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/osdep.h"
22
#include "qapi/error.h"
23
#include "hw/pci/pci_bridge.h"
24
#include "hw/pci/pcie.h"
25
#include "hw/pci/msix.h"
26
#include "hw/pci/msi.h"
27
#include "hw/pci/pci_bus.h"
28
#include "hw/pci/pcie_regs.h"
29
#include "hw/pci/pcie_port.h"
30
#include "qemu/range.h"
35
# define PCIE_DPRINTF(fmt, ...) \
36
fprintf(stderr, "%s:%d " fmt, __func__, __LINE__, ## __VA_ARGS__)
38
# define PCIE_DPRINTF(fmt, ...) do {} while (0)
40
#define PCIE_DEV_PRINTF(dev, fmt, ...) \
41
PCIE_DPRINTF("%s:%x "fmt, (dev)->name, (dev)->devfn, ## __VA_ARGS__)
43
static bool pcie_sltctl_powered_off(uint16_t sltctl)
45
return (sltctl & PCI_EXP_SLTCTL_PCC) == PCI_EXP_SLTCTL_PWR_OFF
46
&& (sltctl & PCI_EXP_SLTCTL_PIC) == PCI_EXP_SLTCTL_PWR_IND_OFF;
49
static const char *pcie_led_state_to_str(uint16_t value)
52
case PCI_EXP_SLTCTL_PWR_IND_ON:
53
case PCI_EXP_SLTCTL_ATTN_IND_ON:
55
case PCI_EXP_SLTCTL_PWR_IND_BLINK:
56
case PCI_EXP_SLTCTL_ATTN_IND_BLINK:
58
case PCI_EXP_SLTCTL_PWR_IND_OFF:
59
case PCI_EXP_SLTCTL_ATTN_IND_OFF:
66
/***************************************************************************
67
* pci express capability helper functions
71
pcie_cap_v1_fill(PCIDevice *dev, uint8_t port, uint8_t type, uint8_t version)
73
uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
74
uint8_t *cmask = dev->cmask + dev->exp.exp_cap;
76
/* capability register
77
interrupt message number defaults to 0 */
78
pci_set_word(exp_cap + PCI_EXP_FLAGS,
79
((type << PCI_EXP_FLAGS_TYPE_SHIFT) & PCI_EXP_FLAGS_TYPE) |
82
/* device capability register
84
* roll based error reporting bit must be set by all
85
* Functions conforming to the ECN, PCI Express Base
86
* Specification, Revision 1.1., or subsequent PCI Express Base
87
* Specification revisions.
89
pci_set_long(exp_cap + PCI_EXP_DEVCAP, PCI_EXP_DEVCAP_RBER);
91
pci_set_long(exp_cap + PCI_EXP_LNKCAP,
92
(port << PCI_EXP_LNKCAP_PN_SHIFT) |
93
PCI_EXP_LNKCAP_ASPMS_0S |
94
QEMU_PCI_EXP_LNKCAP_MLW(QEMU_PCI_EXP_LNK_X1) |
95
QEMU_PCI_EXP_LNKCAP_MLS(QEMU_PCI_EXP_LNK_2_5GT));
97
pci_set_word(exp_cap + PCI_EXP_LNKSTA,
98
QEMU_PCI_EXP_LNKSTA_NLW(QEMU_PCI_EXP_LNK_X1) |
99
QEMU_PCI_EXP_LNKSTA_CLS(QEMU_PCI_EXP_LNK_2_5GT));
101
/* We changed link status bits over time, and changing them across
102
* migrations is generally fine as hardware changes them too.
103
* Let's not bother checking.
105
pci_set_word(cmask + PCI_EXP_LNKSTA, 0);
108
static void pcie_cap_fill_slot_lnk(PCIDevice *dev)
110
PCIESlot *s = (PCIESlot *)object_dynamic_cast(OBJECT(dev), TYPE_PCIE_SLOT);
111
uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
113
/* Skip anything that isn't a PCIESlot */
118
/* Clear and fill LNKCAP from what was configured above */
119
pci_long_test_and_clear_mask(exp_cap + PCI_EXP_LNKCAP,
120
PCI_EXP_LNKCAP_MLW | PCI_EXP_LNKCAP_SLS);
121
pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP,
122
QEMU_PCI_EXP_LNKCAP_MLW(s->width) |
123
QEMU_PCI_EXP_LNKCAP_MLS(s->speed));
126
* Link bandwidth notification is required for all root ports and
127
* downstream ports supporting links wider than x1 or multiple link
130
if (s->width > QEMU_PCI_EXP_LNK_X1 ||
131
s->speed > QEMU_PCI_EXP_LNK_2_5GT) {
132
pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP,
133
PCI_EXP_LNKCAP_LBNC);
136
if (s->speed > QEMU_PCI_EXP_LNK_2_5GT) {
138
* Hot-plug capable downstream ports and downstream ports supporting
139
* link speeds greater than 5GT/s must hardwire PCI_EXP_LNKCAP_DLLLARC
140
* to 1b. PCI_EXP_LNKCAP_DLLLARC implies PCI_EXP_LNKSTA_DLLLA, which
141
* we also hardwire to 1b here. 2.5GT/s hot-plug slots should also
142
* technically implement this, but it's not done here for compatibility.
