]> git.proxmox.com Git - mirror_qemu.git/blob - hw/pcie.c
Merge branch 'pci' into for_anthony
[mirror_qemu.git] / hw / pcie.c
1 /*
2 * pcie.c
3 *
4 * Copyright (c) 2010 Isaku Yamahata <yamahata at valinux co jp>
5 * VA Linux Systems Japan K.K.
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 as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License along
18 * with this program; if not, see <http://www.gnu.org/licenses/>.
19 */
20
21 #include "sysemu.h"
22 #include "range.h"
23 #include "pci_bridge.h"
24 #include "pcie.h"
25 #include "msix.h"
26 #include "msi.h"
27 #include "pci_internals.h"
28 #include "pcie_regs.h"
29 #include "range.h"
30
31 //#define DEBUG_PCIE
32 #ifdef DEBUG_PCIE
33 # define PCIE_DPRINTF(fmt, ...) \
34 fprintf(stderr, "%s:%d " fmt, __func__, __LINE__, ## __VA_ARGS__)
35 #else
36 # define PCIE_DPRINTF(fmt, ...) do {} while (0)
37 #endif
38 #define PCIE_DEV_PRINTF(dev, fmt, ...) \
39 PCIE_DPRINTF("%s:%x "fmt, (dev)->name, (dev)->devfn, ## __VA_ARGS__)
40
41
42 /***************************************************************************
43 * pci express capability helper functions
44 */
45 int pcie_cap_init(PCIDevice *dev, uint8_t offset, uint8_t type, uint8_t port)
46 {
47 int pos;
48 uint8_t *exp_cap;
49
50 assert(pci_is_express(dev));
51
52 pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
53 PCI_EXP_VER2_SIZEOF);
54 if (pos < 0) {
55 return pos;
56 }
57 dev->exp.exp_cap = pos;
58 exp_cap = dev->config + pos;
59
60 /* capability register
61 interrupt message number defaults to 0 */
62 pci_set_word(exp_cap + PCI_EXP_FLAGS,
63 ((type << PCI_EXP_FLAGS_TYPE_SHIFT) & PCI_EXP_FLAGS_TYPE) |
64 PCI_EXP_FLAGS_VER2);
65
66 /* device capability register
67 * table 7-12:
68 * roll based error reporting bit must be set by all
69 * Functions conforming to the ECN, PCI Express Base
70 * Specification, Revision 1.1., or subsequent PCI Express Base
71 * Specification revisions.
72 */
73 pci_set_long(exp_cap + PCI_EXP_DEVCAP, PCI_EXP_DEVCAP_RBER);
74
75 pci_set_long(exp_cap + PCI_EXP_LNKCAP,
76 (port << PCI_EXP_LNKCAP_PN_SHIFT) |
77 PCI_EXP_LNKCAP_ASPMS_0S |
78 PCI_EXP_LNK_MLW_1 |
79 PCI_EXP_LNK_LS_25);
80
81 pci_set_word(exp_cap + PCI_EXP_LNKSTA,
82 PCI_EXP_LNK_MLW_1 | PCI_EXP_LNK_LS_25);
83
84 pci_set_long(exp_cap + PCI_EXP_DEVCAP2,
85 PCI_EXP_DEVCAP2_EFF | PCI_EXP_DEVCAP2_EETLPP);
86
87 pci_set_word(dev->wmask + pos, PCI_EXP_DEVCTL2_EETLPPB);
88 return pos;
89 }
90
91 void pcie_cap_exit(PCIDevice *dev)
92 {
93 pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER2_SIZEOF);
94 }
95
96 uint8_t pcie_cap_get_type(const PCIDevice *dev)
97 {
98 uint32_t pos = dev->exp.exp_cap;
99 assert(pos > 0);
100 return (pci_get_word(dev->config + pos + PCI_EXP_FLAGS) &
101 PCI_EXP_FLAGS_TYPE) >> PCI_EXP_FLAGS_TYPE_SHIFT;
102 }
103
104 /* MSI/MSI-X */
105 /* pci express interrupt message number */
106 /* 7.8.2 PCI Express Capabilities Register: Interrupt Message Number */
107 void pcie_cap_flags_set_vector(PCIDevice *dev, uint8_t vector)
108 {
109 uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
110 assert(vector < 32);
111 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_FLAGS, PCI_EXP_FLAGS_IRQ);
112 pci_word_test_and_set_mask(exp_cap + PCI_EXP_FLAGS,
113 vector << PCI_EXP_FLAGS_IRQ_SHIFT);
114 }
115
116 uint8_t pcie_cap_flags_get_vector(PCIDevice *dev)
117 {
118 return (pci_get_word(dev->config + dev->exp.exp_cap + PCI_EXP_FLAGS) &
