = pci_user_write_config_dword(pdev, offset, tmp_val);
171 		break;
172 	}
173 
174 	return ret;
175 }
176 
177 static int vfio_default_config_read(struct vfio_pci_device *vdev, int pos,
178 				    int count, struct perm_bits *perm,
179 				    int offset, __le32 *val)
180 {
181 	__le32 virt = 0;
182 
183 	memcpy(val, vdev->vconfig + pos, count);
184 
185 	memcpy(&virt, perm->virt + offset, count);
186 
187 	/* Any non-virtualized bits? */
188 	if (cpu_to_le32(~0U >> (32 - (count * 8))) != virt) {
189 		struct pci_dev *pdev = vdev->pdev;
190 		__le32 phys_val = 0;
191 		int ret;
192 
193 		ret = vfio_user_config_read(pdev, pos, &phys_val, count);
194 		if (ret)
195 			return ret;
196 
197 		*val = (phys_val & ~virt) | (*val & virt);
198 	}
199 
200 	return count;
201 }
202 
203 static int vfio_default_config_write(struct vfio_pci_device *vdev, int pos,
204 				     int count, struct perm_bits *perm,
205 				     int offset, __le32 val)
206 {
207 	__le32 virt = 0, write = 0;
208 
209 	memcpy(&write, perm->write + offset, count);
210 
211 	if (!write)
212 		return count; /* drop, no writable bits */
213 
214 	memcpy(&virt, perm->virt + offset, count);
215 
216 	/* Virtualized and writable bits go to vconfig */
217 	if (write & virt) {
218 		__le32 virt_val = 0;
219 
220 		memcpy(&virt_val, vdev->vconfig + pos, count);
221 
222 		virt_val &= ~(write & virt);
223 		virt_val |= (val & (write & virt));
224 
225 		memcpy(vdev->vconfig + pos, &virt_val, count);
226 	}
227 
228 	/* Non-virtualzed and writable bits go to hardware */
229 	if (write & ~virt) {
230 		struct pci_dev *pdev = vdev->pdev;
231 		__le32 phys_val = 0;
232 		int ret;
233 
234 		ret = vfio_user_config_read(pdev, pos, &phys_val, count);
235 		if (ret)
236 			return ret;
237 
238 		phys_val &= ~(write & ~virt);
239 		phys_val |= (val & (write & ~virt));
240 
241 		ret = vfio_user_config_write(pdev, pos, phys_val, count);
242 		if (ret)
243 			return ret;
244 	}
245 
246 	return count;
247 }
248 
249 /* Allow direct read from hardware, except for capability next pointer */
250 static int vfio_direct_config_read(struct vfio_pci_device *vdev, int pos,
251 				   int count, struct perm_bits *perm,
252 				   int offset, __le32 *val)
253 {
254 	int ret;
255 
256 	ret = vfio_user_config_read(vdev->pdev, pos, val, count);
257 	if (ret)
258 		return ret;
259 
260 	if (pos >= PCI_CFG_SPACE_SIZE) { /* Extended cap header mangling */
261 		if (offset < 4)
262 			memcpy(val, vdev->vconfig + pos, count);
263 	} else if (pos >= PCI_STD_HEADER_SIZEOF) { /* Std cap mangling */
264 		if (offset == PCI_CAP_LIST_ID && count > 1)
265 			memcpy(val, vdev->vconfig + pos,
266 			       min(PCI_CAP_FLAGS, count));
267 		else if (offset == PCI_CAP_LIST_NEXT)
268 			memcpy(val, vdev->vconfig + pos, 1);
269 	}
270 
271 	return count;
272 }
273 
274 /* Raw access skips any kind of virtualization */
275 static int vfio_raw_config_write(struct vfio_pci_device *vdev, int pos,
276 				 int count, struct perm_bits *perm,
277 				 int offset, __le32 val)
278 {
279 	int ret;
280 
281 	ret = vfio_user_config_write(vdev->pdev, pos, val, count);
282 	if (ret)
283 		return ret;
284 
285 	return count;
286 }
287 
288 static int vfio_raw_config_read(struct vfio_pci_device *vdev, int pos,
289 				int count, struct perm_bits *perm,
290 				int offset, __le32 *val)
291 {
292 	int ret;
293 
294 	ret = vfio_user_config_read(vdev->pdev, pos, val, count);
295 	if (ret)
296 		return ret;
297 
298 	return count;
299 }
300 
301 /* Virt access uses only virtualization */
302 static int vfio_virt_config_write(struct vfio_pci_device *vdev, int pos,
303 				  int count, struct perm_bits *perm,
304 				  int offset, __le32 val)
305 {
306 	memcpy(vdev->vconfig + pos, &val, count);
307 	return count;
308 }
309 
310 static int vfio_virt_config_read(struct vfio_pci_device *vdev, int pos,
311 				 int count, struct perm_bits *perm,
312 				 int offset, __le32 *val)
313 {
314 	memcpy(val, vdev->vconfig + pos, count);
315 	return count;
316 }
317 
318 /* Default capability regions to read-only, no-virtualization */
319 static struct perm_bits cap_perms[PCI_CAP_ID_MAX + 1] = {
320 	[0 ... PCI_CAP_ID_MAX] = { .readfn = vfio_direct_config_read }
321 };
322 static struct perm_bits ecap_perms[PCI_EXT_CAP_ID_MAX + 1] = {
323 	[0 ... PCI_EXT_CAP_ID_MAX] = { .readfn = vfio_direct_config_read }
324 };
325 /*
326  * Default unassigned regions to raw read-write access.  Some devices
327  * require this to function as they hide registers between the gaps in
328  * config space (be2net).  Like MMIO and I/O port registers, we have
329  * to trust the hardware isolation.
330  */
331 static struct perm_bits unassigned_perms = {
332 	.readfn = vfio_raw_config_read,
333 	.writefn = vfio_raw_config_write
334 };
335 
336 static struct perm_bits virt_perms = {
337 	.readfn = vfio_virt_config_read,
338 	.writefn = vfio_virt_config_write
339 };
340 
341 static void free_perm_bits(struct perm_bits *perm)
342 {
343 	kfree(perm->virt);
344 	kfree(perm->write);
345 	perm->virt = NULL;
346 	perm->write = NULL;
347 }
348 
349 static int alloc_perm_bits(struct perm_bits *perm, int size)
350 {
351 	/*
352 	 * Round up all permission bits to the next dword, this lets us
353 	 * ignore whether a read/write exceeds the defined capability
354 	 * structure.  We can do this because:
355 	 *  - Standard config space is already dword aligned
356 	 *  - Capabilities are all dword aligned (bits 0:1 of next reserved)
357 	 *  - Express capabilities defined as dword aligned
358 	 */
359 	size = round_up(size, 4);
360 
361 	/*
362 	 * Zero state is
363 	 * - All Readable, None Writeable, None Virtualized
364 	 */
365 	perm->virt = kzalloc(size, GFP_KERNEL);
366 	perm->write = kzalloc(size, GFP_KERNEL);
367 	if (!perm->virt || !perm->write) {
368 		free_perm_bits(perm);
369 		return -ENOMEM;
370 	}
371 
372 	perm->readfn = vfio_default_config_read;
373 	perm->writefn = vfio_default_config_write;
374 
375 	return 0;
376 }
377 
378 /*
379  * Helper functions for filling in permission tables
380  */
381 static inline void p_setb(struct perm_bits *p, int off, u8 virt, u8 write)
382 {
383 	p->virt[off] = virt;
384 	p->write[off] = write;
385 }
386 
387 /* Handle endian-ness - pci and tables are little-endian */
388 static inline void p_setw(struct perm_bits *p, int off, u16 virt, u16 write)
389 {
390 	*(__le16 *)(&p->virt[off]) = cpu_to_le16(virt);
391 	*(__le16 *)(&p->write[off]) = cpu_to_le16(write);
392 }
393 
394 /* Handle endian-ness - pci and tables are little-endian */
395 static inline void p_setd(struct perm_bits *p, int off, u32 virt, u32 write)
396 {
397 	*(__le32 *)(&p->virt[off]) = cpu_to_le32(virt);
398 	*(__le32 *)(&p->write[off]) = cpu_to_le32(write);
399 }
400 
401 /*
402  * Restore the *real* BARs after we detect a FLR or backdoor reset.
