ivshmem: Implement shm=... with a memory backend
[qemu/kevin.git] / hw / xen / xen_pt.c
blob657bf6cdc1304ee3fb09d46633b35cb485c50936
1 /*
2 * Copyright (c) 2007, Neocleus Corporation.
3 * Copyright (c) 2007, Intel Corporation.
5 * This work is licensed under the terms of the GNU GPL, version 2. See
6 * the COPYING file in the top-level directory.
8 * Alex Novik <alex@neocleus.com>
9 * Allen Kay <allen.m.kay@intel.com>
10 * Guy Zana <guy@neocleus.com>
12 * This file implements direct PCI assignment to a HVM guest
16 * Interrupt Disable policy:
18 * INTx interrupt:
19 * Initialize(register_real_device)
20 * Map INTx(xc_physdev_map_pirq):
21 * <fail>
22 * - Set real Interrupt Disable bit to '1'.
23 * - Set machine_irq and assigned_device->machine_irq to '0'.
24 * * Don't bind INTx.
26 * Bind INTx(xc_domain_bind_pt_pci_irq):
27 * <fail>
28 * - Set real Interrupt Disable bit to '1'.
29 * - Unmap INTx.
30 * - Decrement xen_pt_mapped_machine_irq[machine_irq]
31 * - Set assigned_device->machine_irq to '0'.
33 * Write to Interrupt Disable bit by guest software(xen_pt_cmd_reg_write)
34 * Write '0'
35 * - Set real bit to '0' if assigned_device->machine_irq isn't '0'.
37 * Write '1'
38 * - Set real bit to '1'.
40 * MSI interrupt:
41 * Initialize MSI register(xen_pt_msi_setup, xen_pt_msi_update)
42 * Bind MSI(xc_domain_update_msi_irq)
43 * <fail>
44 * - Unmap MSI.
45 * - Set dev->msi->pirq to '-1'.
47 * MSI-X interrupt:
48 * Initialize MSI-X register(xen_pt_msix_update_one)
49 * Bind MSI-X(xc_domain_update_msi_irq)
50 * <fail>
51 * - Unmap MSI-X.
52 * - Set entry->pirq to '-1'.
55 #include "qemu/osdep.h"
56 #include <sys/ioctl.h>
58 #include "hw/pci/pci.h"
59 #include "hw/xen/xen.h"
60 #include "hw/i386/pc.h"
61 #include "hw/xen/xen_backend.h"
62 #include "xen_pt.h"
63 #include "qemu/range.h"
64 #include "exec/address-spaces.h"
66 #define XEN_PT_NR_IRQS (256)
67 static uint8_t xen_pt_mapped_machine_irq[XEN_PT_NR_IRQS] = {0};
69 void xen_pt_log(const PCIDevice *d, const char *f, ...)
71 va_list ap;
73 va_start(ap, f);
74 if (d) {
75 fprintf(stderr, "[%02x:%02x.%d] ", pci_bus_num(d->bus),
76 PCI_SLOT(d->devfn), PCI_FUNC(d->devfn));
78 vfprintf(stderr, f, ap);
79 va_end(ap);
82 /* Config Space */
84 static int xen_pt_pci_config_access_check(PCIDevice *d, uint32_t addr, int len)
86 /* check offset range */
87 if (addr >= 0xFF) {
88 XEN_PT_ERR(d, "Failed to access register with offset exceeding 0xFF. "
89 "(addr: 0x%02x, len: %d)\n", addr, len);
90 return -1;
93 /* check read size */
94 if ((len != 1) && (len != 2) && (len != 4)) {
95 XEN_PT_ERR(d, "Failed to access register with invalid access length. "
96 "(addr: 0x%02x, len: %d)\n", addr, len);
97 return -1;
100 /* check offset alignment */
101 if (addr & (len - 1)) {
102 XEN_PT_ERR(d, "Failed to access register with invalid access size "
103 "alignment. (addr: 0x%02x, len: %d)\n", addr, len);
104 return -1;
107 return 0;
110 int xen_pt_bar_offset_to_index(uint32_t offset)
112 int index = 0;
114 /* check Exp ROM BAR */
115 if (offset == PCI_ROM_ADDRESS) {
116 return PCI_ROM_SLOT;
119 /* calculate BAR index */
120 index = (offset - PCI_BASE_ADDRESS_0) >> 2;
121 if (index >= PCI_NUM_REGIONS) {
122 return -1;
125 return index;
128 static uint32_t xen_pt_pci_read_config(PCIDevice *d, uint32_t addr, int len)