144
pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP,
145
PCI_EXP_LNKCAP_DLLLARC);
146
/* the PCI_EXP_LNKSTA_DLLLA will be set in the hotplug function */
149
* Target Link Speed defaults to the highest link speed supported by
150
* the component. 2.5GT/s devices are permitted to hardwire to zero.
152
pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKCTL2,
153
PCI_EXP_LNKCTL2_TLS);
154
pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKCTL2,
155
QEMU_PCI_EXP_LNKCAP_MLS(s->speed) &
156
PCI_EXP_LNKCTL2_TLS);
160
* 2.5 & 5.0GT/s can be fully described by LNKCAP, but 8.0GT/s is
161
* actually a reference to the highest bit supported in this register.
162
* We assume the device supports all link speeds.
164
if (s->speed > QEMU_PCI_EXP_LNK_5GT) {
165
pci_long_test_and_clear_mask(exp_cap + PCI_EXP_LNKCAP2, ~0U);
166
pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP2,
167
PCI_EXP_LNKCAP2_SLS_2_5GB |
168
PCI_EXP_LNKCAP2_SLS_5_0GB |
169
PCI_EXP_LNKCAP2_SLS_8_0GB);
170
if (s->speed > QEMU_PCI_EXP_LNK_8GT) {
171
pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP2,
172
PCI_EXP_LNKCAP2_SLS_16_0GB);
174
if (s->speed > QEMU_PCI_EXP_LNK_16GT) {
175
pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP2,
176
PCI_EXP_LNKCAP2_SLS_32_0GB);
178
if (s->speed > QEMU_PCI_EXP_LNK_32GT) {
179
pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP2,
180
PCI_EXP_LNKCAP2_SLS_64_0GB);
185
int pcie_cap_init(PCIDevice *dev, uint8_t offset,
186
uint8_t type, uint8_t port,
189
/* PCIe cap v2 init */
193
assert(pci_is_express(dev));
195
pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
196
PCI_EXP_VER2_SIZEOF, errp);
200
dev->exp.exp_cap = pos;
201
exp_cap = dev->config + pos;
203
/* Filling values common with v1 */
204
pcie_cap_v1_fill(dev, port, type, PCI_EXP_FLAGS_VER2);
206
/* Fill link speed and width options */
207
pcie_cap_fill_slot_lnk(dev);
209
/* Filling v2 specific values */
210
pci_set_long(exp_cap + PCI_EXP_DEVCAP2,
211
PCI_EXP_DEVCAP2_EFF | PCI_EXP_DEVCAP2_EETLPP);
213
pci_set_word(dev->wmask + pos + PCI_EXP_DEVCTL2, PCI_EXP_DEVCTL2_EETLPPB);
215
if (dev->cap_present & QEMU_PCIE_EXTCAP_INIT) {
216
/* read-only to behave like a 'NULL' Extended Capability Header */
217
pci_set_long(dev->wmask + PCI_CONFIG_SPACE_SIZE, 0);
223
int pcie_cap_v1_init(PCIDevice *dev, uint8_t offset, uint8_t type,
226
/* PCIe cap v1 init */
228
Error *local_err = NULL;
230
assert(pci_is_express(dev));
232
pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
233
PCI_EXP_VER1_SIZEOF, &local_err);
235
error_report_err(local_err);
238
dev->exp.exp_cap = pos;
240
pcie_cap_v1_fill(dev, port, type, PCI_EXP_FLAGS_VER1);
246
pcie_endpoint_cap_common_init(PCIDevice *dev, uint8_t offset, uint8_t cap_size)
248
uint8_t type = PCI_EXP_TYPE_ENDPOINT;
249
Error *local_err = NULL;
253
* Windows guests will report Code 10, device cannot start, if
254
* a regular Endpoint type is exposed on a root complex. These
255
* should instead be Root Complex Integrated Endpoints.