403  * (backdoor = some device specific technique that we didn't catch)
404  */
405 static void vfio_bar_restore(struct vfio_pci_device *vdev)
406 {
407 	struct pci_dev *pdev = vdev->pdev;
408 	u32 *rbar = vdev->rbar;
409 	u16 cmd;
410 	int i;
411 
412 	if (pdev->is_virtfn)
413 		return;
414 
415 	pci_info(pdev, "%s: reset recovery - restoring BARs\n", __func__);
416 
417 	for (i = PCI_BASE_ADDRESS_0; i <= PCI_BASE_ADDRESS_5; i += 4, rbar++)
418 		pci_user_write_config_dword(pdev, i, *rbar);
419 
420 	pci_user_write_config_dword(pdev, PCI_ROM_ADDRESS, *rbar);
421 
422 	if (vdev->nointx) {
423 		pci_user_read_config_word(pdev, PCI_COMMAND, &cmd);
424 		cmd |= PCI_COMMAND_INTX_DISABLE;
425 		pci_user_write_config_word(pdev, PCI_COMMAND, cmd);
426 	}
427 }
428 
429 static __le32 vfio_generate_bar_flags(struct pci_dev *pdev, int bar)
430 {
431 	unsigned long flags = pci_resource_flags(pdev, bar);
432 	u32 val;
433 
434 	if (flags & IORESOURCE_IO)
435 		return cpu_to_le32(PCI_BASE_ADDRESS_SPACE_IO);
436 
437 	val = PCI_BASE_ADDRESS_SPACE_MEMORY;
438 
439 	if (flags & IORESOURCE_PREFETCH)
440 		val |= PCI_BASE_ADDRESS_MEM_PREFETCH;
441 
442 	if (flags & IORESOURCE_MEM_64)
443 		val |= PCI_BASE_ADDRESS_MEM_TYPE_64;
444 
445 	return cpu_to_le32(val);
446 }
447 
448 /*
449  * Pretend we're hardware and tweak the values of the *virtual* PCI BARs
450  * to reflect the hardware capabilities.  This implements BAR sizing.
451  */
452 static void vfio_bar_fixup(struct vfio_pci_device *vdev)
453 {
454 	struct pci_dev *pdev = vdev->pdev;
455 	int i;
456 	__le32 *bar;
457 	u64 mask;
458 
459 	bar = (__le32 *)&vdev->vconfig[PCI_BASE_ADDRESS_0];
460 
461 	for (i = PCI_STD_RESOURCES; i <= PCI_STD_RESOURCE_END; i++, bar++) {
462 		if (!pci_resource_start(pdev, i)) {
463 			*bar = 0; /* Unmapped by host = unimplemented to user */
464 			continue;
465 		}
466 
467 		mask = ~(pci_resource_len(pdev, i) - 1);
468 
469 		*bar &= cpu_to_le32((u32)mask);
470 		*bar |= vfio_generate_bar_flags(pdev, i);
471 
472 		if (*bar & cpu_to_le32(PCI_BASE_ADDRESS_MEM_TYPE_64)) {
473 			bar++;
474 			*bar &= cpu_to_le32((u32)(mask >> 32));
475 			i++;
476 		}
477 	}
478 
479 	bar = (__le32 *)&vdev->vconfig[PCI_ROM_ADDRESS];
480 
481 	/*
482 	 * NB. REGION_INFO will have reported zero size if we weren't able
483 	 * to read the ROM, but we still return the actual BAR size here if
484 	 * it exists (or the shadow ROM space).
485 	 */
486 	if (pci_resource_start(pdev, PCI_ROM_RESOURCE)) {
487 		mask = ~(pci_resource_len(pdev, PCI_ROM_RESOURCE) - 1);
488 		mask |= PCI_ROM_ADDRESS_ENABLE;
489 		*bar &= cpu_to_le32((u32)mask);
490 	} else if (pdev->resource[PCI_ROM_RESOURCE].flags &
491 					IORESOURCE_ROM_SHADOW) {
492 		mask = ~(0x20000 - 1);
493 		mask |= PCI_ROM_ADDRESS_ENABLE;
494 		*bar &= cpu_to_le32((u32)mask);
495 	} else
496 		*bar = 0;
497 
498 	vdev->bardirty = false;
499 }
500 
501 static int vfio_basic_config_read(struct vfio_pci_device *vdev, int pos,
502 				  int count, struct perm_bits *perm,
503 				  int offset, __le32 *val)
504 {
505 	if (is_bar(offset)) /* pos == offset for basic config */
506 		vfio_bar_fixup(vdev);
507 
508 	count = vfio_default_config_read(vdev, pos, count, perm, offset, val);
509 
510 	/* Mask in virtual memory enable for SR-IOV devices */
511 	if (offset == PCI_COMMAND && vdev->pdev->is_virtfn) {
512 		u16 cmd = le16_to_cpu(*(__le16 *)&vdev->vconfig[PCI_COMMAND]);
513 		u32 tmp_val = le32_to_cpu(*val);
514 
515 		tmp_val |= cmd & PCI_COMMAND_MEMORY;
516 		*val = cpu_to_le32(tmp_val);
517 	}
518 
519 	return count;
520 }
521 
522 /* Test whether BARs match the value we think they should contain */
523 static bool vfio_need_bar_restore(struct vfio_pci_device *vdev)
524 {
525 	int i = 0, pos = PCI_BASE_ADDRESS_0, ret;
526 	u32 bar;
527 
528 	for (; pos <= PCI_BASE_ADDRESS_5; i++, pos += 4) {
529 		if (vdev->rbar[i]) {
530 			ret = pci_user_read_config_dword(vdev->pdev, pos, &bar);
531 			if (ret || vdev->rbar[i] != bar)
532 				return true;
533 		}
534 	}
535 
536 	return false;
537 }
538 
539 static int vfio_basic_config_write(struct vfio_pci_device *vdev, int pos,
540 				   int count, struct perm_bits *perm,
541 				   int offset, __le32 val)
542 {
543 	struct pci_dev *pdev = vdev->pdev;
544 	__le16 *virt_cmd;
545 	u16 new_cmd = 0;
546 	int ret;
547 
548 	virt_cmd = (__le16 *)&vdev->vconfig[PCI_COMMAND];
549 
550 	if (offset == PCI_COMMAND) {
551 		bool phys_mem, virt_mem, new_mem, phys_io, virt_io, new_io;
552 		u16 phys_cmd;
553 
554 		ret = pci_user_read_config_word(pdev, PCI_COMMAND, &phys_cmd);
555 		if (ret)
556 			return ret;
557 
558 		new_cmd = le32_to_cpu(val);
559 
560 		phys_mem = !!(phys_cmd & PCI_COMMAND_MEMORY);
561 		virt_mem = !!(le16_to_cpu(*virt_cmd) & PCI_COMMAND_MEMORY);
562 		new_mem = !!(new_cmd & PCI_COMMAND_MEMORY);
563 
564 		phys_io = !!(phys_cmd & PCI_COMMAND_IO);
565 		virt_io = !!(le16_to_cpu(*virt_cmd) & PCI_COMMAND_IO);
566 		new_io = !!(new_cmd & PCI_COMMAND_IO);
567 
568 		/*
569 		 * If the user is writing mem/io enable (new_mem/io) and we
570 		 * think it's already enabled (virt_mem/io), but the hardware
571 		 * shows it disabled (phys_mem/io, then the device has
572 		 * undergone some kind of backdoor reset and needs to be
573 		 * restored before we allow it to enable the bars.
574 		 * SR-IOV devices will trigger this, but we catch them later
575 		 */
576 		if ((new_mem && virt_mem && !phys_mem) ||
577 		    (new_io && virt_io && !phys_io) ||
578 		    vfio_need_bar_restore(vdev))
579 			vfio_bar_restore(vdev);
580 	}
581 
582 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
583 	if (count < 0)
584 		return count;
585 
586 	/*
587 	 * Save current memory/io enable bits in vconfig to allow for
588 	 * the test above next time.
589 	 */
590 	if (offset == PCI_COMMAND) {
591 		u16 mask = PCI_COMMAND_MEMORY | PCI_COMMAND_IO;
592 
593 		*virt_cmd &= cpu_to_le16(~mask);
594 		*virt_cmd |= cpu_to_le16(new_cmd & mask);
595 	}
596 
597 	/* Emulate INTx disable */
598 	if (offset >= PCI_COMMAND && offset <= PCI_COMMAND + 1) {
599 		bool virt_intx_disable;
600 
601 		virt_intx_disable = !!(le16_to_cpu(*virt_cmd) &
602 				       PCI_COMMAND_INTX_DISABLE);
603 
604 		if (virt_intx_disable && !vdev->virq_disabled) {
605 			vdev->virq_disabled = true;
606 			vfio_pci_intx_mask(vdev);
607 		} else if (!virt_intx_disable && vdev->virq_disabled) {
608 			vdev->virq_disabled = false;
609 			vfio_pci_intx_unmask(vdev);
610 		}
611 	}
612 
613 	if (is_bar(offset))
614 		vdev->bardirty = true;
615 
616 	return count;
617 }
618 
619 /* Permissions for the Basic PCI Header */
620 static int __init init_pci_cap_basic_perm(struct perm_bits *perm)
621 {
622 	if (alloc_perm_bits(perm, PCI_STD_HEADER_SIZEOF))
623 		return -ENOMEM;
624 
625 	perm->readfn = vfio_basic_config_read;
626 	perm->writefn = vfio_basic_config_write;
627 
628 	/* Virtualized for SR-IOV functions, which just have FFFF */
629 	p_setw(perm, PCI_VENDOR_ID, (u16)ALL_VIRT, NO_WRITE);
630 	p_setw(perm, PCI_DEVICE_ID, (u16)ALL_VIRT, NO_WRITE);
631 
632 	/*
633 	 * Virtualize INTx disable, we use it internally for interrupt
634 	 * control and can emulate it for non-PCI 2.3 devices.
