xref: /openbmc/qemu/hw/pci/pcie.c (revision f3635813)
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     PCIDevice *pci_dev = PCI_DEVICE(dev);
419 
420     /* Don't send event when device is enabled during qemu machine creation:
421      * it is present on boot, no hotplug event is necessary. We do send an
422      * event when the device is disabled later. */
423     if (!dev->hotplugged) {
424         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
425                                    PCI_EXP_SLTSTA_PDS);
426         if (pci_dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA) {
427             pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA,
428                                        PCI_EXP_LNKSTA_DLLLA);
429         }
430         return;
431     }
432 
433     /* To enable multifunction hot-plug, we just ensure the function
434      * 0 added last. When function 0 is added, we set the sltsta and
435      * inform OS via event notification.
436      */
437     if (pci_get_function_0(pci_dev)) {
438         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
439                                    PCI_EXP_SLTSTA_PDS);
440         if (pci_dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA) {
441             pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA,
442                                        PCI_EXP_LNKSTA_DLLLA);
443         }
444         pcie_cap_slot_event(PCI_DEVICE(hotplug_dev),
445                             PCI_EXP_HP_EV_PDC | PCI_EXP_HP_EV_ABP);
446     }
447 }
448 
449 void pcie_cap_slot_unplug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
450                              Error **errp)
451 {
452     object_property_set_bool(OBJECT(dev), false, "realized", NULL);
453 }
454 
455 static void pcie_unplug_device(PCIBus *bus, PCIDevice *dev, void *opaque)
456 {
457     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(DEVICE(dev));
458 
459     if (dev->partially_hotplugged) {
460         dev->qdev.pending_deleted_event = false;
461         return;
462     }
463     hotplug_handler_unplug(hotplug_ctrl, DEVICE(dev), &error_abort);
464     object_unparent(OBJECT(dev));
465 }
466 
467 void pcie_cap_slot_unplug_request_cb(HotplugHandler *hotplug_dev,
468                                      DeviceState *dev, Error **errp)
469 {
470     Error *local_err = NULL;
471     PCIDevice *pci_dev = PCI_DEVICE(dev);
472     PCIBus *bus = pci_get_bus(pci_dev);
473 
474     pcie_cap_slot_plug_common(PCI_DEVICE(hotplug_dev), dev, &local_err);
475     if (local_err) {
476         error_propagate(errp, local_err);
477         return;
478     }
479 
480     dev->pending_deleted_event = true;
481 
482     /* In case user cancel the operation of multi-function hot-add,
483      * remove the function that is unexposed to guest individually,
484      * without interaction with guest.
485      */
486     if (pci_dev->devfn &&
487         !bus->devices[0]) {
488         pcie_unplug_device(bus, pci_dev, NULL);
489 
490         return;
491     }
492 
493     pcie_cap_slot_push_attention_button(PCI_DEVICE(hotplug_dev));
494 }
495 
496 /* pci express slot for pci express root/downstream port
497    PCI express capability slot registers */
498 void pcie_cap_slot_init(PCIDevice *dev, uint16_t slot)
499 {
500     uint32_t pos = dev->exp.exp_cap;
501 
502     pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_FLAGS,
503                                PCI_EXP_FLAGS_SLOT);
504 
505     pci_long_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCAP,
506                                  ~PCI_EXP_SLTCAP_PSN);
507     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
508                                (slot << PCI_EXP_SLTCAP_PSN_SHIFT) |
509                                PCI_EXP_SLTCAP_EIP |
510                                PCI_EXP_SLTCAP_HPS |
511                                PCI_EXP_SLTCAP_HPC |
512                                PCI_EXP_SLTCAP_PIP |
513                                PCI_EXP_SLTCAP_AIP |
514                                PCI_EXP_SLTCAP_ABP);
515 
516     if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
517         pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
518                                    PCI_EXP_SLTCAP_PCP);
519         pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
520                                      PCI_EXP_SLTCTL_PCC);
521         pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
522                                    PCI_EXP_SLTCTL_PCC);
523     }
524 
525     pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
526                                  PCI_EXP_SLTCTL_PIC |
527                                  PCI_EXP_SLTCTL_AIC);
528     pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCTL,
529                                PCI_EXP_SLTCTL_PIC_OFF |
530                                PCI_EXP_SLTCTL_AIC_OFF);
531     pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
532                                PCI_EXP_SLTCTL_PIC |
533                                PCI_EXP_SLTCTL_AIC |
534                                PCI_EXP_SLTCTL_HPIE |
535                                PCI_EXP_SLTCTL_CCIE |
536                                PCI_EXP_SLTCTL_PDCE |
537                                PCI_EXP_SLTCTL_ABPE);
538     /* Although reading PCI_EXP_SLTCTL_EIC returns always 0,
539      * make the bit writable here in order to detect 1b is written.
