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