xref: /qemu/hw/pci/pcie.c (revision d072cdf3)
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-common.h"
22 #include "hw/pci/pci_bridge.h"
23 #include "hw/pci/pcie.h"
24 #include "hw/pci/msix.h"
25 #include "hw/pci/msi.h"
26 #include "hw/pci/pci_bus.h"
27 #include "hw/pci/pcie_regs.h"
28 #include "qemu/range.h"
29 #include "qapi/qmp/qerror.h"
30 
31 //#define DEBUG_PCIE
32 #ifdef DEBUG_PCIE
33 # define PCIE_DPRINTF(fmt, ...)                                         \
34     fprintf(stderr, "%s:%d " fmt, __func__, __LINE__, ## __VA_ARGS__)
35 #else
36 # define PCIE_DPRINTF(fmt, ...) do {} while (0)
37 #endif
38 #define PCIE_DEV_PRINTF(dev, fmt, ...)                                  \
39     PCIE_DPRINTF("%s:%x "fmt, (dev)->name, (dev)->devfn, ## __VA_ARGS__)
40 
41 
42 /***************************************************************************
43  * pci express capability helper functions
44  */
45 int pcie_cap_init(PCIDevice *dev, uint8_t offset, uint8_t type, uint8_t port)
46 {
47     int pos;
48     uint8_t *exp_cap;
49 
50     assert(pci_is_express(dev));
51 
52     pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
53                                  PCI_EXP_VER2_SIZEOF);
54     if (pos < 0) {
55         return pos;
56     }
57     dev->exp.exp_cap = pos;
58     exp_cap = dev->config + pos;
59 
60     /* capability register
61        interrupt message number defaults to 0 */
62     pci_set_word(exp_cap + PCI_EXP_FLAGS,
63                  ((type << PCI_EXP_FLAGS_TYPE_SHIFT) & PCI_EXP_FLAGS_TYPE) |
64                  PCI_EXP_FLAGS_VER2);
65 
66     /* device capability register
67      * table 7-12:
68      * roll based error reporting bit must be set by all
69      * Functions conforming to the ECN, PCI Express Base
70      * Specification, Revision 1.1., or subsequent PCI Express Base
71      * Specification revisions.
72      */
73     pci_set_long(exp_cap + PCI_EXP_DEVCAP, PCI_EXP_DEVCAP_RBER);
74 
75     pci_set_long(exp_cap + PCI_EXP_LNKCAP,
76                  (port << PCI_EXP_LNKCAP_PN_SHIFT) |
77                  PCI_EXP_LNKCAP_ASPMS_0S |
78                  PCI_EXP_LNK_MLW_1 |
79                  PCI_EXP_LNK_LS_25);
80 
81     pci_set_word(exp_cap + PCI_EXP_LNKSTA,
82                  PCI_EXP_LNK_MLW_1 | PCI_EXP_LNK_LS_25);
83 
84     pci_set_long(exp_cap + PCI_EXP_DEVCAP2,
85                  PCI_EXP_DEVCAP2_EFF | PCI_EXP_DEVCAP2_EETLPP);
86 
87     pci_set_word(dev->wmask + pos, PCI_EXP_DEVCTL2_EETLPPB);
88     return pos;
89 }
90 
91 int pcie_endpoint_cap_init(PCIDevice *dev, uint8_t offset)
92 {
93     uint8_t type = PCI_EXP_TYPE_ENDPOINT;
94 
95     /*
96      * Windows guests will report Code 10, device cannot start, if
97      * a regular Endpoint type is exposed on a root complex.  These
98      * should instead be Root Complex Integrated Endpoints.