130 XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
131 uint32_t val = 0;
132 XenPTRegGroup *reg_grp_entry = NULL;
133 XenPTReg *reg_entry = NULL;
134 int rc = 0;
135 int emul_len = 0;
136 uint32_t find_addr = addr;
138 if (xen_pt_pci_config_access_check(d, addr, len)) {
139 goto exit;
142 /* find register group entry */
143 reg_grp_entry = xen_pt_find_reg_grp(s, addr);
144 if (reg_grp_entry) {
145 /* check 0-Hardwired register group */
146 if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
147 /* no need to emulate, just return 0 */
148 val = 0;
149 goto exit;
153 /* read I/O device register value */
154 rc = xen_host_pci_get_block(&s->real_device, addr, (uint8_t *)&val, len);
155 if (rc < 0) {
156 XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
157 memset(&val, 0xff, len);
160 /* just return the I/O device register value for
161 * passthrough type register group */
162 if (reg_grp_entry == NULL) {
163 goto exit;
166 /* adjust the read value to appropriate CFC-CFF window */
167 val <<= (addr & 3) << 3;
168 emul_len = len;
170 /* loop around the guest requested size */
171 while (emul_len > 0) {
172 /* find register entry to be emulated */
173 reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
174 if (reg_entry) {
175 XenPTRegInfo *reg = reg_entry->reg;
176 uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
177 uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
178 uint8_t *ptr_val = NULL;
180 valid_mask <<= (find_addr - real_offset) << 3;
181 ptr_val = (uint8_t *)&val + (real_offset & 3);
183 /* do emulation based on register size */
184 switch (reg->size) {
185 case 1:
186 if (reg->u.b.read) {
187 rc = reg->u.b.read(s, reg_entry, ptr_val, valid_mask);
189 break;
190 case 2:
191 if (reg->u.w.read) {
192 rc = reg->u.w.read(s, reg_entry,
193 (uint16_t *)ptr_val, valid_mask);
195 break;
196 case 4:
197 if (reg->u.dw.read) {
198 rc = reg->u.dw.read(s, reg_entry,
199 (uint32_t *)ptr_val, valid_mask);
201 break;
204 if (rc < 0) {
205 xen_shutdown_fatal_error("Internal error: Invalid read "
206 "emulation. (%s, rc: %d)\n",
207 __func__, rc);
208 return 0;
211 /* calculate next address to find */
212 emul_len -= reg->size;
213 if (emul_len > 0) {
214 find_addr = real_offset + reg->size;
216 } else {
217 /* nothing to do with passthrough type register,
218 * continue to find next byte */
219 emul_len--;
220 find_addr++;
224 /* need to shift back before returning them to pci bus emulator */
225 val >>= ((addr & 3) << 3);
227 exit:
228 XEN_PT_LOG_CONFIG(d, addr, val, len);
229 return val;
232 static void xen_pt_pci_write_config(PCIDevice *d, uint32_t addr,
233 uint32_t val, int len)
235 XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
236 int index = 0;
237 XenPTRegGroup *reg_grp_entry = NULL;
238 int rc = 0;
239 uint32_t read_val = 0, wb_mask;
240 int emul_len = 0;
241 XenPTReg *reg_entry = NULL;
242 uint32_t find_addr = addr;
243 XenPTRegInfo *reg = NULL;
244 bool wp_flag = false;
246 if (xen_pt_pci_config_access_check(d, addr, len)) {
247 return;
250 XEN_PT_LOG_CONFIG(d, addr, val, len);
252 /* check unused BAR register */
253 index = xen_pt_bar_offset_to_index(addr);
254 if ((index >= 0) && (val != 0)) {
255 uint32_t chk = val;
