1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VFIO PCI config space virtualization
4  *
5  * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
6  *     Author: Alex Williamson <alex.williamson@redhat.com>
7  *
8  * Derived from original vfio:
9  * Copyright 2010 Cisco Systems, Inc.  All rights reserved.
10  * Author: Tom Lyon, pugs@cisco.com
11  */
12 
13 /*
14  * This code handles reading and writing of PCI configuration registers.
15  * This is hairy because we want to allow a lot of flexibility to the
16  * user driver, but cannot trust it with all of the config fields.
17  * Tables determine which fields can be read and written, as well as
18  * which fields are 'virtualized' - special actions and translations to
19  * make it appear to the user that he has control, when in fact things
20  * must be negotiated with the underlying OS.
21  */
22 
23 #include <linux/fs.h>
24 #include <linux/pci.h>
25 #include <linux/uaccess.h>
26 #include <linux/vfio.h>
27 #include <linux/slab.h>
28 
29 #include "vfio_pci_private.h"
30 
31 /* Fake capability ID for standard config space */
32 #define PCI_CAP_ID_BASIC	0
33 
34 #define is_bar(offset)	\
35 	((offset >= PCI_BASE_ADDRESS_0 && offset < PCI_BASE_ADDRESS_5 + 4) || \
36 	 (offset >= PCI_ROM_ADDRESS && offset < PCI_ROM_ADDRESS + 4))
37 
38 /*
39  * Lengths of PCI Config Capabilities
40  *   0: Removed from the user visible capability list
41  *   FF: Variable length
42  */
43 static const u8 pci_cap_length[PCI_CAP_ID_MAX + 1] = {
44 	[PCI_CAP_ID_BASIC]	= PCI_STD_HEADER_SIZEOF, /* pci config header */
45 	[PCI_CAP_ID_PM]		= PCI_PM_SIZEOF,
46 	[PCI_CAP_ID_AGP]	= PCI_AGP_SIZEOF,
47 	[PCI_CAP_ID_VPD]	= PCI_CAP_VPD_SIZEOF,
48 	[PCI_CAP_ID_SLOTID]	= 0,		/* bridge - don't care */
49 	[PCI_CAP_ID_MSI]	= 0xFF,		/* 10, 14, 20, or 24 */
50 	[PCI_CAP_ID_CHSWP]	= 0,		/* cpci - not yet */
51 	[PCI_CAP_ID_PCIX]	= 0xFF,		/* 8 or 24 */
52 	[PCI_CAP_ID_HT]		= 0xFF,		/* hypertransport */
53 	[PCI_CAP_ID_VNDR]	= 0xFF,		/* variable */
54 	[PCI_CAP_ID_DBG]	= 0,		/* debug - don't care */
55 	[PCI_CAP_ID_CCRC]	= 0,		/* cpci - not yet */
56 	[PCI_CAP_ID_SHPC]	= 0,		/* hotswap - not yet */
57 	[PCI_CAP_ID_SSVID]	= 0,		/* bridge - don't care */
58 	[PCI_CAP_ID_AGP3]	= 0,		/* AGP8x - not yet */
59 	[PCI_CAP_ID_SECDEV]	= 0,		/* secure device not yet */
60 	[PCI_CAP_ID_EXP]	= 0xFF,		/* 20 or 44 */
61 	[PCI_CAP_ID_MSIX]	= PCI_CAP_MSIX_SIZEOF,
62 	[PCI_CAP_ID_SATA]	= 0xFF,
63 	[PCI_CAP_ID_AF]		= PCI_CAP_AF_SIZEOF,
64 };
65 
66 /*
67  * Lengths of PCIe/PCI-X Extended Config Capabilities
68  *   0: Removed or masked from the user visible capability list
69  *   FF: Variable length
70  */
71 static const u16 pci_ext_cap_length[PCI_EXT_CAP_ID_MAX + 1] = {
72 	[PCI_EXT_CAP_ID_ERR]	=	PCI_ERR_ROOT_COMMAND,
73 	[PCI_EXT_CAP_ID_VC]	=	0xFF,
74 	[PCI_EXT_CAP_ID_DSN]	=	PCI_EXT_CAP_DSN_SIZEOF,
75 	[PCI_EXT_CAP_ID_PWR]	=	PCI_EXT_CAP_PWR_SIZEOF,
76 	[PCI_EXT_CAP_ID_RCLD]	=	0,	/* root only - don't care */
77 	[PCI_EXT_CAP_ID_RCILC]	=	0,	/* root only - don't care */
78 	[PCI_EXT_CAP_ID_RCEC]	=	0,	/* root only - don't care */
79 	[PCI_EXT_CAP_ID_MFVC]	=	0xFF,
80 	[PCI_EXT_CAP_ID_VC9]	=	0xFF,	/* same as CAP_ID_VC */
81 	[PCI_EXT_CAP_ID_RCRB]	=	0,	/* root only - don't care */
82 	[PCI_EXT_CAP_ID_VNDR]	=	0xFF,
83 	[PCI_EXT_CAP_ID_CAC]	=	0,	/* obsolete */
84 	[PCI_EXT_CAP_ID_ACS]	=	0xFF,
85 	[PCI_EXT_CAP_ID_ARI]	=	PCI_EXT_CAP_ARI_SIZEOF,
86 	[PCI_EXT_CAP_ID_ATS]	=	PCI_EXT_CAP_ATS_SIZEOF,
87 	[PCI_EXT_CAP_ID_SRIOV]	=	PCI_EXT_CAP_SRIOV_SIZEOF,
88 	[PCI_EXT_CAP_ID_MRIOV]	=	0,	/* not yet */
89 	[PCI_EXT_CAP_ID_MCAST]	=	PCI_EXT_CAP_MCAST_ENDPOINT_SIZEOF,
90 	[PCI_EXT_CAP_ID_PRI]	=	PCI_EXT_CAP_PRI_SIZEOF,
91 	[PCI_EXT_CAP_ID_AMD_XXX] =	0,	/* not yet */
92 	[PCI_EXT_CAP_ID_REBAR]	=	0xFF,
93 	[PCI_EXT_CAP_ID_DPA]	=	0xFF,
94 	[PCI_EXT_CAP_ID_TPH]	=	0xFF,
95 	[PCI_EXT_CAP_ID_LTR]	=	PCI_EXT_CAP_LTR_SIZEOF,
96 	[PCI_EXT_CAP_ID_SECPCI]	=	0,	/* not yet */
97 	[PCI_EXT_CAP_ID_PMUX]	=	0,	/* not yet */
98 	[PCI_EXT_CAP_ID_PASID]	=	0,	/* not yet */
99 };
100 
101 /*
102  * Read/Write Permission Bits - one bit for each bit in capability
103  * Any field can be read if it exists, but what is read depends on
104  * whether the field is 'virtualized', or just pass thru to the
105  * hardware.  Any virtualized field is also virtualized for writes.
106  * Writes are only permitted if they have a 1 bit here.
107  */
108 struct perm_bits {
109 	u8	*virt;		/* read/write virtual data, not hw */
110 	u8	*write;		/* writeable bits */
111 	int	(*readfn)(struct vfio_pci_device *vdev, int pos, int count,
112 			  struct perm_bits *perm, int offset, __le32 *val);
113 	int	(*writefn)(struct vfio_pci_device *vdev, int pos, int count,
114 			   struct perm_bits *perm, int offset, __le32 val);
115 };
116 
117 #define	NO_VIRT		0
118 #define	ALL_VIRT	0xFFFFFFFFU
119 #define	NO_WRITE	0
120 #define	ALL_WRITE	0xFFFFFFFFU
121 
122 static int vfio_user_config_read(struct pci_dev *pdev, int offset,
123 				 __le32 *val, int count)
124 {
125 	int ret = -EINVAL;
126 	u32 tmp_val = 0;
127 
128 	switch (count) {
129 	case 1:
130 	{
131 		u8 tmp;
132 		ret = pci_user_read_config_byte(pdev, offset, &tmp);
133 		tmp_val = tmp;
134 		break;
135 	}
136 	case 2:
137 	{
138 		u16 tmp;
139 		ret = pci_user_read_config_word(pdev, offset, &tmp);
140 		tmp_val = tmp;
141 		break;
142 	}
143 	case 4:
144 		ret = pci_user_read_config_dword(pdev, offset, &tmp_val);
145 		break;
146 	}
147 
148 	*val = cpu_to_le32(tmp_val);
149 
150 	return ret;
151 }
152 
153 static int vfio_user_config_write(struct pci_dev *pdev, int offset,
154 				  __le32 val, int count)
155 {
156 	int ret = -EINVAL;
157 	u32 tmp_val = le32_to_cpu(val);
158 
159 	switch (count) {
160 	case 1:
161 		ret = pci_user_write_config_byte(pdev, offset, tmp_val);
162 		break;
163 	case 2:
164 		ret = pci_user_write_config_word(pdev, offset, tmp_val);
165 		break;
166 	case 4:
167 		ret = pci_user_write_config_dword(pdev, offset, tmp_val);
168 		break;
169 	}
170 
171 	return ret;
172 }
173 
174 static int vfio_default_config_read(struct vfio_pci_device *vdev, int pos,
175 				    int count, struct perm_bits *perm,
176 				    int offset, __le32 *val)
177 {
178 	__le32 virt = 0;
179 
180 	memcpy(val, vdev->vconfig + pos, count);
181 
182 	memcpy(&virt, perm->virt + offset, count);
183 
184 	/* Any non-virtualized bits? */
185 	if (cpu_to_le32(~0U >> (32 - (count * 8))) != virt) {
186 		struct pci_dev *pdev = vdev->pdev;
187 		__le32 phys_val = 0;
188 		int ret;
189 
190 		ret = vfio_user_config_read(pdev, pos, &phys_val, count);
191 		if (ret)
192 			return ret;
193 
194 		*val = (phys_val & ~virt) | (*val & virt);
195 	}
196 
197 	return count;
198 }
199 
200 static int vfio_default_config_write(struct vfio_pci_device *vdev, int pos,
201 				     int count, struct perm_bits *perm,
202 				     int offset, __le32 val)
203 {
204 	__le32 virt = 0, write = 0;
205 
206 	memcpy(&write, perm->write + offset, count);
207 
208 	if (!write)
209 		return count; /* drop, no writable bits */
210 
211 	memcpy(&virt, perm->virt + offset, count);
212 
213 	/* Virtualized and writable bits go to vconfig */
214 	if (write & virt) {
215 		__le32 virt_val = 0;
216 
217 		memcpy(&virt_val, vdev->vconfig + pos, count);
218 
219 		virt_val &= ~(write & virt);
220 		virt_val |= (val & (write & virt));
221 
