qapi/error: Check format string argument in error_*prepend()
[qemu/armbru.git] / hw / pci / msix.c
blob67e34f34d6584eff4eb55ffe053b5bb677032600
1 /*
2 * MSI-X device support
4 * This module includes support for MSI-X in pci devices.
6 * Author: Michael S. Tsirkin <mst@redhat.com>
8 * Copyright (c) 2009, Red Hat Inc, Michael S. Tsirkin (mst@redhat.com)
10 * This work is licensed under the terms of the GNU GPL, version 2. See
11 * the COPYING file in the top-level directory.
13 * Contributions after 2012-01-13 are licensed under the terms of the
14 * GNU GPL, version 2 or (at your option) any later version.
17 #include "qemu/osdep.h"
18 #include "hw/pci/msi.h"
19 #include "hw/pci/msix.h"
20 #include "hw/pci/pci.h"
21 #include "hw/xen/xen.h"
22 #include "sysemu/xen.h"
23 #include "migration/qemu-file-types.h"
24 #include "migration/vmstate.h"
25 #include "qemu/range.h"
26 #include "qapi/error.h"
27 #include "trace.h"
29 /* MSI enable bit and maskall bit are in byte 1 in FLAGS register */
30 #define MSIX_CONTROL_OFFSET (PCI_MSIX_FLAGS + 1)
31 #define MSIX_ENABLE_MASK (PCI_MSIX_FLAGS_ENABLE >> 8)
32 #define MSIX_MASKALL_MASK (PCI_MSIX_FLAGS_MASKALL >> 8)
34 MSIMessage msix_get_message(PCIDevice *dev, unsigned vector)
36 uint8_t *table_entry = dev->msix_table + vector * PCI_MSIX_ENTRY_SIZE;
37 MSIMessage msg;
39 msg.address = pci_get_quad(table_entry + PCI_MSIX_ENTRY_LOWER_ADDR);
40 msg.data = pci_get_long(table_entry + PCI_MSIX_ENTRY_DATA);
41 return msg;
45 * Special API for POWER to configure the vectors through
46 * a side channel. Should never be used by devices.
48 void msix_set_message(PCIDevice *dev, int vector, struct MSIMessage msg)
50 uint8_t *table_entry = dev->msix_table + vector * PCI_MSIX_ENTRY_SIZE;
52 pci_set_quad(table_entry + PCI_MSIX_ENTRY_LOWER_ADDR, msg.address);
53 pci_set_long(table_entry + PCI_MSIX_ENTRY_DATA, msg.data);
54 table_entry[PCI_MSIX_ENTRY_VECTOR_CTRL] &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
57 static uint8_t msix_pending_mask(int vector)
59 return 1 << (vector % 8);
62 static uint8_t *msix_pending_byte(PCIDevice *dev, int vector)
64 return dev->msix_pba + vector / 8;
67 static int msix_is_pending(PCIDevice *dev, int vector)
69 return *msix_pending_byte(dev, vector) & msix_pending_mask(vector);
72 void msix_set_pending(PCIDevice *dev, unsigned int vector)
74 *msix_pending_byte(dev, vector) |= msix_pending_mask(vector);
77 void msix_clr_pending(PCIDevice *dev, int vector)
79 *msix_pending_byte(dev, vector) &= ~msix_pending_mask(vector);
82 static bool msix_vector_masked(PCIDevice *dev, unsigned int vector, bool fmask)
84 unsigned offset = vector * PCI_MSIX_ENTRY_SIZE;
85 uint8_t *data = &dev->msix_table[offset + PCI_MSIX_ENTRY_DATA];
86 /* MSIs on Xen can be remapped into pirqs. In those cases, masking
87 * and unmasking go through the PV evtchn path. */
88 if (xen_enabled() && xen_is_pirq_msi(pci_get_long(data))) {
89 return false;
91 return fmask || dev->msix_table[offset + PCI_MSIX_ENTRY_VECTOR_CTRL] &
92 PCI_MSIX_ENTRY_CTRL_MASKBIT;
95 bool msix_is_masked(PCIDevice *dev, unsigned int vector)
97 return msix_vector_masked(dev, vector, dev->msix_function_masked);
100 static void msix_fire_vector_notifier(PCIDevice *dev,
101 unsigned int vector, bool is_masked)
103 MSIMessage msg;
104 int ret;
106 if (!dev->msix_vector_use_notifier) {
107 return;
109 if (is_masked) {
110 dev->msix_vector_release_notifier(dev, vector);
111 } else {
112 msg = msix_get_message(dev, vector);