540      * pcie_cap_slot_write_config() test-and-clear the bit, so
541      * this bit always returns 0 to the guest.
542      */
543     pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
544                                PCI_EXP_SLTCTL_EIC);
545 
546     pci_word_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_SLTSTA,
547                                PCI_EXP_HP_EV_SUPPORTED);
548 
549     dev->exp.hpev_notified = false;
550 
551     qbus_set_hotplug_handler(BUS(pci_bridge_get_sec_bus(PCI_BRIDGE(dev))),
552                              OBJECT(dev), NULL);
553 }
554 
555 void pcie_cap_slot_reset(PCIDevice *dev)
556 {
557     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
558     uint8_t port_type = pcie_cap_get_type(dev);
559 
560     assert(port_type == PCI_EXP_TYPE_DOWNSTREAM ||
561            port_type == PCI_EXP_TYPE_ROOT_PORT);
562 
563     PCIE_DEV_PRINTF(dev, "reset\n");
564 
565     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
566                                  PCI_EXP_SLTCTL_EIC |
567                                  PCI_EXP_SLTCTL_PIC |
568                                  PCI_EXP_SLTCTL_AIC |
569                                  PCI_EXP_SLTCTL_HPIE |
570                                  PCI_EXP_SLTCTL_CCIE |
571                                  PCI_EXP_SLTCTL_PDCE |
572                                  PCI_EXP_SLTCTL_ABPE);
573     pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
574                                PCI_EXP_SLTCTL_AIC_OFF);
575 
576     if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
577         /* Downstream ports enforce device number 0. */
578         bool populated = pci_bridge_get_sec_bus(PCI_BRIDGE(dev))->devices[0];
579         uint16_t pic;
580 
581         if (populated) {
582             pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
583                                          PCI_EXP_SLTCTL_PCC);
584         } else {
585             pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
586                                        PCI_EXP_SLTCTL_PCC);
587         }
588 
589         pic = populated ? PCI_EXP_SLTCTL_PIC_ON : PCI_EXP_SLTCTL_PIC_OFF;
590         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL, pic);
591     }
592 
593     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
594                                  PCI_EXP_SLTSTA_EIS |/* on reset,
595                                                         the lock is released */
596                                  PCI_EXP_SLTSTA_CC |
597                                  PCI_EXP_SLTSTA_PDC |
598                                  PCI_EXP_SLTSTA_ABP);
599 
600     hotplug_event_update_event_status(dev);
601 }
602 
603 void pcie_cap_slot_get(PCIDevice *dev, uint16_t *slt_ctl, uint16_t *slt_sta)
604 {
605     uint32_t pos = dev->exp.exp_cap;
606     uint8_t *exp_cap = dev->config + pos;
607     *slt_ctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
608     *slt_sta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
609 }
610 
611 void pcie_cap_slot_write_config(PCIDevice *dev,
612                                 uint16_t old_slt_ctl, uint16_t old_slt_sta,
613                                 uint32_t addr, uint32_t val, int len)
614 {
615     uint32_t pos = dev->exp.exp_cap;
616     uint8_t *exp_cap = dev->config + pos;
617     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
618 
619     if (ranges_overlap(addr, len, pos + PCI_EXP_SLTSTA, 2)) {
620         /*
621          * Guests tend to clears all bits during init.
622          * If they clear bits that weren't set this is racy and will lose events:
623          * not a big problem for manual button presses, but a problem for us.
