xref: /openbmc/qemu/hw/xen/xen_pt.c (revision 93e0932b)
1 /*
2  * Copyright (c) 2007, Neocleus Corporation.
3  * Copyright (c) 2007, Intel Corporation.
4  *
5  * This work is licensed under the terms of the GNU GPL, version 2.  See
6  * the COPYING file in the top-level directory.
7  *
8  * Alex Novik <alex@neocleus.com>
9  * Allen Kay <allen.m.kay@intel.com>
10  * Guy Zana <guy@neocleus.com>
11  *
12  * This file implements direct PCI assignment to a HVM guest
13  */
14 
15 /*
16  * Interrupt Disable policy:
17  *
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.
25  *
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'.
32  *
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'.
36  *
37  *     Write '1'
38  *       - Set real bit to '1'.
39  *
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'.
46  *
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'.
53  */
54 
55 #include "qemu/osdep.h"
56 #include "qapi/error.h"
57 #include <sys/ioctl.h>
58 
59 #include "hw/pci/pci.h"
60 #include "hw/qdev-properties.h"
61 #include "hw/qdev-properties-system.h"
62 #include "hw/xen/xen.h"
63 #include "hw/xen/xen-legacy-backend.h"
64 #include "xen_pt.h"
65 #include "qemu/range.h"
66 
67 static bool has_igd_gfx_passthru;
68 
69 bool xen_igd_gfx_pt_enabled(void)
70 {
71     return has_igd_gfx_passthru;
72 }
73 
74 void xen_igd_gfx_pt_set(bool value, Error **errp)
75 {
76     has_igd_gfx_passthru = value;
77 }
78 
79 #define XEN_PT_NR_IRQS (256)
80 static uint8_t xen_pt_mapped_machine_irq[XEN_PT_NR_IRQS] = {0};
81 
82 void xen_pt_log(const PCIDevice *d, const char *f, ...)
83 {
84     va_list ap;
85 
86     va_start(ap, f);
87     if (d) {
88         fprintf(stderr, "[%02x:%02x.%d] ", pci_dev_bus_num(d),
89                 PCI_SLOT(d->devfn), PCI_FUNC(d->devfn));
90     }
91     vfprintf(stderr, f, ap);
92     va_end(ap);
93 }
94 
95 /* Config Space */
96 
97 static int xen_pt_pci_config_access_check(PCIDevice *d, uint32_t addr, int len)
98 {
99     /* check offset range */
100     if (addr > 0xFF) {
101         XEN_PT_ERR(d, "Failed to access register with offset exceeding 0xFF. "
102                    "(addr: 0x%02x, len: %d)\n", addr, len);
103         return -1;
104     }
105 
106     /* check read size */
107     if ((len != 1) && (len != 2) && (len != 4)) {
108         XEN_PT_ERR(d, "Failed to access register with invalid access length. "
109                    "(addr: 0x%02x, len: %d)\n", addr, len);
110         return -1;
111     }
112 
113     /* check offset alignment */
114     if (addr & (len - 1)) {
115         XEN_PT_ERR(d, "Failed to access register with invalid access size "
116                    "alignment. (addr: 0x%02x, len: %d)\n", addr, len);
117         return -1;
118     }
119 
120     return 0;
121 }
122 
123 int xen_pt_bar_offset_to_index(uint32_t offset)
124 {
125     int index = 0;
126 
127     /* check Exp ROM BAR */
128     if (offset == PCI_ROM_ADDRESS) {
129         return PCI_ROM_SLOT;
130     }
131 
132     /* calculate BAR index */
133     index = (offset - PCI_BASE_ADDRESS_0) >> 2;
134     if (index >= PCI_NUM_REGIONS) {
135         return -1;
136     }
137 
138     return index;
139 }
140 
141 static uint32_t xen_pt_pci_read_config(PCIDevice *d, uint32_t addr, int len)
142 {
143     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
144     uint32_t val = 0;
145     XenPTRegGroup *reg_grp_entry = NULL;
146     XenPTReg *reg_entry = NULL;
147     int rc = 0;
148     int emul_len = 0;
149     uint32_t find_addr = addr;
150 
151     if (xen_pt_pci_config_access_check(d, addr, len)) {
152         goto exit;
153     }
154 
155     /* find register group entry */
156     reg_grp_entry = xen_pt_find_reg_grp(s, addr);
157     if (reg_grp_entry) {
158         /* check 0-Hardwired register group */
159         if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
160             /* no need to emulate, just return 0 */
161             val = 0;
162             goto exit;
163         }
164     }
165 
166     /* read I/O device register value */
167     rc = xen_host_pci_get_block(&s->real_device, addr, (uint8_t *)&val, len);
168     if (rc < 0) {
169         XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
