xref: /openbmc/qemu/hw/pci/pcie.c (revision bac4711b)
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 "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"
31 
32 //#define DEBUG_PCIE
33 #ifdef DEBUG_PCIE
34 # define PCIE_DPRINTF(fmt, ...)                                         \
35     fprintf(stderr, "%s:%d " fmt, __func__, __LINE__, ## __VA_ARGS__)
36 #else
37 # define PCIE_DPRINTF(fmt, ...) do {} while (0)
38 #endif
39 #define PCIE_DEV_PRINTF(dev, fmt, ...)                                  \
40     PCIE_DPRINTF("%s:%x "fmt, (dev)->name, (dev)->devfn, ## __VA_ARGS__)
41 
42 static bool pcie_sltctl_powered_off(uint16_t sltctl)
43 {
44     return (sltctl & PCI_EXP_SLTCTL_PCC) == PCI_EXP_SLTCTL_PWR_OFF
45         && (sltctl & PCI_EXP_SLTCTL_PIC) == PCI_EXP_SLTCTL_PWR_IND_OFF;
46 }
47 
48 /***************************************************************************
49  * pci express capability helper functions
50  */
51 
52 static void
53 pcie_cap_v1_fill(PCIDevice *dev, uint8_t port, uint8_t type, uint8_t version)
54 {
55     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
56     uint8_t *cmask = dev->cmask + dev->exp.exp_cap;
57 
58     /* capability register
59     interrupt message number defaults to 0 */
60     pci_set_word(exp_cap + PCI_EXP_FLAGS,
61                  ((type << PCI_EXP_FLAGS_TYPE_SHIFT) & PCI_EXP_FLAGS_TYPE) |
62                  version);
63 
64     /* device capability register
65      * table 7-12:
66      * roll based error reporting bit must be set by all
67      * Functions conforming to the ECN, PCI Express Base
68      * Specification, Revision 1.1., or subsequent PCI Express Base
69      * Specification revisions.
70      */
71     pci_set_long(exp_cap + PCI_EXP_DEVCAP, PCI_EXP_DEVCAP_RBER);
72 
73     pci_set_long(exp_cap + PCI_EXP_LNKCAP,
74                  (port << PCI_EXP_LNKCAP_PN_SHIFT) |
75                  PCI_EXP_LNKCAP_ASPMS_0S |
76                  QEMU_PCI_EXP_LNKCAP_MLW(QEMU_PCI_EXP_LNK_X1) |
77                  QEMU_PCI_EXP_LNKCAP_MLS(QEMU_PCI_EXP_LNK_2_5GT));
78 
79     pci_set_word(exp_cap + PCI_EXP_LNKSTA,
80                  QEMU_PCI_EXP_LNKSTA_NLW(QEMU_PCI_EXP_LNK_X1) |
81                  QEMU_PCI_EXP_LNKSTA_CLS(QEMU_PCI_EXP_LNK_2_5GT));
82 
83     /* We changed link status bits over time, and changing them across
84      * migrations is generally fine as hardware changes them too.
85      * Let's not bother checking.
86      */
87     pci_set_word(cmask + PCI_EXP_LNKSTA, 0);
88 }
89 
90 static void pcie_cap_fill_slot_lnk(PCIDevice *dev)
91 {
92     PCIESlot *s = (PCIESlot *)object_dynamic_cast(OBJECT(dev), TYPE_PCIE_SLOT);
93     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
94 
95     /* Skip anything that isn't a PCIESlot */
96     if (!s) {
97         return;
98     }
99 
100     /* Clear and fill LNKCAP from what was configured above */
101     pci_long_test_and_clear_mask(exp_cap + PCI_EXP_LNKCAP,
102                                  PCI_EXP_LNKCAP_MLW | PCI_EXP_LNKCAP_SLS);
103     pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP,
104                                QEMU_PCI_EXP_LNKCAP_MLW(s->width) |
105                                QEMU_PCI_EXP_LNKCAP_MLS(s->speed));
106 
107     /*
108      * Link bandwidth notification is required for all root ports and
109      * downstream ports supporting links wider than x1 or multiple link
110      * speeds.
111      */
112     if (s->width > QEMU_PCI_EXP_LNK_X1 ||
113         s->speed > QEMU_PCI_EXP_LNK_2_5GT) {
114         pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP,
115                                    PCI_EXP_LNKCAP_LBNC);
116     }
117 
118     if (s->speed > QEMU_PCI_EXP_LNK_2_5GT) {
119         /*
120          * Hot-plug capable downstream ports and downstream ports supporting
121          * link speeds greater than 5GT/s must hardwire PCI_EXP_LNKCAP_DLLLARC
122          * to 1b.  PCI_EXP_LNKCAP_DLLLARC implies PCI_EXP_LNKSTA_DLLLA, which
123          * we also hardwire to 1b here.  2.5GT/s hot-plug slots should also
124          * technically implement this, but it's not done here for compatibility.
125          */
126         pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP,
127                                    PCI_EXP_LNKCAP_DLLLARC);
128         /* the PCI_EXP_LNKSTA_DLLLA will be set in the hotplug function */
129 
130         /*
131          * Target Link Speed defaults to the highest link speed supported by
132          * the component.  2.5GT/s devices are permitted to hardwire to zero.
133          */
134         pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKCTL2,
135                                      PCI_EXP_LNKCTL2_TLS);
136         pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKCTL2,
137                                    QEMU_PCI_EXP_LNKCAP_MLS(s->speed) &
138                                    PCI_EXP_LNKCTL2_TLS);
139     }
140 
141     /*
142      * 2.5 & 5.0GT/s can be fully described by LNKCAP, but 8.0GT/s is
143      * actually a reference to the highest bit supported in this register.
144      * We assume the device supports all link speeds.