99      */
100     if (pci_bus_is_express(dev->bus) && pci_bus_is_root(dev->bus)) {
101         type = PCI_EXP_TYPE_RC_END;
102     }
103 
104     return pcie_cap_init(dev, offset, type, 0);
105 }
106 
107 void pcie_cap_exit(PCIDevice *dev)
108 {
109     pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER2_SIZEOF);
110 }
111 
112 uint8_t pcie_cap_get_type(const PCIDevice *dev)
113 {
114     uint32_t pos = dev->exp.exp_cap;
115     assert(pos > 0);
116     return (pci_get_word(dev->config + pos + PCI_EXP_FLAGS) &
117             PCI_EXP_FLAGS_TYPE) >> PCI_EXP_FLAGS_TYPE_SHIFT;
118 }
119 
120 /* MSI/MSI-X */
121 /* pci express interrupt message number */
122 /* 7.8.2 PCI Express Capabilities Register: Interrupt Message Number */
123 void pcie_cap_flags_set_vector(PCIDevice *dev, uint8_t vector)
124 {
125     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
126     assert(vector < 32);
127     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_FLAGS, PCI_EXP_FLAGS_IRQ);
128     pci_word_test_and_set_mask(exp_cap + PCI_EXP_FLAGS,
129                                vector << PCI_EXP_FLAGS_IRQ_SHIFT);
130 }
131 
132 uint8_t pcie_cap_flags_get_vector(PCIDevice *dev)
133 {
134     return (pci_get_word(dev->config + dev->exp.exp_cap + PCI_EXP_FLAGS) &
135             PCI_EXP_FLAGS_IRQ) >> PCI_EXP_FLAGS_IRQ_SHIFT;
136 }
137 
138 void pcie_cap_deverr_init(PCIDevice *dev)
139 {
140     uint32_t pos = dev->exp.exp_cap;
141     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP,
142                                PCI_EXP_DEVCAP_RBER);
143     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL,
144                                PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
145                                PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
146     pci_long_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_DEVSTA,
147                                PCI_EXP_DEVSTA_CED | PCI_EXP_DEVSTA_NFED |
148                                PCI_EXP_DEVSTA_URD | PCI_EXP_DEVSTA_URD);
149 }
150 
151 void pcie_cap_deverr_reset(PCIDevice *dev)
152 {
153     uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
154     pci_long_test_and_clear_mask(devctl,
155                                  PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
156                                  PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
157 }
158 
159 static void hotplug_event_update_event_status(PCIDevice *dev)
160 {
161     uint32_t pos = dev->exp.exp_cap;
162     uint8_t *exp_cap = dev->config + pos;
163     uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
164     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
165 
166     dev->exp.hpev_notified = (sltctl & PCI_EXP_SLTCTL_HPIE) &&
167         (sltsta & sltctl & PCI_EXP_HP_EV_SUPPORTED);
168 }
169 
170 static void hotplug_event_notify(PCIDevice *dev)
171 {
172     bool prev = dev->exp.hpev_notified;
173 
174     hotplug_event_update_event_status(dev);
175 
176     if (prev == dev->exp.hpev_notified) {
177         return;
178     }
179 
180     /* Note: the logic above does not take into account whether interrupts
181      * are masked. The result is that interrupt will be sent when it is
182      * subsequently unmasked. This appears to be legal: Section 6.7.3.4:
183      * The Port may optionally send an MSI when there are hot-plug events that
184      * occur while interrupt generation is disabled, and interrupt generation is
185      * subsequently enabled. */
186     if (msix_enabled(dev)) {
187         msix_notify(dev, pcie_cap_flags_get_vector(dev));
188     } else if (msi_enabled(dev)) {
189         msi_notify(dev, pcie_cap_flags_get_vector(dev));
190     } else {
191         pci_set_irq(dev, dev->exp.hpev_notified);
192     }
193 }
194 
195 static void hotplug_event_clear(PCIDevice *dev)
196 {
197     hotplug_event_update_event_status(dev);
198     if (!msix_enabled(dev) && !msi_enabled(dev) && !dev->exp.hpev_notified) {
199         pci_irq_deassert(dev);
200     }
201 }
202 
203 /*
204  * A PCI Express Hot-Plug Event has occurred, so update slot status register
205  * and notify OS of the event if necessary.
