xref: /openbmc/qemu/hw/pci/pcie.c (revision 6f03770d)
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 
43 /***************************************************************************
44  * pci express capability helper functions
45  */
46 
47 static void
48 pcie_cap_v1_fill(PCIDevice *dev, uint8_t port, uint8_t type, uint8_t version)
49 {
50     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
51     uint8_t *cmask = dev->cmask + dev->exp.exp_cap;
52 
53     /* capability register
54     interrupt message number defaults to 0 */
55     pci_set_word(exp_cap + PCI_EXP_FLAGS,
56                  ((type << PCI_EXP_FLAGS_TYPE_SHIFT) & PCI_EXP_FLAGS_TYPE) |
57                  version);
58 
59     /* device capability register
60      * table 7-12:
61      * roll based error reporting bit must be set by all
62      * Functions conforming to the ECN, PCI Express Base
63      * Specification, Revision 1.1., or subsequent PCI Express Base
64      * Specification revisions.
65      */
66     pci_set_long(exp_cap + PCI_EXP_DEVCAP, PCI_EXP_DEVCAP_RBER);
67 
68     pci_set_long(exp_cap + PCI_EXP_LNKCAP,
69                  (port << PCI_EXP_LNKCAP_PN_SHIFT) |
70                  PCI_EXP_LNKCAP_ASPMS_0S |
71                  QEMU_PCI_EXP_LNKCAP_MLW(QEMU_PCI_EXP_LNK_X1) |
72                  QEMU_PCI_EXP_LNKCAP_MLS(QEMU_PCI_EXP_LNK_2_5GT));
73 
74     pci_set_word(exp_cap + PCI_EXP_LNKSTA,
75                  QEMU_PCI_EXP_LNKSTA_NLW(QEMU_PCI_EXP_LNK_X1) |
76                  QEMU_PCI_EXP_LNKSTA_CLS(QEMU_PCI_EXP_LNK_2_5GT));
77 
78     /* We changed link status bits over time, and changing them across
79      * migrations is generally fine as hardware changes them too.
80      * Let's not bother checking.
81      */
82     pci_set_word(cmask + PCI_EXP_LNKSTA, 0);
83 }
84 
85 static void pcie_cap_fill_slot_lnk(PCIDevice *dev)
86 {
87     PCIESlot *s = (PCIESlot *)object_dynamic_cast(OBJECT(dev), TYPE_PCIE_SLOT);
88     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
89 
90     /* Skip anything that isn't a PCIESlot */
91     if (!s) {
92         return;
93     }
94 
95     /* Clear and fill LNKCAP from what was configured above */
96     pci_long_test_and_clear_mask(exp_cap + PCI_EXP_LNKCAP,
97                                  PCI_EXP_LNKCAP_MLW | PCI_EXP_LNKCAP_SLS);
98     pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP,
99                                QEMU_PCI_EXP_LNKCAP_MLW(s->width) |
100                                QEMU_PCI_EXP_LNKCAP_MLS(s->speed));
101 
102     /*
103      * Link bandwidth notification is required for all root ports and
104      * downstream ports supporting links wider than x1 or multiple link
105      * speeds.
106      */
107     if (s->width > QEMU_PCI_EXP_LNK_X1 ||
108         s->speed > QEMU_PCI_EXP_LNK_2_5GT) {
109         pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP,
110                                    PCI_EXP_LNKCAP_LBNC);
111     }
112 
113     if (s->speed > QEMU_PCI_EXP_LNK_2_5GT) {
114         /*
115          * Hot-plug capable downstream ports and downstream ports supporting
116          * link speeds greater than 5GT/s must hardwire PCI_EXP_LNKCAP_DLLLARC
117          * to 1b.  PCI_EXP_LNKCAP_DLLLARC implies PCI_EXP_LNKSTA_DLLLA, which
118          * we also hardwire to 1b here.  2.5GT/s hot-plug slots should also
119          * technically implement this, but it's not done here for compatibility.
120          */
121         pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP,
122                                    PCI_EXP_LNKCAP_DLLLARC);
123         /* the PCI_EXP_LNKSTA_DLLLA will be set in the hotplug function */
124 
125         /*
126          * Target Link Speed defaults to the highest link speed supported by
127          * the component.  2.5GT/s devices are permitted to hardwire to zero.
128          */
129         pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKCTL2,
130                                      PCI_EXP_LNKCTL2_TLS);
131         pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKCTL2,
132                                    QEMU_PCI_EXP_LNKCAP_MLS(s->speed) &
133                                    PCI_EXP_LNKCTL2_TLS);
134     }
135 
136     /*
137      * 2.5 & 5.0GT/s can be fully described by LNKCAP, but 8.0GT/s is
138      * actually a reference to the highest bit supported in this register.
139      * We assume the device supports all link speeds.