257 if (index == PCI_ROM_SLOT)
258 chk |= (uint32_t)~PCI_ROM_ADDRESS_MASK;
260 if ((chk != XEN_PT_BAR_ALLF) &&
261 (s->bases[index].bar_flag == XEN_PT_BAR_FLAG_UNUSED)) {
262 XEN_PT_WARN(d, "Guest attempt to set address to unused "
263 "Base Address Register. (addr: 0x%02x, len: %d)\n",
264 addr, len);
268 /* find register group entry */
269 reg_grp_entry = xen_pt_find_reg_grp(s, addr);
270 if (reg_grp_entry) {
271 /* check 0-Hardwired register group */
272 if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
273 /* ignore silently */
274 XEN_PT_WARN(d, "Access to 0-Hardwired register. "
275 "(addr: 0x%02x, len: %d)\n", addr, len);
276 return;
280 rc = xen_host_pci_get_block(&s->real_device, addr,
281 (uint8_t *)&read_val, len);
282 if (rc < 0) {
283 XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
284 memset(&read_val, 0xff, len);
285 wb_mask = 0;
286 } else {
287 wb_mask = 0xFFFFFFFF >> ((4 - len) << 3);
290 /* pass directly to the real device for passthrough type register group */
291 if (reg_grp_entry == NULL) {
292 if (!s->permissive) {
293 wb_mask = 0;
294 wp_flag = true;
296 goto out;
299 memory_region_transaction_begin();
300 pci_default_write_config(d, addr, val, len);
302 /* adjust the read and write value to appropriate CFC-CFF window */
303 read_val <<= (addr & 3) << 3;
304 val <<= (addr & 3) << 3;
305 emul_len = len;
307 /* loop around the guest requested size */
308 while (emul_len > 0) {
309 /* find register entry to be emulated */
310 reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
311 if (reg_entry) {
312 reg = reg_entry->reg;
313 uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
314 uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
315 uint8_t *ptr_val = NULL;
316 uint32_t wp_mask = reg->emu_mask | reg->ro_mask;
318 valid_mask <<= (find_addr - real_offset) << 3;
319 ptr_val = (uint8_t *)&val + (real_offset & 3);
320 if (!s->permissive) {
321 wp_mask |= reg->res_mask;
323 if (wp_mask == (0xFFFFFFFF >> ((4 - reg->size) << 3))) {
324 wb_mask &= ~((wp_mask >> ((find_addr - real_offset) << 3))
325 << ((len - emul_len) << 3));
328 /* do emulation based on register size */
329 switch (reg->size) {
330 case 1:
331 if (reg->u.b.write) {
332 rc = reg->u.b.write(s, reg_entry, ptr_val,
333 read_val >> ((real_offset & 3) << 3),
334 valid_mask);
336 break;
337 case 2:
338 if (reg->u.w.write) {
339 rc = reg->u.w.write(s, reg_entry, (uint16_t *)ptr_val,
340 (read_val >> ((real_offset & 3) << 3)),
341 valid_mask);
343 break;
344 case 4:
345 if (reg->u.dw.write) {
346 rc = reg->u.dw.write(s, reg_entry, (uint32_t *)ptr_val,
347 (read_val >> ((real_offset & 3) << 3)),
348 valid_mask);
350 break;
353 if (rc < 0) {
354 xen_shutdown_fatal_error("Internal error: Invalid write"
355 " emulation. (%s, rc: %d)\n",
356 __func__, rc);
357 return;
360 /* calculate next address to find */
361 emul_len -= reg->size;
362 if (emul_len > 0) {
363 find_addr = real_offset + reg->size;
365 } else {
366 /* nothing to do with passthrough type register,
367 * continue to find next byte */
368 if (!s->permissive) {
369 wb_mask &= ~(0xff << ((len - emul_len) << 3));
370 /* Unused BARs will make it here, but we don't want to issue
371 * warnings for writes to them (bogus writes get dealt with
372 * above).