635 	 */
636 	p_setw(perm, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE, (u16)ALL_WRITE);
637 
638 	/* Virtualize capability list, we might want to skip/disable */
639 	p_setw(perm, PCI_STATUS, PCI_STATUS_CAP_LIST, NO_WRITE);
640 
641 	/* No harm to write */
642 	p_setb(perm, PCI_CACHE_LINE_SIZE, NO_VIRT, (u8)ALL_WRITE);
643 	p_setb(perm, PCI_LATENCY_TIMER, NO_VIRT, (u8)ALL_WRITE);
644 	p_setb(perm, PCI_BIST, NO_VIRT, (u8)ALL_WRITE);
645 
646 	/* Virtualize all bars, can't touch the real ones */
647 	p_setd(perm, PCI_BASE_ADDRESS_0, ALL_VIRT, ALL_WRITE);
648 	p_setd(perm, PCI_BASE_ADDRESS_1, ALL_VIRT, ALL_WRITE);
649 	p_setd(perm, PCI_BASE_ADDRESS_2, ALL_VIRT, ALL_WRITE);
650 	p_setd(perm, PCI_BASE_ADDRESS_3, ALL_VIRT, ALL_WRITE);
651 	p_setd(perm, PCI_BASE_ADDRESS_4, ALL_VIRT, ALL_WRITE);
652 	p_setd(perm, PCI_BASE_ADDRESS_5, ALL_VIRT, ALL_WRITE);
653 	p_setd(perm, PCI_ROM_ADDRESS, ALL_VIRT, ALL_WRITE);
654 
655 	/* Allow us to adjust capability chain */
656 	p_setb(perm, PCI_CAPABILITY_LIST, (u8)ALL_VIRT, NO_WRITE);
657 
658 	/* Sometimes used by sw, just virtualize */
659 	p_setb(perm, PCI_INTERRUPT_LINE, (u8)ALL_VIRT, (u8)ALL_WRITE);
660 
661 	/* Virtualize interrupt pin to allow hiding INTx */
662 	p_setb(perm, PCI_INTERRUPT_PIN, (u8)ALL_VIRT, (u8)NO_WRITE);
663 
664 	return 0;
665 }
666 
667 static int vfio_pm_config_write(struct vfio_pci_device *vdev, int pos,
668 				int count, struct perm_bits *perm,
669 				int offset, __le32 val)
670 {
671 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
672 	if (count < 0)
673 		return count;
674 
675 	if (offset == PCI_PM_CTRL) {
676 		pci_power_t state;
677 
678 		switch (le32_to_cpu(val) & PCI_PM_CTRL_STATE_MASK) {
679 		case 0:
680 			state = PCI_D0;
681 			break;
682 		case 1:
683 			state = PCI_D1;
684 			break;
685 		case 2:
686 			state = PCI_D2;
687 			break;
688 		case 3:
689 			state = PCI_D3hot;
690 			break;
691 		}
692 
693 		vfio_pci_set_power_state(vdev, state);
694 	}
695 
696 	return count;
697 }
698 
699 /* Permissions for the Power Management capability */
700 static int __init init_pci_cap_pm_perm(struct perm_bits *perm)
701 {
702 	if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_PM]))
703 		return -ENOMEM;
704 
705 	perm->writefn = vfio_pm_config_write;
706 
707 	/*
708 	 * We always virtualize the next field so we can remove
709 	 * capabilities from the chain if we want to.
710 	 */
711 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
712 
713 	/*
714 	 * Power management is defined *per function*, so we can let
715 	 * the user change power state, but we trap and initiate the
716 	 * change ourselves, so the state bits are read-only.
717 	 */
718 	p_setd(perm, PCI_PM_CTRL, NO_VIRT, ~PCI_PM_CTRL_STATE_MASK);
719 	return 0;
720 }
721 
722 static int vfio_vpd_config_write(struct vfio_pci_device *vdev, int pos,
723 				 int count, struct perm_bits *perm,
724 				 int offset, __le32 val)
725 {
726 	struct pci_dev *pdev = vdev->pdev;
727 	__le16 *paddr = (__le16 *)(vdev->vconfig + pos - offset + PCI_VPD_ADDR);
728 	__le32 *pdata = (__le32 *)(vdev->vconfig + pos - offset + PCI_VPD_DATA);
729 	u16 addr;
730 	u32 data;
731 
732 	/*
733 	 * Write through to emulation.  If the write includes the upper byte
734 	 * of PCI_VPD_ADDR, then the PCI_VPD_ADDR_F bit is written and we
735 	 * have work to do.
736 	 */
737 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
738 	if (count < 0 || offset > PCI_VPD_ADDR + 1 ||
739 	    offset + count <= PCI_VPD_ADDR + 1)
740 		return count;
741 
742 	addr = le16_to_cpu(*paddr);
743 
744 	if (addr & PCI_VPD_ADDR_F) {
745 		data = le32_to_cpu(*pdata);
746 		if (pci_write_vpd(pdev, addr & ~PCI_VPD_ADDR_F, 4, &data) != 4)
747 			return count;
748 	} else {
749 		data = 0;
750 		if (pci_read_vpd(pdev, addr, 4, &data) < 0)
751 			return count;
752 		*pdata = cpu_to_le32(data);
753 	}
754 
755 	/*
756 	 * Toggle PCI_VPD_ADDR_F in the emulated PCI_VPD_ADDR register to
757 	 * signal completion.  If an error occurs above, we assume that not
758 	 * toggling this bit will induce a driver timeout.
759 	 */
760 	addr ^= PCI_VPD_ADDR_F;
761 	*paddr = cpu_to_le16(addr);
762 
763 	return count;
764 }
765 
766 /* Permissions for Vital Product Data capability */
767 static int __init init_pci_cap_vpd_perm(struct perm_bits *perm)
768 {
769 	if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_VPD]))
770 		return -ENOMEM;
771 
772 	perm->writefn = vfio_vpd_config_write;
773 
774 	/*
775 	 * We always virtualize the next field so we can remove
776 	 * capabilities from the chain if we want to.
777 	 */
778 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
779 
780 	/*
781 	 * Both the address and data registers are virtualized to
782 	 * enable access through the pci_vpd_read/write functions
783 	 */
784 	p_setw(perm, PCI_VPD_ADDR, (u16)ALL_VIRT, (u16)ALL_WRITE);
785 	p_setd(perm, PCI_VPD_DATA, ALL_VIRT, ALL_WRITE);
786 
787 	return 0;
788 }
789 
790 /* Permissions for PCI-X capability */
791 static int __init init_pci_cap_pcix_perm(struct perm_bits *perm)
792 {
793 	/* Alloc 24, but only 8 are used in v0 */
794 	if (alloc_perm_bits(perm, PCI_CAP_PCIX_SIZEOF_V2))
795 		return -ENOMEM;
796 
797 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
798 
799 	p_setw(perm, PCI_X_CMD, NO_VIRT, (u16)ALL_WRITE);
800 	p_setd(perm, PCI_X_ECC_CSR, NO_VIRT, ALL_WRITE);
801 	return 0;
802 }
803 
804 static int vfio_exp_config_write(struct vfio_pci_device *vdev, int pos,
805 				 int count, struct perm_bits *perm,
806 				 int offset, __le32 val)
807 {
808 	__le16 *ctrl = (__le16 *)(vdev->vconfig + pos -
809 				  offset + PCI_EXP_DEVCTL);
810 	int readrq = le16_to_cpu(*ctrl) & PCI_EXP_DEVCTL_READRQ;
811 
812 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
813 	if (count < 0)
814 		return count;
815 
816 	/*
817 	 * The FLR bit is virtualized, if set and the device supports PCIe
818 	 * FLR, issue a reset_function.  Regardless, clear the bit, the spec
819 	 * requires it to be always read as zero.  NB, reset_function might
820 	 * not use a PCIe FLR, we don't have that level of granularity.
821 	 */
822 	if (*ctrl & cpu_to_le16(PCI_EXP_DEVCTL_BCR_FLR)) {
823 		u32 cap;
824 		int ret;
825 
826 		*ctrl &= ~cpu_to_le16(PCI_EXP_DEVCTL_BCR_FLR);
827 
828 		ret = pci_user_read_config_dword(vdev->pdev,
829 						 pos - offset + PCI_EXP_DEVCAP,
830 						 &cap);
831 
832 		if (!ret && (cap & PCI_EXP_DEVCAP_FLR))
833 			pci_try_reset_function(vdev->pdev);
834 	}
835 
836 	/*
837 	 * MPS is virtualized to the user, writes do not change the physical
838 	 * register since determining a proper MPS value requires a system wide
839 	 * device view.  The MRRS is largely independent of MPS, but since the
840 	 * user does not have that system-wide view, they might set a safe, but
841 	 * inefficiently low value.  Here we allow writes through to hardware,
842 	 * but we set the floor to the physical device MPS setting, so that
843 	 * we can at least use full TLPs, as defined by the MPS value.
