xref: /openbmc/qemu/hw/pci/pcie.c (revision f1f7e4bf)
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 
30 //#define DEBUG_PCIE
31 #ifdef DEBUG_PCIE
32 # define PCIE_DPRINTF(fmt, ...)                                         \
33     fprintf(stderr, "%s:%d " fmt, __func__, __LINE__, ## __VA_ARGS__)
34 #else
35 # define PCIE_DPRINTF(fmt, ...) do {} while (0)
36 #endif
37 #define PCIE_DEV_PRINTF(dev, fmt, ...)                                  \
38     PCIE_DPRINTF("%s:%x "fmt, (dev)->name, (dev)->devfn, ## __VA_ARGS__)
39 
40 
41 /***************************************************************************
42  * pci express capability helper functions
43  */
44 int pcie_cap_init(PCIDevice *dev, uint8_t offset, uint8_t type, uint8_t port)
45 {
46     int pos;
47     uint8_t *exp_cap;
48 
49     assert(pci_is_express(dev));
50 
51     pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
52                                  PCI_EXP_VER2_SIZEOF);
53     if (pos < 0) {
54         return pos;
55     }
56     dev->exp.exp_cap = pos;
57     exp_cap = dev->config + pos;
58 
59     /* capability register
60        interrupt message number defaults to 0 */
61     pci_set_word(exp_cap + PCI_EXP_FLAGS,
62                  ((type << PCI_EXP_FLAGS_TYPE_SHIFT) & PCI_EXP_FLAGS_TYPE) |
63                  PCI_EXP_FLAGS_VER2);
64 
65     /* device capability register
66      * table 7-12:
67      * roll based error reporting bit must be set by all
68      * Functions conforming to the ECN, PCI Express Base
69      * Specification, Revision 1.1., or subsequent PCI Express Base
70      * Specification revisions.
71      */
72     pci_set_long(exp_cap + PCI_EXP_DEVCAP, PCI_EXP_DEVCAP_RBER);
73 
74     pci_set_long(exp_cap + PCI_EXP_LNKCAP,
75                  (port << PCI_EXP_LNKCAP_PN_SHIFT) |
76                  PCI_EXP_LNKCAP_ASPMS_0S |
77                  PCI_EXP_LNK_MLW_1 |
78                  PCI_EXP_LNK_LS_25);
79 
80     pci_set_word(exp_cap + PCI_EXP_LNKSTA,
81                  PCI_EXP_LNK_MLW_1 | PCI_EXP_LNK_LS_25 |PCI_EXP_LNKSTA_DLLLA);
82 
83     pci_set_long(exp_cap + PCI_EXP_DEVCAP2,
84                  PCI_EXP_DEVCAP2_EFF | PCI_EXP_DEVCAP2_EETLPP);
85 
86     pci_set_word(dev->wmask + pos + PCI_EXP_DEVCTL2, PCI_EXP_DEVCTL2_EETLPPB);
87     return pos;
88 }
89 
90 int pcie_endpoint_cap_init(PCIDevice *dev, uint8_t offset)
91 {
92     uint8_t type = PCI_EXP_TYPE_ENDPOINT;
93 
94     /*
95      * Windows guests will report Code 10, device cannot start, if
96      * a regular Endpoint type is exposed on a root complex.  These
97      * should instead be Root Complex Integrated Endpoints.
