xref: /openbmc/qemu/hw/xen/xen_pt.c (revision 135b03cb)
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 #define XEN_PT_NR_IRQS (256)
69 static uint8_t xen_pt_mapped_machine_irq[XEN_PT_NR_IRQS] = {0};
70 
71 void xen_pt_log(const PCIDevice *d, const char *f, ...)
72 {
73     va_list ap;
74 
75     va_start(ap, f);
76     if (d) {
77         fprintf(stderr, "[%02x:%02x.%d] ", pci_dev_bus_num(d),
78                 PCI_SLOT(d->devfn), PCI_FUNC(d->devfn));
79     }
80     vfprintf(stderr, f, ap);
81     va_end(ap);
82 }
83 
84 /* Config Space */
85 
86 static int xen_pt_pci_config_access_check(PCIDevice *d, uint32_t addr, int len)
87 {
88     /* check offset range */
89     if (addr > 0xFF) {
90         XEN_PT_ERR(d, "Failed to access register with offset exceeding 0xFF. "
91                    "(addr: 0x%02x, len: %d)\n", addr, len);
92         return -1;
93     }
94 
95     /* check read size */
96     if ((len != 1) && (len != 2) && (len != 4)) {
97         XEN_PT_ERR(d, "Failed to access register with invalid access length. "
98                    "(addr: 0x%02x, len: %d)\n", addr, len);
99         return -1;
100     }
101 
102     /* check offset alignment */
103     if (addr & (len - 1)) {
104         XEN_PT_ERR(d, "Failed to access register with invalid access size "
105                    "alignment. (addr: 0x%02x, len: %d)\n", addr, len);
106         return -1;
107     }
108 
109     return 0;
110 }
111 
112 int xen_pt_bar_offset_to_index(uint32_t offset)
113 {
114     int index = 0;
115 
116     /* check Exp ROM BAR */
117     if (offset == PCI_ROM_ADDRESS) {
118         return PCI_ROM_SLOT;
119     }
120 
121     /* calculate BAR index */
122     index = (offset - PCI_BASE_ADDRESS_0) >> 2;
123     if (index >= PCI_NUM_REGIONS) {
124         return -1;
125     }
126 
127     return index;
128 }
129 
130 static uint32_t xen_pt_pci_read_config(PCIDevice *d, uint32_t addr, int len)
131 {
132     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
133     uint32_t val = 0;
134     XenPTRegGroup *reg_grp_entry = NULL;
135     XenPTReg *reg_entry = NULL;
136     int rc = 0;
137     int emul_len = 0;
138     uint32_t find_addr = addr;
139 
140     if (xen_pt_pci_config_access_check(d, addr, len)) {
141         goto exit;
142     }
143 
144     /* find register group entry */
145     reg_grp_entry = xen_pt_find_reg_grp(s, addr);
146     if (reg_grp_entry) {
147         /* check 0-Hardwired register group */
148         if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
149             /* no need to emulate, just return 0 */
150             val = 0;
151             goto exit;
152         }
153     }
154 
155     /* read I/O device register value */
156     rc = xen_host_pci_get_block(&s->real_device, addr, (uint8_t *)&val, len);
157     if (rc < 0) {
158         XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
159         memset(&val, 0xff, len);
160     }
161 
162     /* just return the I/O device register value for
163      * passthrough type register group */
164     if (reg_grp_entry == NULL) {
165         goto exit;
166     }
167 
168     /* adjust the read value to appropriate CFC-CFF window */
169     val <<= (addr & 3) << 3;
170     emul_len = len;
171 
172     /* loop around the guest requested size */
173     while (emul_len > 0) {
174         /* find register entry to be emulated */
175         reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
176         if (reg_entry) {
177             XenPTRegInfo *reg = reg_entry->reg;
178             uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
179             uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
180             uint8_t *ptr_val = NULL;
181 
182             valid_mask <<= (find_addr - real_offset) << 3;
183             ptr_val = (uint8_t *)&val + (real_offset & 3);
184 
185             /* do emulation based on register size */
186             switch (reg->size) {
187             case 1:
188                 if (reg->u.b.read) {
189                     rc = reg->u.b.read(s, reg_entry, ptr_val, valid_mask);
190                 }
191                 break;
192             case 2:
193                 if (reg->u.w.read) {
