hw/arm/virt: Support using SMC for PSCI
[qemu/ar7.git] / hw / xen / xen_pt.c
blobb6d71bb52aa44bd4f342c893fcebe8808c15fb09
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 "qapi/error.h"
57 #include <sys/ioctl.h>
59 #include "hw/pci/pci.h"
60 #include "hw/xen/xen.h"
61 #include "hw/i386/pc.h"
62 #include "hw/xen/xen_backend.h"
63 #include "xen_pt.h"
64 #include "qemu/range.h"
65 #include "exec/address-spaces.h"
67 #define XEN_PT_NR_IRQS (256)
68 static uint8_t xen_pt_mapped_machine_irq[XEN_PT_NR_IRQS] = {0};
70 void xen_pt_log(const PCIDevice *d, const char *f, ...)
72 va_list ap;
74 va_start(ap, f);
75 if (d) {
76 fprintf(stderr, "[%02x:%02x.%d] ", pci_bus_num(d->bus),
77 PCI_SLOT(d->devfn), PCI_FUNC(d->devfn));
79 vfprintf(stderr, f, ap);
80 va_end(ap);
83 /* Config Space */
85 static int xen_pt_pci_config_access_check(PCIDevice *d, uint32_t addr, int len)
87 /* check offset range */
88 if (addr >= 0xFF) {
89 XEN_PT_ERR(d, "Failed to access register with offset exceeding 0xFF. "
90 "(addr: 0x%02x, len: %d)\n", addr, len);
91 return -1;
94 /* check read size */
95 if ((len != 1) && (len != 2) && (len != 4)) {
96 XEN_PT_ERR(d, "Failed to access register with invalid access length. "
97 "(addr: 0x%02x, len: %d)\n", addr, len);
98 return -1;
101 /* check offset alignment */
102 if (addr & (len - 1)) {
103 XEN_PT_ERR(d, "Failed to access register with invalid access size "
104 "alignment. (addr: 0x%02x, len: %d)\n", addr, len);
105 return -1;
108 return 0;
111 int xen_pt_bar_offset_to_index(uint32_t offset)
113 int index = 0;
115 /* check Exp ROM BAR */
116 if (offset == PCI_ROM_ADDRESS) {
117 return PCI_ROM_SLOT;
120 /* calculate BAR index */
121 index = (offset - PCI_BASE_ADDRESS_0) >> 2;
122 if (index >= PCI_NUM_REGIONS) {
123 return -1;
126 return index;
129 static uint32_t xen_pt_pci_read_config(PCIDevice *d, uint32_t addr, int len)
131 XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
132 uint32_t val = 0;
133 XenPTRegGroup *reg_grp_entry = NULL;
134 XenPTReg *reg_entry = NULL;
135 int rc = 0;
136 int emul_len = 0;
137 uint32_t find_addr = addr;
139 if (xen_pt_pci_config_access_check(d, addr, len)) {
140 goto exit;
143 /* find register group entry */
144 reg_grp_entry = xen_pt_find_reg_grp(s, addr);
145 if (reg_grp_entry) {
146 /* check 0-Hardwired register group */
147 if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
148 /* no need to emulate, just return 0 */
149 val = 0;
150 goto exit;
154 /* read I/O device register value */
155 rc = xen_host_pci_get_block(&s->real_device, addr, (uint8_t *)&val, len);
156 if (rc < 0) {
157 XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
158 memset(&val, 0xff, len);
161 /* just return the I/O device register value for
162 * passthrough type register group */
163 if (reg_grp_entry == NULL) {
164 goto exit;
167 /* adjust the read value to appropriate CFC-CFF window */
168 val <<= (addr & 3) << 3;
169 emul_len = len;
171 /* loop around the guest requested size */
172 while (emul_len > 0) {
173 /* find register entry to be emulated */
174 reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
175 if (reg_entry) {