98      */
99     if (pci_bus_is_express(dev->bus) && pci_bus_is_root(dev->bus)) {
100         type = PCI_EXP_TYPE_RC_END;
101     }
102 
103     return pcie_cap_init(dev, offset, type, 0);
104 }
105 
106 void pcie_cap_exit(PCIDevice *dev)
107 {
108     pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER2_SIZEOF);
109 }
110 
111 uint8_t pcie_cap_get_type(const PCIDevice *dev)
112 {
113     uint32_t pos = dev->exp.exp_cap;
114     assert(pos > 0);
115     return (pci_get_word(dev->config + pos + PCI_EXP_FLAGS) &
116             PCI_EXP_FLAGS_TYPE) >> PCI_EXP_FLAGS_TYPE_SHIFT;
117 }
118 
119 /* MSI/MSI-X */
120 /* pci express interrupt message number */
121 /* 7.8.2 PCI Express Capabilities Register: Interrupt Message Number */
122 void pcie_cap_flags_set_vector(PCIDevice *dev, uint8_t vector)
123 {
124     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
125     assert(vector < 32);
126     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_FLAGS, PCI_EXP_FLAGS_IRQ);
127     pci_word_test_and_set_mask(exp_cap + PCI_EXP_FLAGS,
128                                vector << PCI_EXP_FLAGS_IRQ_SHIFT);
129 }
130 
131 uint8_t pcie_cap_flags_get_vector(PCIDevice *dev)
132 {
133     return (pci_get_word(dev->config + dev->exp.exp_cap + PCI_EXP_FLAGS) &
134             PCI_EXP_FLAGS_IRQ) >> PCI_EXP_FLAGS_IRQ_SHIFT;
135 }
136 
137 void pcie_cap_deverr_init(PCIDevice *dev)
138 {
139     uint32_t pos = dev->exp.exp_cap;
140     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP,
141                                PCI_EXP_DEVCAP_RBER);
142     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL,
143                                PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
144                                PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
145     pci_long_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_DEVSTA,
146                                PCI_EXP_DEVSTA_CED | PCI_EXP_DEVSTA_NFED |
147                                PCI_EXP_DEVSTA_FED | PCI_EXP_DEVSTA_URD);
148 }
149 
150 void pcie_cap_deverr_reset(PCIDevice *dev)
151 {
152     uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
153     pci_long_test_and_clear_mask(devctl,
154                                  PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
155                                  PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
156 }
157 
158 static void hotplug_event_update_event_status(PCIDevice *dev)
159 {
160     uint32_t pos = dev->exp.exp_cap;
161     uint8_t *exp_cap = dev->config + pos;
162     uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
163     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
164 
165     dev->exp.hpev_notified = (sltctl & PCI_EXP_SLTCTL_HPIE) &&
166         (sltsta & sltctl & PCI_EXP_HP_EV_SUPPORTED);
167 }
168 
169 static void hotplug_event_notify(PCIDevice *dev)
170 {
171     bool prev = dev->exp.hpev_notified;
172 
173     hotplug_event_update_event_status(dev);
174 
175     if (prev == dev->exp.hpev_notified) {
176         return;
177     }
178 
179     /* Note: the logic above does not take into account whether interrupts
180      * are masked. The result is that interrupt will be sent when it is
181      * subsequently unmasked. This appears to be legal: Section 6.7.3.4:
182      * The Port may optionally send an MSI when there are hot-plug events that
183      * occur while interrupt generation is disabled, and interrupt generation is
184      * subsequently enabled. */
185     if (msix_enabled(dev)) {
186         msix_notify(dev, pcie_cap_flags_get_vector(dev));
187     } else if (msi_enabled(dev)) {
188         msi_notify(dev, pcie_cap_flags_get_vector(dev));
189     } else {
190         pci_set_irq(dev, dev->exp.hpev_notified);
191     }
192 }
193 
194 static void hotplug_event_clear(PCIDevice *dev)
195 {
196     hotplug_event_update_event_status(dev);
197     if (!msix_enabled(dev) && !msi_enabled(dev) && !dev->exp.hpev_notified) {
198         pci_irq_deassert(dev);
199     }
200 }
201 
202 /*
203  * A PCI Express Hot-Plug Event has occurred, so update slot status register
204  * and notify OS of the event if necessary.
205  *
206  * 6.7.3 PCI Express Hot-Plug Events
207  * 6.7.3.4 Software Notification of Hot-Plug Events
208  */
209 static void pcie_cap_slot_event(PCIDevice *dev, PCIExpressHotPlugEvent event)
210 {
211     /* Minor optimization: if nothing changed - no event is needed. */
212     if (pci_word_test_and_set_mask(dev->config + dev->exp.exp_cap +
213                                    PCI_EXP_SLTSTA, event)) {
214         return;
215     }
216     hotplug_event_notify(dev);
217 }
218 
219 static void pcie_cap_slot_hotplug_common(PCIDevice *hotplug_dev,
220                                          DeviceState *dev,
221                                          uint8_t **exp_cap, Error **errp)
222 {
223     *exp_cap = hotplug_dev->config + hotplug_dev->exp.exp_cap;
224     uint16_t sltsta = pci_get_word(*exp_cap + PCI_EXP_SLTSTA);
225 
226     PCIE_DEV_PRINTF(PCI_DEVICE(dev), "hotplug state: 0x%x\n", sltsta);
227     if (sltsta & PCI_EXP_SLTSTA_EIS) {
228         /* the slot is electromechanically locked.
