xref: /openbmc/qemu/hw/pci/pcie.c (revision 65a117da)
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     if (dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA) {
79         pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA,
80                                    PCI_EXP_LNKSTA_DLLLA);
81     }
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         pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA,
129                                    PCI_EXP_LNKSTA_DLLLA);
130 
131         /*
132          * Target Link Speed defaults to the highest link speed supported by
133          * the component.  2.5GT/s devices are permitted to hardwire to zero.
134          */
135         pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKCTL2,
136                                      PCI_EXP_LNKCTL2_TLS);
137         pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKCTL2,
138                                    QEMU_PCI_EXP_LNKCAP_MLS(s->speed) &
139                                    PCI_EXP_LNKCTL2_TLS);
140     }
141 
142     /*
143      * 2.5 & 5.0GT/s can be fully described by LNKCAP, but 8.0GT/s is
144      * actually a reference to the highest bit supported in this register.
145      * We assume the device supports all link speeds.
146      */
147     if (s->speed > QEMU_PCI_EXP_LNK_5GT) {
148         pci_long_test_and_clear_mask(exp_cap + PCI_EXP_LNKCAP2, ~0U);
149         pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP2,
150                                    PCI_EXP_LNKCAP2_SLS_2_5GB |
151                                    PCI_EXP_LNKCAP2_SLS_5_0GB |
152                                    PCI_EXP_LNKCAP2_SLS_8_0GB);
153         if (s->speed > QEMU_PCI_EXP_LNK_8GT) {
154             pci_long_test_and_set_mask(exp_cap + PCI_EXP_LNKCAP2,
155                                        PCI_EXP_LNKCAP2_SLS_16_0GB);
156         }
157     }
158 }
159 
160 int pcie_cap_init(PCIDevice *dev, uint8_t offset,
161                   uint8_t type, uint8_t port,
162                   Error **errp)
163 {
164     /* PCIe cap v2 init */
165     int pos;
166     uint8_t *exp_cap;
167 
168     assert(pci_is_express(dev));
169 
170     pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
171                              PCI_EXP_VER2_SIZEOF, errp);
172     if (pos < 0) {
173         return pos;
174     }
175     dev->exp.exp_cap = pos;
176     exp_cap = dev->config + pos;
177 
178     /* Filling values common with v1 */
179     pcie_cap_v1_fill(dev, port, type, PCI_EXP_FLAGS_VER2);
180 
181     /* Fill link speed and width options */
182     pcie_cap_fill_slot_lnk(dev);
183 
184     /* Filling v2 specific values */
185     pci_set_long(exp_cap + PCI_EXP_DEVCAP2,
186                  PCI_EXP_DEVCAP2_EFF | PCI_EXP_DEVCAP2_EETLPP);
187 
188     pci_set_word(dev->wmask + pos + PCI_EXP_DEVCTL2, PCI_EXP_DEVCTL2_EETLPPB);
189 
190     if (dev->cap_present & QEMU_PCIE_EXTCAP_INIT) {
191         /* read-only to behave like a 'NULL' Extended Capability Header */
192         pci_set_long(dev->wmask + PCI_CONFIG_SPACE_SIZE, 0);
193     }
194 
195     return pos;
196 }
197 
198 int pcie_cap_v1_init(PCIDevice *dev, uint8_t offset, uint8_t type,
199                      uint8_t port)
200 {
201     /* PCIe cap v1 init */
202     int pos;
203     Error *local_err = NULL;
204 
205     assert(pci_is_express(dev));
206 
207     pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
208                              PCI_EXP_VER1_SIZEOF, &local_err);
209     if (pos < 0) {
210         error_report_err(local_err);
211         return pos;
212     }
213     dev->exp.exp_cap = pos;
214 
215     pcie_cap_v1_fill(dev, port, type, PCI_EXP_FLAGS_VER1);
216 
217     return pos;
218 }
219 
220 static int
221 pcie_endpoint_cap_common_init(PCIDevice *dev, uint8_t offset, uint8_t cap_size)
222 {
223     uint8_t type = PCI_EXP_TYPE_ENDPOINT;
224     Error *local_err = NULL;
225     int ret;
226 
227     /*
228      * Windows guests will report Code 10, device cannot start, if
229      * a regular Endpoint type is exposed on a root complex.  These
230      * should instead be Root Complex Integrated Endpoints.