194                     rc = reg->u.w.read(s, reg_entry,
195                                        (uint16_t *)ptr_val, valid_mask);
196                 }
197                 break;
198             case 4:
199                 if (reg->u.dw.read) {
200                     rc = reg->u.dw.read(s, reg_entry,
201                                         (uint32_t *)ptr_val, valid_mask);
202                 }
203                 break;
204             }
205 
206             if (rc < 0) {
207                 xen_shutdown_fatal_error("Internal error: Invalid read "
208                                          "emulation. (%s, rc: %d)\n",
209                                          __func__, rc);
210                 return 0;
211             }
212 
213             /* calculate next address to find */
214             emul_len -= reg->size;
215             if (emul_len > 0) {
216                 find_addr = real_offset + reg->size;
217             }
218         } else {
219             /* nothing to do with passthrough type register,
220              * continue to find next byte */
221             emul_len--;
222             find_addr++;
223         }
224     }
225 
226     /* need to shift back before returning them to pci bus emulator */
227     val >>= ((addr & 3) << 3);
228 
229 exit:
230     XEN_PT_LOG_CONFIG(d, addr, val, len);
231     return val;
232 }
233 
234 static void xen_pt_pci_write_config(PCIDevice *d, uint32_t addr,
235                                     uint32_t val, int len)
236 {
237     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
238     int index = 0;
239     XenPTRegGroup *reg_grp_entry = NULL;
240     int rc = 0;
241     uint32_t read_val = 0, wb_mask;
242     int emul_len = 0;
243     XenPTReg *reg_entry = NULL;
244     uint32_t find_addr = addr;
245     XenPTRegInfo *reg = NULL;
246     bool wp_flag = false;
247 
248     if (xen_pt_pci_config_access_check(d, addr, len)) {
249         return;
250     }
251 
252     XEN_PT_LOG_CONFIG(d, addr, val, len);
253 
254     /* check unused BAR register */
255     index = xen_pt_bar_offset_to_index(addr);
256     if ((index >= 0) && (val != 0)) {
257         uint32_t chk = val;
258 
259         if (index == PCI_ROM_SLOT)
260             chk |= (uint32_t)~PCI_ROM_ADDRESS_MASK;
261 
262         if ((chk != XEN_PT_BAR_ALLF) &&
263             (s->bases[index].bar_flag == XEN_PT_BAR_FLAG_UNUSED)) {
264             XEN_PT_WARN(d, "Guest attempt to set address to unused "
265                         "Base Address Register. (addr: 0x%02x, len: %d)\n",
266                         addr, len);
267         }
268     }
269 
270     /* find register group entry */
271     reg_grp_entry = xen_pt_find_reg_grp(s, addr);
272     if (reg_grp_entry) {
273         /* check 0-Hardwired register group */
274         if (reg_grp_entry->reg_grp->grp_type == XEN_PT_GRP_TYPE_HARDWIRED) {
275             /* ignore silently */
276             XEN_PT_WARN(d, "Access to 0-Hardwired register. "
277                         "(addr: 0x%02x, len: %d)\n", addr, len);
278             return;
279         }
280     }
281 
282     rc = xen_host_pci_get_block(&s->real_device, addr,
283                                 (uint8_t *)&read_val, len);
284     if (rc < 0) {
285         XEN_PT_ERR(d, "pci_read_block failed. return value: %d.\n", rc);
286         memset(&read_val, 0xff, len);
287         wb_mask = 0;
288     } else {
289         wb_mask = 0xFFFFFFFF >> ((4 - len) << 3);
290     }
291 
292     /* pass directly to the real device for passthrough type register group */
293     if (reg_grp_entry == NULL) {
294         if (!s->permissive) {
295             wb_mask = 0;
296             wp_flag = true;
297         }
298         goto out;
299     }
300 
301     memory_region_transaction_begin();
302     pci_default_write_config(d, addr, val, len);
303 
304     /* adjust the read and write value to appropriate CFC-CFF window */
305     read_val <<= (addr & 3) << 3;
306     val <<= (addr & 3) << 3;
307     emul_len = len;
308 
309     /* loop around the guest requested size */
310     while (emul_len > 0) {
311         /* find register entry to be emulated */
312         reg_entry = xen_pt_find_reg(reg_grp_entry, find_addr);
313         if (reg_entry) {
