• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 const u8 pci_cap_length[PCI_CAP_ID_MAX + 1] = {
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 const u16 pci_ext_cap_length[PCI_EXT_CAP_ID_MAX + 1] = {
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 
vfio_user_config_read(struct pci_dev * pdev,int offset,__le32 * val,int count)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 
vfio_user_config_write(struct pci_dev * pdev,int offset,__le32 val,int count)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 
vfio_default_config_read(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 * val)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 
vfio_default_config_write(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 val)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 */
vfio_direct_config_read(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 * val)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 */
vfio_raw_config_write(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 val)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 
vfio_raw_config_read(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 * val)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 
free_perm_bits(struct perm_bits * perm)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 
alloc_perm_bits(struct perm_bits * perm,int size)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  */
p_setb(struct perm_bits * p,int off,u8 virt,u8 write)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 */
p_setw(struct perm_bits * p,int off,u16 virt,u16 write)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 */
p_setd(struct perm_bits * p,int off,u32 virt,u32 write)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  */
vfio_bar_restore(struct vfio_pci_device * vdev)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 
vfio_generate_bar_flags(struct pci_dev * pdev,int bar)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  */
vfio_bar_fixup(struct vfio_pci_device * vdev)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 
vfio_basic_config_read(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 * val)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 
vfio_basic_config_write(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 val)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 */
init_pci_cap_basic_perm(struct perm_bits * perm)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 
vfio_pm_config_write(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 val)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 */
init_pci_cap_pm_perm(struct perm_bits * perm)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 
vfio_vpd_config_write(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 val)674 static int vfio_vpd_config_write(struct vfio_pci_device *vdev, int pos,
675 				 int count, struct perm_bits *perm,
676 				 int offset, __le32 val)
677 {
678 	struct pci_dev *pdev = vdev->pdev;
679 	__le16 *paddr = (__le16 *)(vdev->vconfig + pos - offset + PCI_VPD_ADDR);
680 	__le32 *pdata = (__le32 *)(vdev->vconfig + pos - offset + PCI_VPD_DATA);
681 	u16 addr;
682 	u32 data;
683 
684 	/*
685 	 * Write through to emulation.  If the write includes the upper byte
686 	 * of PCI_VPD_ADDR, then the PCI_VPD_ADDR_F bit is written and we
687 	 * have work to do.
688 	 */
689 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
690 	if (count < 0 || offset > PCI_VPD_ADDR + 1 ||
691 	    offset + count <= PCI_VPD_ADDR + 1)
692 		return count;
693 
694 	addr = le16_to_cpu(*paddr);
695 
696 	if (addr & PCI_VPD_ADDR_F) {
697 		data = le32_to_cpu(*pdata);
698 		if (pci_write_vpd(pdev, addr & ~PCI_VPD_ADDR_F, 4, &data) != 4)
699 			return count;
700 	} else {
701 		data = 0;
702 		if (pci_read_vpd(pdev, addr, 4, &data) < 0)
703 			return count;
704 		*pdata = cpu_to_le32(data);
705 	}
706 
707 	/*
708 	 * Toggle PCI_VPD_ADDR_F in the emulated PCI_VPD_ADDR register to
709 	 * signal completion.  If an error occurs above, we assume that not
710 	 * toggling this bit will induce a driver timeout.
711 	 */
712 	addr ^= PCI_VPD_ADDR_F;
713 	*paddr = cpu_to_le16(addr);
714 
715 	return count;
716 }
717 
718 /* Permissions for Vital Product Data capability */
init_pci_cap_vpd_perm(struct perm_bits * perm)719 static int __init init_pci_cap_vpd_perm(struct perm_bits *perm)
720 {
721 	if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_VPD]))
722 		return -ENOMEM;
723 
724 	perm->writefn = vfio_vpd_config_write;
725 
726 	/*
727 	 * We always virtualize the next field so we can remove
728 	 * capabilities from the chain if we want to.
729 	 */
730 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
731 
732 	/*
733 	 * Both the address and data registers are virtualized to
734 	 * enable access through the pci_vpd_read/write functions
735 	 */
736 	p_setw(perm, PCI_VPD_ADDR, (u16)ALL_VIRT, (u16)ALL_WRITE);
737 	p_setd(perm, PCI_VPD_DATA, ALL_VIRT, ALL_WRITE);
738 
739 	return 0;
740 }
741 
742 /* Permissions for PCI-X capability */
init_pci_cap_pcix_perm(struct perm_bits * perm)743 static int __init init_pci_cap_pcix_perm(struct perm_bits *perm)
744 {
745 	/* Alloc 24, but only 8 are used in v0 */
746 	if (alloc_perm_bits(perm, PCI_CAP_PCIX_SIZEOF_V2))
747 		return -ENOMEM;
748 
749 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
750 
751 	p_setw(perm, PCI_X_CMD, NO_VIRT, (u16)ALL_WRITE);
752 	p_setd(perm, PCI_X_ECC_CSR, NO_VIRT, ALL_WRITE);
753 	return 0;
754 }
755 
vfio_exp_config_write(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 val)756 static int vfio_exp_config_write(struct vfio_pci_device *vdev, int pos,
757 				 int count, struct perm_bits *perm,
758 				 int offset, __le32 val)
759 {
760 	__le16 *ctrl = (__le16 *)(vdev->vconfig + pos -
761 				  offset + PCI_EXP_DEVCTL);
762 	int readrq = le16_to_cpu(*ctrl) & PCI_EXP_DEVCTL_READRQ;
763 
764 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
765 	if (count < 0)
766 		return count;
767 
768 	/*
769 	 * The FLR bit is virtualized, if set and the device supports PCIe
770 	 * FLR, issue a reset_function.  Regardless, clear the bit, the spec
771 	 * requires it to be always read as zero.  NB, reset_function might
772 	 * not use a PCIe FLR, we don't have that level of granularity.
773 	 */
774 	if (*ctrl & cpu_to_le16(PCI_EXP_DEVCTL_BCR_FLR)) {
775 		u32 cap;
776 		int ret;
777 
778 		*ctrl &= ~cpu_to_le16(PCI_EXP_DEVCTL_BCR_FLR);
779 
780 		ret = pci_user_read_config_dword(vdev->pdev,
781 						 pos - offset + PCI_EXP_DEVCAP,
782 						 &cap);
783 
784 		if (!ret && (cap & PCI_EXP_DEVCAP_FLR))
785 			pci_try_reset_function(vdev->pdev);
786 	}
787 
788 	/*
789 	 * MPS is virtualized to the user, writes do not change the physical
790 	 * register since determining a proper MPS value requires a system wide
791 	 * device view.  The MRRS is largely independent of MPS, but since the
792 	 * user does not have that system-wide view, they might set a safe, but
793 	 * inefficiently low value.  Here we allow writes through to hardware,
794 	 * but we set the floor to the physical device MPS setting, so that
795 	 * we can at least use full TLPs, as defined by the MPS value.