145      */
146     if (s->speed > QEMU_PCI_EXP_LNK_5GT) {
147         pci_long_test_and_clear_mask(exp_cap + PCI_EXP_LNKCAP2, ~0U);
148         pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP2,
149                                    PCI_EXP_LNKCAP2_SLS_2_5GB |
150                                    PCI_EXP_LNKCAP2_SLS_5_0GB |
151                                    PCI_EXP_LNKCAP2_SLS_8_0GB);
152         if (s->speed > QEMU_PCI_EXP_LNK_8GT) {
153             pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP2,
154                                        PCI_EXP_LNKCAP2_SLS_16_0GB);
155         }
156     }
157 }
158 
159 int pcie_cap_init(PCIDevice *dev, uint8_t offset,
160                   uint8_t type, uint8_t port,
161                   Error **errp)
162 {
163     /* PCIe cap v2 init */
164     int pos;
165     uint8_t *exp_cap;
166 
167     assert(pci_is_express(dev));
168 
169     pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
170                              PCI_EXP_VER2_SIZEOF, errp);
171     if (pos < 0) {
172         return pos;
173     }
174     dev->exp.exp_cap = pos;
175     exp_cap = dev->config + pos;
176 
177     /* Filling values common with v1 */
178     pcie_cap_v1_fill(dev, port, type, PCI_EXP_FLAGS_VER2);
179 
180     /* Fill link speed and width options */
181     pcie_cap_fill_slot_lnk(dev);
182 
183     /* Filling v2 specific values */
184     pci_set_long(exp_cap + PCI_EXP_DEVCAP2,
185                  PCI_EXP_DEVCAP2_EFF | PCI_EXP_DEVCAP2_EETLPP);
186 
187     pci_set_word(dev->wmask + pos + PCI_EXP_DEVCTL2, PCI_EXP_DEVCTL2_EETLPPB);
188 
189     if (dev->cap_present & QEMU_PCIE_EXTCAP_INIT) {
190         /* read-only to behave like a 'NULL' Extended Capability Header */
191         pci_set_long(dev->wmask + PCI_CONFIG_SPACE_SIZE, 0);
192     }
193 
194     return pos;
195 }
196 
197 int pcie_cap_v1_init(PCIDevice *dev, uint8_t offset, uint8_t type,
198                      uint8_t port)
199 {
200     /* PCIe cap v1 init */
201     int pos;
202     Error *local_err = NULL;
203 
204     assert(pci_is_express(dev));
205 
206     pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
207                              PCI_EXP_VER1_SIZEOF, &local_err);
208     if (pos < 0) {
209         error_report_err(local_err);
210         return pos;
211     }
212     dev->exp.exp_cap = pos;
213 
214     pcie_cap_v1_fill(dev, port, type, PCI_EXP_FLAGS_VER1);
215 
216     return pos;
217 }
218 
219 static int
220 pcie_endpoint_cap_common_init(PCIDevice *dev, uint8_t offset, uint8_t cap_size)
221 {
222     uint8_t type = PCI_EXP_TYPE_ENDPOINT;
223     Error *local_err = NULL;
224     int ret;
225 
226     /*
227      * Windows guests will report Code 10, device cannot start, if
228      * a regular Endpoint type is exposed on a root complex.  These
229      * should instead be Root Complex Integrated Endpoints.
230      */
231     if (pci_bus_is_express(pci_get_bus(dev))
232         && pci_bus_is_root(pci_get_bus(dev))) {
233         type = PCI_EXP_TYPE_RC_END;
234     }
235 
236     if (cap_size == PCI_EXP_VER1_SIZEOF) {
237         return pcie_cap_v1_init(dev, offset, type, 0);
238     } else {
239         ret = pcie_cap_init(dev, offset, type, 0, &local_err);
240 
241         if (ret < 0) {
242             error_report_err(local_err);
243         }
244 
245         return ret;
246     }
247 }
248 
249 int pcie_endpoint_cap_init(PCIDevice *dev, uint8_t offset)
250 {
251     return pcie_endpoint_cap_common_init(dev, offset, PCI_EXP_VER2_SIZEOF);
252 }
253 
254 int pcie_endpoint_cap_v1_init(PCIDevice *dev, uint8_t offset)
255 {
256     return pcie_endpoint_cap_common_init(dev, offset, PCI_EXP_VER1_SIZEOF);
257 }
258 
259 void pcie_cap_exit(PCIDevice *dev)
260 {
261     pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER2_SIZEOF);
262 }
263 
264 void pcie_cap_v1_exit(PCIDevice *dev)
265 {
266     pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER1_SIZEOF);
267 }
268 
269 uint8_t pcie_cap_get_type(const PCIDevice *dev)
270 {
271     uint32_t pos = dev->exp.exp_cap;
272     assert(pos > 0);
273     return (pci_get_word(dev->config + pos + PCI_EXP_FLAGS) &
274             PCI_EXP_FLAGS_TYPE) >> PCI_EXP_FLAGS_TYPE_SHIFT;
275 }
276 
277 uint8_t pcie_cap_get_version(const PCIDevice *dev)
278 {
279     uint32_t pos = dev->exp.exp_cap;
280     assert(pos > 0);
281     return pci_get_word(dev->config + pos + PCI_EXP_FLAGS) & PCI_EXP_FLAGS_VERS;
282 }
283 
284 /* MSI/MSI-X */
285 /* pci express interrupt message number */
286 /* 7.8.2 PCI Express Capabilities Register: Interrupt Message Number */
287 void pcie_cap_flags_set_vector(PCIDevice *dev, uint8_t vector)
288 {
289     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
290     assert(vector < 32);
291     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_FLAGS, PCI_EXP_FLAGS_IRQ);
292     pci_word_test_and_set_mask(exp_cap + PCI_EXP_FLAGS,
293                                vector << PCI_EXP_FLAGS_IRQ_SHIFT);
294 }
295 
296 uint8_t pcie_cap_flags_get_vector(PCIDevice *dev)
297 {
298     return (pci_get_word(dev->config + dev->exp.exp_cap + PCI_EXP_FLAGS) &
299             PCI_EXP_FLAGS_IRQ) >> PCI_EXP_FLAGS_IRQ_SHIFT;
300 }
301 
302 void pcie_cap_deverr_init(PCIDevice *dev)
303 {
304     uint32_t pos = dev->exp.exp_cap;
305     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP,
306                                PCI_EXP_DEVCAP_RBER);