374 if (index < 0) {
375 wp_flag = true;
378 emul_len--;
379 find_addr++;
383 /* need to shift back before passing them to xen_host_pci_set_block. */
384 val >>= (addr & 3) << 3;
386 memory_region_transaction_commit();
388 out:
389 if (wp_flag && !s->permissive_warned) {
390 s->permissive_warned = true;
391 xen_pt_log(d, "Write-back to unknown field 0x%02x (partially) inhibited (0x%0*x)\n",
392 addr, len * 2, wb_mask);
393 xen_pt_log(d, "If the device doesn't work, try enabling permissive mode\n");
394 xen_pt_log(d, "(unsafe) and if it helps report the problem to xen-devel\n");
396 for (index = 0; wb_mask; index += len) {
397 /* unknown regs are passed through */
398 while (!(wb_mask & 0xff)) {
399 index++;
400 wb_mask >>= 8;
402 len = 0;
403 do {
404 len++;
405 wb_mask >>= 8;
406 } while (wb_mask & 0xff);
407 rc = xen_host_pci_set_block(&s->real_device, addr + index,
408 (uint8_t *)&val + index, len);
410 if (rc < 0) {
411 XEN_PT_ERR(d, "xen_host_pci_set_block failed. return value: %d.\n", rc);
416 /* register regions */
418 static uint64_t xen_pt_bar_read(void *o, hwaddr addr,
419 unsigned size)
421 PCIDevice *d = o;
422 /* if this function is called, that probably means that there is a
423 * misconfiguration of the IOMMU. */
424 XEN_PT_ERR(d, "Should not read BAR through QEMU. @0x"TARGET_FMT_plx"\n",
425 addr);
426 return 0;
428 static void xen_pt_bar_write(void *o, hwaddr addr, uint64_t val,
429 unsigned size)
431 PCIDevice *d = o;
432 /* Same comment as xen_pt_bar_read function */
433 XEN_PT_ERR(d, "Should not write BAR through QEMU. @0x"TARGET_FMT_plx"\n",
434 addr);
437 static const MemoryRegionOps ops = {
438 .endianness = DEVICE_NATIVE_ENDIAN,
439 .read = xen_pt_bar_read,
440 .write = xen_pt_bar_write,
443 static int xen_pt_register_regions(XenPCIPassthroughState *s, uint16_t *cmd)
445 int i = 0;
446 XenHostPCIDevice *d = &s->real_device;
448 /* Register PIO/MMIO BARs */
449 for (i = 0; i < PCI_ROM_SLOT; i++) {
450 XenHostPCIIORegion *r = &d->io_regions[i];
451 uint8_t type;
453 if (r->base_addr == 0 || r->size == 0) {
454 continue;
457 s->bases[i].access.u = r->base_addr;
459 if (r->type & XEN_HOST_PCI_REGION_TYPE_IO) {
460 type = PCI_BASE_ADDRESS_SPACE_IO;
461 *cmd |= PCI_COMMAND_IO;
462 } else {
463 type = PCI_BASE_ADDRESS_SPACE_MEMORY;
464 if (r->type & XEN_HOST_PCI_REGION_TYPE_PREFETCH) {
465 type |= PCI_BASE_ADDRESS_MEM_PREFETCH;
467 if (r->type & XEN_HOST_PCI_REGION_TYPE_MEM_64) {
468 type |= PCI_BASE_ADDRESS_MEM_TYPE_64;
470 *cmd |= PCI_COMMAND_MEMORY;
473 memory_region_init_io(&s->bar[i], OBJECT(s), &ops, &s->dev,
474 "xen-pci-pt-bar", r->size);
475 pci_register_bar(&s->dev, i, type, &s->bar[i]);
477 XEN_PT_LOG(&s->dev, "IO region %i registered (size=0x%08"PRIx64
478 " base_addr=0x%08"PRIx64" type: %#x)\n",
479 i, r->size, r->base_addr, type);
482 /* Register expansion ROM address */
483 if (d->rom.base_addr && d->rom.size) {
484 uint32_t bar_data = 0;
486 /* Re-set BAR reported by OS, otherwise ROM can't be read. */
487 if (xen_host_pci_get_long(d, PCI_ROM_ADDRESS, &bar_data)) {
488 return 0;
