pci: allow hotplug removal of cold-plugged devices
[qemu-kvm/stefanha.git] / hw / pcie.c
blobf461c1cfbe813f9737e9e27eeb4b9ade1632cfa6
1 /*
2 * pcie.c
4 * Copyright (c) 2010 Isaku Yamahata <yamahata at valinux co jp>
5 * VA Linux Systems Japan K.K.
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.
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.
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/>.
21 #include "sysemu.h"
22 #include "range.h"
23 #include "pci_bridge.h"
24 #include "pcie.h"
25 #include "msix.h"
26 #include "msi.h"
27 #include "pci_internals.h"
28 #include "pcie_regs.h"
29 #include "range.h"
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__)
42 /***************************************************************************
43 * pci express capability helper functions
45 int pcie_cap_init(PCIDevice *dev, uint8_t offset, uint8_t type, uint8_t port)
47 int pos;
48 uint8_t *exp_cap;
50 assert(pci_is_express(dev));
52 pos = pci_add_capability(dev, PCI_CAP_ID_EXP, offset,
53 PCI_EXP_VER2_SIZEOF);
54 if (pos < 0) {
55 return pos;
57 dev->exp.exp_cap = pos;
58 exp_cap = dev->config + pos;
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);
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.
73 pci_set_long(exp_cap + PCI_EXP_DEVCAP, PCI_EXP_DEVCAP_RBER);
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);
81 pci_set_word(exp_cap + PCI_EXP_LNKSTA,
82 PCI_EXP_LNK_MLW_1 | PCI_EXP_LNK_LS_25);
84 pci_set_long(exp_cap + PCI_EXP_DEVCAP2,
85 PCI_EXP_DEVCAP2_EFF | PCI_EXP_DEVCAP2_EETLPP);
87 pci_set_word(dev->wmask + pos, PCI_EXP_DEVCTL2_EETLPPB);
88 return pos;
91 void pcie_cap_exit(PCIDevice *dev)
93 pci_del_capability(dev, PCI_CAP_ID_EXP, PCI_EXP_VER2_SIZEOF);
96 uint8_t pcie_cap_get_type(const PCIDevice *dev)
98 uint32_t pos = dev->exp.exp_cap;
99 assert(pos > 0);
100 return (pci_get_word(dev->config + pos + PCI_EXP_FLAGS) &
101 PCI_EXP_FLAGS_TYPE) >> PCI_EXP_FLAGS_TYPE_SHIFT;
104 /* MSI/MSI-X */
105 /* pci express interrupt message number */
106 /* 7.8.2 PCI Express Capabilities Register: Interrupt Message Number */
107 void pcie_cap_flags_set_vector(PCIDevice *dev, uint8_t vector)
109 uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
110 assert(vector < 32);
111 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_FLAGS, PCI_EXP_FLAGS_IRQ);
112 pci_word_test_and_set_mask(exp_cap + PCI_EXP_FLAGS,
113 vector << PCI_EXP_FLAGS_IRQ_SHIFT);
116 uint8_t pcie_cap_flags_get_vector(PCIDevice *dev)
118 return (pci_get_word(dev->config + dev->exp.exp_cap + PCI_EXP_FLAGS) &
119 PCI_EXP_FLAGS_IRQ) >> PCI_EXP_FLAGS_IRQ_SHIFT;
122 void pcie_cap_deverr_init(PCIDevice *dev)
124 uint32_t pos = dev->exp.exp_cap;
125 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP,
126 PCI_EXP_DEVCAP_RBER);
127 pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL,
128 PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
129 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
130 pci_long_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_DEVSTA,