170         memset(&val, 0xff, len);
171     }
172 
173     /* just return the I/O device register value for
174      * passthrough type register group */
175     if (reg_grp_entry == NULL) {
176         goto exit;
177     }
178 
179     /* adjust the read value to appropriate CFC-CFF window */
180     val <<= (addr & 3) << 3;
181     emul_len = len;
182 
183     /* loop around the guest requested size */
184     while (emul_len > 0) {
185         /* find register entry to be emulated */
186         reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
187         if (reg_entry) {
188             XenPTRegInfo *reg = reg_entry->reg;
189             uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
190             uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
191             uint8_t *ptr_val = NULL;
192 
193             valid_mask <<= (find_addr - real_offset) << 3;
194             ptr_val = (uint8_t *)&val + (real_offset & 3);
195 
196             /* do emulation based on register size */
197             switch (reg->size) {
198             case 1:
199                 if (reg->u.b.read) {
200                     rc = reg->u.b.read(s, reg_entry, ptr_val, valid_mask);
201                 }
202                 break;
203             case 2:
204                 if (reg->u.w.read) {
205                     rc = reg->u.w.read(s, reg_entry,
206                                        (uint16_t *)ptr_val, valid_mask);
207                 }
208                 break;
209             case 4:
210                 if (reg->u.dw.read) {
211                     rc = reg->u.dw.read(s, reg_entry,
212                                         (uint32_t *)ptr_val, valid_mask);
213                 }
214                 break;
215             }
216 
217             if (rc < 0) {
218                 xen_shutdown_fatal_error("Internal error: Invalid read "
219                                          "emulation. (%s, rc: %d)\n",
220                                          __func__, rc);
221                 return 0;
222             }
223 
224             /* calculate next address to find */
225             emul_len -= reg->size;
226             if (emul_len > 0) {
227                 find_addr = real_offset + reg->size;
228             }
229         } else {
230             /* nothing to do with passthrough type register,
231              * continue to find next byte */
232             emul_len--;
233             find_addr++;
234         }
235     }
236 
237     /* need to shift back before returning them to pci bus emulator */
238     val >>= ((addr & 3) << 3);
239 
240 exit:
241     XEN_PT_LOG_CONFIG(d, addr, val, len);
242     return val;
243 }
244 
245 static void xen_pt_pci_write_config(PCIDevice *d, uint32_t addr,
246                                     uint32_t val, int len)
247 {
248     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
249     int index = 0;
250     XenPTRegGroup *reg_grp_entry = NULL;
251     int rc = 0;
252     uint32_t read_val = 0, wb_mask;
253     int emul_len = 0;
254     XenPTReg *reg_entry = NULL;
255     uint32_t find_addr = addr;
256     XenPTRegInfo *reg = NULL;
257     bool wp_flag = false;
258 
259     if (xen_pt_pci_config_access_check(d, addr, len)) {
260         return;
261     }
262 
263     XEN_PT_LOG_CONFIG(d, addr, val, len);
264 
265     /* check unused BAR register */
266     index = xen_pt_bar_offset_to_index(addr);
267     if ((index >= 0) && (val != 0)) {
268         uint32_t chk = val;
269 
270         if (index == PCI_ROM_SLOT)
271             chk |= (uint32_t)~PCI_ROM_ADDRESS_MASK;
272 
273         if ((chk != XEN_PT_BAR_ALLF) &&
274             (s->bases[index].bar_flag == XEN_PT_BAR_FLAG_UNUSED)) {
275             XEN_PT_WARN(d, "Guest attempt to set address to unused "
276                         "Base Address Register. (addr: 0x%02x, len: %d)\n",
277                         addr, len);
278         }
279     }
280 
281     /* find register group entry */
282     reg_grp_entry = xen_pt_find_reg_grp(s, addr);
283     if (reg_grp_entry) {
284         /* check 0-Hardwired register group */
285         if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
