Merge remote-tracking branch 'mdroth/qga-pull-6-21-12' into staging
[qemu-kvm.git] / hw / pcie.c
blob7c92f193e7d2cb9e152c01fdd22d04ab4a36d048
1 /*
2 * pcie.c
4 * Copyright (c) 2010 Isaku Yamahata <yamahata at valinux co jp>
5 * VA Linux Systems Japan K.K.
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.
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.
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/>.
21 #include "qemu-common.h"
22 #include "pci_bridge.h"
23 #include "pcie.h"
24 #include "msix.h"
25 #include "msi.h"
26 #include "pci_internals.h"
27 #include "pcie_regs.h"
28 #include "range.h"
30 //#define DEBUG_PCIE
31 #ifdef DEBUG_PCIE
32 # define PCIE_DPRINTF(fmt, ...) \
33 fprintf(stderr, "%s:%d " fmt, __func__, __LINE__, ## __VA_ARGS__)
34 #else
35 # define PCIE_DPRINTF(fmt, ...) do {} while (0)
36 #endif
37 #define PCIE_DEV_PRINTF(dev, fmt, ...) \
38 PCIE_DPRINTF("%s:%x "fmt, (dev)->name, (dev)->devfn, ## __VA_ARGS__)
41 /***************************************************************************
42 * pci express capability helper functions
44 int pcie_cap_init(PCIDevice *dev, uint8_t offset, uint8_t type, uint8_t port)
46 int pos;
47 uint8_t *exp_cap;
49 assert(pci_is_express(dev));
51 pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
52 PCI_EXP_VER2_SIZEOF);
53 if (pos < 0) {
54 return pos;
56 dev->exp.exp_cap = pos;
57 exp_cap = dev->config + pos;
59 /* capability register
60 interrupt message number defaults to 0 */
61 pci_set_word(exp_cap + PCI_EXP_FLAGS,
62 ((type << PCI_EXP_FLAGS_TYPE_SHIFT) & PCI_EXP_FLAGS_TYPE) |
63 PCI_EXP_FLAGS_VER2);
65 /* device capability register
66 * table 7-12:
67 * roll based error reporting bit must be set by all
68 * Functions conforming to the ECN, PCI Express Base
69 * Specification, Revision 1.1., or subsequent PCI Express Base
70 * Specification revisions.
72 pci_set_long(exp_cap + PCI_EXP_DEVCAP, PCI_EXP_DEVCAP_RBER);
74 pci_set_long(exp_cap + PCI_EXP_LNKCAP,
75 (port << PCI_EXP_LNKCAP_PN_SHIFT) |
76 PCI_EXP_LNKCAP_ASPMS_0S |
77 PCI_EXP_LNK_MLW_1 |
78 PCI_EXP_LNK_LS_25);
80 pci_set_word(exp_cap + PCI_EXP_LNKSTA,
81 PCI_EXP_LNK_MLW_1 | PCI_EXP_LNK_LS_25);
83 pci_set_long(exp_cap + PCI_EXP_DEVCAP2,
84 PCI_EXP_DEVCAP2_EFF | PCI_EXP_DEVCAP2_EETLPP);
86 pci_set_word(dev->wmask + pos, PCI_EXP_DEVCTL2_EETLPPB);
87 return pos;
90 void pcie_cap_exit(PCIDevice *dev)
92 pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER2_SIZEOF);
95 uint8_t pcie_cap_get_type(const PCIDevice *dev)
97 uint32_t pos = dev->exp.exp_cap;
98 assert(pos > 0);
99 return (pci_get_word(dev->config + pos + PCI_EXP_FLAGS) &
100 PCI_EXP_FLAGS_TYPE) >> PCI_EXP_FLAGS_TYPE_SHIFT;
103 /* MSI/MSI-X */
104 /* pci express interrupt message number */
105 /* 7.8.2 PCI Express Capabilities Register: Interrupt Message Number */
106 void pcie_cap_flags_set_vector(PCIDevice *dev, uint8_t vector)
108 uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
109 assert(vector < 32);
110 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_FLAGS, PCI_EXP_FLAGS_IRQ);
111 pci_word_test_and_set_mask(exp_cap + PCI_EXP_FLAGS,
112 vector << PCI_EXP_FLAGS_IRQ_SHIFT);
115 uint8_t pcie_cap_flags_get_vector(PCIDevice *dev)
117 return (pci_get_word(dev->config + dev->exp.exp_cap + PCI_EXP_FLAGS) &
118 PCI_EXP_FLAGS_IRQ) >> PCI_EXP_FLAGS_IRQ_SHIFT;
121 void pcie_cap_deverr_init(PCIDevice *dev)
123 uint32_t pos = dev->exp.exp_cap;