229          * This error is propagated up to qdev and then to HMP/QMP.
230          */
231         error_setg_errno(errp, EBUSY, "slot is electromechanically locked");
232     }
233 }
234 
235 void pcie_cap_slot_hotplug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
236                               Error **errp)
237 {
238     uint8_t *exp_cap;
239     PCIDevice *pci_dev = PCI_DEVICE(dev);
240 
241     pcie_cap_slot_hotplug_common(PCI_DEVICE(hotplug_dev), dev, &exp_cap, errp);
242 
243     /* Don't send event when device is enabled during qemu machine creation:
244      * it is present on boot, no hotplug event is necessary. We do send an
245      * event when the device is disabled later. */
246     if (!dev->hotplugged) {
247         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
248                                    PCI_EXP_SLTSTA_PDS);
249         return;
250     }
251 
252     /* To enable multifunction hot-plug, we just ensure the function
253      * 0 added last. When function 0 is added, we set the sltsta and
254      * inform OS via event notification.
255      */
256     if (pci_get_function_0(pci_dev)) {
257         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
258                                    PCI_EXP_SLTSTA_PDS);
259         pcie_cap_slot_event(PCI_DEVICE(hotplug_dev),
260                             PCI_EXP_HP_EV_PDC | PCI_EXP_HP_EV_ABP);
261     }
262 }
263 
264 static void pcie_unplug_device(PCIBus *bus, PCIDevice *dev, void *opaque)
265 {
266     object_unparent(OBJECT(dev));
267 }
268 
269 void pcie_cap_slot_hot_unplug_request_cb(HotplugHandler *hotplug_dev,
270                                          DeviceState *dev, Error **errp)
271 {
272     uint8_t *exp_cap;
273     PCIDevice *pci_dev = PCI_DEVICE(dev);
274     PCIBus *bus = pci_dev->bus;
275 
276     pcie_cap_slot_hotplug_common(PCI_DEVICE(hotplug_dev), dev, &exp_cap, errp);
277 
278     /* In case user cancel the operation of multi-function hot-add,
279      * remove the function that is unexposed to guest individually,
280      * without interaction with guest.
281      */
282     if (pci_dev->devfn &&
283         !bus->devices[0]) {
284         pcie_unplug_device(bus, pci_dev, NULL);
285 
286         return;
287     }
288 
289     pcie_cap_slot_push_attention_button(PCI_DEVICE(hotplug_dev));
290 }
291 
292 /* pci express slot for pci express root/downstream port
293    PCI express capability slot registers */
294 void pcie_cap_slot_init(PCIDevice *dev, uint16_t slot)
295 {
296     uint32_t pos = dev->exp.exp_cap;
297 
298     pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_FLAGS,
299                                PCI_EXP_FLAGS_SLOT);
300 
301     pci_long_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCAP,
302                                  ~PCI_EXP_SLTCAP_PSN);
303     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
304                                (slot << PCI_EXP_SLTCAP_PSN_SHIFT) |
305                                PCI_EXP_SLTCAP_EIP |
306                                PCI_EXP_SLTCAP_HPS |
307                                PCI_EXP_SLTCAP_HPC |
308                                PCI_EXP_SLTCAP_PIP |
309                                PCI_EXP_SLTCAP_AIP |
310                                PCI_EXP_SLTCAP_ABP);
311 
312     if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
313         pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
314                                    PCI_EXP_SLTCAP_PCP);
315         pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
316                                      PCI_EXP_SLTCTL_PCC);
317         pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
318                                    PCI_EXP_SLTCTL_PCC);
319     }
320 
321     pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
322                                  PCI_EXP_SLTCTL_PIC |
323                                  PCI_EXP_SLTCTL_AIC);
324     pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCTL,
325                                PCI_EXP_SLTCTL_PIC_OFF |
326                                PCI_EXP_SLTCTL_AIC_OFF);
327     pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
328                                PCI_EXP_SLTCTL_PIC |
329                                PCI_EXP_SLTCTL_AIC |
330                                PCI_EXP_SLTCTL_HPIE |
331                                PCI_EXP_SLTCTL_CCIE |
332                                PCI_EXP_SLTCTL_PDCE |
333                                PCI_EXP_SLTCTL_ABPE);
334     /* Although reading PCI_EXP_SLTCTL_EIC returns always 0,
335      * make the bit writable here in order to detect 1b is written.
