xref: /openbmc/qemu/hw/xen/xen_pt.c (revision 38472890)
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/xen/xen.h"
62 #include "hw/i386/pc.h"
63 #include "hw/xen/xen-legacy-backend.h"
64 #include "xen_pt.h"
65 #include "qemu/range.h"
66 #include "exec/address-spaces.h"
67 
68 bool has_igd_gfx_passthru;
69 
70 #define XEN_PT_NR_IRQS (256)
71 static uint8_t xen_pt_mapped_machine_irq[XEN_PT_NR_IRQS] = {0};
72 
73 void xen_pt_log(const PCIDevice *d, const char *f, ...)
74 {
75     va_list ap;
76 
77     va_start(ap, f);
78     if (d) {
79         fprintf(stderr, "[%02x:%02x.%d] ", pci_dev_bus_num(d),
80                 PCI_SLOT(d->devfn), PCI_FUNC(d->devfn));
81     }
82     vfprintf(stderr, f, ap);
83     va_end(ap);
84 }
85 
86 /* Config Space */
87 
88 static int xen_pt_pci_config_access_check(PCIDevice *d, uint32_t addr, int len)
89 {
90     /* check offset range */
91     if (addr > 0xFF) {
92         XEN_PT_ERR(d, "Failed to access register with offset exceeding 0xFF. "
93                    "(addr: 0x%02x, len: %d)\n", addr, len);
94         return -1;
95     }
96 
97     /* check read size */
98     if ((len != 1) && (len != 2) && (len != 4)) {
99         XEN_PT_ERR(d, "Failed to access register with invalid access length. "
100                    "(addr: 0x%02x, len: %d)\n", addr, len);
101         return -1;
102     }
103 
104     /* check offset alignment */
105     if (addr & (len - 1)) {
106         XEN_PT_ERR(d, "Failed to access register with invalid access size "
107                    "alignment. (addr: 0x%02x, len: %d)\n", addr, len);
108         return -1;
109     }
110 
111     return 0;
112 }
113 
114 int xen_pt_bar_offset_to_index(uint32_t offset)
115 {
116     int index = 0;
117 
118     /* check Exp ROM BAR */
119     if (offset == PCI_ROM_ADDRESS) {
120         return PCI_ROM_SLOT;
121     }
122 
123     /* calculate BAR index */
124     index = (offset - PCI_BASE_ADDRESS_0) >> 2;
125     if (index >= PCI_NUM_REGIONS) {
126         return -1;
127     }
128 
129     return index;
130 }
131 
132 static uint32_t xen_pt_pci_read_config(PCIDevice *d, uint32_t addr, int len)
133 {
134     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
135     uint32_t val = 0;
136     XenPTRegGroup *reg_grp_entry = NULL;
137     XenPTReg *reg_entry = NULL;
138     int rc = 0;
139     int emul_len = 0;
140     uint32_t find_addr = addr;
141 
142     if (xen_pt_pci_config_access_check(d, addr, len)) {
143         goto exit;
144     }
145 
146     /* find register group entry */
147     reg_grp_entry = xen_pt_find_reg_grp(s, addr);
148     if (reg_grp_entry) {
149         /* check 0-Hardwired register group */
150         if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
151             /* no need to emulate, just return 0 */
152             val = 0;
153             goto exit;
154         }
155     }
156 
157     /* read I/O device register value */
158     rc = xen_host_pci_get_block(&s->real_device, addr, (uint8_t *)&val, len);
159     if (rc < 0) {
160         XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
161         memset(&val, 0xff, len);
162     }
163 
164     /* just return the I/O device register value for
165      * passthrough type register group */
166     if (reg_grp_entry == NULL) {
167         goto exit;
168     }
169 
170     /* adjust the read value to appropriate CFC-CFF window */
171     val <<= (addr & 3) << 3;
172     emul_len = len;
173 
174     /* loop around the guest requested size */
175     while (emul_len > 0) {
176         /* find register entry to be emulated */
177         reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
178         if (reg_entry) {
179             XenPTRegInfo *reg = reg_entry->reg;
180             uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