796 	 *
797 	 * NB, if any devices actually depend on an artificially low MRRS
798 	 * setting, this will need to be revisited, perhaps with a quirk
799 	 * though pcie_set_readrq().
800 	 */
801 	if (readrq != (le16_to_cpu(*ctrl) & PCI_EXP_DEVCTL_READRQ)) {
802 		readrq = 128 <<
803 			((le16_to_cpu(*ctrl) & PCI_EXP_DEVCTL_READRQ) >> 12);
804 		readrq = max(readrq, pcie_get_mps(vdev->pdev));
805 
806 		pcie_set_readrq(vdev->pdev, readrq);
807 	}
808 
809 	return count;
810 }
811 
812 /* Permissions for PCI Express capability */
init_pci_cap_exp_perm(struct perm_bits * perm)813 static int __init init_pci_cap_exp_perm(struct perm_bits *perm)
814 {
815 	/* Alloc larger of two possible sizes */
816 	if (alloc_perm_bits(perm, PCI_CAP_EXP_ENDPOINT_SIZEOF_V2))
817 		return -ENOMEM;
818 
819 	perm->writefn = vfio_exp_config_write;
820 
821 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
822 
823 	/*
824 	 * Allow writes to device control fields, except devctl_phantom,
825 	 * which could confuse IOMMU, MPS, which can break communication
826 	 * with other physical devices, and the ARI bit in devctl2, which
827 	 * is set at probe time.  FLR and MRRS get virtualized via our
828 	 * writefn.
829 	 */
830 	p_setw(perm, PCI_EXP_DEVCTL,
831 	       PCI_EXP_DEVCTL_BCR_FLR | PCI_EXP_DEVCTL_PAYLOAD |
832 	       PCI_EXP_DEVCTL_READRQ, ~PCI_EXP_DEVCTL_PHANTOM);
833 	p_setw(perm, PCI_EXP_DEVCTL2, NO_VIRT, ~PCI_EXP_DEVCTL2_ARI);
834 	return 0;
835 }
836 
vfio_af_config_write(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 val)837 static int vfio_af_config_write(struct vfio_pci_device *vdev, int pos,
838 				int count, struct perm_bits *perm,
839 				int offset, __le32 val)
840 {
841 	u8 *ctrl = vdev->vconfig + pos - offset + PCI_AF_CTRL;
842 
843 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
844 	if (count < 0)
845 		return count;
846 
847 	/*
848 	 * The FLR bit is virtualized, if set and the device supports AF
849 	 * FLR, issue a reset_function.  Regardless, clear the bit, the spec
850 	 * requires it to be always read as zero.  NB, reset_function might
851 	 * not use an AF FLR, we don't have that level of granularity.
852 	 */
853 	if (*ctrl & PCI_AF_CTRL_FLR) {
854 		u8 cap;
855 		int ret;
856 
857 		*ctrl &= ~PCI_AF_CTRL_FLR;
858 
859 		ret = pci_user_read_config_byte(vdev->pdev,
860 						pos - offset + PCI_AF_CAP,
861 						&cap);
862 
863 		if (!ret && (cap & PCI_AF_CAP_FLR) && (cap & PCI_AF_CAP_TP))
864 			pci_try_reset_function(vdev->pdev);
865 	}
866 
867 	return count;
868 }
869 
870 /* Permissions for Advanced Function capability */
init_pci_cap_af_perm(struct perm_bits * perm)871 static int __init init_pci_cap_af_perm(struct perm_bits *perm)
872 {
873 	if (alloc_perm_bits(perm, pci_cap_length[PCI_CAP_ID_AF]))
874 		return -ENOMEM;
875 
876 	perm->writefn = vfio_af_config_write;
877 
878 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
879 	p_setb(perm, PCI_AF_CTRL, PCI_AF_CTRL_FLR, PCI_AF_CTRL_FLR);
880 	return 0;
881 }
882 
883 /* Permissions for Advanced Error Reporting extended capability */
init_pci_ext_cap_err_perm(struct perm_bits * perm)884 static int __init init_pci_ext_cap_err_perm(struct perm_bits *perm)
885 {
886 	u32 mask;
887 
888 	if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_ERR]))
889 		return -ENOMEM;
890 
891 	/*
892 	 * Virtualize the first dword of all express capabilities
893 	 * because it includes the next pointer.  This lets us later
894 	 * remove capabilities from the chain if we need to.