257
if (pci_bus_is_express(pci_get_bus(dev))
258
&& pci_bus_is_root(pci_get_bus(dev))) {
259
type = PCI_EXP_TYPE_RC_END;
262
if (cap_size == PCI_EXP_VER1_SIZEOF) {
263
return pcie_cap_v1_init(dev, offset, type, 0);
265
ret = pcie_cap_init(dev, offset, type, 0, &local_err);
268
error_report_err(local_err);
275
int pcie_endpoint_cap_init(PCIDevice *dev, uint8_t offset)
277
return pcie_endpoint_cap_common_init(dev, offset, PCI_EXP_VER2_SIZEOF);
280
int pcie_endpoint_cap_v1_init(PCIDevice *dev, uint8_t offset)
282
return pcie_endpoint_cap_common_init(dev, offset, PCI_EXP_VER1_SIZEOF);
285
void pcie_cap_exit(PCIDevice *dev)
287
pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER2_SIZEOF);
290
void pcie_cap_v1_exit(PCIDevice *dev)
292
pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER1_SIZEOF);
295
uint8_t pcie_cap_get_type(const PCIDevice *dev)
297
uint32_t pos = dev->exp.exp_cap;
299
return (pci_get_word(dev->config + pos + PCI_EXP_FLAGS) &
300
PCI_EXP_FLAGS_TYPE) >> PCI_EXP_FLAGS_TYPE_SHIFT;
303
uint8_t pcie_cap_get_version(const PCIDevice *dev)
305
uint32_t pos = dev->exp.exp_cap;
307
return pci_get_word(dev->config + pos + PCI_EXP_FLAGS) & PCI_EXP_FLAGS_VERS;
311
/* pci express interrupt message number */
312
/* 7.8.2 PCI Express Capabilities Register: Interrupt Message Number */
313
void pcie_cap_flags_set_vector(PCIDevice *dev, uint8_t vector)
315
uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
317
pci_word_test_and_clear_mask(exp_cap + PCI_EXP_FLAGS, PCI_EXP_FLAGS_IRQ);
318
pci_word_test_and_set_mask(exp_cap + PCI_EXP_FLAGS,
319
vector << PCI_EXP_FLAGS_IRQ_SHIFT);
322
uint8_t pcie_cap_flags_get_vector(PCIDevice *dev)
324
return (pci_get_word(dev->config + dev->exp.exp_cap + PCI_EXP_FLAGS) &
325
PCI_EXP_FLAGS_IRQ) >> PCI_EXP_FLAGS_IRQ_SHIFT;
328
void pcie_cap_deverr_init(PCIDevice *dev)
330
uint32_t pos = dev->exp.exp_cap;
331
pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP,
332
PCI_EXP_DEVCAP_RBER);
333
pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL,
334
PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
335
PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
336
pci_long_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_DEVSTA,
337
PCI_EXP_DEVSTA_CED | PCI_EXP_DEVSTA_NFED |
338
PCI_EXP_DEVSTA_FED | PCI_EXP_DEVSTA_URD);
341
void pcie_cap_deverr_reset(PCIDevice *dev)
343
uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
344
pci_long_test_and_clear_mask(devctl,
345
PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
346
PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
349
void pcie_cap_lnkctl_init(PCIDevice *dev)
351
uint32_t pos = dev->exp.exp_cap;
352
pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_LNKCTL,
353
PCI_EXP_LNKCTL_CCC | PCI_EXP_LNKCTL_ES);
356
void pcie_cap_lnkctl_reset(PCIDevice *dev)
358
uint8_t *lnkctl = dev->config + dev->exp.exp_cap + PCI_EXP_LNKCTL;
359
pci_long_test_and_clear_mask(lnkctl,
360
PCI_EXP_LNKCTL_CCC | PCI_EXP_LNKCTL_ES);
363
static void hotplug_event_update_event_status(PCIDevice *dev)
365
uint32_t pos = dev->exp.exp_cap;
366
uint8_t *exp_cap = dev->config + pos;
367
uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
368
uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
370
dev->exp.hpev_notified = (sltctl & PCI_EXP_SLTCTL_HPIE) &&
371
(sltsta & sltctl & PCI_EXP_HP_EV_SUPPORTED);
374
static void hotplug_event_notify(PCIDevice *dev)
376
bool prev = dev->exp.hpev_notified;
378
hotplug_event_update_event_status(dev);
380
if (prev == dev->exp.hpev_notified) {
384
/* Note: the logic above does not take into account whether interrupts
385
* are masked. The result is that interrupt will be sent when it is
386
* subsequently unmasked. This appears to be legal: Section 6.7.3.4:
387
* The Port may optionally send an MSI when there are hot-plug events that
388
* occur while interrupt generation is disabled, and interrupt generation is
389
* subsequently enabled. */
390
if (msix_enabled(dev)) {
391
msix_notify(dev, pcie_cap_flags_get_vector(dev));
392
} else if (msi_enabled(dev)) {
393
msi_notify(dev, pcie_cap_flags_get_vector(dev));
394
} else if (pci_intx(dev) != -1) {
395
pci_set_irq(dev, dev->exp.hpev_notified);
399
static void hotplug_event_clear(PCIDevice *dev)
401
hotplug_event_update_event_status(dev);
402
if (!msix_enabled(dev) && !msi_enabled(dev) && pci_intx(dev) != -1 &&
403
!dev->exp.hpev_notified) {
404
pci_irq_deassert(dev);
408
void pcie_cap_slot_enable_power(PCIDevice *dev)
410
uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
411
uint32_t sltcap = pci_get_long(exp_cap + PCI_EXP_SLTCAP);
413
if (sltcap & PCI_EXP_SLTCAP_PCP) {
414
pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
419
static void pcie_set_power_device(PCIBus *bus, PCIDevice *dev, void *opaque)
421
bool *power = opaque;
423
pci_set_power(dev, *power);
426
static void pcie_cap_update_power(PCIDevice *hotplug_dev)
428
uint8_t *exp_cap = hotplug_dev->config + hotplug_dev->exp.exp_cap;
429
PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(hotplug_dev));
430
uint32_t sltcap = pci_get_long(exp_cap + PCI_EXP_SLTCAP);
431
uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
434
if (sltcap & PCI_EXP_SLTCAP_PCP) {
435
power = (sltctl & PCI_EXP_SLTCTL_PCC) == PCI_EXP_SLTCTL_PWR_ON;
436
/* Don't we need to check also (sltctl & PCI_EXP_SLTCTL_PIC) ? */
439
pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
440
pcie_set_power_device, &power);
444
* A PCI Express Hot-Plug Event has occurred, so update slot status register
445
* and notify OS of the event if necessary.
