xref: /openbmc/qemu/hw/xen/xen_pt.c (revision 729cc683)
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/i386/pc.h"
64 #include "hw/xen/xen-legacy-backend.h"
65 #include "xen_pt.h"
66 #include "qemu/range.h"
67 #include "exec/address-spaces.h"
68 
69 static bool has_igd_gfx_passthru;
70 
71 bool xen_igd_gfx_pt_enabled(void)
72 {
73     return has_igd_gfx_passthru;
74 }
75 
76 void xen_igd_gfx_pt_set(bool value, Error **errp)
77 {
78     has_igd_gfx_passthru = value;
79 }
80 
81 #define XEN_PT_NR_IRQS (256)
82 static uint8_t xen_pt_mapped_machine_irq[XEN_PT_NR_IRQS] = {0};
83 
84 void xen_pt_log(const PCIDevice *d, const char *f, ...)
85 {
86     va_list ap;
87 
88     va_start(ap, f);
89     if (d) {
90         fprintf(stderr, "[%02x:%02x.%d] ", pci_dev_bus_num(d),
91                 PCI_SLOT(d->devfn), PCI_FUNC(d->devfn));
92     }
93     vfprintf(stderr, f, ap);
94     va_end(ap);
95 }
96 
97 /* Config Space */
98 
99 static int xen_pt_pci_config_access_check(PCIDevice *d, uint32_t addr, int len)
100 {
101     /* check offset range */
102     if (addr > 0xFF) {
103         XEN_PT_ERR(d, "Failed to access register with offset exceeding 0xFF. "
104                    "(addr: 0x%02x, len: %d)\n", addr, len);
105         return -1;
106     }
107 
108     /* check read size */
109     if ((len != 1) && (len != 2) && (len != 4)) {
110         XEN_PT_ERR(d, "Failed to access register with invalid access length. "
111                    "(addr: 0x%02x, len: %d)\n", addr, len);
112         return -1;
113     }
114 
115     /* check offset alignment */
116     if (addr & (len - 1)) {
117         XEN_PT_ERR(d, "Failed to access register with invalid access size "
118                    "alignment. (addr: 0x%02x, len: %d)\n", addr, len);
119         return -1;
120     }
121 
122     return 0;
123 }
124 
125 int xen_pt_bar_offset_to_index(uint32_t offset)
126 {
127     int index = 0;
128 
129     /* check Exp ROM BAR */
130     if (offset == PCI_ROM_ADDRESS) {
131         return PCI_ROM_SLOT;
132     }
133 
134     /* calculate BAR index */
135     index = (offset - PCI_BASE_ADDRESS_0) >> 2;
136     if (index >= PCI_NUM_REGIONS) {
137         return -1;
138     }
139 
140     return index;
141 }
142 
143 static uint32_t xen_pt_pci_read_config(PCIDevice *d, uint32_t addr, int len)
144 {
145     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
146     uint32_t val = 0;
147     XenPTRegGroup *reg_grp_entry = NULL;
148     XenPTReg *reg_entry = NULL;
149     int rc = 0;
150     int emul_len = 0;
151     uint32_t find_addr = addr;
152 
153     if (xen_pt_pci_config_access_check(d, addr, len)) {
154         goto exit;
155     }
156 
157     /* find register group entry */
158     reg_grp_entry = xen_pt_find_reg_grp(s, addr);
159     if (reg_grp_entry) {
160         /* check 0-Hardwired register group */
161         if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
162             /* no need to emulate, just return 0 */
163             val = 0;
164             goto exit;
165         }
166     }
167 
168     /* read I/O device register value */
169     rc = xen_host_pci_get_block(&s->real_device, addr, (uint8_t *)&val, len);
170     if (rc < 0) {
171         XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
172         memset(&val, 0xff, len);
173     }
174 
175     /* just return the I/O device register value for
176      * passthrough type register group */
177     if (reg_grp_entry == NULL) {
178         goto exit;
179     }
180 
181     /* adjust the read value to appropriate CFC-CFF window */
182     val <<= (addr & 3) << 3;
183     emul_len = len;
184 
185     /* loop around the guest requested size */
186     while (emul_len > 0) {
187         /* find register entry to be emulated */
188         reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
189         if (reg_entry) {
190             XenPTRegInfo *reg = reg_entry->reg;
191             uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
192             uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
193             uint8_t *ptr_val = NULL;
194 
195             valid_mask <<= (find_addr - real_offset) << 3;
196             ptr_val = (uint8_t *)&val + (real_offset & 3);
197 
198             /* do emulation based on register size */