231      */
232     if (pci_bus_is_express(pci_get_bus(dev))
233         && pci_bus_is_root(pci_get_bus(dev))) {
234         type = PCI_EXP_TYPE_RC_END;
235     }
236 
237     if (cap_size == PCI_EXP_VER1_SIZEOF) {
238         return pcie_cap_v1_init(dev, offset, type, 0);
239     } else {
240         ret = pcie_cap_init(dev, offset, type, 0, &local_err);
241 
242         if (ret < 0) {
243             error_report_err(local_err);
244         }
245 
246         return ret;
247     }
248 }
249 
250 int pcie_endpoint_cap_init(PCIDevice *dev, uint8_t offset)
251 {
252     return pcie_endpoint_cap_common_init(dev, offset, PCI_EXP_VER2_SIZEOF);
253 }
254 
255 int pcie_endpoint_cap_v1_init(PCIDevice *dev, uint8_t offset)
256 {
257     return pcie_endpoint_cap_common_init(dev, offset, PCI_EXP_VER1_SIZEOF);
258 }
259 
260 void pcie_cap_exit(PCIDevice *dev)
261 {
262     pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER2_SIZEOF);
263 }
264 
265 void pcie_cap_v1_exit(PCIDevice *dev)
266 {
267     pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER1_SIZEOF);
268 }
269 
270 uint8_t pcie_cap_get_type(const PCIDevice *dev)
271 {
272     uint32_t pos = dev->exp.exp_cap;
273     assert(pos > 0);
274     return (pci_get_word(dev->config + pos + PCI_EXP_FLAGS) &
275             PCI_EXP_FLAGS_TYPE) >> PCI_EXP_FLAGS_TYPE_SHIFT;
276 }
277 
278 /* MSI/MSI-X */
279 /* pci express interrupt message number */
280 /* 7.8.2 PCI Express Capabilities Register: Interrupt Message Number */
281 void pcie_cap_flags_set_vector(PCIDevice *dev, uint8_t vector)
282 {
283     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
284     assert(vector < 32);
285     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_FLAGS, PCI_EXP_FLAGS_IRQ);
286     pci_word_test_and_set_mask(exp_cap + PCI_EXP_FLAGS,
287                                vector << PCI_EXP_FLAGS_IRQ_SHIFT);
288 }
289 
290 uint8_t pcie_cap_flags_get_vector(PCIDevice *dev)
291 {
292     return (pci_get_word(dev->config + dev->exp.exp_cap + PCI_EXP_FLAGS) &
293             PCI_EXP_FLAGS_IRQ) >> PCI_EXP_FLAGS_IRQ_SHIFT;
294 }
295 
296 void pcie_cap_deverr_init(PCIDevice *dev)
297 {
298     uint32_t pos = dev->exp.exp_cap;
299     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP,
300                                PCI_EXP_DEVCAP_RBER);
301     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL,
302                                PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
303                                PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
304     pci_long_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_DEVSTA,
305                                PCI_EXP_DEVSTA_CED | PCI_EXP_DEVSTA_NFED |
306                                PCI_EXP_DEVSTA_FED | PCI_EXP_DEVSTA_URD);
307 }
308 
309 void pcie_cap_deverr_reset(PCIDevice *dev)
310 {
311     uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
312     pci_long_test_and_clear_mask(devctl,
313                                  PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
314                                  PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
315 }
316 
317 void pcie_cap_lnkctl_init(PCIDevice *dev)
318 {
319     uint32_t pos = dev->exp.exp_cap;
320     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_LNKCTL,
321                                PCI_EXP_LNKCTL_CCC | PCI_EXP_LNKCTL_ES);
322 }
323 
324 void pcie_cap_lnkctl_reset(PCIDevice *dev)
325 {
326     uint8_t *lnkctl = dev->config + dev->exp.exp_cap + PCI_EXP_LNKCTL;
327     pci_long_test_and_clear_mask(lnkctl,
328                                  PCI_EXP_LNKCTL_CCC | PCI_EXP_LNKCTL_ES);
329 }
330 
331 static void hotplug_event_update_event_status(PCIDevice *dev)
332 {
333     uint32_t pos = dev->exp.exp_cap;
334     uint8_t *exp_cap = dev->config + pos;
335     uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
336     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
337 
338     dev->exp.hpev_notified = (sltctl & PCI_EXP_SLTCTL_HPIE) &&
339         (sltsta & sltctl & PCI_EXP_HP_EV_SUPPORTED);
340 }
341 
342 static void hotplug_event_notify(PCIDevice *dev)