844 	 *
845 	 * NB, if any devices actually depend on an artificially low MRRS
846 	 * setting, this will need to be revisited, perhaps with a quirk
847 	 * though pcie_set_readrq().
848 	 */
849 	if (readrq != (le16_to_cpu(*ctrl) & PCI_EXP_DEVCTL_READRQ)) {
850 		readrq = 128 <<
851 			((le16_to_cpu(*ctrl) & PCI_EXP_DEVCTL_READRQ) >> 12);
852 		readrq = max(readrq, pcie_get_mps(vdev->pdev));
853 
854 		pcie_set_readrq(vdev->pdev, readrq);
855 	}
856 
857 	return count;
858 }
859 
860 /* Permissions for PCI Express capability */
861 static int __init init_pci_cap_exp_perm(struct perm_bits *perm)
862 {
863 	/* Alloc largest of possible sizes */
864 	if (alloc_perm_bits(perm, PCI_CAP_EXP_ENDPOINT_SIZEOF_V2))
865 		return -ENOMEM;
866 
867 	perm->writefn = vfio_exp_config_write;
868 
869 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
870 
871 	/*
872 	 * Allow writes to device control fields, except devctl_phantom,
873 	 * which could confuse IOMMU, MPS, which can break communication
874 	 * with other physical devices, and the ARI bit in devctl2, which
875 	 * is set at probe time.  FLR and MRRS get virtualized via our
876 	 * writefn.
877 	 */
878 	p_setw(perm, PCI_EXP_DEVCTL,
879 	       PCI_EXP_DEVCTL_BCR_FLR | PCI_EXP_DEVCTL_PAYLOAD |
880 	       PCI_EXP_DEVCTL_READRQ, ~PCI_EXP_DEVCTL_PHANTOM);
881 	p_setw(perm, PCI_EXP_DEVCTL2, NO_VIRT, ~PCI_EXP_DEVCTL2_ARI);
882 	return 0;
883 }
884 
885 static int vfio_af_config_write(struct vfio_pci_device *vdev, int pos,
886 				int count, struct perm_bits *perm,
887 				int offset, __le32 val)
888 {
889 	u8 *ctrl = vdev->vconfig + pos - offset + PCI_AF_CTRL;
890 
891 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
892 	if (count < 0)
893 		return count;
894 
895 	/*
896 	 * The FLR bit is virtualized, if set and the device supports AF
897 	 * FLR, issue a reset_function.  Regardless, clear the bit, the spec
898 	 * requires it to be always read as zero.  NB, reset_function might
899 	 * not use an AF FLR, we don't have that level of granularity.
900 	 */
901 	if (*ctrl & PCI_AF_CTRL_FLR) {
902 		u8 cap;
903 		int ret;
904 
905 		*ctrl &= ~PCI_AF_CTRL_FLR;
906 
907 		ret = pci_user_read_config_byte(vdev->pdev,
908 						pos - offset + PCI_AF_CAP,
909 						&cap);
910 
911 		if (!ret && (cap & PCI_AF_CAP_FLR) && (cap & PCI_AF_CAP_TP))
912 			pci_try_reset_function(vdev->pdev);
913 	}
914 
915 	return count;
916 }
917 
918 /* Permissions for Advanced Function capability */
919 static int __init init_pci_cap_af_perm(struct perm_bits *perm)
920 {
921 	if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_AF]))
922 		return -ENOMEM;
923 
924 	perm->writefn = vfio_af_config_write;
925 
926 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
927 	p_setb(perm, PCI_AF_CTRL, PCI_AF_CTRL_FLR, PCI_AF_CTRL_FLR);
928 	return 0;
929 }
930 
931 /* Permissions for Advanced Error Reporting extended capability */
932 static int __init init_pci_ext_cap_err_perm(struct perm_bits *perm)
933 {
934 	u32 mask;
935 
936 	if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_ERR]))
937 		return -ENOMEM;
938 
939 	/*
940 	 * Virtualize the first dword of all express capabilities
941 	 * because it includes the next pointer.  This lets us later
942 	 * remove capabilities from the chain if we need to.
943 	 */
944 	p_setd(perm, 0, ALL_VIRT, NO_WRITE);
945 
946 	/* Writable bits mask */
947 	mask =	PCI_ERR_UNC_UND |		/* Undefined */
948 		PCI_ERR_UNC_DLP |		/* Data Link Protocol */
949 		PCI_ERR_UNC_SURPDN |		/* Surprise Down */
950 		PCI_ERR_UNC_POISON_TLP |	/* Poisoned TLP */
951 		PCI_ERR_UNC_FCP |		/* Flow Control Protocol */
952 		PCI_ERR_UNC_COMP_TIME |		/* Completion Timeout */
953 		PCI_ERR_UNC_COMP_ABORT |	/* Completer Abort */
954 		PCI_ERR_UNC_UNX_COMP |		/* Unexpected Completion */
955 		PCI_ERR_UNC_RX_OVER |		/* Receiver Overflow */
956 		PCI_ERR_UNC_MALF_TLP |		/* Malformed TLP */
957 		PCI_ERR_UNC_ECRC |		/* ECRC Error Status */
958 		PCI_ERR_UNC_UNSUP |		/* Unsupported Request */
959 		PCI_ERR_UNC_ACSV |		/* ACS Violation */
960 		PCI_ERR_UNC_INTN |		/* internal error */
961 		PCI_ERR_UNC_MCBTLP |		/* MC blocked TLP */
962 		PCI_ERR_UNC_ATOMEG |		/* Atomic egress blocked */
963 		PCI_ERR_UNC_TLPPRE;		/* TLP prefix blocked */
964 	p_setd(perm, PCI_ERR_UNCOR_STATUS, NO_VIRT, mask);
965 	p_setd(perm, PCI_ERR_UNCOR_MASK, NO_VIRT, mask);
966 	p_setd(perm, PCI_ERR_UNCOR_SEVER, NO_VIRT, mask);
967 
968 	mask =	PCI_ERR_COR_RCVR |		/* Receiver Error Status */
969 		PCI_ERR_COR_BAD_TLP |		/* Bad TLP Status */
970 		PCI_ERR_COR_BAD_DLLP |		/* Bad DLLP Status */
971 		PCI_ERR_COR_REP_ROLL |		/* REPLAY_NUM Rollover */
972 		PCI_ERR_COR_REP_TIMER |		/* Replay Timer Timeout */
973 		PCI_ERR_COR_ADV_NFAT |		/* Advisory Non-Fatal */
974 		PCI_ERR_COR_INTERNAL |		/* Corrected Internal */
975 		PCI_ERR_COR_LOG_OVER;		/* Header Log Overflow */
976 	p_setd(perm, PCI_ERR_COR_STATUS, NO_VIRT, mask);
977 	p_setd(perm, PCI_ERR_COR_MASK, NO_VIRT, mask);
978 
979 	mask =	PCI_ERR_CAP_ECRC_GENE |		/* ECRC Generation Enable */
980 		PCI_ERR_CAP_ECRC_CHKE;		/* ECRC Check Enable */
981 	p_setd(perm, PCI_ERR_CAP, NO_VIRT, mask);
982 	return 0;
983 }
984 
985 /* Permissions for Power Budgeting extended capability */
986 static int __init init_pci_ext_cap_pwr_perm(struct perm_bits *perm)
987 {
988 	if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_PWR]))
989 		return -ENOMEM;
990 
991 	p_setd(perm, 0, ALL_VIRT, NO_WRITE);
992 
993 	/* Writing the data selector is OK, the info is still read-only */
994 	p_setb(perm, PCI_PWR_DATA, NO_VIRT, (u8)ALL_WRITE);
995 	return 0;
996 }
997 
998 /*
999  * Initialize the shared permission tables
1000  */
1001 void vfio_pci_uninit_perm_bits(void)
1002 {
1003 	free_perm_bits(&cap_perms[PCI_CAP_ID_BASIC]);
1004 
1005 	free_perm_bits(&cap_perms[PCI_CAP_ID_PM]);
1006 	free_perm_bits(&cap_perms[PCI_CAP_ID_VPD]);
1007 	free_perm_bits(&cap_perms[PCI_CAP_ID_PCIX]);
1008 	free_perm_bits(&cap_perms[PCI_CAP_ID_EXP]);
1009 	free_perm_bits(&cap_perms[PCI_CAP_ID_AF]);
1010 
1011 	free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_ERR]);
1012 	free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_PWR]);
1013 }
1014 
1015 int __init vfio_pci_init_perm_bits(void)
1016 {
1017 	int ret;
1018 
1019 	/* Basic config space */
1020 	ret = init_pci_cap_basic_perm(&cap_perms[PCI_CAP_ID_BASIC]);