113 ret = dev->msix_vector_use_notifier(dev, vector, msg);
114 assert(ret >= 0);
118 static void msix_handle_mask_update(PCIDevice *dev, int vector, bool was_masked)
120 bool is_masked = msix_is_masked(dev, vector);
122 if (is_masked == was_masked) {
123 return;
126 msix_fire_vector_notifier(dev, vector, is_masked);
128 if (!is_masked && msix_is_pending(dev, vector)) {
129 msix_clr_pending(dev, vector);
130 msix_notify(dev, vector);
134 static bool msix_masked(PCIDevice *dev)
136 return dev->config[dev->msix_cap + MSIX_CONTROL_OFFSET] & MSIX_MASKALL_MASK;
139 static void msix_update_function_masked(PCIDevice *dev)
141 dev->msix_function_masked = !msix_enabled(dev) || msix_masked(dev);
144 /* Handle MSI-X capability config write. */
145 void msix_write_config(PCIDevice *dev, uint32_t addr,
146 uint32_t val, int len)
148 unsigned enable_pos = dev->msix_cap + MSIX_CONTROL_OFFSET;
149 int vector;
150 bool was_masked;
152 if (!msix_present(dev) || !range_covers_byte(addr, len, enable_pos)) {
153 return;
156 trace_msix_write_config(dev->name, msix_enabled(dev), msix_masked(dev));
158 was_masked = dev->msix_function_masked;
159 msix_update_function_masked(dev);
161 if (!msix_enabled(dev)) {
162 return;
165 pci_device_deassert_intx(dev);
167 if (dev->msix_function_masked == was_masked) {
168 return;
171 for (vector = 0; vector < dev->msix_entries_nr; ++vector) {
172 msix_handle_mask_update(dev, vector,
173 msix_vector_masked(dev, vector, was_masked));
177 static uint64_t msix_table_mmio_read(void *opaque, hwaddr addr,
178 unsigned size)
180 PCIDevice *dev = opaque;
182 return pci_get_long(dev->msix_table + addr);
185 static void msix_table_mmio_write(void *opaque, hwaddr addr,
186 uint64_t val, unsigned size)
188 PCIDevice *dev = opaque;
189 int vector = addr / PCI_MSIX_ENTRY_SIZE;
190 bool was_masked;
192 was_masked = msix_is_masked(dev, vector);
193 pci_set_long(dev->msix_table + addr, val);
194 msix_handle_mask_update(dev, vector, was_masked);
197 static const MemoryRegionOps msix_table_mmio_ops = {
198 .read = msix_table_mmio_read,
199 .write = msix_table_mmio_write,
200 .endianness = DEVICE_LITTLE_ENDIAN,
201 .valid = {
202 .min_access_size = 4,
203 .max_access_size = 8,
205 .impl = {
206 .max_access_size = 4,
210 static uint64_t msix_pba_mmio_read(void *opaque, hwaddr addr,
211 unsigned size)
213 PCIDevice *dev = opaque;
214 if (dev->msix_vector_poll_notifier) {
215 unsigned vector_start = addr * 8;
216 unsigned vector_end = MIN(addr + size * 8, dev->msix_entries_nr);
217 dev->msix_vector_poll_notifier(dev, vector_start, vector_end);
220 return pci_get_long(dev->msix_pba + addr);
223 static void msix_pba_mmio_write(void *opaque, hwaddr addr,
224 uint64_t val, unsigned size)
228 static const MemoryRegionOps msix_pba_mmio_ops = {
229 .read = msix_pba_mmio_read,
230 .write = msix_pba_mmio_write,
231 .endianness = DEVICE_LITTLE_ENDIAN,
232 .valid = {
233 .min_access_size = 4,
234 .max_access_size = 8,
236 .impl = {
237 .max_access_size = 4,
241 static void msix_mask_all(struct PCIDevice *dev, unsigned nentries)
243 int vector;
245 for (vector = 0; vector < nentries; ++vector) {
246 unsigned offset =
247 vector * PCI_MSIX_ENTRY_SIZE + PCI_MSIX_ENTRY_VECTOR_CTRL;
248 bool was_masked = msix_is_masked(dev, vector);
250 dev->msix_table[offset] |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
251 msix_handle_mask_update(dev, vector, was_masked);
256 * Make PCI device @dev MSI-X capable
257 * @nentries is the max number of MSI-X vectors that the device support.