181             uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
182             uint8_t *ptr_val = NULL;
183 
184             valid_mask <<= (find_addr - real_offset) << 3;
185             ptr_val = (uint8_t *)&val + (real_offset & 3);
186 
187             /* do emulation based on register size */
188             switch (reg->size) {
189             case 1:
190                 if (reg->u.b.read) {
191                     rc = reg->u.b.read(s, reg_entry, ptr_val, valid_mask);
192                 }
193                 break;
194             case 2:
195                 if (reg->u.w.read) {
196                     rc = reg->u.w.read(s, reg_entry,
197                                        (uint16_t *)ptr_val, valid_mask);
198                 }
199                 break;
200             case 4:
201                 if (reg->u.dw.read) {
202                     rc = reg->u.dw.read(s, reg_entry,
203                                         (uint32_t *)ptr_val, valid_mask);
204                 }
205                 break;
206             }
207 
208             if (rc < 0) {
209                 xen_shutdown_fatal_error("Internal error: Invalid read "
210                                          "emulation. (%s, rc: %d)\n",
211                                          __func__, rc);
212                 return 0;
213             }
214 
215             /* calculate next address to find */
216             emul_len -= reg->size;
217             if (emul_len > 0) {
218                 find_addr = real_offset + reg->size;
219             }
220         } else {
221             /* nothing to do with passthrough type register,
222              * continue to find next byte */
223             emul_len--;
224             find_addr++;
225         }
226     }
227 
228     /* need to shift back before returning them to pci bus emulator */
229     val >>= ((addr & 3) << 3);
230 
231 exit:
232     XEN_PT_LOG_CONFIG(d, addr, val, len);
233     return val;
234 }
235 
236 static void xen_pt_pci_write_config(PCIDevice *d, uint32_t addr,
237                                     uint32_t val, int len)
238 {
239     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
240     int index = 0;
241     XenPTRegGroup *reg_grp_entry = NULL;
242     int rc = 0;
243     uint32_t read_val = 0, wb_mask;
244     int emul_len = 0;
245     XenPTReg *reg_entry = NULL;
246     uint32_t find_addr = addr;
247     XenPTRegInfo *reg = NULL;
248     bool wp_flag = false;
249 
250     if (xen_pt_pci_config_access_check(d, addr, len)) {
251         return;
252     }
253 
254     XEN_PT_LOG_CONFIG(d, addr, val, len);
255 
256     /* check unused BAR register */
257     index = xen_pt_bar_offset_to_index(addr);
258     if ((index >= 0) && (val != 0)) {
259         uint32_t chk = val;
260 
261         if (index == PCI_ROM_SLOT)
262             chk |= (uint32_t)~PCI_ROM_ADDRESS_MASK;
263 
264         if ((chk != XEN_PT_BAR_ALLF) &&
265             (s->bases[index].bar_flag == XEN_PT_BAR_FLAG_UNUSED)) {
266             XEN_PT_WARN(d, "Guest attempt to set address to unused "
267                         "Base Address Register. (addr: 0x%02x, len: %d)\n",
268                         addr, len);
269         }
270     }
271 
272     /* find register group entry */
273     reg_grp_entry = xen_pt_find_reg_grp(s, addr);
274     if (reg_grp_entry) {
275         /* check 0-Hardwired register group */
276         if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
277             /* ignore silently */
278             XEN_PT_WARN(d, "Access to 0-Hardwired register. "
279                         "(addr: 0x%02x, len: %d)\n", addr, len);
280             return;
281         }
282     }
283 
284     rc = xen_host_pci_get_block(&s->real_device, addr,
285                                 (uint8_t *)&read_val, len);