286             /* ignore silently */
287             XEN_PT_WARN(d, "Access to 0-Hardwired register. "
288                         "(addr: 0x%02x, len: %d)\n", addr, len);
289             return;
290         }
291     }
292 
293     rc = xen_host_pci_get_block(&s->real_device, addr,
294                                 (uint8_t *)&read_val, len);
295     if (rc < 0) {
296         XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
297         memset(&read_val, 0xff, len);
298         wb_mask = 0;
299     } else {
300         wb_mask = 0xFFFFFFFF >> ((4 - len) << 3);
301     }
302 
303     /* pass directly to the real device for passthrough type register group */
304     if (reg_grp_entry == NULL) {
305         if (!s->permissive) {
306             wb_mask = 0;
307             wp_flag = true;
308         }
309         goto out;
310     }
311 
312     memory_region_transaction_begin();
313     pci_default_write_config(d, addr, val, len);
314 
315     /* adjust the read and write value to appropriate CFC-CFF window */
316     read_val <<= (addr & 3) << 3;
317     val <<= (addr & 3) << 3;
318     emul_len = len;
319 
320     /* loop around the guest requested size */
321     while (emul_len > 0) {
322         /* find register entry to be emulated */
323         reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
324         if (reg_entry) {
325             reg = reg_entry->reg;
326             uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
327             uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
328             uint8_t *ptr_val = NULL;
329             uint32_t wp_mask = reg->emu_mask | reg->ro_mask;
330 
331             valid_mask <<= (find_addr - real_offset) << 3;
332             ptr_val = (uint8_t *)&val + (real_offset & 3);
333             if (!s->permissive) {
334                 wp_mask |= reg->res_mask;
335             }
336             if (wp_mask == (0xFFFFFFFF >> ((4 - reg->size) << 3))) {
337                 wb_mask &= ~((wp_mask >> ((find_addr - real_offset) << 3))
338                              << ((len - emul_len) << 3));
339             }
340 
341             /* do emulation based on register size */
342             switch (reg->size) {
343             case 1:
344                 if (reg->u.b.write) {
345                     rc = reg->u.b.write(s, reg_entry, ptr_val,
346                                         read_val >> ((real_offset & 3) << 3),
347                                         valid_mask);
348                 }
349                 break;
350             case 2:
351                 if (reg->u.w.write) {
352                     rc = reg->u.w.write(s, reg_entry, (uint16_t *)ptr_val,
353                                         (read_val >> ((real_offset & 3) << 3)),
354                                         valid_mask);
355                 }
356                 break;
357             case 4:
358                 if (reg->u.dw.write) {
359                     rc = reg->u.dw.write(s, reg_entry, (uint32_t *)ptr_val,
360                                          (read_val >> ((real_offset & 3) << 3)),
361                                          valid_mask);
362                 }
363                 break;
364             }
365 
366             if (rc < 0) {
367                 xen_shutdown_fatal_error("Internal error: Invalid write"
368                                          " emulation. (%s, rc: %d)\n",
369                                          __func__, rc);
370                 return;
371             }
372 
373             /* calculate next address to find */
374             emul_len -= reg->size;
375             if (emul_len > 0) {
376                 find_addr = real_offset + reg->size;
377             }
378         } else {
379             /* nothing to do with passthrough type register,
380              * continue to find next byte */
381             if (!s->permissive) {
382                 wb_mask &= ~(0xff << ((len - emul_len) << 3));
383                 /* Unused BARs will make it here, but we don't want to issue
384                  * warnings for writes to them (bogus writes get dealt with
385                  * above).