314             reg = reg_entry->reg;
315             uint32_t real_offset = reg_grp_entry->base_offset + reg->offset;
316             uint32_t valid_mask = 0xFFFFFFFF >> ((4 - emul_len) << 3);
317             uint8_t *ptr_val = NULL;
318             uint32_t wp_mask = reg->emu_mask | reg->ro_mask;
319 
320             valid_mask <<= (find_addr - real_offset) << 3;
321             ptr_val = (uint8_t *)&val + (real_offset & 3);
322             if (!s->permissive) {
323                 wp_mask |= reg->res_mask;
324             }
325             if (wp_mask == (0xFFFFFFFF >> ((4 - reg->size) << 3))) {
326                 wb_mask &= ~((wp_mask >> ((find_addr - real_offset) << 3))
327                              << ((len - emul_len) << 3));
328             }
329 
330             /* do emulation based on register size */
331             switch (reg->size) {
332             case 1:
333                 if (reg->u.b.write) {
334                     rc = reg->u.b.write(s, reg_entry, ptr_val,
335                                         read_val >> ((real_offset & 3) << 3),
336                                         valid_mask);
337                 }
338                 break;
339             case 2:
340                 if (reg->u.w.write) {
341                     rc = reg->u.w.write(s, reg_entry, (uint16_t *)ptr_val,
342                                         (read_val >> ((real_offset & 3) << 3)),
343                                         valid_mask);
344                 }
345                 break;
346             case 4:
347                 if (reg->u.dw.write) {
348                     rc = reg->u.dw.write(s, reg_entry, (uint32_t *)ptr_val,
349                                          (read_val >> ((real_offset & 3) << 3)),
350                                          valid_mask);
351                 }
352                 break;
353             }
354 
355             if (rc < 0) {
356                 xen_shutdown_fatal_error("Internal error: Invalid write"
357                                          " emulation. (%s, rc: %d)\n",
358                                          __func__, rc);
359                 return;
360             }
361 
362             /* calculate next address to find */
363             emul_len -= reg->size;
364             if (emul_len > 0) {
365                 find_addr = real_offset + reg->size;
366             }
367         } else {
368             /* nothing to do with passthrough type register,
369              * continue to find next byte */
370             if (!s->permissive) {
371                 wb_mask &= ~(0xff << ((len - emul_len) << 3));
372                 /* Unused BARs will make it here, but we don't want to issue
373                  * warnings for writes to them (bogus writes get dealt with
374                  * above).
375                  */
376                 if (index < 0) {
377                     wp_flag = true;
378                 }
379             }
380             emul_len--;
381             find_addr++;
382         }
383     }
384 
385     /* need to shift back before passing them to xen_host_pci_set_block. */
386     val >>= (addr & 3) << 3;
387 
388     memory_region_transaction_commit();
389 
390 out:
391     if (wp_flag && !s->permissive_warned) {
392         s->permissive_warned = true;
393         xen_pt_log(d, "Write-back to unknown field 0x%02x (partially) inhibited (0x%0*x)\n",
394                    addr, len * 2, wb_mask);
395         xen_pt_log(d, "If the device doesn't work, try enabling permissive mode\n");
396         xen_pt_log(d, "(unsafe) and if it helps report the problem to xen-devel\n");
397     }
398     for (index = 0; wb_mask; index += len) {
399         /* unknown regs are passed through */
400         while (!(wb_mask & 0xff)) {
401             index++;
402             wb_mask >>= 8;
403         }
404         len = 0;
405         do {
406             len++;
407             wb_mask >>= 8;
408         } while (wb_mask & 0xff);
409         rc = xen_host_pci_set_block(&s->real_device, addr + index,
410                                     (uint8_t *)&val + index, len);
411 
412         if (rc < 0) {
413             XEN_PT_ERR(d, "xen_host_pci_set_block failed. return value: %d.\n", rc);