490 if ((bar_data & PCI_ROM_ADDRESS_MASK) == 0) {
491 bar_data |= d->rom.base_addr & PCI_ROM_ADDRESS_MASK;
492 xen_host_pci_set_long(d, PCI_ROM_ADDRESS, bar_data);
495 s->bases[PCI_ROM_SLOT].access.maddr = d->rom.base_addr;
497 memory_region_init_io(&s->rom, OBJECT(s), &ops, &s->dev,
498 "xen-pci-pt-rom", d->rom.size);
499 pci_register_bar(&s->dev, PCI_ROM_SLOT, PCI_BASE_ADDRESS_MEM_PREFETCH,
500 &s->rom);
502 XEN_PT_LOG(&s->dev, "Expansion ROM registered (size=0x%08"PRIx64
503 " base_addr=0x%08"PRIx64")\n",
504 d->rom.size, d->rom.base_addr);
507 xen_pt_register_vga_regions(d);
508 return 0;
511 /* region mapping */
513 static int xen_pt_bar_from_region(XenPCIPassthroughState *s, MemoryRegion *mr)
515 int i = 0;
517 for (i = 0; i < PCI_NUM_REGIONS - 1; i++) {
518 if (mr == &s->bar[i]) {
519 return i;
522 if (mr == &s->rom) {
523 return PCI_ROM_SLOT;
525 return -1;
529 * This function checks if an io_region overlaps an io_region from another
530 * device. The io_region to check is provided with (addr, size and type)
531 * A callback can be provided and will be called for every region that is
532 * overlapped.
533 * The return value indicates if the region is overlappsed */
534 struct CheckBarArgs {
535 XenPCIPassthroughState *s;
536 pcibus_t addr;
537 pcibus_t size;
538 uint8_t type;
539 bool rc;
541 static void xen_pt_check_bar_overlap(PCIBus *bus, PCIDevice *d, void *opaque)
543 struct CheckBarArgs *arg = opaque;
544 XenPCIPassthroughState *s = arg->s;
545 uint8_t type = arg->type;
546 int i;
548 if (d->devfn == s->dev.devfn) {
549 return;
552 /* xxx: This ignores bridges. */
553 for (i = 0; i < PCI_NUM_REGIONS; i++) {
554 const PCIIORegion *r = &d->io_regions[i];
556 if (!r->size) {
557 continue;
559 if ((type & PCI_BASE_ADDRESS_SPACE_IO)
560 != (r->type & PCI_BASE_ADDRESS_SPACE_IO)) {
561 continue;
564 if (ranges_overlap(arg->addr, arg->size, r->addr, r->size)) {
565 XEN_PT_WARN(&s->dev,
566 "Overlapped to device [%02x:%02x.%d] Region: %i"
567 " (addr: %#"FMT_PCIBUS", len: %#"FMT_PCIBUS")\n",
568 pci_bus_num(bus), PCI_SLOT(d->devfn),
569 PCI_FUNC(d->devfn), i, r->addr, r->size);
570 arg->rc = true;
575 static void xen_pt_region_update(XenPCIPassthroughState *s,
576 MemoryRegionSection *sec, bool adding)
578 PCIDevice *d = &s->dev;
579 MemoryRegion *mr = sec->mr;
580 int bar = -1;
581 int rc;
582 int op = adding ? DPCI_ADD_MAPPING : DPCI_REMOVE_MAPPING;
583 struct CheckBarArgs args = {
584 .s = s,
585 .addr = sec->offset_within_address_space,
586 .size = int128_get64(sec->size),
587 .rc = false,
590 bar = xen_pt_bar_from_region(s, mr);
591 if (bar == -1 && (!s->msix || &s->msix->mmio != mr)) {
592 return;
595 if (s->msix && &s->msix->mmio == mr) {
596 if (adding) {
597 s->msix->mmio_base_addr = sec->offset_within_address_space;
598 rc = xen_pt_msix_update_remap(s, s->msix->bar_index);
600 return;
603 args.type = d->io_regions[bar].type;
604 pci_for_each_device(d->bus, pci_bus_num(d->bus),
605 xen_pt_check_bar_overlap, &args);
606 if (args.rc) {
607 XEN_PT_WARN(d, "Region: %d (addr: %#"FMT_PCIBUS
608 ", len: %#"FMT_PCIBUS") is overlapped.\n",