447
* 6.7.3 PCI Express Hot-Plug Events
448
* 6.7.3.4 Software Notification of Hot-Plug Events
450
static void pcie_cap_slot_event(PCIDevice *dev, PCIExpressHotPlugEvent event)
452
/* Minor optimization: if nothing changed - no event is needed. */
453
if (pci_word_test_and_set_mask(dev->config + dev->exp.exp_cap +
454
PCI_EXP_SLTSTA, event) == event) {
457
hotplug_event_notify(dev);
460
static void pcie_cap_slot_plug_common(PCIDevice *hotplug_dev, DeviceState *dev,
463
uint8_t *exp_cap = hotplug_dev->config + hotplug_dev->exp.exp_cap;
464
uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
466
PCIE_DEV_PRINTF(PCI_DEVICE(dev), "hotplug state: 0x%x\n", sltsta);
467
if (sltsta & PCI_EXP_SLTSTA_EIS) {
468
/* the slot is electromechanically locked.
469
* This error is propagated up to qdev and then to HMP/QMP.
471
error_setg_errno(errp, EBUSY, "slot is electromechanically locked");
475
void pcie_cap_slot_pre_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
478
PCIDevice *hotplug_pdev = PCI_DEVICE(hotplug_dev);
479
uint8_t *exp_cap = hotplug_pdev->config + hotplug_pdev->exp.exp_cap;
480
uint32_t sltcap = pci_get_word(exp_cap + PCI_EXP_SLTCAP);
482
/* Check if hot-plug is disabled on the slot */
483
if (dev->hotplugged && (sltcap & PCI_EXP_SLTCAP_HPC) == 0) {
484
error_setg(errp, "Hot-plug failed: unsupported by the port device '%s'",
485
DEVICE(hotplug_pdev)->id);
489
pcie_cap_slot_plug_common(PCI_DEVICE(hotplug_dev), dev, errp);
492
void pcie_cap_slot_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
495
PCIDevice *hotplug_pdev = PCI_DEVICE(hotplug_dev);
496
uint8_t *exp_cap = hotplug_pdev->config + hotplug_pdev->exp.exp_cap;
497
PCIDevice *pci_dev = PCI_DEVICE(dev);
498
uint32_t lnkcap = pci_get_long(exp_cap + PCI_EXP_LNKCAP);
500
if (pci_is_vf(pci_dev)) {
501
/* Virtual function cannot be physically disconnected */
505
/* Don't send event when device is enabled during qemu machine creation:
506
* it is present on boot, no hotplug event is necessary. We do send an
507
* event when the device is disabled later. */
508
if (!dev->hotplugged) {
509
pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
511
if (pci_dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA ||
512
(lnkcap & PCI_EXP_LNKCAP_DLLLARC)) {
513
pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA,
514
PCI_EXP_LNKSTA_DLLLA);
516
pcie_cap_update_power(hotplug_pdev);
520
/* To enable multifunction hot-plug, we just ensure the function
521
* 0 added last. When function 0 is added, we set the sltsta and
522
* inform OS via event notification.
524
if (pci_get_function_0(pci_dev)) {
525
pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
527
if (pci_dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA ||
528
(lnkcap & PCI_EXP_LNKCAP_DLLLARC)) {
529
pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA,
530
PCI_EXP_LNKSTA_DLLLA);
532
pcie_cap_slot_event(hotplug_pdev,
533
PCI_EXP_HP_EV_PDC | PCI_EXP_HP_EV_ABP);
534
pcie_cap_update_power(hotplug_pdev);
538
void pcie_cap_slot_unplug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
544
static void pcie_unplug_device(PCIBus *bus, PCIDevice *dev, void *opaque)
546
HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(DEVICE(dev));
548
if (dev->partially_hotplugged) {
549
dev->qdev.pending_deleted_event = false;
552
hotplug_handler_unplug(hotplug_ctrl, DEVICE(dev), &error_abort);
553
object_unparent(OBJECT(dev));
556
static void pcie_cap_slot_do_unplug(PCIDevice *dev)
558
PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
559
uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
560
uint32_t lnkcap = pci_get_long(exp_cap + PCI_EXP_LNKCAP);
562
pci_for_each_device_under_bus(sec_bus, pcie_unplug_device, NULL);
564
pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
566
if (dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA ||
567
(lnkcap & PCI_EXP_LNKCAP_DLLLARC)) {
568
pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKSTA,
569
PCI_EXP_LNKSTA_DLLLA);
571
pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
575
void pcie_cap_slot_unplug_request_cb(HotplugHandler *hotplug_dev,
576
DeviceState *dev, Error **errp)
578
Error *local_err = NULL;
579
PCIDevice *pci_dev = PCI_DEVICE(dev);
580
PCIBus *bus = pci_get_bus(pci_dev);
581
PCIDevice *hotplug_pdev = PCI_DEVICE(hotplug_dev);
582
uint8_t *exp_cap = hotplug_pdev->config + hotplug_pdev->exp.exp_cap;
583
uint32_t sltcap = pci_get_word(exp_cap + PCI_EXP_SLTCAP);
584
uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
586
/* Check if hot-unplug is disabled on the slot */
587
if ((sltcap & PCI_EXP_SLTCAP_HPC) == 0) {
588
error_setg(errp, "Hot-unplug failed: "
589
"unsupported by the port device '%s'",
590
DEVICE(hotplug_pdev)->id);
594
pcie_cap_slot_plug_common(hotplug_pdev, dev, &local_err);
596
error_propagate(errp, local_err);
600
if ((sltctl & PCI_EXP_SLTCTL_PIC) == PCI_EXP_SLTCTL_PWR_IND_BLINK) {
601
error_setg(errp, "Hot-unplug failed: "
602
"guest is busy (power indicator blinking)");
606
dev->pending_deleted_event = true;
607
dev->pending_deleted_expires_ms =
608
qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 5000; /* 5 secs */
610
/* In case user cancel the operation of multi-function hot-add,
611
* remove the function that is unexposed to guest individually,
612
* without interaction with guest.