131 PCI_EXP_DEVSTA_CED | PCI_EXP_DEVSTA_NFED |
132 PCI_EXP_DEVSTA_URD | PCI_EXP_DEVSTA_URD);
135 void pcie_cap_deverr_reset(PCIDevice *dev)
137 uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
138 pci_long_test_and_clear_mask(devctl,
139 PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE |
140 PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE);
143 static void hotplug_event_update_event_status(PCIDevice *dev)
145 uint32_t pos = dev->exp.exp_cap;
146 uint8_t *exp_cap = dev->config + pos;
147 uint16_t sltctl = pci_get_word(exp_cap + PCI_EXP_SLTCTL);
148 uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
150 dev->exp.hpev_notified = (sltctl & PCI_EXP_SLTCTL_HPIE) &&
151 (sltsta & sltctl & PCI_EXP_HP_EV_SUPPORTED);
154 static void hotplug_event_notify(PCIDevice *dev)
156 bool prev = dev->exp.hpev_notified;
158 hotplug_event_update_event_status(dev);
160 if (prev == dev->exp.hpev_notified) {
161 return;
164 /* Note: the logic above does not take into account whether interrupts
165 * are masked. The result is that interrupt will be sent when it is
166 * subsequently unmasked. This appears to be legal: Section 6.7.3.4:
167 * The Port may optionally send an MSI when there are hot-plug events that
168 * occur while interrupt generation is disabled, and interrupt generation is
169 * subsequently enabled. */
170 if (!pci_msi_enabled(dev)) {
171 qemu_set_irq(dev->irq[dev->exp.hpev_intx], dev->exp.hpev_notified);
172 } else if (dev->exp.hpev_notified) {
173 pci_msi_notify(dev, pcie_cap_flags_get_vector(dev));
178 * A PCI Express Hot-Plug Event has occured, so update slot status register
179 * and notify OS of the event if necessary.
181 * 6.7.3 PCI Express Hot-Plug Events
182 * 6.7.3.4 Software Notification of Hot-Plug Events
184 static void pcie_cap_slot_event(PCIDevice *dev, PCIExpressHotPlugEvent event)
186 /* Minor optimization: if nothing changed - no event is needed. */
187 if (pci_word_test_and_set_mask(dev->config + dev->exp.exp_cap +
188 PCI_EXP_SLTSTA, event)) {
189 return;
191 hotplug_event_notify(dev);
194 static int pcie_cap_slot_hotplug(DeviceState *qdev,
195 PCIDevice *pci_dev, PCIHotplugState state)
197 PCIDevice *d = DO_UPCAST(PCIDevice, qdev, qdev);
198 uint8_t *exp_cap = d->config + d->exp.exp_cap;
199 uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
201 /* Don't send event when device is enabled during qemu machine creation:
202 * it is present on boot, no hotplug event is necessary. We do send an
203 * event when the device is disabled later. */
204 if (state == PCI_COLDPLUG_ENABLED) {
205 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
206 PCI_EXP_SLTSTA_PDS);
207 return 0;
210 PCIE_DEV_PRINTF(pci_dev, "hotplug state: %d\n", state);
211 if (sltsta & PCI_EXP_SLTSTA_EIS) {
212 /* the slot is electromechanically locked.
213 * This error is propagated up to qdev and then to HMP/QMP.
215 return -EBUSY;
218 /* TODO: multifunction hot-plug.
219 * Right now, only a device of function = 0 is allowed to be
220 * hot plugged/unplugged.