336      * pcie_cap_slot_write_config() test-and-clear the bit, so
337      * this bit always returns 0 to the guest.
338      */
339     pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
340                                PCI_EXP_SLTCTL_EIC);
341 
342     pci_word_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_SLTSTA,
343                                PCI_EXP_HP_EV_SUPPORTED);
344 
345     dev->exp.hpev_notified = false;
346 
347     qbus_set_hotplug_handler(BUS(pci_bridge_get_sec_bus(PCI_BRIDGE(dev))),
348                              DEVICE(dev), NULL);
349 }
350 
351 void pcie_cap_slot_reset(PCIDevice *dev)
352 {
353     uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
354     uint8_t port_type = pcie_cap_get_type(dev);
355 
356     assert(port_type == PCI_EXP_TYPE_DOWNSTREAM ||
357            port_type == PCI_EXP_TYPE_ROOT_PORT);
358 
359     PCIE_DEV_PRINTF(dev, "reset\n");
360 
361     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
362                                  PCI_EXP_SLTCTL_EIC |
363                                  PCI_EXP_SLTCTL_PIC |
364                                  PCI_EXP_SLTCTL_AIC |
365                                  PCI_EXP_SLTCTL_HPIE |
366                                  PCI_EXP_SLTCTL_CCIE |
367                                  PCI_EXP_SLTCTL_PDCE |
368                                  PCI_EXP_SLTCTL_ABPE);
369     pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
370                                PCI_EXP_SLTCTL_AIC_OFF);
371 
372     if (dev->cap_present & QEMU_PCIE_SLTCAP_PCP) {
373         /* Downstream ports enforce device number 0. */
374         bool populated = pci_bridge_get_sec_bus(PCI_BRIDGE(dev))->devices[0];
375         uint16_t pic;
376 
377         if (populated) {
378             pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
379                                          PCI_EXP_SLTCTL_PCC);
380         } else {
381             pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
382                                        PCI_EXP_SLTCTL_PCC);
383         }
384 
385         pic = populated ? PCI_EXP_SLTCTL_PIC_ON : PCI_EXP_SLTCTL_PIC_OFF;
386         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL, pic);
387     }
388 
389     pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
390                                  PCI_EXP_SLTSTA_EIS |/* on reset,
391                                                         the lock is released */
392                                  PCI_EXP_SLTSTA_CC |
393                                  PCI_EXP_SLTSTA_PDC |
394                                  PCI_EXP_SLTSTA_ABP);
395 
396     hotplug_event_update_event_status(dev);
397 }
398 
399 void pcie_cap_slot_write_config(PCIDevice *dev,
400                                 uint32_t addr, uint32_t val, int len)
401 {
402     uint32_t pos = dev->exp.exp_cap;
403     uint8_t *exp_cap = dev->config + pos;
404     uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
405 
406     if (ranges_overlap(addr, len, pos + PCI_EXP_SLTSTA, 2)) {
407         hotplug_event_clear(dev);
408     }
409 
410     if (!ranges_overlap(addr, len, pos + PCI_EXP_SLTCTL, 2)) {
411         return;
412     }
413 
414     if (pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
415                                      PCI_EXP_SLTCTL_EIC)) {
416         sltsta ^= PCI_EXP_SLTSTA_EIS; /* toggle PCI_EXP_SLTSTA_EIS bit */
417         pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
418         PCIE_DEV_PRINTF(dev, "PCI_EXP_SLTCTL_EIC: "
419                         "sltsta -> 0x%02"PRIx16"\n",
420                         sltsta);
421     }
422 
423     /*
424      * If the slot is polulated, power indicator is off and power
425      * controller is off, it is safe to detach the devices.