614
if (pci_dev->devfn &&
616
pcie_unplug_device(bus, pci_dev, NULL);
621
if (pcie_sltctl_powered_off(sltctl)) {
622
/* slot is powered off -> unplug without round-trip to the guest */
623
pcie_cap_slot_do_unplug(hotplug_pdev);
624
hotplug_event_notify(hotplug_pdev);
625
pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
630
pcie_cap_slot_push_attention_button(hotplug_pdev);
633
/* pci express slot for pci express root/downstream port
634
PCI express capability slot registers */
635
void pcie_cap_slot_init(PCIDevice *dev, PCIESlot *s)
637
uint32_t pos = dev->exp.exp_cap;
639
pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_FLAGS,
642
pci_long_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCAP,
643
~PCI_EXP_SLTCAP_PSN);
644
pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
645
(s->slot << PCI_EXP_SLTCAP_PSN_SHIFT) |
652
* Expose native hot-plug on all bridges if hot-plug is enabled on the slot.
653
* (unless broken 6.1 ABI is enforced for compat reasons)
656
(!s->hide_native_hotplug_cap || DEVICE(dev)->hotplugged)) {
657
pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
662
if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
663
pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
665
pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
667
pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
671
pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
674
pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCTL,
675
PCI_EXP_SLTCTL_PWR_IND_OFF |
676
PCI_EXP_SLTCTL_ATTN_IND_OFF);
677
pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
680
PCI_EXP_SLTCTL_HPIE |
681
PCI_EXP_SLTCTL_CCIE |
682
PCI_EXP_SLTCTL_PDCE |
683
PCI_EXP_SLTCTL_ABPE);
684
/* Although reading PCI_EXP_SLTCTL_EIC returns always 0,
685
* make the bit writable here in order to detect 1b is written.
686
* pcie_cap_slot_write_config() test-and-clear the bit, so
687
* this bit always returns 0 to the guest.
689
pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
692
pci_word_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_SLTSTA,
693
PCI_EXP_HP_EV_SUPPORTED);
695
/* Avoid migration abortion when this device hot-removed by guest */
696
pci_word_test_and_clear_mask(dev->cmask + pos + PCI_EXP_SLTSTA,
699
dev->exp.hpev_notified = false;
701
qbus_set_hotplug_handler(BUS(pci_bridge_get_sec_bus(PCI_BRIDGE(dev))),
705
void pcie_cap_slot_reset(PCIDevice *dev)
707
uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
708
uint8_t port_type = pcie_cap_get_type(dev);
710
assert(port_type == PCI_EXP_TYPE_DOWNSTREAM ||
711
port_type == PCI_EXP_TYPE_ROOT_PORT);
713
PCIE_DEV_PRINTF(dev, "reset\n");
715
pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
719
PCI_EXP_SLTCTL_HPIE |
720
PCI_EXP_SLTCTL_CCIE |
721
PCI_EXP_SLTCTL_PDCE |
722
PCI_EXP_SLTCTL_ABPE);
723
pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
724
PCI_EXP_SLTCTL_PWR_IND_OFF |
725
PCI_EXP_SLTCTL_ATTN_IND_OFF);
727
if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
728
/* Downstream ports enforce device number 0. */
729
bool populated = pci_bridge_get_sec_bus(PCI_BRIDGE(dev))->devices[0];
733
pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
736
pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
741
PCI_EXP_SLTCTL_PWR_IND_ON : PCI_EXP_SLTCTL_PWR_IND_OFF;
742
pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL, pic);
745
pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
746
PCI_EXP_SLTSTA_EIS |/* on reset,
747
the lock is released */
752
pcie_cap_update_power(dev);
753
hotplug_event_update_event_status(dev);
756
void pcie_cap_slot_get(PCIDevice *dev, uint16_t *slt_ctl, uint16_t *slt_sta)
758
uint32_t pos = dev->exp.exp_cap;
759
uint8_t *exp_cap = dev->config + pos;
760
*slt_ctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
761
*slt_sta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
764
static void find_child_fn(PCIBus *bus, PCIDevice *dev, void *opaque)
766
PCIDevice **child = opaque;
774
* Returns the plugged device or first function of multifunction plugged device
776
static PCIDevice *pcie_cap_slot_find_child(PCIDevice *dev)
778
PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
779
PCIDevice *child = NULL;
781
pci_for_each_device(sec_bus, pci_bus_num(sec_bus), find_child_fn, &child);
786
void pcie_cap_slot_write_config(PCIDevice *dev,
787
uint16_t old_slt_ctl, uint16_t old_slt_sta,
788
uint32_t addr, uint32_t val, int len)
790
uint32_t pos = dev->exp.exp_cap;
791
uint8_t *exp_cap = dev->config + pos;
792
uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
794
if (ranges_overlap(addr, len, pos + PCI_EXP_SLTSTA, 2)) {
796
* Guests tend to clears all bits during init.