895 	 */
896 	p_setd(perm, 0, ALL_VIRT, NO_WRITE);
897 
898 	/* Writable bits mask */
899 	mask =	PCI_ERR_UNC_UND |		/* Undefined */
900 		PCI_ERR_UNC_DLP |		/* Data Link Protocol */
901 		PCI_ERR_UNC_SURPDN |		/* Surprise Down */
902 		PCI_ERR_UNC_POISON_TLP |	/* Poisoned TLP */
903 		PCI_ERR_UNC_FCP |		/* Flow Control Protocol */
904 		PCI_ERR_UNC_COMP_TIME |		/* Completion Timeout */
905 		PCI_ERR_UNC_COMP_ABORT |	/* Completer Abort */
906 		PCI_ERR_UNC_UNX_COMP |		/* Unexpected Completion */
907 		PCI_ERR_UNC_RX_OVER |		/* Receiver Overflow */
908 		PCI_ERR_UNC_MALF_TLP |		/* Malformed TLP */
909 		PCI_ERR_UNC_ECRC |		/* ECRC Error Status */
910 		PCI_ERR_UNC_UNSUP |		/* Unsupported Request */
911 		PCI_ERR_UNC_ACSV |		/* ACS Violation */
912 		PCI_ERR_UNC_INTN |		/* internal error */
913 		PCI_ERR_UNC_MCBTLP |		/* MC blocked TLP */
914 		PCI_ERR_UNC_ATOMEG |		/* Atomic egress blocked */
915 		PCI_ERR_UNC_TLPPRE;		/* TLP prefix blocked */
916 	p_setd(perm, PCI_ERR_UNCOR_STATUS, NO_VIRT, mask);
917 	p_setd(perm, PCI_ERR_UNCOR_MASK, NO_VIRT, mask);
918 	p_setd(perm, PCI_ERR_UNCOR_SEVER, NO_VIRT, mask);
919 
920 	mask =	PCI_ERR_COR_RCVR |		/* Receiver Error Status */
921 		PCI_ERR_COR_BAD_TLP |		/* Bad TLP Status */
922 		PCI_ERR_COR_BAD_DLLP |		/* Bad DLLP Status */
923 		PCI_ERR_COR_REP_ROLL |		/* REPLAY_NUM Rollover */
924 		PCI_ERR_COR_REP_TIMER |		/* Replay Timer Timeout */
925 		PCI_ERR_COR_ADV_NFAT |		/* Advisory Non-Fatal */
926 		PCI_ERR_COR_INTERNAL |		/* Corrected Internal */
927 		PCI_ERR_COR_LOG_OVER;		/* Header Log Overflow */
928 	p_setd(perm, PCI_ERR_COR_STATUS, NO_VIRT, mask);
929 	p_setd(perm, PCI_ERR_COR_MASK, NO_VIRT, mask);
930 
931 	mask =	PCI_ERR_CAP_ECRC_GENE |		/* ECRC Generation Enable */
932 		PCI_ERR_CAP_ECRC_CHKE;		/* ECRC Check Enable */
933 	p_setd(perm, PCI_ERR_CAP, NO_VIRT, mask);
934 	return 0;
935 }
936 
937 /* Permissions for Power Budgeting extended capability */
init_pci_ext_cap_pwr_perm(struct perm_bits * perm)938 static int __init init_pci_ext_cap_pwr_perm(struct perm_bits *perm)
939 {
940 	if (alloc_perm_bits(perm, pci_ext_cap_length[PCI_EXT_CAP_ID_PWR]))
941 		return -ENOMEM;
942 
943 	p_setd(perm, 0, ALL_VIRT, NO_WRITE);
944 
945 	/* Writing the data selector is OK, the info is still read-only */
946 	p_setb(perm, PCI_PWR_DATA, NO_VIRT, (u8)ALL_WRITE);
947 	return 0;
948 }
949 
950 /*
951  * Initialize the shared permission tables
952  */
vfio_pci_uninit_perm_bits(void)953 void vfio_pci_uninit_perm_bits(void)
954 {
955 	free_perm_bits(&cap_perms[PCI_CAP_ID_BASIC]);
956 
957 	free_perm_bits(&cap_perms[PCI_CAP_ID_PM]);
958 	free_perm_bits(&cap_perms[PCI_CAP_ID_VPD]);
959 	free_perm_bits(&cap_perms[PCI_CAP_ID_PCIX]);
960 	free_perm_bits(&cap_perms[PCI_CAP_ID_EXP]);
961 	free_perm_bits(&cap_perms[PCI_CAP_ID_AF]);
962 
963 	free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_ERR]);
964 	free_perm_bits(&ecap_perms[PCI_EXT_CAP_ID_PWR]);
965 }
966 
vfio_pci_init_perm_bits(void)967 int __init vfio_pci_init_perm_bits(void)
968 {
969 	int ret;
970 
971 	/* Basic config space */
972 	ret = init_pci_cap_basic_perm(&cap_perms[PCI_CAP_ID_BASIC]);
973 
974 	/* Capabilities */
975 	ret |= init_pci_cap_pm_perm(&cap_perms[PCI_CAP_ID_PM]);
976 	ret |= init_pci_cap_vpd_perm(&cap_perms[PCI_CAP_ID_VPD]);
977 	ret |= init_pci_cap_pcix_perm(&cap_perms[PCI_CAP_ID_PCIX]);
978 	cap_perms[PCI_CAP_ID_VNDR].writefn = vfio_raw_config_write;
979 	ret |= init_pci_cap_exp_perm(&cap_perms[PCI_CAP_ID_EXP]);
980 	ret |= init_pci_cap_af_perm(&cap_perms[PCI_CAP_ID_AF]);
981 
982 	/* Extended capabilities */
983 	ret |= init_pci_ext_cap_err_perm(&ecap_perms[PCI_EXT_CAP_ID_ERR]);
984 	ret |= init_pci_ext_cap_pwr_perm(&ecap_perms[PCI_EXT_CAP_ID_PWR]);
985 	ecap_perms[PCI_EXT_CAP_ID_VNDR].writefn = vfio_raw_config_write;
986 
987 	if (ret)
988 		vfio_pci_uninit_perm_bits();
989 
990 	return ret;
991 }
992 
vfio_find_cap_start(struct vfio_pci_device * vdev,int pos)993 static int vfio_find_cap_start(struct vfio_pci_device *vdev, int pos)
994 {
995 	u8 cap;
996 	int base = (pos >= PCI_CFG_SPACE_SIZE) ? PCI_CFG_SPACE_SIZE :
997 						 PCI_STD_HEADER_SIZEOF;
998 	cap = vdev->pci_config_map[pos];
999 
1000 	if (cap == PCI_CAP_ID_BASIC)
1001 		return 0;
1002 
1003 	/* XXX Can we have to abutting capabilities of the same type? */
1004 	while (pos - 1 >= base && vdev->pci_config_map[pos - 1] == cap)
1005 		pos--;
1006 
1007 	return pos;
1008 }
1009 
vfio_msi_config_read(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 * val)1010 static int vfio_msi_config_read(struct vfio_pci_device *vdev, int pos,
1011 				int count, struct perm_bits *perm,
1012 				int offset, __le32 *val)
1013 {
1014 	/* Update max available queue size from msi_qmax */
1015 	if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
1016 		__le16 *flags;
1017 		int start;
1018 
1019 		start = vfio_find_cap_start(vdev, pos);
1020 
1021 		flags = (__le16 *)&vdev->vconfig[start];
1022 