176 XenPTRegInfo *reg = reg_entry->reg;
177 uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
178 uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
179 uint8_t *ptr_val = NULL;
181 valid_mask <<= (find_addr - real_offset) << 3;
182 ptr_val = (uint8_t *)&val + (real_offset & 3);
184 /* do emulation based on register size */
185 switch (reg->size) {
186 case 1:
187 if (reg->u.b.read) {
188 rc = reg->u.b.read(s, reg_entry, ptr_val, valid_mask);
190 break;
191 case 2:
192 if (reg->u.w.read) {
193 rc = reg->u.w.read(s, reg_entry,
194 (uint16_t *)ptr_val, valid_mask);
196 break;
197 case 4:
198 if (reg->u.dw.read) {
199 rc = reg->u.dw.read(s, reg_entry,
200 (uint32_t *)ptr_val, valid_mask);
202 break;
205 if (rc < 0) {
206 xen_shutdown_fatal_error("Internal error: Invalid read "
207 "emulation. (%s, rc: %d)\n",
208 __func__, rc);
209 return 0;
212 /* calculate next address to find */
213 emul_len -= reg->size;
214 if (emul_len > 0) {
215 find_addr = real_offset + reg->size;
217 } else {
218 /* nothing to do with passthrough type register,
219 * continue to find next byte */
220 emul_len--;
221 find_addr++;
225 /* need to shift back before returning them to pci bus emulator */
226 val >>= ((addr & 3) << 3);
228 exit:
229 XEN_PT_LOG_CONFIG(d, addr, val, len);
230 return val;
233 static void xen_pt_pci_write_config(PCIDevice *d, uint32_t addr,
234 uint32_t val, int len)
236 XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
237 int index = 0;
238 XenPTRegGroup *reg_grp_entry = NULL;
239 int rc = 0;
240 uint32_t read_val = 0, wb_mask;
241 int emul_len = 0;
242 XenPTReg *reg_entry = NULL;
243 uint32_t find_addr = addr;
244 XenPTRegInfo *reg = NULL;
245 bool wp_flag = false;
247 if (xen_pt_pci_config_access_check(d, addr, len)) {
248 return;
251 XEN_PT_LOG_CONFIG(d, addr, val, len);
253 /* check unused BAR register */
254 index = xen_pt_bar_offset_to_index(addr);
255 if ((index >= 0) && (val != 0)) {
256 uint32_t chk = val;
258 if (index == PCI_ROM_SLOT)
259 chk |= (uint32_t)~PCI_ROM_ADDRESS_MASK;
261 if ((chk != XEN_PT_BAR_ALLF) &&
262 (s->bases[index].bar_flag == XEN_PT_BAR_FLAG_UNUSED)) {
263 XEN_PT_WARN(d, "Guest attempt to set address to unused "
264 "Base Address Register. (addr: 0x%02x, len: %d)\n",
265 addr, len);
269 /* find register group entry */
270 reg_grp_entry = xen_pt_find_reg_grp(s, addr);
271 if (reg_grp_entry) {
272 /* check 0-Hardwired register group */
273 if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
274 /* ignore silently */
275 XEN_PT_WARN(d, "Access to 0-Hardwired register. "
276 "(addr: 0x%02x, len: %d)\n", addr, len);
277 return;
281 rc = xen_host_pci_get_block(&s->real_device, addr,
282 (uint8_t *)&read_val, len);
283 if (rc < 0) {
284 XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
285 memset(&read_val, 0xff, len);
286 wb_mask = 0;
287 } else {
288 wb_mask = 0xFFFFFFFF >> ((4 - len) << 3);
291 /* pass directly to the real device for passthrough type register group */
292 if (reg_grp_entry == NULL) {
293 if (!s->permissive) {
294 wb_mask = 0;
295 wp_flag = true;
297 goto out;
300 memory_region_transaction_begin();
301 pci_default_write_config(d, addr, val, len);