119 PCI_EXP_FLAGS_IRQ) >> PCI_EXP_FLAGS_IRQ_SHIFT;
120 }
121
122 void pcie_cap_deverr_init(PCIDevice *dev)
123 {
124 uint32_t pos = dev->exp.exp_cap;
125 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP,
126 PCI_EXP_DEVCAP_RBER);
127 pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL,
128 PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
129 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
130 pci_long_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_DEVSTA,
131 PCI_EXP_DEVSTA_CED | PCI_EXP_DEVSTA_NFED |
132 PCI_EXP_DEVSTA_URD | PCI_EXP_DEVSTA_URD);
133 }
134
135 void pcie_cap_deverr_reset(PCIDevice *dev)
136 {
137 uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
138 pci_long_test_and_clear_mask(devctl,
139 PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
140 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
141 }
142
143 static void hotplug_event_update_event_status(PCIDevice *dev)
144 {
145 uint32_t pos = dev->exp.exp_cap;
146 uint8_t *exp_cap = dev->config + pos;
147 uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
148 uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
149
150 dev->exp.hpev_notified = (sltctl & PCI_EXP_SLTCTL_HPIE) &&
151 (sltsta & sltctl & PCI_EXP_HP_EV_SUPPORTED);
152 }
153
154 static void hotplug_event_notify(PCIDevice *dev)
155 {
156 bool prev = dev->exp.hpev_notified;
157
158 hotplug_event_update_event_status(dev);
159
160 if (prev == dev->exp.hpev_notified) {
161 return;
162 }
163
164 /* Note: the logic above does not take into account whether interrupts
165 * are masked. The result is that interrupt will be sent when it is
166 * subsequently unmasked. This appears to be legal: Section 6.7.3.4:
167 * The Port may optionally send an MSI when there are hot-plug events that
168 * occur while interrupt generation is disabled, and interrupt generation is
169 * subsequently enabled. */
170 if (!pci_msi_enabled(dev)) {
171 qemu_set_irq(dev->irq[dev->exp.hpev_intx], dev->exp.hpev_notified);
172 } else if (dev->exp.hpev_notified) {
173 pci_msi_notify(dev, pcie_cap_flags_get_vector(dev));
174 }
175 }
176
177 /*
178 * A PCI Express Hot-Plug Event has occured, so update slot status register
179 * and notify OS of the event if necessary.
180 *
181 * 6.7.3 PCI Express Hot-Plug Events
182 * 6.7.3.4 Software Notification of Hot-Plug Events
183 */
184 static void pcie_cap_slot_event(PCIDevice *dev, PCIExpressHotPlugEvent event)
185 {
186 /* Minor optimization: if nothing changed - no event is needed. */
187 if (pci_word_test_and_set_mask(dev->config + dev->exp.exp_cap +
188 PCI_EXP_SLTSTA, event)) {
189 return;
190 }
191 hotplug_event_notify(dev);
192 }
193
194 static int pcie_cap_slot_hotplug(DeviceState *qdev,
195 PCIDevice *pci_dev, int state)
196 {
197 PCIDevice *d = DO_UPCAST(PCIDevice, qdev, qdev);
198 uint8_t *exp_cap = d->config + d->exp.exp_cap;
199 uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
200
201 if (!pci_dev->qdev.hotplugged) {
202 assert(state); /* this case only happens at machine creation. */
203 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
204 PCI_EXP_SLTSTA_PDS);
205 return 0;
206 }
207
208 PCIE_DEV_PRINTF(pci_dev, "hotplug state: %d\n", state);
209 if (sltsta & PCI_EXP_SLTSTA_EIS) {
210 /* the slot is electromechanically locked.
211 * This error is propagated up to qdev and then to HMP/QMP.