199             switch (reg->size) {
200             case 1:
201                 if (reg->u.b.read) {
202                     rc = reg->u.b.read(s, reg_entry, ptr_val, valid_mask);
203                 }
204                 break;
205             case 2:
206                 if (reg->u.w.read) {
207                     rc = reg->u.w.read(s, reg_entry,
208                                        (uint16_t *)ptr_val, valid_mask);
209                 }
210                 break;
211             case 4:
212                 if (reg->u.dw.read) {
213                     rc = reg->u.dw.read(s, reg_entry,
214                                         (uint32_t *)ptr_val, valid_mask);
215                 }
216                 break;
217             }
218 
219             if (rc < 0) {
220                 xen_shutdown_fatal_error("Internal error: Invalid read "
221                                          "emulation. (%s, rc: %d)\n",
222                                          __func__, rc);
223                 return 0;
224             }
225 
226             /* calculate next address to find */
227             emul_len -= reg->size;
228             if (emul_len > 0) {
229                 find_addr = real_offset + reg->size;
230             }
231         } else {
232             /* nothing to do with passthrough type register,
233              * continue to find next byte */
234             emul_len--;
235             find_addr++;
236         }
237     }
238 
239     /* need to shift back before returning them to pci bus emulator */
240     val >>= ((addr & 3) << 3);
241 
242 exit:
243     XEN_PT_LOG_CONFIG(d, addr, val, len);
244     return val;
245 }
246 
247 static void xen_pt_pci_write_config(PCIDevice *d, uint32_t addr,
248                                     uint32_t val, int len)
249 {
250     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
251     int index = 0;
252     XenPTRegGroup *reg_grp_entry = NULL;
253     int rc = 0;
254     uint32_t read_val = 0, wb_mask;
255     int emul_len = 0;
256     XenPTReg *reg_entry = NULL;
257     uint32_t find_addr = addr;
258     XenPTRegInfo *reg = NULL;
259     bool wp_flag = false;
260 
261     if (xen_pt_pci_config_access_check(d, addr, len)) {
262         return;
263     }
264 
265     XEN_PT_LOG_CONFIG(d, addr, val, len);
266 
267     /* check unused BAR register */
268     index = xen_pt_bar_offset_to_index(addr);
269     if ((index >= 0) && (val != 0)) {
270         uint32_t chk = val;
271 
272         if (index == PCI_ROM_SLOT)
273             chk |= (uint32_t)~PCI_ROM_ADDRESS_MASK;
274 
275         if ((chk != XEN_PT_BAR_ALLF) &&
276             (s->bases[index].bar_flag == XEN_PT_BAR_FLAG_UNUSED)) {
277             XEN_PT_WARN(d, "Guest attempt to set address to unused "
278                         "Base Address Register. (addr: 0x%02x, len: %d)\n",
279                         addr, len);
280         }
281     }
282 
283     /* find register group entry */
284     reg_grp_entry = xen_pt_find_reg_grp(s, addr);
285     if (reg_grp_entry) {
286         /* check 0-Hardwired register group */
287         if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
288             /* ignore silently */
289             XEN_PT_WARN(d, "Access to 0-Hardwired register. "
290                         "(addr: 0x%02x, len: %d)\n", addr, len);
291             return;
292         }
293     }
294 
295     rc = xen_host_pci_get_block(&s->real_device, addr,
296                                 (uint8_t *)&read_val, len);
297     if (rc < 0) {
298         XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
299         memset(&read_val, 0xff, len);
300         wb_mask = 0;
301     } else {
302         wb_mask = 0xFFFFFFFF >> ((4 - len) << 3);
303     }
304 
305     /* pass directly to the real device for passthrough type register group */
306     if (reg_grp_entry == NULL) {
307         if (!s->permissive) {
308             wb_mask = 0;
309             wp_flag = true;
310         }
311         goto out;
312     }
313 
314     memory_region_transaction_begin();
315     pci_default_write_config(d, addr, val, len);
316 
317     /* adjust the read and write value to appropriate CFC-CFF window */
318     read_val <<= (addr & 3) << 3;
319     val <<= (addr & 3) << 3;
320     emul_len = len;
321 
322     /* loop around the guest requested size */