307     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL,
308                                PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
309                                PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
310     pci_long_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_DEVSTA,
311                                PCI_EXP_DEVSTA_CED | PCI_EXP_DEVSTA_NFED |
312                                PCI_EXP_DEVSTA_FED | PCI_EXP_DEVSTA_URD);
313 }
314 
315 void pcie_cap_deverr_reset(PCIDevice *dev)
316 {
317     uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
318     pci_long_test_and_clear_mask(devctl,
319                                  PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
320                                  PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
321 }
322 
323 void pcie_cap_lnkctl_init(PCIDevice *dev)
324 {
325     uint32_t pos = dev->exp.exp_cap;
326     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_LNKCTL,
327                                PCI_EXP_LNKCTL_CCC | PCI_EXP_LNKCTL_ES);
328 }
329 
330 void pcie_cap_lnkctl_reset(PCIDevice *dev)
331 {
332     uint8_t *lnkctl = dev->config + dev->exp.exp_cap + PCI_EXP_LNKCTL;
333     pci_long_test_and_clear_mask(lnkctl,
334                                  PCI_EXP_LNKCTL_CCC | PCI_EXP_LNKCTL_ES);
335 }
336 
337 static void hotplug_event_update_event_status(PCIDevice *dev)
338 {
339     uint32_t pos = dev->exp.exp_cap;
340     uint8_t *exp_cap = dev->config + pos;
341     uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
342     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
343 
344     dev->exp.hpev_notified = (sltctl & PCI_EXP_SLTCTL_HPIE) &&
345         (sltsta & sltctl & PCI_EXP_HP_EV_SUPPORTED);
346 }
347 
348 static void hotplug_event_notify(PCIDevice *dev)
349 {
350     bool prev = dev->exp.hpev_notified;
351 
352     hotplug_event_update_event_status(dev);
353 
354     if (prev == dev->exp.hpev_notified) {
355         return;
356     }
357 
358     /* Note: the logic above does not take into account whether interrupts
359      * are masked. The result is that interrupt will be sent when it is
360      * subsequently unmasked. This appears to be legal: Section 6.7.3.4:
361      * The Port may optionally send an MSI when there are hot-plug events that
362      * occur while interrupt generation is disabled, and interrupt generation is
363      * subsequently enabled. */
364     if (msix_enabled(dev)) {
365         msix_notify(dev, pcie_cap_flags_get_vector(dev));
366     } else if (msi_enabled(dev)) {
367         msi_notify(dev, pcie_cap_flags_get_vector(dev));
368     } else if (pci_intx(dev) != -1) {
369         pci_set_irq(dev, dev->exp.hpev_notified);
370     }
371 }
372 
373 static void hotplug_event_clear(PCIDevice *dev)
374 {
375     hotplug_event_update_event_status(dev);
376     if (!msix_enabled(dev) && !msi_enabled(dev) && pci_intx(dev) != -1 &&
377         !dev->exp.hpev_notified) {
378         pci_irq_deassert(dev);
379     }
380 }
381 
382 void pcie_cap_slot_enable_power(PCIDevice *dev)
383 {
384     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
385     uint32_t sltcap = pci_get_long(exp_cap + PCI_EXP_SLTCAP);
386 
387     if (sltcap & PCI_EXP_SLTCAP_PCP) {
388         pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
389                                      PCI_EXP_SLTCTL_PCC);
390     }
391 }
392 
393 static void pcie_set_power_device(PCIBus *bus, PCIDevice *dev, void *opaque)
394 {
395     bool *power = opaque;
396 
397     pci_set_power(dev, *power);
398 }
399 
400 static void pcie_cap_update_power(PCIDevice *hotplug_dev)
401 {
402     uint8_t *exp_cap = hotplug_dev->config + hotplug_dev->exp.exp_cap;
403     PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(hotplug_dev));
404     uint32_t sltcap = pci_get_long(exp_cap + PCI_EXP_SLTCAP);
405     uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
406     bool power = true;
407 
408     if (sltcap & PCI_EXP_SLTCAP_PCP) {
409         power = (sltctl & PCI_EXP_SLTCTL_PCC) == PCI_EXP_SLTCTL_PWR_ON;
410         /* Don't we need to check also (sltctl & PCI_EXP_SLTCTL_PIC) ? */
411     }
412 
413     pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
414                         pcie_set_power_device, &power);
415 }
416 
417 /*
418  * A PCI Express Hot-Plug Event has occurred, so update slot status register
419  * and notify OS of the event if necessary.
420  *
421  * 6.7.3 PCI Express Hot-Plug Events
422  * 6.7.3.4 Software Notification of Hot-Plug Events
423  */
424 static void pcie_cap_slot_event(PCIDevice *dev, PCIExpressHotPlugEvent event)
425 {
426     /* Minor optimization: if nothing changed - no event is needed. */
427     if (pci_word_test_and_set_mask(dev->config + dev->exp.exp_cap +
428                                    PCI_EXP_SLTSTA, event) == event) {
429         return;
430     }
431     hotplug_event_notify(dev);
432 }
433 
434 static void pcie_cap_slot_plug_common(PCIDevice *hotplug_dev, DeviceState *dev,
435                                       Error **errp)
436 {
437     uint8_t *exp_cap = hotplug_dev->config + hotplug_dev->exp.exp_cap;
438     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
439 
440     PCIE_DEV_PRINTF(PCI_DEVICE(dev), "hotplug state: 0x%x\n", sltsta);
441     if (sltsta & PCI_EXP_SLTSTA_EIS) {
442         /* the slot is electromechanically locked.