140      */
141     if (s->speed > QEMU_PCI_EXP_LNK_5GT) {
142         pci_long_test_and_clear_mask(exp_cap + PCI_EXP_LNKCAP2, ~0U);
143         pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP2,
144                                    PCI_EXP_LNKCAP2_SLS_2_5GB |
145                                    PCI_EXP_LNKCAP2_SLS_5_0GB |
146                                    PCI_EXP_LNKCAP2_SLS_8_0GB);
147         if (s->speed > QEMU_PCI_EXP_LNK_8GT) {
148             pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP2,
149                                        PCI_EXP_LNKCAP2_SLS_16_0GB);
150         }
151     }
152 }
153 
154 int pcie_cap_init(PCIDevice *dev, uint8_t offset,
155                   uint8_t type, uint8_t port,
156                   Error **errp)
157 {
158     /* PCIe cap v2 init */
159     int pos;
160     uint8_t *exp_cap;
161 
162     assert(pci_is_express(dev));
163 
164     pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
165                              PCI_EXP_VER2_SIZEOF, errp);
166     if (pos < 0) {
167         return pos;
168     }
169     dev->exp.exp_cap = pos;
170     exp_cap = dev->config + pos;
171 
172     /* Filling values common with v1 */
173     pcie_cap_v1_fill(dev, port, type, PCI_EXP_FLAGS_VER2);
174 
175     /* Fill link speed and width options */
176     pcie_cap_fill_slot_lnk(dev);
177 
178     /* Filling v2 specific values */
179     pci_set_long(exp_cap + PCI_EXP_DEVCAP2,
180                  PCI_EXP_DEVCAP2_EFF | PCI_EXP_DEVCAP2_EETLPP);
181 
182     pci_set_word(dev->wmask + pos + PCI_EXP_DEVCTL2, PCI_EXP_DEVCTL2_EETLPPB);
183 
184     if (dev->cap_present & QEMU_PCIE_EXTCAP_INIT) {
185         /* read-only to behave like a 'NULL' Extended Capability Header */
186         pci_set_long(dev->wmask + PCI_CONFIG_SPACE_SIZE, 0);
187     }
188 
189     return pos;
190 }
191 
192 int pcie_cap_v1_init(PCIDevice *dev, uint8_t offset, uint8_t type,
193                      uint8_t port)
194 {
195     /* PCIe cap v1 init */
196     int pos;
197     Error *local_err = NULL;
198 
199     assert(pci_is_express(dev));
200 
201     pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
202                              PCI_EXP_VER1_SIZEOF, &local_err);
203     if (pos < 0) {
204         error_report_err(local_err);
205         return pos;
206     }
207     dev->exp.exp_cap = pos;
208 
209     pcie_cap_v1_fill(dev, port, type, PCI_EXP_FLAGS_VER1);
210 
211     return pos;
212 }
213 
214 static int
215 pcie_endpoint_cap_common_init(PCIDevice *dev, uint8_t offset, uint8_t cap_size)
216 {
217     uint8_t type = PCI_EXP_TYPE_ENDPOINT;
218     Error *local_err = NULL;
219     int ret;
220 
221     /*
222      * Windows guests will report Code 10, device cannot start, if
223      * a regular Endpoint type is exposed on a root complex.  These
224      * should instead be Root Complex Integrated Endpoints.
225      */
226     if (pci_bus_is_express(pci_get_bus(dev))
227         && pci_bus_is_root(pci_get_bus(dev))) {
228         type = PCI_EXP_TYPE_RC_END;
229     }
230 
231     if (cap_size == PCI_EXP_VER1_SIZEOF) {
232         return pcie_cap_v1_init(dev, offset, type, 0);
233     } else {
234         ret = pcie_cap_init(dev, offset, type, 0, &local_err);
235 
236         if (ret < 0) {
237             error_report_err(local_err);
238         }
239 
240         return ret;
241     }
242 }
243 
244 int pcie_endpoint_cap_init(PCIDevice *dev, uint8_t offset)
245 {
246     return pcie_endpoint_cap_common_init(dev, offset, PCI_EXP_VER2_SIZEOF);
247 }
248 
249 int pcie_endpoint_cap_v1_init(PCIDevice *dev, uint8_t offset)
250 {
251     return pcie_endpoint_cap_common_init(dev, offset, PCI_EXP_VER1_SIZEOF);
252 }
253 
254 void pcie_cap_exit(PCIDevice *dev)
255 {
256     pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER2_SIZEOF);
257 }
258 
259 void pcie_cap_v1_exit(PCIDevice *dev)
260 {
261     pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER1_SIZEOF);
262 }
263 
264 uint8_t pcie_cap_get_type(const PCIDevice *dev)
265 {
266     uint32_t pos = dev->exp.exp_cap;
267     assert(pos > 0);
268     return (pci_get_word(dev->config + pos + PCI_EXP_FLAGS) &
269             PCI_EXP_FLAGS_TYPE) >> PCI_EXP_FLAGS_TYPE_SHIFT;
270 }
271 
272 /* MSI/MSI-X */
273 /* pci express interrupt message number */
274 /* 7.8.2 PCI Express Capabilities Register: Interrupt Message Number */
275 void pcie_cap_flags_set_vector(PCIDevice *dev, uint8_t vector)
276 {
277     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
278     assert(vector < 32);
279     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_FLAGS, PCI_EXP_FLAGS_IRQ);
280     pci_word_test_and_set_mask(exp_cap + PCI_EXP_FLAGS,