222 assert(PCI_FUNC(pci_dev->devfn) == 0);
224 if (state == PCI_HOTPLUG_ENABLED) {
225 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTSTA,
226 PCI_EXP_SLTSTA_PDS);
227 pcie_cap_slot_event(d, PCI_EXP_HP_EV_PDC);
228 } else {
229 qdev_free(&pci_dev->qdev);
230 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
231 PCI_EXP_SLTSTA_PDS);
232 pcie_cap_slot_event(d, PCI_EXP_HP_EV_PDC);
234 return 0;
237 /* pci express slot for pci express root/downstream port
238 PCI express capability slot registers */
239 void pcie_cap_slot_init(PCIDevice *dev, uint16_t slot)
241 uint32_t pos = dev->exp.exp_cap;
243 pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_FLAGS,
244 PCI_EXP_FLAGS_SLOT);
246 pci_long_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCAP,
247 ~PCI_EXP_SLTCAP_PSN);
248 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCAP,
249 (slot << PCI_EXP_SLTCAP_PSN_SHIFT) |
250 PCI_EXP_SLTCAP_EIP |
251 PCI_EXP_SLTCAP_HPS |
252 PCI_EXP_SLTCAP_HPC |
253 PCI_EXP_SLTCAP_PIP |
254 PCI_EXP_SLTCAP_AIP |
255 PCI_EXP_SLTCAP_ABP);
257 pci_word_test_and_clear_mask(dev->config + pos + PCI_EXP_SLTCTL,
258 PCI_EXP_SLTCTL_PIC |
259 PCI_EXP_SLTCTL_AIC);
260 pci_word_test_and_set_mask(dev->config + pos + PCI_EXP_SLTCTL,
261 PCI_EXP_SLTCTL_PIC_OFF |
262 PCI_EXP_SLTCTL_AIC_OFF);
263 pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
264 PCI_EXP_SLTCTL_PIC |
265 PCI_EXP_SLTCTL_AIC |
266 PCI_EXP_SLTCTL_HPIE |
267 PCI_EXP_SLTCTL_CCIE |
268 PCI_EXP_SLTCTL_PDCE |
269 PCI_EXP_SLTCTL_ABPE);
270 /* Although reading PCI_EXP_SLTCTL_EIC returns always 0,
271 * make the bit writable here in order to detect 1b is written.
272 * pcie_cap_slot_write_config() test-and-clear the bit, so
273 * this bit always returns 0 to the guest.
275 pci_word_test_and_set_mask(dev->wmask + pos + PCI_EXP_SLTCTL,
276 PCI_EXP_SLTCTL_EIC);
278 pci_word_test_and_set_mask(dev->w1cmask + pos + PCI_EXP_SLTSTA,
279 PCI_EXP_HP_EV_SUPPORTED);
281 dev->exp.hpev_notified = false;
283 pci_bus_hotplug(pci_bridge_get_sec_bus(DO_UPCAST(PCIBridge, dev, dev)),
284 pcie_cap_slot_hotplug, &dev->qdev);
287 void pcie_cap_slot_reset(PCIDevice *dev)
289 uint8_t *exp_cap = dev->config + dev->exp.exp_cap;
291 PCIE_DEV_PRINTF(dev, "reset\n");
293 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
294 PCI_EXP_SLTCTL_EIC |
295 PCI_EXP_SLTCTL_PIC |
296 PCI_EXP_SLTCTL_AIC |
297 PCI_EXP_SLTCTL_HPIE |
298 PCI_EXP_SLTCTL_CCIE |
299 PCI_EXP_SLTCTL_PDCE |
300 PCI_EXP_SLTCTL_ABPE);
301 pci_word_test_and_set_mask(exp_cap + PCI_EXP_SLTCTL,
302 PCI_EXP_SLTCTL_PIC_OFF |
303 PCI_EXP_SLTCTL_AIC_OFF);
305 pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTSTA,
306 PCI_EXP_SLTSTA_EIS |/* on reset,
307 the lock is released */
308 PCI_EXP_SLTSTA_CC |
309 PCI_EXP_SLTSTA_PDC |
310 PCI_EXP_SLTSTA_ABP);
312 hotplug_event_update_event_status(dev);
315 void pcie_cap_slot_write_config(PCIDevice *dev,
316 uint32_t addr, uint32_t val, int len)
318 uint32_t pos = dev->exp.exp_cap;
319 uint8_t *exp_cap = dev->config + pos;
320 uint16_t sltsta = pci_get_word(exp_cap + PCI_EXP_SLTSTA);
322 if (!ranges_overlap(addr, len, pos + PCI_EXP_SLTCTL, 2)) {
323 return;
326 if (pci_word_test_and_clear_mask(exp_cap + PCI_EXP_SLTCTL,
327 PCI_EXP_SLTCTL_EIC)) {
328 sltsta ^= PCI_EXP_SLTSTA_EIS; /* toggle PCI_EXP_SLTSTA_EIS bit */
329 pci_set_word(exp_cap + PCI_EXP_SLTSTA, sltsta);
330 PCIE_DEV_PRINTF(dev, "PCI_EXP_SLTCTL_EIC: "
331 "sltsta -> 0x%02"PRIx16"\n",
332 sltsta);
335 hotplug_event_notify(dev);
338 * 6.7.3.2 Command Completed Events
340 * Software issues a command to a hot-plug capable Downstream Port by
341 * issuing a write transaction that targets any portion of the Port’s Slot
342 * Control register. A single write to the Slot Control register is
343 * considered to be a single command, even if the write affects more than
344 * one field in the Slot Control register. In response to this transaction,
345 * the Port must carry out the requested actions and then set the
346 * associated status field for the command completed event. */
348 /* Real hardware might take a while to complete requested command because
349 * physical movement would be involved like locking the electromechanical
350 * lock. However in our case, command is completed instantaneously above,
351 * so send a command completion event right now.