124 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP,
125 PCI_EXP_DEVCAP_RBER);
126 pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL,
127 PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
128 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
129 pci_long_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_DEVSTA,
130 PCI_EXP_DEVSTA_CED | PCI_EXP_DEVSTA_NFED |
131 PCI_EXP_DEVSTA_URD | PCI_EXP_DEVSTA_URD);
134 void pcie_cap_deverr_reset(PCIDevice *dev)
136 uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
137 pci_long_test_and_clear_mask(devctl,
138 PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
139 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
142 static void hotplug_event_update_event_status(PCIDevice *dev)
144 uint32_t pos = dev->exp.exp_cap;
145 uint8_t *exp_cap = dev->config + pos;
146 uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
147 uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
149 dev->exp.hpev_notified = (sltctl & PCI_EXP_SLTCTL_HPIE) &&
150 (sltsta & sltctl & PCI_EXP_HP_EV_SUPPORTED);
153 static void hotplug_event_notify(PCIDevice *dev)
155 bool prev = dev->exp.hpev_notified;
157 hotplug_event_update_event_status(dev);
159 if (prev == dev->exp.hpev_notified) {
160 return;
163 /* Note: the logic above does not take into account whether interrupts
164 * are masked. The result is that interrupt will be sent when it is
165 * subsequently unmasked. This appears to be legal: Section 6.7.3.4:
166 * The Port may optionally send an MSI when there are hot-plug events that
167 * occur while interrupt generation is disabled, and interrupt generation is
168 * subsequently enabled. */
169 if (msix_enabled(dev)) {
170 msix_notify(dev, pcie_cap_flags_get_vector(dev));
171 } else if (msi_enabled(dev)) {
172 msi_notify(dev, pcie_cap_flags_get_vector(dev));
173 } else {
174 qemu_set_irq(dev->irq[dev->exp.hpev_intx], dev->exp.hpev_notified);
178 static void hotplug_event_clear(PCIDevice *dev)
180 hotplug_event_update_event_status(dev);
181 if (!msix_enabled(dev) && !msi_enabled(dev) && !dev->exp.hpev_notified) {
182 qemu_set_irq(dev->irq[dev->exp.hpev_intx], 0);
187 * A PCI Express Hot-Plug Event has occurred, so update slot status register
188 * and notify OS of the event if necessary.
190 * 6.7.3 PCI Express Hot-Plug Events
191 * 6.7.3.4 Software Notification of Hot-Plug Events
193 static void pcie_cap_slot_event(PCIDevice *dev, PCIExpressHotPlugEvent event)
195 /* Minor optimization: if nothing changed - no event is needed. */
196 if (pci_word_test_and_set_mask(dev->config + dev->exp.exp_cap +
197 PCI_EXP_SLTSTA, event)) {
198 return;
200 hotplug_event_notify(dev);
203 static int pcie_cap_slot_hotplug(DeviceState *qdev,
204 PCIDevice *pci_dev, PCIHotplugState state)
206 PCIDevice *d = PCI_DEVICE(qdev);
207 uint8_t *exp_cap = d->config + d->exp.exp_cap;
208 uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
210 /* Don't send event when device is enabled during qemu machine creation:
211 * it is present on boot, no hotplug event is necessary. We do send an
212 * event when the device is disabled later. */
213 if (state == PCI_COLDPLUG_ENABLED) {
214 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
215 PCI_EXP_SLTSTA_PDS);
216 return 0;
219 PCIE_DEV_PRINTF(pci_dev, "hotplug state: %d\n", state);
220 if (sltsta & PCI_EXP_SLTSTA_EIS) {
221 /* the slot is electromechanically locked.