426      */
427     if ((sltsta & PCI_EXP_SLTSTA_PDS) && (val & PCI_EXP_SLTCTL_PCC) &&
428         ((val & PCI_EXP_SLTCTL_PIC_OFF) == PCI_EXP_SLTCTL_PIC_OFF)) {
429         PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev));
430         pci_for_each_device(sec_bus, pci_bus_num(sec_bus),
431                             pcie_unplug_device, NULL);
432 
433         pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
434                                      PCI_EXP_SLTSTA_PDS);
435         pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
436                                        PCI_EXP_SLTSTA_PDC);
437     }
438 
439     hotplug_event_notify(dev);
440 
441     /*
442      * 6.7.3.2 Command Completed Events
443      *
444      * Software issues a command to a hot-plug capable Downstream Port by
445      * issuing a write transaction that targets any portion of the Port’s Slot
446      * Control register. A single write to the Slot Control register is
447      * considered to be a single command, even if the write affects more than
448      * one field in the Slot Control register. In response to this transaction,
449      * the Port must carry out the requested actions and then set the
450      * associated status field for the command completed event. */
451 
452     /* Real hardware might take a while to complete requested command because
453      * physical movement would be involved like locking the electromechanical
454      * lock.  However in our case, command is completed instantaneously above,
455      * so send a command completion event right now.
456      */
457     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_CCI);
458 }
459 
460 int pcie_cap_slot_post_load(void *opaque, int version_id)
461 {
462     PCIDevice *dev = opaque;
463     hotplug_event_update_event_status(dev);
464     return 0;
465 }
466 
467 void pcie_cap_slot_push_attention_button(PCIDevice *dev)
468 {
469     pcie_cap_slot_event(dev, PCI_EXP_HP_EV_ABP);
470 }
471 
472 /* root control/capabilities/status. PME isn't emulated for now */
473 void pcie_cap_root_init(PCIDevice *dev)
474 {
475     pci_set_word(dev->wmask + dev->exp.exp_cap + PCI_EXP_RTCTL,
476                  PCI_EXP_RTCTL_SECEE | PCI_EXP_RTCTL_SENFEE |
477                  PCI_EXP_RTCTL_SEFEE);
478 }
479 
480 void pcie_cap_root_reset(PCIDevice *dev)
481 {
482     pci_set_word(dev->config + dev->exp.exp_cap + PCI_EXP_RTCTL, 0);
483 }
484 
485 /* function level reset(FLR) */
486 void pcie_cap_flr_init(PCIDevice *dev)
487 {
488     pci_long_test_and_set_mask(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCAP,
489                                PCI_EXP_DEVCAP_FLR);
490 
491     /* Although reading BCR_FLR returns always 0,
492      * the bit is made writable here in order to detect the 1b is written
493      * pcie_cap_flr_write_config() test-and-clear the bit, so
494      * this bit always returns 0 to the guest.
495      */
496     pci_word_test_and_set_mask(dev->wmask + dev->exp.exp_cap + PCI_EXP_DEVCTL,
497                                PCI_EXP_DEVCTL_BCR_FLR);
498 }
499 
500 void pcie_cap_flr_write_config(PCIDevice *dev,
501                                uint32_t addr, uint32_t val, int len)
502 {
503     uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
504     if (pci_get_word(devctl) & PCI_EXP_DEVCTL_BCR_FLR) {
505         /* Clear PCI_EXP_DEVCTL_BCR_FLR after invoking the reset handler
506            so the handler can detect FLR by looking at this bit. */
507         pci_device_reset(dev);
508         pci_word_test_and_clear_mask(devctl, PCI_EXP_DEVCTL_BCR_FLR);
509     }
510 }
511 
512 /* Alternative Routing-ID Interpretation (ARI)
513  * forwarding support for root and downstream ports
514  */
515 void pcie_cap_arifwd_init(PCIDevice *dev)
516 {
517     uint32_t pos = dev->exp.exp_cap;
518     pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP2,
519                                PCI_EXP_DEVCAP2_ARI);
520     pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL2,
521                                PCI_EXP_DEVCTL2_ARI);
522 }
523 
524 void pcie_cap_arifwd_reset(PCIDevice *dev)