222 		memcpy(vdev->vconfig + pos, &virt_val, count);
223 	}
224 
225 	/* Non-virtualzed and writable bits go to hardware */
226 	if (write & ~virt) {
227 		struct pci_dev *pdev = vdev->pdev;
228 		__le32 phys_val = 0;
229 		int ret;
230 
231 		ret = vfio_user_config_read(pdev, pos, &phys_val, count);
232 		if (ret)
233 			return ret;
234 
235 		phys_val &= ~(write & ~virt);
236 		phys_val |= (val & (write & ~virt));
237 
238 		ret = vfio_user_config_write(pdev, pos, phys_val, count);
239 		if (ret)
240 			return ret;
241 	}
242 
243 	return count;
244 }
245 
246 /* Allow direct read from hardware, except for capability next pointer */
247 static int vfio_direct_config_read(struct vfio_pci_device *vdev, int pos,
248 				   int count, struct perm_bits *perm,
249 				   int offset, __le32 *val)
250 {
251 	int ret;
252 
253 	ret = vfio_user_config_read(vdev->pdev, pos, val, count);
254 	if (ret)
255 		return ret;
256 
257 	if (pos >= PCI_CFG_SPACE_SIZE) { /* Extended cap header mangling */
258 		if (offset < 4)
259 			memcpy(val, vdev->vconfig + pos, count);
260 	} else if (pos >= PCI_STD_HEADER_SIZEOF) { /* Std cap mangling */
261 		if (offset == PCI_CAP_LIST_ID && count > 1)
262 			memcpy(val, vdev->vconfig + pos,
263 			       min(PCI_CAP_FLAGS, count));
264 		else if (offset == PCI_CAP_LIST_NEXT)
265 			memcpy(val, vdev->vconfig + pos, 1);
266 	}
267 
268 	return count;
269 }
270 
271 /* Raw access skips any kind of virtualization */
272 static int vfio_raw_config_write(struct vfio_pci_device *vdev, int pos,
273 				 int count, struct perm_bits *perm,
274 				 int offset, __le32 val)
275 {
276 	int ret;
277 
278 	ret = vfio_user_config_write(vdev->pdev, pos, val, count);
279 	if (ret)
280 		return ret;
281 
282 	return count;
283 }
284 
285 static int vfio_raw_config_read(struct vfio_pci_device *vdev, int pos,
286 				int count, struct perm_bits *perm,
287 				int offset, __le32 *val)
288 {
289 	int ret;
290 
291 	ret = vfio_user_config_read(vdev->pdev, pos, val, count);
292 	if (ret)
293 		return ret;
294 
295 	return count;
296 }
297 
298 /* Virt access uses only virtualization */
299 static int vfio_virt_config_write(struct vfio_pci_device *vdev, int pos,
300 				  int count, struct perm_bits *perm,
301 				  int offset, __le32 val)
302 {
303 	memcpy(vdev->vconfig + pos, &val, count);
304 	return count;
305 }
306 
307 static int vfio_virt_config_read(struct vfio_pci_device *vdev, int pos,
308 				 int count, struct perm_bits *perm,
309 				 int offset, __le32 *val)
310 {
311 	memcpy(val, vdev->vconfig + pos, count);
312 	return count;
313 }
314 
315 /* Default capability regions to read-only, no-virtualization */
316 static struct perm_bits cap_perms[PCI_CAP_ID_MAX + 1] = {
317 	[0 ... PCI_CAP_ID_MAX] = { .readfn = vfio_direct_config_read }
318 };
319 static struct perm_bits ecap_perms[PCI_EXT_CAP_ID_MAX + 1] = {
320 	[0 ... PCI_EXT_CAP_ID_MAX] = { .readfn = vfio_direct_config_read }
321 };
322 /*
323  * Default unassigned regions to raw read-write access.  Some devices
324  * require this to function as they hide registers between the gaps in
325  * config space (be2net).  Like MMIO and I/O port registers, we have
326  * to trust the hardware isolation.
327  */
328 static struct perm_bits unassigned_perms = {
329 	.readfn = vfio_raw_config_read,
330 	.writefn = vfio_raw_config_write
331 };
332 
333 static struct perm_bits virt_perms = {
334 	.readfn = vfio_virt_config_read,
335 	.writefn = vfio_virt_config_write
336 };
337 
338 static void free_perm_bits(struct perm_bits *perm)
339 {
340 	kfree(perm->virt);
341 	kfree(perm->write);
342 	perm->virt = NULL;
343 	perm->write = NULL;
344 }
345 
346 static int alloc_perm_bits(struct perm_bits *perm, int size)
347 {
348 	/*
349 	 * Round up all permission bits to the next dword, this lets us
350 	 * ignore whether a read/write exceeds the defined capability
351 	 * structure.  We can do this because:
352 	 *  - Standard config space is already dword aligned
353 	 *  - Capabilities are all dword aligned (bits 0:1 of next reserved)
354 	 *  - Express capabilities defined as dword aligned
355 	 */
356 	size = round_up(size, 4);
357 
358 	/*
359 	 * Zero state is
360 	 * - All Readable, None Writeable, None Virtualized
361 	 */
362 	perm->virt = kzalloc(size, GFP_KERNEL);
363 	perm->write = kzalloc(size, GFP_KERNEL);
364 	if (!perm->virt || !perm->write) {
365 		free_perm_bits(perm);
366 		return -ENOMEM;
367 	}
368 
369 	perm->readfn = vfio_default_config_read;
370 	perm->writefn = vfio_default_config_write;
371 
372 	return 0;
373 }
374 
375 /*
376  * Helper functions for filling in permission tables
377  */
378 static inline void p_setb(struct perm_bits *p, int off, u8 virt, u8 write)
379 {
380 	p->virt[off] = virt;
381 	p->write[off] = write;
382 }
383 
384 /* Handle endian-ness - pci and tables are little-endian */
385 static inline void p_setw(struct perm_bits *p, int off, u16 virt, u16 write)
386 {
387 	*(__le16 *)(&p->virt[off]) = cpu_to_le16(virt);
388 	*(__le16 *)(&p->write[off]) = cpu_to_le16(write);
389 }
390 
391 /* Handle endian-ness - pci and tables are little-endian */
392 static inline void p_setd(struct perm_bits *p, int off, u32 virt, u32 write)
393 {
394 	*(__le32 *)(&p->virt[off]) = cpu_to_le32(virt);
395 	*(__le32 *)(&p->write[off]) = cpu_to_le32(write);
396 }
397 
398 /* Caller should hold memory_lock semaphore */
399 bool __vfio_pci_memory_enabled(struct vfio_pci_device *vdev)
400 {
401 	struct pci_dev *pdev = vdev->pdev;
402 	u16 cmd = le16_to_cpu(*(__le16 *)&vdev->vconfig[PCI_COMMAND]);
403 
404 	/*
405 	 * SR-IOV VF memory enable is handled by the MSE bit in the
406 	 * PF SR-IOV capability, there's therefore no need to trigger
407 	 * faults based on the virtual value.
408 	 */
409 	return pdev->is_virtfn || (cmd & PCI_COMMAND_MEMORY);
410 }
411 
412 /*
413  * Restore the *real* BARs after we detect a FLR or backdoor reset.
414  * (backdoor = some device specific technique that we didn't catch)
415  */
416 static void vfio_bar_restore(struct vfio_pci_device *vdev)
417 {
418 	struct pci_dev *pdev = vdev->pdev;
419 	u32 *rbar = vdev->rbar;
420 	u16 cmd;
421 	int i;
422 
423 	if (pdev->is_virtfn)
424 		return;
425 
426 	pci_info(pdev, "%s: reset recovery - restoring BARs\n", __func__);
427 
428 	for (i = PCI_BASE_ADDRESS_0; i <= PCI_BASE_ADDRESS_5; i += 4, rbar++)
429 		pci_user_write_config_dword(pdev, i, *rbar);
430 
431 	pci_user_write_config_dword(pdev, PCI_ROM_ADDRESS, *rbar);
432 
433 	if (vdev->nointx) {
434 		pci_user_read_config_word(pdev, PCI_COMMAND, &cmd);
435 		cmd |= PCI_COMMAND_INTX_DISABLE;
436 		pci_user_write_config_word(pdev, PCI_COMMAND, cmd);
437 	}
438 }
439 
440 static __le32 vfio_generate_bar_flags(struct pci_dev *pdev, int bar)
441 {
442 	unsigned long flags = pci_resource_flags(pdev, bar);
443 	u32 val;
444 
445 	if (flags & IORESOURCE_IO)
446 		return cpu_to_le32(PCI_BASE_ADDRESS_SPACE_IO);
447 
448 	val = PCI_BASE_ADDRESS_SPACE_MEMORY;
449 
450 	if (flags & IORESOURCE_PREFETCH)
451 		val |= PCI_BASE_ADDRESS_MEM_PREFETCH;
452 
453 	if (flags & IORESOURCE_MEM_64)
454 		val |= PCI_BASE_ADDRESS_MEM_TYPE_64;
455 
456 	return cpu_to_le32(val);
457 }
458 
459 /*
460  * Pretend we're hardware and tweak the values of the *virtual* PCI BARs
461  * to reflect the hardware capabilities.  This implements BAR sizing.
