xref: /openbmc/qemu/hw/xen/xen_pt.c (revision da054c64)
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/xen/xen.h"
61 #include "hw/i386/pc.h"
62 #include "hw/xen/xen-legacy-backend.h"
63 #include "xen_pt.h"
64 #include "qemu/range.h"
65 #include "exec/address-spaces.h"
66 
67 #define XEN_PT_NR_IRQS (256)
68 static uint8_t xen_pt_mapped_machine_irq[XEN_PT_NR_IRQS] = {0};
69 
70 void xen_pt_log(const PCIDevice *d, const char *f, ...)
71 {
72     va_list ap;
73 
74     va_start(ap, f);
75     if (d) {
76         fprintf(stderr, "[%02x:%02x.%d] ", pci_dev_bus_num(d),
77                 PCI_SLOT(d->devfn), PCI_FUNC(d->devfn));
78     }
79     vfprintf(stderr, f, ap);
80     va_end(ap);
81 }
82 
83 /* Config Space */
84 
85 static int xen_pt_pci_config_access_check(PCIDevice *d, uint32_t addr, int len)
86 {
87     /* check offset range */
88     if (addr > 0xFF) {
89         XEN_PT_ERR(d, "Failed to access register with offset exceeding 0xFF. "
90                    "(addr: 0x%02x, len: %d)\n", addr, len);
91         return -1;
92     }
93 
94     /* check read size */
95     if ((len != 1) && (len != 2) && (len != 4)) {
96         XEN_PT_ERR(d, "Failed to access register with invalid access length. "
97                    "(addr: 0x%02x, len: %d)\n", addr, len);
98         return -1;
99     }
100 
101     /* check offset alignment */
102     if (addr & (len - 1)) {
103         XEN_PT_ERR(d, "Failed to access register with invalid access size "
104                    "alignment. (addr: 0x%02x, len: %d)\n", addr, len);
105         return -1;
106     }
107 
108     return 0;
109 }
110 
111 int xen_pt_bar_offset_to_index(uint32_t offset)
112 {
113     int index = 0;
114 
115     /* check Exp ROM BAR */
116     if (offset == PCI_ROM_ADDRESS) {
117         return PCI_ROM_SLOT;
118     }
119 
120     /* calculate BAR index */
121     index = (offset - PCI_BASE_ADDRESS_0) >> 2;
122     if (index >= PCI_NUM_REGIONS) {
123         return -1;
124     }
125 
126     return index;
127 }
128 
129 static uint32_t xen_pt_pci_read_config(PCIDevice *d, uint32_t addr, int len)
130 {
131     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
132     uint32_t val = 0;
133     XenPTRegGroup *reg_grp_entry = NULL;
134     XenPTReg *reg_entry = NULL;
135     int rc = 0;
136     int emul_len = 0;
137     uint32_t find_addr = addr;
138 
139     if (xen_pt_pci_config_access_check(d, addr, len)) {
140         goto exit;
141     }
142 
143     /* find register group entry */
144     reg_grp_entry = xen_pt_find_reg_grp(s, addr);
145     if (reg_grp_entry) {
146         /* check 0-Hardwired register group */
147         if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
148             /* no need to emulate, just return 0 */
149             val = 0;
150             goto exit;
151         }
152     }
153 
154     /* read I/O device register value */
155     rc = xen_host_pci_get_block(&s->real_device, addr, (uint8_t *)&val, len);
156     if (rc < 0) {
157         XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
158         memset(&val, 0xff, len);
159     }
160 
161     /* just return the I/O device register value for
162      * passthrough type register group */
163     if (reg_grp_entry == NULL) {
164         goto exit;
165     }
166 
167     /* adjust the read value to appropriate CFC-CFF window */
168     val <<= (addr & 3) << 3;
169     emul_len = len;
170 
171     /* loop around the guest requested size */
172     while (emul_len > 0) {
173         /* find register entry to be emulated */
174         reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
175         if (reg_entry) {
176             XenPTRegInfo *reg = reg_entry->reg;
177             uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
178             uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
179             uint8_t *ptr_val = NULL;
180 
181             valid_mask <<= (find_addr - real_offset) << 3;
182             ptr_val = (uint8_t *)&val + (real_offset & 3);
183 
184             /* do emulation based on register size */
185             switch (reg->size) {
186             case 1:
187                 if (reg->u.b.read) {
188                     rc = reg->u.b.read(s, reg_entry, ptr_val, valid_mask);