258 * @table_bar is the MemoryRegion that MSI-X table structure resides.
259 * @table_bar_nr is number of base address register corresponding to @table_bar.
260 * @table_offset indicates the offset that the MSI-X table structure starts with
261 * in @table_bar.
262 * @pba_bar is the MemoryRegion that the Pending Bit Array structure resides.
263 * @pba_bar_nr is number of base address register corresponding to @pba_bar.
264 * @pba_offset indicates the offset that the Pending Bit Array structure
265 * starts with in @pba_bar.
266 * Non-zero @cap_pos puts capability MSI-X at that offset in PCI config space.
267 * @errp is for returning errors.
269 * Return 0 on success; set @errp and return -errno on error:
270 * -ENOTSUP means lacking msi support for a msi-capable platform.
271 * -EINVAL means capability overlap, happens when @cap_pos is non-zero,
272 * also means a programming error, except device assignment, which can check
273 * if a real HW is broken.
275 int msix_init(struct PCIDevice *dev, unsigned short nentries,
276 MemoryRegion *table_bar, uint8_t table_bar_nr,
277 unsigned table_offset, MemoryRegion *pba_bar,
278 uint8_t pba_bar_nr, unsigned pba_offset, uint8_t cap_pos,
279 Error **errp)
281 int cap;
282 unsigned table_size, pba_size;
283 uint8_t *config;
285 /* Nothing to do if MSI is not supported by interrupt controller */
286 if (!msi_nonbroken) {
287 error_setg(errp, "MSI-X is not supported by interrupt controller");
288 return -ENOTSUP;
291 if (nentries < 1 || nentries > PCI_MSIX_FLAGS_QSIZE + 1) {
292 error_setg(errp, "The number of MSI-X vectors is invalid");
293 return -EINVAL;
296 table_size = nentries * PCI_MSIX_ENTRY_SIZE;
297 pba_size = QEMU_ALIGN_UP(nentries, 64) / 8;
299 /* Sanity test: table & pba don't overlap, fit within BARs, min aligned */
300 if ((table_bar_nr == pba_bar_nr &&
301 ranges_overlap(table_offset, table_size, pba_offset, pba_size)) ||
302 table_offset + table_size > memory_region_size(table_bar) ||
303 pba_offset + pba_size > memory_region_size(pba_bar) ||
304 (table_offset | pba_offset) & PCI_MSIX_FLAGS_BIRMASK) {
305 error_setg(errp, "table & pba overlap, or they don't fit in BARs,"
306 " or don't align");
307 return -EINVAL;
310 cap = pci_add_capability(dev, PCI_CAP_ID_MSIX,
311 cap_pos, MSIX_CAP_LENGTH, errp);
312 if (cap < 0) {
313 return cap;
316 dev->msix_cap = cap;
317 dev->cap_present |= QEMU_PCI_CAP_MSIX;
318 config = dev->config + cap;
320 pci_set_word(config + PCI_MSIX_FLAGS, nentries - 1);
321 dev->msix_entries_nr = nentries;
322 dev->msix_function_masked = true;
324 pci_set_long(config + PCI_MSIX_TABLE, table_offset | table_bar_nr);