624          * As a work-around, detect this and revert status to what it was
625          * before the write.
626          *
627          * Note: in theory this can be detected as a duplicate button press
628          * which cancels the previous press. Does not seem to happen in
629          * practice as guests seem to only have this bug during init.
630          */
631 #define PCIE_SLOT_EVENTS (PCI_EXP_SLTSTA_ABP | PCI_EXP_SLTSTA_PFD | \
632                           PCI_EXP_SLTSTA_MRLSC | PCI_EXP_SLTSTA_PDC | \
633                           PCI_EXP_SLTSTA_CC)
634 
635         if (val & ~old_slt_sta & PCIE_SLOT_EVENTS) {
636             sltsta = (sltsta & ~PCIE_SLOT_EVENTS) | (old_slt_sta & PCIE_SLOT_EVENTS);
637             pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
638         }
639         hotplug_event_clear(dev);
640     }
641 
642     if (!ranges_overlap(addr, len, pos + PCI_EXP_SLTCTL, 2)) {
643         return;
644     }
645 
646     if (pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
647                                      PCI_EXP_SLTCTL_EIC)) {
648         sltsta ^= PCI_EXP_SLTSTA_EIS; /* toggle PCI_EXP_SLTSTA_EIS bit */
649         pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
650         PCIE_DEV_PRINTF(dev, "PCI_EXP_SLTCTL_EIC: "
651                         "sltsta -> 0x%02"PRIx16"\n",
652                         sltsta);
653     }
654 
655     /*
656      * If the slot is populated, power indicator is off and power
657      * controller is off, it is safe to detach the devices.
658      *
659      * Note: don't detach if condition was already true:
660      * this is a work around for guests that overwrite
661      * control of powered off slots before powering them on.
662      */
663     if ((sltsta & PCI_EXP_SLTSTA_PDS) && (val & PCI_EXP_SLTCTL_PCC) &&
664         (val & PCI_EXP_SLTCTL_PIC_OFF) == PCI_EXP_SLTCTL_PIC_OFF &&
665         (!(old_slt_ctl & PCI_EXP_SLTCTL_PCC) ||
666         (old_slt_ctl & PCI_EXP_SLTCTL_PIC_OFF) != PCI_EXP_SLTCTL_PIC_OFF)) {
667         PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
668         pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
669                             pcie_unplug_device, NULL);
670 
671         pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
672                                      PCI_EXP_SLTSTA_PDS);
673         if (dev->cap_present & QEMU_PCIE_LNKSTA_DLLLA) {
674             pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKSTA,
675                                          PCI_EXP_LNKSTA_DLLLA);
676         }
677         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
678                                        PCI_EXP_SLTSTA_PDC);
679     }
680 
681     hotplug_event_notify(dev);
682 
683     /*
684      * 6.7.3.2 Command Completed Events
685      *
686      * Software issues a command to a hot-plug capable Downstream Port by
687      * issuing a write transaction that targets any portion of the Port’s Slot
688      * Control register. A single write to the Slot Control register is
689      * considered to be a single command, even if the write affects more than
690      * one field in the Slot Control register. In response to this transaction,
691      * the Port must carry out the requested actions and then set the
692      * associated status field for the command completed event. */
693 
694     /* Real hardware might take a while to complete requested command because
695      * physical movement would be involved like locking the electromechanical
696      * lock.  However in our case, command is completed instantaneously above,
697      * so send a command completion event right now.