286     if (rc < 0) {
287         XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
288         memset(&read_val, 0xff, len);
289         wb_mask = 0;
290     } else {
291         wb_mask = 0xFFFFFFFF >> ((4 - len) << 3);
292     }
293 
294     /* pass directly to the real device for passthrough type register group */
295     if (reg_grp_entry == NULL) {
296         if (!s->permissive) {
297             wb_mask = 0;
298             wp_flag = true;
299         }
300         goto out;
301     }
302 
303     memory_region_transaction_begin();
304     pci_default_write_config(d, addr, val, len);
305 
306     /* adjust the read and write value to appropriate CFC-CFF window */
307     read_val <<= (addr & 3) << 3;
308     val <<= (addr & 3) << 3;
309     emul_len = len;
310 
311     /* loop around the guest requested size */
312     while (emul_len > 0) {
313         /* find register entry to be emulated */
314         reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
315         if (reg_entry) {
316             reg = reg_entry->reg;
317             uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
318             uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
319             uint8_t *ptr_val = NULL;
320             uint32_t wp_mask = reg->emu_mask | reg->ro_mask;
321 
322             valid_mask <<= (find_addr - real_offset) << 3;
323             ptr_val = (uint8_t *)&val + (real_offset & 3);
324             if (!s->permissive) {
325                 wp_mask |= reg->res_mask;
326             }
327             if (wp_mask == (0xFFFFFFFF >> ((4 - reg->size) << 3))) {
328                 wb_mask &= ~((wp_mask >> ((find_addr - real_offset) << 3))
329                              << ((len - emul_len) << 3));
330             }
331 
332             /* do emulation based on register size */
333             switch (reg->size) {
334             case 1:
335                 if (reg->u.b.write) {
336                     rc = reg->u.b.write(s, reg_entry, ptr_val,
337                                         read_val >> ((real_offset & 3) << 3),
338                                         valid_mask);
339                 }
340                 break;
341             case 2:
342                 if (reg->u.w.write) {
343                     rc = reg->u.w.write(s, reg_entry, (uint16_t *)ptr_val,
344                                         (read_val >> ((real_offset & 3) << 3)),
345                                         valid_mask);
346                 }
347                 break;
348             case 4:
349                 if (reg->u.dw.write) {
350                     rc = reg->u.dw.write(s, reg_entry, (uint32_t *)ptr_val,
351                                          (read_val >> ((real_offset & 3) << 3)),
352                                          valid_mask);
353                 }
354                 break;
355             }
356 
357             if (rc < 0) {
358                 xen_shutdown_fatal_error("Internal error: Invalid write"
359                                          " emulation. (%s, rc: %d)\n",
360                                          __func__, rc);
361                 return;
362             }
363 
364             /* calculate next address to find */
365             emul_len -= reg->size;
366             if (emul_len > 0) {
367                 find_addr = real_offset + reg->size;
368             }
369         } else {
370             /* nothing to do with passthrough type register,
371              * continue to find next byte */
372             if (!s->permissive) {
373                 wb_mask &= ~(0xff << ((len - emul_len) << 3));
374                 /* Unused BARs will make it here, but we don't want to issue
375                  * warnings for writes to them (bogus writes get dealt with
376                  * above).