303 /* adjust the read and write value to appropriate CFC-CFF window */
304 read_val <<= (addr & 3) << 3;
305 val <<= (addr & 3) << 3;
306 emul_len = len;
308 /* loop around the guest requested size */
309 while (emul_len > 0) {
310 /* find register entry to be emulated */
311 reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
312 if (reg_entry) {
313 reg = reg_entry->reg;
314 uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
315 uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
316 uint8_t *ptr_val = NULL;
317 uint32_t wp_mask = reg->emu_mask | reg->ro_mask;
319 valid_mask <<= (find_addr - real_offset) << 3;
320 ptr_val = (uint8_t *)&val + (real_offset & 3);
321 if (!s->permissive) {
322 wp_mask |= reg->res_mask;
324 if (wp_mask == (0xFFFFFFFF >> ((4 - reg->size) << 3))) {
325 wb_mask &= ~((wp_mask >> ((find_addr - real_offset) << 3))
326 << ((len - emul_len) << 3));
329 /* do emulation based on register size */
330 switch (reg->size) {
331 case 1:
332 if (reg->u.b.write) {
333 rc = reg->u.b.write(s, reg_entry, ptr_val,
334 read_val >> ((real_offset & 3) << 3),
335 valid_mask);
337 break;
338 case 2:
339 if (reg->u.w.write) {
340 rc = reg->u.w.write(s, reg_entry, (uint16_t *)ptr_val,
341 (read_val >> ((real_offset & 3) << 3)),
342 valid_mask);
344 break;
345 case 4:
346 if (reg->u.dw.write) {
347 rc = reg->u.dw.write(s, reg_entry, (uint32_t *)ptr_val,
348 (read_val >> ((real_offset & 3) << 3)),
349 valid_mask);
351 break;
354 if (rc < 0) {
355 xen_shutdown_fatal_error("Internal error: Invalid write"
356 " emulation. (%s, rc: %d)\n",
357 __func__, rc);
358 return;
361 /* calculate next address to find */
362 emul_len -= reg->size;
363 if (emul_len > 0) {
364 find_addr = real_offset + reg->size;
366 } else {
367 /* nothing to do with passthrough type register,
368 * continue to find next byte */
369 if (!s->permissive) {
370 wb_mask &= ~(0xff << ((len - emul_len) << 3));
371 /* Unused BARs will make it here, but we don't want to issue
372 * warnings for writes to them (bogus writes get dealt with
373 * above).
375 if (index < 0) {
376 wp_flag = true;
379 emul_len--;
380 find_addr++;
384 /* need to shift back before passing them to xen_host_pci_set_block. */
385 val >>= (addr & 3) << 3;
387 memory_region_transaction_commit();
389 out:
390 if (wp_flag && !s->permissive_warned) {
391 s->permissive_warned = true;
392 xen_pt_log(d, "Write-back to unknown field 0x%02x (partially) inhibited (0x%0*x)\n",
393 addr, len * 2, wb_mask);
394 xen_pt_log(d, "If the device doesn't work, try enabling permissive mode\n");
395 xen_pt_log(d, "(unsafe) and if it helps report the problem to xen-devel\n");
397 for (index = 0; wb_mask; index += len) {
398 /* unknown regs are passed through */
399 while (!(wb_mask & 0xff)) {
400 index++;
401 wb_mask >>= 8;
403 len = 0;
404 do {
405 len++;
406 wb_mask >>= 8;
407 } while (wb_mask & 0xff);
408 rc = xen_host_pci_set_block(&s->real_device, addr + index,
409 (uint8_t *)&val + index, len);
411 if (rc < 0) {
412 XEN_PT_ERR(d, "xen_host_pci_set_block failed. return value: %d.\n", rc);
417 /* register regions */