325 pci_set_long(config + PCI_MSIX_PBA, pba_offset | pba_bar_nr);
327 /* Make flags bit writable. */
328 dev->wmask[cap + MSIX_CONTROL_OFFSET] |= MSIX_ENABLE_MASK |
329 MSIX_MASKALL_MASK;
331 dev->msix_table = g_malloc0(table_size);
332 dev->msix_pba = g_malloc0(pba_size);
333 dev->msix_entry_used = g_malloc0(nentries * sizeof *dev->msix_entry_used);
335 msix_mask_all(dev, nentries);
337 memory_region_init_io(&dev->msix_table_mmio, OBJECT(dev), &msix_table_mmio_ops, dev,
338 "msix-table", table_size);
339 memory_region_add_subregion(table_bar, table_offset, &dev->msix_table_mmio);
340 memory_region_init_io(&dev->msix_pba_mmio, OBJECT(dev), &msix_pba_mmio_ops, dev,
341 "msix-pba", pba_size);
342 memory_region_add_subregion(pba_bar, pba_offset, &dev->msix_pba_mmio);
344 return 0;
347 int msix_init_exclusive_bar(PCIDevice *dev, unsigned short nentries,
348 uint8_t bar_nr, Error **errp)
350 int ret;
351 char *name;
352 uint32_t bar_size = 4096;
353 uint32_t bar_pba_offset = bar_size / 2;
354 uint32_t bar_pba_size = QEMU_ALIGN_UP(nentries, 64) / 8;
357 * Migration compatibility dictates that this remains a 4k
358 * BAR with the vector table in the lower half and PBA in
359 * the upper half for nentries which is lower or equal to 128.
360 * No need to care about using more than 65 entries for legacy
361 * machine types who has at most 64 queues.
363 if (nentries * PCI_MSIX_ENTRY_SIZE > bar_pba_offset) {
364 bar_pba_offset = nentries * PCI_MSIX_ENTRY_SIZE;
367 if (bar_pba_offset + bar_pba_size > 4096) {
368 bar_size = bar_pba_offset + bar_pba_size;
371 bar_size = pow2ceil(bar_size);
373 name = g_strdup_printf("%s-msix", dev->name);
374 memory_region_init(&dev->msix_exclusive_bar, OBJECT(dev), name, bar_size);
375 g_free(name);
377 ret = msix_init(dev, nentries, &dev->msix_exclusive_bar, bar_nr,
378 0, &dev->msix_exclusive_bar,
379 bar_nr, bar_pba_offset,
380 0, errp);
381 if (ret) {
382 return ret;
385 pci_register_bar(dev, bar_nr, PCI_BASE_ADDRESS_SPACE_MEMORY,
386 &dev->msix_exclusive_bar);
388 return 0;
391 static void msix_free_irq_entries(PCIDevice *dev)
393 int vector;
395 for (vector = 0; vector < dev->msix_entries_nr; ++vector) {
396 dev->msix_entry_used[vector] = 0;
397 msix_clr_pending(dev, vector);
401 static void msix_clear_all_vectors(PCIDevice *dev)
403 int vector;
405 for (vector = 0; vector < dev->msix_entries_nr; ++vector) {
406 msix_clr_pending(dev, vector);
410 /* Clean up resources for the device. */
411 void msix_uninit(PCIDevice *dev, MemoryRegion *table_bar, MemoryRegion *pba_bar)
413 if (!msix_present(dev)) {
414 return;
416 pci_del_capability(dev, PCI_CAP_ID_MSIX, MSIX_CAP_LENGTH);