377                  */
378                 if (index < 0) {
379                     wp_flag = true;
380                 }
381             }
382             emul_len--;
383             find_addr++;
384         }
385     }
386 
387     /* need to shift back before passing them to xen_host_pci_set_block. */
388     val >>= (addr & 3) << 3;
389 
390     memory_region_transaction_commit();
391 
392 out:
393     if (wp_flag && !s->permissive_warned) {
394         s->permissive_warned = true;
395         xen_pt_log(d, "Write-back to unknown field 0x%02x (partially) inhibited (0x%0*x)\n",
396                    addr, len * 2, wb_mask);
397         xen_pt_log(d, "If the device doesn't work, try enabling permissive mode\n");
398         xen_pt_log(d, "(unsafe) and if it helps report the problem to xen-devel\n");
399     }
400     for (index = 0; wb_mask; index += len) {
401         /* unknown regs are passed through */
402         while (!(wb_mask & 0xff)) {
403             index++;
404             wb_mask >>= 8;
405         }
406         len = 0;
407         do {
408             len++;
409             wb_mask >>= 8;
410         } while (wb_mask & 0xff);
411         rc = xen_host_pci_set_block(&s->real_device, addr + index,
412                                     (uint8_t *)&val + index, len);
413 
414         if (rc < 0) {
415             XEN_PT_ERR(d, "xen_host_pci_set_block failed. return value: %d.\n", rc);
416         }
417     }
418 }
419 
420 /* register regions */
421 
422 static uint64_t xen_pt_bar_read(void *o, hwaddr addr,
423                                 unsigned size)
424 {
425     PCIDevice *d = o;
426     /* if this function is called, that probably means that there is a
427      * misconfiguration of the IOMMU. */
428     XEN_PT_ERR(d, "Should not read BAR through QEMU. @0x"TARGET_FMT_plx"\n",
429                addr);
430     return 0;
431 }
432 static void xen_pt_bar_write(void *o, hwaddr addr, uint64_t val,
433                              unsigned size)
434 {
435     PCIDevice *d = o;
436     /* Same comment as xen_pt_bar_read function */
437     XEN_PT_ERR(d, "Should not write BAR through QEMU. @0x"TARGET_FMT_plx"\n",
438                addr);
439 }
440 
441 static const MemoryRegionOps ops = {
442     .endianness = DEVICE_NATIVE_ENDIAN,
443     .read = xen_pt_bar_read,
444     .write = xen_pt_bar_write,
445 };
446 
447 static int xen_pt_register_regions(XenPCIPassthroughState *s, uint16_t *cmd)
448 {
449     int i = 0;
450     XenHostPCIDevice *d = &s->real_device;
451 
452     /* Register PIO/MMIO BARs */
453     for (i = 0; i < PCI_ROM_SLOT; i++) {
454         XenHostPCIIORegion *r = &d->io_regions[i];
455         uint8_t type;
456 
457         if (r->base_addr == 0 || r->size == 0) {
458             continue;
459         }
460 
461         s->bases[i].access.u = r->base_addr;
462 
463         if (r->type & XEN_HOST_PCI_REGION_TYPE_IO) {
464             type = PCI_BASE_ADDRESS_SPACE_IO;
465             *cmd |= PCI_COMMAND_IO;
466         } else {
467             type = PCI_BASE_ADDRESS_SPACE_MEMORY;
468             if (r->type & XEN_HOST_PCI_REGION_TYPE_PREFETCH) {
469                 type |= PCI_BASE_ADDRESS_MEM_PREFETCH;
470             }
471             if (r->type & XEN_HOST_PCI_REGION_TYPE_MEM_64) {
472                 type |= PCI_BASE_ADDRESS_MEM_TYPE_64;
473             }
474             *cmd |= PCI_COMMAND_MEMORY;
475         }
476 
477         memory_region_init_io(&s->bar[i], OBJECT(s), &ops, &s->dev,
478                               "xen-pci-pt-bar", r->size);
479         pci_register_bar(&s->dev, i, type, &s->bar[i]);
480 
481         XEN_PT_LOG(&s->dev, "IO region %i registered (size=0x%08"PRIx64
482                    " base_addr=0x%08"PRIx64" type: %#x)\n",
483                    i, r->size, r->base_addr, type);
484     }
485 
486     /* Register expansion ROM address */
487     if (d->rom.base_addr && d->rom.size) {
488         uint32_t bar_data = 0;
489 
490         /* Re-set BAR reported by OS, otherwise ROM can't be read. */
491         if (xen_host_pci_get_long(d, PCI_ROM_ADDRESS, &bar_data)) {
492             return 0;
493         }
494         if ((bar_data & PCI_ROM_ADDRESS_MASK) == 0) {
495             bar_data |= d->rom.base_addr & PCI_ROM_ADDRESS_MASK;
496             xen_host_pci_set_long(d, PCI_ROM_ADDRESS, bar_data);
497         }
498 
499         s->bases[PCI_ROM_SLOT].access.maddr = d->rom.base_addr;
500 
501         memory_region_init_io(&s->rom, OBJECT(s), &ops, &s->dev,
502                               "xen-pci-pt-rom", d->rom.size);
503         pci_register_bar(&s->dev, PCI_ROM_SLOT, PCI_BASE_ADDRESS_MEM_PREFETCH,
504                          &s->rom);
505 
506         XEN_PT_LOG(&s->dev, "Expansion ROM registered (size=0x%08"PRIx64
507                    " base_addr=0x%08"PRIx64")\n",
508                    d->rom.size, d->rom.base_addr);
509     }
510 
511     xen_pt_register_vga_regions(d);
512     return 0;
513 }
514 
515 /* region mapping */
516 
517 static int xen_pt_bar_from_region(XenPCIPassthroughState *s, MemoryRegion *mr)
518 {
519     int i = 0;
520 
521     for (i = 0; i < PCI_NUM_REGIONS - 1; i++) {
522         if (mr == &s->bar[i]) {
523             return i;
524         }
525     }
526     if (mr == &s->rom) {
527         return PCI_ROM_SLOT;
528     }
529     return -1;
530 }
531 
532 /*
533  * This function checks if an io_region overlaps an io_region from another
534  * device.  The io_region to check is provided with (addr, size and type)
535  * A callback can be provided and will be called for every region that is
536  * overlapped.