419 static uint64_t xen_pt_bar_read(void *o, hwaddr addr,
420 unsigned size)
422 PCIDevice *d = o;
423 /* if this function is called, that probably means that there is a
424 * misconfiguration of the IOMMU. */
425 XEN_PT_ERR(d, "Should not read BAR through QEMU. @0x"TARGET_FMT_plx"\n",
426 addr);
427 return 0;
429 static void xen_pt_bar_write(void *o, hwaddr addr, uint64_t val,
430 unsigned size)
432 PCIDevice *d = o;
433 /* Same comment as xen_pt_bar_read function */
434 XEN_PT_ERR(d, "Should not write BAR through QEMU. @0x"TARGET_FMT_plx"\n",
435 addr);
438 static const MemoryRegionOps ops = {
439 .endianness = DEVICE_NATIVE_ENDIAN,
440 .read = xen_pt_bar_read,
441 .write = xen_pt_bar_write,
444 static int xen_pt_register_regions(XenPCIPassthroughState *s, uint16_t *cmd)
446 int i = 0;
447 XenHostPCIDevice *d = &s->real_device;
449 /* Register PIO/MMIO BARs */
450 for (i = 0; i < PCI_ROM_SLOT; i++) {
451 XenHostPCIIORegion *r = &d->io_regions[i];
452 uint8_t type;
454 if (r->base_addr == 0 || r->size == 0) {
455 continue;
458 s->bases[i].access.u = r->base_addr;
460 if (r->type & XEN_HOST_PCI_REGION_TYPE_IO) {
461 type = PCI_BASE_ADDRESS_SPACE_IO;
462 *cmd |= PCI_COMMAND_IO;
463 } else {
464 type = PCI_BASE_ADDRESS_SPACE_MEMORY;
465 if (r->type & XEN_HOST_PCI_REGION_TYPE_PREFETCH) {
466 type |= PCI_BASE_ADDRESS_MEM_PREFETCH;
468 if (r->type & XEN_HOST_PCI_REGION_TYPE_MEM_64) {
469 type |= PCI_BASE_ADDRESS_MEM_TYPE_64;
471 *cmd |= PCI_COMMAND_MEMORY;
474 memory_region_init_io(&s->bar[i], OBJECT(s), &ops, &s->dev,
475 "xen-pci-pt-bar", r->size);
476 pci_register_bar(&s->dev, i, type, &s->bar[i]);
478 XEN_PT_LOG(&s->dev, "IO region %i registered (size=0x%08"PRIx64
479 " base_addr=0x%08"PRIx64" type: %#x)\n",
480 i, r->size, r->base_addr, type);
483 /* Register expansion ROM address */
484 if (d->rom.base_addr && d->rom.size) {
485 uint32_t bar_data = 0;
487 /* Re-set BAR reported by OS, otherwise ROM can't be read. */
488 if (xen_host_pci_get_long(d, PCI_ROM_ADDRESS, &bar_data)) {
489 return 0;
491 if ((bar_data & PCI_ROM_ADDRESS_MASK) == 0) {
492 bar_data |= d->rom.base_addr & PCI_ROM_ADDRESS_MASK;
493 xen_host_pci_set_long(d, PCI_ROM_ADDRESS, bar_data);
496 s->bases[PCI_ROM_SLOT].access.maddr = d->rom.base_addr;
498 memory_region_init_io(&s->rom, OBJECT(s), &ops, &s->dev,
499 "xen-pci-pt-rom", d->rom.size);
500 pci_register_bar(&s->dev, PCI_ROM_SLOT, PCI_BASE_ADDRESS_MEM_PREFETCH,
501 &s->rom);
503 XEN_PT_LOG(&s->dev, "Expansion ROM registered (size=0x%08"PRIx64
504 " base_addr=0x%08"PRIx64")\n",
505 d->rom.size, d->rom.base_addr);
508 xen_pt_register_vga_regions(d);
509 return 0;
512 /* region mapping */
514 static int xen_pt_bar_from_region(XenPCIPassthroughState *s, MemoryRegion *mr)
516 int i = 0;
518 for (i = 0; i < PCI_NUM_REGIONS - 1; i++) {
519 if (mr == &s->bar[i]) {
520 return i;
523 if (mr == &s->rom) {
524 return PCI_ROM_SLOT;
526 return -1;
530 * This function checks if an io_region overlaps an io_region from another
531 * device. The io_region to check is provided with (addr, size and type)
532 * A callback can be provided and will be called for every region that is
533 * overlapped.