343 {
344     bool prev = dev->exp.hpev_notified;
345 
346     hotplug_event_update_event_status(dev);
347 
348     if (prev == dev->exp.hpev_notified) {
349         return;
350     }
351 
352     /* Note: the logic above does not take into account whether interrupts
353      * are masked. The result is that interrupt will be sent when it is
354      * subsequently unmasked. This appears to be legal: Section 6.7.3.4:
355      * The Port may optionally send an MSI when there are hot-plug events that
356      * occur while interrupt generation is disabled, and interrupt generation is
357      * subsequently enabled. */
358     if (msix_enabled(dev)) {
359         msix_notify(dev, pcie_cap_flags_get_vector(dev));
360     } else if (msi_enabled(dev)) {
361         msi_notify(dev, pcie_cap_flags_get_vector(dev));
362     } else {
363         pci_set_irq(dev, dev->exp.hpev_notified);
364     }
365 }
366 
367 static void hotplug_event_clear(PCIDevice *dev)
368 {
369     hotplug_event_update_event_status(dev);
370     if (!msix_enabled(dev) && !msi_enabled(dev) && !dev->exp.hpev_notified) {
371         pci_irq_deassert(dev);
372     }
373 }
374 
375 /*
376  * A PCI Express Hot-Plug Event has occurred, so update slot status register
377  * and notify OS of the event if necessary.
378  *
379  * 6.7.3 PCI Express Hot-Plug Events
380  * 6.7.3.4 Software Notification of Hot-Plug Events
381  */
382 static void pcie_cap_slot_event(PCIDevice *dev, PCIExpressHotPlugEvent event)
383 {
384     /* Minor optimization: if nothing changed - no event is needed. */
385     if (pci_word_test_and_set_mask(dev->config + dev->exp.exp_cap +
386                                    PCI_EXP_SLTSTA, event) == event) {
387         return;
388     }
389     hotplug_event_notify(dev);
390 }
391 
392 static void pcie_cap_slot_plug_common(PCIDevice *hotplug_dev, DeviceState *dev,
393                                       Error **errp)
394 {
395     uint8_t *exp_cap = hotplug_dev->config + hotplug_dev->exp.exp_cap;
396     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
397 
398     PCIE_DEV_PRINTF(PCI_DEVICE(dev), "hotplug state: 0x%x\n", sltsta);
399     if (sltsta & PCI_EXP_SLTSTA_EIS) {
400         /* the slot is electromechanically locked.
401          * This error is propagated up to qdev and then to HMP/QMP.
402          */
403         error_setg_errno(errp, EBUSY, "slot is electromechanically locked");
404     }
405 }
406 
407 void pcie_cap_slot_pre_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
408                                Error **errp)
409 {
410     pcie_cap_slot_plug_common(PCI_DEVICE(hotplug_dev), dev, errp);
411 }
412 
413 void pcie_cap_slot_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
414                            Error **errp)
415 {
416     PCIDevice *hotplug_pdev = PCI_DEVICE(hotplug_dev);
417     uint8_t *exp_cap = hotplug_pdev->config + hotplug_pdev->exp.exp_cap;
418     uint32_t sltcap = pci_get_word(exp_cap + PCI_EXP_SLTCAP);
419     PCIDevice *pci_dev = PCI_DEVICE(dev);
420 
421     /* Don't send event when device is enabled during qemu machine creation:
422      * it is present on boot, no hotplug event is necessary. We do send an
423      * event when the device is disabled later. */
424     if (!dev->hotplugged) {
425         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
426                                    PCI_EXP_SLTSTA_PDS);
427         if (pci_dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA) {
428             pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA,
429                                        PCI_EXP_LNKSTA_DLLLA);
430         }
431         return;
432     }
433 
434     /* Check if hot-plug is disabled on the slot */
435     if ((sltcap & PCI_EXP_SLTCAP_HPC) == 0) {
436         error_setg(errp, "Hot-plug failed: unsupported by the port device '%s'",
437                          DEVICE(hotplug_pdev)->id);
438         return;
439     }
440 
441     /* To enable multifunction hot-plug, we just ensure the function
442      * 0 added last. When function 0 is added, we set the sltsta and
443      * inform OS via event notification.