462  */
463 static void vfio_bar_fixup(struct vfio_pci_device *vdev)
464 {
465 	struct pci_dev *pdev = vdev->pdev;
466 	int i;
467 	__le32 *vbar;
468 	u64 mask;
469 
470 	vbar = (__le32 *)&vdev->vconfig[PCI_BASE_ADDRESS_0];
471 
472 	for (i = 0; i < PCI_STD_NUM_BARS; i++, vbar++) {
473 		int bar = i + PCI_STD_RESOURCES;
474 
475 		if (!pci_resource_start(pdev, bar)) {
476 			*vbar = 0; /* Unmapped by host = unimplemented to user */
477 			continue;
478 		}
479 
480 		mask = ~(pci_resource_len(pdev, bar) - 1);
481 
482 		*vbar &= cpu_to_le32((u32)mask);
483 		*vbar |= vfio_generate_bar_flags(pdev, bar);
484 
485 		if (*vbar & cpu_to_le32(PCI_BASE_ADDRESS_MEM_TYPE_64)) {
486 			vbar++;
487 			*vbar &= cpu_to_le32((u32)(mask >> 32));
488 			i++;
489 		}
490 	}
491 
492 	vbar = (__le32 *)&vdev->vconfig[PCI_ROM_ADDRESS];
493 
494 	/*
495 	 * NB. REGION_INFO will have reported zero size if we weren't able
496 	 * to read the ROM, but we still return the actual BAR size here if
497 	 * it exists (or the shadow ROM space).
498 	 */
499 	if (pci_resource_start(pdev, PCI_ROM_RESOURCE)) {
500 		mask = ~(pci_resource_len(pdev, PCI_ROM_RESOURCE) - 1);
501 		mask |= PCI_ROM_ADDRESS_ENABLE;
502 		*vbar &= cpu_to_le32((u32)mask);
503 	} else if (pdev->resource[PCI_ROM_RESOURCE].flags &
504 					IORESOURCE_ROM_SHADOW) {
505 		mask = ~(0x20000 - 1);
506 		mask |= PCI_ROM_ADDRESS_ENABLE;
507 		*vbar &= cpu_to_le32((u32)mask);
508 	} else
509 		*vbar = 0;
510 
511 	vdev->bardirty = false;
512 }
513 
514 static int vfio_basic_config_read(struct vfio_pci_device *vdev, int pos,
515 				  int count, struct perm_bits *perm,
516 				  int offset, __le32 *val)
517 {
518 	if (is_bar(offset)) /* pos == offset for basic config */
519 		vfio_bar_fixup(vdev);
520 
521 	count = vfio_default_config_read(vdev, pos, count, perm, offset, val);
522 
523 	/* Mask in virtual memory enable for SR-IOV devices */
524 	if (offset == PCI_COMMAND && vdev->pdev->is_virtfn) {
525 		u16 cmd = le16_to_cpu(*(__le16 *)&vdev->vconfig[PCI_COMMAND]);
526 		u32 tmp_val = le32_to_cpu(*val);
527 
528 		tmp_val |= cmd & PCI_COMMAND_MEMORY;
529 		*val = cpu_to_le32(tmp_val);
530 	}
531 
532 	return count;
533 }
534 
535 /* Test whether BARs match the value we think they should contain */
536 static bool vfio_need_bar_restore(struct vfio_pci_device *vdev)
537 {
538 	int i = 0, pos = PCI_BASE_ADDRESS_0, ret;
539 	u32 bar;
540 
541 	for (; pos <= PCI_BASE_ADDRESS_5; i++, pos += 4) {
542 		if (vdev->rbar[i]) {
543 			ret = pci_user_read_config_dword(vdev->pdev, pos, &bar);
544 			if (ret || vdev->rbar[i] != bar)
545 				return true;
546 		}
547 	}
548 
549 	return false;
550 }
551 
552 static int vfio_basic_config_write(struct vfio_pci_device *vdev, int pos,
553 				   int count, struct perm_bits *perm,
554 				   int offset, __le32 val)
555 {
556 	struct pci_dev *pdev = vdev->pdev;
557 	__le16 *virt_cmd;
558 	u16 new_cmd = 0;
559 	int ret;
560 
561 	virt_cmd = (__le16 *)&vdev->vconfig[PCI_COMMAND];
562 
563 	if (offset == PCI_COMMAND) {
564 		bool phys_mem, virt_mem, new_mem, phys_io, virt_io, new_io;
565 		u16 phys_cmd;
566 
567 		ret = pci_user_read_config_word(pdev, PCI_COMMAND, &phys_cmd);
568 		if (ret)
569 			return ret;
570 
571 		new_cmd = le32_to_cpu(val);
572 
573 		phys_io = !!(phys_cmd & PCI_COMMAND_IO);
574 		virt_io = !!(le16_to_cpu(*virt_cmd) & PCI_COMMAND_IO);
575 		new_io = !!(new_cmd & PCI_COMMAND_IO);
576 
577 		phys_mem = !!(phys_cmd & PCI_COMMAND_MEMORY);
578 		virt_mem = !!(le16_to_cpu(*virt_cmd) & PCI_COMMAND_MEMORY);
579 		new_mem = !!(new_cmd & PCI_COMMAND_MEMORY);
580 
581 		if (!new_mem)
582 			vfio_pci_zap_and_down_write_memory_lock(vdev);
583 		else
584 			down_write(&vdev->memory_lock);
585 
586 		/*
587 		 * If the user is writing mem/io enable (new_mem/io) and we
588 		 * think it's already enabled (virt_mem/io), but the hardware
589 		 * shows it disabled (phys_mem/io, then the device has
590 		 * undergone some kind of backdoor reset and needs to be
591 		 * restored before we allow it to enable the bars.
592 		 * SR-IOV devices will trigger this, but we catch them later
593 		 */
594 		if ((new_mem && virt_mem && !phys_mem) ||
595 		    (new_io && virt_io && !phys_io) ||
596 		    vfio_need_bar_restore(vdev))
597 			vfio_bar_restore(vdev);
598 	}
599 
600 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
601 	if (count < 0) {
602 		if (offset == PCI_COMMAND)
603 			up_write(&vdev->memory_lock);
604 		return count;
605 	}
606 
607 	/*
608 	 * Save current memory/io enable bits in vconfig to allow for
609 	 * the test above next time.
610 	 */
611 	if (offset == PCI_COMMAND) {
612 		u16 mask = PCI_COMMAND_MEMORY | PCI_COMMAND_IO;
613 
614 		*virt_cmd &= cpu_to_le16(~mask);
615 		*virt_cmd |= cpu_to_le16(new_cmd & mask);
616 
617 		up_write(&vdev->memory_lock);
618 	}
619 
620 	/* Emulate INTx disable */
621 	if (offset >= PCI_COMMAND && offset <= PCI_COMMAND + 1) {
622 		bool virt_intx_disable;
623 
624 		virt_intx_disable = !!(le16_to_cpu(*virt_cmd) &
625 				       PCI_COMMAND_INTX_DISABLE);
626 
627 		if (virt_intx_disable && !vdev->virq_disabled) {
628 			vdev->virq_disabled = true;
629 			vfio_pci_intx_mask(vdev);
630 		} else if (!virt_intx_disable && vdev->virq_disabled) {
631 			vdev->virq_disabled = false;
632 			vfio_pci_intx_unmask(vdev);
633 		}
634 	}
635 
636 	if (is_bar(offset))
637 		vdev->bardirty = true;
638 
639 	return count;
640 }
641 
642 /* Permissions for the Basic PCI Header */
643 static int __init init_pci_cap_basic_perm(struct perm_bits *perm)
644 {
645 	if (alloc_perm_bits(perm, PCI_STD_HEADER_SIZEOF))
646 		return -ENOMEM;
647 
648 	perm->readfn = vfio_basic_config_read;
649 	perm->writefn = vfio_basic_config_write;
650 
651 	/* Virtualized for SR-IOV functions, which just have FFFF */
652 	p_setw(perm, PCI_VENDOR_ID, (u16)ALL_VIRT, NO_WRITE);
653 	p_setw(perm, PCI_DEVICE_ID, (u16)ALL_VIRT, NO_WRITE);
654 
655 	/*
656 	 * Virtualize INTx disable, we use it internally for interrupt
657 	 * control and can emulate it for non-PCI 2.3 devices.
658 	 */
659 	p_setw(perm, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE, (u16)ALL_WRITE);
660 
661 	/* Virtualize capability list, we might want to skip/disable */
662 	p_setw(perm, PCI_STATUS, PCI_STATUS_CAP_LIST, NO_WRITE);
663 
664 	/* No harm to write */
665 	p_setb(perm, PCI_CACHE_LINE_SIZE, NO_VIRT, (u8)ALL_WRITE);
666 	p_setb(perm, PCI_LATENCY_TIMER, NO_VIRT, (u8)ALL_WRITE);
667 	p_setb(perm, PCI_BIST, NO_VIRT, (u8)ALL_WRITE);
668 
669 	/* Virtualize all bars, can't touch the real ones */
670 	p_setd(perm, PCI_BASE_ADDRESS_0, ALL_VIRT, ALL_WRITE);
671 	p_setd(perm, PCI_BASE_ADDRESS_1, ALL_VIRT, ALL_WRITE);
672 	p_setd(perm, PCI_BASE_ADDRESS_2, ALL_VIRT, ALL_WRITE);
673 	p_setd(perm, PCI_BASE_ADDRESS_3, ALL_VIRT, ALL_WRITE);
674 	p_setd(perm, PCI_BASE_ADDRESS_4, ALL_VIRT, ALL_WRITE);
675 	p_setd(perm, PCI_BASE_ADDRESS_5, ALL_VIRT, ALL_WRITE);
676 	p_setd(perm, PCI_ROM_ADDRESS, ALL_VIRT, ALL_WRITE);
677 
678 	/* Allow us to adjust capability chain */
679 	p_setb(perm, PCI_CAPABILITY_LIST, (u8)ALL_VIRT, NO_WRITE);
680 
681 	/* Sometimes used by sw, just virtualize */
682 	p_setb(perm, PCI_INTERRUPT_LINE, (u8)ALL_VIRT, (u8)ALL_WRITE);
683 
684 	/* Virtualize interrupt pin to allow hiding INTx */
685 	p_setb(perm, PCI_INTERRUPT_PIN, (u8)ALL_VIRT, (u8)NO_WRITE);
686 
687 	return 0;
688 }
689 
690 static int vfio_pm_config_write(struct vfio_pci_device *vdev, int pos,
691 				int count, struct perm_bits *perm,
692 				int offset, __le32 val)
693 {
694 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
695 	if (count < 0)
696 		return count;
697 
698 	if (offset == PCI_PM_CTRL) {
699 		pci_power_t state;
700 
701 		switch (le32_to_cpu(val) & PCI_PM_CTRL_STATE_MASK) {
702 		case 0:
703 			state = PCI_D0;
704 			break;
705 		case 1:
706 			state = PCI_D1;
707 			break;
708 		case 2:
709 			state = PCI_D2;
710 			break;
711 		case 3:
712 			state = PCI_D3hot;
713 			break;
714 		}
715 
716 		vfio_pci_set_power_state(vdev, state);
717 	}
718 
719 	return count;
720 }
721 
722 /* Permissions for the Power Management capability */
723 static int __init init_pci_cap_pm_perm(struct perm_bits *perm)
724 {
725 	if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_PM]))
726 		return -ENOMEM;
727 
728 	perm->writefn = vfio_pm_config_write;
729 
730 	/*
731 	 * We always virtualize the next field so we can remove
732 	 * capabilities from the chain if we want to.