1021 
1022 	/* Capabilities */
1023 	ret |= init_pci_cap_pm_perm(&cap_perms[PCI_CAP_ID_PM]);
1024 	ret |= init_pci_cap_vpd_perm(&cap_perms[PCI_CAP_ID_VPD]);
1025 	ret |= init_pci_cap_pcix_perm(&cap_perms[PCI_CAP_ID_PCIX]);
1026 	cap_perms[PCI_CAP_ID_VNDR].writefn = vfio_raw_config_write;
1027 	ret |= init_pci_cap_exp_perm(&cap_perms[PCI_CAP_ID_EXP]);
1028 	ret |= init_pci_cap_af_perm(&cap_perms[PCI_CAP_ID_AF]);
1029 
1030 	/* Extended capabilities */
1031 	ret |= init_pci_ext_cap_err_perm(&ecap_perms[PCI_EXT_CAP_ID_ERR]);
1032 	ret |= init_pci_ext_cap_pwr_perm(&ecap_perms[PCI_EXT_CAP_ID_PWR]);
1033 	ecap_perms[PCI_EXT_CAP_ID_VNDR].writefn = vfio_raw_config_write;
1034 
1035 	if (ret)
1036 		vfio_pci_uninit_perm_bits();
1037 
1038 	return ret;
1039 }
1040 
1041 static int vfio_find_cap_start(struct vfio_pci_device *vdev, int pos)
1042 {
1043 	u8 cap;
1044 	int base = (pos >= PCI_CFG_SPACE_SIZE) ? PCI_CFG_SPACE_SIZE :
1045 						 PCI_STD_HEADER_SIZEOF;
1046 	cap = vdev->pci_config_map[pos];
1047 
1048 	if (cap == PCI_CAP_ID_BASIC)
1049 		return 0;
1050 
1051 	/* XXX Can we have to abutting capabilities of the same type? */
1052 	while (pos - 1 >= base && vdev->pci_config_map[pos - 1] == cap)
1053 		pos--;
1054 
1055 	return pos;
1056 }
1057 
1058 static int vfio_msi_config_read(struct vfio_pci_device *vdev, int pos,
1059 				int count, struct perm_bits *perm,
1060 				int offset, __le32 *val)
1061 {
1062 	/* Update max available queue size from msi_qmax */
1063 	if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
1064 		__le16 *flags;
1065 		int start;
1066 
1067 		start = vfio_find_cap_start(vdev, pos);
1068 
1069 		flags = (__le16 *)&vdev->vconfig[start];
1070 
1071 		*flags &= cpu_to_le16(~PCI_MSI_FLAGS_QMASK);
1072 		*flags |= cpu_to_le16(vdev->msi_qmax << 1);
1073 	}
1074 
1075 	return vfio_default_config_read(vdev, pos, count, perm, offset, val);
1076 }
1077 
1078 static int vfio_msi_config_write(struct vfio_pci_device *vdev, int pos,
1079 				 int count, struct perm_bits *perm,
1080 				 int offset, __le32 val)
1081 {
1082 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
1083 	if (count < 0)
1084 		return count;
1085 
1086 	/* Fixup and write configured queue size and enable to hardware */
1087 	if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
1088 		__le16 *pflags;
1089 		u16 flags;
1090 		int start, ret;
1091 
1092 		start = vfio_find_cap_start(vdev, pos);
1093 
1094 		pflags = (__le16 *)&vdev->vconfig[start + PCI_MSI_FLAGS];
1095 
1096 		flags = le16_to_cpu(*pflags);
1097 
1098 		/* MSI is enabled via ioctl */
1099 		if  (!is_msi(vdev))
1100 			flags &= ~PCI_MSI_FLAGS_ENABLE;
1101 
1102 		/* Check queue size */
1103 		if ((flags & PCI_MSI_FLAGS_QSIZE) >> 4 > vdev->msi_qmax) {
1104 			flags &= ~PCI_MSI_FLAGS_QSIZE;
1105 			flags |= vdev->msi_qmax << 4;
1106 		}
1107 
1108 		/* Write back to virt and to hardware */
1109 		*pflags = cpu_to_le16(flags);
1110 		ret = pci_user_write_config_word(vdev->pdev,
1111 						 start + PCI_MSI_FLAGS,
1112 						 flags);
1113 		if (ret)
1114 			return ret;
1115 	}
1116 
1117 	return count;
1118 }
1119 
1120 /*
1121  * MSI determination is per-device, so this routine gets used beyond
1122  * initialization time. Don't add __init
1123  */
1124 static int init_pci_cap_msi_perm(struct perm_bits *perm, int len, u16 flags)
1125 {
1126 	if (alloc_perm_bits(perm, len))
1127 		return -ENOMEM;
1128 
1129 	perm->readfn = vfio_msi_config_read;
1130 	perm->writefn = vfio_msi_config_write;
1131 
1132 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
1133 
1134 	/*
1135 	 * The upper byte of the control register is reserved,
1136 	 * just setup the lower byte.
1137 	 */
1138 	p_setb(perm, PCI_MSI_FLAGS, (u8)ALL_VIRT, (u8)ALL_WRITE);
1139 	p_setd(perm, PCI_MSI_ADDRESS_LO, ALL_VIRT, ALL_WRITE);
1140 	if (flags & PCI_MSI_FLAGS_64BIT) {
1141 		p_setd(perm, PCI_MSI_ADDRESS_HI, ALL_VIRT, ALL_WRITE);
1142 		p_setw(perm, PCI_MSI_DATA_64, (u16)ALL_VIRT, (u16)ALL_WRITE);
1143 		if (flags & PCI_MSI_FLAGS_MASKBIT) {
1144 			p_setd(perm, PCI_MSI_MASK_64, NO_VIRT, ALL_WRITE);
1145 			p_setd(perm, PCI_MSI_PENDING_64, NO_VIRT, ALL_WRITE);
1146 		}
1147 	} else {
1148 		p_setw(perm, PCI_MSI_DATA_32, (u16)ALL_VIRT, (u16)ALL_WRITE);
1149 		if (flags & PCI_MSI_FLAGS_MASKBIT) {
1150 			p_setd(perm, PCI_MSI_MASK_32, NO_VIRT, ALL_WRITE);
1151 			p_setd(perm, PCI_MSI_PENDING_32, NO_VIRT, ALL_WRITE);
1152 		}
1153 	}
1154 	return 0;
1155 }
1156 
1157 /* Determine MSI CAP field length; initialize msi_perms on 1st call per vdev */
1158 static int vfio_msi_cap_len(struct vfio_pci_device *vdev, u8 pos)
1159 {
1160 	struct pci_dev *pdev = vdev->pdev;
1161 	int len, ret;
1162 	u16 flags;
1163 
1164 	ret = pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &flags);
1165 	if (ret)
1166 		return pcibios_err_to_errno(ret);
1167 
1168 	len = 10; /* Minimum size */
1169 	if (flags & PCI_MSI_FLAGS_64BIT)
1170 		len += 4;
1171 	if (flags & PCI_MSI_FLAGS_MASKBIT)
1172 		len += 10;
1173 
1174 	if (vdev->msi_perm)
1175 		return len;
1176 
1177 	vdev->msi_perm = kmalloc(sizeof(struct perm_bits), GFP_KERNEL);
1178 	if (!vdev->msi_perm)
1179 		return -ENOMEM;
1180 
1181 	ret = init_pci_cap_msi_perm(vdev->msi_perm, len, flags);
1182 	if (ret) {
1183 		kfree(vdev->msi_perm);
1184 		return ret;
1185 	}
1186 
1187 	return len;
1188 }
1189 
1190 /* Determine extended capability length for VC (2 & 9) and MFVC */
1191 static int vfio_vc_cap_len(struct vfio_pci_device *vdev, u16 pos)
1192 {
1193 	struct pci_dev *pdev = vdev->pdev;
1194 	u32 tmp;
1195 	int ret, evcc, phases, vc_arb;
1196 	int len = PCI_CAP_VC_BASE_SIZEOF;
1197 
1198 	ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_CAP1, &tmp);
1199 	if (ret)
1200 		return pcibios_err_to_errno(ret);
1201 
1202 	evcc = tmp & PCI_VC_CAP1_EVCC; /* extended vc count */
1203 	ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_CAP2, &tmp);
1204 	if (ret)
1205 		return pcibios_err_to_errno(ret);
1206 
1207 	if (tmp & PCI_VC_CAP2_128_PHASE)
1208 		phases = 128;
1209 	else if (tmp & PCI_VC_CAP2_64_PHASE)
1210 		phases = 64;
1211 	else if (tmp & PCI_VC_CAP2_32_PHASE)
1212 		phases = 32;
1213 	else
1214 		phases = 0;
1215 
1216 	vc_arb = phases * 4;
1217 
1218 	/*
1219 	 * Port arbitration tables are root & switch only;
1220 	 * function arbitration tables are function 0 only.