797
* If they clear bits that weren't set this is racy and will lose events:
798
* not a big problem for manual button presses, but a problem for us.
799
* As a work-around, detect this and revert status to what it was
802
* Note: in theory this can be detected as a duplicate button press
803
* which cancels the previous press. Does not seem to happen in
804
* practice as guests seem to only have this bug during init.
806
#define PCIE_SLOT_EVENTS (PCI_EXP_SLTSTA_ABP | PCI_EXP_SLTSTA_PFD | \
807
PCI_EXP_SLTSTA_MRLSC | PCI_EXP_SLTSTA_PDC | \
810
if (val & ~old_slt_sta & PCIE_SLOT_EVENTS) {
811
sltsta = (sltsta & ~PCIE_SLOT_EVENTS) | (old_slt_sta & PCIE_SLOT_EVENTS);
812
pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
814
hotplug_event_clear(dev);
817
if (!ranges_overlap(addr, len, pos + PCI_EXP_SLTCTL, 2)) {
821
if (pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
822
PCI_EXP_SLTCTL_EIC)) {
823
sltsta ^= PCI_EXP_SLTSTA_EIS; /* toggle PCI_EXP_SLTSTA_EIS bit */
824
pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
825
PCIE_DEV_PRINTF(dev, "PCI_EXP_SLTCTL_EIC: "
826
"sltsta -> 0x%02"PRIx16"\n",
830
if (trace_event_get_state_backends(TRACE_PCIE_CAP_SLOT_WRITE_CONFIG)) {
831
DeviceState *parent = DEVICE(dev);
832
DeviceState *child = DEVICE(pcie_cap_slot_find_child(dev));
834
trace_pcie_cap_slot_write_config(
835
parent->canonical_path,
836
child ? child->canonical_path : "no-child",
837
(sltsta & PCI_EXP_SLTSTA_PDS) ? "present" : "not present",
838
pcie_led_state_to_str(old_slt_ctl & PCI_EXP_SLTCTL_PIC),
839
pcie_led_state_to_str(val & PCI_EXP_SLTCTL_PIC),
840
pcie_led_state_to_str(old_slt_ctl & PCI_EXP_SLTCTL_AIC),
841
pcie_led_state_to_str(val & PCI_EXP_SLTCTL_AIC),
842
(old_slt_ctl & PCI_EXP_SLTCTL_PWR_OFF) ? "off" : "on",
843
(val & PCI_EXP_SLTCTL_PWR_OFF) ? "off" : "on");
847
* If the slot is populated, power indicator is off and power
848
* controller is off, it is safe to detach the devices.
850
* Note: don't detach if condition was already true:
851
* this is a work around for guests that overwrite
852
* control of powered off slots before powering them on.
854
if ((sltsta & PCI_EXP_SLTSTA_PDS) && pcie_sltctl_powered_off(val) &&
855
!pcie_sltctl_powered_off(old_slt_ctl))
857
pcie_cap_slot_do_unplug(dev);
859
pcie_cap_update_power(dev);
861
hotplug_event_notify(dev);
864
* 6.7.3.2 Command Completed Events
866
* Software issues a command to a hot-plug capable Downstream Port by
867
* issuing a write transaction that targets any portion of the Port’s Slot
868
* Control register. A single write to the Slot Control register is
869
* considered to be a single command, even if the write affects more than
870
* one field in the Slot Control register. In response to this transaction,
871
* the Port must carry out the requested actions and then set the
872
* associated status field for the command completed event. */
874
/* Real hardware might take a while to complete requested command because
875
* physical movement would be involved like locking the electromechanical
876
* lock. However in our case, command is completed instantaneously above,
877
* so send a command completion event right now.
879
pcie_cap_slot_event(dev, PCI_EXP_HP_EV_CCI);
882
int pcie_cap_slot_post_load(void *opaque, int version_id)
884
PCIDevice *dev = opaque;
885
hotplug_event_update_event_status(dev);
886
pcie_cap_update_power(dev);
890
void pcie_cap_slot_push_attention_button(PCIDevice *dev)
892
pcie_cap_slot_event(dev, PCI_EXP_HP_EV_ABP);
895
/* root control/capabilities/status. PME isn't emulated for now */
896
void pcie_cap_root_init(PCIDevice *dev)
898
pci_set_word(dev->wmask + dev->exp.exp_cap + PCI_EXP_RTCTL,
899
PCI_EXP_RTCTL_SECEE | PCI_EXP_RTCTL_SENFEE |
900
PCI_EXP_RTCTL_SEFEE);
903
void pcie_cap_root_reset(PCIDevice *dev)
905
pci_set_word(dev->config + dev->exp.exp_cap + PCI_EXP_RTCTL, 0);
908
/* function level reset(FLR) */
909
void pcie_cap_flr_init(PCIDevice *dev)
911
pci_long_test_and_set_mask(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCAP,
914
/* Although reading BCR_FLR returns always 0,
915
* the bit is made writable here in order to detect the 1b is written
916
* pcie_cap_flr_write_config() test-and-clear the bit, so
917
* this bit always returns 0 to the guest.