414         }
415     }
416 }
417 
418 /* register regions */
419 
420 static uint64_t xen_pt_bar_read(void *o, hwaddr addr,
421                                 unsigned size)
422 {
423     PCIDevice *d = o;
424     /* if this function is called, that probably means that there is a
425      * misconfiguration of the IOMMU. */
426     XEN_PT_ERR(d, "Should not read BAR through QEMU. @0x"TARGET_FMT_plx"\n",
427                addr);
428     return 0;
429 }
430 static void xen_pt_bar_write(void *o, hwaddr addr, uint64_t val,
431                              unsigned size)
432 {
433     PCIDevice *d = o;
434     /* Same comment as xen_pt_bar_read function */
435     XEN_PT_ERR(d, "Should not write BAR through QEMU. @0x"TARGET_FMT_plx"\n",
436                addr);
437 }
438 
439 static const MemoryRegionOps ops = {
440     .endianness = DEVICE_NATIVE_ENDIAN,
441     .read = xen_pt_bar_read,
442     .write = xen_pt_bar_write,
443 };
444 
445 static int xen_pt_register_regions(XenPCIPassthroughState *s, uint16_t *cmd)
446 {
447     int i = 0;
448     XenHostPCIDevice *d = &s->real_device;
449 
450     /* Register PIO/MMIO BARs */
451     for (i = 0; i < PCI_ROM_SLOT; i++) {
452         XenHostPCIIORegion *r = &d->io_regions[i];
453         uint8_t type;
454 
455         if (r->base_addr == 0 || r->size == 0) {
456             continue;
457         }
458 
459         s->bases[i].access.u = r->base_addr;
460 
461         if (r->type & XEN_HOST_PCI_REGION_TYPE_IO) {
462             type = PCI_BASE_ADDRESS_SPACE_IO;
463             *cmd |= PCI_COMMAND_IO;
464         } else {
465             type = PCI_BASE_ADDRESS_SPACE_MEMORY;
466             if (r->type & XEN_HOST_PCI_REGION_TYPE_PREFETCH) {
467                 type |= PCI_BASE_ADDRESS_MEM_PREFETCH;
468             }
469             if (r->type & XEN_HOST_PCI_REGION_TYPE_MEM_64) {
470                 type |= PCI_BASE_ADDRESS_MEM_TYPE_64;
471             }
472             *cmd |= PCI_COMMAND_MEMORY;
473         }
474 
475         memory_region_init_io(&s->bar[i], OBJECT(s), &ops, &s->dev,
476                               "xen-pci-pt-bar", r->size);
477         pci_register_bar(&s->dev, i, type, &s->bar[i]);
478 
479         XEN_PT_LOG(&s->dev, "IO region %i registered (size=0x%08"PRIx64
480                    " base_addr=0x%08"PRIx64" type: %#x)\n",
481                    i, r->size, r->base_addr, type);
482     }
483 
484     /* Register expansion ROM address */
485     if (d->rom.base_addr && d->rom.size) {
486         uint32_t bar_data = 0;
487 
488         /* Re-set BAR reported by OS, otherwise ROM can't be read. */
489         if (xen_host_pci_get_long(d, PCI_ROM_ADDRESS, &bar_data)) {
490             return 0;
491         }
492         if ((bar_data & PCI_ROM_ADDRESS_MASK) == 0) {
493             bar_data |= d->rom.base_addr & PCI_ROM_ADDRESS_MASK;
494             xen_host_pci_set_long(d, PCI_ROM_ADDRESS, bar_data);
495         }
496 
497         s->bases[PCI_ROM_SLOT].access.maddr = d->rom.base_addr;
498 
499         memory_region_init_io(&s->rom, OBJECT(s), &ops, &s->dev,
500                               "xen-pci-pt-rom", d->rom.size);
501         pci_register_bar(&s->dev, PCI_ROM_SLOT, PCI_BASE_ADDRESS_MEM_PREFETCH,
502                          &s->rom);
503 
504         XEN_PT_LOG(&s->dev, "Expansion ROM registered (size=0x%08"PRIx64
505                    " base_addr=0x%08"PRIx64")\n",
506                    d->rom.size, d->rom.base_addr);
507     }
508 
509     xen_pt_register_vga_regions(d);
510     return 0;
511 }
512 
513 /* region mapping */
514 
515 static int xen_pt_bar_from_region(XenPCIPassthroughState *s, MemoryRegion *mr)
516 {
517     int i = 0;
518 
519     for (i = 0; i < PCI_NUM_REGIONS - 1; i++) {
520         if (mr == &s->bar[i]) {
521             return i;
522         }
523     }
524     if (mr == &s->rom) {
525         return PCI_ROM_SLOT;
526     }
527     return -1;
528 }
529 
530 /*
531  * This function checks if an io_region overlaps an io_region from another
532  * device.  The io_region to check is provided with (addr, size and type)
533  * A callback can be provided and will be called for every region that is
534  * overlapped.