417 dev->msix_cap = 0;
418 msix_free_irq_entries(dev);
419 dev->msix_entries_nr = 0;
420 memory_region_del_subregion(pba_bar, &dev->msix_pba_mmio);
421 g_free(dev->msix_pba);
422 dev->msix_pba = NULL;
423 memory_region_del_subregion(table_bar, &dev->msix_table_mmio);
424 g_free(dev->msix_table);
425 dev->msix_table = NULL;
426 g_free(dev->msix_entry_used);
427 dev->msix_entry_used = NULL;
428 dev->cap_present &= ~QEMU_PCI_CAP_MSIX;
431 void msix_uninit_exclusive_bar(PCIDevice *dev)
433 if (msix_present(dev)) {
434 msix_uninit(dev, &dev->msix_exclusive_bar, &dev->msix_exclusive_bar);
438 void msix_save(PCIDevice *dev, QEMUFile *f)
440 unsigned n = dev->msix_entries_nr;
442 if (!msix_present(dev)) {
443 return;
446 qemu_put_buffer(f, dev->msix_table, n * PCI_MSIX_ENTRY_SIZE);
447 qemu_put_buffer(f, dev->msix_pba, DIV_ROUND_UP(n, 8));
450 /* Should be called after restoring the config space. */
451 void msix_load(PCIDevice *dev, QEMUFile *f)
453 unsigned n = dev->msix_entries_nr;
454 unsigned int vector;
456 if (!msix_present(dev)) {
457 return;
460 msix_clear_all_vectors(dev);
461 qemu_get_buffer(f, dev->msix_table, n * PCI_MSIX_ENTRY_SIZE);
462 qemu_get_buffer(f, dev->msix_pba, DIV_ROUND_UP(n, 8));
463 msix_update_function_masked(dev);
465 for (vector = 0; vector < n; vector++) {
466 msix_handle_mask_update(dev, vector, true);
470 /* Does device support MSI-X? */
471 int msix_present(PCIDevice *dev)
473 return dev->cap_present & QEMU_PCI_CAP_MSIX;
476 /* Is MSI-X enabled? */
477 int msix_enabled(PCIDevice *dev)
479 return (dev->cap_present & QEMU_PCI_CAP_MSIX) &&
480 (dev->config[dev->msix_cap + MSIX_CONTROL_OFFSET] &
481 MSIX_ENABLE_MASK);
484 /* Send an MSI-X message */
485 void msix_notify(PCIDevice *dev, unsigned vector)
487 MSIMessage msg;
489 if (vector >= dev->msix_entries_nr || !dev->msix_entry_used[vector]) {
490 return;
493 if (msix_is_masked(dev, vector)) {
494 msix_set_pending(dev, vector);
495 return;
498 msg = msix_get_message(dev, vector);
500 msi_send_message(dev, msg);
503 void msix_reset(PCIDevice *dev)
505 if (!msix_present(dev)) {
506 return;
508 msix_clear_all_vectors(dev);
509 dev->config[dev->msix_cap + MSIX_CONTROL_OFFSET] &=
510 ~dev->wmask[dev->msix_cap + MSIX_CONTROL_OFFSET];
511 memset(dev->msix_table, 0, dev->msix_entries_nr * PCI_MSIX_ENTRY_SIZE);
512 memset(dev->msix_pba, 0, QEMU_ALIGN_UP(dev->msix_entries_nr, 64) / 8);
513 msix_mask_all(dev, dev->msix_entries_nr);
516 /* PCI spec suggests that devices make it possible for software to configure
517 * less vectors than supported by the device, but does not specify a standard
518 * mechanism for devices to do so.
