1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * PCI Backend - Functions for creating a virtual configuration space for
4 * exported PCI Devices.
5 * It's dangerous to allow PCI Driver Domains to change their
6 * device's resources (memory, i/o ports, interrupts). We need to
7 * restrict changes to certain PCI Configuration registers:
8 * BARs, INTERRUPT_PIN, most registers in the header...
9 *
10 * Author: Ryan Wilson <hap9@epoch.ncsc.mil>
11 */
12
13 #define dev_fmt(fmt) DRV_NAME ": " fmt
14
15 #include <linux/kernel.h>
16 #include <linux/moduleparam.h>
17 #include <linux/pci.h>
18 #include "pciback.h"
19 #include "conf_space.h"
20 #include "conf_space_quirks.h"
21
22 bool xen_pcibk_permissive;
23 module_param_named(permissive, xen_pcibk_permissive, bool, 0644);
24
25 /* This is where xen_pcibk_read_config_byte, xen_pcibk_read_config_word,
26 * xen_pcibk_write_config_word, and xen_pcibk_write_config_byte are created. */
27 #define DEFINE_PCI_CONFIG(op, size, type) \
28 int xen_pcibk_##op##_config_##size \
29 (struct pci_dev *dev, int offset, type value, void *data) \
30 { \
31 return pci_##op##_config_##size(dev, offset, value); \
32 }
33
DEFINE_PCI_CONFIG(read,byte,u8 *)34 DEFINE_PCI_CONFIG(read, byte, u8 *)
35 DEFINE_PCI_CONFIG(read, word, u16 *)
36 DEFINE_PCI_CONFIG(read, dword, u32 *)
37
38 DEFINE_PCI_CONFIG(write, byte, u8)
39 DEFINE_PCI_CONFIG(write, word, u16)
40 DEFINE_PCI_CONFIG(write, dword, u32)
41
42 static int conf_space_read(struct pci_dev *dev,
43 const struct config_field_entry *entry,
44 int offset, u32 *value)
45 {
46 int ret = 0;
47 const struct config_field *field = entry->field;
48
49 *value = 0;
50
51 switch (field->size) {
52 case 1:
53 if (field->u.b.read)
54 ret = field->u.b.read(dev, offset, (u8 *) value,
55 entry->data);
56 break;
57 case 2:
58 if (field->u.w.read)
59 ret = field->u.w.read(dev, offset, (u16 *) value,
60 entry->data);
61 break;
62 case 4:
63 if (field->u.dw.read)
64 ret = field->u.dw.read(dev, offset, value, entry->data);
65 break;
66 }
67 return ret;
68 }
69
conf_space_write(struct pci_dev * dev,const struct config_field_entry * entry,int offset,u32 value)70 static int conf_space_write(struct pci_dev *dev,
71 const struct config_field_entry *entry,
72 int offset, u32 value)
73 {
74 int ret = 0;
75 const struct config_field *field = entry->field;
76
77 switch (field->size) {
78 case 1:
79 if (field->u.b.write)
80 ret = field->u.b.write(dev, offset, (u8) value,
81 entry->data);
82 break;
83 case 2:
84 if (field->u.w.write)
85 ret = field->u.w.write(dev, offset, (u16) value,
86 entry->data);
87 break;
88 case 4:
89 if (field->u.dw.write)
90 ret = field->u.dw.write(dev, offset, value,
91 entry->data);
92 break;
93 }
94 return ret;
95 }
96
get_mask(int size)97 static inline u32 get_mask(int size)
98 {
99 if (size == 1)
100 return 0xff;
101 else if (size == 2)
102 return 0xffff;
103 else
104 return 0xffffffff;
105 }
106
valid_request(int offset,int size)107 static inline int valid_request(int offset, int size)
108 {
109 /* Validate request (no un-aligned requests) */
110 if ((size == 1 || size == 2 || size == 4) && (offset % size) == 0)
111 return 1;
112 return 0;
113 }
114
merge_value(u32 val,u32 new_val,u32 new_val_mask,int offset)115 static inline u32 merge_value(u32 val, u32 new_val, u32 new_val_mask,
116 int offset)
117 {
118 if (offset >= 0) {
119 new_val_mask <<= (offset * 8);
120 new_val <<= (offset * 8);
121 } else {
122 new_val_mask >>= (offset * -8);
123 new_val >>= (offset * -8);
124 }