919
pci_word_test_and_set_mask(dev->wmask + dev->exp.exp_cap + PCI_EXP_DEVCTL,
920
PCI_EXP_DEVCTL_BCR_FLR);
923
void pcie_cap_flr_write_config(PCIDevice *dev,
924
uint32_t addr, uint32_t val, int len)
926
uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
927
if (pci_get_word(devctl) & PCI_EXP_DEVCTL_BCR_FLR) {
928
/* Clear PCI_EXP_DEVCTL_BCR_FLR after invoking the reset handler
929
so the handler can detect FLR by looking at this bit. */
930
pci_device_reset(dev);
931
pci_word_test_and_clear_mask(devctl, PCI_EXP_DEVCTL_BCR_FLR);
935
/* Alternative Routing-ID Interpretation (ARI)
936
* forwarding support for root and downstream ports
938
void pcie_cap_arifwd_init(PCIDevice *dev)
940
uint32_t pos = dev->exp.exp_cap;
941
pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP2,
942
PCI_EXP_DEVCAP2_ARI);
943
pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL2,
944
PCI_EXP_DEVCTL2_ARI);
947
void pcie_cap_arifwd_reset(PCIDevice *dev)
949
uint8_t *devctl2 = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2;
950
pci_long_test_and_clear_mask(devctl2, PCI_EXP_DEVCTL2_ARI);
953
bool pcie_cap_is_arifwd_enabled(const PCIDevice *dev)
955
if (!pci_is_express(dev)) {
958
if (!dev->exp.exp_cap) {
962
return pci_get_long(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2) &
966
/**************************************************************************
967
* pci express extended capability list management functions
968
* uint16_t ext_cap_id (16 bit)
969
* uint8_t cap_ver (4 bit)
970
* uint16_t cap_offset (12 bit)
971
* uint16_t ext_cap_size
974
/* Passing a cap_id value > 0xffff will return 0 and put end of list in prev */
975
static uint16_t pcie_find_capability_list(PCIDevice *dev, uint32_t cap_id,
980
uint32_t header = pci_get_long(dev->config + PCI_CONFIG_SPACE_SIZE);
983
/* no extended capability */
987
for (next = PCI_CONFIG_SPACE_SIZE; next;
988
prev = next, next = PCI_EXT_CAP_NEXT(header)) {
990
assert(next >= PCI_CONFIG_SPACE_SIZE);
991
assert(next <= PCIE_CONFIG_SPACE_SIZE - 8);
993
header = pci_get_long(dev->config + next);
994
if (PCI_EXT_CAP_ID(header) == cap_id) {
1006
uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id)
1008
return pcie_find_capability_list(dev, cap_id, NULL);
1011
static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next)
1013
uint32_t header = pci_get_long(dev->config + pos);
1014
assert(!(next & (PCI_EXT_CAP_ALIGN - 1)));
1015
header = (header & ~PCI_EXT_CAP_NEXT_MASK) |
1016
((next << PCI_EXT_CAP_NEXT_SHIFT) & PCI_EXT_CAP_NEXT_MASK);
1017
pci_set_long(dev->config + pos, header);
1021
* Caller must supply valid (offset, size) such that the range wouldn't
1022
* overlap with other capability or other registers.
1023
* This function doesn't check it.
1025
void pcie_add_capability(PCIDevice *dev,
1026
uint16_t cap_id, uint8_t cap_ver,
1027
uint16_t offset, uint16_t size)
1029
assert(offset >= PCI_CONFIG_SPACE_SIZE);
1030
assert(offset < (uint16_t)(offset + size));
1031
assert((uint16_t)(offset + size) <= PCIE_CONFIG_SPACE_SIZE);
1033
assert(pci_is_express(dev));
1035
if (offset != PCI_CONFIG_SPACE_SIZE) {
1039
* 0xffffffff is not a valid cap id (it's a 16 bit field). use
1040
* internally to find the last capability in the linked list.
1042
pcie_find_capability_list(dev, 0xffffffff, &prev);
1043
assert(prev >= PCI_CONFIG_SPACE_SIZE);
1044
pcie_ext_cap_set_next(dev, prev, offset);
1046
pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, 0));
1048
/* Make capability read-only by default */
1049
memset(dev->wmask + offset, 0, size);
1050
memset(dev->w1cmask + offset, 0, size);
1051
/* Check capability by default */
1052
memset(dev->cmask + offset, 0xFF, size);
1056
* Sync the PCIe Link Status negotiated speed and width of a bridge with the
1057
* downstream device. If downstream device is not present, re-write with the
1058
* Link Capability fields. If downstream device reports invalid width or
1059
* speed, replace with minimum values (LnkSta fields are RsvdZ on VFs but such
1060
* values interfere with PCIe native hotplug detecting new devices). Limit
1061
* width and speed to bridge capabilities for compatibility. Use config_read
1062
* to access the downstream device since it could be an assigned device with
1063
* volatile link information.