698      */
699     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_CCI);
700 }
701 
702 int pcie_cap_slot_post_load(void *opaque, int version_id)
703 {
704     PCIDevice *dev = opaque;
705     hotplug_event_update_event_status(dev);
706     return 0;
707 }
708 
709 void pcie_cap_slot_push_attention_button(PCIDevice *dev)
710 {
711     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_ABP);
712 }
713 
714 /* root control/capabilities/status. PME isn't emulated for now */
715 void pcie_cap_root_init(PCIDevice *dev)
716 {
717     pci_set_word(dev->wmask + dev->exp.exp_cap + PCI_EXP_RTCTL,
718                  PCI_EXP_RTCTL_SECEE | PCI_EXP_RTCTL_SENFEE |
719                  PCI_EXP_RTCTL_SEFEE);
720 }
721 
722 void pcie_cap_root_reset(PCIDevice *dev)
723 {
724     pci_set_word(dev->config + dev->exp.exp_cap + PCI_EXP_RTCTL, 0);
725 }
726 
727 /* function level reset(FLR) */
728 void pcie_cap_flr_init(PCIDevice *dev)
729 {
730     pci_long_test_and_set_mask(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCAP,
731                                PCI_EXP_DEVCAP_FLR);
732 
733     /* Although reading BCR_FLR returns always 0,
734      * the bit is made writable here in order to detect the 1b is written
735      * pcie_cap_flr_write_config() test-and-clear the bit, so
736      * this bit always returns 0 to the guest.
737      */
738     pci_word_test_and_set_mask(dev->wmask + dev->exp.exp_cap + PCI_EXP_DEVCTL,
739                                PCI_EXP_DEVCTL_BCR_FLR);
740 }
741 
742 void pcie_cap_flr_write_config(PCIDevice *dev,
743                                uint32_t addr, uint32_t val, int len)
744 {
745     uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
746     if (pci_get_word(devctl) & PCI_EXP_DEVCTL_BCR_FLR) {
747         /* Clear PCI_EXP_DEVCTL_BCR_FLR after invoking the reset handler
748            so the handler can detect FLR by looking at this bit. */
749         pci_device_reset(dev);
750         pci_word_test_and_clear_mask(devctl, PCI_EXP_DEVCTL_BCR_FLR);
751     }
752 }
753 
754 /* Alternative Routing-ID Interpretation (ARI)
755  * forwarding support for root and downstream ports
756  */
757 void pcie_cap_arifwd_init(PCIDevice *dev)
758 {
759     uint32_t pos = dev->exp.exp_cap;
760     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP2,
761                                PCI_EXP_DEVCAP2_ARI);
762     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL2,
763                                PCI_EXP_DEVCTL2_ARI);
764 }
765 
766 void pcie_cap_arifwd_reset(PCIDevice *dev)
767 {
768     uint8_t *devctl2 = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2;
769     pci_long_test_and_clear_mask(devctl2, PCI_EXP_DEVCTL2_ARI);
770 }
771 
772 bool pcie_cap_is_arifwd_enabled(const PCIDevice *dev)
773 {
774     if (!pci_is_express(dev)) {
775         return false;
776     }
777     if (!dev->exp.exp_cap) {
778         return false;
779     }
780 
781     return pci_get_long(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2) &
782         PCI_EXP_DEVCTL2_ARI;
783 }
784 
785 /**************************************************************************
786  * pci express extended capability list management functions
787  * uint16_t ext_cap_id (16 bit)
788  * uint8_t cap_ver (4 bit)
789  * uint16_t cap_offset (12 bit)
790  * uint16_t ext_cap_size
791  */
792 
793 /* Passing a cap_id value > 0xffff will return 0 and put end of list in prev */
794 static uint16_t pcie_find_capability_list(PCIDevice *dev, uint32_t cap_id,
795                                           uint16_t *prev_p)
796 {
797     uint16_t prev = 0;
798     uint16_t next;
799     uint32_t header = pci_get_long(dev->config + PCI_CONFIG_SPACE_SIZE);
800 
801     if (!header) {
802         /* no extended capability */
803         next = 0;
804         goto out;
805     }
806     for (next = PCI_CONFIG_SPACE_SIZE; next;
807          prev = next, next = PCI_EXT_CAP_NEXT(header)) {
808 
809         assert(next >= PCI_CONFIG_SPACE_SIZE);
810         assert(next <= PCIE_CONFIG_SPACE_SIZE - 8);
811 
812         header = pci_get_long(dev->config + next);
813         if (PCI_EXT_CAP_ID(header) == cap_id) {
814             break;
815         }
816     }
817 
818 out:
819     if (prev_p) {
820         *prev_p = prev;
821     }
822     return next;
823 }
824 
825 uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id)
826 {
827     return pcie_find_capability_list(dev, cap_id, NULL);
828 }
829 
830 static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next)
831 {
832     uint32_t header = pci_get_long(dev->config + pos);
833     assert(!(next & (PCI_EXT_CAP_ALIGN - 1)));
834     header = (header & ~PCI_EXT_CAP_NEXT_MASK) |
835         ((next << PCI_EXT_CAP_NEXT_SHIFT) & PCI_EXT_CAP_NEXT_MASK);
836     pci_set_long(dev->config + pos, header);
837 }
838 
839 /*
840  * Caller must supply valid (offset, size) such that the range wouldn't
841  * overlap with other capability or other registers.