386                  */
387                 if (index < 0) {
388                     wp_flag = true;
389                 }
390             }
391             emul_len--;
392             find_addr++;
393         }
394     }
395 
396     /* need to shift back before passing them to xen_host_pci_set_block. */
397     val >>= (addr & 3) << 3;
398 
399     memory_region_transaction_commit();
400 
401 out:
402     if (wp_flag && !s->permissive_warned) {
403         s->permissive_warned = true;
404         xen_pt_log(d, "Write-back to unknown field 0x%02x (partially) inhibited (0x%0*x)\n",
405                    addr, len * 2, wb_mask);
406         xen_pt_log(d, "If the device doesn't work, try enabling permissive mode\n");
407         xen_pt_log(d, "(unsafe) and if it helps report the problem to xen-devel\n");
408     }
409     for (index = 0; wb_mask; index += len) {
410         /* unknown regs are passed through */
411         while (!(wb_mask & 0xff)) {
412             index++;
413             wb_mask >>= 8;
414         }
415         len = 0;
416         do {
417             len++;
418             wb_mask >>= 8;
419         } while (wb_mask & 0xff);
420         rc = xen_host_pci_set_block(&s->real_device, addr + index,
421                                     (uint8_t *)&val + index, len);
422 
423         if (rc < 0) {
424             XEN_PT_ERR(d, "xen_host_pci_set_block failed. return value: %d.\n", rc);
425         }
426     }
427 }
428 
429 /* register regions */
430 
431 static uint64_t xen_pt_bar_read(void *o, hwaddr addr,
432                                 unsigned size)
433 {
434     PCIDevice *d = o;
435     /* if this function is called, that probably means that there is a
436      * misconfiguration of the IOMMU. */
437     XEN_PT_ERR(d, "Should not read BAR through QEMU. @0x"HWADDR_FMT_plx"\n",
438                addr);
439     return 0;
440 }
441 static void xen_pt_bar_write(void *o, hwaddr addr, uint64_t val,
442                              unsigned size)
443 {
444     PCIDevice *d = o;
445     /* Same comment as xen_pt_bar_read function */
446     XEN_PT_ERR(d, "Should not write BAR through QEMU. @0x"HWADDR_FMT_plx"\n",
447                addr);
448 }
449 
450 static const MemoryRegionOps ops = {
451     .endianness = DEVICE_NATIVE_ENDIAN,
452     .read = xen_pt_bar_read,
453     .write = xen_pt_bar_write,
454 };
455 
456 static int xen_pt_register_regions(XenPCIPassthroughState *s, uint16_t *cmd)
457 {
458     int i = 0;
459     XenHostPCIDevice *d = &s->real_device;
460 
461     /* Register PIO/MMIO BARs */
462     for (i = 0; i < PCI_ROM_SLOT; i++) {
463         XenHostPCIIORegion *r = &d->io_regions[i];
464         uint8_t type;
465 
466         if (r->base_addr == 0 || r->size == 0) {
467             continue;
468         }
469 
470         s->bases[i].access.u = r->base_addr;
471 
472         if (r->type & XEN_HOST_PCI_REGION_TYPE_IO) {
473             type = PCI_BASE_ADDRESS_SPACE_IO;
474             *cmd |= PCI_COMMAND_IO;
475         } else {
476             type = PCI_BASE_ADDRESS_SPACE_MEMORY;
477             if (r->type & XEN_HOST_PCI_REGION_TYPE_PREFETCH) {
478                 type |= PCI_BASE_ADDRESS_MEM_PREFETCH;
479             }
480             if (r->type & XEN_HOST_PCI_REGION_TYPE_MEM_64) {
481                 type |= PCI_BASE_ADDRESS_MEM_TYPE_64;
482             }
483             *cmd |= PCI_COMMAND_MEMORY;
484         }
485 
486         memory_region_init_io(&s->bar[i], OBJECT(s), &ops, &s->dev,
487                               "xen-pci-pt-bar", r->size);
488         pci_register_bar(&s->dev, i, type, &s->bar[i]);
489 
490         XEN_PT_LOG(&s->dev, "IO region %i registered (size=0x%08"PRIx64
491                    " base_addr=0x%08"PRIx64" type: 0x%x)\n",
492                    i, r->size, r->base_addr, type);
493     }
494 
495     /* Register expansion ROM address */
496     if (d->rom.base_addr && d->rom.size) {
497         uint32_t bar_data = 0;
498 
499         /* Re-set BAR reported by OS, otherwise ROM can't be read. */
500         if (xen_host_pci_get_long(d, PCI_ROM_ADDRESS, &bar_data)) {
501             return 0;
502         }
503         if ((bar_data & PCI_ROM_ADDRESS_MASK) == 0) {
504             bar_data |= d->rom.base_addr & PCI_ROM_ADDRESS_MASK;
505             xen_host_pci_set_long(d, PCI_ROM_ADDRESS, bar_data);
506         }
507 
508         s->bases[PCI_ROM_SLOT].access.maddr = d->rom.base_addr;
509 
510         memory_region_init_io(&s->rom, OBJECT(s), &ops, &s->dev,
511                               "xen-pci-pt-rom", d->rom.size);
512         pci_register_bar(&s->dev, PCI_ROM_SLOT, PCI_BASE_ADDRESS_MEM_PREFETCH,
513                          &s->rom);
514 
515         XEN_PT_LOG(&s->dev, "Expansion ROM registered (size=0x%08"PRIx64
516                    " base_addr=0x%08"PRIx64")\n",
517                    d->rom.size, d->rom.base_addr);
518     }
519 
520     xen_pt_register_vga_regions(d);
521     return 0;
522 }
523 
524 /* region mapping */
525 
526 static int xen_pt_bar_from_region(XenPCIPassthroughState *s, MemoryRegion *mr)
527 {
528     int i = 0;
529 
530     for (i = 0; i < PCI_NUM_REGIONS - 1; i++) {
531         if (mr == &s->bar[i]) {
532             return i;
533         }
534     }
535     if (mr == &s->rom) {
536         return PCI_ROM_SLOT;
537     }
538     return -1;
539 }
540 
541 /*
542  * This function checks if an io_region overlaps an io_region from another
543  * device.  The io_region to check is provided with (addr, size and type)
544  * A callback can be provided and will be called for every region that is
545  * overlapped.