537  * The return value indicates if the region is overlappsed */
538 struct CheckBarArgs {
539     XenPCIPassthroughState *s;
540     pcibus_t addr;
541     pcibus_t size;
542     uint8_t type;
543     bool rc;
544 };
545 static void xen_pt_check_bar_overlap(PCIBus *bus, PCIDevice *d, void *opaque)
546 {
547     struct CheckBarArgs *arg = opaque;
548     XenPCIPassthroughState *s = arg->s;
549     uint8_t type = arg->type;
550     int i;
551 
552     if (d->devfn == s->dev.devfn) {
553         return;
554     }
555 
556     /* xxx: This ignores bridges. */
557     for (i = 0; i < PCI_NUM_REGIONS; i++) {
558         const PCIIORegion *r = &d->io_regions[i];
559 
560         if (!r->size) {
561             continue;
562         }
563         if ((type & PCI_BASE_ADDRESS_SPACE_IO)
564             != (r->type & PCI_BASE_ADDRESS_SPACE_IO)) {
565             continue;
566         }
567 
568         if (ranges_overlap(arg->addr, arg->size, r->addr, r->size)) {
569             XEN_PT_WARN(&s->dev,
570                         "Overlapped to device [%02x:%02x.%d] Region: %i"
571                         " (addr: %#"FMT_PCIBUS", len: %#"FMT_PCIBUS")\n",
572                         pci_bus_num(bus), PCI_SLOT(d->devfn),
573                         PCI_FUNC(d->devfn), i, r->addr, r->size);
574             arg->rc = true;
575         }
576     }
577 }
578 
579 static void xen_pt_region_update(XenPCIPassthroughState *s,
580                                  MemoryRegionSection *sec, bool adding)
581 {
582     PCIDevice *d = &s->dev;
583     MemoryRegion *mr = sec->mr;
584     int bar = -1;
585     int rc;
586     int op = adding ? DPCI_ADD_MAPPING : DPCI_REMOVE_MAPPING;
587     struct CheckBarArgs args = {
588         .s = s,
589         .addr = sec->offset_within_address_space,
590         .size = int128_get64(sec->size),
591         .rc = false,
592     };
593 
594     bar = xen_pt_bar_from_region(s, mr);
595     if (bar == -1 && (!s->msix || &s->msix->mmio != mr)) {
596         return;
597     }
598 
599     if (s->msix && &s->msix->mmio == mr) {
600         if (adding) {
601             s->msix->mmio_base_addr = sec->offset_within_address_space;
602             rc = xen_pt_msix_update_remap(s, s->msix->bar_index);
603         }
604         return;
605     }
606 
607     args.type = d->io_regions[bar].type;
608     pci_for_each_device(pci_get_bus(d), pci_dev_bus_num(d),
609                         xen_pt_check_bar_overlap, &args);
610     if (args.rc) {
611         XEN_PT_WARN(d, "Region: %d (addr: %#"FMT_PCIBUS
612                     ", len: %#"FMT_PCIBUS") is overlapped.\n",
613                     bar, sec->offset_within_address_space,
614                     int128_get64(sec->size));
615     }
616 
617     if (d->io_regions[bar].type & PCI_BASE_ADDRESS_SPACE_IO) {
618         uint32_t guest_port = sec->offset_within_address_space;
619         uint32_t machine_port = s->bases[bar].access.pio_base;
620         uint32_t size = int128_get64(sec->size);
621         rc = xc_domain_ioport_mapping(xen_xc, xen_domid,
622                                       guest_port, machine_port, size,
623                                       op);
624         if (rc) {
625             XEN_PT_ERR(d, "%s ioport mapping failed! (err: %i)\n",
626                        adding ? "create new" : "remove old", errno);
627         }
628     } else {
629         pcibus_t guest_addr = sec->offset_within_address_space;
630         pcibus_t machine_addr = s->bases[bar].access.maddr
631             + sec->offset_within_region;