842  * This function doesn't check it.
843  */
844 void pcie_add_capability(PCIDevice *dev,
845                          uint16_t cap_id, uint8_t cap_ver,
846                          uint16_t offset, uint16_t size)
847 {
848     assert(offset >= PCI_CONFIG_SPACE_SIZE);
849     assert(offset < offset + size);
850     assert(offset + size <= PCIE_CONFIG_SPACE_SIZE);
851     assert(size >= 8);
852     assert(pci_is_express(dev));
853 
854     if (offset != PCI_CONFIG_SPACE_SIZE) {
855         uint16_t prev;
856 
857         /*
858          * 0xffffffff is not a valid cap id (it's a 16 bit field). use
859          * internally to find the last capability in the linked list.
860          */
861         pcie_find_capability_list(dev, 0xffffffff, &prev);
862         assert(prev >= PCI_CONFIG_SPACE_SIZE);
863         pcie_ext_cap_set_next(dev, prev, offset);
864     }
865     pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, 0));
866 
867     /* Make capability read-only by default */
868     memset(dev->wmask + offset, 0, size);
869     memset(dev->w1cmask + offset, 0, size);
870     /* Check capability by default */
871     memset(dev->cmask + offset, 0xFF, size);
872 }
873 
874 /*
875  * Sync the PCIe Link Status negotiated speed and width of a bridge with the
876  * downstream device.  If downstream device is not present, re-write with the
877  * Link Capability fields.  If downstream device reports invalid width or
878  * speed, replace with minimum values (LnkSta fields are RsvdZ on VFs but such
879  * values interfere with PCIe native hotplug detecting new devices).  Limit
880  * width and speed to bridge capabilities for compatibility.  Use config_read
881  * to access the downstream device since it could be an assigned device with
882  * volatile link information.
883  */
884 void pcie_sync_bridge_lnk(PCIDevice *bridge_dev)
885 {
886     PCIBridge *br = PCI_BRIDGE(bridge_dev);
887     PCIBus *bus = pci_bridge_get_sec_bus(br);
888     PCIDevice *target = bus->devices[0];
889     uint8_t *exp_cap = bridge_dev->config + bridge_dev->exp.exp_cap;
890     uint16_t lnksta, lnkcap = pci_get_word(exp_cap + PCI_EXP_LNKCAP);
891 
892     if (!target || !target->exp.exp_cap) {
893         lnksta = lnkcap;
894     } else {
895         lnksta = target->config_read(target,
896                                      target->exp.exp_cap + PCI_EXP_LNKSTA,
897                                      sizeof(lnksta));
898 
899         if ((lnksta & PCI_EXP_LNKSTA_NLW) > (lnkcap & PCI_EXP_LNKCAP_MLW)) {
900             lnksta &= ~PCI_EXP_LNKSTA_NLW;
901             lnksta |= lnkcap & PCI_EXP_LNKCAP_MLW;
902         } else if (!(lnksta & PCI_EXP_LNKSTA_NLW)) {
903             lnksta |= QEMU_PCI_EXP_LNKSTA_NLW(QEMU_PCI_EXP_LNK_X1);
904         }
905 
906         if ((lnksta & PCI_EXP_LNKSTA_CLS) > (lnkcap & PCI_EXP_LNKCAP_SLS)) {
907             lnksta &= ~PCI_EXP_LNKSTA_CLS;
908             lnksta |= lnkcap & PCI_EXP_LNKCAP_SLS;
909         } else if (!(lnksta & PCI_EXP_LNKSTA_CLS)) {
910             lnksta |= QEMU_PCI_EXP_LNKSTA_CLS(QEMU_PCI_EXP_LNK_2_5GT);
911         }
912     }
913 
