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 "hw/pci/pci_bridge.h"
23 #include "hw/pci/pcie.h"
24 #include "hw/pci/msix.h"
25 #include "hw/pci/msi.h"
26 #include "hw/pci/pci_bus.h"
27 #include "hw/pci/pcie_regs.h"
28 #include "qemu/range.h"
29 #include "qapi/qmp/qerror.h"
33 # define PCIE_DPRINTF(fmt, ...) \
34 fprintf(stderr, "%s:%d " fmt, __func__, __LINE__, ## __VA_ARGS__)
36 # define PCIE_DPRINTF(fmt, ...) do {} while (0)
38 #define PCIE_DEV_PRINTF(dev, fmt, ...) \
39 PCIE_DPRINTF("%s:%x "fmt, (dev)->name, (dev)->devfn, ## __VA_ARGS__)
42 /***************************************************************************
43 * pci express capability helper functions
45 int pcie_cap_init(PCIDevice *dev, uint8_t offset, uint8_t type, uint8_t port)
50 assert(pci_is_express(dev));
52 pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
57 dev->exp.exp_cap = pos;
58 exp_cap = dev->config + pos;
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) |
66 /* device capability register
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.
73 pci_set_long(exp_cap + PCI_EXP_DEVCAP, PCI_EXP_DEVCAP_RBER);
75 pci_set_long(exp_cap + PCI_EXP_LNKCAP,
76 (port << PCI_EXP_LNKCAP_PN_SHIFT) |
77 PCI_EXP_LNKCAP_ASPMS_0S |
81 pci_set_word(exp_cap + PCI_EXP_LNKSTA,
82 PCI_EXP_LNK_MLW_1 | PCI_EXP_LNK_LS_25);
84 pci_set_long(exp_cap + PCI_EXP_DEVCAP2,
85 PCI_EXP_DEVCAP2_EFF | PCI_EXP_DEVCAP2_EETLPP);
87 pci_set_word(dev->wmask + pos, PCI_EXP_DEVCTL2_EETLPPB);
91 int pcie_endpoint_cap_init(PCIDevice *dev, uint8_t offset)
93 uint8_t type = PCI_EXP_TYPE_ENDPOINT;
96 * Windows guests will report Code 10, device cannot start, if
97 * a regular Endpoint type is exposed on a root complex. These
98 * should instead be Root Complex Integrated Endpoints.
100 if (pci_bus_is_express(dev->bus) && pci_bus_is_root(dev->bus)) {
101 type = PCI_EXP_TYPE_RC_END;
104 return pcie_cap_init(dev, offset, type, 0);
107 void pcie_cap_exit(PCIDevice *dev)
109 pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER2_SIZEOF);
112 uint8_t pcie_cap_get_type(const PCIDevice *dev)
114 uint32_t pos = dev->exp.exp_cap;
116 return (pci_get_word(dev->config + pos + PCI_EXP_FLAGS) &
117 PCI_EXP_FLAGS_TYPE) >> PCI_EXP_FLAGS_TYPE_SHIFT;
121 /* pci express interrupt message number */
122 /* 7.8.2 PCI Express Capabilities Register: Interrupt Message Number */
123 void pcie_cap_flags_set_vector(PCIDevice *dev, uint8_t vector)
125 uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
127 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_FLAGS, PCI_EXP_FLAGS_IRQ);
128 pci_word_test_and_set_mask(exp_cap + PCI_EXP_FLAGS,
129 vector << PCI_EXP_FLAGS_IRQ_SHIFT);
132 uint8_t pcie_cap_flags_get_vector(PCIDevice *dev)
134 return (pci_get_word(dev->config + dev->exp.exp_cap + PCI_EXP_FLAGS) &
135 PCI_EXP_FLAGS_IRQ) >> PCI_EXP_FLAGS_IRQ_SHIFT;
138 void pcie_cap_deverr_init(PCIDevice *dev)
140 uint32_t pos = dev->exp.exp_cap;
141 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP,
142 PCI_EXP_DEVCAP_RBER);
143 pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL,
144 PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
145 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
146 pci_long_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_DEVSTA,