206  *
207  * 6.7.3 PCI Express Hot-Plug Events
208  * 6.7.3.4 Software Notification of Hot-Plug Events
209  */
210 static void pcie_cap_slot_event(PCIDevice *dev, PCIExpressHotPlugEvent event)
211 {
212     /* Minor optimization: if nothing changed - no event is needed. */
213     if (pci_word_test_and_set_mask(dev->config + dev->exp.exp_cap +
214                                    PCI_EXP_SLTSTA, event)) {
215         return;
216     }
217     hotplug_event_notify(dev);
218 }
219 
220 static void pcie_cap_slot_hotplug_common(PCIDevice *hotplug_dev,
221                                          DeviceState *dev,
222                                          uint8_t **exp_cap, Error **errp)
223 {
224     *exp_cap = hotplug_dev->config + hotplug_dev->exp.exp_cap;
225     uint16_t sltsta = pci_get_word(*exp_cap + PCI_EXP_SLTSTA);
226 
227     PCIE_DEV_PRINTF(PCI_DEVICE(dev), "hotplug state: 0x%x\n", sltsta);
228     if (sltsta & PCI_EXP_SLTSTA_EIS) {
229         /* the slot is electromechanically locked.
230          * This error is propagated up to qdev and then to HMP/QMP.
231          */
232         error_setg_errno(errp, -EBUSY, "slot is electromechanically locked");
233     }
234 }
235 
236 void pcie_cap_slot_hotplug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
237                               Error **errp)
238 {
239     uint8_t *exp_cap;
240     PCIDevice *pci_dev = PCI_DEVICE(dev);
241 
242     pcie_cap_slot_hotplug_common(PCI_DEVICE(hotplug_dev), dev, &exp_cap, errp);
243 
244     /* Don't send event when device is enabled during qemu machine creation:
245      * it is present on boot, no hotplug event is necessary. We do send an
246      * event when the device is disabled later. */
247     if (!dev->hotplugged) {
248         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
249                                    PCI_EXP_SLTSTA_PDS);
250         return;
251     }
252 
253     /* TODO: multifunction hot-plug.
254      * Right now, only a device of function = 0 is allowed to be
255      * hot plugged/unplugged.
256      */
257     assert(PCI_FUNC(pci_dev->devfn) == 0);
258 
259     pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
260                                PCI_EXP_SLTSTA_PDS);
261     pcie_cap_slot_event(PCI_DEVICE(hotplug_dev),
262                         PCI_EXP_HP_EV_PDC | PCI_EXP_HP_EV_ABP);
263 }
264 
265 void pcie_cap_slot_hot_unplug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
266                                  Error **errp)
267 {
268     uint8_t *exp_cap;
269 
270     pcie_cap_slot_hotplug_common(PCI_DEVICE(hotplug_dev), dev, &exp_cap, errp);
271 
272     pcie_cap_slot_push_attention_button(PCI_DEVICE(hotplug_dev));
273 }
274 
275 /* pci express slot for pci express root/downstream port
276    PCI express capability slot registers */
277 void pcie_cap_slot_init(PCIDevice *dev, uint16_t slot)
278 {
279     uint32_t pos = dev->exp.exp_cap;
280 
281     pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_FLAGS,
282                                PCI_EXP_FLAGS_SLOT);
283 
284     pci_long_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCAP,
285                                  ~PCI_EXP_SLTCAP_PSN);
286     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
287                                (slot << PCI_EXP_SLTCAP_PSN_SHIFT) |
288                                PCI_EXP_SLTCAP_EIP |
289                                PCI_EXP_SLTCAP_HPS |
290                                PCI_EXP_SLTCAP_HPC |
291                                PCI_EXP_SLTCAP_PIP |
292                                PCI_EXP_SLTCAP_AIP |
293                                PCI_EXP_SLTCAP_ABP);
294 
295     if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
296         pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
297                                    PCI_EXP_SLTCAP_PCP);
298         pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
299                                      PCI_EXP_SLTCTL_PCC);
300         pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
301                                    PCI_EXP_SLTCTL_PCC);
302     }
303 
304     pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
305                                  PCI_EXP_SLTCTL_PIC |
306                                  PCI_EXP_SLTCTL_AIC);
307     pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCTL,
308                                PCI_EXP_SLTCTL_PIC_OFF |
309                                PCI_EXP_SLTCTL_AIC_OFF);
310     pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
311                                PCI_EXP_SLTCTL_PIC |
312                                PCI_EXP_SLTCTL_AIC |
313                                PCI_EXP_SLTCTL_HPIE |
314                                PCI_EXP_SLTCTL_CCIE |
315                                PCI_EXP_SLTCTL_PDCE |
316                                PCI_EXP_SLTCTL_ABPE);
317     /* Although reading PCI_EXP_SLTCTL_EIC returns always 0,
318      * make the bit writable here in order to detect 1b is written.