1023 		*flags &= cpu_to_le16(~PCI_MSI_FLAGS_QMASK);
1024 		*flags |= cpu_to_le16(vdev->msi_qmax << 1);
1025 	}
1026 
1027 	return vfio_default_config_read(vdev, pos, count, perm, offset, val);
1028 }
1029 
vfio_msi_config_write(struct vfio_pci_device * vdev,int pos,int count,struct perm_bits * perm,int offset,__le32 val)1030 static int vfio_msi_config_write(struct vfio_pci_device *vdev, int pos,
1031 				 int count, struct perm_bits *perm,
1032 				 int offset, __le32 val)
1033 {
1034 	count = vfio_default_config_write(vdev, pos, count, perm, offset, val);
1035 	if (count < 0)
1036 		return count;
1037 
1038 	/* Fixup and write configured queue size and enable to hardware */
1039 	if (offset <= PCI_MSI_FLAGS && offset + count >= PCI_MSI_FLAGS) {
1040 		__le16 *pflags;
1041 		u16 flags;
1042 		int start, ret;
1043 
1044 		start = vfio_find_cap_start(vdev, pos);
1045 
1046 		pflags = (__le16 *)&vdev->vconfig[start + PCI_MSI_FLAGS];
1047 
1048 		flags = le16_to_cpu(*pflags);
1049 
1050 		/* MSI is enabled via ioctl */
1051 		if  (!is_msi(vdev))
1052 			flags &= ~PCI_MSI_FLAGS_ENABLE;
1053 
1054 		/* Check queue size */
1055 		if ((flags & PCI_MSI_FLAGS_QSIZE) >> 4 > vdev->msi_qmax) {
1056 			flags &= ~PCI_MSI_FLAGS_QSIZE;
1057 			flags |= vdev->msi_qmax << 4;
1058 		}
1059 
1060 		/* Write back to virt and to hardware */
1061 		*pflags = cpu_to_le16(flags);
1062 		ret = pci_user_write_config_word(vdev->pdev,
1063 						 start + PCI_MSI_FLAGS,
1064 						 flags);
1065 		if (ret)
1066 			return pcibios_err_to_errno(ret);
1067 	}
1068 
1069 	return count;
1070 }
1071 
1072 /*
1073  * MSI determination is per-device, so this routine gets used beyond
1074  * initialization time. Don't add __init
1075  */
init_pci_cap_msi_perm(struct perm_bits * perm,int len,u16 flags)1076 static int init_pci_cap_msi_perm(struct perm_bits *perm, int len, u16 flags)
1077 {
1078 	if (alloc_perm_bits(perm, len))
1079 		return -ENOMEM;
1080 
1081 	perm->readfn = vfio_msi_config_read;
1082 	perm->writefn = vfio_msi_config_write;
1083 
1084 	p_setb(perm, PCI_CAP_LIST_NEXT, (u8)ALL_VIRT, NO_WRITE);
1085 
1086 	/*
1087 	 * The upper byte of the control register is reserved,
1088 	 * just setup the lower byte.
1089 	 */
1090 	p_setb(perm, PCI_MSI_FLAGS, (u8)ALL_VIRT, (u8)ALL_WRITE);
1091 	p_setd(perm, PCI_MSI_ADDRESS_LO, ALL_VIRT, ALL_WRITE);
1092 	if (flags & PCI_MSI_FLAGS_64BIT) {
1093 		p_setd(perm, PCI_MSI_ADDRESS_HI, ALL_VIRT, ALL_WRITE);
1094 		p_setw(perm, PCI_MSI_DATA_64, (u16)ALL_VIRT, (u16)ALL_WRITE);
1095 		if (flags & PCI_MSI_FLAGS_MASKBIT) {
1096 			p_setd(perm, PCI_MSI_MASK_64, NO_VIRT, ALL_WRITE);
1097 			p_setd(perm, PCI_MSI_PENDING_64, NO_VIRT, ALL_WRITE);
1098 		}
1099 	} else {
1100 		p_setw(perm, PCI_MSI_DATA_32, (u16)ALL_VIRT, (u16)ALL_WRITE);
1101 		if (flags & PCI_MSI_FLAGS_MASKBIT) {
1102 			p_setd(perm, PCI_MSI_MASK_32, NO_VIRT, ALL_WRITE);
1103 			p_setd(perm, PCI_MSI_PENDING_32, NO_VIRT, ALL_WRITE);
1104 		}
1105 	}
1106 	return 0;
1107 }
1108 
1109 /* Determine MSI CAP field length; initialize msi_perms on 1st call per vdev */
vfio_msi_cap_len(struct vfio_pci_device * vdev,u8 pos)1110 static int vfio_msi_cap_len(struct vfio_pci_device *vdev, u8 pos)
1111 {
1112 	struct pci_dev *pdev = vdev->pdev;
1113 	int len, ret;
1114 	u16 flags;
1115 
1116 	ret = pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &flags);
1117 	if (ret)
1118 		return pcibios_err_to_errno(ret);
1119 
1120 	len = 10; /* Minimum size */
1121 	if (flags & PCI_MSI_FLAGS_64BIT)
1122 		len += 4;
1123 	if (flags & PCI_MSI_FLAGS_MASKBIT)
1124 		len += 10;
1125 
1126 	if (vdev->msi_perm)
1127 		return len;
1128 
1129 	vdev->msi_perm = kmalloc(sizeof(struct perm_bits), GFP_KERNEL);
1130 	if (!vdev->msi_perm)
1131 		return -ENOMEM;
1132 
1133 	ret = init_pci_cap_msi_perm(vdev->msi_perm, len, flags);
1134 	if (ret) {
1135 		kfree(vdev->msi_perm);
1136 		return ret;
1137 	}
1138 
1139 	return len;
1140 }
1141 
1142 /* Determine extended capability length for VC (2 & 9) and MFVC */
vfio_vc_cap_len(struct vfio_pci_device * vdev,u16 pos)1143 static int vfio_vc_cap_len(struct vfio_pci_device *vdev, u16 pos)
1144 {
1145 	struct pci_dev *pdev = vdev->pdev;
1146 	u32 tmp;
1147 	int ret, evcc, phases, vc_arb;
1148 	int len = PCI_CAP_VC_BASE_SIZEOF;
1149 
1150 	ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_CAP1, &tmp);
1151 	if (ret)
1152 		return pcibios_err_to_errno(ret);
1153 
1154 	evcc = tmp & PCI_VC_CAP1_EVCC; /* extended vc count */
1155 	ret = pci_read_config_dword(pdev, pos + PCI_VC_PORT_CAP2, &tmp);
1156 	if (ret)
1157 		return pcibios_err_to_errno(ret);
1158 
1159 	if (tmp & PCI_VC_CAP2_128_PHASE)
1160 		phases = 128;
1161 	else if (tmp & PCI_VC_CAP2_64_PHASE)
1162 		phases = 64;
1163 	else if (tmp & PCI_VC_CAP2_32_PHASE)
1164 		phases = 32;
1165 	else
1166 		phases = 0;
1167 
1168 	vc_arb = phases * 4;
1169 
1170 	/*
1171 	 * Port arbitration tables are root & switch only;
1172 	 * function arbitration tables are function 0 only.