147 PCI_EXP_DEVSTA_CED | PCI_EXP_DEVSTA_NFED |
148 PCI_EXP_DEVSTA_URD | PCI_EXP_DEVSTA_URD);
151 void pcie_cap_deverr_reset(PCIDevice *dev)
153 uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
154 pci_long_test_and_clear_mask(devctl,
155 PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
156 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
159 static void hotplug_event_update_event_status(PCIDevice *dev)
161 uint32_t pos = dev->exp.exp_cap;
162 uint8_t *exp_cap = dev->config + pos;
163 uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
164 uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
166 dev->exp.hpev_notified = (sltctl & PCI_EXP_SLTCTL_HPIE) &&
167 (sltsta & sltctl & PCI_EXP_HP_EV_SUPPORTED);
170 static void hotplug_event_notify(PCIDevice *dev)
172 bool prev = dev->exp.hpev_notified;
174 hotplug_event_update_event_status(dev);
176 if (prev == dev->exp.hpev_notified) {
180 /* Note: the logic above does not take into account whether interrupts
181 * are masked. The result is that interrupt will be sent when it is
182 * subsequently unmasked. This appears to be legal: Section 6.7.3.4:
183 * The Port may optionally send an MSI when there are hot-plug events that
184 * occur while interrupt generation is disabled, and interrupt generation is
185 * subsequently enabled. */
186 if (msix_enabled(dev)) {
187 msix_notify(dev, pcie_cap_flags_get_vector(dev));
188 } else if (msi_enabled(dev)) {
189 msi_notify(dev, pcie_cap_flags_get_vector(dev));
191 pci_set_irq(dev, dev->exp.hpev_notified);
195 static void hotplug_event_clear(PCIDevice *dev)
197 hotplug_event_update_event_status(dev);
198 if (!msix_enabled(dev) && !msi_enabled(dev) && !dev->exp.hpev_notified) {
199 pci_irq_deassert(dev);
204 * A PCI Express Hot-Plug Event has occurred, so update slot status register
205 * and notify OS of the event if necessary.
207 * 6.7.3 PCI Express Hot-Plug Events
208 * 6.7.3.4 Software Notification of Hot-Plug Events
210 static void pcie_cap_slot_event(PCIDevice *dev, PCIExpressHotPlugEvent event)
212 /* Minor optimization: if nothing changed - no event is needed. */
213 if (pci_word_test_and_set_mask(dev->config + dev->exp.exp_cap +
214 PCI_EXP_SLTSTA, event)) {
217 hotplug_event_notify(dev);
220 static void pcie_cap_slot_hotplug_common(PCIDevice *hotplug_dev,
222 uint8_t **exp_cap, Error **errp)
224 *exp_cap = hotplug_dev->config + hotplug_dev->exp.exp_cap;
225 uint16_t sltsta = pci_get_word(*exp_cap + PCI_EXP_SLTSTA);
227 PCIE_DEV_PRINTF(PCI_DEVICE(dev), "hotplug state: 0x%x\n", sltsta);
228 if (sltsta & PCI_EXP_SLTSTA_EIS) {
229 /* the slot is electromechanically locked.
230 * This error is propagated up to qdev and then to HMP/QMP.
232 error_setg_errno(errp, -EBUSY, "slot is electromechanically locked");
236 void pcie_cap_slot_hotplug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
240 PCIDevice *pci_dev = PCI_DEVICE(dev);
242 pcie_cap_slot_hotplug_common(PCI_DEVICE(hotplug_dev), dev, &exp_cap, errp);
244 /* Don't send event when device is enabled during qemu machine creation:
245 * it is present on boot, no hotplug event is necessary. We do send an
246 * event when the device is disabled later. */
247 if (!dev->hotplugged) {
248 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
253 /* TODO: multifunction hot-plug.