444      */
445     if (pci_get_function_0(pci_dev)) {
446         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
447                                    PCI_EXP_SLTSTA_PDS);
448         if (pci_dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA) {
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     object_property_set_bool(OBJECT(dev), false, "realized", &error_abort);
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), NULL);
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 
641     if (ranges_overlap(addr, len, pos + PCI_EXP_SLTSTA, 2)) {
642         /*
643          * Guests tend to clears all bits during init.
644          * If they clear bits that weren't set this is racy and will lose events:
645          * not a big problem for manual button presses, but a problem for us.
646          * As a work-around, detect this and revert status to what it was
647          * before the write.
648          *
649          * Note: in theory this can be detected as a duplicate button press
650          * which cancels the previous press. Does not seem to happen in
651          * practice as guests seem to only have this bug during init.
652          */
653 #define PCIE_SLOT_EVENTS (PCI_EXP_SLTSTA_ABP | PCI_EXP_SLTSTA_PFD | \
654                           PCI_EXP_SLTSTA_MRLSC | PCI_EXP_SLTSTA_PDC | \
655                           PCI_EXP_SLTSTA_CC)
656 
657         if (val & ~old_slt_sta & PCIE_SLOT_EVENTS) {
658             sltsta = (sltsta & ~PCIE_SLOT_EVENTS) | (old_slt_sta & PCIE_SLOT_EVENTS);
659             pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
660         }
661         hotplug_event_clear(dev);
662     }
663 
664     if (!ranges_overlap(addr, len, pos + PCI_EXP_SLTCTL, 2)) {
665         return;
666     }
667 
668     if (pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
669                                      PCI_EXP_SLTCTL_EIC)) {
670         sltsta ^= PCI_EXP_SLTSTA_EIS; /* toggle PCI_EXP_SLTSTA_EIS bit */
671         pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
672         PCIE_DEV_PRINTF(dev, "PCI_EXP_SLTCTL_EIC: "
673                         "sltsta -> 0x%02"PRIx16"\n",
674                         sltsta);
675     }
676 
677     /*
678      * If the slot is populated, power indicator is off and power
679      * controller is off, it is safe to detach the devices.
680      *
681      * Note: don't detach if condition was already true:
682      * this is a work around for guests that overwrite
683      * control of powered off slots before powering them on.
684      */
685     if ((sltsta & PCI_EXP_SLTSTA_PDS) && (val & PCI_EXP_SLTCTL_PCC) &&
686         (val & PCI_EXP_SLTCTL_PIC_OFF) == PCI_EXP_SLTCTL_PIC_OFF &&
687         (!(old_slt_ctl & PCI_EXP_SLTCTL_PCC) ||
688         (old_slt_ctl & PCI_EXP_SLTCTL_PIC_OFF) != PCI_EXP_SLTCTL_PIC_OFF)) {
689         PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
690         pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
691                             pcie_unplug_device, NULL);
692 
693         pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
694                                      PCI_EXP_SLTSTA_PDS);
695         if (dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA) {
696             pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKSTA,
697                                          PCI_EXP_LNKSTA_DLLLA);
698         }
699         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
700                                        PCI_EXP_SLTSTA_PDC);
701     }
702 
703     hotplug_event_notify(dev);
704 
705     /*
706      * 6.7.3.2 Command Completed Events
707      *
708      * Software issues a command to a hot-plug capable Downstream Port by
709      * issuing a write transaction that targets any portion of the Port’s Slot
710      * Control register. A single write to the Slot Control register is
711      * considered to be a single command, even if the write affects more than
712      * one field in the Slot Control register. In response to this transaction,
713      * the Port must carry out the requested actions and then set the
714      * associated status field for the command completed event. */
715 
716     /* Real hardware might take a while to complete requested command because
717      * physical movement would be involved like locking the electromechanical
718      * lock.  However in our case, command is completed instantaneously above,
719      * so send a command completion event right now.