319      * pcie_cap_slot_write_config() test-and-clear the bit, so
320      * this bit always returns 0 to the guest.
321      */
322     pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
323                                PCI_EXP_SLTCTL_EIC);
324 
325     pci_word_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_SLTSTA,
326                                PCI_EXP_HP_EV_SUPPORTED);
327 
328     dev->exp.hpev_notified = false;
329 
330     qbus_set_hotplug_handler(BUS(pci_bridge_get_sec_bus(PCI_BRIDGE(dev))),
331                              DEVICE(dev), NULL);
332 }
333 
334 void pcie_cap_slot_reset(PCIDevice *dev)
335 {
336     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
337     uint8_t port_type = pcie_cap_get_type(dev);
338 
339     assert(port_type == PCI_EXP_TYPE_DOWNSTREAM ||
340            port_type == PCI_EXP_TYPE_ROOT_PORT);
341 
342     PCIE_DEV_PRINTF(dev, "reset\n");
343 
344     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
345                                  PCI_EXP_SLTCTL_EIC |
346                                  PCI_EXP_SLTCTL_PIC |
347                                  PCI_EXP_SLTCTL_AIC |
348                                  PCI_EXP_SLTCTL_HPIE |
349                                  PCI_EXP_SLTCTL_CCIE |
350                                  PCI_EXP_SLTCTL_PDCE |
351                                  PCI_EXP_SLTCTL_ABPE);
352     pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
353                                PCI_EXP_SLTCTL_AIC_OFF);
354 
355     if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
356         /* Downstream ports enforce device number 0. */
357         bool populated = pci_bridge_get_sec_bus(PCI_BRIDGE(dev))->devices[0];
358         uint16_t pic;
359 
360         if (populated) {
361             pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
362                                          PCI_EXP_SLTCTL_PCC);
363         } else {
364             pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
365                                        PCI_EXP_SLTCTL_PCC);
366         }
367 
368         pic = populated ? PCI_EXP_SLTCTL_PIC_ON : PCI_EXP_SLTCTL_PIC_OFF;
369         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL, pic);
370     }
371 
372     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
373                                  PCI_EXP_SLTSTA_EIS |/* on reset,
374                                                         the lock is released */
375                                  PCI_EXP_SLTSTA_CC |
376                                  PCI_EXP_SLTSTA_PDC |
377                                  PCI_EXP_SLTSTA_ABP);
378 
379     hotplug_event_update_event_status(dev);
380 }
381 
382 static void pcie_unplug_device(PCIBus *bus, PCIDevice *dev, void *opaque)
383 {
384     object_unparent(OBJECT(dev));
385 }
386 
387 void pcie_cap_slot_write_config(PCIDevice *dev,
388                                 uint32_t addr, uint32_t val, int len)
389 {
390     uint32_t pos = dev->exp.exp_cap;
391     uint8_t *exp_cap = dev->config + pos;
392     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
393 
394     if (ranges_overlap(addr, len, pos + PCI_EXP_SLTSTA, 2)) {
395         hotplug_event_clear(dev);
396     }
397 
398     if (!ranges_overlap(addr, len, pos + PCI_EXP_SLTCTL, 2)) {
399         return;
400     }
401 
402     if (pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
403                                      PCI_EXP_SLTCTL_EIC)) {
404         sltsta ^= PCI_EXP_SLTSTA_EIS; /* toggle PCI_EXP_SLTSTA_EIS bit */
405         pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
406         PCIE_DEV_PRINTF(dev, "PCI_EXP_SLTCTL_EIC: "
407                         "sltsta -> 0x%02"PRIx16"\n",
408                         sltsta);
409     }
410 
411     /*
412      * If the slot is polulated, power indicator is off and power
413      * controller is off, it is safe to detach the devices.