281                                vector << PCI_EXP_FLAGS_IRQ_SHIFT);
282 }
283 
284 uint8_t pcie_cap_flags_get_vector(PCIDevice *dev)
285 {
286     return (pci_get_word(dev->config + dev->exp.exp_cap + PCI_EXP_FLAGS) &
287             PCI_EXP_FLAGS_IRQ) >> PCI_EXP_FLAGS_IRQ_SHIFT;
288 }
289 
290 void pcie_cap_deverr_init(PCIDevice *dev)
291 {
292     uint32_t pos = dev->exp.exp_cap;
293     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP,
294                                PCI_EXP_DEVCAP_RBER);
295     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL,
296                                PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
297                                PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
298     pci_long_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_DEVSTA,
299                                PCI_EXP_DEVSTA_CED | PCI_EXP_DEVSTA_NFED |
300                                PCI_EXP_DEVSTA_FED | PCI_EXP_DEVSTA_URD);
301 }
302 
303 void pcie_cap_deverr_reset(PCIDevice *dev)
304 {
305     uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
306     pci_long_test_and_clear_mask(devctl,
307                                  PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
308                                  PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
309 }
310 
311 void pcie_cap_lnkctl_init(PCIDevice *dev)
312 {
313     uint32_t pos = dev->exp.exp_cap;
314     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_LNKCTL,
315                                PCI_EXP_LNKCTL_CCC | PCI_EXP_LNKCTL_ES);
316 }
317 
318 void pcie_cap_lnkctl_reset(PCIDevice *dev)
319 {
320     uint8_t *lnkctl = dev->config + dev->exp.exp_cap + PCI_EXP_LNKCTL;
321     pci_long_test_and_clear_mask(lnkctl,
322                                  PCI_EXP_LNKCTL_CCC | PCI_EXP_LNKCTL_ES);
323 }
324 
325 static void hotplug_event_update_event_status(PCIDevice *dev)
326 {
327     uint32_t pos = dev->exp.exp_cap;
328     uint8_t *exp_cap = dev->config + pos;
329     uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
330     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
331 
332     dev->exp.hpev_notified = (sltctl & PCI_EXP_SLTCTL_HPIE) &&
333         (sltsta & sltctl & PCI_EXP_HP_EV_SUPPORTED);
334 }
335 
336 static void hotplug_event_notify(PCIDevice *dev)
337 {
338     bool prev = dev->exp.hpev_notified;
339 
340     hotplug_event_update_event_status(dev);
341 
342     if (prev == dev->exp.hpev_notified) {
343         return;
344     }
345 
346     /* Note: the logic above does not take into account whether interrupts
347      * are masked. The result is that interrupt will be sent when it is
348      * subsequently unmasked. This appears to be legal: Section 6.7.3.4:
349      * The Port may optionally send an MSI when there are hot-plug events that
350      * occur while interrupt generation is disabled, and interrupt generation is
351      * subsequently enabled. */
352     if (msix_enabled(dev)) {
353         msix_notify(dev, pcie_cap_flags_get_vector(dev));
354     } else if (msi_enabled(dev)) {
355         msi_notify(dev, pcie_cap_flags_get_vector(dev));
356     } else {
357         pci_set_irq(dev, dev->exp.hpev_notified);
358     }
359 }
360 
361 static void hotplug_event_clear(PCIDevice *dev)
362 {
363     hotplug_event_update_event_status(dev);
364     if (!msix_enabled(dev) && !msi_enabled(dev) && !dev->exp.hpev_notified) {
365         pci_irq_deassert(dev);
366     }
367 }
368 
369 /*
370  * A PCI Express Hot-Plug Event has occurred, so update slot status register
371  * and notify OS of the event if necessary.
372  *
373  * 6.7.3 PCI Express Hot-Plug Events
374  * 6.7.3.4 Software Notification of Hot-Plug Events
375  */
376 static void pcie_cap_slot_event(PCIDevice *dev, PCIExpressHotPlugEvent event)
377 {
378     /* Minor optimization: if nothing changed - no event is needed. */
379     if (pci_word_test_and_set_mask(dev->config + dev->exp.exp_cap +
380                                    PCI_EXP_SLTSTA, event) == event) {
381         return;
382     }
383     hotplug_event_notify(dev);
384 }
385 
386 static void pcie_cap_slot_plug_common(PCIDevice *hotplug_dev, DeviceState *dev,
387                                       Error **errp)
388 {
389     uint8_t *exp_cap = hotplug_dev->config + hotplug_dev->exp.exp_cap;
390     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
391 
392     PCIE_DEV_PRINTF(PCI_DEVICE(dev), "hotplug state: 0x%x\n", sltsta);
393     if (sltsta & PCI_EXP_SLTSTA_EIS) {
394         /* the slot is electromechanically locked.
395          * This error is propagated up to qdev and then to HMP/QMP.