323     while (emul_len > 0) {
324         /* find register entry to be emulated */
325         reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
326         if (reg_entry) {
327             reg = reg_entry->reg;
328             uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
329             uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
330             uint8_t *ptr_val = NULL;
331             uint32_t wp_mask = reg->emu_mask | reg->ro_mask;
332 
333             valid_mask <<= (find_addr - real_offset) << 3;
334             ptr_val = (uint8_t *)&val + (real_offset & 3);
335             if (!s->permissive) {
336                 wp_mask |= reg->res_mask;
337             }
338             if (wp_mask == (0xFFFFFFFF >> ((4 - reg->size) << 3))) {
339                 wb_mask &= ~((wp_mask >> ((find_addr - real_offset) << 3))
340                              << ((len - emul_len) << 3));
341             }
342 
343             /* do emulation based on register size */
344             switch (reg->size) {
345             case 1:
346                 if (reg->u.b.write) {
347                     rc = reg->u.b.write(s, reg_entry, ptr_val,
348                                         read_val >> ((real_offset & 3) << 3),
349                                         valid_mask);
350                 }
351                 break;
352             case 2:
353                 if (reg->u.w.write) {
354                     rc = reg->u.w.write(s, reg_entry, (uint16_t *)ptr_val,
355                                         (read_val >> ((real_offset & 3) << 3)),
356                                         valid_mask);
357                 }
358                 break;
359             case 4:
360                 if (reg->u.dw.write) {
361                     rc = reg->u.dw.write(s, reg_entry, (uint32_t *)ptr_val,
362                                          (read_val >> ((real_offset & 3) << 3)),
363                                          valid_mask);
364                 }
365                 break;
366             }
367 
368             if (rc < 0) {
369                 xen_shutdown_fatal_error("Internal error: Invalid write"
370                                          " emulation. (%s, rc: %d)\n",
371                                          __func__, rc);
372                 return;
373             }
374 
375             /* calculate next address to find */
376             emul_len -= reg->size;
377             if (emul_len > 0) {
378                 find_addr = real_offset + reg->size;
379             }
380         } else {
381             /* nothing to do with passthrough type register,
382              * continue to find next byte */
383             if (!s->permissive) {
384                 wb_mask &= ~(0xff << ((len - emul_len) << 3));
385                 /* Unused BARs will make it here, but we don't want to issue
386                  * warnings for writes to them (bogus writes get dealt with
387                  * above).
388                  */
389                 if (index < 0) {
390                     wp_flag = true;
391                 }
392             }
393             emul_len--;
394             find_addr++;
395         }
396     }
397 
398     /* need to shift back before passing them to xen_host_pci_set_block. */
399     val >>= (addr & 3) << 3;
400 
401     memory_region_transaction_commit();
402 
403 out:
404     if (wp_flag && !s->permissive_warned) {
405         s->permissive_warned = true;
406         xen_pt_log(d, "Write-back to unknown field 0x%02x (partially) inhibited (0x%0*x)\n",
407                    addr, len * 2, wb_mask);
408         xen_pt_log(d, "If the device doesn't work, try enabling permissive mode\n");
409         xen_pt_log(d, "(unsafe) and if it helps report the problem to xen-devel\n");
410     }
411     for (index = 0; wb_mask; index += len) {
412         /* unknown regs are passed through */
413         while (!(wb_mask & 0xff)) {
414             index++;
415             wb_mask >>= 8;
416         }
417         len = 0;
418         do {
419             len++;
420             wb_mask >>= 8;
421         } while (wb_mask & 0xff);
422         rc = xen_host_pci_set_block(&s->real_device, addr + index,
423                                     (uint8_t *)&val + index, len);