222 * This error is propagated up to qdev and then to HMP/QMP.
224 return -EBUSY;
227 /* TODO: multifunction hot-plug.
228 * Right now, only a device of function = 0 is allowed to be
229 * hot plugged/unplugged.
231 assert(PCI_FUNC(pci_dev->devfn) == 0);
233 if (state == PCI_HOTPLUG_ENABLED) {
234 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
235 PCI_EXP_SLTSTA_PDS);
236 pcie_cap_slot_event(d, PCI_EXP_HP_EV_PDC);
237 } else {
238 qdev_free(&pci_dev->qdev);
239 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
240 PCI_EXP_SLTSTA_PDS);
241 pcie_cap_slot_event(d, PCI_EXP_HP_EV_PDC);
243 return 0;
246 /* pci express slot for pci express root/downstream port
247 PCI express capability slot registers */
248 void pcie_cap_slot_init(PCIDevice *dev, uint16_t slot)
250 uint32_t pos = dev->exp.exp_cap;
252 pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_FLAGS,
253 PCI_EXP_FLAGS_SLOT);
255 pci_long_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCAP,
256 ~PCI_EXP_SLTCAP_PSN);
257 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
258 (slot << PCI_EXP_SLTCAP_PSN_SHIFT) |
259 PCI_EXP_SLTCAP_EIP |
260 PCI_EXP_SLTCAP_HPS |
261 PCI_EXP_SLTCAP_HPC |
262 PCI_EXP_SLTCAP_PIP |
263 PCI_EXP_SLTCAP_AIP |
264 PCI_EXP_SLTCAP_ABP);
266 pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
267 PCI_EXP_SLTCTL_PIC |
268 PCI_EXP_SLTCTL_AIC);
269 pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCTL,
270 PCI_EXP_SLTCTL_PIC_OFF |
271 PCI_EXP_SLTCTL_AIC_OFF);
272 pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
273 PCI_EXP_SLTCTL_PIC |
274 PCI_EXP_SLTCTL_AIC |
275 PCI_EXP_SLTCTL_HPIE |
276 PCI_EXP_SLTCTL_CCIE |
277 PCI_EXP_SLTCTL_PDCE |
278 PCI_EXP_SLTCTL_ABPE);
279 /* Although reading PCI_EXP_SLTCTL_EIC returns always 0,
280 * make the bit writable here in order to detect 1b is written.
281 * pcie_cap_slot_write_config() test-and-clear the bit, so
282 * this bit always returns 0 to the guest.
284 pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
285 PCI_EXP_SLTCTL_EIC);
287 pci_word_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_SLTSTA,
288 PCI_EXP_HP_EV_SUPPORTED);
290 dev->exp.hpev_notified = false;
292 pci_bus_hotplug(pci_bridge_get_sec_bus(DO_UPCAST(PCIBridge, dev, dev)),
293 pcie_cap_slot_hotplug, &dev->qdev);
296 void pcie_cap_slot_reset(PCIDevice *dev)
298 uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
300 PCIE_DEV_PRINTF(dev, "reset\n");
302 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
303 PCI_EXP_SLTCTL_EIC |
304 PCI_EXP_SLTCTL_PIC |
305 PCI_EXP_SLTCTL_AIC |
306 PCI_EXP_SLTCTL_HPIE |
307 PCI_EXP_SLTCTL_CCIE |
308 PCI_EXP_SLTCTL_PDCE |
309 PCI_EXP_SLTCTL_ABPE);
310 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
311 PCI_EXP_SLTCTL_PIC_OFF |
312 PCI_EXP_SLTCTL_AIC_OFF);
314 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