396          */
397         error_setg_errno(errp, EBUSY, "slot is electromechanically locked");
398     }
399 }
400 
401 void pcie_cap_slot_pre_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
402                                Error **errp)
403 {
404     PCIDevice *hotplug_pdev = PCI_DEVICE(hotplug_dev);
405     uint8_t *exp_cap = hotplug_pdev->config + hotplug_pdev->exp.exp_cap;
406     uint32_t sltcap = pci_get_word(exp_cap + PCI_EXP_SLTCAP);
407 
408     /* Check if hot-plug is disabled on the slot */
409     if (dev->hotplugged && (sltcap & PCI_EXP_SLTCAP_HPC) == 0) {
410         error_setg(errp, "Hot-plug failed: unsupported by the port device '%s'",
411                          DEVICE(hotplug_pdev)->id);
412         return;
413     }
414 
415     pcie_cap_slot_plug_common(PCI_DEVICE(hotplug_dev), dev, errp);
416 }
417 
418 void pcie_cap_slot_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
419                            Error **errp)
420 {
421     PCIDevice *hotplug_pdev = PCI_DEVICE(hotplug_dev);
422     uint8_t *exp_cap = hotplug_pdev->config + hotplug_pdev->exp.exp_cap;
423     PCIDevice *pci_dev = PCI_DEVICE(dev);
424     uint32_t lnkcap = pci_get_long(exp_cap + PCI_EXP_LNKCAP);
425 
426     /* Don't send event when device is enabled during qemu machine creation:
427      * it is present on boot, no hotplug event is necessary. We do send an
428      * event when the device is disabled later. */
429     if (!dev->hotplugged) {
430         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
431                                    PCI_EXP_SLTSTA_PDS);
432         if (pci_dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA ||
433             (lnkcap & PCI_EXP_LNKCAP_DLLLARC)) {
434             pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA,
435                                        PCI_EXP_LNKSTA_DLLLA);
436         }
437         return;
438     }
439 
440     /* To enable multifunction hot-plug, we just ensure the function
441      * 0 added last. When function 0 is added, we set the sltsta and
442      * inform OS via event notification.
443      */
444     if (pci_get_function_0(pci_dev)) {
445         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
446                                    PCI_EXP_SLTSTA_PDS);
447         if (pci_dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA ||
448             (lnkcap & PCI_EXP_LNKCAP_DLLLARC)) {
449             pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA,
450                                        PCI_EXP_LNKSTA_DLLLA);
451         }
452         pcie_cap_slot_event(hotplug_pdev,
453                             PCI_EXP_HP_EV_PDC | PCI_EXP_HP_EV_ABP);
454     }
455 }
456 
457 void pcie_cap_slot_unplug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
458                              Error **errp)
459 {
460     qdev_unrealize(dev);
461 }
462 
463 static void pcie_unplug_device(PCIBus *bus, PCIDevice *dev, void *opaque)
464 {
465     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(DEVICE(dev));
466 
467     if (dev->partially_hotplugged) {
468         dev->qdev.pending_deleted_event = false;
469         return;
470     }
471     hotplug_handler_unplug(hotplug_ctrl, DEVICE(dev), &error_abort);
472     object_unparent(OBJECT(dev));
473 }
474 
475 void pcie_cap_slot_unplug_request_cb(HotplugHandler *hotplug_dev,
476                                      DeviceState *dev, Error **errp)
477 {
478     Error *local_err = NULL;
479     PCIDevice *pci_dev = PCI_DEVICE(dev);
480     PCIBus *bus = pci_get_bus(pci_dev);
481     PCIDevice *hotplug_pdev = PCI_DEVICE(hotplug_dev);
482     uint8_t *exp_cap = hotplug_pdev->config + hotplug_pdev->exp.exp_cap;
483     uint32_t sltcap = pci_get_word(exp_cap + PCI_EXP_SLTCAP);
484 
485     /* Check if hot-unplug is disabled on the slot */
486     if ((sltcap & PCI_EXP_SLTCAP_HPC) == 0) {
487         error_setg(errp, "Hot-unplug failed: "
488                          "unsupported by the port device '%s'",
489                          DEVICE(hotplug_pdev)->id);
490         return;
491     }
492 
493     pcie_cap_slot_plug_common(hotplug_pdev, dev, &local_err);
494     if (local_err) {
495         error_propagate(errp, local_err);
496         return;
497     }
498 
499     dev->pending_deleted_event = true;
500 
501     /* In case user cancel the operation of multi-function hot-add,
502      * remove the function that is unexposed to guest individually,
503      * without interaction with guest.
504      */
505     if (pci_dev->devfn &&
506         !bus->devices[0]) {
507         pcie_unplug_device(bus, pci_dev, NULL);
508 
509         return;
510     }
511 
512     pcie_cap_slot_push_attention_button(hotplug_pdev);
513 }
514 
515 /* pci express slot for pci express root/downstream port
516    PCI express capability slot registers */
517 void pcie_cap_slot_init(PCIDevice *dev, PCIESlot *s)
518 {
519     uint32_t pos = dev->exp.exp_cap;
520 
521     pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_FLAGS,
522                                PCI_EXP_FLAGS_SLOT);
523 
524     pci_long_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCAP,
525                                  ~PCI_EXP_SLTCAP_PSN);
526     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
527                                (s->slot << PCI_EXP_SLTCAP_PSN_SHIFT) |
528                                PCI_EXP_SLTCAP_EIP |
529                                PCI_EXP_SLTCAP_PIP |
530                                PCI_EXP_SLTCAP_AIP |
531                                PCI_EXP_SLTCAP_ABP);
532     if (s->hotplug) {
533         pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
534                                    PCI_EXP_SLTCAP_HPS |
535                                    PCI_EXP_SLTCAP_HPC);
536     }
537 
538     if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
539         pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
540                                    PCI_EXP_SLTCAP_PCP);
541         pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
542                                      PCI_EXP_SLTCTL_PCC);
543         pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
544                                    PCI_EXP_SLTCTL_PCC);
545     }
546 
547     pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
548                                  PCI_EXP_SLTCTL_PIC |
549                                  PCI_EXP_SLTCTL_AIC);
550     pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCTL,
551                                PCI_EXP_SLTCTL_PIC_OFF |
552                                PCI_EXP_SLTCTL_AIC_OFF);
553     pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
554                                PCI_EXP_SLTCTL_PIC |
555                                PCI_EXP_SLTCTL_AIC |
556                                PCI_EXP_SLTCTL_HPIE |
557                                PCI_EXP_SLTCTL_CCIE |
558                                PCI_EXP_SLTCTL_PDCE |
559                                PCI_EXP_SLTCTL_ABPE);
560     /* Although reading PCI_EXP_SLTCTL_EIC returns always 0,
561      * make the bit writable here in order to detect 1b is written.