424 
425         if (rc < 0) {
426             XEN_PT_ERR(d, "xen_host_pci_set_block failed. return value: %d.\n", rc);
427         }
428     }
429 }
430 
431 /* register regions */
432 
433 static uint64_t xen_pt_bar_read(void *o, hwaddr addr,
434                                 unsigned size)
435 {
436     PCIDevice *d = o;
437     /* if this function is called, that probably means that there is a
438      * misconfiguration of the IOMMU. */
439     XEN_PT_ERR(d, "Should not read BAR through QEMU. @0x"TARGET_FMT_plx"\n",
440                addr);
441     return 0;
442 }
443 static void xen_pt_bar_write(void *o, hwaddr addr, uint64_t val,
444                              unsigned size)
445 {
446     PCIDevice *d = o;
447     /* Same comment as xen_pt_bar_read function */
448     XEN_PT_ERR(d, "Should not write BAR through QEMU. @0x"TARGET_FMT_plx"\n",
449                addr);
450 }
451 
452 static const MemoryRegionOps ops = {
453     .endianness = DEVICE_NATIVE_ENDIAN,
454     .read = xen_pt_bar_read,
455     .write = xen_pt_bar_write,
456 };
457 
458 static int xen_pt_register_regions(XenPCIPassthroughState *s, uint16_t *cmd)
459 {
460     int i = 0;
461     XenHostPCIDevice *d = &s->real_device;
462 
463     /* Register PIO/MMIO BARs */
464     for (i = 0; i < PCI_ROM_SLOT; i++) {
465         XenHostPCIIORegion *r = &d->io_regions[i];
466         uint8_t type;
467 
468         if (r->base_addr == 0 || r->size == 0) {
469             continue;
470         }
471 
472         s->bases[i].access.u = r->base_addr;
473 
474         if (r->type & XEN_HOST_PCI_REGION_TYPE_IO) {
475             type = PCI_BASE_ADDRESS_SPACE_IO;
476             *cmd |= PCI_COMMAND_IO;
477         } else {
478             type = PCI_BASE_ADDRESS_SPACE_MEMORY;
479             if (r->type & XEN_HOST_PCI_REGION_TYPE_PREFETCH) {
480                 type |= PCI_BASE_ADDRESS_MEM_PREFETCH;
481             }
482             if (r->type & XEN_HOST_PCI_REGION_TYPE_MEM_64) {
483                 type |= PCI_BASE_ADDRESS_MEM_TYPE_64;
484             }
485             *cmd |= PCI_COMMAND_MEMORY;
486         }
487 
488         memory_region_init_io(&s->bar[i], OBJECT(s), &ops, &s->dev,
489                               "xen-pci-pt-bar", r->size);
490         pci_register_bar(&s->dev, i, type, &s->bar[i]);
491 
492         XEN_PT_LOG(&s->dev, "IO region %i registered (size=0x%08"PRIx64
493                    " base_addr=0x%08"PRIx64" type: 0x%x)\n",
494                    i, r->size, r->base_addr, type);
495     }
496 
497     /* Register expansion ROM address */
498     if (d->rom.base_addr && d->rom.size) {
499         uint32_t bar_data = 0;
500 
501         /* Re-set BAR reported by OS, otherwise ROM can't be read. */
502         if (xen_host_pci_get_long(d, PCI_ROM_ADDRESS, &bar_data)) {
503             return 0;
504         }
505         if ((bar_data & PCI_ROM_ADDRESS_MASK) == 0) {
506             bar_data |= d->rom.base_addr & PCI_ROM_ADDRESS_MASK;
507             xen_host_pci_set_long(d, PCI_ROM_ADDRESS, bar_data);
508         }
509 
510         s->bases[PCI_ROM_SLOT].access.maddr = d->rom.base_addr;
511 
512         memory_region_init_io(&s->rom, OBJECT(s), &ops, &s->dev,
513                               "xen-pci-pt-rom", d->rom.size);
514         pci_register_bar(&s->dev, PCI_ROM_SLOT, PCI_BASE_ADDRESS_MEM_PREFETCH,
515                          &s->rom);
516 
517         XEN_PT_LOG(&s->dev, "Expansion ROM registered (size=0x%08"PRIx64
518                    " base_addr=0x%08"PRIx64")\n",
519                    d->rom.size, d->rom.base_addr);
520     }
521 
522     xen_pt_register_vga_regions(d);
523     return 0;
524 }
525 
526 /* region mapping */
527 
528 static int xen_pt_bar_from_region(XenPCIPassthroughState *s, MemoryRegion *mr)
529 {
530     int i = 0;
531 
532     for (i = 0; i < PCI_NUM_REGIONS - 1; i++) {
533         if (mr == &s->bar[i]) {
534             return i;
535         }
536     }
537     if (mr == &s->rom) {
538         return PCI_ROM_SLOT;
539     }
540     return -1;
541 }
542 
543 /*
544  * This function checks if an io_region overlaps an io_region from another
545  * device.  The io_region to check is provided with (addr, size and type)
546  * A callback can be provided and will be called for every region that is
547  * overlapped.