733 	 */
734 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
735 
736 	/*
737 	 * Power management is defined *per function*, so we can let
738 	 * the user change power state, but we trap and initiate the
739 	 * change ourselves, so the state bits are read-only.
740 	 */
741 	p_setd(perm, PCI_PM_CTRL, NO_VIRT, ~PCI_PM_CTRL_STATE_MASK);
742 	return 0;
743 }
744 
745 static int vfio_vpd_config_write(struct vfio_pci_device *vdev, int pos,
746 				 int count, struct perm_bits *perm,
747 				 int offset, __le32 val)
748 {
749 	struct pci_dev *pdev = vdev->pdev;
750 	__le16 *paddr = (__le16 *)(vdev->vconfig + pos - offset + PCI_VPD_ADDR);
751 	__le32 *pdata = (__le32 *)(vdev->vconfig + pos - offset + PCI_VPD_DATA);
752 	u16 addr;
753 	u32 data;
754 
755 	/*
756 	 * Write through to emulation.  If the write includes the upper byte
757 	 * of PCI_VPD_ADDR, then the PCI_VPD_ADDR_F bit is written and we
758 	 * have work to do.
759 	 */
760 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
761 	if (count < 0 || offset > PCI_VPD_ADDR + 1 ||
762 	    offset + count <= PCI_VPD_ADDR + 1)
763 		return count;
764 
765 	addr = le16_to_cpu(*paddr);
766 
767 	if (addr & PCI_VPD_ADDR_F) {
768 		data = le32_to_cpu(*pdata);
769 		if (pci_write_vpd(pdev, addr & ~PCI_VPD_ADDR_F, 4, &data) != 4)
770 			return count;
771 	} else {
772 		data = 0;
773 		if (pci_read_vpd(pdev, addr, 4, &data) < 0)
774 			return count;
775 		*pdata = cpu_to_le32(data);
776 	}
777 
778 	/*
779 	 * Toggle PCI_VPD_ADDR_F in the emulated PCI_VPD_ADDR register to
780 	 * signal completion.  If an error occurs above, we assume that not
781 	 * toggling this bit will induce a driver timeout.
782 	 */
783 	addr ^= PCI_VPD_ADDR_F;
784 	*paddr = cpu_to_le16(addr);
785 
786 	return count;
787 }
788 
789 /* Permissions for Vital Product Data capability */
790 static int __init init_pci_cap_vpd_perm(struct perm_bits *perm)
791 {
792 	if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_VPD]))
793 		return -ENOMEM;
794 
795 	perm->writefn = vfio_vpd_config_write;
796 
797 	/*
798 	 * We always virtualize the next field so we can remove
799 	 * capabilities from the chain if we want to.
800 	 */
801 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
802 
803 	/*
804 	 * Both the address and data registers are virtualized to
805 	 * enable access through the pci_vpd_read/write functions
806 	 */
807 	p_setw(perm, PCI_VPD_ADDR, (u16)ALL_VIRT, (u16)ALL_WRITE);
808 	p_setd(perm, PCI_VPD_DATA, ALL_VIRT, ALL_WRITE);
809 
810 	return 0;
811 }
812 
813 /* Permissions for PCI-X capability */
814 static int __init init_pci_cap_pcix_perm(struct perm_bits *perm)
815 {
816 	/* Alloc 24, but only 8 are used in v0 */
817 	if (alloc_perm_bits(perm, PCI_CAP_PCIX_SIZEOF_V2))
818 		return -ENOMEM;
819 
820 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
821 
822 	p_setw(perm, PCI_X_CMD, NO_VIRT, (u16)ALL_WRITE);
823 	p_setd(perm, PCI_X_ECC_CSR, NO_VIRT, ALL_WRITE);
824 	return 0;
825 }
826 
827 static int vfio_exp_config_write(struct vfio_pci_device *vdev, int pos,
828 				 int count, struct perm_bits *perm,
829 				 int offset, __le32 val)
830 {
831 	__le16 *ctrl = (__le16 *)(vdev->vconfig + pos -
832 				  offset + PCI_EXP_DEVCTL);
833 	int readrq = le16_to_cpu(*ctrl) & PCI_EXP_DEVCTL_READRQ;
834 
835 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
836 	if (count < 0)
837 		return count;
838 
839 	/*
840 	 * The FLR bit is virtualized, if set and the device supports PCIe
841 	 * FLR, issue a reset_function.  Regardless, clear the bit, the spec
842 	 * requires it to be always read as zero.  NB, reset_function might
843 	 * not use a PCIe FLR, we don't have that level of granularity.
844 	 */
845 	if (*ctrl & cpu_to_le16(PCI_EXP_DEVCTL_BCR_FLR)) {
846 		u32 cap;
847 		int ret;
848 
849 		*ctrl &= ~cpu_to_le16(PCI_EXP_DEVCTL_BCR_FLR);
850 
851 		ret = pci_user_read_config_dword(vdev->pdev,
852 						 pos - offset + PCI_EXP_DEVCAP,
853 						 &cap);
854 
855 		if (!ret && (cap & PCI_EXP_DEVCAP_FLR)) {
856 			vfio_pci_zap_and_down_write_memory_lock(vdev);
857 			pci_try_reset_function(vdev->pdev);
858 			up_write(&vdev->memory_lock);
859 		}
860 	}
861 
862 	/*
863 	 * MPS is virtualized to the user, writes do not change the physical
864 	 * register since determining a proper MPS value requires a system wide
865 	 * device view.  The MRRS is largely independent of MPS, but since the
866 	 * user does not have that system-wide view, they might set a safe, but
867 	 * inefficiently low value.  Here we allow writes through to hardware,
868 	 * but we set the floor to the physical device MPS setting, so that
869 	 * we can at least use full TLPs, as defined by the MPS value.
870 	 *
871 	 * NB, if any devices actually depend on an artificially low MRRS
872 	 * setting, this will need to be revisited, perhaps with a quirk
873 	 * though pcie_set_readrq().
874 	 */
875 	if (readrq != (le16_to_cpu(*ctrl) & PCI_EXP_DEVCTL_READRQ)) {
876 		readrq = 128 <<
877 			((le16_to_cpu(*ctrl) & PCI_EXP_DEVCTL_READRQ) >> 12);
878 		readrq = max(readrq, pcie_get_mps(vdev->pdev));
879 
880 		pcie_set_readrq(vdev->pdev, readrq);
881 	}
882 
883 	return count;
884 }
885 
886 /* Permissions for PCI Express capability */
887 static int __init init_pci_cap_exp_perm(struct perm_bits *perm)
888 {
889 	/* Alloc largest of possible sizes */
890 	if (alloc_perm_bits(perm, PCI_CAP_EXP_ENDPOINT_SIZEOF_V2))
891 		return -ENOMEM;
892 
893 	perm->writefn = vfio_exp_config_write;
894 
895 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
896 
897 	/*
898 	 * Allow writes to device control fields, except devctl_phantom,
899 	 * which could confuse IOMMU, MPS, which can break communication
900 	 * with other physical devices, and the ARI bit in devctl2, which
901 	 * is set at probe time.  FLR and MRRS get virtualized via our
902 	 * writefn.
903 	 */
904 	p_setw(perm, PCI_EXP_DEVCTL,
905 	       PCI_EXP_DEVCTL_BCR_FLR | PCI_EXP_DEVCTL_PAYLOAD |
906 	       PCI_EXP_DEVCTL_READRQ, ~PCI_EXP_DEVCTL_PHANTOM);
907 	p_setw(perm, PCI_EXP_DEVCTL2, NO_VIRT, ~PCI_EXP_DEVCTL2_ARI);
908 	return 0;
909 }
910 
911 static int vfio_af_config_write(struct vfio_pci_device *vdev, int pos,
912 				int count, struct perm_bits *perm,
913 				int offset, __le32 val)
914 {
915 	u8 *ctrl = vdev->vconfig + pos - offset + PCI_AF_CTRL;
916 
917 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
918 	if (count < 0)
919 		return count;
920 
921 	/*
922 	 * The FLR bit is virtualized, if set and the device supports AF
923 	 * FLR, issue a reset_function.  Regardless, clear the bit, the spec
924 	 * requires it to be always read as zero.  NB, reset_function might
925 	 * not use an AF FLR, we don't have that level of granularity.
926 	 */
927 	if (*ctrl & PCI_AF_CTRL_FLR) {
928 		u8 cap;
929 		int ret;
930 
931 		*ctrl &= ~PCI_AF_CTRL_FLR;
932 
933 		ret = pci_user_read_config_byte(vdev->pdev,
934 						pos - offset + PCI_AF_CAP,
935 						&cap);
936 
937 		if (!ret && (cap & PCI_AF_CAP_FLR) && (cap & PCI_AF_CAP_TP)) {
938 			vfio_pci_zap_and_down_write_memory_lock(vdev);
939 			pci_try_reset_function(vdev->pdev);
940 			up_write(&vdev->memory_lock);
941 		}
942 	}
943 
944 	return count;
945 }
946 
947 /* Permissions for Advanced Function capability */
948 static int __init init_pci_cap_af_perm(struct perm_bits *perm)
949 {
950 	if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_AF]))
951 		return -ENOMEM;
952 
953 	perm->writefn = vfio_af_config_write;
954 
955 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
956 	p_setb(perm, PCI_AF_CTRL, PCI_AF_CTRL_FLR, PCI_AF_CTRL_FLR);
957 	return 0;
958 }
959 
960 /* Permissions for Advanced Error Reporting extended capability */
961 static int __init init_pci_ext_cap_err_perm(struct perm_bits *perm)
962 {
963 	u32 mask;
964 
965 	if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_ERR]))
966 		return -ENOMEM;
967 
968 	/*
969 	 * Virtualize the first dword of all express capabilities
970 	 * because it includes the next pointer.  This lets us later
971 	 * remove capabilities from the chain if we need to.