632         pcibus_t size = int128_get64(sec->size);
633         rc = xc_domain_memory_mapping(xen_xc, xen_domid,
634                                       XEN_PFN(guest_addr + XC_PAGE_SIZE - 1),
635                                       XEN_PFN(machine_addr + XC_PAGE_SIZE - 1),
636                                       XEN_PFN(size + XC_PAGE_SIZE - 1),
637                                       op);
638         if (rc) {
639             XEN_PT_ERR(d, "%s mem mapping failed! (err: %i)\n",
640                        adding ? "create new" : "remove old", errno);
641         }
642     }
643 }
644 
645 static void xen_pt_region_add(MemoryListener *l, MemoryRegionSection *sec)
646 {
647     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
648                                              memory_listener);
649 
650     memory_region_ref(sec->mr);
651     xen_pt_region_update(s, sec, true);
652 }
653 
654 static void xen_pt_region_del(MemoryListener *l, MemoryRegionSection *sec)
655 {
656     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
657                                              memory_listener);
658 
659     xen_pt_region_update(s, sec, false);
660     memory_region_unref(sec->mr);
661 }
662 
663 static void xen_pt_io_region_add(MemoryListener *l, MemoryRegionSection *sec)
664 {
665     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
666                                              io_listener);
667 
668     memory_region_ref(sec->mr);
669     xen_pt_region_update(s, sec, true);
670 }
671 
672 static void xen_pt_io_region_del(MemoryListener *l, MemoryRegionSection *sec)
673 {
674     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
675                                              io_listener);
676 
677     xen_pt_region_update(s, sec, false);
678     memory_region_unref(sec->mr);
679 }
680 
681 static const MemoryListener xen_pt_memory_listener = {
682     .region_add = xen_pt_region_add,
683     .region_del = xen_pt_region_del,
684     .priority = 10,
685 };
686 
687 static const MemoryListener xen_pt_io_listener = {
688     .region_add = xen_pt_io_region_add,
689     .region_del = xen_pt_io_region_del,
690     .priority = 10,
691 };
692 
693 static void
694 xen_igd_passthrough_isa_bridge_create(XenPCIPassthroughState *s,
695                                       XenHostPCIDevice *dev)
696 {
697     uint16_t gpu_dev_id;
698     PCIDevice *d = &s->dev;
699 
700     gpu_dev_id = dev->device_id;
701     igd_passthrough_isa_bridge_create(pci_get_bus(d), gpu_dev_id);
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     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
771     int i, rc = 0;
772     uint8_t machine_irq = 0, scratch;
773     uint16_t cmd = 0;
774     int pirq = XEN_PT_UNASSIGNED_PIRQ;
775     Error *err = NULL;
776 
777     /* register real device */
778     XEN_PT_LOG(d, "Assigning real physical device %02x:%02x.%d"
779                " to devfn %#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                             &err);
787     if (err) {
788         error_append_hint(&err, "Failed to \"open\" the real pci device");
789         error_propagate(errp, err);
790         return;
791     }
792 
793     s->is_virtfn = s->real_device.is_virtfn;
794     if (s->is_virtfn) {
795         XEN_PT_LOG(d, "%04x:%02x:%02x.%d is a SR-IOV Virtual Function\n",
796                    s->real_device.domain, s->real_device.bus,
797                    s->real_device.dev, s->real_device.func);
798     }
799 
800     /* Initialize virtualized PCI configuration (Extended 256 Bytes) */