443          * This error is propagated up to qdev and then to HMP/QMP.
444          */
445         error_setg_errno(errp, EBUSY, "slot is electromechanically locked");
446     }
447 }
448 
449 void pcie_cap_slot_pre_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
450                                Error **errp)
451 {
452     PCIDevice *hotplug_pdev = PCI_DEVICE(hotplug_dev);
453     uint8_t *exp_cap = hotplug_pdev->config + hotplug_pdev->exp.exp_cap;
454     uint32_t sltcap = pci_get_word(exp_cap + PCI_EXP_SLTCAP);
455 
456     /* Check if hot-plug is disabled on the slot */
457     if (dev->hotplugged && (sltcap & PCI_EXP_SLTCAP_HPC) == 0) {
458         error_setg(errp, "Hot-plug failed: unsupported by the port device '%s'",
459                          DEVICE(hotplug_pdev)->id);
460         return;
461     }
462 
463     pcie_cap_slot_plug_common(PCI_DEVICE(hotplug_dev), dev, errp);
464 }
465 
466 void pcie_cap_slot_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
467                            Error **errp)
468 {
469     PCIDevice *hotplug_pdev = PCI_DEVICE(hotplug_dev);
470     uint8_t *exp_cap = hotplug_pdev->config + hotplug_pdev->exp.exp_cap;
471     PCIDevice *pci_dev = PCI_DEVICE(dev);
472     uint32_t lnkcap = pci_get_long(exp_cap + PCI_EXP_LNKCAP);
473 
474     if (pci_is_vf(pci_dev)) {
475         /* Virtual function cannot be physically disconnected */
476         return;
477     }
478 
479     /* Don't send event when device is enabled during qemu machine creation:
480      * it is present on boot, no hotplug event is necessary. We do send an
481      * event when the device is disabled later. */
482     if (!dev->hotplugged) {
483         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
484                                    PCI_EXP_SLTSTA_PDS);
485         if (pci_dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA ||
486             (lnkcap & PCI_EXP_LNKCAP_DLLLARC)) {
487             pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA,
488                                        PCI_EXP_LNKSTA_DLLLA);
489         }
490         pcie_cap_update_power(hotplug_pdev);
491         return;
492     }
493 
494     /* To enable multifunction hot-plug, we just ensure the function
495      * 0 added last. When function 0 is added, we set the sltsta and
496      * inform OS via event notification.
497      */
498     if (pci_get_function_0(pci_dev)) {
499         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
500                                    PCI_EXP_SLTSTA_PDS);
501         if (pci_dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA ||
502             (lnkcap & PCI_EXP_LNKCAP_DLLLARC)) {
503             pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA,
504                                        PCI_EXP_LNKSTA_DLLLA);
505         }
506         pcie_cap_slot_event(hotplug_pdev,
507                             PCI_EXP_HP_EV_PDC | PCI_EXP_HP_EV_ABP);
508         pcie_cap_update_power(hotplug_pdev);
509     }
510 }
511 
512 void pcie_cap_slot_unplug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
513                              Error **errp)
514 {
515     qdev_unrealize(dev);
516 }
517 
518 static void pcie_unplug_device(PCIBus *bus, PCIDevice *dev, void *opaque)
519 {
520     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(DEVICE(dev));
521 
522     if (dev->partially_hotplugged) {
523         dev->qdev.pending_deleted_event = false;
524         return;
525     }
526     hotplug_handler_unplug(hotplug_ctrl, DEVICE(dev), &error_abort);
527     object_unparent(OBJECT(dev));
528 }
529 
530 static void pcie_cap_slot_do_unplug(PCIDevice *dev)
531 {
532     PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
533     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
534     uint32_t lnkcap = pci_get_long(exp_cap + PCI_EXP_LNKCAP);
535 
536     pci_for_each_device_under_bus(sec_bus, pcie_unplug_device, NULL);
537 
538     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
539                                  PCI_EXP_SLTSTA_PDS);
540     if (dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA ||
541         (lnkcap & PCI_EXP_LNKCAP_DLLLARC)) {
542         pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKSTA,
543                                      PCI_EXP_LNKSTA_DLLLA);
544     }
545     pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
546                                PCI_EXP_SLTSTA_PDC);
547 }
548 
549 void pcie_cap_slot_unplug_request_cb(HotplugHandler *hotplug_dev,
550                                      DeviceState *dev, Error **errp)
551 {
552     Error *local_err = NULL;
553     PCIDevice *pci_dev = PCI_DEVICE(dev);
554     PCIBus *bus = pci_get_bus(pci_dev);
555     PCIDevice *hotplug_pdev = PCI_DEVICE(hotplug_dev);
556     uint8_t *exp_cap = hotplug_pdev->config + hotplug_pdev->exp.exp_cap;
557     uint32_t sltcap = pci_get_word(exp_cap + PCI_EXP_SLTCAP);
558     uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
559 
560     /* Check if hot-unplug is disabled on the slot */
561     if ((sltcap & PCI_EXP_SLTCAP_HPC) == 0) {
562         error_setg(errp, "Hot-unplug failed: "
563                          "unsupported by the port device '%s'",
564                          DEVICE(hotplug_pdev)->id);
565         return;
566     }
567 
568     pcie_cap_slot_plug_common(hotplug_pdev, dev, &local_err);
569     if (local_err) {
570         error_propagate(errp, local_err);
571         return;
572     }
573 
574     if ((sltctl & PCI_EXP_SLTCTL_PIC) == PCI_EXP_SLTCTL_PWR_IND_BLINK) {
575         error_setg(errp, "Hot-unplug failed: "
576                    "guest is busy (power indicator blinking)");
577         return;
578     }
579 
580     dev->pending_deleted_event = true;
581     dev->pending_deleted_expires_ms =
582         qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + 5000; /* 5 secs */
583 
584     /* In case user cancel the operation of multi-function hot-add,
585      * remove the function that is unexposed to guest individually,
586      * without interaction with guest.