353 pcie_cap_slot_event(dev, PCI_EXP_HP_EV_CCI);
356 int pcie_cap_slot_post_load(void *opaque, int version_id)
358 PCIDevice *dev = opaque;
359 hotplug_event_update_event_status(dev);
360 return 0;
363 void pcie_cap_slot_push_attention_button(PCIDevice *dev)
365 pcie_cap_slot_event(dev, PCI_EXP_HP_EV_ABP);
368 /* root control/capabilities/status. PME isn't emulated for now */
369 void pcie_cap_root_init(PCIDevice *dev)
371 pci_set_word(dev->wmask + dev->exp.exp_cap + PCI_EXP_RTCTL,
372 PCI_EXP_RTCTL_SECEE | PCI_EXP_RTCTL_SENFEE |
373 PCI_EXP_RTCTL_SEFEE);
376 void pcie_cap_root_reset(PCIDevice *dev)
378 pci_set_word(dev->config + dev->exp.exp_cap + PCI_EXP_RTCTL, 0);
382 * TODO: implement FLR:
383 * Right now sets the bit which indicates FLR is supported.
385 /* function level reset(FLR) */
386 void pcie_cap_flr_init(PCIDevice *dev)
388 pci_long_test_and_set_mask(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCAP,
389 PCI_EXP_DEVCAP_FLR);
391 /* Although reading BCR_FLR returns always 0,
392 * the bit is made writable here in order to detect the 1b is written
393 * pcie_cap_flr_write_config() test-and-clear the bit, so
394 * this bit always returns 0 to the guest.
396 pci_word_test_and_set_mask(dev->wmask + dev->exp.exp_cap + PCI_EXP_DEVCTL,
397 PCI_EXP_DEVCTL_BCR_FLR);
400 void pcie_cap_flr_write_config(PCIDevice *dev,
401 uint32_t addr, uint32_t val, int len)
403 uint8_t *devctl = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL;
404 if (pci_word_test_and_clear_mask(devctl, PCI_EXP_DEVCTL_BCR_FLR)) {
405 /* TODO: implement FLR */
409 /* Alternative Routing-ID Interpretation (ARI) */
410 /* ari forwarding support for down stream port */
411 void pcie_cap_ari_init(PCIDevice *dev)
413 uint32_t pos = dev->exp.exp_cap;
414 pci_long_test_and_set_mask(dev->config + pos + PCI_EXP_DEVCAP2,
415 PCI_EXP_DEVCAP2_ARI);
416 pci_long_test_and_set_mask(dev->wmask + pos + PCI_EXP_DEVCTL2,
417 PCI_EXP_DEVCTL2_ARI);
420 void pcie_cap_ari_reset(PCIDevice *dev)
422 uint8_t *devctl2 = dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2;
423 pci_long_test_and_clear_mask(devctl2, PCI_EXP_DEVCTL2_ARI);
426 bool pcie_cap_is_ari_enabled(const PCIDevice *dev)
428 if (!pci_is_express(dev)) {
429 return false;
431 if (!dev->exp.exp_cap) {
432 return false;
435 return pci_get_long(dev->config + dev->exp.exp_cap + PCI_EXP_DEVCTL2) &
436 PCI_EXP_DEVCTL2_ARI;
439 /**************************************************************************
440 * pci express extended capability allocation functions
441 * uint16_t ext_cap_id (16 bit)
442 * uint8_t cap_ver (4 bit)
443 * uint16_t cap_offset (12 bit)
444 * uint16_t ext_cap_size
447 static uint16_t pcie_find_capability_list(PCIDevice *dev, uint16_t cap_id,