534 * The return value indicates if the region is overlappsed */
535 struct CheckBarArgs {
536 XenPCIPassthroughState *s;
537 pcibus_t addr;
538 pcibus_t size;
539 uint8_t type;
540 bool rc;
542 static void xen_pt_check_bar_overlap(PCIBus *bus, PCIDevice *d, void *opaque)
544 struct CheckBarArgs *arg = opaque;
545 XenPCIPassthroughState *s = arg->s;
546 uint8_t type = arg->type;
547 int i;
549 if (d->devfn == s->dev.devfn) {
550 return;
553 /* xxx: This ignores bridges. */
554 for (i = 0; i < PCI_NUM_REGIONS; i++) {
555 const PCIIORegion *r = &d->io_regions[i];
557 if (!r->size) {
558 continue;
560 if ((type & PCI_BASE_ADDRESS_SPACE_IO)
561 != (r->type & PCI_BASE_ADDRESS_SPACE_IO)) {
562 continue;
565 if (ranges_overlap(arg->addr, arg->size, r->addr, r->size)) {
566 XEN_PT_WARN(&s->dev,
567 "Overlapped to device [%02x:%02x.%d] Region: %i"
568 " (addr: %#"FMT_PCIBUS", len: %#"FMT_PCIBUS")\n",
569 pci_bus_num(bus), PCI_SLOT(d->devfn),
570 PCI_FUNC(d->devfn), i, r->addr, r->size);
571 arg->rc = true;
576 static void xen_pt_region_update(XenPCIPassthroughState *s,
577 MemoryRegionSection *sec, bool adding)
579 PCIDevice *d = &s->dev;
580 MemoryRegion *mr = sec->mr;
581 int bar = -1;
582 int rc;
583 int op = adding ? DPCI_ADD_MAPPING : DPCI_REMOVE_MAPPING;
584 struct CheckBarArgs args = {
585 .s = s,
586 .addr = sec->offset_within_address_space,
587 .size = int128_get64(sec->size),
588 .rc = false,
591 bar = xen_pt_bar_from_region(s, mr);
592 if (bar == -1 && (!s->msix || &s->msix->mmio != mr)) {
593 return;
596 if (s->msix && &s->msix->mmio == mr) {
597 if (adding) {
598 s->msix->mmio_base_addr = sec->offset_within_address_space;
599 rc = xen_pt_msix_update_remap(s, s->msix->bar_index);
601 return;
604 args.type = d->io_regions[bar].type;
605 pci_for_each_device(d->bus, pci_bus_num(d->bus),
606 xen_pt_check_bar_overlap, &args);
607 if (args.rc) {
608 XEN_PT_WARN(d, "Region: %d (addr: %#"FMT_PCIBUS
609 ", len: %#"FMT_PCIBUS") is overlapped.\n",
610 bar, sec->offset_within_address_space,
611 int128_get64(sec->size));
614 if (d->io_regions[bar].type & PCI_BASE_ADDRESS_SPACE_IO) {
615 uint32_t guest_port = sec->offset_within_address_space;
616 uint32_t machine_port = s->bases[bar].access.pio_base;
617 uint32_t size = int128_get64(sec->size);
618 rc = xc_domain_ioport_mapping(xen_xc, xen_domid,
619 guest_port, machine_port, size,
620 op);
621 if (rc) {
622 XEN_PT_ERR(d, "%s ioport mapping failed! (err: %i)\n",
623 adding ? "create new" : "remove old", errno);
625 } else {