525 {
526     uint8_t *devctl2 = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2;
527     pci_long_test_and_clear_mask(devctl2, PCI_EXP_DEVCTL2_ARI);
528 }
529 
530 bool pcie_cap_is_arifwd_enabled(const PCIDevice *dev)
531 {
532     if (!pci_is_express(dev)) {
533         return false;
534     }
535     if (!dev->exp.exp_cap) {
536         return false;
537     }
538 
539     return pci_get_long(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2) &
540         PCI_EXP_DEVCTL2_ARI;
541 }
542 
543 /**************************************************************************
544  * pci express extended capability list management functions
545  * uint16_t ext_cap_id (16 bit)
546  * uint8_t cap_ver (4 bit)
547  * uint16_t cap_offset (12 bit)
548  * uint16_t ext_cap_size
549  */
550 
551 static uint16_t pcie_find_capability_list(PCIDevice *dev, uint16_t cap_id,
552                                           uint16_t *prev_p)
553 {
554     uint16_t prev = 0;
555     uint16_t next;
556     uint32_t header = pci_get_long(dev->config + PCI_CONFIG_SPACE_SIZE);
557 
558     if (!header) {
559         /* no extended capability */
560         next = 0;
561         goto out;
562     }
563     for (next = PCI_CONFIG_SPACE_SIZE; next;
564          prev = next, next = PCI_EXT_CAP_NEXT(header)) {
565 
566         assert(next >= PCI_CONFIG_SPACE_SIZE);
567         assert(next <= PCIE_CONFIG_SPACE_SIZE - 8);
568 
569         header = pci_get_long(dev->config + next);
570         if (PCI_EXT_CAP_ID(header) == cap_id) {
571             break;
572         }
573     }
574 
575 out:
576     if (prev_p) {
577         *prev_p = prev;
578     }
579     return next;
580 }
581 
582 uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id)
583 {
584     return pcie_find_capability_list(dev, cap_id, NULL);
585 }
586 
587 static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next)
588 {
589     uint32_t header = pci_get_long(dev->config + pos);
590     assert(!(next & (PCI_EXT_CAP_ALIGN - 1)));
591     header = (header & ~PCI_EXT_CAP_NEXT_MASK) |
592         ((next << PCI_EXT_CAP_NEXT_SHIFT) & PCI_EXT_CAP_NEXT_MASK);
593     pci_set_long(dev->config + pos, header);
594 }
595 
596 /*
597  * caller must supply valid (offset, size) * such that the range shouldn't
598  * overlap with other capability or other registers.
599  * This function doesn't check it.
600  */
601 void pcie_add_capability(PCIDevice *dev,
602                          uint16_t cap_id, uint8_t cap_ver,
603                          uint16_t offset, uint16_t size)
604 {
605     uint32_t header;
606     uint16_t next;
607 
608     assert(offset >= PCI_CONFIG_SPACE_SIZE);
609     assert(offset < offset + size);
610     assert(offset + size < PCIE_CONFIG_SPACE_SIZE);
611     assert(size >= 8);
612     assert(pci_is_express(dev));
613 
614     if (offset == PCI_CONFIG_SPACE_SIZE) {
615         header = pci_get_long(dev->config + offset);
616         next = PCI_EXT_CAP_NEXT(header);
617     } else {
618         uint16_t prev;
619 
620         /* 0 is reserved cap id. use internally to find the last capability
621            in the linked list */
622         next = pcie_find_capability_list(dev, 0, &prev);
623 
624         assert(prev >= PCI_CONFIG_SPACE_SIZE);
625         assert(next == 0);
626         pcie_ext_cap_set_next(dev, prev, offset);
627     }
628     pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, next));
629 
630     /* Make capability read-only by default */
631     memset(dev->wmask + offset, 0, size);
632     memset(dev->w1cmask + offset, 0, size);
633     /* Check capability by default */
634     memset(dev->cmask + offset, 0xFF, size);
635 }
636 
637 /**************************************************************************
638  * pci express extended capability helper functions
639  */
640 
641 /* ARI */
642 void pcie_ari_init(PCIDevice *dev, uint16_t offset, uint16_t nextfn)
643 {
644     pcie_add_capability(dev, PCI_EXT_CAP_ID_ARI, PCI_ARI_VER,
645                         offset, PCI_ARI_SIZEOF);
646     pci_set_long(dev->config + offset + PCI_ARI_CAP, (nextfn & 0xff) << 8);
647 }
648