1221 	 * In either case, we'll never let user write them so
1222 	 * we don't care how big they are
1223 	 */
1224 	len += (1 + evcc) * PCI_CAP_VC_PER_VC_SIZEOF;
1225 	if (vc_arb) {
1226 		len = round_up(len, 16);
1227 		len += vc_arb / 8;
1228 	}
1229 	return len;
1230 }
1231 
1232 static int vfio_cap_len(struct vfio_pci_device *vdev, u8 cap, u8 pos)
1233 {
1234 	struct pci_dev *pdev = vdev->pdev;
1235 	u32 dword;
1236 	u16 word;
1237 	u8 byte;
1238 	int ret;
1239 
1240 	switch (cap) {
1241 	case PCI_CAP_ID_MSI:
1242 		return vfio_msi_cap_len(vdev, pos);
1243 	case PCI_CAP_ID_PCIX:
1244 		ret = pci_read_config_word(pdev, pos + PCI_X_CMD, &word);
1245 		if (ret)
1246 			return pcibios_err_to_errno(ret);
1247 
1248 		if (PCI_X_CMD_VERSION(word)) {
1249 			if (pdev->cfg_size > PCI_CFG_SPACE_SIZE) {
1250 				/* Test for extended capabilities */
1251 				pci_read_config_dword(pdev, PCI_CFG_SPACE_SIZE,
1252 						      &dword);
1253 				vdev->extended_caps = (dword != 0);
1254 			}
1255 			return PCI_CAP_PCIX_SIZEOF_V2;
1256 		} else
1257 			return PCI_CAP_PCIX_SIZEOF_V0;
1258 	case PCI_CAP_ID_VNDR:
1259 		/* length follows next field */
1260 		ret = pci_read_config_byte(pdev, pos + PCI_CAP_FLAGS, &byte);
1261 		if (ret)
1262 			return pcibios_err_to_errno(ret);
1263 
1264 		return byte;
1265 	case PCI_CAP_ID_EXP:
1266 		if (pdev->cfg_size > PCI_CFG_SPACE_SIZE) {
1267 			/* Test for extended capabilities */
1268 			pci_read_config_dword(pdev, PCI_CFG_SPACE_SIZE, &dword);
1269 			vdev->extended_caps = (dword != 0);
1270 		}
1271 
1272 		/* length based on version and type */
1273 		if ((pcie_caps_reg(pdev) & PCI_EXP_FLAGS_VERS) == 1) {
1274 			if (pci_pcie_type(pdev) == PCI_EXP_TYPE_RC_END)
1275 				return 0xc; /* "All Devices" only, no link */
1276 			return PCI_CAP_EXP_ENDPOINT_SIZEOF_V1;
1277 		} else {
1278 			if (pci_pcie_type(pdev) == PCI_EXP_TYPE_RC_END)
1279 				return 0x2c; /* No link */
1280 			return PCI_CAP_EXP_ENDPOINT_SIZEOF_V2;
1281 		}
1282 	case PCI_CAP_ID_HT:
1283 		ret = pci_read_config_byte(pdev, pos + 3, &byte);
1284 		if (ret)
1285 			return pcibios_err_to_errno(ret);
1286 
1287 		return (byte & HT_3BIT_CAP_MASK) ?
1288 			HT_CAP_SIZEOF_SHORT : HT_CAP_SIZEOF_LONG;
1289 	case PCI_CAP_ID_SATA:
1290 		ret = pci_read_config_byte(pdev, pos + PCI_SATA_REGS, &byte);
1291 		if (ret)
1292 			return pcibios_err_to_errno(ret);
1293 
1294 		byte &= PCI_SATA_REGS_MASK;
1295 		if (byte == PCI_SATA_REGS_INLINE)
1296 			return PCI_SATA_SIZEOF_LONG;
1297 		else
1298 			return PCI_SATA_SIZEOF_SHORT;
1299 	default:
1300 		pci_warn(pdev, "%s: unknown length for PCI cap %#x@%#x\n",
1301 			 __func__, cap, pos);
1302 	}
1303 
1304 	return 0;
1305 }
1306 
1307 static int vfio_ext_cap_len(struct vfio_pci_device *vdev, u16 ecap, u16 epos)
1308 {
1309 	struct pci_dev *pdev = vdev->pdev;
1310 	u8 byte;
1311 	u32 dword;
1312 	int ret;
1313 
1314 	switch (ecap) {
1315 	case PCI_EXT_CAP_ID_VNDR:
1316 		ret = pci_read_config_dword(pdev, epos + PCI_VSEC_HDR, &dword);
1317 		if (ret)
1318 			return pcibios_err_to_errno(ret);
1319 
1320 		return dword >> PCI_VSEC_HDR_LEN_SHIFT;
1321 	case PCI_EXT_CAP_ID_VC:
1322 	case PCI_EXT_CAP_ID_VC9:
1323 	case PCI_EXT_CAP_ID_MFVC:
1324 		return vfio_vc_cap_len(vdev, epos);
1325 	case PCI_EXT_CAP_ID_ACS:
1326 		ret = pci_read_config_byte(pdev, epos + PCI_ACS_CAP, &byte);
1327 		if (ret)
1328 			return pcibios_err_to_errno(ret);
1329 
1330 		if (byte & PCI_ACS_EC) {
1331 			int bits;
1332 
1333 			ret = pci_read_config_byte(pdev,
1334 						   epos + PCI_ACS_EGRESS_BITS,
1335 						   &byte);
1336 			if (ret)
1337 				return pcibios_err_to_errno(ret);
1338 
1339 			bits = byte ? round_up(byte, 32) : 256;
1340 			return 8 + (bits / 8);
1341 		}
1342 		return 8;
1343 
1344 	case PCI_EXT_CAP_ID_REBAR:
1345 		ret = pci_read_config_byte(pdev, epos + PCI_REBAR_CTRL, &byte);
1346 		if (ret)
1347 			return pcibios_err_to_errno(ret);
1348 
1349 		byte &= PCI_REBAR_CTRL_NBAR_MASK;
1350 		byte >>= PCI_REBAR_CTRL_NBAR_SHIFT;
1351 
1352 		return 4 + (byte * 8);
1353 	case PCI_EXT_CAP_ID_DPA:
1354 		ret = pci_read_config_byte(pdev, epos + PCI_DPA_CAP, &byte);
1355 		if (ret)
1356 			return pcibios_err_to_errno(ret);
1357 
1358 		byte &= PCI_DPA_CAP_SUBSTATE_MASK;
1359 		return PCI_DPA_BASE_SIZEOF + byte + 1;
1360 	case PCI_EXT_CAP_ID_TPH:
1361 		ret = pci_read_config_dword(pdev, epos + PCI_TPH_CAP, &dword);
1362 		if (ret)
1363 			return pcibios_err_to_errno(ret);
1364 
1365 		if ((dword & PCI_TPH_CAP_LOC_MASK) == PCI_TPH_LOC_CAP) {
1366 			int sts;
1367 
1368 			sts = dword & PCI_TPH_CAP_ST_MASK;
1369 			sts >>= PCI_TPH_CAP_ST_SHIFT;
1370 			return PCI_TPH_BASE_SIZEOF + (sts * 2) + 2;
1371 		}
1372 		return PCI_TPH_BASE_SIZEOF;
1373 	default:
1374 		pci_warn(pdev, "%s: unknown length for PCI ecap %#x@%#x\n",
1375 			 __func__, ecap, epos);
1376 	}
1377 
1378 	return 0;
1379 }
1380 
1381 static int vfio_fill_vconfig_bytes(struct vfio_pci_device *vdev,
1382 				   int offset, int size)
1383 {
1384 	struct pci_dev *pdev = vdev->pdev;
1385 	int ret = 0;
1386 
1387 	/*
1388 	 * We try to read physical config space in the largest chunks
1389 	 * we can, assuming that all of the fields support dword access.
1390 	 * pci_save_state() makes this same assumption and seems to do ok.