720      */
721     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_CCI);
722 }
723 
724 int pcie_cap_slot_post_load(void *opaque, int version_id)
725 {
726     PCIDevice *dev = opaque;
727     hotplug_event_update_event_status(dev);
728     return 0;
729 }
730 
731 void pcie_cap_slot_push_attention_button(PCIDevice *dev)
732 {
733     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_ABP);
734 }
735 
736 /* root control/capabilities/status. PME isn't emulated for now */
737 void pcie_cap_root_init(PCIDevice *dev)
738 {
739     pci_set_word(dev->wmask + dev->exp.exp_cap + PCI_EXP_RTCTL,
740                  PCI_EXP_RTCTL_SECEE | PCI_EXP_RTCTL_SENFEE |
741                  PCI_EXP_RTCTL_SEFEE);
742 }
743 
744 void pcie_cap_root_reset(PCIDevice *dev)
745 {
746     pci_set_word(dev->config + dev->exp.exp_cap + PCI_EXP_RTCTL, 0);
747 }
748 
749 /* function level reset(FLR) */
750 void pcie_cap_flr_init(PCIDevice *dev)
751 {
752     pci_long_test_and_set_mask(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCAP,
753                                PCI_EXP_DEVCAP_FLR);
754 
755     /* Although reading BCR_FLR returns always 0,
756      * the bit is made writable here in order to detect the 1b is written
757      * pcie_cap_flr_write_config() test-and-clear the bit, so
758      * this bit always returns 0 to the guest.
759      */
760     pci_word_test_and_set_mask(dev->wmask + dev->exp.exp_cap + PCI_EXP_DEVCTL,
761                                PCI_EXP_DEVCTL_BCR_FLR);
762 }
763 
764 void pcie_cap_flr_write_config(PCIDevice *dev,
765                                uint32_t addr, uint32_t val, int len)
766 {
767     uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
768     if (pci_get_word(devctl) & PCI_EXP_DEVCTL_BCR_FLR) {
769         /* Clear PCI_EXP_DEVCTL_BCR_FLR after invoking the reset handler
770            so the handler can detect FLR by looking at this bit. */
771         pci_device_reset(dev);
772         pci_word_test_and_clear_mask(devctl, PCI_EXP_DEVCTL_BCR_FLR);
773     }
774 }
775 
776 /* Alternative Routing-ID Interpretation (ARI)
777  * forwarding support for root and downstream ports
778  */
779 void pcie_cap_arifwd_init(PCIDevice *dev)
780 {
781     uint32_t pos = dev->exp.exp_cap;
782     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP2,
783                                PCI_EXP_DEVCAP2_ARI);
784     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL2,
785                                PCI_EXP_DEVCTL2_ARI);
786 }
787 
788 void pcie_cap_arifwd_reset(PCIDevice *dev)
789 {
790     uint8_t *devctl2 = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2;
791     pci_long_test_and_clear_mask(devctl2, PCI_EXP_DEVCTL2_ARI);
792 }
793 
794 bool pcie_cap_is_arifwd_enabled(const PCIDevice *dev)
795 {
796     if (!pci_is_express(dev)) {
797         return false;
798     }
799     if (!dev->exp.exp_cap) {
800         return false;
801     }
802 
803     return pci_get_long(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2) &
804         PCI_EXP_DEVCTL2_ARI;
805 }
806 
807 /**************************************************************************
808  * pci express extended capability list management functions
809  * uint16_t ext_cap_id (16 bit)
810  * uint8_t cap_ver (4 bit)
811  * uint16_t cap_offset (12 bit)
812  * uint16_t ext_cap_size
813  */
814 
815 /* Passing a cap_id value > 0xffff will return 0 and put end of list in prev */
816 static uint16_t pcie_find_capability_list(PCIDevice *dev, uint32_t cap_id,
817                                           uint16_t *prev_p)
818 {
819     uint16_t prev = 0;
820     uint16_t next;
821     uint32_t header = pci_get_long(dev->config + PCI_CONFIG_SPACE_SIZE);
822 
823     if (!header) {
824         /* no extended capability */
825         next = 0;
826         goto out;
827     }
828     for (next = PCI_CONFIG_SPACE_SIZE; next;
829          prev = next, next = PCI_EXT_CAP_NEXT(header)) {
830 
831         assert(next >= PCI_CONFIG_SPACE_SIZE);
832         assert(next <= PCIE_CONFIG_SPACE_SIZE - 8);
833 
834         header = pci_get_long(dev->config + next);
835         if (PCI_EXT_CAP_ID(header) == cap_id) {
836             break;
837         }
838     }
839 
840 out:
841     if (prev_p) {
842         *prev_p = prev;
843     }
844     return next;
845 }
846 
847 uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id)
848 {
849     return pcie_find_capability_list(dev, cap_id, NULL);
850 }
851 
852 static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next)
853 {
854     uint32_t header = pci_get_long(dev->config + pos);
855     assert(!(next & (PCI_EXT_CAP_ALIGN - 1)));
856     header = (header & ~PCI_EXT_CAP_NEXT_MASK) |
857         ((next << PCI_EXT_CAP_NEXT_SHIFT) & PCI_EXT_CAP_NEXT_MASK);
858     pci_set_long(dev->config + pos, header);
859 }
860 
861 /*
862  * Caller must supply valid (offset, size) such that the range wouldn't
863  * overlap with other capability or other registers.