1173 	 * In either case, we'll never let user write them so
1174 	 * we don't care how big they are
1175 	 */
1176 	len += (1 + evcc) * PCI_CAP_VC_PER_VC_SIZEOF;
1177 	if (vc_arb) {
1178 		len = round_up(len, 16);
1179 		len += vc_arb / 8;
1180 	}
1181 	return len;
1182 }
1183 
vfio_cap_len(struct vfio_pci_device * vdev,u8 cap,u8 pos)1184 static int vfio_cap_len(struct vfio_pci_device *vdev, u8 cap, u8 pos)
1185 {
1186 	struct pci_dev *pdev = vdev->pdev;
1187 	u32 dword;
1188 	u16 word;
1189 	u8 byte;
1190 	int ret;
1191 
1192 	switch (cap) {
1193 	case PCI_CAP_ID_MSI:
1194 		return vfio_msi_cap_len(vdev, pos);
1195 	case PCI_CAP_ID_PCIX:
1196 		ret = pci_read_config_word(pdev, pos + PCI_X_CMD, &word);
1197 		if (ret)
1198 			return pcibios_err_to_errno(ret);
1199 
1200 		if (PCI_X_CMD_VERSION(word)) {
1201 			/* Test for extended capabilities */
1202 			pci_read_config_dword(pdev, PCI_CFG_SPACE_SIZE, &dword);
1203 			vdev->extended_caps = (dword != 0);
1204 			return PCI_CAP_PCIX_SIZEOF_V2;
1205 		} else
1206 			return PCI_CAP_PCIX_SIZEOF_V0;
1207 	case PCI_CAP_ID_VNDR:
1208 		/* length follows next field */
1209 		ret = pci_read_config_byte(pdev, pos + PCI_CAP_FLAGS, &byte);
1210 		if (ret)
1211 			return pcibios_err_to_errno(ret);
1212 
1213 		return byte;
1214 	case PCI_CAP_ID_EXP:
1215 		/* Test for extended capabilities */
1216 		pci_read_config_dword(pdev, PCI_CFG_SPACE_SIZE, &dword);
1217 		vdev->extended_caps = (dword != 0);
1218 
1219 		/* length based on version */
1220 		if ((pcie_caps_reg(pdev) & PCI_EXP_FLAGS_VERS) == 1)
1221 			return PCI_CAP_EXP_ENDPOINT_SIZEOF_V1;
1222 		else
1223 			return PCI_CAP_EXP_ENDPOINT_SIZEOF_V2;
1224 	case PCI_CAP_ID_HT:
1225 		ret = pci_read_config_byte(pdev, pos + 3, &byte);
1226 		if (ret)
1227 			return pcibios_err_to_errno(ret);
1228 
1229 		return (byte & HT_3BIT_CAP_MASK) ?
1230 			HT_CAP_SIZEOF_SHORT : HT_CAP_SIZEOF_LONG;
1231 	case PCI_CAP_ID_SATA:
1232 		ret = pci_read_config_byte(pdev, pos + PCI_SATA_REGS, &byte);
1233 		if (ret)
1234 			return pcibios_err_to_errno(ret);
1235 
1236 		byte &= PCI_SATA_REGS_MASK;
1237 		if (byte == PCI_SATA_REGS_INLINE)
1238 			return PCI_SATA_SIZEOF_LONG;
1239 		else
1240 			return PCI_SATA_SIZEOF_SHORT;
1241 	default:
1242 		pr_warn("%s: %s unknown length for pci cap 0x%x@0x%x\n",
1243 			dev_name(&pdev->dev), __func__, cap, pos);
1244 	}
1245 
1246 	return 0;
1247 }
1248 
vfio_ext_cap_len(struct vfio_pci_device * vdev,u16 ecap,u16 epos)1249 static int vfio_ext_cap_len(struct vfio_pci_device *vdev, u16 ecap, u16 epos)
1250 {
1251 	struct pci_dev *pdev = vdev->pdev;
1252 	u8 byte;
1253 	u32 dword;
1254 	int ret;
1255 
1256 	switch (ecap) {
1257 	case PCI_EXT_CAP_ID_VNDR:
1258 		ret = pci_read_config_dword(pdev, epos + PCI_VSEC_HDR, &dword);
1259 		if (ret)
1260 			return pcibios_err_to_errno(ret);
1261 
1262 		return dword >> PCI_VSEC_HDR_LEN_SHIFT;
1263 	case PCI_EXT_CAP_ID_VC:
1264 	case PCI_EXT_CAP_ID_VC9:
1265 	case PCI_EXT_CAP_ID_MFVC:
1266 		return vfio_vc_cap_len(vdev, epos);
1267 	case PCI_EXT_CAP_ID_ACS:
1268 		ret = pci_read_config_byte(pdev, epos + PCI_ACS_CAP, &byte);
1269 		if (ret)
1270 			return pcibios_err_to_errno(ret);
1271 
1272 		if (byte & PCI_ACS_EC) {
1273 			int bits;
1274 
1275 			ret = pci_read_config_byte(pdev,
1276 						   epos + PCI_ACS_EGRESS_BITS,
1277 						   &byte);
1278 			if (ret)
1279 				return pcibios_err_to_errno(ret);
1280 
1281 			bits = byte ? round_up(byte, 32) : 256;
1282 			return 8 + (bits / 8);
1283 		}
1284 		return 8;
1285 
1286 	case PCI_EXT_CAP_ID_REBAR:
1287 		ret = pci_read_config_byte(pdev, epos + PCI_REBAR_CTRL, &byte);
1288 		if (ret)
1289 			return pcibios_err_to_errno(ret);
1290 
1291 		byte &= PCI_REBAR_CTRL_NBAR_MASK;
1292 		byte >>= PCI_REBAR_CTRL_NBAR_SHIFT;
1293 
1294 		return 4 + (byte * 8);
1295 	case PCI_EXT_CAP_ID_DPA:
1296 		ret = pci_read_config_byte(pdev, epos + PCI_DPA_CAP, &byte);
1297 		if (ret)
1298 			return pcibios_err_to_errno(ret);
1299 
1300 		byte &= PCI_DPA_CAP_SUBSTATE_MASK;
1301 		return PCI_DPA_BASE_SIZEOF + byte + 1;
1302 	case PCI_EXT_CAP_ID_TPH:
1303 		ret = pci_read_config_dword(pdev, epos + PCI_TPH_CAP, &dword);
1304 		if (ret)
1305 			return pcibios_err_to_errno(ret);
1306 
1307 		if ((dword & PCI_TPH_CAP_LOC_MASK) == PCI_TPH_LOC_CAP) {
1308 			int sts;
1309 
1310 			sts = dword & PCI_TPH_CAP_ST_MASK;
1311 			sts >>= PCI_TPH_CAP_ST_SHIFT;
1312 			return PCI_TPH_BASE_SIZEOF + (sts * 2) + 2;
1313 		}
1314 		return PCI_TPH_BASE_SIZEOF;
1315 	default:
1316 		pr_warn("%s: %s unknown length for pci ecap 0x%x@0x%x\n",
1317 			dev_name(&pdev->dev), __func__, ecap, epos);
1318 	}
1319 
1320 	return 0;
1321 }
1322 
vfio_fill_vconfig_bytes(struct vfio_pci_device * vdev,int offset,int size)1323 static int vfio_fill_vconfig_bytes(struct vfio_pci_device *vdev,
1324 				   int offset, int size)
1325 {
1326 	struct pci_dev *pdev = vdev->pdev;
1327 	int ret = 0;
1328 
1329 	/*
1330 	 * We try to read physical config space in the largest chunks
1331 	 * we can, assuming that all of the fields support dword access.