315 PCI_EXP_SLTSTA_EIS |/* on reset,
316 the lock is released */
317 PCI_EXP_SLTSTA_CC |
318 PCI_EXP_SLTSTA_PDC |
319 PCI_EXP_SLTSTA_ABP);
321 hotplug_event_update_event_status(dev);
324 void pcie_cap_slot_write_config(PCIDevice *dev,
325 uint32_t addr, uint32_t val, int len)
327 uint32_t pos = dev->exp.exp_cap;
328 uint8_t *exp_cap = dev->config + pos;
329 uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
331 if (ranges_overlap(addr, len, pos + PCI_EXP_SLTSTA, 2)) {
332 hotplug_event_clear(dev);
335 if (!ranges_overlap(addr, len, pos + PCI_EXP_SLTCTL, 2)) {
336 return;
339 if (pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
340 PCI_EXP_SLTCTL_EIC)) {
341 sltsta ^= PCI_EXP_SLTSTA_EIS; /* toggle PCI_EXP_SLTSTA_EIS bit */
342 pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
343 PCIE_DEV_PRINTF(dev, "PCI_EXP_SLTCTL_EIC: "
344 "sltsta -> 0x%02"PRIx16"\n",
345 sltsta);
348 hotplug_event_notify(dev);
351 * 6.7.3.2 Command Completed Events
353 * Software issues a command to a hot-plug capable Downstream Port by
354 * issuing a write transaction that targets any portion of the Port’s Slot
355 * Control register. A single write to the Slot Control register is
356 * considered to be a single command, even if the write affects more than
357 * one field in the Slot Control register. In response to this transaction,
358 * the Port must carry out the requested actions and then set the
359 * associated status field for the command completed event. */
361 /* Real hardware might take a while to complete requested command because
362 * physical movement would be involved like locking the electromechanical
363 * lock. However in our case, command is completed instantaneously above,
364 * so send a command completion event right now.
366 pcie_cap_slot_event(dev, PCI_EXP_HP_EV_CCI);
369 int pcie_cap_slot_post_load(void *opaque, int version_id)
371 PCIDevice *dev = opaque;
372 hotplug_event_update_event_status(dev);
373 return 0;
376 void pcie_cap_slot_push_attention_button(PCIDevice *dev)
378 pcie_cap_slot_event(dev, PCI_EXP_HP_EV_ABP);
381 /* root control/capabilities/status. PME isn't emulated for now */
382 void pcie_cap_root_init(PCIDevice *dev)
384 pci_set_word(dev->wmask + dev->exp.exp_cap + PCI_EXP_RTCTL,
385 PCI_EXP_RTCTL_SECEE | PCI_EXP_RTCTL_SENFEE |
386 PCI_EXP_RTCTL_SEFEE);
389 void pcie_cap_root_reset(PCIDevice *dev)
391 pci_set_word(dev->config + dev->exp.exp_cap + PCI_EXP_RTCTL, 0);
394 /* function level reset(FLR) */
395 void pcie_cap_flr_init(PCIDevice *dev)
397 pci_long_test_and_set_mask(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCAP,
398 PCI_EXP_DEVCAP_FLR);
400 /* Although reading BCR_FLR returns always 0,
401 * the bit is made writable here in order to detect the 1b is written
402 * pcie_cap_flr_write_config() test-and-clear the bit, so
403 * this bit always returns 0 to the guest.