864  * This function doesn't check it.
865  */
866 void pcie_add_capability(PCIDevice *dev,
867                          uint16_t cap_id, uint8_t cap_ver,
868                          uint16_t offset, uint16_t size)
869 {
870     assert(offset >= PCI_CONFIG_SPACE_SIZE);
871     assert(offset < offset + size);
872     assert(offset + size <= PCIE_CONFIG_SPACE_SIZE);
873     assert(size >= 8);
874     assert(pci_is_express(dev));
875 
876     if (offset != PCI_CONFIG_SPACE_SIZE) {
877         uint16_t prev;
878 
879         /*
880          * 0xffffffff is not a valid cap id (it's a 16 bit field). use
881          * internally to find the last capability in the linked list.
882          */
883         pcie_find_capability_list(dev, 0xffffffff, &prev);
884         assert(prev >= PCI_CONFIG_SPACE_SIZE);
885         pcie_ext_cap_set_next(dev, prev, offset);
886     }
887     pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, 0));
888 
889     /* Make capability read-only by default */
890     memset(dev->wmask + offset, 0, size);
891     memset(dev->w1cmask + offset, 0, size);
892     /* Check capability by default */
893     memset(dev->cmask + offset, 0xFF, size);
894 }
895 
896 /*
897  * Sync the PCIe Link Status negotiated speed and width of a bridge with the
898  * downstream device.  If downstream device is not present, re-write with the
899  * Link Capability fields.  If downstream device reports invalid width or
900  * speed, replace with minimum values (LnkSta fields are RsvdZ on VFs but such
901  * values interfere with PCIe native hotplug detecting new devices).  Limit
902  * width and speed to bridge capabilities for compatibility.  Use config_read
903  * to access the downstream device since it could be an assigned device with
904  * volatile link information.
905  */
906 void pcie_sync_bridge_lnk(PCIDevice *bridge_dev)
907 {
908     PCIBridge *br = PCI_BRIDGE(bridge_dev);
909     PCIBus *bus = pci_bridge_get_sec_bus(br);
910     PCIDevice *target = bus->devices[0];
911     uint8_t *exp_cap = bridge_dev->config + bridge_dev->exp.exp_cap;
912     uint16_t lnksta, lnkcap = pci_get_word(exp_cap + PCI_EXP_LNKCAP);
913 
914     if (!target || !target->exp.exp_cap) {
915         lnksta = lnkcap;
916     } else {
917         lnksta = target->config_read(target,
918                                      target->exp.exp_cap + PCI_EXP_LNKSTA,
919                                      sizeof(lnksta));
920 
921         if ((lnksta & PCI_EXP_LNKSTA_NLW) > (lnkcap & PCI_EXP_LNKCAP_MLW)) {
922             lnksta &= ~PCI_EXP_LNKSTA_NLW;
923             lnksta |= lnkcap & PCI_EXP_LNKCAP_MLW;
924         } else if (!(lnksta & PCI_EXP_LNKSTA_NLW)) {
925             lnksta |= QEMU_PCI_EXP_LNKSTA_NLW(QEMU_PCI_EXP_LNK_X1);
926         }
927 
928         if ((lnksta & PCI_EXP_LNKSTA_CLS) > (lnkcap & PCI_EXP_LNKCAP_SLS)) {
929             lnksta &= ~PCI_EXP_LNKSTA_CLS;
930             lnksta |= lnkcap & PCI_EXP_LNKCAP_SLS;
931         } else if (!(lnksta & PCI_EXP_LNKSTA_CLS)) {
932             lnksta |= QEMU_PCI_EXP_LNKSTA_CLS(QEMU_PCI_EXP_LNK_2_5GT);
933         }
934     }
935 