562      * pcie_cap_slot_write_config() test-and-clear the bit, so
563      * this bit always returns 0 to the guest.
564      */
565     pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
566                                PCI_EXP_SLTCTL_EIC);
567 
568     pci_word_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_SLTSTA,
569                                PCI_EXP_HP_EV_SUPPORTED);
570 
571     dev->exp.hpev_notified = false;
572 
573     qbus_set_hotplug_handler(BUS(pci_bridge_get_sec_bus(PCI_BRIDGE(dev))),
574                              OBJECT(dev));
575 }
576 
577 void pcie_cap_slot_reset(PCIDevice *dev)
578 {
579     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
580     uint8_t port_type = pcie_cap_get_type(dev);
581 
582     assert(port_type == PCI_EXP_TYPE_DOWNSTREAM ||
583            port_type == PCI_EXP_TYPE_ROOT_PORT);
584 
585     PCIE_DEV_PRINTF(dev, "reset\n");
586 
587     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
588                                  PCI_EXP_SLTCTL_EIC |
589                                  PCI_EXP_SLTCTL_PIC |
590                                  PCI_EXP_SLTCTL_AIC |
591                                  PCI_EXP_SLTCTL_HPIE |
592                                  PCI_EXP_SLTCTL_CCIE |
593                                  PCI_EXP_SLTCTL_PDCE |
594                                  PCI_EXP_SLTCTL_ABPE);
595     pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
596                                PCI_EXP_SLTCTL_AIC_OFF);
597 
598     if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
599         /* Downstream ports enforce device number 0. */
600         bool populated = pci_bridge_get_sec_bus(PCI_BRIDGE(dev))->devices[0];
601         uint16_t pic;
602 
603         if (populated) {
604             pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
605                                          PCI_EXP_SLTCTL_PCC);
606         } else {
607             pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
608                                        PCI_EXP_SLTCTL_PCC);
609         }
610 
611         pic = populated ? PCI_EXP_SLTCTL_PIC_ON : PCI_EXP_SLTCTL_PIC_OFF;
612         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL, pic);
613     }
614 
615     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
616                                  PCI_EXP_SLTSTA_EIS |/* on reset,
617                                                         the lock is released */
618                                  PCI_EXP_SLTSTA_CC |
619                                  PCI_EXP_SLTSTA_PDC |
620                                  PCI_EXP_SLTSTA_ABP);
621 
622     hotplug_event_update_event_status(dev);
623 }
624 
625 void pcie_cap_slot_get(PCIDevice *dev, uint16_t *slt_ctl, uint16_t *slt_sta)
626 {
627     uint32_t pos = dev->exp.exp_cap;
628     uint8_t *exp_cap = dev->config + pos;
629     *slt_ctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
630     *slt_sta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
631 }
632 
633 void pcie_cap_slot_write_config(PCIDevice *dev,
634                                 uint16_t old_slt_ctl, uint16_t old_slt_sta,
635                                 uint32_t addr, uint32_t val, int len)
636 {
637     uint32_t pos = dev->exp.exp_cap;
638     uint8_t *exp_cap = dev->config + pos;
639     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
640     uint32_t lnkcap = pci_get_long(exp_cap + PCI_EXP_LNKCAP);
641 
642     if (ranges_overlap(addr, len, pos + PCI_EXP_SLTSTA, 2)) {
643         /*
644          * Guests tend to clears all bits during init.
645          * If they clear bits that weren't set this is racy and will lose events:
646          * not a big problem for manual button presses, but a problem for us.
647          * As a work-around, detect this and revert status to what it was
648          * before the write.
649          *
650          * Note: in theory this can be detected as a duplicate button press
651          * which cancels the previous press. Does not seem to happen in
652          * practice as guests seem to only have this bug during init.