189                 }
190                 break;
191             case 2:
192                 if (reg->u.w.read) {
193                     rc = reg->u.w.read(s, reg_entry,
194                                        (uint16_t *)ptr_val, valid_mask);
195                 }
196                 break;
197             case 4:
198                 if (reg->u.dw.read) {
199                     rc = reg->u.dw.read(s, reg_entry,
200                                         (uint32_t *)ptr_val, valid_mask);
201                 }
202                 break;
203             }
204 
205             if (rc < 0) {
206                 xen_shutdown_fatal_error("Internal error: Invalid read "
207                                          "emulation. (%s, rc: %d)\n",
208                                          __func__, rc);
209                 return 0;
210             }
211 
212             /* calculate next address to find */
213             emul_len -= reg->size;
214             if (emul_len > 0) {
215                 find_addr = real_offset + reg->size;
216             }
217         } else {
218             /* nothing to do with passthrough type register,
219              * continue to find next byte */
220             emul_len--;
221             find_addr++;
222         }
223     }
224 
225     /* need to shift back before returning them to pci bus emulator */
226     val >>= ((addr & 3) << 3);
227 
228 exit:
229     XEN_PT_LOG_CONFIG(d, addr, val, len);
230     return val;
231 }
232 
233 static void xen_pt_pci_write_config(PCIDevice *d, uint32_t addr,
234                                     uint32_t val, int len)
235 {
236     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
237     int index = 0;
238     XenPTRegGroup *reg_grp_entry = NULL;
239     int rc = 0;
240     uint32_t read_val = 0, wb_mask;
241     int emul_len = 0;
242     XenPTReg *reg_entry = NULL;
243     uint32_t find_addr = addr;
244     XenPTRegInfo *reg = NULL;
245     bool wp_flag = false;
246 
247     if (xen_pt_pci_config_access_check(d, addr, len)) {
248         return;
249     }
250 
251     XEN_PT_LOG_CONFIG(d, addr, val, len);
252 
253     /* check unused BAR register */
254     index = xen_pt_bar_offset_to_index(addr);
255     if ((index >= 0) && (val != 0)) {
256         uint32_t chk = val;
257 
258         if (index == PCI_ROM_SLOT)
259             chk |= (uint32_t)~PCI_ROM_ADDRESS_MASK;
260 
261         if ((chk != XEN_PT_BAR_ALLF) &&
262             (s->bases[index].bar_flag == XEN_PT_BAR_FLAG_UNUSED)) {
263             XEN_PT_WARN(d, "Guest attempt to set address to unused "
264                         "Base Address Register. (addr: 0x%02x, len: %d)\n",
265                         addr, len);
266         }
267     }
268 
269     /* find register group entry */
270     reg_grp_entry = xen_pt_find_reg_grp(s, addr);
271     if (reg_grp_entry) {
272         /* check 0-Hardwired register group */
273         if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
274             /* ignore silently */
275             XEN_PT_WARN(d, "Access to 0-Hardwired register. "
276                         "(addr: 0x%02x, len: %d)\n", addr, len);
277             return;
278         }
279     }
280 
281     rc = xen_host_pci_get_block(&s->real_device, addr,
282                                 (uint8_t *)&read_val, len);
283     if (rc < 0) {
284         XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
285         memset(&read_val, 0xff, len);
286         wb_mask = 0;
287     } else {
288         wb_mask = 0xFFFFFFFF >> ((4 - len) << 3);
289     }
290 
291     /* pass directly to the real device for passthrough type register group */
292     if (reg_grp_entry == NULL) {
293         if (!s->permissive) {
294             wb_mask = 0;
295             wp_flag = true;
296         }
297         goto out;
298     }
299 
300     memory_region_transaction_begin();
301     pci_default_write_config(d, addr, val, len);
302 
303     /* adjust the read and write value to appropriate CFC-CFF window */
304     read_val <<= (addr & 3) << 3;
305     val <<= (addr & 3) << 3;
306     emul_len = len;
307 
308     /* loop around the guest requested size */
309     while (emul_len > 0) {
310         /* find register entry to be emulated */