520 * We support this by asking devices to declare vectors software is going to
521 * actually use, and checking this on the notification path. Devices that
522 * don't want to follow the spec suggestion can declare all vectors as used. */
524 /* Mark vector as used. */
525 int msix_vector_use(PCIDevice *dev, unsigned vector)
527 if (vector >= dev->msix_entries_nr) {
528 return -EINVAL;
531 dev->msix_entry_used[vector]++;
532 return 0;
535 /* Mark vector as unused. */
536 void msix_vector_unuse(PCIDevice *dev, unsigned vector)
538 if (vector >= dev->msix_entries_nr || !dev->msix_entry_used[vector]) {
539 return;
541 if (--dev->msix_entry_used[vector]) {
542 return;
544 msix_clr_pending(dev, vector);
547 void msix_unuse_all_vectors(PCIDevice *dev)
549 if (!msix_present(dev)) {
550 return;
552 msix_free_irq_entries(dev);
555 unsigned int msix_nr_vectors_allocated(const PCIDevice *dev)
557 return dev->msix_entries_nr;
560 static int msix_set_notifier_for_vector(PCIDevice *dev, unsigned int vector)
562 MSIMessage msg;
564 if (msix_is_masked(dev, vector)) {
565 return 0;
567 msg = msix_get_message(dev, vector);
568 return dev->msix_vector_use_notifier(dev, vector, msg);
571 static void msix_unset_notifier_for_vector(PCIDevice *dev, unsigned int vector)
573 if (msix_is_masked(dev, vector)) {
574 return;
576 dev->msix_vector_release_notifier(dev, vector);
579 int msix_set_vector_notifiers(PCIDevice *dev,
580 MSIVectorUseNotifier use_notifier,
581 MSIVectorReleaseNotifier release_notifier,
582 MSIVectorPollNotifier poll_notifier)
584 int vector, ret;
586 assert(use_notifier && release_notifier);
588 dev->msix_vector_use_notifier = use_notifier;
589 dev->msix_vector_release_notifier = release_notifier;
590 dev->msix_vector_poll_notifier = poll_notifier;
592 if ((dev->config[dev->msix_cap + MSIX_CONTROL_OFFSET] &
593 (MSIX_ENABLE_MASK | MSIX_MASKALL_MASK)) == MSIX_ENABLE_MASK) {
594 for (vector = 0; vector < dev->msix_entries_nr; vector++) {
595 ret = msix_set_notifier_for_vector(dev, vector);
596 if (ret < 0) {
597 goto undo;
601 if (dev->msix_vector_poll_notifier) {
602 dev->msix_vector_poll_notifier(dev, 0, dev->msix_entries_nr);
604 return 0;
606 undo:
607 while (--vector >= 0) {
608 msix_unset_notifier_for_vector(dev, vector);
610 dev->msix_vector_use_notifier = NULL;
611 dev->msix_vector_release_notifier = NULL;
612 return ret;
615 void msix_unset_vector_notifiers(PCIDevice *dev)
617 int vector;
619 assert(dev->msix_vector_use_notifier &&
620 dev->msix_vector_release_notifier);
622 if ((dev->config[dev->msix_cap + MSIX_CONTROL_OFFSET] &
623 (MSIX_ENABLE_MASK | MSIX_MASKALL_MASK)) == MSIX_ENABLE_MASK) {
624 for (vector = 0; vector < dev->msix_entries_nr; vector++) {
625 msix_unset_notifier_for_vector(dev, vector);
628 dev->msix_vector_use_notifier = NULL;
629 dev->msix_vector_release_notifier = NULL;
630 dev->msix_vector_poll_notifier = NULL;
633 static int put_msix_state(QEMUFile *f, void *pv, size_t size,
634 const VMStateField *field, QJSON *vmdesc)
636 msix_save(pv, f);
638 return 0;
641 static int get_msix_state(QEMUFile *f, void *pv, size_t size,
642 const VMStateField *field)
644 msix_load(pv, f);
645 return 0;
648 static VMStateInfo vmstate_info_msix = {
649 .name = "msix state",
650 .get = get_msix_state,
651 .put = put_msix_state,
654 const VMStateDescription vmstate_msix = {
655 .name = "msix",
656 .fields = (VMStateField[]) {
658 .name = "msix",
659 .version_id = 0,
660 .field_exists = NULL,
661 .size = 0, /* ouch */
662 .info = &vmstate_info_msix,
663 .flags = VMS_SINGLE,
664 .offset = 0,
666 VMSTATE_END_OF_LIST()