626 pcibus_t guest_addr = sec->offset_within_address_space;
627 pcibus_t machine_addr = s->bases[bar].access.maddr
628 + sec->offset_within_region;
629 pcibus_t size = int128_get64(sec->size);
630 rc = xc_domain_memory_mapping(xen_xc, xen_domid,
631 XEN_PFN(guest_addr + XC_PAGE_SIZE - 1),
632 XEN_PFN(machine_addr + XC_PAGE_SIZE - 1),
633 XEN_PFN(size + XC_PAGE_SIZE - 1),
634 op);
635 if (rc) {
636 XEN_PT_ERR(d, "%s mem mapping failed! (err: %i)\n",
637 adding ? "create new" : "remove old", errno);
642 static void xen_pt_region_add(MemoryListener *l, MemoryRegionSection *sec)
644 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
645 memory_listener);
647 memory_region_ref(sec->mr);
648 xen_pt_region_update(s, sec, true);
651 static void xen_pt_region_del(MemoryListener *l, MemoryRegionSection *sec)
653 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
654 memory_listener);
656 xen_pt_region_update(s, sec, false);
657 memory_region_unref(sec->mr);
660 static void xen_pt_io_region_add(MemoryListener *l, MemoryRegionSection *sec)
662 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
663 io_listener);
665 memory_region_ref(sec->mr);
666 xen_pt_region_update(s, sec, true);
669 static void xen_pt_io_region_del(MemoryListener *l, MemoryRegionSection *sec)
671 XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
672 io_listener);
674 xen_pt_region_update(s, sec, false);
675 memory_region_unref(sec->mr);
678 static const MemoryListener xen_pt_memory_listener = {
679 .region_add = xen_pt_region_add,
680 .region_del = xen_pt_region_del,
681 .priority = 10,
684 static const MemoryListener xen_pt_io_listener = {
685 .region_add = xen_pt_io_region_add,
686 .region_del = xen_pt_io_region_del,
687 .priority = 10,
690 static void
691 xen_igd_passthrough_isa_bridge_create(XenPCIPassthroughState *s,
692 XenHostPCIDevice *dev)
694 uint16_t gpu_dev_id;
695 PCIDevice *d = &s->dev;
697 gpu_dev_id = dev->device_id;
698 igd_passthrough_isa_bridge_create(d->bus, gpu_dev_id);
701 /* destroy. */
702 static void xen_pt_destroy(PCIDevice *d) {
704 XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
705 XenHostPCIDevice *host_dev = &s->real_device;
706 uint8_t machine_irq = s->machine_irq;
707 uint8_t intx;
708 int rc;
710 if (machine_irq && !xen_host_pci_device_closed(&s->real_device)) {
711 intx = xen_pt_pci_intx(s);
712 rc = xc_domain_unbind_pt_irq(xen_xc, xen_domid, machine_irq,
713 PT_IRQ_TYPE_PCI,
714 pci_bus_num(d->bus),
715 PCI_SLOT(s->dev.devfn),
716 intx,
717 0 /* isa_irq */);