548  * The return value indicates if the region is overlappsed */
549 struct CheckBarArgs {
550     XenPCIPassthroughState *s;
551     pcibus_t addr;
552     pcibus_t size;
553     uint8_t type;
554     bool rc;
555 };
556 static void xen_pt_check_bar_overlap(PCIBus *bus, PCIDevice *d, void *opaque)
557 {
558     struct CheckBarArgs *arg = opaque;
559     XenPCIPassthroughState *s = arg->s;
560     uint8_t type = arg->type;
561     int i;
562 
563     if (d->devfn == s->dev.devfn) {
564         return;
565     }
566 
567     /* xxx: This ignores bridges. */
568     for (i = 0; i < PCI_NUM_REGIONS; i++) {
569         const PCIIORegion *r = &d->io_regions[i];
570 
571         if (!r->size) {
572             continue;
573         }
574         if ((type & PCI_BASE_ADDRESS_SPACE_IO)
575             != (r->type & PCI_BASE_ADDRESS_SPACE_IO)) {
576             continue;
577         }
578 
579         if (ranges_overlap(arg->addr, arg->size, r->addr, r->size)) {
580             XEN_PT_WARN(&s->dev,
581                         "Overlapped to device [%02x:%02x.%d] Region: %i"
582                         " (addr: 0x%"FMT_PCIBUS", len: 0x%"FMT_PCIBUS")\n",
583                         pci_bus_num(bus), PCI_SLOT(d->devfn),
584                         PCI_FUNC(d->devfn), i, r->addr, r->size);
585             arg->rc = true;
586         }
587     }
588 }
589 
590 static void xen_pt_region_update(XenPCIPassthroughState *s,
591                                  MemoryRegionSection *sec, bool adding)
592 {
593     PCIDevice *d = &s->dev;
594     MemoryRegion *mr = sec->mr;
595     int bar = -1;
596     int rc;
597     int op = adding ? DPCI_ADD_MAPPING : DPCI_REMOVE_MAPPING;
598     struct CheckBarArgs args = {
599         .s = s,
600         .addr = sec->offset_within_address_space,
601         .size = int128_get64(sec->size),
602         .rc = false,
603     };
604 
605     bar = xen_pt_bar_from_region(s, mr);
606     if (bar == -1 && (!s->msix || &s->msix->mmio != mr)) {
607         return;
608     }
609 
610     if (s->msix && &s->msix->mmio == mr) {
611         if (adding) {
612             s->msix->mmio_base_addr = sec->offset_within_address_space;
613             rc = xen_pt_msix_update_remap(s, s->msix->bar_index);
614         }
615         return;
616     }
617 
618     args.type = d->io_regions[bar].type;
619     pci_for_each_device(pci_get_bus(d), pci_dev_bus_num(d),
620                         xen_pt_check_bar_overlap, &args);
621     if (args.rc) {
622         XEN_PT_WARN(d, "Region: %d (addr: 0x%"FMT_PCIBUS
623                     ", len: 0x%"FMT_PCIBUS") is overlapped.\n",
624                     bar, sec->offset_within_address_space,
625                     int128_get64(sec->size));
626     }
627 
628     if (d->io_regions[bar].type & PCI_BASE_ADDRESS_SPACE_IO) {
629         uint32_t guest_port = sec->offset_within_address_space;
630         uint32_t machine_port = s->bases[bar].access.pio_base;
631         uint32_t size = int128_get64(sec->size);
632         rc = xc_domain_ioport_mapping(xen_xc, xen_domid,
633                                       guest_port, machine_port, size,
634                                       op);
635         if (rc) {
636             XEN_PT_ERR(d, "%s ioport mapping failed! (err: %i)\n",
637                        adding ? "create new" : "remove old", errno);
638         }
639     } else {
640         pcibus_t guest_addr = sec->offset_within_address_space;
641         pcibus_t machine_addr = s->bases[bar].access.maddr
642             + sec->offset_within_region;
643         pcibus_t size = int128_get64(sec->size);
644         rc = xc_domain_memory_mapping(xen_xc, xen_domid,
645                                       XEN_PFN(guest_addr + XC_PAGE_SIZE - 1),
646                                       XEN_PFN(machine_addr + XC_PAGE_SIZE - 1),
647                                       XEN_PFN(size + XC_PAGE_SIZE - 1),
648                                       op);
649         if (rc) {
650             XEN_PT_ERR(d, "%s mem mapping failed! (err: %i)\n",
651                        adding ? "create new" : "remove old", errno);
652         }
653     }
654 }
655 
656 static void xen_pt_region_add(MemoryListener *l, MemoryRegionSection *sec)
657 {
658     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
659                                              memory_listener);
660 