414      */
415     if ((sltsta & PCI_EXP_SLTSTA_PDS) && (val & PCI_EXP_SLTCTL_PCC) &&
416         ((val & PCI_EXP_SLTCTL_PIC_OFF) == PCI_EXP_SLTCTL_PIC_OFF)) {
417             PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
418             pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
419                                 pcie_unplug_device, NULL);
420 
421             pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
422                                          PCI_EXP_SLTSTA_PDS);
423             pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
424                                        PCI_EXP_SLTSTA_PDC);
425     }
426 
427     hotplug_event_notify(dev);
428 
429     /*
430      * 6.7.3.2 Command Completed Events
431      *
432      * Software issues a command to a hot-plug capable Downstream Port by
433      * issuing a write transaction that targets any portion of the Port’s Slot
434      * Control register. A single write to the Slot Control register is
435      * considered to be a single command, even if the write affects more than
436      * one field in the Slot Control register. In response to this transaction,
437      * the Port must carry out the requested actions and then set the
438      * associated status field for the command completed event. */
439 
440     /* Real hardware might take a while to complete requested command because
441      * physical movement would be involved like locking the electromechanical
442      * lock.  However in our case, command is completed instantaneously above,
443      * so send a command completion event right now.
444      */
445     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_CCI);
446 }
447 
448 int pcie_cap_slot_post_load(void *opaque, int version_id)
449 {
450     PCIDevice *dev = opaque;
451     hotplug_event_update_event_status(dev);
452     return 0;
453 }
454 
455 void pcie_cap_slot_push_attention_button(PCIDevice *dev)
456 {
457     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_ABP);
458 }
459 
460 /* root control/capabilities/status. PME isn't emulated for now */
461 void pcie_cap_root_init(PCIDevice *dev)
462 {
463     pci_set_word(dev->wmask + dev->exp.exp_cap + PCI_EXP_RTCTL,
464                  PCI_EXP_RTCTL_SECEE | PCI_EXP_RTCTL_SENFEE |
465                  PCI_EXP_RTCTL_SEFEE);
466 }
467 
468 void pcie_cap_root_reset(PCIDevice *dev)
469 {
470     pci_set_word(dev->config + dev->exp.exp_cap + PCI_EXP_RTCTL, 0);
471 }
472 
473 /* function level reset(FLR) */
474 void pcie_cap_flr_init(PCIDevice *dev)
475 {
476     pci_long_test_and_set_mask(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCAP,
477                                PCI_EXP_DEVCAP_FLR);
478 
479     /* Although reading BCR_FLR returns always 0,
480      * the bit is made writable here in order to detect the 1b is written
481      * pcie_cap_flr_write_config() test-and-clear the bit, so
482      * this bit always returns 0 to the guest.
483      */
484     pci_word_test_and_set_mask(dev->wmask + dev->exp.exp_cap + PCI_EXP_DEVCTL,
485                                PCI_EXP_DEVCTL_BCR_FLR);
486 }
487 
488 void pcie_cap_flr_write_config(PCIDevice *dev,
489                                uint32_t addr, uint32_t val, int len)
490 {
491     uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
492     if (pci_get_word(devctl) & PCI_EXP_DEVCTL_BCR_FLR) {
493         /* Clear PCI_EXP_DEVCTL_BCR_FLR after invoking the reset handler
494            so the handler can detect FLR by looking at this bit. */
495         pci_device_reset(dev);
496         pci_word_test_and_clear_mask(devctl, PCI_EXP_DEVCTL_BCR_FLR);
497     }
498 }
499 
500 /* Alternative Routing-ID Interpretation (ARI) */
501 /* ari forwarding support for down stream port */
502 void pcie_cap_ari_init(PCIDevice *dev)
503 {
504     uint32_t pos = dev->exp.exp_cap;
505     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP2,
506                                PCI_EXP_DEVCAP2_ARI);
507     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL2,
508                                PCI_EXP_DEVCTL2_ARI);
509 }
510 
511 void pcie_cap_ari_reset(PCIDevice *dev)