212 */
213 return -EBUSY;
214 }
215
216 /* TODO: multifunction hot-plug.
217 * Right now, only a device of function = 0 is allowed to be
218 * hot plugged/unplugged.
219 */
220 assert(PCI_FUNC(pci_dev->devfn) == 0);
221
222 if (state) {
223 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
224 PCI_EXP_SLTSTA_PDS);
225 pcie_cap_slot_event(d, PCI_EXP_HP_EV_PDC);
226 } else {
227 qdev_free(&pci_dev->qdev);
228 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
229 PCI_EXP_SLTSTA_PDS);
230 pcie_cap_slot_event(d, PCI_EXP_HP_EV_PDC);
231 }
232 return 0;
233 }
234
235 /* pci express slot for pci express root/downstream port
236 PCI express capability slot registers */
237 void pcie_cap_slot_init(PCIDevice *dev, uint16_t slot)
238 {
239 uint32_t pos = dev->exp.exp_cap;
240
241 pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_FLAGS,
242 PCI_EXP_FLAGS_SLOT);
243
244 pci_long_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCAP,
245 ~PCI_EXP_SLTCAP_PSN);
246 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
247 (slot << PCI_EXP_SLTCAP_PSN_SHIFT) |
248 PCI_EXP_SLTCAP_EIP |
249 PCI_EXP_SLTCAP_HPS |
250 PCI_EXP_SLTCAP_HPC |
251 PCI_EXP_SLTCAP_PIP |
252 PCI_EXP_SLTCAP_AIP |
253 PCI_EXP_SLTCAP_ABP);
254
255 pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
256 PCI_EXP_SLTCTL_PIC |
257 PCI_EXP_SLTCTL_AIC);
258 pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCTL,
259 PCI_EXP_SLTCTL_PIC_OFF |
260 PCI_EXP_SLTCTL_AIC_OFF);
261 pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
262 PCI_EXP_SLTCTL_PIC |
263 PCI_EXP_SLTCTL_AIC |
264 PCI_EXP_SLTCTL_HPIE |
265 PCI_EXP_SLTCTL_CCIE |
266 PCI_EXP_SLTCTL_PDCE |
267 PCI_EXP_SLTCTL_ABPE);
268 /* Although reading PCI_EXP_SLTCTL_EIC returns always 0,
269 * make the bit writable here in order to detect 1b is written.
270 * pcie_cap_slot_write_config() test-and-clear the bit, so
271 * this bit always returns 0 to the guest.
272 */
273 pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
274 PCI_EXP_SLTCTL_EIC);
275
276 pci_word_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_SLTSTA,
277 PCI_EXP_HP_EV_SUPPORTED);
278
279 dev->exp.hpev_notified = false;
280
281 pci_bus_hotplug(pci_bridge_get_sec_bus(DO_UPCAST(PCIBridge, dev, dev)),
282 pcie_cap_slot_hotplug, &dev->qdev);
283 }
284
285 void pcie_cap_slot_reset(PCIDevice *dev)
286 {
287 uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
288
289 PCIE_DEV_PRINTF(dev, "reset\n");
290
291 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
292 PCI_EXP_SLTCTL_EIC |
293 PCI_EXP_SLTCTL_PIC |
294 PCI_EXP_SLTCTL_AIC |
295 PCI_EXP_SLTCTL_HPIE |
296 PCI_EXP_SLTCTL_CCIE |
297 PCI_EXP_SLTCTL_PDCE |
298 PCI_EXP_SLTCTL_ABPE);
299 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
300 PCI_EXP_SLTCTL_PIC_OFF |
301 PCI_EXP_SLTCTL_AIC_OFF);
302
303 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
304 PCI_EXP_SLTSTA_EIS |/* on reset,
305 the lock is released */
306 PCI_EXP_SLTSTA_CC |