535  * The return value indicates if the region is overlappsed */
536 struct CheckBarArgs {
537     XenPCIPassthroughState *s;
538     pcibus_t addr;
539     pcibus_t size;
540     uint8_t type;
541     bool rc;
542 };
543 static void xen_pt_check_bar_overlap(PCIBus *bus, PCIDevice *d, void *opaque)
544 {
545     struct CheckBarArgs *arg = opaque;
546     XenPCIPassthroughState *s = arg->s;
547     uint8_t type = arg->type;
548     int i;
549 
550     if (d->devfn == s->dev.devfn) {
551         return;
552     }
553 
554     /* xxx: This ignores bridges. */
555     for (i = 0; i < PCI_NUM_REGIONS; i++) {
556         const PCIIORegion *r = &d->io_regions[i];
557 
558         if (!r->size) {
559             continue;
560         }
561         if ((type & PCI_BASE_ADDRESS_SPACE_IO)
562             != (r->type & PCI_BASE_ADDRESS_SPACE_IO)) {
563             continue;
564         }
565 
566         if (ranges_overlap(arg->addr, arg->size, r->addr, r->size)) {
567             XEN_PT_WARN(&s->dev,
568                         "Overlapped to device [%02x:%02x.%d] Region: %i"
569                         " (addr: %#"FMT_PCIBUS", len: %#"FMT_PCIBUS")\n",
570                         pci_bus_num(bus), PCI_SLOT(d->devfn),
571                         PCI_FUNC(d->devfn), i, r->addr, r->size);
572             arg->rc = true;
573         }
574     }
575 }
576 
577 static void xen_pt_region_update(XenPCIPassthroughState *s,
578                                  MemoryRegionSection *sec, bool adding)
579 {
580     PCIDevice *d = &s->dev;
581     MemoryRegion *mr = sec->mr;
582     int bar = -1;
583     int rc;
584     int op = adding ? DPCI_ADD_MAPPING : DPCI_REMOVE_MAPPING;
585     struct CheckBarArgs args = {
586         .s = s,
587         .addr = sec->offset_within_address_space,
588         .size = int128_get64(sec->size),
589         .rc = false,
590     };
591 
592     bar = xen_pt_bar_from_region(s, mr);
593     if (bar == -1 && (!s->msix || &s->msix->mmio != mr)) {
594         return;
595     }
596 
597     if (s->msix && &s->msix->mmio == mr) {
598         if (adding) {
599             s->msix->mmio_base_addr = sec->offset_within_address_space;
600             rc = xen_pt_msix_update_remap(s, s->msix->bar_index);
601         }
602         return;
603     }
604 
605     args.type = d->io_regions[bar].type;
606     pci_for_each_device(pci_get_bus(d), pci_dev_bus_num(d),
607                         xen_pt_check_bar_overlap, &args);
608     if (args.rc) {
609         XEN_PT_WARN(d, "Region: %d (addr: %#"FMT_PCIBUS
610                     ", len: %#"FMT_PCIBUS") is overlapped.\n",
611                     bar, sec->offset_within_address_space,
612                     int128_get64(sec->size));
613     }
614 
615     if (d->io_regions[bar].type & PCI_BASE_ADDRESS_SPACE_IO) {
616         uint32_t guest_port = sec->offset_within_address_space;
617         uint32_t machine_port = s->bases[bar].access.pio_base;
618         uint32_t size = int128_get64(sec->size);
619         rc = xc_domain_ioport_mapping(xen_xc, xen_domid,
620                                       guest_port, machine_port, size,
621                                       op);
622         if (rc) {
623             XEN_PT_ERR(d, "%s ioport mapping failed! (err: %i)\n",
624                        adding ? "create new" : "remove old", errno);
625         }
626     } else {
627         pcibus_t guest_addr = sec->offset_within_address_space;
628         pcibus_t machine_addr = s->bases[bar].access.maddr
629             + sec->offset_within_region;
630         pcibus_t size = int128_get64(sec->size);