609 bar, sec->offset_within_address_space,
610 int128_get64(sec->size));
613 if (d->io_regions[bar].type & PCI_BASE_ADDRESS_SPACE_IO) {
614 uint32_t guest_port = sec->offset_within_address_space;
615 uint32_t machine_port = s->bases[bar].access.pio_base;
616 uint32_t size = int128_get64(sec->size);
617 rc = xc_domain_ioport_mapping(xen_xc, xen_domid,
618 guest_port, machine_port, size,
619 op);
620 if (rc) {
621 XEN_PT_ERR(d, "%s ioport mapping failed! (err: %i)\n",
622 adding ? "create new" : "remove old", errno);
624 } else {
625 pcibus_t guest_addr = sec->offset_within_address_space;
626 pcibus_t machine_addr = s->bases[bar].access.maddr
627 + sec->offset_within_region;
628 pcibus_t size = int128_get64(sec->size);
629 rc = xc_domain_memory_mapping(xen_xc, xen_domid,
630 XEN_PFN(guest_addr + XC_PAGE_SIZE - 1),
631 XEN_PFN(machine_addr + XC_PAGE_SIZE - 1),
632 XEN_PFN(size + XC_PAGE_SIZE - 1),
633 op);
634 if (rc) {
635 XEN_PT_ERR(d, "%s mem mapping failed! (err: %i)\n",
636 adding ? "create new" : "remove old", errno);
641 static void xen_pt_region_add(MemoryListener *l, MemoryRegionSection *sec)
643 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
644 memory_listener);
646 memory_region_ref(sec->mr);
647 xen_pt_region_update(s, sec, true);
650 static void xen_pt_region_del(MemoryListener *l, MemoryRegionSection *sec)
652 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
653 memory_listener);
655 xen_pt_region_update(s, sec, false);
656 memory_region_unref(sec->mr);
659 static void xen_pt_io_region_add(MemoryListener *l, MemoryRegionSection *sec)
661 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
662 io_listener);
664 memory_region_ref(sec->mr);
665 xen_pt_region_update(s, sec, true);
668 static void xen_pt_io_region_del(MemoryListener *l, MemoryRegionSection *sec)
670 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
671 io_listener);
673 xen_pt_region_update(s, sec, false);
674 memory_region_unref(sec->mr);
677 static const MemoryListener xen_pt_memory_listener = {
678 .region_add = xen_pt_region_add,
679 .region_del = xen_pt_region_del,
680 .priority = 10,
683 static const MemoryListener xen_pt_io_listener = {
684 .region_add = xen_pt_io_region_add,
685 .region_del = xen_pt_io_region_del,
686 .priority = 10,
689 static void
690 xen_igd_passthrough_isa_bridge_create(XenPCIPassthroughState *s,
691 XenHostPCIDevice *dev)
693 uint16_t gpu_dev_id;
694 PCIDevice *d = &s->dev;
696 gpu_dev_id = dev->device_id;
697 igd_passthrough_isa_bridge_create(d->bus, gpu_dev_id);
700 /* destroy. */
701 static void xen_pt_destroy(PCIDevice *d) {
703 XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
704 XenHostPCIDevice *host_dev = &s->real_device;
705 uint8_t machine_irq = s->machine_irq;
706 uint8_t intx;
707 int rc;
709 if (machine_irq && !xen_host_pci_device_closed(&s->real_device)) {
710 intx = xen_pt_pci_intx(s);
711 rc = xc_domain_unbind_pt_irq(xen_xc, xen_domid, machine_irq,
712 PT_IRQ_TYPE_PCI,
713 pci_bus_num(d->bus),
714 PCI_SLOT(s->dev.devfn),
715 intx,
716 0 /* isa_irq */);