405 pci_word_test_and_set_mask(dev->wmask + dev->exp.exp_cap + PCI_EXP_DEVCTL,
406 PCI_EXP_DEVCTL_BCR_FLR);
409 void pcie_cap_flr_write_config(PCIDevice *dev,
410 uint32_t addr, uint32_t val, int len)
412 uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
413 if (pci_get_word(devctl) & PCI_EXP_DEVCTL_BCR_FLR) {
414 /* Clear PCI_EXP_DEVCTL_BCR_FLR after invoking the reset handler
415 so the handler can detect FLR by looking at this bit. */
416 pci_device_reset(dev);
417 pci_word_test_and_clear_mask(devctl, PCI_EXP_DEVCTL_BCR_FLR);
421 /* Alternative Routing-ID Interpretation (ARI) */
422 /* ari forwarding support for down stream port */
423 void pcie_cap_ari_init(PCIDevice *dev)
425 uint32_t pos = dev->exp.exp_cap;
426 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP2,
427 PCI_EXP_DEVCAP2_ARI);
428 pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL2,
429 PCI_EXP_DEVCTL2_ARI);
432 void pcie_cap_ari_reset(PCIDevice *dev)
434 uint8_t *devctl2 = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2;
435 pci_long_test_and_clear_mask(devctl2, PCI_EXP_DEVCTL2_ARI);
438 bool pcie_cap_is_ari_enabled(const PCIDevice *dev)
440 if (!pci_is_express(dev)) {
441 return false;
443 if (!dev->exp.exp_cap) {
444 return false;
447 return pci_get_long(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2) &
448 PCI_EXP_DEVCTL2_ARI;
451 /**************************************************************************
452 * pci express extended capability allocation functions
453 * uint16_t ext_cap_id (16 bit)
454 * uint8_t cap_ver (4 bit)
455 * uint16_t cap_offset (12 bit)
456 * uint16_t ext_cap_size
459 static uint16_t pcie_find_capability_list(PCIDevice *dev, uint16_t cap_id,
460 uint16_t *prev_p)
462 uint16_t prev = 0;
463 uint16_t next;
464 uint32_t header = pci_get_long(dev->config + PCI_CONFIG_SPACE_SIZE);
466 if (!header) {
467 /* no extended capability */
468 next = 0;
469 goto out;
471 for (next = PCI_CONFIG_SPACE_SIZE; next;
472 prev = next, next = PCI_EXT_CAP_NEXT(header)) {
474 assert(next >= PCI_CONFIG_SPACE_SIZE);
475 assert(next <= PCIE_CONFIG_SPACE_SIZE - 8);
477 header = pci_get_long(dev->config + next);
478 if (PCI_EXT_CAP_ID(header) == cap_id) {
479 break;
483 out:
484 if (prev_p) {
485 *prev_p = prev;
487 return next;
490 uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id)
492 return pcie_find_capability_list(dev, cap_id, NULL);
495 static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next)
497 uint16_t header = pci_get_long(dev->config + pos);
498 assert(!(next & (PCI_EXT_CAP_ALIGN - 1)));
499 header = (header & ~PCI_EXT_CAP_NEXT_MASK) |
500 ((next << PCI_EXT_CAP_NEXT_SHIFT) & PCI_EXT_CAP_NEXT_MASK);
501 pci_set_long(dev->config + pos, header);
505 * caller must supply valid (offset, size) * such that the range shouldn't
506 * overlap with other capability or other registers.
507 * This function doesn't check it.
509 void pcie_add_capability(PCIDevice *dev,
510 uint16_t cap_id, uint8_t cap_ver,
511 uint16_t offset, uint16_t size)
513 uint32_t header;
514 uint16_t next;
516 assert(offset >= PCI_CONFIG_SPACE_SIZE);
517 assert(offset < offset + size);
518 assert(offset + size < PCIE_CONFIG_SPACE_SIZE);
519 assert(size >= 8);
520 assert(pci_is_express(dev));
522 if (offset == PCI_CONFIG_SPACE_SIZE) {
523 header = pci_get_long(dev->config + offset);
524 next = PCI_EXT_CAP_NEXT(header);
525 } else {
526 uint16_t prev;
528 /* 0 is reserved cap id. use internally to find the last capability
529 in the linked list */
530 next = pcie_find_capability_list(dev, 0, &prev);
532 assert(prev >= PCI_CONFIG_SPACE_SIZE);
533 assert(next == 0);
534 pcie_ext_cap_set_next(dev, prev, offset);
536 pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, next));
538 /* Make capability read-only by default */
539 memset(dev->wmask + offset, 0, size);
540 memset(dev->w1cmask + offset, 0, size);
541 /* Check capability by default */
542 memset(dev->cmask + offset, 0xFF, size);
545 /**************************************************************************
546 * pci express extended capability helper functions
549 /* ARI */
550 void pcie_ari_init(PCIDevice *dev, uint16_t offset, uint16_t nextfn)
552 pcie_add_capability(dev, PCI_EXT_CAP_ID_ARI, PCI_ARI_VER,
553 offset, PCI_ARI_SIZEOF);
554 pci_set_long(dev->config + offset + PCI_ARI_CAP, PCI_ARI_CAP_NFN(nextfn));