311         reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
312         if (reg_entry) {
313             reg = reg_entry->reg;
314             uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
315             uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
316             uint8_t *ptr_val = NULL;
317             uint32_t wp_mask = reg->emu_mask | reg->ro_mask;
318 
319             valid_mask <<= (find_addr - real_offset) << 3;
320             ptr_val = (uint8_t *)&val + (real_offset & 3);
321             if (!s->permissive) {
322                 wp_mask |= reg->res_mask;
323             }
324             if (wp_mask == (0xFFFFFFFF >> ((4 - reg->size) << 3))) {
325                 wb_mask &= ~((wp_mask >> ((find_addr - real_offset) << 3))
326                              << ((len - emul_len) << 3));
327             }
328 
329             /* do emulation based on register size */
330             switch (reg->size) {
331             case 1:
332                 if (reg->u.b.write) {
333                     rc = reg->u.b.write(s, reg_entry, ptr_val,
334                                         read_val >> ((real_offset & 3) << 3),
335                                         valid_mask);
336                 }
337                 break;
338             case 2:
339                 if (reg->u.w.write) {
340                     rc = reg->u.w.write(s, reg_entry, (uint16_t *)ptr_val,
341                                         (read_val >> ((real_offset & 3) << 3)),
342                                         valid_mask);
343                 }
344                 break;
345             case 4:
346                 if (reg->u.dw.write) {
347                     rc = reg->u.dw.write(s, reg_entry, (uint32_t *)ptr_val,
348                                          (read_val >> ((real_offset & 3) << 3)),
349                                          valid_mask);
350                 }
351                 break;
352             }
353 
354             if (rc < 0) {
355                 xen_shutdown_fatal_error("Internal error: Invalid write"
356                                          " emulation. (%s, rc: %d)\n",
357                                          __func__, rc);
358                 return;
359             }
360 
361             /* calculate next address to find */
362             emul_len -= reg->size;
363             if (emul_len > 0) {
364                 find_addr = real_offset + reg->size;
365             }
366         } else {
367             /* nothing to do with passthrough type register,
368              * continue to find next byte */
369             if (!s->permissive) {
370                 wb_mask &= ~(0xff << ((len - emul_len) << 3));
371                 /* Unused BARs will make it here, but we don't want to issue
372                  * warnings for writes to them (bogus writes get dealt with
373                  * above).
374                  */
375                 if (index < 0) {
376                     wp_flag = true;
377                 }
378             }
379             emul_len--;
380             find_addr++;
381         }
382     }
383 
384     /* need to shift back before passing them to xen_host_pci_set_block. */
385     val >>= (addr & 3) << 3;
386 
387     memory_region_transaction_commit();
388 
389 out:
390     if (wp_flag && !s->permissive_warned) {
391         s->permissive_warned = true;
392         xen_pt_log(d, "Write-back to unknown field 0x%02x (partially) inhibited (0x%0*x)\n",
393                    addr, len * 2, wb_mask);
394         xen_pt_log(d, "If the device doesn't work, try enabling permissive mode\n");
395         xen_pt_log(d, "(unsafe) and if it helps report the problem to xen-devel\n");
396     }
397     for (index = 0; wb_mask; index += len) {
398         /* unknown regs are passed through */
399         while (!(wb_mask & 0xff)) {
400             index++;
401             wb_mask >>= 8;
402         }
403         len = 0;
404         do {
405             len++;
406             wb_mask >>= 8;
407         } while (wb_mask & 0xff);
408         rc = xen_host_pci_set_block(&s->real_device, addr + index,
409                                     (uint8_t *)&val + index, len);
410 
411         if (rc < 0) {