307 PCI_EXP_SLTSTA_PDC |
308 PCI_EXP_SLTSTA_ABP);
309
310 hotplug_event_update_event_status(dev);
311 }
312
313 void pcie_cap_slot_write_config(PCIDevice *dev,
314 uint32_t addr, uint32_t val, int len)
315 {
316 uint32_t pos = dev->exp.exp_cap;
317 uint8_t *exp_cap = dev->config + pos;
318 uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
319
320 if (!ranges_overlap(addr, len, pos + PCI_EXP_SLTCTL, 2)) {
321 return;
322 }
323
324 if (pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
325 PCI_EXP_SLTCTL_EIC)) {
326 sltsta ^= PCI_EXP_SLTSTA_EIS; /* toggle PCI_EXP_SLTSTA_EIS bit */
327 pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
328 PCIE_DEV_PRINTF(dev, "PCI_EXP_SLTCTL_EIC: "
329 "sltsta -> 0x%02"PRIx16"\n",
330 sltsta);
331 }
332
333 hotplug_event_notify(dev);
334
335 /*
336 * 6.7.3.2 Command Completed Events
337 *
338 * Software issues a command to a hot-plug capable Downstream Port by
339 * issuing a write transaction that targets any portion of the Port’s Slot
340 * Control register. A single write to the Slot Control register is
341 * considered to be a single command, even if the write affects more than
342 * one field in the Slot Control register. In response to this transaction,
343 * the Port must carry out the requested actions and then set the
344 * associated status field for the command completed event. */
345
346 /* Real hardware might take a while to complete requested command because
347 * physical movement would be involved like locking the electromechanical
348 * lock. However in our case, command is completed instantaneously above,
349 * so send a command completion event right now.
350 */
351 pcie_cap_slot_event(dev, PCI_EXP_HP_EV_CCI);
352 }
353
354 int pcie_cap_slot_post_load(void *opaque, int version_id)
355 {
356 PCIDevice *dev = opaque;
357 hotplug_event_update_event_status(dev);
358 return 0;
359 }
360
361 void pcie_cap_slot_push_attention_button(PCIDevice *dev)
362 {
363 pcie_cap_slot_event(dev, PCI_EXP_HP_EV_ABP);
364 }
365
366 /* root control/capabilities/status. PME isn't emulated for now */
367 void pcie_cap_root_init(PCIDevice *dev)
368 {
369 pci_set_word(dev->wmask + dev->exp.exp_cap + PCI_EXP_RTCTL,
370 PCI_EXP_RTCTL_SECEE | PCI_EXP_RTCTL_SENFEE |
371 PCI_EXP_RTCTL_SEFEE);
372 }
373
374 void pcie_cap_root_reset(PCIDevice *dev)
375 {
376 pci_set_word(dev->config + dev->exp.exp_cap + PCI_EXP_RTCTL, 0);
377 }
378
379 /*
380 * TODO: implement FLR:
381 * Right now sets the bit which indicates FLR is supported.
382 */
383 /* function level reset(FLR) */
384 void pcie_cap_flr_init(PCIDevice *dev)
385 {
386 pci_long_test_and_set_mask(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCAP,
387 PCI_EXP_DEVCAP_FLR);
388
389 /* Although reading BCR_FLR returns always 0,
390 * the bit is made writable here in order to detect the 1b is written
391 * pcie_cap_flr_write_config() test-and-clear the bit, so
392 * this bit always returns 0 to the guest.