661     memory_region_ref(sec->mr);
662     xen_pt_region_update(s, sec, true);
663 }
664 
665 static void xen_pt_region_del(MemoryListener *l, MemoryRegionSection *sec)
666 {
667     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
668                                              memory_listener);
669 
670     xen_pt_region_update(s, sec, false);
671     memory_region_unref(sec->mr);
672 }
673 
674 static void xen_pt_io_region_add(MemoryListener *l, MemoryRegionSection *sec)
675 {
676     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
677                                              io_listener);
678 
679     memory_region_ref(sec->mr);
680     xen_pt_region_update(s, sec, true);
681 }
682 
683 static void xen_pt_io_region_del(MemoryListener *l, MemoryRegionSection *sec)
684 {
685     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
686                                              io_listener);
687 
688     xen_pt_region_update(s, sec, false);
689     memory_region_unref(sec->mr);
690 }
691 
692 static const MemoryListener xen_pt_memory_listener = {
693     .region_add = xen_pt_region_add,
694     .region_del = xen_pt_region_del,
695     .priority = 10,
696 };
697 
698 static const MemoryListener xen_pt_io_listener = {
699     .region_add = xen_pt_io_region_add,
700     .region_del = xen_pt_io_region_del,
701     .priority = 10,
702 };
703 
704 static void
705 xen_igd_passthrough_isa_bridge_create(XenPCIPassthroughState *s,
706                                       XenHostPCIDevice *dev)
707 {
708     uint16_t gpu_dev_id;
709     PCIDevice *d = &s->dev;
710 
711     gpu_dev_id = dev->device_id;
712     igd_passthrough_isa_bridge_create(pci_get_bus(d), gpu_dev_id);
713 }
714 
715 /* destroy. */
716 static void xen_pt_destroy(PCIDevice *d) {
717 
718     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
719     XenHostPCIDevice *host_dev = &s->real_device;
720     uint8_t machine_irq = s->machine_irq;
721     uint8_t intx;
722     int rc;
723 
724     if (machine_irq && !xen_host_pci_device_closed(&s->real_device)) {
725         intx = xen_pt_pci_intx(s);
726         rc = xc_domain_unbind_pt_irq(xen_xc, xen_domid, machine_irq,
727                                      PT_IRQ_TYPE_PCI,
728                                      pci_dev_bus_num(d),
729                                      PCI_SLOT(s->dev.devfn),
730                                      intx,
731                                      0 /* isa_irq */);
732         if (rc < 0) {
733             XEN_PT_ERR(d, "unbinding of interrupt INT%c failed."
734                        " (machine irq: %i, err: %d)"
735                        " But bravely continuing on..\n",
736                        'a' + intx, machine_irq, errno);
737         }
738     }
739 
740     /* N.B. xen_pt_config_delete takes care of freeing them. */
741     if (s->msi) {
742         xen_pt_msi_disable(s);
743     }
744     if (s->msix) {
745         xen_pt_msix_disable(s);
746     }
747 
748     if (machine_irq) {
749         xen_pt_mapped_machine_irq[machine_irq]--;
750 
751         if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
752             rc = xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq);
753 
754             if (rc < 0) {
755                 XEN_PT_ERR(d, "unmapping of interrupt %i failed. (err: %d)"
756                            " But bravely continuing on..\n",
757                            machine_irq, errno);
758             }
759         }
760         s->machine_irq = 0;
761     }
762 
763     /* delete all emulated config registers */
764     xen_pt_config_delete(s);
765 
766     xen_pt_unregister_vga_regions(host_dev);
767 
768     if (s->listener_set) {
769         memory_listener_unregister(&s->memory_listener);
770         memory_listener_unregister(&s->io_listener);
771         s->listener_set = false;
772     }
773     if (!xen_host_pci_device_closed(&s->real_device)) {
774         xen_host_pci_device_put(&s->real_device);
775     }
776 }
777 /* init */
778 
779 static void xen_pt_realize(PCIDevice *d, Error **errp)
780 {
781     ERRP_GUARD();
782     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
783     int i, rc = 0;
784     uint8_t machine_irq = 0, scratch;
785     uint16_t cmd = 0;
786     int pirq = XEN_PT_UNASSIGNED_PIRQ;
787 
788     /* register real device */
789     XEN_PT_LOG(d, "Assigning real physical device %02x:%02x.%d"