1065
void pcie_sync_bridge_lnk(PCIDevice *bridge_dev)
1067
PCIBridge *br = PCI_BRIDGE(bridge_dev);
1068
PCIBus *bus = pci_bridge_get_sec_bus(br);
1069
PCIDevice *target = bus->devices[0];
1070
uint8_t *exp_cap = bridge_dev->config + bridge_dev->exp.exp_cap;
1071
uint16_t lnksta, lnkcap = pci_get_word(exp_cap + PCI_EXP_LNKCAP);
1073
if (!target || !target->exp.exp_cap) {
1076
lnksta = target->config_read(target,
1077
target->exp.exp_cap + PCI_EXP_LNKSTA,
1080
if ((lnksta & PCI_EXP_LNKSTA_NLW) > (lnkcap & PCI_EXP_LNKCAP_MLW)) {
1081
lnksta &= ~PCI_EXP_LNKSTA_NLW;
1082
lnksta |= lnkcap & PCI_EXP_LNKCAP_MLW;
1083
} else if (!(lnksta & PCI_EXP_LNKSTA_NLW)) {
1084
lnksta |= QEMU_PCI_EXP_LNKSTA_NLW(QEMU_PCI_EXP_LNK_X1);
1087
if ((lnksta & PCI_EXP_LNKSTA_CLS) > (lnkcap & PCI_EXP_LNKCAP_SLS)) {
1088
lnksta &= ~PCI_EXP_LNKSTA_CLS;
1089
lnksta |= lnkcap & PCI_EXP_LNKCAP_SLS;
1090
} else if (!(lnksta & PCI_EXP_LNKSTA_CLS)) {
1091
lnksta |= QEMU_PCI_EXP_LNKSTA_CLS(QEMU_PCI_EXP_LNK_2_5GT);
1095
pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKSTA,
1096
PCI_EXP_LNKSTA_CLS | PCI_EXP_LNKSTA_NLW);
1097
pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA, lnksta &
1098
(PCI_EXP_LNKSTA_CLS | PCI_EXP_LNKSTA_NLW));
1101
/**************************************************************************
1102
* pci express extended capability helper functions
1106
void pcie_ari_init(PCIDevice *dev, uint16_t offset)
1108
uint16_t nextfn = dev->cap_present & QEMU_PCIE_ARI_NEXTFN_1 ? 1 : 0;
1110
pcie_add_capability(dev, PCI_EXT_CAP_ID_ARI, PCI_ARI_VER,
1111
offset, PCI_ARI_SIZEOF);
1112
pci_set_long(dev->config + offset + PCI_ARI_CAP, (nextfn & 0xff) << 8);
1115
void pcie_dev_ser_num_init(PCIDevice *dev, uint16_t offset, uint64_t ser_num)
1117
static const int pci_dsn_ver = 1;
1118
static const int pci_dsn_cap = 4;
1120
pcie_add_capability(dev, PCI_EXT_CAP_ID_DSN, pci_dsn_ver, offset,
1121
PCI_EXT_CAP_DSN_SIZEOF);
1122
pci_set_quad(dev->config + offset + pci_dsn_cap, ser_num);
1125
void pcie_ats_init(PCIDevice *dev, uint16_t offset, bool aligned)
1127
pcie_add_capability(dev, PCI_EXT_CAP_ID_ATS, 0x1,
1128
offset, PCI_EXT_CAP_ATS_SIZEOF);
1130
dev->exp.ats_cap = offset;
1132
/* Invalidate Queue Depth 0 */
1134
pci_set_word(dev->config + offset + PCI_ATS_CAP,
1135
PCI_ATS_CAP_PAGE_ALIGNED);
1137
/* STU 0, Disabled by default */
1138
pci_set_word(dev->config + offset + PCI_ATS_CTRL, 0);
1140
pci_set_word(dev->wmask + dev->exp.ats_cap + PCI_ATS_CTRL, 0x800f);
1143
/* ACS (Access Control Services) */
1144
void pcie_acs_init(PCIDevice *dev, uint16_t offset)
1146
bool is_downstream = pci_is_express_downstream_port(dev);
1147
uint16_t cap_bits = 0;
1149
/* For endpoints, only multifunction devs may have an ACS capability: */
1150
assert(is_downstream ||
1151
(dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) ||
1152
PCI_FUNC(dev->devfn));
1154
pcie_add_capability(dev, PCI_EXT_CAP_ID_ACS, PCI_ACS_VER, offset,
1156
dev->exp.acs_cap = offset;
1158
if (is_downstream) {
1160
* Downstream ports must implement SV, TB, RR, CR, UF, and DT (with
1161
* caveats on the latter four that we ignore for simplicity).
1162
* Endpoints may also implement a subset of ACS capabilities,
1163
* but these are optional if the endpoint does not support
1164
* peer-to-peer between functions and thus omitted here.
1166
cap_bits = PCI_ACS_SV | PCI_ACS_TB | PCI_ACS_RR |
1167
PCI_ACS_CR | PCI_ACS_UF | PCI_ACS_DT;
1170
pci_set_word(dev->config + offset + PCI_ACS_CAP, cap_bits);
1171
pci_set_word(dev->wmask + offset + PCI_ACS_CTRL, cap_bits);
1174
void pcie_acs_reset(PCIDevice *dev)
1176
if (dev->exp.acs_cap) {
1177
pci_set_word(dev->config + dev->exp.acs_cap + PCI_ACS_CTRL, 0);