587      */
588     if (pci_dev->devfn &&
589         !bus->devices[0]) {
590         pcie_unplug_device(bus, pci_dev, NULL);
591 
592         return;
593     }
594 
595     if (pcie_sltctl_powered_off(sltctl)) {
596         /* slot is powered off -> unplug without round-trip to the guest */
597         pcie_cap_slot_do_unplug(hotplug_pdev);
598         hotplug_event_notify(hotplug_pdev);
599         pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
600                                      PCI_EXP_SLTSTA_ABP);
601         return;
602     }
603 
604     pcie_cap_slot_push_attention_button(hotplug_pdev);
605 }
606 
607 /* pci express slot for pci express root/downstream port
608    PCI express capability slot registers */
609 void pcie_cap_slot_init(PCIDevice *dev, PCIESlot *s)
610 {
611     uint32_t pos = dev->exp.exp_cap;
612 
613     pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_FLAGS,
614                                PCI_EXP_FLAGS_SLOT);
615 
616     pci_long_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCAP,
617                                  ~PCI_EXP_SLTCAP_PSN);
618     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
619                                (s->slot << PCI_EXP_SLTCAP_PSN_SHIFT) |
620                                PCI_EXP_SLTCAP_EIP |
621                                PCI_EXP_SLTCAP_PIP |
622                                PCI_EXP_SLTCAP_AIP |
623                                PCI_EXP_SLTCAP_ABP);
624 
625     /*
626      * Expose native hot-plug on all bridges if hot-plug is enabled on the slot.
627      * (unless broken 6.1 ABI is enforced for compat reasons)
628      */
629     if (s->hotplug &&
630         (!s->hide_native_hotplug_cap || DEVICE(dev)->hotplugged)) {
631         pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
632                                    PCI_EXP_SLTCAP_HPS |
633                                    PCI_EXP_SLTCAP_HPC);
634     }
635 
636     if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
637         pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
638                                    PCI_EXP_SLTCAP_PCP);
639         pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
640                                      PCI_EXP_SLTCTL_PCC);
641         pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
642                                    PCI_EXP_SLTCTL_PCC);
643     }
644 
645     pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
646                                  PCI_EXP_SLTCTL_PIC |
647                                  PCI_EXP_SLTCTL_AIC);
648     pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCTL,
649                                PCI_EXP_SLTCTL_PWR_IND_OFF |
650                                PCI_EXP_SLTCTL_ATTN_IND_OFF);
651     pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
652                                PCI_EXP_SLTCTL_PIC |
653                                PCI_EXP_SLTCTL_AIC |
654                                PCI_EXP_SLTCTL_HPIE |
655                                PCI_EXP_SLTCTL_CCIE |
656                                PCI_EXP_SLTCTL_PDCE |
657                                PCI_EXP_SLTCTL_ABPE);
658     /* Although reading PCI_EXP_SLTCTL_EIC returns always 0,
659      * make the bit writable here in order to detect 1b is written.
660      * pcie_cap_slot_write_config() test-and-clear the bit, so
661      * this bit always returns 0 to the guest.
662      */
663     pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
664                                PCI_EXP_SLTCTL_EIC);
665 
666     pci_word_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_SLTSTA,
667                                PCI_EXP_HP_EV_SUPPORTED);
668 
669     dev->exp.hpev_notified = false;
670 
671     qbus_set_hotplug_handler(BUS(pci_bridge_get_sec_bus(PCI_BRIDGE(dev))),
672                              OBJECT(dev));
673 }
674 
675 void pcie_cap_slot_reset(PCIDevice *dev)
676 {
677     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
678     uint8_t port_type = pcie_cap_get_type(dev);
679 
680     assert(port_type == PCI_EXP_TYPE_DOWNSTREAM ||
681            port_type == PCI_EXP_TYPE_ROOT_PORT);
682 
683     PCIE_DEV_PRINTF(dev, "reset\n");
684 
685     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
686                                  PCI_EXP_SLTCTL_EIC |
687                                  PCI_EXP_SLTCTL_PIC |
688                                  PCI_EXP_SLTCTL_AIC |
689                                  PCI_EXP_SLTCTL_HPIE |
690                                  PCI_EXP_SLTCTL_CCIE |
691                                  PCI_EXP_SLTCTL_PDCE |
692                                  PCI_EXP_SLTCTL_ABPE);
693     pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
694                                PCI_EXP_SLTCTL_PWR_IND_OFF |
695                                PCI_EXP_SLTCTL_ATTN_IND_OFF);
696 
697     if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
698         /* Downstream ports enforce device number 0. */
699         bool populated = pci_bridge_get_sec_bus(PCI_BRIDGE(dev))->devices[0];
700         uint16_t pic;
701 
702         if (populated) {
703             pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
704                                          PCI_EXP_SLTCTL_PCC);
705         } else {
706             pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
707                                        PCI_EXP_SLTCTL_PCC);
708         }
709 
710         pic = populated ?