1332 	 * pci_save_state() makes this same assumption and seems to do ok.
1333 	 */
1334 	while (size) {
1335 		int filled;
1336 
1337 		if (size >= 4 && !(offset % 4)) {
1338 			__le32 *dwordp = (__le32 *)&vdev->vconfig[offset];
1339 			u32 dword;
1340 
1341 			ret = pci_read_config_dword(pdev, offset, &dword);
1342 			if (ret)
1343 				return ret;
1344 			*dwordp = cpu_to_le32(dword);
1345 			filled = 4;
1346 		} else if (size >= 2 && !(offset % 2)) {
1347 			__le16 *wordp = (__le16 *)&vdev->vconfig[offset];
1348 			u16 word;
1349 
1350 			ret = pci_read_config_word(pdev, offset, &word);
1351 			if (ret)
1352 				return ret;
1353 			*wordp = cpu_to_le16(word);
1354 			filled = 2;
1355 		} else {
1356 			u8 *byte = &vdev->vconfig[offset];
1357 			ret = pci_read_config_byte(pdev, offset, byte);
1358 			if (ret)
1359 				return ret;
1360 			filled = 1;
1361 		}
1362 
1363 		offset += filled;
1364 		size -= filled;
1365 	}
1366 
1367 	return ret;
1368 }
1369 
vfio_cap_init(struct vfio_pci_device * vdev)1370 static int vfio_cap_init(struct vfio_pci_device *vdev)
1371 {
1372 	struct pci_dev *pdev = vdev->pdev;
1373 	u8 *map = vdev->pci_config_map;
1374 	u16 status;
1375 	u8 pos, *prev, cap;
1376 	int loops, ret, caps = 0;
1377 
1378 	/* Any capabilities? */
1379 	ret = pci_read_config_word(pdev, PCI_STATUS, &status);
1380 	if (ret)
1381 		return ret;
1382 
1383 	if (!(status & PCI_STATUS_CAP_LIST))
1384 		return 0; /* Done */
1385 
1386 	ret = pci_read_config_byte(pdev, PCI_CAPABILITY_LIST, &pos);
1387 	if (ret)
1388 		return ret;
1389 
1390 	/* Mark the previous position in case we want to skip a capability */
1391 	prev = &vdev->vconfig[PCI_CAPABILITY_LIST];
1392 
1393 	/* We can bound our loop, capabilities are dword aligned */
1394 	loops = (PCI_CFG_SPACE_SIZE - PCI_STD_HEADER_SIZEOF) / PCI_CAP_SIZEOF;
1395 	while (pos && loops--) {
1396 		u8 next;
1397 		int i, len = 0;
1398 
1399 		ret = pci_read_config_byte(pdev, pos, &cap);
1400 		if (ret)
1401 			return ret;
1402 
1403 		ret = pci_read_config_byte(pdev,
1404 					   pos + PCI_CAP_LIST_NEXT, &next);
1405 		if (ret)
1406 			return ret;
1407 
1408 		/*
1409 		 * ID 0 is a NULL capability, conflicting with our fake
1410 		 * PCI_CAP_ID_BASIC.  As it has no content, consider it
1411 		 * hidden for now.
1412 		 */
1413 		if (cap && cap <= PCI_CAP_ID_MAX) {
1414 			len = pci_cap_length[cap];
1415 			if (len == 0xFF) { /* Variable length */
1416 				len = vfio_cap_len(vdev, cap, pos);
1417 				if (len < 0)
1418 					return len;
1419 			}
1420 		}
1421 
1422 		if (!len) {
1423 			pr_info("%s: %s hiding cap 0x%x\n",
1424 				__func__, dev_name(&pdev->dev), cap);
1425 			*prev = next;
1426 			pos = next;
1427 			continue;
1428 		}
1429 
1430 		/* Sanity check, do we overlap other capabilities? */
1431 		for (i = 0; i < len; i++) {
1432 			if (likely(map[pos + i] == PCI_CAP_ID_INVALID))
1433 				continue;
1434 
1435 			pr_warn("%s: %s pci config conflict @0x%x, was cap 0x%x now cap 0x%x\n",
1436 				__func__, dev_name(&pdev->dev),
1437 				pos + i, map[pos + i], cap);
1438 		}
1439 
1440 		memset(map + pos, cap, len);
1441 		ret = vfio_fill_vconfig_bytes(vdev, pos, len);
1442 		if (ret)
1443 			return ret;
1444 
1445 		prev = &vdev->vconfig[pos + PCI_CAP_LIST_NEXT];
1446 		pos = next;
1447 		caps++;
1448 	}
1449 
1450 	/* If we didn't fill any capabilities, clear the status flag */
1451 	if (!caps) {
1452 		__le16 *vstatus = (__le16 *)&vdev->vconfig[PCI_STATUS];
1453 		*vstatus &= ~cpu_to_le16(PCI_STATUS_CAP_LIST);
1454 	}
1455 
1456 	return 0;
1457 }
1458 
vfio_ecap_init(struct vfio_pci_device * vdev)1459 static int vfio_ecap_init(struct vfio_pci_device *vdev)
1460 {
1461 	struct pci_dev *pdev = vdev->pdev;
1462 	u8 *map = vdev->pci_config_map;
1463 	u16 epos;
1464 	__le32 *prev = NULL;
1465 	int loops, ret, ecaps = 0;
1466 
1467 	if (!vdev->extended_caps)
1468 		return 0;
1469 
1470 	epos = PCI_CFG_SPACE_SIZE;
1471 
1472 	loops = (pdev->cfg_size - PCI_CFG_SPACE_SIZE) / PCI_CAP_SIZEOF;
1473 
1474 	while (loops-- && epos >= PCI_CFG_SPACE_SIZE) {
1475 		u32 header;
1476 		u16 ecap;
1477 		int i, len = 0;
1478 		bool hidden = false;
1479 
1480 		ret = pci_read_config_dword(pdev, epos, &header);
