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