711                 PCI_EXP_SLTCTL_PWR_IND_ON : PCI_EXP_SLTCTL_PWR_IND_OFF;
712         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL, pic);
713     }
714 
715     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
716                                  PCI_EXP_SLTSTA_EIS |/* on reset,
717                                                         the lock is released */
718                                  PCI_EXP_SLTSTA_CC |
719                                  PCI_EXP_SLTSTA_PDC |
720                                  PCI_EXP_SLTSTA_ABP);
721 
722     pcie_cap_update_power(dev);
723     hotplug_event_update_event_status(dev);
724 }
725 
726 void pcie_cap_slot_get(PCIDevice *dev, uint16_t *slt_ctl, uint16_t *slt_sta)
727 {
728     uint32_t pos = dev->exp.exp_cap;
729     uint8_t *exp_cap = dev->config + pos;
730     *slt_ctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
731     *slt_sta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
732 }
733 
734 void pcie_cap_slot_write_config(PCIDevice *dev,
735                                 uint16_t old_slt_ctl, uint16_t old_slt_sta,
736                                 uint32_t addr, uint32_t val, int len)
737 {
738     uint32_t pos = dev->exp.exp_cap;
739     uint8_t *exp_cap = dev->config + pos;
740     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
741 
742     if (ranges_overlap(addr, len, pos + PCI_EXP_SLTSTA, 2)) {
743         /*
744          * Guests tend to clears all bits during init.
745          * If they clear bits that weren't set this is racy and will lose events:
746          * not a big problem for manual button presses, but a problem for us.
747          * As a work-around, detect this and revert status to what it was
748          * before the write.
749          *
750          * Note: in theory this can be detected as a duplicate button press
751          * which cancels the previous press. Does not seem to happen in
752          * practice as guests seem to only have this bug during init.
753          */
754 #define PCIE_SLOT_EVENTS (PCI_EXP_SLTSTA_ABP | PCI_EXP_SLTSTA_PFD | \
755                           PCI_EXP_SLTSTA_MRLSC | PCI_EXP_SLTSTA_PDC | \
756                           PCI_EXP_SLTSTA_CC)
757 
758         if (val & ~old_slt_sta & PCIE_SLOT_EVENTS) {
759             sltsta = (sltsta & ~PCIE_SLOT_EVENTS) | (old_slt_sta & PCIE_SLOT_EVENTS);
760             pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
761         }
762         hotplug_event_clear(dev);
763     }
764 
765     if (!ranges_overlap(addr, len, pos + PCI_EXP_SLTCTL, 2)) {
766         return;
767     }
768 
769     if (pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
770                                      PCI_EXP_SLTCTL_EIC)) {
771         sltsta ^= PCI_EXP_SLTSTA_EIS; /* toggle PCI_EXP_SLTSTA_EIS bit */
772         pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
773         PCIE_DEV_PRINTF(dev, "PCI_EXP_SLTCTL_EIC: "
774                         "sltsta -> 0x%02"PRIx16"\n",
775                         sltsta);
776     }
777 
778     /*
779      * If the slot is populated, power indicator is off and power
780      * controller is off, it is safe to detach the devices.
781      *
782      * Note: don't detach if condition was already true:
783      * this is a work around for guests that overwrite
784      * control of powered off slots before powering them on.
785      */
786     if ((sltsta & PCI_EXP_SLTSTA_PDS) && pcie_sltctl_powered_off(val) &&
787         !pcie_sltctl_powered_off(old_slt_ctl))
788     {
789         pcie_cap_slot_do_unplug(dev);
790     }
791     pcie_cap_update_power(dev);
792 
793     hotplug_event_notify(dev);
794 
795     /*
796      * 6.7.3.2 Command Completed Events
797      *
798      * Software issues a command to a hot-plug capable Downstream Port by
799      * issuing a write transaction that targets any portion of the Port’s Slot
800      * Control register. A single write to the Slot Control register is
801      * considered to be a single command, even if the write affects more than
802      * one field in the Slot Control register. In response to this transaction,
803      * the Port must carry out the requested actions and then set the
804      * associated status field for the command completed event. */
805 
806     /* Real hardware might take a while to complete requested command because
807      * physical movement would be involved like locking the electromechanical
808      * lock.  However in our case, command is completed instantaneously above,
809      * so send a command completion event right now.
810      */
811     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_CCI);
812 }
813 
814 int pcie_cap_slot_post_load(void *opaque, int version_id)
815 {
816     PCIDevice *dev = opaque;
817     hotplug_event_update_event_status(dev);
818     pcie_cap_update_power(dev);
819     return 0;
820 }
821 
822 void pcie_cap_slot_push_attention_button(PCIDevice *dev)
823 {
824     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_ABP);
825 }
826 
827 /* root control/capabilities/status. PME isn't emulated for now */
828 void pcie_cap_root_init(PCIDevice *dev)
829 {
830     pci_set_word(dev->wmask + dev->exp.exp_cap + PCI_EXP_RTCTL,
831                  PCI_EXP_RTCTL_SECEE | PCI_EXP_RTCTL_SENFEE |
832                  PCI_EXP_RTCTL_SEFEE);
833 }
834 
835 void pcie_cap_root_reset(PCIDevice *dev)
836 {
837     pci_set_word(dev->config + dev->exp.exp_cap + PCI_EXP_RTCTL, 0);
838 }
839 
840 /* function level reset(FLR) */
841 void pcie_cap_flr_init(PCIDevice *dev)
842 {
843     pci_long_test_and_set_mask(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCAP,
844                                PCI_EXP_DEVCAP_FLR);
845 
846     /* Although reading BCR_FLR returns always 0,
847      * the bit is made writable here in order to detect the 1b is written
848      * pcie_cap_flr_write_config() test-and-clear the bit, so
849      * this bit always returns 0 to the guest.