448 uint16_t *prev_p)
450 uint16_t prev = 0;
451 uint16_t next;
452 uint32_t header = pci_get_long(dev->config + PCI_CONFIG_SPACE_SIZE);
454 if (!header) {
455 /* no extended capability */
456 next = 0;
457 goto out;
459 for (next = PCI_CONFIG_SPACE_SIZE; next;
460 prev = next, next = PCI_EXT_CAP_NEXT(header)) {
462 assert(next >= PCI_CONFIG_SPACE_SIZE);
463 assert(next <= PCIE_CONFIG_SPACE_SIZE - 8);
465 header = pci_get_long(dev->config + next);
466 if (PCI_EXT_CAP_ID(header) == cap_id) {
467 break;
471 out:
472 if (prev_p) {
473 *prev_p = prev;
475 return next;
478 uint16_t pcie_find_capability(PCIDevice *dev, uint16_t cap_id)
480 return pcie_find_capability_list(dev, cap_id, NULL);
483 static void pcie_ext_cap_set_next(PCIDevice *dev, uint16_t pos, uint16_t next)
485 uint16_t header = pci_get_long(dev->config + pos);
486 assert(!(next & (PCI_EXT_CAP_ALIGN - 1)));
487 header = (header & ~PCI_EXT_CAP_NEXT_MASK) |
488 ((next << PCI_EXT_CAP_NEXT_SHIFT) & PCI_EXT_CAP_NEXT_MASK);
489 pci_set_long(dev->config + pos, header);
493 * caller must supply valid (offset, size) * such that the range shouldn't
494 * overlap with other capability or other registers.
495 * This function doesn't check it.
497 void pcie_add_capability(PCIDevice *dev,
498 uint16_t cap_id, uint8_t cap_ver,
499 uint16_t offset, uint16_t size)
501 uint32_t header;
502 uint16_t next;
504 assert(offset >= PCI_CONFIG_SPACE_SIZE);
505 assert(offset < offset + size);
506 assert(offset + size < PCIE_CONFIG_SPACE_SIZE);
507 assert(size >= 8);
508 assert(pci_is_express(dev));
510 if (offset == PCI_CONFIG_SPACE_SIZE) {
511 header = pci_get_long(dev->config + offset);
512 next = PCI_EXT_CAP_NEXT(header);
513 } else {
514 uint16_t prev;
516 /* 0 is reserved cap id. use internally to find the last capability
517 in the linked list */
518 next = pcie_find_capability_list(dev, 0, &prev);
520 assert(prev >= PCI_CONFIG_SPACE_SIZE);
521 assert(next == 0);
522 pcie_ext_cap_set_next(dev, prev, offset);
524 pci_set_long(dev->config + offset, PCI_EXT_CAP(cap_id, cap_ver, next));
526 /* Make capability read-only by default */
527 memset(dev->wmask + offset, 0, size);
528 memset(dev->w1cmask + offset, 0, size);
529 /* Check capability by default */
530 memset(dev->cmask + offset, 0xFF, size);
533 /**************************************************************************
534 * pci express extended capability helper functions
537 /* ARI */
538 void pcie_ari_init(PCIDevice *dev, uint16_t offset, uint16_t nextfn)
540 pcie_add_capability(dev, PCI_EXT_CAP_ID_ARI, PCI_ARI_VER,
541 offset, PCI_ARI_SIZEOF);
542 pci_set_long(dev->config + offset + PCI_ARI_CAP, PCI_ARI_CAP_NFN(nextfn));