717 if (rc < 0) {
718 XEN_PT_ERR(d, "unbinding of interrupt INT%c failed."
719 " (machine irq: %i, err: %d)"
720 " But bravely continuing on..\n",
721 'a' + intx, machine_irq, errno);
725 /* N.B. xen_pt_config_delete takes care of freeing them. */
726 if (s->msi) {
727 xen_pt_msi_disable(s);
729 if (s->msix) {
730 xen_pt_msix_disable(s);
733 if (machine_irq) {
734 xen_pt_mapped_machine_irq[machine_irq]--;
736 if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
737 rc = xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq);
739 if (rc < 0) {
740 XEN_PT_ERR(d, "unmapping of interrupt %i failed. (err: %d)"
741 " But bravely continuing on..\n",
742 machine_irq, errno);
745 s->machine_irq = 0;
748 /* delete all emulated config registers */
749 xen_pt_config_delete(s);
751 xen_pt_unregister_vga_regions(host_dev);
753 if (s->listener_set) {
754 memory_listener_unregister(&s->memory_listener);
755 memory_listener_unregister(&s->io_listener);
756 s->listener_set = false;
758 if (!xen_host_pci_device_closed(&s->real_device)) {
759 xen_host_pci_device_put(&s->real_device);
762 /* init */
764 static void xen_pt_realize(PCIDevice *d, Error **errp)
766 XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
767 int i, rc = 0;
768 uint8_t machine_irq = 0, scratch;
769 uint16_t cmd = 0;
770 int pirq = XEN_PT_UNASSIGNED_PIRQ;
771 Error *err = NULL;
773 /* register real device */
774 XEN_PT_LOG(d, "Assigning real physical device %02x:%02x.%d"
775 " to devfn %#x\n",
776 s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function,
777 s->dev.devfn);
779 xen_host_pci_device_get(&s->real_device,
780 s->hostaddr.domain, s->hostaddr.bus,
781 s->hostaddr.slot, s->hostaddr.function,
782 &err);
783 if (err) {
784 error_append_hint(&err, "Failed to \"open\" the real pci device");
785 error_propagate(errp, err);
786 return;
789 s->is_virtfn = s->real_device.is_virtfn;
790 if (s->is_virtfn) {
791 XEN_PT_LOG(d, "%04x:%02x:%02x.%d is a SR-IOV Virtual Function\n",
792 s->real_device.domain, s->real_device.bus,
793 s->real_device.dev, s->real_device.func);
796 /* Initialize virtualized PCI configuration (Extended 256 Bytes) */
797 memset(d->config, 0, PCI_CONFIG_SPACE_SIZE);
799 s->memory_listener = xen_pt_memory_listener;
800 s->io_listener = xen_pt_io_listener;
802 /* Setup VGA bios for passthrough GFX */
803 if ((s->real_device.domain == 0) && (s->real_device.bus == 0) &&
804 (s->real_device.dev == 2) && (s->real_device.func == 0)) {
805 if (!is_igd_vga_passthrough(&s->real_device)) {
806 error_setg(errp, "Need to enable igd-passthru if you're trying"
807 " to passthrough IGD GFX");
808 xen_host_pci_device_put(&s->real_device);
809 return;
812 xen_pt_setup_vga(s, &s->real_device, &err);
813 if (err) {
814 error_append_hint(&err, "Setup VGA BIOS of passthrough"
815 " GFX failed");
816 error_propagate(errp, err);
817 xen_host_pci_device_put(&s->real_device);
818 return;
821 /* Register ISA bridge for passthrough GFX. */
822 xen_igd_passthrough_isa_bridge_create(s, &s->real_device);
825 /* Handle real device's MMIO/PIO BARs */
826 xen_pt_register_regions(s, &cmd);
828 /* reinitialize each config register to be emulated */
829 xen_pt_config_init(s, &err);
830 if (err) {
831 error_append_hint(&err, "PCI Config space initialisation failed");
832 error_report_err(err);
833 rc = -1;
834 goto err_out;
837 /* Bind interrupt */
838 rc = xen_host_pci_get_byte(&s->real_device, PCI_INTERRUPT_PIN, &scratch);
839 if (rc) {
840 error_setg_errno(errp, errno, "Failed to read PCI_INTERRUPT_PIN");
841 goto err_out;
843 if (!scratch) {