546  * The return value indicates if the region is overlappsed */
547 struct CheckBarArgs {
548     XenPCIPassthroughState *s;
549     pcibus_t addr;
550     pcibus_t size;
551     uint8_t type;
552     bool rc;
553 };
554 static void xen_pt_check_bar_overlap(PCIBus *bus, PCIDevice *d, void *opaque)
555 {
556     struct CheckBarArgs *arg = opaque;
557     XenPCIPassthroughState *s = arg->s;
558     uint8_t type = arg->type;
559     int i;
560 
561     if (d->devfn == s->dev.devfn) {
562         return;
563     }
564 
565     /* xxx: This ignores bridges. */
566     for (i = 0; i < PCI_NUM_REGIONS; i++) {
567         const PCIIORegion *r = &d->io_regions[i];
568 
569         if (!r->size) {
570             continue;
571         }
572         if ((type & PCI_BASE_ADDRESS_SPACE_IO)
573             != (r->type & PCI_BASE_ADDRESS_SPACE_IO)) {
574             continue;
575         }
576 
577         if (ranges_overlap(arg->addr, arg->size, r->addr, r->size)) {
578             XEN_PT_WARN(&s->dev,
579                         "Overlapped to device [%02x:%02x.%d] Region: %i"
580                         " (addr: 0x%"FMT_PCIBUS", len: 0x%"FMT_PCIBUS")\n",
581                         pci_bus_num(bus), PCI_SLOT(d->devfn),
582                         PCI_FUNC(d->devfn), i, r->addr, r->size);
583             arg->rc = true;
584         }
585     }
586 }
587 
588 static void xen_pt_region_update(XenPCIPassthroughState *s,
589                                  MemoryRegionSection *sec, bool adding)
590 {
591     PCIDevice *d = &s->dev;
592     MemoryRegion *mr = sec->mr;
593     int bar = -1;
594     int rc;
595     int op = adding ? DPCI_ADD_MAPPING : DPCI_REMOVE_MAPPING;
596     struct CheckBarArgs args = {
597         .s = s,
598         .addr = sec->offset_within_address_space,
599         .size = int128_get64(sec->size),
600         .rc = false,
601     };
602 
603     bar = xen_pt_bar_from_region(s, mr);
604     if (bar == -1 && (!s->msix || &s->msix->mmio != mr)) {
605         return;
606     }
607 
608     if (s->msix && &s->msix->mmio == mr) {
609         if (adding) {
610             s->msix->mmio_base_addr = sec->offset_within_address_space;
611             rc = xen_pt_msix_update_remap(s, s->msix->bar_index);
612         }
613         return;
614     }
615 
616     args.type = d->io_regions[bar].type;
617     pci_for_each_device_under_bus(pci_get_bus(d),
618                                   xen_pt_check_bar_overlap, &args);
619     if (args.rc) {
620         XEN_PT_WARN(d, "Region: %d (addr: 0x%"FMT_PCIBUS
621                     ", len: 0x%"FMT_PCIBUS") is overlapped.\n",
622                     bar, sec->offset_within_address_space,
623                     int128_get64(sec->size));
624     }
625 
626     if (d->io_regions[bar].type & PCI_BASE_ADDRESS_SPACE_IO) {
627         uint32_t guest_port = sec->offset_within_address_space;
628         uint32_t machine_port = s->bases[bar].access.pio_base;
629         uint32_t size = int128_get64(sec->size);
630         rc = xc_domain_ioport_mapping(xen_xc, xen_domid,
631                                       guest_port, machine_port, size,
632                                       op);
633         if (rc) {
634             XEN_PT_ERR(d, "%s ioport mapping failed! (err: %i)\n",