653          */
654 #define PCIE_SLOT_EVENTS (PCI_EXP_SLTSTA_ABP | PCI_EXP_SLTSTA_PFD | \
655                           PCI_EXP_SLTSTA_MRLSC | PCI_EXP_SLTSTA_PDC | \
656                           PCI_EXP_SLTSTA_CC)
657 
658         if (val & ~old_slt_sta & PCIE_SLOT_EVENTS) {
659             sltsta = (sltsta & ~PCIE_SLOT_EVENTS) | (old_slt_sta & PCIE_SLOT_EVENTS);
660             pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
661         }
662         hotplug_event_clear(dev);
663     }
664 
665     if (!ranges_overlap(addr, len, pos + PCI_EXP_SLTCTL, 2)) {
666         return;
667     }
668 
669     if (pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
670                                      PCI_EXP_SLTCTL_EIC)) {
671         sltsta ^= PCI_EXP_SLTSTA_EIS; /* toggle PCI_EXP_SLTSTA_EIS bit */
672         pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
673         PCIE_DEV_PRINTF(dev, "PCI_EXP_SLTCTL_EIC: "
674                         "sltsta -> 0x%02"PRIx16"\n",
675                         sltsta);
676     }
677 
678     /*
679      * If the slot is populated, power indicator is off and power
680      * controller is off, it is safe to detach the devices.
681      *
682      * Note: don't detach if condition was already true:
683      * this is a work around for guests that overwrite
684      * control of powered off slots before powering them on.
685      */
686     if ((sltsta & PCI_EXP_SLTSTA_PDS) && (val & PCI_EXP_SLTCTL_PCC) &&
687         (val & PCI_EXP_SLTCTL_PIC_OFF) == PCI_EXP_SLTCTL_PIC_OFF &&
688         (!(old_slt_ctl & PCI_EXP_SLTCTL_PCC) ||
689         (old_slt_ctl & PCI_EXP_SLTCTL_PIC_OFF) != PCI_EXP_SLTCTL_PIC_OFF)) {
690         PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
691         pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
692                             pcie_unplug_device, NULL);
693 
694         pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
695                                      PCI_EXP_SLTSTA_PDS);
696         if (dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA ||
697             (lnkcap & PCI_EXP_LNKCAP_DLLLARC)) {
698             pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKSTA,
699                                          PCI_EXP_LNKSTA_DLLLA);
700         }
701         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
702                                        PCI_EXP_SLTSTA_PDC);
703     }
704 
705     hotplug_event_notify(dev);
706 
707     /*
708      * 6.7.3.2 Command Completed Events
709      *
710      * Software issues a command to a hot-plug capable Downstream Port by
711      * issuing a write transaction that targets any portion of the Port’s Slot
712      * Control register. A single write to the Slot Control register is
713      * considered to be a single command, even if the write affects more than
714      * one field in the Slot Control register. In response to this transaction,
715      * the Port must carry out the requested actions and then set the
716      * associated status field for the command completed event. */
717 
718     /* Real hardware might take a while to complete requested command because
719      * physical movement would be involved like locking the electromechanical
720      * lock.  However in our case, command is completed instantaneously above,
721      * so send a command completion event right now.
722      */
723     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_CCI);
724 }
725 
726 int pcie_cap_slot_post_load(void *opaque, int version_id)
727 {
728     PCIDevice *dev = opaque;
729     hotplug_event_update_event_status(dev);
730     return 0;
731 }
732 
733 void pcie_cap_slot_push_attention_button(PCIDevice *dev)
734 {
735     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_ABP);
736 }
737 
738 /* root control/capabilities/status. PME isn't emulated for now */
739 void pcie_cap_root_init(PCIDevice *dev)
740 {
741     pci_set_word(dev->wmask + dev->exp.exp_cap + PCI_EXP_RTCTL,
742                  PCI_EXP_RTCTL_SECEE | PCI_EXP_RTCTL_SENFEE |
743                  PCI_EXP_RTCTL_SEFEE);
744 }
745 
746 void pcie_cap_root_reset(PCIDevice *dev)
747 {
748     pci_set_word(dev->config + dev->exp.exp_cap + PCI_EXP_RTCTL, 0);
749 }
750 
751 /* function level reset(FLR) */
752 void pcie_cap_flr_init(PCIDevice *dev)
753 {
754     pci_long_test_and_set_mask(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCAP,
755                                PCI_EXP_DEVCAP_FLR);
756 
757     /* Although reading BCR_FLR returns always 0,
758      * the bit is made writable here in order to detect the 1b is written
759      * pcie_cap_flr_write_config() test-and-clear the bit, so
760      * this bit always returns 0 to the guest.