801     memset(d->config, 0, PCI_CONFIG_SPACE_SIZE);
802 
803     s->memory_listener = xen_pt_memory_listener;
804     s->io_listener = xen_pt_io_listener;
805 
806     /* Setup VGA bios for passthrough GFX */
807     if ((s->real_device.domain == 0) && (s->real_device.bus == 0) &&
808         (s->real_device.dev == 2) && (s->real_device.func == 0)) {
809         if (!is_igd_vga_passthrough(&s->real_device)) {
810             error_setg(errp, "Need to enable igd-passthru if you're trying"
811                     " to passthrough IGD GFX");
812             xen_host_pci_device_put(&s->real_device);
813             return;
814         }
815 
816         xen_pt_setup_vga(s, &s->real_device, &err);
817         if (err) {
818             error_append_hint(&err, "Setup VGA BIOS of passthrough"
819                     " GFX failed");
820             error_propagate(errp, err);
821             xen_host_pci_device_put(&s->real_device);
822             return;
823         }
824 
825         /* Register ISA bridge for passthrough GFX. */
826         xen_igd_passthrough_isa_bridge_create(s, &s->real_device);
827     }
828 
829     /* Handle real device's MMIO/PIO BARs */
830     xen_pt_register_regions(s, &cmd);
831 
832     /* reinitialize each config register to be emulated */
833     xen_pt_config_init(s, &err);
834     if (err) {
835         error_append_hint(&err, "PCI Config space initialisation failed");
836         error_propagate(errp, err);
837         rc = -1;
838         goto err_out;
839     }
840 
841     /* Bind interrupt */
842     rc = xen_host_pci_get_byte(&s->real_device, PCI_INTERRUPT_PIN, &scratch);
843     if (rc) {
844         error_setg_errno(errp, errno, "Failed to read PCI_INTERRUPT_PIN");
845         goto err_out;
846     }
847     if (!scratch) {
848         XEN_PT_LOG(d, "no pin interrupt\n");
849         goto out;
850     }
851 
852     machine_irq = s->real_device.irq;
853     if (machine_irq == 0) {
854         XEN_PT_LOG(d, "machine irq is 0\n");
855         cmd |= PCI_COMMAND_INTX_DISABLE;
856         goto out;
857     }
858 
859     rc = xc_physdev_map_pirq(xen_xc, xen_domid, machine_irq, &pirq);
860     if (rc < 0) {
861         XEN_PT_ERR(d, "Mapping machine irq %u to pirq %i failed, (err: %d)\n",
862                    machine_irq, pirq, errno);
863 
864         /* Disable PCI intx assertion (turn on bit10 of devctl) */
865         cmd |= PCI_COMMAND_INTX_DISABLE;
866         machine_irq = 0;
867         s->machine_irq = 0;
868     } else {
869         machine_irq = pirq;
870         s->machine_irq = pirq;
871         xen_pt_mapped_machine_irq[machine_irq]++;
872     }
873 
874     /* bind machine_irq to device */
875     if (machine_irq != 0) {
876         uint8_t e_intx = xen_pt_pci_intx(s);
877 
878         rc = xc_domain_bind_pt_pci_irq(xen_xc, xen_domid, machine_irq,
879                                        pci_dev_bus_num(d),
880                                        PCI_SLOT(d->devfn),
881                                        e_intx);
882         if (rc < 0) {
883             XEN_PT_ERR(d, "Binding of interrupt %i failed! (err: %d)\n",
884                        e_intx, errno);
885 
886             /* Disable PCI intx assertion (turn on bit10 of devctl) */
887             cmd |= PCI_COMMAND_INTX_DISABLE;
888             xen_pt_mapped_machine_irq[machine_irq]--;
889 
890             if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
891                 if (xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq)) {