718 if (rc < 0) {
719 XEN_PT_ERR(d, "unbinding of interrupt INT%c failed."
720 " (machine irq: %i, err: %d)"
721 " But bravely continuing on..\n",
722 'a' + intx, machine_irq, errno);
726 /* N.B. xen_pt_config_delete takes care of freeing them. */
727 if (s->msi) {
728 xen_pt_msi_disable(s);
730 if (s->msix) {
731 xen_pt_msix_disable(s);
734 if (machine_irq) {
735 xen_pt_mapped_machine_irq[machine_irq]--;
737 if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
738 rc = xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq);
740 if (rc < 0) {
741 XEN_PT_ERR(d, "unmapping of interrupt %i failed. (err: %d)"
742 " But bravely continuing on..\n",
743 machine_irq, errno);
746 s->machine_irq = 0;
749 /* delete all emulated config registers */
750 xen_pt_config_delete(s);
752 xen_pt_unregister_vga_regions(host_dev);
754 if (s->listener_set) {
755 memory_listener_unregister(&s->memory_listener);
756 memory_listener_unregister(&s->io_listener);
757 s->listener_set = false;
759 if (!xen_host_pci_device_closed(&s->real_device)) {
760 xen_host_pci_device_put(&s->real_device);
763 /* init */
765 static void xen_pt_realize(PCIDevice *d, Error **errp)
767 XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
768 int i, rc = 0;
769 uint8_t machine_irq = 0, scratch;
770 uint16_t cmd = 0;
771 int pirq = XEN_PT_UNASSIGNED_PIRQ;
772 Error *err = NULL;
774 /* register real device */
775 XEN_PT_LOG(d, "Assigning real physical device %02x:%02x.%d"
776 " to devfn %#x\n",
777 s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function,
778 s->dev.devfn);
780 xen_host_pci_device_get(&s->real_device,
781 s->hostaddr.domain, s->hostaddr.bus,
782 s->hostaddr.slot, s->hostaddr.function,
783 &err);
784 if (err) {
785 error_append_hint(&err, "Failed to \"open\" the real pci device");
786 error_propagate(errp, err);
787 return;
790 s->is_virtfn = s->real_device.is_virtfn;
791 if (s->is_virtfn) {
792 XEN_PT_LOG(d, "%04x:%02x:%02x.%d is a SR-IOV Virtual Function\n",
793 s->real_device.domain, s->real_device.bus,
794 s->real_device.dev, s->real_device.func);
797 /* Initialize virtualized PCI configuration (Extended 256 Bytes) */
798 memset(d->config, 0, PCI_CONFIG_SPACE_SIZE);
800 s->memory_listener = xen_pt_memory_listener;
801 s->io_listener = xen_pt_io_listener;
803 /* Setup VGA bios for passthrough GFX */
804 if ((s->real_device.domain == 0) && (s->real_device.bus == 0) &&
805 (s->real_device.dev == 2) && (s->real_device.func == 0)) {
806 if (!is_igd_vga_passthrough(&s->real_device)) {
807 error_setg(errp, "Need to enable igd-passthru if you're trying"
808 " to passthrough IGD GFX");
809 xen_host_pci_device_put(&s->real_device);
810 return;
813 xen_pt_setup_vga(s, &s->real_device, &err);
814 if (err) {
815 error_append_hint(&err, "Setup VGA BIOS of passthrough"
816 " GFX failed");
817 error_propagate(errp, err);
818 xen_host_pci_device_put(&s->real_device);
819 return;
822 /* Register ISA bridge for passthrough GFX. */
823 xen_igd_passthrough_isa_bridge_create(s, &s->real_device);
826 /* Handle real device's MMIO/PIO BARs */
827 xen_pt_register_regions(s, &cmd);
829 /* reinitialize each config register to be emulated */
830 xen_pt_config_init(s, &err);
831 if (err) {
832 error_append_hint(&err, "PCI Config space initialisation failed");
833 error_report_err(err);
834 rc = -1;
835 goto err_out;
838 /* Bind interrupt */
839 rc = xen_host_pci_get_byte(&s->real_device, PCI_INTERRUPT_PIN, &scratch);
840 if (rc) {
841 error_setg_errno(errp, errno, "Failed to read PCI_INTERRUPT_PIN");
842 goto err_out;
844 if (!scratch) {