761      */
762     pci_word_test_and_set_mask(dev->wmask + dev->exp.exp_cap + PCI_EXP_DEVCTL,
763                                PCI_EXP_DEVCTL_BCR_FLR);
764 }
765 
766 void pcie_cap_flr_write_config(PCIDevice *dev,
767                                uint32_t addr, uint32_t val, int len)
768 {
769     uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
770     if (pci_get_word(devctl) & PCI_EXP_DEVCTL_BCR_FLR) {
771         /* Clear PCI_EXP_DEVCTL_BCR_FLR after invoking the reset handler
772            so the handler can detect FLR by looking at this bit. */
773         pci_device_reset(dev);
774         pci_word_test_and_clear_mask(devctl, PCI_EXP_DEVCTL_BCR_FLR);
775     }
776 }
777 
778 /* Alternative Routing-ID Interpretation (ARI)
779  * forwarding support for root and downstream ports
780  */
781 void pcie_cap_arifwd_init(PCIDevice *dev)
782 {
783     uint32_t pos = dev->exp.exp_cap;
784     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP2,
785                                PCI_EXP_DEVCAP2_ARI);
786     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL2,
787                                PCI_EXP_DEVCTL2_ARI);
788 }
789 
790 void pcie_cap_arifwd_reset(PCIDevice *dev)
791 {
792     uint8_t *devctl2 = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2;
793     pci_long_test_and_clear_mask(devctl2, PCI_EXP_DEVCTL2_ARI);
794 }
795 
796 bool pcie_cap_is_arifwd_enabled(const PCIDevice *dev)
797 {
798     if (!pci_is_express(dev)) {
799         return false;
800     }
801     if (!dev->exp.exp_cap) {
802         return false;
803     }
804 
805     return pci_get_long(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2) &
806         PCI_EXP_DEVCTL2_ARI;
807 }
808 
809 /**************************************************************************
810  * pci express extended capability list management functions
811  * uint16_t ext_cap_id (16 bit)
812  * uint8_t cap_ver (4 bit)
813  * uint16_t cap_offset (12 bit)
814  * uint16_t ext_cap_size
815  */
816 
817 /* Passing a cap_id value > 0xffff will return 0 and put end of list in prev */
818 static uint16_t pcie_find_capability_list(PCIDevice *dev, uint32_t cap_id,
819                                           uint16_t *prev_p)
820 {
821     uint16_t prev = 0;
822     uint16_t next;
823     uint32_t header = pci_get_long(dev->config + PCI_CONFIG_SPACE_SIZE);
824 
825     if (!header) {
826         /* no extended capability */
827         next = 0;
828         goto out;
829     }
830     for (next = PCI_CONFIG_SPACE_SIZE; next;
831          prev = next, next = PCI_EXT_CAP_NEXT(header)) {
832 
833         assert(next >= PCI_CONFIG_SPACE_SIZE);
834         assert(next <= PCIE_CONFIG_SPACE_SIZE - 8);
835 
836         header = pci_get_long(dev->config + next);
837         if (PCI_EXT_CAP_ID(header) == cap_id) {
838             break;
839         }
840     }
841 
842 out:
843     if (prev_p) {
844         *prev_p = prev;
845     }
846     return next;
847 }
848 
849 uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id)
850 {
851     return pcie_find_capability_list(dev, cap_id, NULL);
852 }
853 
854 static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next)
855 {
856     uint32_t header = pci_get_long(dev->config + pos);
857     assert(!(next & (PCI_EXT_CAP_ALIGN - 1)));
858     header = (header & ~PCI_EXT_CAP_NEXT_MASK) |
859         ((next << PCI_EXT_CAP_NEXT_SHIFT) & PCI_EXT_CAP_NEXT_MASK);
860     pci_set_long(dev->config + pos, header);
861 }
862 
863 /*
864  * Caller must supply valid (offset, size) such that the range wouldn't
865  * overlap with other capability or other registers.
866  * This function doesn't check it.
867  */
868 void pcie_add_capability(PCIDevice *dev,
869                          uint16_t cap_id, uint8_t cap_ver,
870                          uint16_t offset, uint16_t size)
871 {
872     assert(offset >= PCI_CONFIG_SPACE_SIZE);
873     assert(offset < offset + size);
874     assert(offset + size <= PCIE_CONFIG_SPACE_SIZE);
875     assert(size >= 8);
876     assert(pci_is_express(dev));
877 
878     if (offset != PCI_CONFIG_SPACE_SIZE) {
879         uint16_t prev;
880 
881         /*
882          * 0xffffffff is not a valid cap id (it's a 16 bit field). use
883          * internally to find the last capability in the linked list.
884          */
885         pcie_find_capability_list(dev, 0xffffffff, &prev);
886         assert(prev >= PCI_CONFIG_SPACE_SIZE);
887         pcie_ext_cap_set_next(dev, prev, offset);
888     }
889     pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, 0));
890 
891     /* Make capability read-only by default */
892     memset(dev->wmask + offset, 0, size);
893     memset(dev->w1cmask + offset, 0, size);
894     /* Check capability by default */
895     memset(dev->cmask + offset, 0xFF, size);
896 }
897 
898 /*
899  * Sync the PCIe Link Status negotiated speed and width of a bridge with the
900  * downstream device.  If downstream device is not present, re-write with the
901  * Link Capability fields.  If downstream device reports invalid width or
902  * speed, replace with minimum values (LnkSta fields are RsvdZ on VFs but such
903  * values interfere with PCIe native hotplug detecting new devices).  Limit
904  * width and speed to bridge capabilities for compatibility.  Use config_read
905  * to access the downstream device since it could be an assigned device with
906  * volatile link information.