254 * Right now, only a device of function = 0 is allowed to be
255 * hot plugged/unplugged.
257 assert(PCI_FUNC(pci_dev->devfn) == 0);
259 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
261 pcie_cap_slot_event(PCI_DEVICE(hotplug_dev),
262 PCI_EXP_HP_EV_PDC | PCI_EXP_HP_EV_ABP);
265 void pcie_cap_slot_hot_unplug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
270 pcie_cap_slot_hotplug_common(PCI_DEVICE(hotplug_dev), dev, &exp_cap, errp);
272 pcie_cap_slot_push_attention_button(PCI_DEVICE(hotplug_dev));
275 /* pci express slot for pci express root/downstream port
276 PCI express capability slot registers */
277 void pcie_cap_slot_init(PCIDevice *dev, uint16_t slot)
279 uint32_t pos = dev->exp.exp_cap;
281 pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_FLAGS,
284 pci_long_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCAP,
285 ~PCI_EXP_SLTCAP_PSN);
286 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
287 (slot << PCI_EXP_SLTCAP_PSN_SHIFT) |
295 if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
296 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
298 pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
300 pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
304 pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
307 pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCTL,
308 PCI_EXP_SLTCTL_PIC_OFF |
309 PCI_EXP_SLTCTL_AIC_OFF);
310 pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
313 PCI_EXP_SLTCTL_HPIE |
314 PCI_EXP_SLTCTL_CCIE |
315 PCI_EXP_SLTCTL_PDCE |
316 PCI_EXP_SLTCTL_ABPE);
317 /* Although reading PCI_EXP_SLTCTL_EIC returns always 0,
318 * make the bit writable here in order to detect 1b is written.
319 * pcie_cap_slot_write_config() test-and-clear the bit, so
320 * this bit always returns 0 to the guest.
322 pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
325 pci_word_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_SLTSTA,
326 PCI_EXP_HP_EV_SUPPORTED);
328 dev->exp.hpev_notified = false;
330 qbus_set_hotplug_handler(BUS(pci_bridge_get_sec_bus(PCI_BRIDGE(dev))),
334 void pcie_cap_slot_reset(PCIDevice *dev)
336 uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
337 uint8_t port_type = pcie_cap_get_type(dev);
339 assert(port_type == PCI_EXP_TYPE_DOWNSTREAM ||
340 port_type == PCI_EXP_TYPE_ROOT_PORT);
342 PCIE_DEV_PRINTF(dev, "reset\n");
344 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
348 PCI_EXP_SLTCTL_HPIE |
349 PCI_EXP_SLTCTL_CCIE |
350 PCI_EXP_SLTCTL_PDCE |
351 PCI_EXP_SLTCTL_ABPE);
352 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
353 PCI_EXP_SLTCTL_AIC_OFF);
355 if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
359 /* Downstream ports enforce device number 0. */
360 populated = (pci_bridge_get_sec_bus(PCI_BRIDGE(dev))->devices[0] != NULL);
363 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
366 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
370 pic = populated ? PCI_EXP_SLTCTL_PIC_ON : PCI_EXP_SLTCTL_PIC_OFF;
371 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL, pic);
374 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
375 PCI_EXP_SLTSTA_EIS |/* on reset,
376 the lock is released */
381 hotplug_event_update_event_status(dev);
384 static void pcie_unplug_device(PCIBus *bus, PCIDevice *dev, void *opaque)
386 object_unparent(OBJECT(dev));
389 void pcie_cap_slot_write_config(PCIDevice *dev,
390 uint32_t addr, uint32_t val, int len)
392 uint32_t pos = dev->exp.exp_cap;
393 uint8_t *exp_cap = dev->config + pos;
394 uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
396 if (ranges_overlap(addr, len, pos + PCI_EXP_SLTSTA, 2)) {
397 hotplug_event_clear(dev);
400 if (!ranges_overlap(addr, len, pos + PCI_EXP_SLTCTL, 2)) {
404 if (pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
405 PCI_EXP_SLTCTL_EIC)) {
406 sltsta ^= PCI_EXP_SLTSTA_EIS; /* toggle PCI_EXP_SLTSTA_EIS bit */
407 pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
408 PCIE_DEV_PRINTF(dev, "PCI_EXP_SLTCTL_EIC: "
409 "sltsta -> 0x%02"PRIx16"\n",
414 * If the slot is polulated, power indicator is off and power
415 * controller is off, it is safe to detach the devices.