790                " to devfn 0x%x\n",
791                s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function,
792                s->dev.devfn);
793 
794     xen_host_pci_device_get(&s->real_device,
795                             s->hostaddr.domain, s->hostaddr.bus,
796                             s->hostaddr.slot, s->hostaddr.function,
797                             errp);
798     if (*errp) {
799         error_append_hint(errp, "Failed to \"open\" the real pci device");
800         return;
801     }
802 
803     s->is_virtfn = s->real_device.is_virtfn;
804     if (s->is_virtfn) {
805         XEN_PT_LOG(d, "%04x:%02x:%02x.%d is a SR-IOV Virtual Function\n",
806                    s->real_device.domain, s->real_device.bus,
807                    s->real_device.dev, s->real_device.func);
808     }
809 
810     /* Initialize virtualized PCI configuration (Extended 256 Bytes) */
811     memset(d->config, 0, PCI_CONFIG_SPACE_SIZE);
812 
813     s->memory_listener = xen_pt_memory_listener;
814     s->io_listener = xen_pt_io_listener;
815 
816     /* Setup VGA bios for passthrough GFX */
817     if ((s->real_device.domain == 0) && (s->real_device.bus == 0) &&
818         (s->real_device.dev == 2) && (s->real_device.func == 0)) {
819         if (!is_igd_vga_passthrough(&s->real_device)) {
820             error_setg(errp, "Need to enable igd-passthru if you're trying"
821                     " to passthrough IGD GFX");
822             xen_host_pci_device_put(&s->real_device);
823             return;
824         }
825 
826         xen_pt_setup_vga(s, &s->real_device, errp);
827         if (*errp) {
828             error_append_hint(errp, "Setup VGA BIOS of passthrough"
829                               " GFX failed");
830             xen_host_pci_device_put(&s->real_device);
831             return;
832         }
833 
834         /* Register ISA bridge for passthrough GFX. */
835         xen_igd_passthrough_isa_bridge_create(s, &s->real_device);
836     }
837 
838     /* Handle real device's MMIO/PIO BARs */
839     xen_pt_register_regions(s, &cmd);
840 
841     /* reinitialize each config register to be emulated */
842     xen_pt_config_init(s, errp);
843     if (*errp) {
844         error_append_hint(errp, "PCI Config space initialisation failed");
845         rc = -1;
846         goto err_out;
847     }
848 
849     /* Bind interrupt */
850     rc = xen_host_pci_get_byte(&s->real_device, PCI_INTERRUPT_PIN, &scratch);
851     if (rc) {
852         error_setg_errno(errp, errno, "Failed to read PCI_INTERRUPT_PIN");
853         goto err_out;
854     }
855     if (!scratch) {
856         XEN_PT_LOG(d, "no pin interrupt\n");
857         goto out;
858     }
859 
860     machine_irq = s->real_device.irq;
861     if (machine_irq == 0) {
862         XEN_PT_LOG(d, "machine irq is 0\n");
863         cmd |= PCI_COMMAND_INTX_DISABLE;
864         goto out;
865     }
866 
867     rc = xc_physdev_map_pirq(xen_xc, xen_domid, machine_irq, &pirq);
868     if (rc < 0) {
869         XEN_PT_ERR(d, "Mapping machine irq %u to pirq %i failed, (err: %d)\n",
870                    machine_irq, pirq, errno);
871 
872         /* Disable PCI intx assertion (turn on bit10 of devctl) */
873         cmd |= PCI_COMMAND_INTX_DISABLE;
874         machine_irq = 0;
875         s->machine_irq = 0;
876     } else {
877         machine_irq = pirq;
878         s->machine_irq = pirq;
879         xen_pt_mapped_machine_irq[machine_irq]++;
880     }
881 
882     /* bind machine_irq to device */
883     if (machine_irq != 0) {
884         uint8_t e_intx = xen_pt_pci_intx(s);
885 
886         rc = xc_domain_bind_pt_pci_irq(xen_xc, xen_domid, machine_irq,
887                                        pci_dev_bus_num(d),
888                                        PCI_SLOT(d->devfn),
889                                        e_intx);
890         if (rc < 0) {
891             XEN_PT_ERR(d, "Binding of interrupt %i failed! (err: %d)\n",
892                        e_intx, errno);
893 
894             /* Disable PCI intx assertion (turn on bit10 of devctl) */
895             cmd |= PCI_COMMAND_INTX_DISABLE;
896             xen_pt_mapped_machine_irq[machine_irq]--;
897 
898             if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
899                 if (xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq)) {