412             XEN_PT_ERR(d, "xen_host_pci_set_block failed. return value: %d.\n", rc);
413         }
414     }
415 }
416 
417 /* register regions */
418 
419 static uint64_t xen_pt_bar_read(void *o, hwaddr addr,
420                                 unsigned size)
421 {
422     PCIDevice *d = o;
423     /* if this function is called, that probably means that there is a
424      * misconfiguration of the IOMMU. */
425     XEN_PT_ERR(d, "Should not read BAR through QEMU. @0x"TARGET_FMT_plx"\n",
426                addr);
427     return 0;
428 }
429 static void xen_pt_bar_write(void *o, hwaddr addr, uint64_t val,
430                              unsigned size)
431 {
432     PCIDevice *d = o;
433     /* Same comment as xen_pt_bar_read function */
434     XEN_PT_ERR(d, "Should not write BAR through QEMU. @0x"TARGET_FMT_plx"\n",
435                addr);
436 }
437 
438 static const MemoryRegionOps ops = {
439     .endianness = DEVICE_NATIVE_ENDIAN,
440     .read = xen_pt_bar_read,
441     .write = xen_pt_bar_write,
442 };
443 
444 static int xen_pt_register_regions(XenPCIPassthroughState *s, uint16_t *cmd)
445 {
446     int i = 0;
447     XenHostPCIDevice *d = &s->real_device;
448 
449     /* Register PIO/MMIO BARs */
450     for (i = 0; i < PCI_ROM_SLOT; i++) {
451         XenHostPCIIORegion *r = &d->io_regions[i];
452         uint8_t type;
453 
454         if (r->base_addr == 0 || r->size == 0) {
455             continue;
456         }
457 
458         s->bases[i].access.u = r->base_addr;
459 
460         if (r->type & XEN_HOST_PCI_REGION_TYPE_IO) {
461             type = PCI_BASE_ADDRESS_SPACE_IO;
462             *cmd |= PCI_COMMAND_IO;
463         } else {
464             type = PCI_BASE_ADDRESS_SPACE_MEMORY;
465             if (r->type & XEN_HOST_PCI_REGION_TYPE_PREFETCH) {
466                 type |= PCI_BASE_ADDRESS_MEM_PREFETCH;
467             }
468             if (r->type & XEN_HOST_PCI_REGION_TYPE_MEM_64) {
469                 type |= PCI_BASE_ADDRESS_MEM_TYPE_64;
470             }
471             *cmd |= PCI_COMMAND_MEMORY;
472         }
473 
474         memory_region_init_io(&s->bar[i], OBJECT(s), &ops, &s->dev,
475                               "xen-pci-pt-bar", r->size);
476         pci_register_bar(&s->dev, i, type, &s->bar[i]);
477 
478         XEN_PT_LOG(&s->dev, "IO region %i registered (size=0x%08"PRIx64
479                    " base_addr=0x%08"PRIx64" type: %#x)\n",
480                    i, r->size, r->base_addr, type);
481     }
482 
483     /* Register expansion ROM address */
484     if (d->rom.base_addr && d->rom.size) {
485         uint32_t bar_data = 0;
486 
487         /* Re-set BAR reported by OS, otherwise ROM can't be read. */
488         if (xen_host_pci_get_long(d, PCI_ROM_ADDRESS, &bar_data)) {
489             return 0;
490         }
491         if ((bar_data & PCI_ROM_ADDRESS_MASK) == 0) {
492             bar_data |= d->rom.base_addr & PCI_ROM_ADDRESS_MASK;
493             xen_host_pci_set_long(d, PCI_ROM_ADDRESS, bar_data);
494         }
495 
496         s->bases[PCI_ROM_SLOT].access.maddr = d->rom.base_addr;
497 
498         memory_region_init_io(&s->rom, OBJECT(s), &ops, &s->dev,
499                               "xen-pci-pt-rom", d->rom.size);
500         pci_register_bar(&s->dev, PCI_ROM_SLOT, PCI_BASE_ADDRESS_MEM_PREFETCH,
501                          &s->rom);
502 
503         XEN_PT_LOG(&s->dev, "Expansion ROM registered (size=0x%08"PRIx64
504                    " base_addr=0x%08"PRIx64")\n",
505                    d->rom.size, d->rom.base_addr);
506     }
507 
508     xen_pt_register_vga_regions(d);
509     return 0;
510 }
511 
512 /* region mapping */
513 
514 static int xen_pt_bar_from_region(XenPCIPassthroughState *s, MemoryRegion *mr)
515 {
516     int i = 0;
517 
518     for (i = 0; i < PCI_NUM_REGIONS - 1; i++) {
519         if (mr == &s->bar[i]) {
520             return i;
521         }
522     }
523     if (mr == &s->rom) {
524         return PCI_ROM_SLOT;
525     }
526     return -1;
527 }
528 
529 /*
530  * This function checks if an io_region overlaps an io_region from another
531  * device.  The io_region to check is provided with (addr, size and type)
532  * A callback can be provided and will be called for every region that is
533  * overlapped.