972 	 */
973 	p_setd(perm, 0, ALL_VIRT, NO_WRITE);
974 
975 	/* Writable bits mask */
976 	mask =	PCI_ERR_UNC_UND |		/* Undefined */
977 		PCI_ERR_UNC_DLP |		/* Data Link Protocol */
978 		PCI_ERR_UNC_SURPDN |		/* Surprise Down */
979 		PCI_ERR_UNC_POISON_TLP |	/* Poisoned TLP */
980 		PCI_ERR_UNC_FCP |		/* Flow Control Protocol */
981 		PCI_ERR_UNC_COMP_TIME |		/* Completion Timeout */
982 		PCI_ERR_UNC_COMP_ABORT |	/* Completer Abort */
983 		PCI_ERR_UNC_UNX_COMP |		/* Unexpected Completion */
984 		PCI_ERR_UNC_RX_OVER |		/* Receiver Overflow */
985 		PCI_ERR_UNC_MALF_TLP |		/* Malformed TLP */
986 		PCI_ERR_UNC_ECRC |		/* ECRC Error Status */
987 		PCI_ERR_UNC_UNSUP |		/* Unsupported Request */
988 		PCI_ERR_UNC_ACSV |		/* ACS Violation */
989 		PCI_ERR_UNC_INTN |		/* internal error */
990 		PCI_ERR_UNC_MCBTLP |		/* MC blocked TLP */
991 		PCI_ERR_UNC_ATOMEG |		/* Atomic egress blocked */
992 		PCI_ERR_UNC_TLPPRE;		/* TLP prefix blocked */
993 	p_setd(perm, PCI_ERR_UNCOR_STATUS, NO_VIRT, mask);
994 	p_setd(perm, PCI_ERR_UNCOR_MASK, NO_VIRT, mask);
995 	p_setd(perm, PCI_ERR_UNCOR_SEVER, NO_VIRT, mask);
996 
997 	mask =	PCI_ERR_COR_RCVR |		/* Receiver Error Status */
998 		PCI_ERR_COR_BAD_TLP |		/* Bad TLP Status */
999 		PCI_ERR_COR_BAD_DLLP |		/* Bad DLLP Status */
1000 		PCI_ERR_COR_REP_ROLL |		/* REPLAY_NUM Rollover */
1001 		PCI_ERR_COR_REP_TIMER |		/* Replay Timer Timeout */
1002 		PCI_ERR_COR_ADV_NFAT |		/* Advisory Non-Fatal */
1003 		PCI_ERR_COR_INTERNAL |		/* Corrected Internal */
1004 		PCI_ERR_COR_LOG_OVER;		/* Header Log Overflow */
1005 	p_setd(perm, PCI_ERR_COR_STATUS, NO_VIRT, mask);
1006 	p_setd(perm, PCI_ERR_COR_MASK, NO_VIRT, mask);
1007 
1008 	mask =	PCI_ERR_CAP_ECRC_GENE |		/* ECRC Generation Enable */
1009 		PCI_ERR_CAP_ECRC_CHKE;		/* ECRC Check Enable */
1010 	p_setd(perm, PCI_ERR_CAP, NO_VIRT, mask);
1011 	return 0;
1012 }
1013 
1014 /* Permissions for Power Budgeting extended capability */
1015 static int __init init_pci_ext_cap_pwr_perm(struct perm_bits *perm)
1016 {
1017 	if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_PWR]))
1018 		return -ENOMEM;
1019 
1020 	p_setd(perm, 0, ALL_VIRT, NO_WRITE);
1021 
1022 	/* Writing the data selector is OK, the info is still read-only */
1023 	p_setb(perm, PCI_PWR_DATA, NO_VIRT, (u8)ALL_WRITE);
1024 	return 0;
1025 }
1026 
1027 /*
1028  * Initialize the shared permission tables
1029  */
1030 void vfio_pci_uninit_perm_bits(void)
1031 {
1032 	free_perm_bits(&cap_perms[PCI_CAP_ID_BASIC]);
1033 
1034 	free_perm_bits(&cap_perms[PCI_CAP_ID_PM]);
1035 	free_perm_bits(&cap_perms[PCI_CAP_ID_VPD]);
1036 	free_perm_bits(&cap_perms[PCI_CAP_ID_PCIX]);
1037 	free_perm_bits(&cap_perms[PCI_CAP_ID_EXP]);
1038 	free_perm_bits(&cap_perms[PCI_CAP_ID_AF]);
1039 
1040 	free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_ERR]);
1041 	free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_PWR]);
1042 }
1043 
1044 int __init vfio_pci_init_perm_bits(void)
1045 {
1046 	int ret;
1047 
1048 	/* Basic config space */
1049 	ret = init_pci_cap_basic_perm(&cap_perms[PCI_CAP_ID_BASIC]);
1050 
1051 	/* Capabilities */
1052 	ret |= init_pci_cap_pm_perm(&cap_perms[PCI_CAP_ID_PM]);
1053 	ret |= init_pci_cap_vpd_perm(&cap_perms[PCI_CAP_ID_VPD]);
1054 	ret |= init_pci_cap_pcix_perm(&cap_perms[PCI_CAP_ID_PCIX]);
1055 	cap_perms[PCI_CAP_ID_VNDR].writefn = vfio_raw_config_write;
1056 	ret |= init_pci_cap_exp_perm(&cap_perms[PCI_CAP_ID_EXP]);
1057 	ret |= init_pci_cap_af_perm(&cap_perms[PCI_CAP_ID_AF]);
1058 
1059 	/* Extended capabilities */
1060 	ret |= init_pci_ext_cap_err_perm(&ecap_perms[PCI_EXT_CAP_ID_ERR]);
1061 	ret |= init_pci_ext_cap_pwr_perm(&ecap_perms[PCI_EXT_CAP_ID_PWR]);
1062 	ecap_perms[PCI_EXT_CAP_ID_VNDR].writefn = vfio_raw_config_write;
1063 
1064 	if (ret)
1065 		vfio_pci_uninit_perm_bits();
1066 
1067 	return ret;
1068 }
1069 
1070 static int vfio_find_cap_start(struct vfio_pci_device *vdev, int pos)
1071 {
1072 	u8 cap;
1073 	int base = (pos >= PCI_CFG_SPACE_SIZE) ? PCI_CFG_SPACE_SIZE :
1074 						 PCI_STD_HEADER_SIZEOF;
1075 	cap = vdev->pci_config_map[pos];
1076 
1077 	if (cap == PCI_CAP_ID_BASIC)
1078 		return 0;
1079 
1080 	/* XXX Can we have to abutting capabilities of the same type? */
1081 	while (pos - 1 >= base && vdev->pci_config_map[pos - 1] == cap)
1082 		pos--;
1083 
1084 	return pos;
1085 }
1086 
1087 static int vfio_msi_config_read(struct vfio_pci_device *vdev, int pos,
1088 				int count, struct perm_bits *perm,
1089 				int offset, __le32 *val)
1090 {
1091 	/* Update max available queue size from msi_qmax */
1092 	if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
1093 		__le16 *flags;
1094 		int start;
1095 
1096 		start = vfio_find_cap_start(vdev, pos);
1097 
1098 		flags = (__le16 *)&vdev->vconfig[start];
1099 
1100 		*flags &= cpu_to_le16(~PCI_MSI_FLAGS_QMASK);
1101 		*flags |= cpu_to_le16(vdev->msi_qmax << 1);
1102 	}
1103 
1104 	return vfio_default_config_read(vdev, pos, count, perm, offset, val);
1105 }
1106 
1107 static int vfio_msi_config_write(struct vfio_pci_device *vdev, int pos,
1108 				 int count, struct perm_bits *perm,
1109 				 int offset, __le32 val)
1110 {
1111 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
1112 	if (count < 0)
1113 		return count;
1114 
1115 	/* Fixup and write configured queue size and enable to hardware */
1116 	if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
1117 		__le16 *pflags;
1118 		u16 flags;
1119 		int start, ret;
1120 
1121 		start = vfio_find_cap_start(vdev, pos);
1122 
1123 		pflags = (__le16 *)&vdev->vconfig[start + PCI_MSI_FLAGS];
1124 
1125 		flags = le16_to_cpu(*pflags);
1126 
1127 		/* MSI is enabled via ioctl */
1128 		if  (!is_msi(vdev))
1129 			flags &= ~PCI_MSI_FLAGS_ENABLE;
1130 
1131 		/* Check queue size */
1132 		if ((flags & PCI_MSI_FLAGS_QSIZE) >> 4 > vdev->msi_qmax) {
1133 			flags &= ~PCI_MSI_FLAGS_QSIZE;
1134 			flags |= vdev->msi_qmax << 4;
1135 		}
1136 
1137 		/* Write back to virt and to hardware */
1138 		*pflags = cpu_to_le16(flags);
1139 		ret = pci_user_write_config_word(vdev->pdev,
1140 						 start + PCI_MSI_FLAGS,
1141 						 flags);
1142 		if (ret)
1143 			return ret;
1144 	}
1145 
1146 	return count;
1147 }
1148 
1149 /*
1150  * MSI determination is per-device, so this routine gets used beyond
1151  * initialization time. Don't add __init
1152  */
1153 static int init_pci_cap_msi_perm(struct perm_bits *perm, int len, u16 flags)
1154 {
1155 	if (alloc_perm_bits(perm, len))
1156 		return -ENOMEM;
1157 
1158 	perm->readfn = vfio_msi_config_read;
1159 	perm->writefn = vfio_msi_config_write;
1160 
1161 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
1162 
1163 	/*
1164 	 * The upper byte of the control register is reserved,
1165 	 * just setup the lower byte.