914     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_LNKSTA,
915                                  PCI_EXP_LNKSTA_CLS | PCI_EXP_LNKSTA_NLW);
916     pci_word_test_and_set_mask(exp_cap + PCI_EXP_LNKSTA, lnksta &
917                                (PCI_EXP_LNKSTA_CLS | PCI_EXP_LNKSTA_NLW));
918 }
919 
920 /**************************************************************************
921  * pci express extended capability helper functions
922  */
923 
924 /* ARI */
925 void pcie_ari_init(PCIDevice *dev, uint16_t offset, uint16_t nextfn)
926 {
927     pcie_add_capability(dev, PCI_EXT_CAP_ID_ARI, PCI_ARI_VER,
928                         offset, PCI_ARI_SIZEOF);
929     pci_set_long(dev->config + offset + PCI_ARI_CAP, (nextfn & 0xff) << 8);
930 }
931 
932 void pcie_dev_ser_num_init(PCIDevice *dev, uint16_t offset, uint64_t ser_num)
933 {
934     static const int pci_dsn_ver = 1;
935     static const int pci_dsn_cap = 4;
936 
937     pcie_add_capability(dev, PCI_EXT_CAP_ID_DSN, pci_dsn_ver, offset,
938                         PCI_EXT_CAP_DSN_SIZEOF);
939     pci_set_quad(dev->config + offset + pci_dsn_cap, ser_num);
940 }
941 
942 void pcie_ats_init(PCIDevice *dev, uint16_t offset)
943 {
944     pcie_add_capability(dev, PCI_EXT_CAP_ID_ATS, 0x1,
945                         offset, PCI_EXT_CAP_ATS_SIZEOF);
946 
947     dev->exp.ats_cap = offset;
948 
949     /* Invalidate Queue Depth 0, Page Aligned Request 0 */
950     pci_set_word(dev->config + offset + PCI_ATS_CAP, 0);
951     /* STU 0, Disabled by default */
952     pci_set_word(dev->config + offset + PCI_ATS_CTRL, 0);
953 
954     pci_set_word(dev->wmask + dev->exp.ats_cap + PCI_ATS_CTRL, 0x800f);
955 }
956 
957 /* ACS (Access Control Services) */
958 void pcie_acs_init(PCIDevice *dev, uint16_t offset)
959 {
960     bool is_downstream = pci_is_express_downstream_port(dev);
961     uint16_t cap_bits = 0;
962 
963     /* For endpoints, only multifunction devs may have an ACS capability: */
964     assert(is_downstream ||
965            (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) ||
966            PCI_FUNC(dev->devfn));
967 
968     pcie_add_capability(dev, PCI_EXT_CAP_ID_ACS, PCI_ACS_VER, offset,
969                         PCI_ACS_SIZEOF);
970     dev->exp.acs_cap = offset;
971 
972     if (is_downstream) {
973         /*
974          * Downstream ports must implement SV, TB, RR, CR, UF, and DT (with
975          * caveats on the latter four that we ignore for simplicity).
976          * Endpoints may also implement a subset of ACS capabilities,
977          * but these are optional if the endpoint does not support
978          * peer-to-peer between functions and thus omitted here.
979          */
980         cap_bits = PCI_ACS_SV | PCI_ACS_TB | PCI_ACS_RR |
981             PCI_ACS_CR | PCI_ACS_UF | PCI_ACS_DT;
982     }
983 
984     pci_set_word(dev->config + offset + PCI_ACS_CAP, cap_bits);
985     pci_set_word(dev->wmask + offset + PCI_ACS_CTRL, cap_bits);
986 }
987 
988 void pcie_acs_reset(PCIDevice *dev)
989 {
990     if (dev->exp.acs_cap) {
991         pci_set_word(dev->config + dev->exp.acs_cap + PCI_ACS_CTRL, 0);
992     }
993 }
994