534  * The return value indicates if the region is overlappsed */
535 struct CheckBarArgs {
536     XenPCIPassthroughState *s;
537     pcibus_t addr;
538     pcibus_t size;
539     uint8_t type;
540     bool rc;
541 };
542 static void xen_pt_check_bar_overlap(PCIBus *bus, PCIDevice *d, void *opaque)
543 {
544     struct CheckBarArgs *arg = opaque;
545     XenPCIPassthroughState *s = arg->s;
546     uint8_t type = arg->type;
547     int i;
548 
549     if (d->devfn == s->dev.devfn) {
550         return;
551     }
552 
553     /* xxx: This ignores bridges. */
554     for (i = 0; i < PCI_NUM_REGIONS; i++) {
555         const PCIIORegion *r = &d->io_regions[i];
556 
557         if (!r->size) {
558             continue;
559         }
560         if ((type & PCI_BASE_ADDRESS_SPACE_IO)
561             != (r->type & PCI_BASE_ADDRESS_SPACE_IO)) {
562             continue;
563         }
564 
565         if (ranges_overlap(arg->addr, arg->size, r->addr, r->size)) {
566             XEN_PT_WARN(&s->dev,
567                         "Overlapped to device [%02x:%02x.%d] Region: %i"
568                         " (addr: %#"FMT_PCIBUS", len: %#"FMT_PCIBUS")\n",
569                         pci_bus_num(bus), PCI_SLOT(d->devfn),
570                         PCI_FUNC(d->devfn), i, r->addr, r->size);
571             arg->rc = true;
572         }
573     }
574 }
575 
576 static void xen_pt_region_update(XenPCIPassthroughState *s,
577                                  MemoryRegionSection *sec, bool adding)
578 {
579     PCIDevice *d = &s->dev;
580     MemoryRegion *mr = sec->mr;
581     int bar = -1;
582     int rc;
583     int op = adding ? DPCI_ADD_MAPPING : DPCI_REMOVE_MAPPING;
584     struct CheckBarArgs args = {
585         .s = s,
586         .addr = sec->offset_within_address_space,
587         .size = int128_get64(sec->size),
588         .rc = false,
589     };
590 
591     bar = xen_pt_bar_from_region(s, mr);
592     if (bar == -1 && (!s->msix || &s->msix->mmio != mr)) {
593         return;
594     }
595 
596     if (s->msix && &s->msix->mmio == mr) {
597         if (adding) {
598             s->msix->mmio_base_addr = sec->offset_within_address_space;
599             rc = xen_pt_msix_update_remap(s, s->msix->bar_index);
600         }
601         return;
602     }
603 
604     args.type = d->io_regions[bar].type;
605     pci_for_each_device(pci_get_bus(d), pci_dev_bus_num(d),
606                         xen_pt_check_bar_overlap, &args);
607     if (args.rc) {
608         XEN_PT_WARN(d, "Region: %d (addr: %#"FMT_PCIBUS
609                     ", len: %#"FMT_PCIBUS") is overlapped.\n",
610                     bar, sec->offset_within_address_space,
611                     int128_get64(sec->size));
612     }
613 
614     if (d->io_regions[bar].type & PCI_BASE_ADDRESS_SPACE_IO) {
615         uint32_t guest_port = sec->offset_within_address_space;
616         uint32_t machine_port = s->bases[bar].access.pio_base;
617         uint32_t size = int128_get64(sec->size);
618         rc = xc_domain_ioport_mapping(xen_xc, xen_domid,
619                                       guest_port, machine_port, size,
620                                       op);
621         if (rc) {
622             XEN_PT_ERR(d, "%s ioport mapping failed! (err: %i)\n",
623                        adding ? "create new" : "remove old", errno);
624         }
625     } else {
626         pcibus_t guest_addr = sec->offset_within_address_space;