1481 		if (ret)
1482 			return ret;
1483 
1484 		ecap = PCI_EXT_CAP_ID(header);
1485 
1486 		if (ecap <= PCI_EXT_CAP_ID_MAX) {
1487 			len = pci_ext_cap_length[ecap];
1488 			if (len == 0xFF) {
1489 				len = vfio_ext_cap_len(vdev, ecap, epos);
1490 				if (len < 0)
1491 					return len;
1492 			}
1493 		}
1494 
1495 		if (!len) {
1496 			pr_info("%s: %s hiding ecap 0x%x@0x%x\n",
1497 				__func__, dev_name(&pdev->dev), ecap, epos);
1498 
1499 			/* If not the first in the chain, we can skip over it */
1500 			if (prev) {
1501 				u32 val = epos = PCI_EXT_CAP_NEXT(header);
1502 				*prev &= cpu_to_le32(~(0xffcU << 20));
1503 				*prev |= cpu_to_le32(val << 20);
1504 				continue;
1505 			}
1506 
1507 			/*
1508 			 * Otherwise, fill in a placeholder, the direct
1509 			 * readfn will virtualize this automatically
1510 			 */
1511 			len = PCI_CAP_SIZEOF;
1512 			hidden = true;
1513 		}
1514 
1515 		for (i = 0; i < len; i++) {
1516 			if (likely(map[epos + i] == PCI_CAP_ID_INVALID))
1517 				continue;
1518 
1519 			pr_warn("%s: %s pci config conflict @0x%x, was ecap 0x%x now ecap 0x%x\n",
1520 				__func__, dev_name(&pdev->dev),
1521 				epos + i, map[epos + i], ecap);
1522 		}
1523 
1524 		/*
1525 		 * Even though ecap is 2 bytes, we're currently a long way
1526 		 * from exceeding 1 byte capabilities.  If we ever make it
1527 		 * up to 0xFF we'll need to up this to a two-byte, byte map.
1528 		 */
1529 		BUILD_BUG_ON(PCI_EXT_CAP_ID_MAX >= PCI_CAP_ID_INVALID);
1530 
1531 		memset(map + epos, ecap, len);
1532 		ret = vfio_fill_vconfig_bytes(vdev, epos, len);
1533 		if (ret)
1534 			return ret;
1535 
1536 		/*
1537 		 * If we're just using this capability to anchor the list,
1538 		 * hide the real ID.  Only count real ecaps.  XXX PCI spec
1539 		 * indicates to use cap id = 0, version = 0, next = 0 if
1540 		 * ecaps are absent, hope users check all the way to next.
1541 		 */
1542 		if (hidden)
1543 			*(__le32 *)&vdev->vconfig[epos] &=
1544 				cpu_to_le32((0xffcU << 20));
1545 		else
1546 			ecaps++;
1547 
1548 		prev = (__le32 *)&vdev->vconfig[epos];
1549 		epos = PCI_EXT_CAP_NEXT(header);
1550 	}
1551 
1552 	if (!ecaps)
1553 		*(u32 *)&vdev->vconfig[PCI_CFG_SPACE_SIZE] = 0;
1554 
1555 	return 0;
1556 }
1557 
1558 /*
1559  * For each device we allocate a pci_config_map that indicates the
1560  * capability occupying each dword and thus the struct perm_bits we
1561  * use for read and write.  We also allocate a virtualized config
1562  * space which tracks reads and writes to bits that we emulate for
1563  * the user.  Initial values filled from device.
1564  *
1565  * Using shared stuct perm_bits between all vfio-pci devices saves
1566  * us from allocating cfg_size buffers for virt and write for every
1567  * device.  We could remove vconfig and allocate individual buffers
1568  * for each area requring emulated bits, but the array of pointers
1569  * would be comparable in size (at least for standard config space).
1570  */
vfio_config_init(struct vfio_pci_device * vdev)1571 int vfio_config_init(struct vfio_pci_device *vdev)
1572 {
1573 	struct pci_dev *pdev = vdev->pdev;
1574 	u8 *map, *vconfig;
1575 	int ret;
1576 
1577 	/*
1578 	 * Config space, caps and ecaps are all dword aligned, so we could
1579 	 * use one byte per dword to record the type.  However, there are
1580 	 * no requiremenst on the length of a capability, so the gap between
1581 	 * capabilities needs byte granularity.
1582 	 */
1583 	map = kmalloc(pdev->cfg_size, GFP_KERNEL);
1584 	if (!map)
1585 		return -ENOMEM;
1586 
1587 	vconfig = kmalloc(pdev->cfg_size, GFP_KERNEL);
1588 	if (!vconfig) {
1589 		kfree(map);
1590 		return -ENOMEM;
1591 	}
1592 
1593 	vdev->pci_config_map = map;
1594 	vdev->vconfig = vconfig;
1595 
1596 	memset(map, PCI_CAP_ID_BASIC, PCI_STD_HEADER_SIZEOF);
1597 	memset(map + PCI_STD_HEADER_SIZEOF, PCI_CAP_ID_INVALID,
1598 	       pdev->cfg_size - PCI_STD_HEADER_SIZEOF);
1599 
1600 	ret = vfio_fill_vconfig_bytes(vdev, 0, PCI_STD_HEADER_SIZEOF);
1601 	if (ret)
1602 		goto out;
1603 
1604 	vdev->bardirty = true;
1605 
1606 	/*
1607 	 * XXX can we just pci_load_saved_state/pci_restore_state?