417 if ((sltsta & PCI_EXP_SLTSTA_PDS) && (val & PCI_EXP_SLTCTL_PCC) &&
418 ((val & PCI_EXP_SLTCTL_PIC_OFF) == PCI_EXP_SLTCTL_PIC_OFF)) {
419 PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
420 pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
421 pcie_unplug_device, NULL);
423 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
425 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
429 hotplug_event_notify(dev);
432 * 6.7.3.2 Command Completed Events
434 * Software issues a command to a hot-plug capable Downstream Port by
435 * issuing a write transaction that targets any portion of the Port’s Slot
436 * Control register. A single write to the Slot Control register is
437 * considered to be a single command, even if the write affects more than
438 * one field in the Slot Control register. In response to this transaction,
439 * the Port must carry out the requested actions and then set the
440 * associated status field for the command completed event. */
442 /* Real hardware might take a while to complete requested command because
443 * physical movement would be involved like locking the electromechanical
444 * lock. However in our case, command is completed instantaneously above,
445 * so send a command completion event right now.
447 pcie_cap_slot_event(dev, PCI_EXP_HP_EV_CCI);
450 int pcie_cap_slot_post_load(void *opaque, int version_id)
452 PCIDevice *dev = opaque;
453 hotplug_event_update_event_status(dev);
457 void pcie_cap_slot_push_attention_button(PCIDevice *dev)
459 pcie_cap_slot_event(dev, PCI_EXP_HP_EV_ABP);
462 /* root control/capabilities/status. PME isn't emulated for now */
463 void pcie_cap_root_init(PCIDevice *dev)
465 pci_set_word(dev->wmask + dev->exp.exp_cap + PCI_EXP_RTCTL,
466 PCI_EXP_RTCTL_SECEE | PCI_EXP_RTCTL_SENFEE |
467 PCI_EXP_RTCTL_SEFEE);
470 void pcie_cap_root_reset(PCIDevice *dev)
472 pci_set_word(dev->config + dev->exp.exp_cap + PCI_EXP_RTCTL, 0);
475 /* function level reset(FLR) */
476 void pcie_cap_flr_init(PCIDevice *dev)
478 pci_long_test_and_set_mask(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCAP,
481 /* Although reading BCR_FLR returns always 0,
482 * the bit is made writable here in order to detect the 1b is written
483 * pcie_cap_flr_write_config() test-and-clear the bit, so
484 * this bit always returns 0 to the guest.