900                     XEN_PT_ERR(d, "Unmapping of machine interrupt %i failed!"
901                                " (err: %d)\n", machine_irq, errno);
902                 }
903             }
904             s->machine_irq = 0;
905         }
906     }
907 
908 out:
909     if (cmd) {
910         uint16_t val;
911 
912         rc = xen_host_pci_get_word(&s->real_device, PCI_COMMAND, &val);
913         if (rc) {
914             error_setg_errno(errp, errno, "Failed to read PCI_COMMAND");
915             goto err_out;
916         } else {
917             val |= cmd;
918             rc = xen_host_pci_set_word(&s->real_device, PCI_COMMAND, val);
919             if (rc) {
920                 error_setg_errno(errp, errno, "Failed to write PCI_COMMAND"
921                                  " val = 0x%x", val);
922                 goto err_out;
923             }
924         }
925     }
926 
927     memory_listener_register(&s->memory_listener, &address_space_memory);
928     memory_listener_register(&s->io_listener, &address_space_io);
929     s->listener_set = true;
930     XEN_PT_LOG(d,
931                "Real physical device %02x:%02x.%d registered successfully\n",
932                s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function);
933 
934     return;
935 
936 err_out:
937     for (i = 0; i < PCI_ROM_SLOT; i++) {
938         object_unparent(OBJECT(&s->bar[i]));
939     }
940     object_unparent(OBJECT(&s->rom));
941 
942     xen_pt_destroy(d);
943     assert(rc);
944 }
945 
946 static void xen_pt_unregister_device(PCIDevice *d)
947 {
948     xen_pt_destroy(d);
949 }
950 
951 static Property xen_pci_passthrough_properties[] = {
952     DEFINE_PROP_PCI_HOST_DEVADDR("hostaddr", XenPCIPassthroughState, hostaddr),
953     DEFINE_PROP_BOOL("permissive", XenPCIPassthroughState, permissive, false),
954     DEFINE_PROP_END_OF_LIST(),
955 };
956 
957 static void xen_pci_passthrough_instance_init(Object *obj)
958 {
959     /* QEMU_PCI_CAP_EXPRESS initialization does not depend on QEMU command
960      * line, therefore, no need to wait to realize like other devices */
961     PCI_DEVICE(obj)->cap_present |= QEMU_PCI_CAP_EXPRESS;
962 }
963 
964 static void xen_pci_passthrough_class_init(ObjectClass *klass, void *data)
965 {
966     DeviceClass *dc = DEVICE_CLASS(klass);
967     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
968 
969     k->realize = xen_pt_realize;
970     k->exit = xen_pt_unregister_device;
971     k->config_read = xen_pt_pci_read_config;
972     k->config_write = xen_pt_pci_write_config;
973     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
974     dc->desc = "Assign an host PCI device with Xen";
975     device_class_set_props(dc, xen_pci_passthrough_properties);
976 };
977 
978 static void xen_pci_passthrough_finalize(Object *obj)
979 {
980     XenPCIPassthroughState *s = XEN_PT_DEVICE(obj);
981 
982     xen_pt_msix_delete(s);
983 }
984 
985 static const TypeInfo xen_pci_passthrough_info = {
986     .name = TYPE_XEN_PT_DEVICE,
987     .parent = TYPE_PCI_DEVICE,
988     .instance_size = sizeof(XenPCIPassthroughState),
989     .instance_finalize = xen_pci_passthrough_finalize,
990     .class_init = xen_pci_passthrough_class_init,
991     .instance_init = xen_pci_passthrough_instance_init,
992     .interfaces = (InterfaceInfo[]) {
993         { INTERFACE_CONVENTIONAL_PCI_DEVICE },
994         { INTERFACE_PCIE_DEVICE },
995         { },
996     },
997 };
998 
999 static void xen_pci_passthrough_register_types(void)
1000 {
1001     type_register_static(&xen_pci_passthrough_info);
1002 }
1003 
1004 type_init(xen_pci_passthrough_register_types)
1005