1166 	 */
1167 	p_setb(perm, PCI_MSI_FLAGS, (u8)ALL_VIRT, (u8)ALL_WRITE);
1168 	p_setd(perm, PCI_MSI_ADDRESS_LO, ALL_VIRT, ALL_WRITE);
1169 	if (flags & PCI_MSI_FLAGS_64BIT) {
1170 		p_setd(perm, PCI_MSI_ADDRESS_HI, ALL_VIRT, ALL_WRITE);
1171 		p_setw(perm, PCI_MSI_DATA_64, (u16)ALL_VIRT, (u16)ALL_WRITE);
1172 		if (flags & PCI_MSI_FLAGS_MASKBIT) {
1173 			p_setd(perm, PCI_MSI_MASK_64, NO_VIRT, ALL_WRITE);
1174 			p_setd(perm, PCI_MSI_PENDING_64, NO_VIRT, ALL_WRITE);
1175 		}
1176 	} else {
1177 		p_setw(perm, PCI_MSI_DATA_32, (u16)ALL_VIRT, (u16)ALL_WRITE);
1178 		if (flags & PCI_MSI_FLAGS_MASKBIT) {
1179 			p_setd(perm, PCI_MSI_MASK_32, NO_VIRT, ALL_WRITE);
1180 			p_setd(perm, PCI_MSI_PENDING_32, NO_VIRT, ALL_WRITE);
1181 		}
1182 	}
1183 	return 0;
1184 }
1185 
1186 /* Determine MSI CAP field length; initialize msi_perms on 1st call per vdev */
1187 static int vfio_msi_cap_len(struct vfio_pci_device *vdev, u8 pos)
1188 {
1189 	struct pci_dev *pdev = vdev->pdev;
1190 	int len, ret;
1191 	u16 flags;
1192 
1193 	ret = pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &flags);
1194 	if (ret)
1195 		return pcibios_err_to_errno(ret);
1196 
1197 	len = 10; /* Minimum size */
1198 	if (flags & PCI_MSI_FLAGS_64BIT)
1199 		len += 4;
1200 	if (flags & PCI_MSI_FLAGS_MASKBIT)
1201 		len += 10;
1202 
1203 	if (vdev->msi_perm)
1204 		return len;
1205 
1206 	vdev->msi_perm = kmalloc(sizeof(struct perm_bits), GFP_KERNEL);
1207 	if (!vdev->msi_perm)
1208 		return -ENOMEM;
1209 
1210 	ret = init_pci_cap_msi_perm(vdev->msi_perm, len, flags);
1211 	if (ret) {
1212 		kfree(vdev->msi_perm);
1213 		return ret;
1214 	}
1215 
1216 	return len;
1217 }
1218 
1219 /* Determine extended capability length for VC (2 & 9) and MFVC */
1220 static int vfio_vc_cap_len(struct vfio_pci_device *vdev, u16 pos)
1221 {
1222 	struct pci_dev *pdev = vdev->pdev;
1223 	u32 tmp;
1224 	int ret, evcc, phases, vc_arb;
1225 	int len = PCI_CAP_VC_BASE_SIZEOF;
1226 
1227 	ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_CAP1, &tmp);
1228 	if (ret)
1229 		return pcibios_err_to_errno(ret);
1230 
1231 	evcc = tmp & PCI_VC_CAP1_EVCC; /* extended vc count */
1232 	ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_CAP2, &tmp);
1233 	if (ret)
1234 		return pcibios_err_to_errno(ret);
1235 
1236 	if (tmp & PCI_VC_CAP2_128_PHASE)
1237 		phases = 128;
1238 	else if (tmp & PCI_VC_CAP2_64_PHASE)
1239 		phases = 64;
1240 	else if (tmp & PCI_VC_CAP2_32_PHASE)
1241 		phases = 32;
1242 	else
1243 		phases = 0;
1244 
1245 	vc_arb = phases * 4;
1246 
1247 	/*
1248 	 * Port arbitration tables are root & switch only;
1249 	 * function arbitration tables are function 0 only.
1250 	 * In either case, we'll never let user write them so
1251 	 * we don't care how big they are
1252 	 */
1253 	len += (1 + evcc) * PCI_CAP_VC_PER_VC_SIZEOF;
1254 	if (vc_arb) {
1255 		len = round_up(len, 16);
1256 		len += vc_arb / 8;
1257 	}
1258 	return len;
1259 }
1260 
1261 static int vfio_cap_len(struct vfio_pci_device *vdev, u8 cap, u8 pos)
1262 {
1263 	struct pci_dev *pdev = vdev->pdev;
1264 	u32 dword;
1265 	u16 word;
1266 	u8 byte;
1267 	int ret;
1268 
1269 	switch (cap) {
1270 	case PCI_CAP_ID_MSI:
1271 		return vfio_msi_cap_len(vdev, pos);
1272 	case PCI_CAP_ID_PCIX:
1273 		ret = pci_read_config_word(pdev, pos + PCI_X_CMD, &word);
1274 		if (ret)
1275 			return pcibios_err_to_errno(ret);
1276 
1277 		if (PCI_X_CMD_VERSION(word)) {
1278 			if (pdev->cfg_size > PCI_CFG_SPACE_SIZE) {
1279 				/* Test for extended capabilities */
1280 				pci_read_config_dword(pdev, PCI_CFG_SPACE_SIZE,
1281 						      &dword);
1282 				vdev->extended_caps = (dword != 0);
1283 			}
1284 			return PCI_CAP_PCIX_SIZEOF_V2;
1285 		} else
1286 			return PCI_CAP_PCIX_SIZEOF_V0;
1287 	case PCI_CAP_ID_VNDR:
1288 		/* length follows next field */
1289 		ret = pci_read_config_byte(pdev, pos + PCI_CAP_FLAGS, &byte);
1290 		if (ret)
1291 			return pcibios_err_to_errno(ret);
1292 
1293 		return byte;
1294 	case PCI_CAP_ID_EXP:
1295 		if (pdev->cfg_size > PCI_CFG_SPACE_SIZE) {
1296 			/* Test for extended capabilities */
1297 			pci_read_config_dword(pdev, PCI_CFG_SPACE_SIZE, &dword);
1298 			vdev->extended_caps = (dword != 0);
1299 		}
1300 
1301 		/* length based on version and type */
1302 		if ((pcie_caps_reg(pdev) & PCI_EXP_FLAGS_VERS) == 1) {
1303 			if (pci_pcie_type(pdev) == PCI_EXP_TYPE_RC_END)
1304 				return 0xc; /* "All Devices" only, no link */
1305 			return PCI_CAP_EXP_ENDPOINT_SIZEOF_V1;
1306 		} else {
1307 			if (pci_pcie_type(pdev) == PCI_EXP_TYPE_RC_END)
1308 				return 0x2c; /* No link */
1309 			return PCI_CAP_EXP_ENDPOINT_SIZEOF_V2;
1310 		}
1311 	case PCI_CAP_ID_HT:
1312 		ret = pci_read_config_byte(pdev, pos + 3, &byte);
1313 		if (ret)
1314 			return pcibios_err_to_errno(ret);
1315 
1316 		return (byte & HT_3BIT_CAP_MASK) ?
1317 			HT_CAP_SIZEOF_SHORT : HT_CAP_SIZEOF_LONG;
1318 	case PCI_CAP_ID_SATA:
1319 		ret = pci_read_config_byte(pdev, pos + PCI_SATA_REGS, &byte);
1320 		if (ret)
1321 			return pcibios_err_to_errno(ret);
1322 
1323 		byte &= PCI_SATA_REGS_MASK;
1324 		if (byte == PCI_SATA_REGS_INLINE)
1325 			return PCI_SATA_SIZEOF_LONG;
1326 		else
1327 			return PCI_SATA_SIZEOF_SHORT;
1328 	default:
1329 		pci_warn(pdev, "%s: unknown length for PCI cap %#x@%#x\n",
1330 			 __func__, cap, pos);
1331 	}
1332 
1333 	return 0;
1334 }
1335 
1336 static int vfio_ext_cap_len(struct vfio_pci_device *vdev, u16 ecap, u16 epos)
1337 {
1338 	struct pci_dev *pdev = vdev->pdev;
1339 	u8 byte;
1340 	u32 dword;
1341 	int ret;
1342 
1343 	switch (ecap) {
1344 	case PCI_EXT_CAP_ID_VNDR:
1345 		ret = pci_read_config_dword(pdev, epos + PCI_VSEC_HDR, &dword);
1346 		if (ret)
1347 			return pcibios_err_to_errno(ret);
1348 
1349 		return dword >> PCI_VSEC_HDR_LEN_SHIFT;
1350 	case PCI_EXT_CAP_ID_VC:
1351 	case PCI_EXT_CAP_ID_VC9:
1352 	case PCI_EXT_CAP_ID_MFVC:
1353 		return vfio_vc_cap_len(vdev, epos);
1354 	case PCI_EXT_CAP_ID_ACS:
1355 		ret = pci_read_config_byte(pdev, epos + PCI_ACS_CAP, &byte);
1356 		if (ret)
1357 			return pcibios_err_to_errno(ret);
1358 
1359 		if (byte & PCI_ACS_EC) {
1360 			int bits;
1361 
1362 			ret = pci_read_config_byte(pdev,
1363 						   epos + PCI_ACS_EGRESS_BITS,
1364 						   &byte);
1365 			if (ret)
1366 				return pcibios_err_to_errno(ret);
1367 
1368 			bits = byte ? round_up(byte, 32) : 256;
1369 			return 8 + (bits / 8);
1370 		}
1371 		return 8;
1372 
1373 	case PCI_EXT_CAP_ID_REBAR:
1374 		ret = pci_read_config_byte(pdev, epos + PCI_REBAR_CTRL, &byte);
1375 		if (ret)
1376 			return pcibios_err_to_errno(ret);
1377 
1378 		byte &= PCI_REBAR_CTRL_NBAR_MASK;
1379 		byte >>= PCI_REBAR_CTRL_NBAR_SHIFT;
1380 
1381 		return 4 + (byte * 8);
1382 	case PCI_EXT_CAP_ID_DPA:
1383 		ret = pci_read_config_byte(pdev, epos + PCI_DPA_CAP, &byte);
1384 		if (ret)
1385 			return pcibios_err_to_errno(ret);
1386 
1387 		byte &= PCI_DPA_CAP_SUBSTATE_MASK;
1388 		return PCI_DPA_BASE_SIZEOF + byte + 1;
1389 	case PCI_EXT_CAP_ID_TPH:
1390 		ret = pci_read_config_dword(pdev, epos + PCI_TPH_CAP, &dword);
1391 		if (ret)
1392 			return pcibios_err_to_errno(ret);
1393 
1394 		if ((dword & PCI_TPH_CAP_LOC_MASK) == PCI_TPH_LOC_CAP) {
1395 			int sts;
1396 
1397 			sts = dword & PCI_TPH_CAP_ST_MASK;
1398 			sts >>= PCI_TPH_CAP_ST_SHIFT;
1399 			return PCI_TPH_BASE_SIZEOF + (sts * 2) + 2;
1400 		}
1401 		return PCI_TPH_BASE_SIZEOF;
1402 	default:
1403 		pci_warn(pdev, "%s: unknown length for PCI ecap %#x@%#x\n",
1404 			 __func__, ecap, epos);
1405 	}
1406 
1407 	return 0;
1408 }
1409 
1410 static int vfio_fill_vconfig_bytes(struct vfio_pci_device *vdev,
1411 				   int offset, int size)
1412 {
1413 	struct pci_dev *pdev = vdev->pdev;
1414 	int ret = 0;
1415 
1416 	/*
1417 	 * We try to read physical config space in the largest chunks
1418 	 * we can, assuming that all of the fields support dword access.