892                     XEN_PT_ERR(d, "Unmapping of machine interrupt %i failed!"
893                                " (err: %d)\n", machine_irq, errno);
894                 }
895             }
896             s->machine_irq = 0;
897         }
898     }
899 
900 out:
901     if (cmd) {
902         uint16_t val;
903 
904         rc = xen_host_pci_get_word(&s->real_device, PCI_COMMAND, &val);
905         if (rc) {
906             error_setg_errno(errp, errno, "Failed to read PCI_COMMAND");
907             goto err_out;
908         } else {
909             val |= cmd;
910             rc = xen_host_pci_set_word(&s->real_device, PCI_COMMAND, val);
911             if (rc) {
912                 error_setg_errno(errp, errno, "Failed to write PCI_COMMAND"
913                                  " val = 0x%x", val);
914                 goto err_out;
915             }
916         }
917     }
918 
919     memory_listener_register(&s->memory_listener, &address_space_memory);
920     memory_listener_register(&s->io_listener, &address_space_io);
921     s->listener_set = true;
922     XEN_PT_LOG(d,
923                "Real physical device %02x:%02x.%d registered successfully\n",
924                s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function);
925 
926     return;
927 
928 err_out:
929     for (i = 0; i < PCI_ROM_SLOT; i++) {
930         object_unparent(OBJECT(&s->bar[i]));
931     }
932     object_unparent(OBJECT(&s->rom));
933 
934     xen_pt_destroy(d);
935     assert(rc);
936 }
937 
938 static void xen_pt_unregister_device(PCIDevice *d)
939 {
940     xen_pt_destroy(d);
941 }
942 
943 static Property xen_pci_passthrough_properties[] = {
944     DEFINE_PROP_PCI_HOST_DEVADDR("hostaddr", XenPCIPassthroughState, hostaddr),
945     DEFINE_PROP_BOOL("permissive", XenPCIPassthroughState, permissive, false),
946     DEFINE_PROP_END_OF_LIST(),
947 };
948 
949 static void xen_pci_passthrough_instance_init(Object *obj)
950 {
951     /* QEMU_PCI_CAP_EXPRESS initialization does not depend on QEMU command
952      * line, therefore, no need to wait to realize like other devices */
953     PCI_DEVICE(obj)->cap_present |= QEMU_PCI_CAP_EXPRESS;
954 }
955 
956 static void xen_pci_passthrough_class_init(ObjectClass *klass, void *data)
957 {
958     DeviceClass *dc = DEVICE_CLASS(klass);
959     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
960 
961     k->realize = xen_pt_realize;
962     k->exit = xen_pt_unregister_device;
963     k->config_read = xen_pt_pci_read_config;
964     k->config_write = xen_pt_pci_write_config;
965     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
966     dc->desc = "Assign an host PCI device with Xen";
967     device_class_set_props(dc, xen_pci_passthrough_properties);
968 };
969 
970 static void xen_pci_passthrough_finalize(Object *obj)
971 {
972     XenPCIPassthroughState *s = XEN_PT_DEVICE(obj);
973 
974     xen_pt_msix_delete(s);
975 }
976 
977 static const TypeInfo xen_pci_passthrough_info = {
978     .name = TYPE_XEN_PT_DEVICE,
979     .parent = TYPE_PCI_DEVICE,
980     .instance_size = sizeof(XenPCIPassthroughState),
981     .instance_finalize = xen_pci_passthrough_finalize,
982     .class_init = xen_pci_passthrough_class_init,
983     .instance_init = xen_pci_passthrough_instance_init,
984     .interfaces = (InterfaceInfo[]) {
985         { INTERFACE_CONVENTIONAL_PCI_DEVICE },
986         { INTERFACE_PCIE_DEVICE },
987         { },
988     },
989 };
990 
991 static void xen_pci_passthrough_register_types(void)
992 {
993     type_register_static(&xen_pci_passthrough_info);
994 }
995 
996 type_init(xen_pci_passthrough_register_types)
997