125 val = (val & ~new_val_mask) | (new_val & new_val_mask);
126
127 return val;
128 }
129
xen_pcibios_err_to_errno(int err)130 static int xen_pcibios_err_to_errno(int err)
131 {
132 switch (err) {
133 case PCIBIOS_SUCCESSFUL:
134 return XEN_PCI_ERR_success;
135 case PCIBIOS_DEVICE_NOT_FOUND:
136 return XEN_PCI_ERR_dev_not_found;
137 case PCIBIOS_BAD_REGISTER_NUMBER:
138 return XEN_PCI_ERR_invalid_offset;
139 case PCIBIOS_FUNC_NOT_SUPPORTED:
140 return XEN_PCI_ERR_not_implemented;
141 case PCIBIOS_SET_FAILED:
142 return XEN_PCI_ERR_access_denied;
143 }
144 return err;
145 }
146
xen_pcibk_config_read(struct pci_dev * dev,int offset,int size,u32 * ret_val)147 int xen_pcibk_config_read(struct pci_dev *dev, int offset, int size,
148 u32 *ret_val)
149 {
150 int err = 0;
151 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
152 const struct config_field_entry *cfg_entry;
153 const struct config_field *field;
154 int field_start, field_end;
155 /* if read fails for any reason, return 0
156 * (as if device didn't respond) */
157 u32 value = 0, tmp_val;
158
159 dev_dbg(&dev->dev, "read %d bytes at 0x%x\n", size, offset);
160
161 if (!valid_request(offset, size)) {
162 err = XEN_PCI_ERR_invalid_offset;
163 goto out;
164 }
165
166 /* Get the real value first, then modify as appropriate */
167 switch (size) {
168 case 1:
169 err = pci_read_config_byte(dev, offset, (u8 *) &value);
170 break;
171 case 2:
172 err = pci_read_config_word(dev, offset, (u16 *) &value);
173 break;
174 case 4:
175 err = pci_read_config_dword(dev, offset, &value);
176 break;
177 }
178
179 list_for_each_entry(cfg_entry, &dev_data->config_fields, list) {
180 field = cfg_entry->field;
181
182 field_start = OFFSET(cfg_entry);
183 field_end = OFFSET(cfg_entry) + field->size;
184
185 if (offset + size > field_start && field_end > offset) {
186 err = conf_space_read(dev, cfg_entry, field_start,
187 &tmp_val);
188 if (err)
189 goto out;
190
191 value = merge_value(value, tmp_val,
192 get_mask(field->size),
193 field_start - offset);
194 }
195 }
196
197 out:
198 dev_dbg(&dev->dev, "read %d bytes at 0x%x = %x\n", size, offset, value);
199
200 *ret_val = value;
201 return xen_pcibios_err_to_errno(err);
202 }
203
xen_pcibk_config_write(struct pci_dev * dev,int offset,int size,u32 value)204 int xen_pcibk_config_write(struct pci_dev *dev, int offset, int size, u32 value)
205 {
206 int err = 0, handled = 0;
207 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
208 const struct config_field_entry *cfg_entry;
209 const struct config_field *field;
210 u32 tmp_val;
211 int field_start, field_end;
212
213 dev_dbg(&dev->dev, "write request %d bytes at 0x%x = %x\n",
214 size, offset, value);
215
216 if (!valid_request(offset, size))
217 return XEN_PCI_ERR_invalid_offset;
218
219 list_for_each_entry(cfg_entry, &dev_data->config_fields, list) {
220 field = cfg_entry->field;
221
222 field_start = OFFSET(cfg_entry);
223 field_end = OFFSET(cfg_entry) + field->size;
224
225 if (offset + size > field_start && field_end > offset) {
226 err = conf_space_read(dev, cfg_entry, field_start,
227 &tmp_val);
228 if (err)
229 break;
230
231 tmp_val = merge_value(tmp_val, value, get_mask(size),
232 offset - field_start);
233
234 err = conf_space_write(dev, cfg_entry, field_start,
235 tmp_val);
236
237 /* handled is set true here, but not every byte
238 * may have been written! Properly detecting if
239 * every byte is handled is unnecessary as the
240 * flag is used to detect devices that need
241 * special helpers to work correctly.