844 error_setg(errp, "no pin interrupt");
845 goto out;
848 machine_irq = s->real_device.irq;
849 rc = xc_physdev_map_pirq(xen_xc, xen_domid, machine_irq, &pirq);
850 if (rc < 0) {
851 error_setg_errno(errp, errno, "Mapping machine irq %u to"
852 " pirq %i failed", machine_irq, pirq);
854 /* Disable PCI intx assertion (turn on bit10 of devctl) */
855 cmd |= PCI_COMMAND_INTX_DISABLE;
856 machine_irq = 0;
857 s->machine_irq = 0;
858 } else {
859 machine_irq = pirq;
860 s->machine_irq = pirq;
861 xen_pt_mapped_machine_irq[machine_irq]++;
864 /* bind machine_irq to device */
865 if (machine_irq != 0) {
866 uint8_t e_intx = xen_pt_pci_intx(s);
868 rc = xc_domain_bind_pt_pci_irq(xen_xc, xen_domid, machine_irq,
869 pci_bus_num(d->bus),
870 PCI_SLOT(d->devfn),
871 e_intx);
872 if (rc < 0) {
873 error_setg_errno(errp, errno, "Binding of interrupt %u failed",
874 e_intx);
876 /* Disable PCI intx assertion (turn on bit10 of devctl) */
877 cmd |= PCI_COMMAND_INTX_DISABLE;
878 xen_pt_mapped_machine_irq[machine_irq]--;
880 if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
881 if (xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq)) {
882 error_setg_errno(errp, errno, "Unmapping of machine"
883 " interrupt %u failed", machine_irq);
886 s->machine_irq = 0;
890 out:
891 if (cmd) {
892 uint16_t val;
894 rc = xen_host_pci_get_word(&s->real_device, PCI_COMMAND, &val);
895 if (rc) {
896 error_setg_errno(errp, errno, "Failed to read PCI_COMMAND");
897 goto err_out;
898 } else {
899 val |= cmd;
900 rc = xen_host_pci_set_word(&s->real_device, PCI_COMMAND, val);
901 if (rc) {
902 error_setg_errno(errp, errno, "Failed to write PCI_COMMAND"
903 " val = 0x%x", val);
904 goto err_out;
909 memory_listener_register(&s->memory_listener, &s->dev.bus_master_as);
910 memory_listener_register(&s->io_listener, &address_space_io);
911 s->listener_set = true;
912 XEN_PT_LOG(d,
913 "Real physical device %02x:%02x.%d registered successfully\n",
914 s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function);
916 return;
918 err_out:
919 for (i = 0; i < PCI_ROM_SLOT; i++) {
920 object_unparent(OBJECT(&s->bar[i]));
922 object_unparent(OBJECT(&s->rom));
924 xen_pt_destroy(d);
925 assert(rc);
928 static void xen_pt_unregister_device(PCIDevice *d)
930 xen_pt_destroy(d);
933 static Property xen_pci_passthrough_properties[] = {
934 DEFINE_PROP_PCI_HOST_DEVADDR("hostaddr", XenPCIPassthroughState, hostaddr),
935 DEFINE_PROP_BOOL("permissive", XenPCIPassthroughState, permissive, false),
936 DEFINE_PROP_END_OF_LIST(),
939 static void xen_pci_passthrough_class_init(ObjectClass *klass, void *data)
941 DeviceClass *dc = DEVICE_CLASS(klass);
942 PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
944 k->realize = xen_pt_realize;
945 k->exit = xen_pt_unregister_device;
946 k->config_read = xen_pt_pci_read_config;
947 k->config_write = xen_pt_pci_write_config;
948 set_bit(DEVICE_CATEGORY_MISC, dc->categories);
949 dc->desc = "Assign an host PCI device with Xen";
950 dc->props = xen_pci_passthrough_properties;
953 static void xen_pci_passthrough_finalize(Object *obj)
955 XenPCIPassthroughState *s = XEN_PT_DEVICE(obj);
957 xen_pt_msix_delete(s);
960 static const TypeInfo xen_pci_passthrough_info = {
961 .name = TYPE_XEN_PT_DEVICE,
962 .parent = TYPE_PCI_DEVICE,
963 .instance_size = sizeof(XenPCIPassthroughState),
964 .instance_finalize = xen_pci_passthrough_finalize,
965 .class_init = xen_pci_passthrough_class_init,
968 static void xen_pci_passthrough_register_types(void)
970 type_register_static(&xen_pci_passthrough_info);
973 type_init(xen_pci_passthrough_register_types)