627         pcibus_t machine_addr = s->bases[bar].access.maddr
628             + sec->offset_within_region;
629         pcibus_t size = int128_get64(sec->size);
630         rc = xc_domain_memory_mapping(xen_xc, xen_domid,
631                                       XEN_PFN(guest_addr + XC_PAGE_SIZE - 1),
632                                       XEN_PFN(machine_addr + XC_PAGE_SIZE - 1),
633                                       XEN_PFN(size + XC_PAGE_SIZE - 1),
634                                       op);
635         if (rc) {
636             XEN_PT_ERR(d, "%s mem mapping failed! (err: %i)\n",
637                        adding ? "create new" : "remove old", errno);
638         }
639     }
640 }
641 
642 static void xen_pt_region_add(MemoryListener *l, MemoryRegionSection *sec)
643 {
644     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
645                                              memory_listener);
646 
647     memory_region_ref(sec->mr);
648     xen_pt_region_update(s, sec, true);
649 }
650 
651 static void xen_pt_region_del(MemoryListener *l, MemoryRegionSection *sec)
652 {
653     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
654                                              memory_listener);
655 
656     xen_pt_region_update(s, sec, false);
657     memory_region_unref(sec->mr);
658 }
659 
660 static void xen_pt_io_region_add(MemoryListener *l, MemoryRegionSection *sec)
661 {
662     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
663                                              io_listener);
664 
665     memory_region_ref(sec->mr);
666     xen_pt_region_update(s, sec, true);
667 }
668 
669 static void xen_pt_io_region_del(MemoryListener *l, MemoryRegionSection *sec)
670 {
671     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
672                                              io_listener);
673 
674     xen_pt_region_update(s, sec, false);
675     memory_region_unref(sec->mr);
676 }
677 
678 static const MemoryListener xen_pt_memory_listener = {
679     .region_add = xen_pt_region_add,
680     .region_del = xen_pt_region_del,
681     .priority = 10,
682 };
683 
684 static const MemoryListener xen_pt_io_listener = {
685     .region_add = xen_pt_io_region_add,
686     .region_del = xen_pt_io_region_del,
687     .priority = 10,
688 };
689 
690 static void
691 xen_igd_passthrough_isa_bridge_create(XenPCIPassthroughState *s,
692                                       XenHostPCIDevice *dev)
693 {
694     uint16_t gpu_dev_id;
695     PCIDevice *d = &s->dev;
696 
697     gpu_dev_id = dev->device_id;
698     igd_passthrough_isa_bridge_create(pci_get_bus(d), gpu_dev_id);
699 }
700 
701 /* destroy. */
702 static void xen_pt_destroy(PCIDevice *d) {
703 
704     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
705     XenHostPCIDevice *host_dev = &s->real_device;
706     uint8_t machine_irq = s->machine_irq;
707     uint8_t intx;
708     int rc;
709 
710     if (machine_irq && !xen_host_pci_device_closed(&s->real_device)) {
711         intx = xen_pt_pci_intx(s);
712         rc = xc_domain_unbind_pt_irq(xen_xc, xen_domid, machine_irq,
713                                      PT_IRQ_TYPE_PCI,
714                                      pci_dev_bus_num(d),
715                                      PCI_SLOT(s->dev.devfn),
716                                      intx,
717                                      0 /* isa_irq */);