936     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKSTA,
937                                  PCI_EXP_LNKSTA_CLS | PCI_EXP_LNKSTA_NLW);
938     pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA, lnksta &
939                                (PCI_EXP_LNKSTA_CLS | PCI_EXP_LNKSTA_NLW));
940 }
941 
942 /**************************************************************************
943  * pci express extended capability helper functions
944  */
945 
946 /* ARI */
947 void pcie_ari_init(PCIDevice *dev, uint16_t offset, uint16_t nextfn)
948 {
949     pcie_add_capability(dev, PCI_EXT_CAP_ID_ARI, PCI_ARI_VER,
950                         offset, PCI_ARI_SIZEOF);
951     pci_set_long(dev->config + offset + PCI_ARI_CAP, (nextfn & 0xff) << 8);
952 }
953 
954 void pcie_dev_ser_num_init(PCIDevice *dev, uint16_t offset, uint64_t ser_num)
955 {
956     static const int pci_dsn_ver = 1;
957     static const int pci_dsn_cap = 4;
958 
959     pcie_add_capability(dev, PCI_EXT_CAP_ID_DSN, pci_dsn_ver, offset,
960                         PCI_EXT_CAP_DSN_SIZEOF);
961     pci_set_quad(dev->config + offset + pci_dsn_cap, ser_num);
962 }
963 
964 void pcie_ats_init(PCIDevice *dev, uint16_t offset)
965 {
966     pcie_add_capability(dev, PCI_EXT_CAP_ID_ATS, 0x1,
967                         offset, PCI_EXT_CAP_ATS_SIZEOF);
968 
969     dev->exp.ats_cap = offset;
970 
971     /* Invalidate Queue Depth 0, Page Aligned Request 0 */
972     pci_set_word(dev->config + offset + PCI_ATS_CAP, 0);
973     /* STU 0, Disabled by default */
974     pci_set_word(dev->config + offset + PCI_ATS_CTRL, 0);
975 
976     pci_set_word(dev->wmask + dev->exp.ats_cap + PCI_ATS_CTRL, 0x800f);
977 }
978 
979 /* ACS (Access Control Services) */
980 void pcie_acs_init(PCIDevice *dev, uint16_t offset)
981 {
982     bool is_downstream = pci_is_express_downstream_port(dev);
983     uint16_t cap_bits = 0;
984 
985     /* For endpoints, only multifunction devs may have an ACS capability: */
986     assert(is_downstream ||
987            (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) ||
988            PCI_FUNC(dev->devfn));
989 
990     pcie_add_capability(dev, PCI_EXT_CAP_ID_ACS, PCI_ACS_VER, offset,
991                         PCI_ACS_SIZEOF);
992     dev->exp.acs_cap = offset;
993 
994     if (is_downstream) {
995         /*
996          * Downstream ports must implement SV, TB, RR, CR, UF, and DT (with
997          * caveats on the latter four that we ignore for simplicity).
998          * Endpoints may also implement a subset of ACS capabilities,
999          * but these are optional if the endpoint does not support
1000          * peer-to-peer between functions and thus omitted here.
1001          */
1002         cap_bits = PCI_ACS_SV | PCI_ACS_TB | PCI_ACS_RR |
1003             PCI_ACS_CR | PCI_ACS_UF | PCI_ACS_DT;
1004     }
1005 
1006     pci_set_word(dev->config + offset + PCI_ACS_CAP, cap_bits);
1007     pci_set_word(dev->wmask + offset + PCI_ACS_CTRL, cap_bits);
1008 }
1009 
1010 void pcie_acs_reset(PCIDevice *dev)
1011 {
1012     if (dev->exp.acs_cap) {
1013         pci_set_word(dev->config + dev->exp.acs_cap + PCI_ACS_CTRL, 0);
1014     }
1015 }
1016