1608 	 * may need to rebuild vconfig after that
1609 	 */
1610 
1611 	/* For restore after reset */
1612 	vdev->rbar[0] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_0]);
1613 	vdev->rbar[1] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_1]);
1614 	vdev->rbar[2] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_2]);
1615 	vdev->rbar[3] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_3]);
1616 	vdev->rbar[4] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_4]);
1617 	vdev->rbar[5] = le32_to_cpu(*(__le32 *)&vconfig[PCI_BASE_ADDRESS_5]);
1618 	vdev->rbar[6] = le32_to_cpu(*(__le32 *)&vconfig[PCI_ROM_ADDRESS]);
1619 
1620 	if (pdev->is_virtfn) {
1621 		*(__le16 *)&vconfig[PCI_VENDOR_ID] = cpu_to_le16(pdev->vendor);
1622 		*(__le16 *)&vconfig[PCI_DEVICE_ID] = cpu_to_le16(pdev->device);
1623 	}
1624 
1625 	if (!IS_ENABLED(CONFIG_VFIO_PCI_INTX))
1626 		vconfig[PCI_INTERRUPT_PIN] = 0;
1627 
1628 	ret = vfio_cap_init(vdev);
1629 	if (ret)
1630 		goto out;
1631 
1632 	ret = vfio_ecap_init(vdev);
1633 	if (ret)
1634 		goto out;
1635 
1636 	return 0;
1637 
1638 out:
1639 	kfree(map);
1640 	vdev->pci_config_map = NULL;
1641 	kfree(vconfig);
1642 	vdev->vconfig = NULL;
1643 	return pcibios_err_to_errno(ret);
1644 }
1645 
vfio_config_free(struct vfio_pci_device * vdev)1646 void vfio_config_free(struct vfio_pci_device *vdev)
1647 {
1648 	kfree(vdev->vconfig);
1649 	vdev->vconfig = NULL;
1650 	kfree(vdev->pci_config_map);
1651 	vdev->pci_config_map = NULL;
1652 	if (vdev->msi_perm) {
1653 		free_perm_bits(vdev->msi_perm);
1654 		kfree(vdev->msi_perm);
1655 		vdev->msi_perm = NULL;
1656 	}
1657 }
1658 
1659 /*
1660  * Find the remaining number of bytes in a dword that match the given
1661  * position.  Stop at either the end of the capability or the dword boundary.
1662  */
vfio_pci_cap_remaining_dword(struct vfio_pci_device * vdev,loff_t pos)1663 static size_t vfio_pci_cap_remaining_dword(struct vfio_pci_device *vdev,
1664 					   loff_t pos)
1665 {
1666 	u8 cap = vdev->pci_config_map[pos];
1667 	size_t i;
1668 
1669 	for (i = 1; (pos + i) % 4 && vdev->pci_config_map[pos + i] == cap; i++)
1670 		/* nop */;
1671 
1672 	return i;
1673 }
1674 
vfio_config_do_rw(struct vfio_pci_device * vdev,char __user * buf,size_t count,loff_t * ppos,bool iswrite)1675 static ssize_t vfio_config_do_rw(struct vfio_pci_device *vdev, char __user *buf,
1676 				 size_t count, loff_t *ppos, bool iswrite)
1677 {
1678 	struct pci_dev *pdev = vdev->pdev;
1679 	struct perm_bits *perm;
1680 	__le32 val = 0;
1681 	int cap_start = 0, offset;
1682 	u8 cap_id;
1683 	ssize_t ret;
1684 
1685 	if (*ppos < 0 || *ppos >= pdev->cfg_size ||
1686 	    *ppos + count > pdev->cfg_size)
1687 		return -EFAULT;
1688 
1689 	/*
1690 	 * Chop accesses into aligned chunks containing no more than a
1691 	 * single capability.  Caller increments to the next chunk.
1692 	 */
1693 	count = min(count, vfio_pci_cap_remaining_dword(vdev, *ppos));
1694 	if (count >= 4 && !(*ppos % 4))
1695 		count = 4;
1696 	else if (count >= 2 && !(*ppos % 2))
1697 		count = 2;
1698 	else
1699 		count = 1;
1700 
1701 	ret = count;
1702 
1703 	cap_id = vdev->pci_config_map[*ppos];
1704 
1705 	if (cap_id == PCI_CAP_ID_INVALID) {
1706 		perm = &unassigned_perms;
1707 		cap_start = *ppos;
1708 	} else {
1709 		if (*ppos >= PCI_CFG_SPACE_SIZE) {
1710 			WARN_ON(cap_id > PCI_EXT_CAP_ID_MAX);
1711 
1712 			perm = &ecap_perms[cap_id];
1713 			cap_start = vfio_find_cap_start(vdev, *ppos);
1714 		} else {
1715 			WARN_ON(cap_id > PCI_CAP_ID_MAX);
1716 
1717 			perm = &cap_perms[cap_id];
1718 
1719 			if (cap_id == PCI_CAP_ID_MSI)
1720 				perm = vdev->msi_perm;
1721 
1722 			if (cap_id > PCI_CAP_ID_BASIC)
1723 				cap_start = vfio_find_cap_start(vdev, *ppos);
1724 		}
1725 	}
1726 
1727 	WARN_ON(!cap_start && cap_id != PCI_CAP_ID_BASIC);
1728 	WARN_ON(cap_start > *ppos);
1729 
1730 	offset = *ppos - cap_start;
1731 
1732 	if (iswrite) {
1733 		if (!perm->writefn)
1734 			return ret;
1735 
1736 		if (copy_from_user(&val, buf, count))
1737 			return -EFAULT;
1738 
1739 		ret = perm->writefn(vdev, *ppos, count, perm, offset, val);
1740 	} else {
1741 		if (perm->readfn) {
1742 			ret = perm->readfn(vdev, *ppos, count,
1743 					   perm, offset, &val);
1744 			if (ret < 0)
1745 				return ret;
1746 		}
1747 
1748 		if (copy_to_user(buf, &val, count))
1749 			return -EFAULT;
1750 	}
1751 
1752 	return ret;
1753 }
1754 
vfio_pci_config_rw(struct vfio_pci_device * vdev,char __user * buf,size_t count,loff_t * ppos,bool iswrite)1755 ssize_t vfio_pci_config_rw(struct vfio_pci_device *vdev, char __user *buf,
1756 			   size_t count, loff_t *ppos, bool iswrite)
1757 {
1758 	size_t done = 0;
1759 	int ret = 0;
1760 	loff_t pos = *ppos;
1761 
1762 	pos &= VFIO_PCI_OFFSET_MASK;
1763 
1764 	while (count) {
1765 		ret = vfio_config_do_rw(vdev, buf, count, &pos, iswrite);
1766 		if (ret < 0)
1767 			return ret;
1768 
1769 		count -= ret;
1770 		done += ret;
1771 		buf += ret;
1772 		pos += ret;
1773 	}
1774 
1775 	*ppos += done;
1776 
1777 	return done;
1778 }
1779