907  */
908 void pcie_sync_bridge_lnk(PCIDevice *bridge_dev)
909 {
910     PCIBridge *br = PCI_BRIDGE(bridge_dev);
911     PCIBus *bus = pci_bridge_get_sec_bus(br);
912     PCIDevice *target = bus->devices[0];
913     uint8_t *exp_cap = bridge_dev->config + bridge_dev->exp.exp_cap;
914     uint16_t lnksta, lnkcap = pci_get_word(exp_cap + PCI_EXP_LNKCAP);
915 
916     if (!target || !target->exp.exp_cap) {
917         lnksta = lnkcap;
918     } else {
919         lnksta = target->config_read(target,
920                                      target->exp.exp_cap + PCI_EXP_LNKSTA,
921                                      sizeof(lnksta));
922 
923         if ((lnksta & PCI_EXP_LNKSTA_NLW) > (lnkcap & PCI_EXP_LNKCAP_MLW)) {
924             lnksta &= ~PCI_EXP_LNKSTA_NLW;
925             lnksta |= lnkcap & PCI_EXP_LNKCAP_MLW;
926         } else if (!(lnksta & PCI_EXP_LNKSTA_NLW)) {
927             lnksta |= QEMU_PCI_EXP_LNKSTA_NLW(QEMU_PCI_EXP_LNK_X1);
928         }
929 
930         if ((lnksta & PCI_EXP_LNKSTA_CLS) > (lnkcap & PCI_EXP_LNKCAP_SLS)) {
931             lnksta &= ~PCI_EXP_LNKSTA_CLS;
932             lnksta |= lnkcap & PCI_EXP_LNKCAP_SLS;
933         } else if (!(lnksta & PCI_EXP_LNKSTA_CLS)) {
934             lnksta |= QEMU_PCI_EXP_LNKSTA_CLS(QEMU_PCI_EXP_LNK_2_5GT);
935         }
936     }
937 
938     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKSTA,
939                                  PCI_EXP_LNKSTA_CLS | PCI_EXP_LNKSTA_NLW);
940     pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA, lnksta &
941                                (PCI_EXP_LNKSTA_CLS | PCI_EXP_LNKSTA_NLW));
942 }
943 
944 /**************************************************************************
945  * pci express extended capability helper functions
946  */
947 
948 /* ARI */
949 void pcie_ari_init(PCIDevice *dev, uint16_t offset, uint16_t nextfn)
950 {
951     pcie_add_capability(dev, PCI_EXT_CAP_ID_ARI, PCI_ARI_VER,
952                         offset, PCI_ARI_SIZEOF);
953     pci_set_long(dev->config + offset + PCI_ARI_CAP, (nextfn & 0xff) << 8);
954 }
955 
956 void pcie_dev_ser_num_init(PCIDevice *dev, uint16_t offset, uint64_t ser_num)
957 {
958     static const int pci_dsn_ver = 1;
959     static const int pci_dsn_cap = 4;
960 
961     pcie_add_capability(dev, PCI_EXT_CAP_ID_DSN, pci_dsn_ver, offset,
962                         PCI_EXT_CAP_DSN_SIZEOF);
963     pci_set_quad(dev->config + offset + pci_dsn_cap, ser_num);
964 }
965 
966 void pcie_ats_init(PCIDevice *dev, uint16_t offset)
967 {
968     pcie_add_capability(dev, PCI_EXT_CAP_ID_ATS, 0x1,
969                         offset, PCI_EXT_CAP_ATS_SIZEOF);
970 
971     dev->exp.ats_cap = offset;
972 
973     /* Invalidate Queue Depth 0, Page Aligned Request 1 */
974     pci_set_word(dev->config + offset + PCI_ATS_CAP,
975                  PCI_ATS_CAP_PAGE_ALIGNED);
976     /* STU 0, Disabled by default */
977     pci_set_word(dev->config + offset + PCI_ATS_CTRL, 0);
978 
979     pci_set_word(dev->wmask + dev->exp.ats_cap + PCI_ATS_CTRL, 0x800f);
980 }
981 
982 /* ACS (Access Control Services) */
983 void pcie_acs_init(PCIDevice *dev, uint16_t offset)
984 {
985     bool is_downstream = pci_is_express_downstream_port(dev);
986     uint16_t cap_bits = 0;
987 
988     /* For endpoints, only multifunction devs may have an ACS capability: */
989     assert(is_downstream ||
990            (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) ||
991            PCI_FUNC(dev->devfn));
992 
993     pcie_add_capability(dev, PCI_EXT_CAP_ID_ACS, PCI_ACS_VER, offset,
994                         PCI_ACS_SIZEOF);
995     dev->exp.acs_cap = offset;
996 
997     if (is_downstream) {
998         /*
999          * Downstream ports must implement SV, TB, RR, CR, UF, and DT (with
1000          * caveats on the latter four that we ignore for simplicity).
1001          * Endpoints may also implement a subset of ACS capabilities,
1002          * but these are optional if the endpoint does not support
1003          * peer-to-peer between functions and thus omitted here.
1004          */
1005         cap_bits = PCI_ACS_SV | PCI_ACS_TB | PCI_ACS_RR |
1006             PCI_ACS_CR | PCI_ACS_UF | PCI_ACS_DT;
1007     }
1008 
1009     pci_set_word(dev->config + offset + PCI_ACS_CAP, cap_bits);
1010     pci_set_word(dev->wmask + offset + PCI_ACS_CTRL, cap_bits);
1011 }
1012 
1013 void pcie_acs_reset(PCIDevice *dev)
1014 {
1015     if (dev->exp.acs_cap) {
1016         pci_set_word(dev->config + dev->exp.acs_cap + PCI_ACS_CTRL, 0);
1017     }
1018 }
1019