512 {
513     uint8_t *devctl2 = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2;
514     pci_long_test_and_clear_mask(devctl2, PCI_EXP_DEVCTL2_ARI);
515 }
516 
517 bool pcie_cap_is_ari_enabled(const PCIDevice *dev)
518 {
519     if (!pci_is_express(dev)) {
520         return false;
521     }
522     if (!dev->exp.exp_cap) {
523         return false;
524     }
525 
526     return pci_get_long(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2) &
527         PCI_EXP_DEVCTL2_ARI;
528 }
529 
530 /**************************************************************************
531  * pci express extended capability allocation functions
532  * uint16_t ext_cap_id (16 bit)
533  * uint8_t cap_ver (4 bit)
534  * uint16_t cap_offset (12 bit)
535  * uint16_t ext_cap_size
536  */
537 
538 static uint16_t pcie_find_capability_list(PCIDevice *dev, uint16_t cap_id,
539                                           uint16_t *prev_p)
540 {
541     uint16_t prev = 0;
542     uint16_t next;
543     uint32_t header = pci_get_long(dev->config + PCI_CONFIG_SPACE_SIZE);
544 
545     if (!header) {
546         /* no extended capability */
547         next = 0;
548         goto out;
549     }
550     for (next = PCI_CONFIG_SPACE_SIZE; next;
551          prev = next, next = PCI_EXT_CAP_NEXT(header)) {
552 
553         assert(next >= PCI_CONFIG_SPACE_SIZE);
554         assert(next <= PCIE_CONFIG_SPACE_SIZE - 8);
555 
556         header = pci_get_long(dev->config + next);
557         if (PCI_EXT_CAP_ID(header) == cap_id) {
558             break;
559         }
560     }
561 
562 out:
563     if (prev_p) {
564         *prev_p = prev;
565     }
566     return next;
567 }
568 
569 uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id)
570 {
571     return pcie_find_capability_list(dev, cap_id, NULL);
572 }
573 
574 static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next)
575 {
576     uint32_t header = pci_get_long(dev->config + pos);
577     assert(!(next & (PCI_EXT_CAP_ALIGN - 1)));
578     header = (header & ~PCI_EXT_CAP_NEXT_MASK) |
579         ((next << PCI_EXT_CAP_NEXT_SHIFT) & PCI_EXT_CAP_NEXT_MASK);
580     pci_set_long(dev->config + pos, header);
581 }
582 
583 /*
584  * caller must supply valid (offset, size) * such that the range shouldn't
585  * overlap with other capability or other registers.
586  * This function doesn't check it.
587  */
588 void pcie_add_capability(PCIDevice *dev,
589                          uint16_t cap_id, uint8_t cap_ver,
590                          uint16_t offset, uint16_t size)
591 {
592     uint32_t header;
593     uint16_t next;
594 
595     assert(offset >= PCI_CONFIG_SPACE_SIZE);
596     assert(offset < offset + size);
597     assert(offset + size < PCIE_CONFIG_SPACE_SIZE);
598     assert(size >= 8);
599     assert(pci_is_express(dev));
600 
601     if (offset == PCI_CONFIG_SPACE_SIZE) {
602         header = pci_get_long(dev->config + offset);
603         next = PCI_EXT_CAP_NEXT(header);
604     } else {
605         uint16_t prev;
606 
607         /* 0 is reserved cap id. use internally to find the last capability
608            in the linked list */
609         next = pcie_find_capability_list(dev, 0, &prev);
610 
611         assert(prev >= PCI_CONFIG_SPACE_SIZE);
612         assert(next == 0);
613         pcie_ext_cap_set_next(dev, prev, offset);
614     }
615     pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, next));
616 
617     /* Make capability read-only by default */
618     memset(dev->wmask + offset, 0, size);
619     memset(dev->w1cmask + offset, 0, size);
620     /* Check capability by default */
621     memset(dev->cmask + offset, 0xFF, size);
622 }
623 
624 /**************************************************************************
625  * pci express extended capability helper functions
626  */
627 
628 /* ARI */
629 void pcie_ari_init(PCIDevice *dev, uint16_t offset, uint16_t nextfn)
630 {
631     pcie_add_capability(dev, PCI_EXT_CAP_ID_ARI, PCI_ARI_VER,
632                         offset, PCI_ARI_SIZEOF);
633     pci_set_long(dev->config + offset + PCI_ARI_CAP, PCI_ARI_CAP_NFN(nextfn));
634 }
635