1419 	 * pci_save_state() makes this same assumption and seems to do ok.
1420 	 */
1421 	while (size) {
1422 		int filled;
1423 
1424 		if (size >= 4 && !(offset % 4)) {
1425 			__le32 *dwordp = (__le32 *)&vdev->vconfig[offset];
1426 			u32 dword;
1427 
1428 			ret = pci_read_config_dword(pdev, offset, &dword);
1429 			if (ret)
1430 				return ret;
1431 			*dwordp = cpu_to_le32(dword);
1432 			filled = 4;
1433 		} else if (size >= 2 && !(offset % 2)) {
1434 			__le16 *wordp = (__le16 *)&vdev->vconfig[offset];
1435 			u16 word;
1436 
1437 			ret = pci_read_config_word(pdev, offset, &word);
1438 			if (ret)
1439 				return ret;
1440 			*wordp = cpu_to_le16(word);
1441 			filled = 2;
1442 		} else {
1443 			u8 *byte = &vdev->vconfig[offset];
1444 			ret = pci_read_config_byte(pdev, offset, byte);
1445 			if (ret)
1446 				return ret;
1447 			filled = 1;
1448 		}
1449 
1450 		offset += filled;
1451 		size -= filled;
1452 	}
1453 
1454 	return ret;
1455 }
1456 
1457 static int vfio_cap_init(struct vfio_pci_device *vdev)
1458 {
1459 	struct pci_dev *pdev = vdev->pdev;
1460 	u8 *map = vdev->pci_config_map;
1461 	u16 status;
1462 	u8 pos, *prev, cap;
1463 	int loops, ret, caps = 0;
1464 
1465 	/* Any capabilities? */
1466 	ret = pci_read_config_word(pdev, PCI_STATUS, &status);
1467 	if (ret)
1468 		return ret;
1469 
1470 	if (!(status & PCI_STATUS_CAP_LIST))
1471 		return 0; /* Done */
1472 
1473 	ret = pci_read_config_byte(pdev, PCI_CAPABILITY_LIST, &pos);
1474 	if (ret)
1475 		return ret;
1476 
1477 	/* Mark the previous position in case we want to skip a capability */
1478 	prev = &vdev->vconfig[PCI_CAPABILITY_LIST];
1479 
1480 	/* We can bound our loop, capabilities are dword aligned */
1481 	loops = (PCI_CFG_SPACE_SIZE - PCI_STD_HEADER_SIZEOF) / PCI_CAP_SIZEOF;
1482 	while (pos && loops--) {
1483 		u8 next;
1484 		int i, len = 0;
1485 
1486 		ret = pci_read_config_byte(pdev, pos, &cap);
1487 		if (ret)
1488 			return ret;
1489 
1490 		ret = pci_read_config_byte(pdev,
1491 					   pos + PCI_CAP_LIST_NEXT, &next);
1492 		if (ret)
1493 			return ret;
1494 
1495 		/*
1496 		 * ID 0 is a NULL capability, conflicting with our fake
1497 		 * PCI_CAP_ID_BASIC.  As it has no content, consider it
1498 		 * hidden for now.
1499 		 */
1500 		if (cap && cap <= PCI_CAP_ID_MAX) {
1501 			len = pci_cap_length[cap];
1502 			if (len == 0xFF) { /* Variable length */
1503 				len = vfio_cap_len(vdev, cap, pos);
1504 				if (len < 0)
1505 					return len;
1506 			}
1507 		}
1508 
1509 		if (!len) {
1510 			pci_info(pdev, "%s: hiding cap %#x@%#x\n", __func__,
1511 				 cap, pos);
1512 			*prev = next;
1513 			pos = next;
1514 			continue;
1515 		}
1516 
1517 		/* Sanity check, do we overlap other capabilities? */
1518 		for (i = 0; i < len; i++) {
1519 			if (likely(map[pos + i] == PCI_CAP_ID_INVALID))
1520 				continue;
1521 
1522 			pci_warn(pdev, "%s: PCI config conflict @%#x, was cap %#x now cap %#x\n",
1523 				 __func__, pos + i, map[pos + i], cap);
1524 		}
1525 
1526 		BUILD_BUG_ON(PCI_CAP_ID_MAX >= PCI_CAP_ID_INVALID_VIRT);
1527 
1528 		memset(map + pos, cap, len);
1529 		ret = vfio_fill_vconfig_bytes(vdev, pos, len);
1530 		if (ret)
1531 			return ret;
1532 
1533 		prev = &vdev->vconfig[pos + PCI_CAP_LIST_NEXT];
1534 		pos = next;
1535 		caps++;
1536 	}
1537 
1538 	/* If we didn't fill any capabilities, clear the status flag */
1539 	if (!caps) {
1540 		__le16 *vstatus = (__le16 *)&vdev->vconfig[PCI_STATUS];
1541 		*vstatus &= ~cpu_to_le16(PCI_STATUS_CAP_LIST);
1542 	}
1543 
1544 	return 0;
1545 }
1546 
1547 static int vfio_ecap_init(struct vfio_pci_device *vdev)
1548 {
1549 	struct pci_dev *pdev = vdev->pdev;
1550 	u8 *map = vdev->pci_config_map;
1551 	u16 epos;
1552 	__le32 *prev = NULL;
1553 	int loops, ret, ecaps = 0;
1554 
1555 	if (!vdev->extended_caps)
1556 		return 0;
1557 
1558 	epos = PCI_CFG_SPACE_SIZE;
1559 
1560 	loops = (pdev->cfg_size - PCI_CFG_SPACE_SIZE) / PCI_CAP_SIZEOF;
1561 
1562 	while (loops-- && epos >= PCI_CFG_SPACE_SIZE) {
1563 		u32 header;
1564 		u16 ecap;
1565 		int i, len = 0;
1566 		bool hidden = false;
1567 
1568 		ret = pci_read_config_dword(pdev, epos, &header);
1569 		if (ret)
1570 			return ret;
1571 
1572 		ecap = PCI_EXT_CAP_ID(header);
1573 
1574 		if (ecap <= PCI_EXT_CAP_ID_MAX) {
1575 			len = pci_ext_cap_length[ecap];
1576 			if (len == 0xFF) {
1577 				len = vfio_ext_cap_len(vdev, ecap, epos);
1578 				if (len < 0)
1579 					return ret;
1580 			}
1581 		}
1582 
1583 		if (!len) {
1584 			pci_info(pdev, "%s: hiding ecap %#x@%#x\n",
1585 				 __func__, ecap, epos);
1586 
1587 			/* If not the first in the chain, we can skip over it */
1588 			if (prev) {
1589 				u32 val = epos = PCI_EXT_CAP_NEXT(header);
1590 				*prev &= cpu_to_le32(~(0xffcU << 20));
1591 				*prev |= cpu_to_le32(val << 20);
1592 				continue;
1593 			}
1594 
1595 			/*
1596 			 * Otherwise, fill in a placeholder, the direct
1597 			 * readfn will virtualize this automatically
1598 			 */
1599 			len = PCI_CAP_SIZEOF;
1600 			hidden = true;
1601 		}
1602 
1603 		for (i = 0; i < len; i++) {
1604 			if (likely(map[epos + i] == PCI_CAP_ID_INVALID))
1605 				continue;
1606 
1607 			pci_warn(pdev, "%s: PCI config conflict @%#x, was ecap %#x now ecap %#x\n",
1608 				 __func__, epos + i, map[epos + i], ecap);
1609 		}
1610 
1611 		/*
1612 		 * Even though ecap is 2 bytes, we're currently a long way
1613 		 * from exceeding 1 byte capabilities.  If we ever make it
1614 		 * up to 0xFE we'll need to up this to a two-byte, byte map.
1615 		 */
1616 		BUILD_BUG_ON(PCI_EXT_CAP_ID_MAX >= PCI_CAP_ID_INVALID_VIRT);
1617 
1618 		memset(map + epos, ecap, len);
1619 		ret = vfio_fill_vconfig_bytes(vdev, epos, len);
1620 		if (ret)
1621 			return ret;
1622 
1623 		/*
1624 		 * If we're just using this capability to anchor the list,
1625 		 * hide the real ID.  Only count real ecaps.  XXX PCI spec
1626 		 * indicates to use cap id = 0, version = 0, next = 0 if
1627 		 * ecaps are absent, hope users check all the way to next.
1628 		 */
1629 		if (hidden)
1630 			*(__le32 *)&vdev->vconfig[epos] &=
1631 				cpu_to_le32((0xffcU << 20));
1632 		else
1633 			ecaps++;
1634 
1635 		prev = (__le32 *)&vdev->vconfig[epos];
1636 		epos = PCI_EXT_CAP_NEXT(header);
1637 	}
1638 
1639 	if (!ecaps)
1640 		*(u32 *)&vdev->vconfig[PCI_CFG_SPACE_SIZE] = 0;
1641 
1642 	return 0;
1643 }
1644 
1645 /*
1646  * Nag about hardware bugs, hopefully to have vendors fix them, but at least
1647  * to collect a list of dependencies for the VF INTx pin quirk below.