1391 	 */
1392 	while (size) {
1393 		int filled;
1394 
1395 		if (size >= 4 && !(offset % 4)) {
1396 			__le32 *dwordp = (__le32 *)&vdev->vconfig[offset];
1397 			u32 dword;
1398 
1399 			ret = pci_read_config_dword(pdev, offset, &dword);
1400 			if (ret)
1401 				return ret;
1402 			*dwordp = cpu_to_le32(dword);
1403 			filled = 4;
1404 		} else if (size >= 2 && !(offset % 2)) {
1405 			__le16 *wordp = (__le16 *)&vdev->vconfig[offset];
1406 			u16 word;
1407 
1408 			ret = pci_read_config_word(pdev, offset, &word);
1409 			if (ret)
1410 				return ret;
1411 			*wordp = cpu_to_le16(word);
1412 			filled = 2;
1413 		} else {
1414 			u8 *byte = &vdev->vconfig[offset];
1415 			ret = pci_read_config_byte(pdev, offset, byte);
1416 			if (ret)
1417 				return ret;
1418 			filled = 1;
1419 		}
1420 
1421 		offset += filled;
1422 		size -= filled;
1423 	}
1424 
1425 	return ret;
1426 }
1427 
1428 static int vfio_cap_init(struct vfio_pci_device *vdev)
1429 {
1430 	struct pci_dev *pdev = vdev->pdev;
1431 	u8 *map = vdev->pci_config_map;
1432 	u16 status;
1433 	u8 pos, *prev, cap;
1434 	int loops, ret, caps = 0;
1435 
1436 	/* Any capabilities? */
1437 	ret = pci_read_config_word(pdev, PCI_STATUS, &status);
1438 	if (ret)
1439 		return ret;
1440 
1441 	if (!(status & PCI_STATUS_CAP_LIST))
1442 		return 0; /* Done */
1443 
1444 	ret = pci_read_config_byte(pdev, PCI_CAPABILITY_LIST, &pos);
1445 	if (ret)
1446 		return ret;
1447 
1448 	/* Mark the previous position in case we want to skip a capability */
1449 	prev = &vdev->vconfig[PCI_CAPABILITY_LIST];
1450 
1451 	/* We can bound our loop, capabilities are dword aligned */
1452 	loops = (PCI_CFG_SPACE_SIZE - PCI_STD_HEADER_SIZEOF) / PCI_CAP_SIZEOF;
1453 	while (pos && loops--) {
1454 		u8 next;
1455 		int i, len = 0;
1456 
1457 		ret = pci_read_config_byte(pdev, pos, &cap);
1458 		if (ret)
1459 			return ret;
1460 
1461 		ret = pci_read_config_byte(pdev,
1462 					   pos + PCI_CAP_LIST_NEXT, &next);
1463 		if (ret)
1464 			return ret;
1465 
1466 		if (cap <= PCI_CAP_ID_MAX) {
1467 			len = pci_cap_length[cap];
1468 			if (len == 0xFF) { /* Variable length */
1469 				len = vfio_cap_len(vdev, cap, pos);
1470 				if (len < 0)
1471 					return len;
1472 			}
1473 		}
1474 
1475 		if (!len) {
1476 			pci_info(pdev, "%s: hiding cap %#x@%#x\n", __func__,
1477 				 cap, pos);
1478 			*prev = next;
1479 			pos = next;
1480 			continue;
1481 		}
1482 
1483 		/* Sanity check, do we overlap other capabilities? */
1484 		for (i = 0; i < len; i++) {
1485 			if (likely(map[pos + i] == PCI_CAP_ID_INVALID))
1486 				continue;
1487 
1488 			pci_warn(pdev, "%s: PCI config conflict @%#x, was cap %#x now cap %#x\n",
1489 				 __func__, pos + i, map[pos + i], cap);
1490 		}
1491 
1492 		BUILD_BUG_ON(PCI_CAP_ID_MAX >= PCI_CAP_ID_INVALID_VIRT);
1493 
1494 		memset(map + pos, cap, len);
1495 		ret = vfio_fill_vconfig_bytes(vdev, pos, len);
1496 		if (ret)
1497 			return ret;
1498 
1499 		prev = &vdev->vconfig[pos + PCI_CAP_LIST_NEXT];
1500 		pos = next;
1501 		caps++;
1502 	}
1503 
1504 	/* If we didn't fill any capabilities, clear the status flag */
1505 	if (!caps) {
1506 		__le16 *vstatus = (__le16 *)&vdev->vconfig[PCI_STATUS];
1507 		*vstatus &= ~cpu_to_le16(PCI_STATUS_CAP_LIST);
1508 	}
1509 
1510 	return 0;
1511 }
1512 
1513 static int vfio_ecap_init(struct vfio_pci_device *vdev)
1514 {
1515 	struct pci_dev *pdev = vdev->pdev;
1516 	u8 *map = vdev->pci_config_map;
1517 	u16 epos;
1518 	__le32 *prev = NULL;
1519 	int loops, ret, ecaps = 0;
1520 
1521 	if (!vdev->extended_caps)
1522 		return 0;
1523 
1524 	epos = PCI_CFG_SPACE_SIZE;
1525 
1526 	loops = (pdev->cfg_size - PCI_CFG_SPACE_SIZE) / PCI_CAP_SIZEOF;
1527 
1528 	while (loops-- && epos >= PCI_CFG_SPACE_SIZE) {
1529 		u32 header;
1530 		u16 ecap;
1531 		int i, len = 0;
1532 		bool hidden = false;
1533 
1534 		ret = pci_read_config_dword(pdev, epos, &header);
1535 		if (ret)
1536 			return ret;
1537 
1538 		ecap = PCI_EXT_CAP_ID(header);
1539 
1540 		if (ecap <= PCI_EXT_CAP_ID_MAX) {
1541 			len = pci_ext_cap_length[ecap];
1542 			if (len == 0xFF) {
1543 				len = vfio_ext_cap_len(vdev, ecap, epos);
1544 				if (len < 0)
1545 					return ret;
1546 			}
1547 		}
1548 
1549 		if (!len) {
1550 			pci_info(pdev, "%s: hiding ecap %#x@%#x\n",
1551 				 __func__, ecap, epos);
1552 
1553 			/* If not the first in the chain, we can skip over it */
1554 			if (prev) {
1555 				u32 val = epos = PCI_EXT_CAP_NEXT(header);
1556 				*prev &= cpu_to_le32(~(0xffcU << 20));
1557 				*prev |= cpu_to_le32(val << 20);
1558 				continue;
1559 			}
1560 
1561 			/*
1562 			 * Otherwise, fill in a placeholder, the direct
1563 			 * readfn will virtualize this automatically
1564 			 */
1565 			len = PCI_CAP_SIZEOF;
1566 			hidden = true;
1567 		}
1568 
1569 		for (i = 0; i < len; i++) {
1570 			if (likely(map[epos + i] == PCI_CAP_ID_INVALID))
1571 				continue;
1572 
1573 			pci_warn(pdev, "%s: PCI config conflict @%#x, was ecap %#x now ecap %#x\n",
1574 				 __func__, epos + i, map[epos + i], ecap);
1575 		}
1576 
1577 		/*
1578 		 * Even though ecap is 2 bytes, we're currently a long way
1579 		 * from exceeding 1 byte capabilities.  If we ever make it
1580 		 * up to 0xFE we'll need to up this to a two-byte, byte map.
1581 		 */
1582 		BUILD_BUG_ON(PCI_EXT_CAP_ID_MAX >= PCI_CAP_ID_INVALID_VIRT);
1583 
1584 		memset(map + epos, ecap, len);
1585 		ret = vfio_fill_vconfig_bytes(vdev, epos, len);
1586 		if (ret)
1587 			return ret;
1588 
1589 		/*
1590 		 * If we're just using this capability to anchor the list,
1591 		 * hide the real ID.  Only count real ecaps.  XXX PCI spec
1592 		 * indicates to use cap id = 0, version = 0, next = 0 if
1593 		 * ecaps are absent, hope users check all the way to next.
1594 		 */
1595 		if (hidden)
1596 			*(__le32 *)&vdev->vconfig[epos] &=
1597 				cpu_to_le32((0xffcU << 20));
1598 		else
1599 			ecaps++;
1600 
1601 		prev = (__le32 *)&vdev->vconfig[epos];
1602 		epos = PCI_EXT_CAP_NEXT(header);
1603 	}
1604 
1605 	if (!ecaps)
1606 		*(u32 *)&vdev->vconfig[PCI_CFG_SPACE_SIZE] = 0;
1607 
1608 	return 0;
1609 }
1610 
1611 /*
1612  * Nag about hardware bugs, hopefully to have vendors fix them, but at least
1613  * to collect a list of dependencies for the VF INTx pin quirk below.
1614  */
1615 static const struct pci_device_id known_bogus_vf_intx_pin[] = {
1616 	{ PCI_DEVICE(PCI_VENDOR_ID_INTEL, 0x270c) },
1617 	{}
1618 };
1619 
1620 /*
1621  * For each device we allocate a pci_config_map that indicates the
1622  * capability occupying each dword and thus the struct perm_bits we
1623  * use for read and write.  We also allocate a virtualized config
1624  * space which tracks reads and writes to bits that we emulate for
1625  * the user.  Initial values filled from device.
1626  *
1627  * Using shared struct perm_bits between all vfio-pci devices saves
1628  * us from allocating cfg_size buffers for virt and write for every
1629  * device.  We could remove vconfig and allocate individual buffers
1630  * for each area requiring emulated bits, but the array of pointers
1631  * would be comparable in size (at least for standard config space).