635                        adding ? "create new" : "remove old", errno);
636         }
637     } else {
638         pcibus_t guest_addr = sec->offset_within_address_space;
639         pcibus_t machine_addr = s->bases[bar].access.maddr
640             + sec->offset_within_region;
641         pcibus_t size = int128_get64(sec->size);
642         rc = xc_domain_memory_mapping(xen_xc, xen_domid,
643                                       XEN_PFN(guest_addr + XC_PAGE_SIZE - 1),
644                                       XEN_PFN(machine_addr + XC_PAGE_SIZE - 1),
645                                       XEN_PFN(size + XC_PAGE_SIZE - 1),
646                                       op);
647         if (rc) {
648             XEN_PT_ERR(d, "%s mem mapping failed! (err: %i)\n",
649                        adding ? "create new" : "remove old", errno);
650         }
651     }
652 }
653 
654 static void xen_pt_region_add(MemoryListener *l, MemoryRegionSection *sec)
655 {
656     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
657                                              memory_listener);
658 
659     memory_region_ref(sec->mr);
660     xen_pt_region_update(s, sec, true);
661 }
662 
663 static void xen_pt_region_del(MemoryListener *l, MemoryRegionSection *sec)
664 {
665     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
666                                              memory_listener);
667 
668     xen_pt_region_update(s, sec, false);
669     memory_region_unref(sec->mr);
670 }
671 
672 static void xen_pt_io_region_add(MemoryListener *l, MemoryRegionSection *sec)
673 {
674     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
675                                              io_listener);
676 
677     memory_region_ref(sec->mr);
678     xen_pt_region_update(s, sec, true);
679 }
680 
681 static void xen_pt_io_region_del(MemoryListener *l, MemoryRegionSection *sec)
682 {
683     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
684                                              io_listener);
685 
686     xen_pt_region_update(s, sec, false);
687     memory_region_unref(sec->mr);
688 }
689 
690 static const MemoryListener xen_pt_memory_listener = {
691     .name = "xen-pt-mem",
692     .region_add = xen_pt_region_add,
693     .region_del = xen_pt_region_del,
694     .priority = 10,
695 };
696 
697 static const MemoryListener xen_pt_io_listener = {
698     .name = "xen-pt-io",
699     .region_add = xen_pt_io_region_add,
700     .region_del = xen_pt_io_region_del,
701     .priority = 10,
702 };
703 
704 /* destroy. */
705 static void xen_pt_destroy(PCIDevice *d) {
706 
707     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
708     XenHostPCIDevice *host_dev = &s->real_device;
709     uint8_t machine_irq = s->machine_irq;
710     uint8_t intx;
711     int rc;
712 
713     if (machine_irq && !xen_host_pci_device_closed(&s->real_device)) {
714         intx = xen_pt_pci_intx(s);
715         rc = xc_domain_unbind_pt_irq(xen_xc, xen_domid, machine_irq,
716                                      PT_IRQ_TYPE_PCI,
717                                      pci_dev_bus_num(d),
718                                      PCI_SLOT(s->dev.devfn),
719                                      intx,
720                                      0 /* isa_irq */);