486 pci_word_test_and_set_mask(dev->wmask + dev->exp.exp_cap + PCI_EXP_DEVCTL,
487 PCI_EXP_DEVCTL_BCR_FLR);
490 void pcie_cap_flr_write_config(PCIDevice *dev,
491 uint32_t addr, uint32_t val, int len)
493 uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
494 if (pci_get_word(devctl) & PCI_EXP_DEVCTL_BCR_FLR) {
495 /* Clear PCI_EXP_DEVCTL_BCR_FLR after invoking the reset handler
496 so the handler can detect FLR by looking at this bit. */
497 pci_device_reset(dev);
498 pci_word_test_and_clear_mask(devctl, PCI_EXP_DEVCTL_BCR_FLR);
502 /* Alternative Routing-ID Interpretation (ARI) */
503 /* ari forwarding support for down stream port */
504 void pcie_cap_ari_init(PCIDevice *dev)
506 uint32_t pos = dev->exp.exp_cap;
507 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP2,
508 PCI_EXP_DEVCAP2_ARI);
509 pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL2,
510 PCI_EXP_DEVCTL2_ARI);
513 void pcie_cap_ari_reset(PCIDevice *dev)
515 uint8_t *devctl2 = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2;
516 pci_long_test_and_clear_mask(devctl2, PCI_EXP_DEVCTL2_ARI);
519 bool pcie_cap_is_ari_enabled(const PCIDevice *dev)
521 if (!pci_is_express(dev)) {
524 if (!dev->exp.exp_cap) {
528 return pci_get_long(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2) &
532 /**************************************************************************
533 * pci express extended capability allocation functions
534 * uint16_t ext_cap_id (16 bit)
535 * uint8_t cap_ver (4 bit)
536 * uint16_t cap_offset (12 bit)
537 * uint16_t ext_cap_size
540 static uint16_t pcie_find_capability_list(PCIDevice *dev, uint16_t cap_id,
545 uint32_t header = pci_get_long(dev->config + PCI_CONFIG_SPACE_SIZE);
548 /* no extended capability */
552 for (next = PCI_CONFIG_SPACE_SIZE; next;
553 prev = next, next = PCI_EXT_CAP_NEXT(header)) {
555 assert(next >= PCI_CONFIG_SPACE_SIZE);
556 assert(next <= PCIE_CONFIG_SPACE_SIZE - 8);
558 header = pci_get_long(dev->config + next);
559 if (PCI_EXT_CAP_ID(header) == cap_id) {
571 uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id)
573 return pcie_find_capability_list(dev, cap_id, NULL);
576 static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next)
578 uint32_t header = pci_get_long(dev->config + pos);
579 assert(!(next & (PCI_EXT_CAP_ALIGN - 1)));
580 header = (header & ~PCI_EXT_CAP_NEXT_MASK) |
581 ((next << PCI_EXT_CAP_NEXT_SHIFT) & PCI_EXT_CAP_NEXT_MASK);
582 pci_set_long(dev->config + pos, header);
586 * caller must supply valid (offset, size) * such that the range shouldn't
587 * overlap with other capability or other registers.
588 * This function doesn't check it.
590 void pcie_add_capability(PCIDevice *dev,
591 uint16_t cap_id, uint8_t cap_ver,
592 uint16_t offset, uint16_t size)
597 assert(offset >= PCI_CONFIG_SPACE_SIZE);
598 assert(offset < offset + size);
599 assert(offset + size < PCIE_CONFIG_SPACE_SIZE);
601 assert(pci_is_express(dev));
603 if (offset == PCI_CONFIG_SPACE_SIZE) {
604 header = pci_get_long(dev->config + offset);
605 next = PCI_EXT_CAP_NEXT(header);
609 /* 0 is reserved cap id. use internally to find the last capability
610 in the linked list */
611 next = pcie_find_capability_list(dev, 0, &prev);
613 assert(prev >= PCI_CONFIG_SPACE_SIZE);
615 pcie_ext_cap_set_next(dev, prev, offset);
617 pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, next));
619 /* Make capability read-only by default */
620 memset(dev->wmask + offset, 0, size);
621 memset(dev->w1cmask + offset, 0, size);
622 /* Check capability by default */
623 memset(dev->cmask + offset, 0xFF, size);
626 /**************************************************************************
627 * pci express extended capability helper functions
631 void pcie_ari_init(PCIDevice *dev, uint16_t offset, uint16_t nextfn)
633 pcie_add_capability(dev, PCI_EXT_CAP_ID_ARI, PCI_ARI_VER,
634 offset, PCI_ARI_SIZEOF);
635 pci_set_long(dev->config + offset + PCI_ARI_CAP, PCI_ARI_CAP_NFN(nextfn));