1648  */
1649 static const struct pci_device_id known_bogus_vf_intx_pin[] = {
1650 	{ PCI_DEVICE(PCI_VENDOR_ID_INTEL, 0x270c) },
1651 	{}
1652 };
1653 
1654 /*
1655  * For each device we allocate a pci_config_map that indicates the
1656  * capability occupying each dword and thus the struct perm_bits we
1657  * use for read and write.  We also allocate a virtualized config
1658  * space which tracks reads and writes to bits that we emulate for
1659  * the user.  Initial values filled from device.
1660  *
1661  * Using shared struct perm_bits between all vfio-pci devices saves
1662  * us from allocating cfg_size buffers for virt and write for every
1663  * device.  We could remove vconfig and allocate individual buffers
1664  * for each area requiring emulated bits, but the array of pointers
1665  * would be comparable in size (at least for standard config space).
1666  */
1667 int vfio_config_init(struct vfio_pci_device *vdev)
1668 {
1669 	struct pci_dev *pdev = vdev->pdev;
1670 	u8 *map, *vconfig;
1671 	int ret;
1672 
1673 	/*
1674 	 * Config space, caps and ecaps are all dword aligned, so we could
1675 	 * use one byte per dword to record the type.  However, there are
1676 	 * no requiremenst on the length of a capability, so the gap between
1677 	 * capabilities needs byte granularity.
1678 	 */
1679 	map = kmalloc(pdev->cfg_size, GFP_KERNEL);
1680 	if (!map)
1681 		return -ENOMEM;
1682 
1683 	vconfig = kmalloc(pdev->cfg_size, GFP_KERNEL);
1684 	if (!vconfig) {
1685 		kfree(map);
1686 		return -ENOMEM;
1687 	}
1688 
1689 	vdev->pci_config_map = map;
1690 	vdev->vconfig = vconfig;
1691 
1692 	memset(map, PCI_CAP_ID_BASIC, PCI_STD_HEADER_SIZEOF);
1693 	memset(map + PCI_STD_HEADER_SIZEOF, PCI_CAP_ID_INVALID,
1694 	       pdev->cfg_size - PCI_STD_HEADER_SIZEOF);
1695 
1696 	ret = vfio_fill_vconfig_bytes(vdev, 0, PCI_STD_HEADER_SIZEOF);
1697 	if (ret)
1698 		goto out;
1699 
1700 	vdev->bardirty = true;
1701 
1702 	/*
1703 	 * XXX can we just pci_load_saved_state/pci_restore_state?
1704 	 * may need to rebuild vconfig after that
1705 	 */
1706 
1707 	/* For restore after reset */
1708 	vdev->rbar[0] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_0]);
1709 	vdev->rbar[1] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_1]);
1710 	vdev->rbar[2] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_2]);
1711 	vdev->rbar[3] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_3]);
1712 	vdev->rbar[4] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_4]);
1713 	vdev->rbar[5] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_5]);
1714 	vdev->rbar[6] = le32_to_cpu(*(__le32 *)&vconfig[PCI_ROM_ADDRESS]);
1715 
1716 	if (pdev->is_virtfn) {
1717 		*(__le16 *)&vconfig[PCI_VENDOR_ID] = cpu_to_le16(pdev->vendor);
1718 		*(__le16 *)&vconfig[PCI_DEVICE_ID] = cpu_to_le16(pdev->device);
1719 
1720 		/*
1721 		 * Per SR-IOV spec rev 1.1, 3.4.1.18 the interrupt pin register
1722 		 * does not apply to VFs and VFs must implement this register
1723 		 * as read-only with value zero.  Userspace is not readily able
1724 		 * to identify whether a device is a VF and thus that the pin
1725 		 * definition on the device is bogus should it violate this
1726 		 * requirement.  We already virtualize the pin register for
1727 		 * other purposes, so we simply need to replace the bogus value
1728 		 * and consider VFs when we determine INTx IRQ count.
1729 		 */
1730 		if (vconfig[PCI_INTERRUPT_PIN] &&
1731 		    !pci_match_id(known_bogus_vf_intx_pin, pdev))
1732 			pci_warn(pdev,
1733 				 "Hardware bug: VF reports bogus INTx pin %d\n",
1734 				 vconfig[PCI_INTERRUPT_PIN]);
1735 
1736 		vconfig[PCI_INTERRUPT_PIN] = 0; /* Gratuitous for good VFs */
1737 
1738 		/*
1739 		 * VFs do no implement the memory enable bit of the COMMAND
1740 		 * register therefore we'll not have it set in our initial
1741 		 * copy of config space after pci_enable_device().  For
1742 		 * consistency with PFs, set the virtual enable bit here.
1743 		 */
1744 		*(__le16 *)&vconfig[PCI_COMMAND] |=
1745 					cpu_to_le16(PCI_COMMAND_MEMORY);
1746 	}
1747 
1748 	if (!IS_ENABLED(CONFIG_VFIO_PCI_INTX) || vdev->nointx)
1749 		vconfig[PCI_INTERRUPT_PIN] = 0;
1750 
1751 	ret = vfio_cap_init(vdev);
1752 	if (ret)
1753 		goto out;
1754 
1755 	ret = vfio_ecap_init(vdev);
1756 	if (ret)
1757 		goto out;
1758 
1759 	return 0;
1760 
1761 out:
1762 	kfree(map);
1763 	vdev->pci_config_map = NULL;
1764 	kfree(vconfig);
1765 	vdev->vconfig = NULL;
1766 	return pcibios_err_to_errno(ret);
1767 }
1768 
1769 void vfio_config_free(struct vfio_pci_device *vdev)
1770 {
1771 	kfree(vdev->vconfig);
1772 	vdev->vconfig = NULL;
1773 	kfree(vdev->pci_config_map);
1774 	vdev->pci_config_map = NULL;
1775 	if (vdev->msi_perm) {
1776 		free_perm_bits(vdev->msi_perm);
1777 		kfree(vdev->msi_perm);
1778 		vdev->msi_perm = NULL;
1779 	}
1780 }
1781 
1782 /*
1783  * Find the remaining number of bytes in a dword that match the given
1784  * position.  Stop at either the end of the capability or the dword boundary.
1785  */
1786 static size_t vfio_pci_cap_remaining_dword(struct vfio_pci_device *vdev,
1787 					   loff_t pos)
1788 {
1789 	u8 cap = vdev->pci_config_map[pos];
1790 	size_t i;
1791 
1792 	for (i = 1; (pos + i) % 4 && vdev->pci_config_map[pos + i] == cap; i++)
1793 		/* nop */;
1794 
1795 	return i;
1796 }
1797 
1798 static ssize_t vfio_config_do_rw(struct vfio_pci_device *vdev, char __user *buf,
1799 				 size_t count, loff_t *ppos, bool iswrite)
1800 {
1801 	struct pci_dev *pdev = vdev->pdev;
1802 	struct perm_bits *perm;
1803 	__le32 val = 0;
1804 	int cap_start = 0, offset;
1805 	u8 cap_id;
1806 	ssize_t ret;
1807 
1808 	if (*ppos < 0 || *ppos >= pdev->cfg_size ||
1809 	    *ppos + count > pdev->cfg_size)
1810 		return -EFAULT;
1811 
1812 	/*
1813 	 * Chop accesses into aligned chunks containing no more than a
1814 	 * single capability.  Caller increments to the next chunk.
1815 	 */
1816 	count = min(count, vfio_pci_cap_remaining_dword(vdev, *ppos));
1817 	if (count >= 4 && !(*ppos % 4))
1818 		count = 4;
1819 	else if (count >= 2 && !(*ppos % 2))
1820 		count = 2;
1821 	else
1822 		count = 1;
1823 
1824 	ret = count;
1825 
1826 	cap_id = vdev->pci_config_map[*ppos];
1827 
1828 	if (cap_id == PCI_CAP_ID_INVALID) {
1829 		perm = &unassigned_perms;
1830 		cap_start = *ppos;
1831 	} else if (cap_id == PCI_CAP_ID_INVALID_VIRT) {
1832 		perm = &virt_perms;
1833 		cap_start = *ppos;
1834 	} else {
1835 		if (*ppos >= PCI_CFG_SPACE_SIZE) {
1836 			WARN_ON(cap_id > PCI_EXT_CAP_ID_MAX);
1837 
1838 			perm = &ecap_perms[cap_id];
1839 			cap_start = vfio_find_cap_start(vdev, *ppos);
1840 		} else {
1841 			WARN_ON(cap_id > PCI_CAP_ID_MAX);
1842 
1843 			perm = &cap_perms[cap_id];
1844 
1845 			if (cap_id == PCI_CAP_ID_MSI)
1846 				perm = vdev->msi_perm;
1847 
1848 			if (cap_id > PCI_CAP_ID_BASIC)
1849 				cap_start = vfio_find_cap_start(vdev, *ppos);
1850 		}
1851 	}
1852 
1853 	WARN_ON(!cap_start && cap_id != PCI_CAP_ID_BASIC);
1854 	WARN_ON(cap_start > *ppos);
1855 
1856 	offset = *ppos - cap_start;
1857 
1858 	if (iswrite) {
1859 		if (!perm->writefn)
1860 			return ret;
1861 
1862 		if (copy_from_user(&val, buf, count))
1863 			return -EFAULT;
1864 
1865 		ret = perm->writefn(vdev, *ppos, count, perm, offset, val);
1866 	} else {
1867 		if (perm->readfn) {
1868 			ret = perm->readfn(vdev, *ppos, count,
1869 					   perm, offset, &val);
1870 			if (ret < 0)
1871 				return ret;
1872 		}
1873 
1874 		if (copy_to_user(buf, &val, count))
1875 			return -EFAULT;
1876 	}
1877 
1878 	return ret;
1879 }
1880 
1881 ssize_t vfio_pci_config_rw(struct vfio_pci_device *vdev, char __user *buf,
1882 			   size_t count, loff_t *ppos, bool iswrite)
1883 {
1884 	size_t done = 0;
1885 	int ret = 0;
1886 	loff_t pos = *ppos;
1887 
1888 	pos &= VFIO_PCI_OFFSET_MASK;
1889 
1890 	while (count) {
1891 		ret = vfio_config_do_rw(vdev, buf, count, &pos, iswrite);
1892 		if (ret < 0)
1893 			return ret;
1894 
1895 		count -= ret;
1896 		done += ret;
1897 		buf += ret;
1898 		pos += ret;
1899 	}
1900 
1901 	*ppos += done;
1902 
1903 	return done;
1904 }
1905