721         if (rc < 0) {
722             XEN_PT_ERR(d, "unbinding of interrupt INT%c failed."
723                        " (machine irq: %i, err: %d)"
724                        " But bravely continuing on..\n",
725                        'a' + intx, machine_irq, errno);
726         }
727     }
728 
729     /* N.B. xen_pt_config_delete takes care of freeing them. */
730     if (s->msi) {
731         xen_pt_msi_disable(s);
732     }
733     if (s->msix) {
734         xen_pt_msix_disable(s);
735     }
736 
737     if (machine_irq) {
738         xen_pt_mapped_machine_irq[machine_irq]--;
739 
740         if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
741             rc = xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq);
742 
743             if (rc < 0) {
744                 XEN_PT_ERR(d, "unmapping of interrupt %i failed. (err: %d)"
745                            " But bravely continuing on..\n",
746                            machine_irq, errno);
747             }
748         }
749         s->machine_irq = 0;
750     }
751 
752     /* delete all emulated config registers */
753     xen_pt_config_delete(s);
754 
755     xen_pt_unregister_vga_regions(host_dev);
756 
757     if (s->listener_set) {
758         memory_listener_unregister(&s->memory_listener);
759         memory_listener_unregister(&s->io_listener);
760         s->listener_set = false;
761     }
762     if (!xen_host_pci_device_closed(&s->real_device)) {
763         xen_host_pci_device_put(&s->real_device);
764     }
765 }
766 /* init */
767 
768 static void xen_pt_realize(PCIDevice *d, Error **errp)
769 {
770     ERRP_GUARD();
771     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
772     int i, rc = 0;
773     uint8_t machine_irq = 0, scratch;
774     uint16_t cmd = 0;
775     int pirq = XEN_PT_UNASSIGNED_PIRQ;
776 
777     /* register real device */
778     XEN_PT_LOG(d, "Assigning real physical device %02x:%02x.%d"
779                " to devfn 0x%x\n",
780                s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function,
781                s->dev.devfn);
782 
783     xen_host_pci_device_get(&s->real_device,
784                             s->hostaddr.domain, s->hostaddr.bus,
785                             s->hostaddr.slot, s->hostaddr.function,
786                             errp);
787     if (*errp) {
788         error_append_hint(errp, "Failed to \"open\" the real pci device");
789         return;
790     }
791 
792     s->is_virtfn = s->real_device.is_virtfn;
793     if (s->is_virtfn) {
794         XEN_PT_LOG(d, "%04x:%02x:%02x.%d is a SR-IOV Virtual Function\n",
795                    s->real_device.domain, s->real_device.bus,
796                    s->real_device.dev, s->real_device.func);
797     }
798 
799     /* Initialize virtualized PCI configuration (Extended 256 Bytes) */
800     memset(d->config, 0, PCI_CONFIG_SPACE_SIZE);
801 
802     s->memory_listener = xen_pt_memory_listener;
803     s->io_listener = xen_pt_io_listener;
804 
805     /* Setup VGA bios for passthrough GFX */
806     if ((s->real_device.domain == 0) && (s->real_device.bus == 0) &&
807         (s->real_device.dev == 2) && (s->real_device.func == 0)) {
808         if (!is_igd_vga_passthrough(&s->real_device)) {
809             error_setg(errp, "Need to enable igd-passthru if you're trying"
810                     " to passthrough IGD GFX");
811             xen_host_pci_device_put(&s->real_device);
812             return;
813         }
814 
815         xen_pt_setup_vga(s, &s->real_device, errp);
816         if (*errp) {
817             error_append_hint(errp, "Setup VGA BIOS of passthrough"
818                               " GFX failed");
819             xen_host_pci_device_put(&s->real_device);
820             return;
821         }
822 
823         /* Register ISA bridge for passthrough GFX. */
824         xen_igd_passthrough_isa_bridge_create(s, &s->real_device);
825     }
826 
827     /* Handle real device's MMIO/PIO BARs */
828     xen_pt_register_regions(s, &cmd);
829 
830     /* reinitialize each config register to be emulated */
831     xen_pt_config_init(s, errp);
832     if (*errp) {
833         error_append_hint(errp, "PCI Config space initialisation failed");
834         rc = -1;
835         goto err_out;
836     }
837 
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;
843     }
844     if (!scratch) {
845         XEN_PT_LOG(d, "no pin interrupt\n");
846         goto out;
847     }
848 
849     machine_irq = s->real_device.irq;
850     if (machine_irq == 0) {
851         XEN_PT_LOG(d, "machine irq is 0\n");
852         cmd |= PCI_COMMAND_INTX_DISABLE;
853         goto out;
854     }
855 
856     rc = xc_physdev_map_pirq(xen_xc, xen_domid, machine_irq, &pirq);
857     if (rc < 0) {
858         XEN_PT_ERR(d, "Mapping machine irq %u to pirq %i failed, (err: %d)\n",
859                    machine_irq, pirq, errno);
860 
861         /* Disable PCI intx assertion (turn on bit10 of devctl) */
862         cmd |= PCI_COMMAND_INTX_DISABLE;
863         machine_irq = 0;
864         s->machine_irq = 0;
865     } else {
866         machine_irq = pirq;
867         s->machine_irq = pirq;
868         xen_pt_mapped_machine_irq[machine_irq]++;
869     }
870 
871     /* bind machine_irq to device */
872     if (machine_irq != 0) {
873         uint8_t e_intx = xen_pt_pci_intx(s);
874 
875         rc = xc_domain_bind_pt_pci_irq(xen_xc, xen_domid, machine_irq,
876                                        pci_dev_bus_num(d),
877                                        PCI_SLOT(d->devfn),
878                                        e_intx);
879         if (rc < 0) {
880             XEN_PT_ERR(d, "Binding of interrupt %i failed! (err: %d)\n",
881                        e_intx, errno);
882 
883             /* Disable PCI intx assertion (turn on bit10 of devctl) */
884             cmd |= PCI_COMMAND_INTX_DISABLE;
885             xen_pt_mapped_machine_irq[machine_irq]--;
886 
887             if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
888                 if (xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq)) {