242 */
243 handled = 1;
244 }
245 }
246
247 if (!handled && !err) {
248 /* By default, anything not specificially handled above is
249 * read-only. The permissive flag changes this behavior so
250 * that anything not specifically handled above is writable.
251 * This means that some fields may still be read-only because
252 * they have entries in the config_field list that intercept
253 * the write and do nothing. */
254 if (dev_data->permissive || xen_pcibk_permissive) {
255 switch (size) {
256 case 1:
257 err = pci_write_config_byte(dev, offset,
258 (u8) value);
259 break;
260 case 2:
261 err = pci_write_config_word(dev, offset,
262 (u16) value);
263 break;
264 case 4:
265 err = pci_write_config_dword(dev, offset,
266 (u32) value);
267 break;
268 }
269 } else if (!dev_data->warned_on_write) {
270 dev_data->warned_on_write = 1;
271 dev_warn(&dev->dev, "Driver tried to write to a "
272 "read-only configuration space field at offset"
273 " 0x%x, size %d. This may be harmless, but if "
274 "you have problems with your device:\n"
275 "1) see permissive attribute in sysfs\n"
276 "2) report problems to the xen-devel "
277 "mailing list along with details of your "
278 "device obtained from lspci.\n", offset, size);
279 }
280 }
281
282 return xen_pcibios_err_to_errno(err);
283 }
284
xen_pcibk_get_interrupt_type(struct pci_dev * dev)285 int xen_pcibk_get_interrupt_type(struct pci_dev *dev)
286 {
287 int err;
288 u16 val;
289 int ret = 0;
290
291 /*
292 * Do not trust dev->msi(x)_enabled here, as enabling could be done
293 * bypassing the pci_*msi* functions, by the qemu.
294 */
295 if (dev->msi_cap) {
296 err = pci_read_config_word(dev,
297 dev->msi_cap + PCI_MSI_FLAGS,
298 &val);
299 if (err)
300 return err;
301 if (val & PCI_MSI_FLAGS_ENABLE)
302 ret |= INTERRUPT_TYPE_MSI;
303 }
304 if (dev->msix_cap) {
305 err = pci_read_config_word(dev,
306 dev->msix_cap + PCI_MSIX_FLAGS,
307 &val);
308 if (err)
309 return err;
310 if (val & PCI_MSIX_FLAGS_ENABLE)
311 ret |= INTERRUPT_TYPE_MSIX;
312 }
313
314 /*
315 * PCIe spec says device cannot use INTx if MSI/MSI-X is enabled,
316 * so check for INTx only when both are disabled.