850      */
851     pci_word_test_and_set_mask(dev->wmask + dev->exp.exp_cap + PCI_EXP_DEVCTL,
852                                PCI_EXP_DEVCTL_BCR_FLR);
853 }
854 
855 void pcie_cap_flr_write_config(PCIDevice *dev,
856                                uint32_t addr, uint32_t val, int len)
857 {
858     uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
859     if (pci_get_word(devctl) & PCI_EXP_DEVCTL_BCR_FLR) {
860         /* Clear PCI_EXP_DEVCTL_BCR_FLR after invoking the reset handler
861            so the handler can detect FLR by looking at this bit. */
862         pci_device_reset(dev);
863         pci_word_test_and_clear_mask(devctl, PCI_EXP_DEVCTL_BCR_FLR);
864     }
865 }
866 
867 /* Alternative Routing-ID Interpretation (ARI)
868  * forwarding support for root and downstream ports
869  */
870 void pcie_cap_arifwd_init(PCIDevice *dev)
871 {
872     uint32_t pos = dev->exp.exp_cap;
873     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP2,
874                                PCI_EXP_DEVCAP2_ARI);
875     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL2,
876                                PCI_EXP_DEVCTL2_ARI);
877 }
878 
879 void pcie_cap_arifwd_reset(PCIDevice *dev)
880 {
881     uint8_t *devctl2 = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2;
882     pci_long_test_and_clear_mask(devctl2, PCI_EXP_DEVCTL2_ARI);
883 }
884 
885 bool pcie_cap_is_arifwd_enabled(const PCIDevice *dev)
886 {
887     if (!pci_is_express(dev)) {
888         return false;
889     }
890     if (!dev->exp.exp_cap) {
891         return false;
892     }
893 
894     return pci_get_long(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2) &
895         PCI_EXP_DEVCTL2_ARI;
896 }
897 
898 /**************************************************************************
899  * pci express extended capability list management functions
900  * uint16_t ext_cap_id (16 bit)
901  * uint8_t cap_ver (4 bit)
902  * uint16_t cap_offset (12 bit)
903  * uint16_t ext_cap_size
904  */
905 
906 /* Passing a cap_id value > 0xffff will return 0 and put end of list in prev */
907 static uint16_t pcie_find_capability_list(PCIDevice *dev, uint32_t cap_id,
908                                           uint16_t *prev_p)
909 {
910     uint16_t prev = 0;
911     uint16_t next;
912     uint32_t header = pci_get_long(dev->config + PCI_CONFIG_SPACE_SIZE);
913 
914     if (!header) {
915         /* no extended capability */
916         next = 0;
917         goto out;
918     }
919     for (next = PCI_CONFIG_SPACE_SIZE; next;
920          prev = next, next = PCI_EXT_CAP_NEXT(header)) {
921 
922         assert(next >= PCI_CONFIG_SPACE_SIZE);
923         assert(next <= PCIE_CONFIG_SPACE_SIZE - 8);
924 
925         header = pci_get_long(dev->config + next);
926         if (PCI_EXT_CAP_ID(header) == cap_id) {
927             break;
928         }
929     }
930 
931 out:
932     if (prev_p) {
933         *prev_p = prev;
934     }
935     return next;
936 }
937 
938 uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id)
939 {
940     return pcie_find_capability_list(dev, cap_id, NULL);
941 }
942 
943 static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next)
944 {
945     uint32_t header = pci_get_long(dev->config + pos);
946     assert(!(next & (PCI_EXT_CAP_ALIGN - 1)));
947     header = (header & ~PCI_EXT_CAP_NEXT_MASK) |
948         ((next << PCI_EXT_CAP_NEXT_SHIFT) & PCI_EXT_CAP_NEXT_MASK);
949     pci_set_long(dev->config + pos, header);
950 }
951 
952 /*
953  * Caller must supply valid (offset, size) such that the range wouldn't
954  * overlap with other capability or other registers.
955  * This function doesn't check it.
956  */
957 void pcie_add_capability(PCIDevice *dev,
958                          uint16_t cap_id, uint8_t cap_ver,
959                          uint16_t offset, uint16_t size)
960 {
961     assert(offset >= PCI_CONFIG_SPACE_SIZE);
962     assert(offset < (uint16_t)(offset + size));
963     assert((uint16_t)(offset + size) <= PCIE_CONFIG_SPACE_SIZE);
964     assert(size >= 8);
965     assert(pci_is_express(dev));
966 
967     if (offset != PCI_CONFIG_SPACE_SIZE) {
968         uint16_t prev;
969 
970         /*
971          * 0xffffffff is not a valid cap id (it's a 16 bit field). use
972          * internally to find the last capability in the linked list.
973          */
974         pcie_find_capability_list(dev, 0xffffffff, &prev);
975         assert(prev >= PCI_CONFIG_SPACE_SIZE);
976         pcie_ext_cap_set_next(dev, prev, offset);
977     }
978     pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, 0));
979 
980     /* Make capability read-only by default */
981     memset(dev->wmask + offset, 0, size);
982     memset(dev->w1cmask + offset, 0, size);
983     /* Check capability by default */
984     memset(dev->cmask + offset, 0xFF, size);
985 }
986 
987 /*
988  * Sync the PCIe Link Status negotiated speed and width of a bridge with the
989  * downstream device.  If downstream device is not present, re-write with the
990  * Link Capability fields.  If downstream device reports invalid width or
991  * speed, replace with minimum values (LnkSta fields are RsvdZ on VFs but such
992  * values interfere with PCIe native hotplug detecting new devices).  Limit
993  * width and speed to bridge capabilities for compatibility.  Use config_read
994  * to access the downstream device since it could be an assigned device with
995  * volatile link information.