1632  */
1633 int vfio_config_init(struct vfio_pci_device *vdev)
1634 {
1635 	struct pci_dev *pdev = vdev->pdev;
1636 	u8 *map, *vconfig;
1637 	int ret;
1638 
1639 	/*
1640 	 * Config space, caps and ecaps are all dword aligned, so we could
1641 	 * use one byte per dword to record the type.  However, there are
1642 	 * no requiremenst on the length of a capability, so the gap between
1643 	 * capabilities needs byte granularity.
1644 	 */
1645 	map = kmalloc(pdev->cfg_size, GFP_KERNEL);
1646 	if (!map)
1647 		return -ENOMEM;
1648 
1649 	vconfig = kmalloc(pdev->cfg_size, GFP_KERNEL);
1650 	if (!vconfig) {
1651 		kfree(map);
1652 		return -ENOMEM;
1653 	}
1654 
1655 	vdev->pci_config_map = map;
1656 	vdev->vconfig = vconfig;
1657 
1658 	memset(map, PCI_CAP_ID_BASIC, PCI_STD_HEADER_SIZEOF);
1659 	memset(map + PCI_STD_HEADER_SIZEOF, PCI_CAP_ID_INVALID,
1660 	       pdev->cfg_size - PCI_STD_HEADER_SIZEOF);
1661 
1662 	ret = vfio_fill_vconfig_bytes(vdev, 0, PCI_STD_HEADER_SIZEOF);
1663 	if (ret)
1664 		goto out;
1665 
1666 	vdev->bardirty = true;
1667 
1668 	/*
1669 	 * XXX can we just pci_load_saved_state/pci_restore_state?
1670 	 * may need to rebuild vconfig after that
1671 	 */
1672 
1673 	/* For restore after reset */
1674 	vdev->rbar[0] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_0]);
1675 	vdev->rbar[1] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_1]);
1676 	vdev->rbar[2] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_2]);
1677 	vdev->rbar[3] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_3]);
1678 	vdev->rbar[4] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_4]);
1679 	vdev->rbar[5] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_5]);
1680 	vdev->rbar[6] = le32_to_cpu(*(__le32 *)&vconfig[PCI_ROM_ADDRESS]);
1681 
1682 	if (pdev->is_virtfn) {
1683 		*(__le16 *)&vconfig[PCI_VENDOR_ID] = cpu_to_le16(pdev->vendor);
1684 		*(__le16 *)&vconfig[PCI_DEVICE_ID] = cpu_to_le16(pdev->device);
1685 
1686 		/*
1687 		 * Per SR-IOV spec rev 1.1, 3.4.1.18 the interrupt pin register
1688 		 * does not apply to VFs and VFs must implement this register
1689 		 * as read-only with value zero.  Userspace is not readily able
1690 		 * to identify whether a device is a VF and thus that the pin
1691 		 * definition on the device is bogus should it violate this
1692 		 * requirement.  We already virtualize the pin register for
1693 		 * other purposes, so we simply need to replace the bogus value
1694 		 * and consider VFs when we determine INTx IRQ count.
1695 		 */
1696 		if (vconfig[PCI_INTERRUPT_PIN] &&
1697 		    !pci_match_id(known_bogus_vf_intx_pin, pdev))
1698 			pci_warn(pdev,
1699 				 "Hardware bug: VF reports bogus INTx pin %d\n",
1700 				 vconfig[PCI_INTERRUPT_PIN]);
1701 
1702 		vconfig[PCI_INTERRUPT_PIN] = 0; /* Gratuitous for good VFs */
1703 	}
1704 
1705 	if (!IS_ENABLED(CONFIG_VFIO_PCI_INTX) || vdev->nointx)
1706 		vconfig[PCI_INTERRUPT_PIN] = 0;
1707 
1708 	ret = vfio_cap_init(vdev);
1709 	if (ret)
1710 		goto out;
1711 
1712 	ret = vfio_ecap_init(vdev);
1713 	if (ret)
1714 		goto out;
1715 
1716 	return 0;
1717 
1718 out:
1719 	kfree(map);
1720 	vdev->pci_config_map = NULL;
1721 	kfree(vconfig);
1722 	vdev->vconfig = NULL;
1723 	return pcibios_err_to_errno(ret);
1724 }
1725 
1726 void vfio_config_free(struct vfio_pci_device *vdev)
1727 {
1728 	kfree(vdev->vconfig);
1729 	vdev->vconfig = NULL;
1730 	kfree(vdev->pci_config_map);
1731 	vdev->pci_config_map = NULL;
1732 	kfree(vdev->msi_perm);
1733 	vdev->msi_perm = NULL;
1734 }
1735 
1736 /*
1737  * Find the remaining number of bytes in a dword that match the given
1738  * position.  Stop at either the end of the capability or the dword boundary.
1739  */
1740 static size_t vfio_pci_cap_remaining_dword(struct vfio_pci_device *vdev,
1741 					   loff_t pos)
1742 {
1743 	u8 cap = vdev->pci_config_map[pos];
1744 	size_t i;
1745 
1746 	for (i = 1; (pos + i) % 4 && vdev->pci_config_map[pos + i] == cap; i++)
1747 		/* nop */;
1748 
1749 	return i;
1750 }
1751 
1752 static ssize_t vfio_config_do_rw(struct vfio_pci_device *vdev, char __user *buf,
1753 				 size_t count, loff_t *ppos, bool iswrite)
1754 {
1755 	struct pci_dev *pdev = vdev->pdev;
1756 	struct perm_bits *perm;
1757 	__le32 val = 0;
1758 	int cap_start = 0, offset;
1759 	u8 cap_id;
1760 	ssize_t ret;
1761 
1762 	if (*ppos < 0 || *ppos >= pdev->cfg_size ||
1763 	    *ppos + count > pdev->cfg_size)
1764 		return -EFAULT;
1765 
1766 	/*
1767 	 * Chop accesses into aligned chunks containing no more than a
1768 	 * single capability.  Caller increments to the next chunk.
1769 	 */
1770 	count = min(count, vfio_pci_cap_remaining_dword(vdev, *ppos));
1771 	if (count >= 4 && !(*ppos % 4))
1772 		count = 4;
1773 	else if (count >= 2 && !(*ppos % 2))
1774 		count = 2;
1775 	else
1776 		count = 1;
1777 
1778 	ret = count;
1779 
1780 	cap_id = vdev->pci_config_map[*ppos];
1781 
1782 	if (cap_id == PCI_CAP_ID_INVALID) {
1783 		perm = &unassigned_perms;
1784 		cap_start = *ppos;
1785 	} else if (cap_id == PCI_CAP_ID_INVALID_VIRT) {
1786 		perm = &virt_perms;
1787 		cap_start = *ppos;
1788 	} else {
1789 		if (*ppos >= PCI_CFG_SPACE_SIZE) {
1790 			WARN_ON(cap_id > PCI_EXT_CAP_ID_MAX);
1791 
1792 			perm = &ecap_perms[cap_id];
1793 			cap_start = vfio_find_cap_start(vdev, *ppos);
1794 		} else {
1795 			WARN_ON(cap_id > PCI_CAP_ID_MAX);
1796 
1797 			perm = &cap_perms[cap_id];
1798 
1799 			if (cap_id == PCI_CAP_ID_MSI)
1800 				perm = vdev->msi_perm;
1801 
1802 			if (cap_id > PCI_CAP_ID_BASIC)
1803 				cap_start = vfio_find_cap_start(vdev, *ppos);
1804 		}
1805 	}
1806 
1807 	WARN_ON(!cap_start && cap_id != PCI_CAP_ID_BASIC);
1808 	WARN_ON(cap_start > *ppos);
1809 
1810 	offset = *ppos - cap_start;
1811 
1812 	if (iswrite) {
1813 		if (!perm->writefn)
1814 			return ret;
1815 
1816 		if (copy_from_user(&val, buf, count))
1817 			return -EFAULT;
1818 
1819 		ret = perm->writefn(vdev, *ppos, count, perm, offset, val);
1820 	} else {
1821 		if (perm->readfn) {
1822 			ret = perm->readfn(vdev, *ppos, count,
1823 					   perm, offset, &val);
1824 			if (ret < 0)
1825 				return ret;
1826 		}
1827 
1828 		if (copy_to_user(buf, &val, count))
1829 			return -EFAULT;
1830 	}
1831 
1832 	return ret;
1833 }
1834 
1835 ssize_t vfio_pci_config_rw(struct vfio_pci_device *vdev, char __user *buf,
1836 			   size_t count, loff_t *ppos, bool iswrite)
1837 {
1838 	size_t done = 0;
1839 	int ret = 0;
1840 	loff_t pos = *ppos;
1841 
1842 	pos &= VFIO_PCI_OFFSET_MASK;
1843 
1844 	while (count) {
1845 		ret = vfio_config_do_rw(vdev, buf, count, &pos, iswrite);
1846 		if (ret < 0)
1847 			return ret;
1848 
1849 		count -= ret;
1850 		done += ret;
1851 		buf += ret;
1852 		pos += ret;
1853 	}
1854 
1855 	*ppos += done;
1856 
1857 	return done;
1858 }
1859