317 */
318 if (!ret) {
319 err = pci_read_config_word(dev, PCI_COMMAND, &val);
320 if (err)
321 return err;
322 if (!(val & PCI_COMMAND_INTX_DISABLE))
323 ret |= INTERRUPT_TYPE_INTX;
324 }
325
326 return ret ?: INTERRUPT_TYPE_NONE;
327 }
328
xen_pcibk_config_free_dyn_fields(struct pci_dev * dev)329 void xen_pcibk_config_free_dyn_fields(struct pci_dev *dev)
330 {
331 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
332 struct config_field_entry *cfg_entry, *t;
333 const struct config_field *field;
334
335 dev_dbg(&dev->dev, "free-ing dynamically allocated virtual "
336 "configuration space fields\n");
337 if (!dev_data)
338 return;
339
340 list_for_each_entry_safe(cfg_entry, t, &dev_data->config_fields, list) {
341 field = cfg_entry->field;
342
343 if (field->clean) {
344 field->clean((struct config_field *)field);
345
346 kfree(cfg_entry->data);
347
348 list_del(&cfg_entry->list);
349 kfree(cfg_entry);
350 }
351
352 }
353 }
354
xen_pcibk_config_reset_dev(struct pci_dev * dev)355 void xen_pcibk_config_reset_dev(struct pci_dev *dev)
356 {
357 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
358 const struct config_field_entry *cfg_entry;
359 const struct config_field *field;
360
361 dev_dbg(&dev->dev, "resetting virtual configuration space\n");
362 if (!dev_data)
363 return;
364
365 list_for_each_entry(cfg_entry, &dev_data->config_fields, list) {
366 field = cfg_entry->field;
367
368 if (field->reset)
369 field->reset(dev, OFFSET(cfg_entry), cfg_entry->data);
370 }
371 }
372
xen_pcibk_config_free_dev(struct pci_dev * dev)373 void xen_pcibk_config_free_dev(struct pci_dev *dev)
374 {
375 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
376 struct config_field_entry *cfg_entry, *t;
377 const struct config_field *field;
378
379 dev_dbg(&dev->dev, "free-ing virtual configuration space fields\n");
380 if (!dev_data)
381 return;
382
383 list_for_each_entry_safe(cfg_entry, t, &dev_data->config_fields, list) {
384 list_del(&cfg_entry->list);
385
386 field = cfg_entry->field;
387
388 if (field->release)
389 field->release(dev, OFFSET(cfg_entry), cfg_entry->data);
390
391 kfree(cfg_entry);
392 }
393 }
394
xen_pcibk_config_add_field_offset(struct pci_dev * dev,const struct config_field * field,unsigned int base_offset)395 int xen_pcibk_config_add_field_offset(struct pci_dev *dev,
396 const struct config_field *field,
397 unsigned int base_offset)
398 {
399 int err = 0;
400 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
401 struct config_field_entry *cfg_entry;
402 void *tmp;
403
404 cfg_entry = kmalloc(sizeof(*cfg_entry), GFP_KERNEL);
405 if (!cfg_entry) {
406 err = -ENOMEM;
407 goto out;
408 }
409
410 cfg_entry->data = NULL;
411 cfg_entry->field = field;
412 cfg_entry->base_offset = base_offset;
413
414 /* silently ignore duplicate fields */
415 err = xen_pcibk_field_is_dup(dev, OFFSET(cfg_entry));
416 if (err)
417 goto out;
418
419 if (field->init) {
420 tmp = field->init(dev, OFFSET(cfg_entry));
421
422 if (IS_ERR(tmp)) {
423 err = PTR_ERR(tmp);
424 goto out;
425 }
426
427 cfg_entry->data = tmp;
428 }
429
430 dev_dbg(&dev->dev, "added config field at offset 0x%02x\n",
431 OFFSET(cfg_entry));
432 list_add_tail(&cfg_entry->list, &dev_data->config_fields);
433
434 out:
435 if (err)
436 kfree(cfg_entry);
437
438 return err;
439 }
440
441 /* This sets up the device's virtual configuration space to keep track of
442 * certain registers (like the base address registers (BARs) so that we can
443 * keep the client from manipulating them directly.
444 */
xen_pcibk_config_init_dev(struct pci_dev * dev)445 int xen_pcibk_config_init_dev(struct pci_dev *dev)
446 {
447 int err = 0;
448 struct xen_pcibk_dev_data *dev_data = pci_get_drvdata(dev);
449
450 dev_dbg(&dev->dev, "initializing virtual configuration space\n");
451
452 INIT_LIST_HEAD(&dev_data->config_fields);
453
454 err = xen_pcibk_config_header_add_fields(dev);
455 if (err)
456 goto out;
457
458 err = xen_pcibk_config_capability_add_fields(dev);
459 if (err)
460 goto out;
461
462 err = xen_pcibk_config_quirks_init(dev);
463
464 out:
465 return err;
466 }
467
xen_pcibk_config_init(void)468 int xen_pcibk_config_init(void)
469 {
470 return xen_pcibk_config_capability_init();
471 }
472