718         if (rc < 0) {
719             XEN_PT_ERR(d, "unbinding of interrupt INT%c failed."
720                        " (machine irq: %i, err: %d)"
721                        " But bravely continuing on..\n",
722                        'a' + intx, machine_irq, errno);
723         }
724     }
725 
726     /* N.B. xen_pt_config_delete takes care of freeing them. */
727     if (s->msi) {
728         xen_pt_msi_disable(s);
729     }
730     if (s->msix) {
731         xen_pt_msix_disable(s);
732     }
733 
734     if (machine_irq) {
735         xen_pt_mapped_machine_irq[machine_irq]--;
736 
737         if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
738             rc = xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq);
739 
740             if (rc < 0) {
741                 XEN_PT_ERR(d, "unmapping of interrupt %i failed. (err: %d)"
742                            " But bravely continuing on..\n",
743                            machine_irq, errno);
744             }
745         }
746         s->machine_irq = 0;
747     }
748 
749     /* delete all emulated config registers */
750     xen_pt_config_delete(s);
751 
752     xen_pt_unregister_vga_regions(host_dev);
753 
754     if (s->listener_set) {
755         memory_listener_unregister(&s->memory_listener);
756         memory_listener_unregister(&s->io_listener);
757         s->listener_set = false;
758     }
759     if (!xen_host_pci_device_closed(&s->real_device)) {
760         xen_host_pci_device_put(&s->real_device);
761     }
762 }
763 /* init */
764 
765 static void xen_pt_realize(PCIDevice *d, Error **errp)
766 {
767     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
768     int i, rc = 0;
769     uint8_t machine_irq = 0, scratch;
770     uint16_t cmd = 0;
771     int pirq = XEN_PT_UNASSIGNED_PIRQ;
772     Error *err = NULL;
773 
774     /* register real device */
775     XEN_PT_LOG(d, "Assigning real physical device %02x:%02x.%d"
776                " to devfn %#x\n",
777                s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function,
778                s->dev.devfn);
779 
780     xen_host_pci_device_get(&s->real_device,
781                             s->hostaddr.domain, s->hostaddr.bus,
782                             s->hostaddr.slot, s->hostaddr.function,
783                             &err);
784     if (err) {
785         error_append_hint(&err, "Failed to \"open\" the real pci device");
786         error_propagate(errp, err);
787         return;
788     }
789 
790     s->is_virtfn = s->real_device.is_virtfn;
791     if (s->is_virtfn) {
792         XEN_PT_LOG(d, "%04x:%02x:%02x.%d is a SR-IOV Virtual Function\n",
793                    s->real_device.domain, s->real_device.bus,
794                    s->real_device.dev, s->real_device.func);
795     }
796 
797     /* Initialize virtualized PCI configuration (Extended 256 Bytes) */
798     memset(d->config, 0, PCI_CONFIG_SPACE_SIZE);
799 
800     s->memory_listener = xen_pt_memory_listener;
801     s->io_listener = xen_pt_io_listener;
802 
803     /* Setup VGA bios for passthrough GFX */
804     if ((s->real_device.domain == 0) && (s->real_device.bus == 0) &&
805         (s->real_device.dev == 2) && (s->real_device.func == 0)) {
806         if (!is_igd_vga_passthrough(&s->real_device)) {
807             error_setg(errp, "Need to enable igd-passthru if you're trying"
808                     " to passthrough IGD GFX");
809             xen_host_pci_device_put(&s->real_device);
810             return;
811         }
812 
813         xen_pt_setup_vga(s, &s->real_device, &err);
814         if (err) {
815             error_append_hint(&err, "Setup VGA BIOS of passthrough"
816                     " GFX failed");
817             error_propagate(errp, err);
818             xen_host_pci_device_put(&s->real_device);
819             return;
820         }
821 
822         /* Register ISA bridge for passthrough GFX. */
823         xen_igd_passthrough_isa_bridge_create(s, &s->real_device);
824     }
825 
826     /* Handle real device's MMIO/PIO BARs */
827     xen_pt_register_regions(s, &cmd);
828 
829     /* reinitialize each config register to be emulated */
830     xen_pt_config_init(s, &err);
831     if (err) {
832         error_append_hint(&err, "PCI Config space initialisation failed");
833         error_propagate(errp, err);
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         error_setg_errno(errp, errno, "Mapping machine irq %u to"
859                          " pirq %i failed", machine_irq, pirq);
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             error_setg_errno(errp, errno, "Binding of interrupt %u failed",
881                              e_intx);
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                     error_setg_errno(errp, errno, "Unmapping of machine"
890                             " interrupt %u failed", machine_irq);
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     dc->props = 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