845 XEN_PT_LOG(d, "no pin interrupt\n");
846 goto out;
849 machine_irq = s->real_device.irq;
850 rc = xc_physdev_map_pirq(xen_xc, xen_domid, machine_irq, &pirq);
851 if (rc < 0) {
852 error_setg_errno(errp, errno, "Mapping machine irq %u to"
853 " pirq %i failed", machine_irq, pirq);
855 /* Disable PCI intx assertion (turn on bit10 of devctl) */
856 cmd |= PCI_COMMAND_INTX_DISABLE;
857 machine_irq = 0;
858 s->machine_irq = 0;
859 } else {
860 machine_irq = pirq;
861 s->machine_irq = pirq;
862 xen_pt_mapped_machine_irq[machine_irq]++;
865 /* bind machine_irq to device */
866 if (machine_irq != 0) {
867 uint8_t e_intx = xen_pt_pci_intx(s);
869 rc = xc_domain_bind_pt_pci_irq(xen_xc, xen_domid, machine_irq,
870 pci_bus_num(d->bus),
871 PCI_SLOT(d->devfn),
872 e_intx);
873 if (rc < 0) {
874 error_setg_errno(errp, errno, "Binding of interrupt %u failed",
875 e_intx);
877 /* Disable PCI intx assertion (turn on bit10 of devctl) */
878 cmd |= PCI_COMMAND_INTX_DISABLE;
879 xen_pt_mapped_machine_irq[machine_irq]--;
881 if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
882 if (xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq)) {
883 error_setg_errno(errp, errno, "Unmapping of machine"
884 " interrupt %u failed", machine_irq);
887 s->machine_irq = 0;
891 out:
892 if (cmd) {
893 uint16_t val;
895 rc = xen_host_pci_get_word(&s->real_device, PCI_COMMAND, &val);
896 if (rc) {
897 error_setg_errno(errp, errno, "Failed to read PCI_COMMAND");
898 goto err_out;
899 } else {
900 val |= cmd;
901 rc = xen_host_pci_set_word(&s->real_device, PCI_COMMAND, val);
902 if (rc) {
903 error_setg_errno(errp, errno, "Failed to write PCI_COMMAND"
904 " val = 0x%x", val);
905 goto err_out;
910 memory_listener_register(&s->memory_listener, &s->dev.bus_master_as);
911 memory_listener_register(&s->io_listener, &address_space_io);
912 s->listener_set = true;
913 XEN_PT_LOG(d,
914 "Real physical device %02x:%02x.%d registered successfully\n",
915 s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function);
917 return;
919 err_out:
920 for (i = 0; i < PCI_ROM_SLOT; i++) {
921 object_unparent(OBJECT(&s->bar[i]));
923 object_unparent(OBJECT(&s->rom));
925 xen_pt_destroy(d);
926 assert(rc);
929 static void xen_pt_unregister_device(PCIDevice *d)
931 xen_pt_destroy(d);
934 static Property xen_pci_passthrough_properties[] = {
935 DEFINE_PROP_PCI_HOST_DEVADDR("hostaddr", XenPCIPassthroughState, hostaddr),
936 DEFINE_PROP_BOOL("permissive", XenPCIPassthroughState, permissive, false),
937 DEFINE_PROP_END_OF_LIST(),
940 static void xen_pci_passthrough_class_init(ObjectClass *klass, void *data)
942 DeviceClass *dc = DEVICE_CLASS(klass);
943 PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
945 k->realize = xen_pt_realize;
946 k->exit = xen_pt_unregister_device;
947 k->config_read = xen_pt_pci_read_config;
948 k->config_write = xen_pt_pci_write_config;
949 set_bit(DEVICE_CATEGORY_MISC, dc->categories);
950 dc->desc = "Assign an host PCI device with Xen";
951 dc->props = xen_pci_passthrough_properties;
954 static void xen_pci_passthrough_finalize(Object *obj)
956 XenPCIPassthroughState *s = XEN_PT_DEVICE(obj);
958 xen_pt_msix_delete(s);
961 static const TypeInfo xen_pci_passthrough_info = {
962 .name = TYPE_XEN_PT_DEVICE,
963 .parent = TYPE_PCI_DEVICE,
964 .instance_size = sizeof(XenPCIPassthroughState),
965 .instance_finalize = xen_pci_passthrough_finalize,
966 .class_init = xen_pci_passthrough_class_init,
969 static void xen_pci_passthrough_register_types(void)
971 type_register_static(&xen_pci_passthrough_info);
974 type_init(xen_pci_passthrough_register_types)