996  */
997 void pcie_sync_bridge_lnk(PCIDevice *bridge_dev)
998 {
999     PCIBridge *br = PCI_BRIDGE(bridge_dev);
1000     PCIBus *bus = pci_bridge_get_sec_bus(br);
1001     PCIDevice *target = bus->devices[0];
1002     uint8_t *exp_cap = bridge_dev->config + bridge_dev->exp.exp_cap;
1003     uint16_t lnksta, lnkcap = pci_get_word(exp_cap + PCI_EXP_LNKCAP);
1004 
1005     if (!target || !target->exp.exp_cap) {
1006         lnksta = lnkcap;
1007     } else {
1008         lnksta = target->config_read(target,
1009                                      target->exp.exp_cap + PCI_EXP_LNKSTA,
1010                                      sizeof(lnksta));
1011 
1012         if ((lnksta & PCI_EXP_LNKSTA_NLW) > (lnkcap & PCI_EXP_LNKCAP_MLW)) {
1013             lnksta &= ~PCI_EXP_LNKSTA_NLW;
1014             lnksta |= lnkcap & PCI_EXP_LNKCAP_MLW;
1015         } else if (!(lnksta & PCI_EXP_LNKSTA_NLW)) {
1016             lnksta |= QEMU_PCI_EXP_LNKSTA_NLW(QEMU_PCI_EXP_LNK_X1);
1017         }
1018 
1019         if ((lnksta & PCI_EXP_LNKSTA_CLS) > (lnkcap & PCI_EXP_LNKCAP_SLS)) {
1020             lnksta &= ~PCI_EXP_LNKSTA_CLS;
1021             lnksta |= lnkcap & PCI_EXP_LNKCAP_SLS;
1022         } else if (!(lnksta & PCI_EXP_LNKSTA_CLS)) {
1023             lnksta |= QEMU_PCI_EXP_LNKSTA_CLS(QEMU_PCI_EXP_LNK_2_5GT);
1024         }
1025     }
1026 
1027     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKSTA,
1028                                  PCI_EXP_LNKSTA_CLS | PCI_EXP_LNKSTA_NLW);
1029     pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA, lnksta &
1030                                (PCI_EXP_LNKSTA_CLS | PCI_EXP_LNKSTA_NLW));
1031 }
1032 
1033 /**************************************************************************
1034  * pci express extended capability helper functions
1035  */
1036 
1037 /* ARI */
1038 void pcie_ari_init(PCIDevice *dev, uint16_t offset, uint16_t nextfn)
1039 {
1040     pcie_add_capability(dev, PCI_EXT_CAP_ID_ARI, PCI_ARI_VER,
1041                         offset, PCI_ARI_SIZEOF);
1042     pci_set_long(dev->config + offset + PCI_ARI_CAP, (nextfn & 0xff) << 8);
1043 }
1044 
1045 void pcie_dev_ser_num_init(PCIDevice *dev, uint16_t offset, uint64_t ser_num)
1046 {
1047     static const int pci_dsn_ver = 1;
1048     static const int pci_dsn_cap = 4;
1049 
1050     pcie_add_capability(dev, PCI_EXT_CAP_ID_DSN, pci_dsn_ver, offset,
1051                         PCI_EXT_CAP_DSN_SIZEOF);
1052     pci_set_quad(dev->config + offset + pci_dsn_cap, ser_num);
1053 }
1054 
1055 void pcie_ats_init(PCIDevice *dev, uint16_t offset, bool aligned)
1056 {
1057     pcie_add_capability(dev, PCI_EXT_CAP_ID_ATS, 0x1,
1058                         offset, PCI_EXT_CAP_ATS_SIZEOF);
1059 
1060     dev->exp.ats_cap = offset;
1061 
1062     /* Invalidate Queue Depth 0 */
1063     if (aligned) {
1064         pci_set_word(dev->config + offset + PCI_ATS_CAP,
1065                      PCI_ATS_CAP_PAGE_ALIGNED);
1066     }
1067     /* STU 0, Disabled by default */
1068     pci_set_word(dev->config + offset + PCI_ATS_CTRL, 0);
1069 
1070     pci_set_word(dev->wmask + dev->exp.ats_cap + PCI_ATS_CTRL, 0x800f);
1071 }
1072 
1073 /* ACS (Access Control Services) */
1074 void pcie_acs_init(PCIDevice *dev, uint16_t offset)
1075 {
1076     bool is_downstream = pci_is_express_downstream_port(dev);
1077     uint16_t cap_bits = 0;
1078 
1079     /* For endpoints, only multifunction devs may have an ACS capability: */
1080     assert(is_downstream ||
1081            (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) ||
1082            PCI_FUNC(dev->devfn));
1083 
1084     pcie_add_capability(dev, PCI_EXT_CAP_ID_ACS, PCI_ACS_VER, offset,
1085                         PCI_ACS_SIZEOF);
1086     dev->exp.acs_cap = offset;
1087 
1088     if (is_downstream) {
1089         /*
1090          * Downstream ports must implement SV, TB, RR, CR, UF, and DT (with
1091          * caveats on the latter four that we ignore for simplicity).
1092          * Endpoints may also implement a subset of ACS capabilities,
1093          * but these are optional if the endpoint does not support
1094          * peer-to-peer between functions and thus omitted here.
1095          */
1096         cap_bits = PCI_ACS_SV | PCI_ACS_TB | PCI_ACS_RR |
1097             PCI_ACS_CR | PCI_ACS_UF | PCI_ACS_DT;
1098     }
1099 
1100     pci_set_word(dev->config + offset + PCI_ACS_CAP, cap_bits);
1101     pci_set_word(dev->wmask + offset + PCI_ACS_CTRL, cap_bits);
1102 }
1103 
1104 void pcie_acs_reset(PCIDevice *dev)
1105 {
1106     if (dev->exp.acs_cap) {
1107         pci_set_word(dev->config + dev->exp.acs_cap + PCI_ACS_CTRL, 0);
1108     }
1109 }
1110