393 */
394 pci_word_test_and_set_mask(dev->wmask + dev->exp.exp_cap + PCI_EXP_DEVCTL,
395 PCI_EXP_DEVCTL_BCR_FLR);
396 }
397
398 void pcie_cap_flr_write_config(PCIDevice *dev,
399 uint32_t addr, uint32_t val, int len)
400 {
401 uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
402 if (pci_word_test_and_clear_mask(devctl, PCI_EXP_DEVCTL_BCR_FLR)) {
403 /* TODO: implement FLR */
404 }
405 }
406
407 /* Alternative Routing-ID Interpretation (ARI) */
408 /* ari forwarding support for down stream port */
409 void pcie_cap_ari_init(PCIDevice *dev)
410 {
411 uint32_t pos = dev->exp.exp_cap;
412 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP2,
413 PCI_EXP_DEVCAP2_ARI);
414 pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL2,
415 PCI_EXP_DEVCTL2_ARI);
416 }
417
418 void pcie_cap_ari_reset(PCIDevice *dev)
419 {
420 uint8_t *devctl2 = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2;
421 pci_long_test_and_clear_mask(devctl2, PCI_EXP_DEVCTL2_ARI);
422 }
423
424 bool pcie_cap_is_ari_enabled(const PCIDevice *dev)
425 {
426 if (!pci_is_express(dev)) {
427 return false;
428 }
429 if (!dev->exp.exp_cap) {
430 return false;
431 }
432
433 return pci_get_long(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2) &
434 PCI_EXP_DEVCTL2_ARI;
435 }
436
437 /**************************************************************************
438 * pci express extended capability allocation functions
439 * uint16_t ext_cap_id (16 bit)
440 * uint8_t cap_ver (4 bit)
441 * uint16_t cap_offset (12 bit)
442 * uint16_t ext_cap_size
443 */
444
445 static uint16_t pcie_find_capability_list(PCIDevice *dev, uint16_t cap_id,
446 uint16_t *prev_p)
447 {
448 uint16_t prev = 0;
449 uint16_t next;
450 uint32_t header = pci_get_long(dev->config + PCI_CONFIG_SPACE_SIZE);
451
452 if (!header) {
453 /* no extended capability */
454 next = 0;
455 goto out;
456 }
457 for (next = PCI_CONFIG_SPACE_SIZE; next;
458 prev = next, next = PCI_EXT_CAP_NEXT(header)) {
459
460 assert(next >= PCI_CONFIG_SPACE_SIZE);
461 assert(next <= PCIE_CONFIG_SPACE_SIZE - 8);
462
463 header = pci_get_long(dev->config + next);
464 if (PCI_EXT_CAP_ID(header) == cap_id) {
465 break;
466 }
467 }
468
469 out:
470 if (prev_p) {
471 *prev_p = prev;
472 }
473 return next;
474 }
475
476 uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id)
477 {
478 return pcie_find_capability_list(dev, cap_id, NULL);
479 }
480
481 static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next)
482 {
483 uint16_t header = pci_get_long(dev->config + pos);
484 assert(!(next & (PCI_EXT_CAP_ALIGN - 1)));
485 header = (header & ~PCI_EXT_CAP_NEXT_MASK) |
486 ((next << PCI_EXT_CAP_NEXT_SHIFT) & PCI_EXT_CAP_NEXT_MASK);
487 pci_set_long(dev->config + pos, header);
488 }
489
490 /*
491 * caller must supply valid (offset, size) * such that the range shouldn't
492 * overlap with other capability or other registers.
493 * This function doesn't check it.
494 */
495 void pcie_add_capability(PCIDevice *dev,
496 uint16_t cap_id, uint8_t cap_ver,
497 uint16_t offset, uint16_t size)
498 {
499 uint32_t header;
500 uint16_t next;
501
502 assert(offset >= PCI_CONFIG_SPACE_SIZE);
503 assert(offset < offset + size);
504 assert(offset + size < PCIE_CONFIG_SPACE_SIZE);
505 assert(size >= 8);
506 assert(pci_is_express(dev));
507
508 if (offset == PCI_CONFIG_SPACE_SIZE) {
509 header = pci_get_long(dev->config + offset);
510 next = PCI_EXT_CAP_NEXT(header);
511 } else {
512 uint16_t prev;
513
514 /* 0 is reserved cap id. use internally to find the last capability
515 in the linked list */
516 next = pcie_find_capability_list(dev, 0, &prev);
517
518 assert(prev >= PCI_CONFIG_SPACE_SIZE);
519 assert(next == 0);
520 pcie_ext_cap_set_next(dev, prev, offset);
521 }
522 pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, next));
523
524 /* Make capability read-only by default */
525 memset(dev->wmask + offset, 0, size);
526 memset(dev->w1cmask + offset, 0, size);
527 /* Check capability by default */
528 memset(dev->cmask + offset, 0xFF, size);
529 }
530
531 /**************************************************************************
532 * pci express extended capability helper functions
533 */
534
535 /* ARI */
536 void pcie_ari_init(PCIDevice *dev, uint16_t offset, uint16_t nextfn)
537 {
538 pcie_add_capability(dev, PCI_EXT_CAP_ID_ARI, PCI_ARI_VER,
539 offset, PCI_ARI_SIZEOF);
540 pci_set_long(dev->config + offset + PCI_ARI_CAP, PCI_ARI_CAP_NFN(nextfn));
541 }