889                     XEN_PT_ERR(d, "Unmapping of machine interrupt %i failed!"
890                                " (err: %d)\n", machine_irq, errno);
891                 }
892             }
893             s->machine_irq = 0;
894         }
895     }
896 
897 out:
898     if (cmd) {
899         uint16_t val;
900 
901         rc = xen_host_pci_get_word(&s->real_device, PCI_COMMAND, &val);
902         if (rc) {
903             error_setg_errno(errp, errno, "Failed to read PCI_COMMAND");
904             goto err_out;
905         } else {
906             val |= cmd;
907             rc = xen_host_pci_set_word(&s->real_device, PCI_COMMAND, val);
908             if (rc) {
909                 error_setg_errno(errp, errno, "Failed to write PCI_COMMAND"
910                                  " val = 0x%x", val);
911                 goto err_out;
912             }
913         }
914     }
915 
916     memory_listener_register(&s->memory_listener, &address_space_memory);
917     memory_listener_register(&s->io_listener, &address_space_io);
918     s->listener_set = true;
919     XEN_PT_LOG(d,
920                "Real physical device %02x:%02x.%d registered successfully\n",
921                s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function);
922 
923     return;
924 
925 err_out:
926     for (i = 0; i < PCI_ROM_SLOT; i++) {
927         object_unparent(OBJECT(&s->bar[i]));
928     }
929     object_unparent(OBJECT(&s->rom));
930 
931     xen_pt_destroy(d);
932     assert(rc);
933 }
934 
935 static void xen_pt_unregister_device(PCIDevice *d)
936 {
937     xen_pt_destroy(d);
938 }
939 
940 static Property xen_pci_passthrough_properties[] = {
941     DEFINE_PROP_PCI_HOST_DEVADDR("hostaddr", XenPCIPassthroughState, hostaddr),
942     DEFINE_PROP_BOOL("permissive", XenPCIPassthroughState, permissive, false),
943     DEFINE_PROP_END_OF_LIST(),
944 };
945 
946 static void xen_pci_passthrough_instance_init(Object *obj)
947 {
948     /* QEMU_PCI_CAP_EXPRESS initialization does not depend on QEMU command
949      * line, therefore, no need to wait to realize like other devices */
950     PCI_DEVICE(obj)->cap_present |= QEMU_PCI_CAP_EXPRESS;
951 }
952 
953 static void xen_pci_passthrough_class_init(ObjectClass *klass, void *data)
954 {
955     DeviceClass *dc = DEVICE_CLASS(klass);
956     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
957 
958     k->realize = xen_pt_realize;
959     k->exit = xen_pt_unregister_device;
960     k->config_read = xen_pt_pci_read_config;
961     k->config_write = xen_pt_pci_write_config;
962     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
963     dc->desc = "Assign an host PCI device with Xen";
964     device_class_set_props(dc, xen_pci_passthrough_properties);
965 };
966 
967 static void xen_pci_passthrough_finalize(Object *obj)
968 {
969     XenPCIPassthroughState *s = XEN_PT_DEVICE(obj);
970 
971     xen_pt_msix_delete(s);
972 }
973 
974 static const TypeInfo xen_pci_passthrough_info = {
975     .name = TYPE_XEN_PT_DEVICE,
976     .parent = TYPE_PCI_DEVICE,
977     .instance_size = sizeof(XenPCIPassthroughState),
978     .instance_finalize = xen_pci_passthrough_finalize,
979     .class_init = xen_pci_passthrough_class_init,
980     .instance_init = xen_pci_passthrough_instance_init,
981     .interfaces = (InterfaceInfo[]) {
982         { INTERFACE_CONVENTIONAL_PCI_DEVICE },
983         { INTERFACE_PCIE_DEVICE },
984         { },
985     },
986 };
987 
988 static void xen_pci_passthrough_register_types(void)
989 {
990     type_register_static(&xen_pci_passthrough_info);
991 }
992 
993 type_init(xen_pci_passthrough_register_types)
994