631         rc = xc_domain_memory_mapping(xen_xc, xen_domid,
632                                       XEN_PFN(guest_addr + XC_PAGE_SIZE - 1),
633                                       XEN_PFN(machine_addr + XC_PAGE_SIZE - 1),
634                                       XEN_PFN(size + XC_PAGE_SIZE - 1),
635                                       op);
636         if (rc) {
637             XEN_PT_ERR(d, "%s mem mapping failed! (err: %i)\n",
638                        adding ? "create new" : "remove old", errno);
639         }
640     }
641 }
642 
643 static void xen_pt_region_add(MemoryListener *l, MemoryRegionSection *sec)
644 {
645     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
646                                              memory_listener);
647 
648     memory_region_ref(sec->mr);
649     xen_pt_region_update(s, sec, true);
650 }
651 
652 static void xen_pt_region_del(MemoryListener *l, MemoryRegionSection *sec)
653 {
654     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
655                                              memory_listener);
656 
657     xen_pt_region_update(s, sec, false);
658     memory_region_unref(sec->mr);
659 }
660 
661 static void xen_pt_io_region_add(MemoryListener *l, MemoryRegionSection *sec)
662 {
663     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
664                                              io_listener);
665 
666     memory_region_ref(sec->mr);
667     xen_pt_region_update(s, sec, true);
668 }
669 
670 static void xen_pt_io_region_del(MemoryListener *l, MemoryRegionSection *sec)
671 {
672     XenPCIPassthroughState *s = container_of(l, XenPCIPassthroughState,
673                                              io_listener);
674 
675     xen_pt_region_update(s, sec, false);
676     memory_region_unref(sec->mr);
677 }
678 
679 static const MemoryListener xen_pt_memory_listener = {
680     .region_add = xen_pt_region_add,
681     .region_del = xen_pt_region_del,
682     .priority = 10,
683 };
684 
685 static const MemoryListener xen_pt_io_listener = {
686     .region_add = xen_pt_io_region_add,
687     .region_del = xen_pt_io_region_del,
688     .priority = 10,
689 };
690 
691 static void
692 xen_igd_passthrough_isa_bridge_create(XenPCIPassthroughState *s,
693                                       XenHostPCIDevice *dev)
694 {
695     uint16_t gpu_dev_id;
696     PCIDevice *d = &s->dev;
697 
698     gpu_dev_id = dev->device_id;
699     igd_passthrough_isa_bridge_create(pci_get_bus(d), gpu_dev_id);
700 }
701 
702 /* destroy. */
703 static void xen_pt_destroy(PCIDevice *d) {
704 
705     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
706     XenHostPCIDevice *host_dev = &s->real_device;
707     uint8_t machine_irq = s->machine_irq;
708     uint8_t intx;
709     int rc;
710 
711     if (machine_irq && !xen_host_pci_device_closed(&s->real_device)) {
712         intx = xen_pt_pci_intx(s);
713         rc = xc_domain_unbind_pt_irq(xen_xc, xen_domid, machine_irq,
714                                      PT_IRQ_TYPE_PCI,
715                                      pci_dev_bus_num(d),
716                                      PCI_SLOT(s->dev.devfn),
717                                      intx,
718                                      0 /* isa_irq */);
719         if (rc < 0) {
720             XEN_PT_ERR(d, "unbinding of interrupt INT%c failed."
721                        " (machine irq: %i, err: %d)"
722                        " But bravely continuing on..\n",
723                        'a' + intx, machine_irq, errno);
724         }
725     }
726 
727     /* N.B. xen_pt_config_delete takes care of freeing them. */
728     if (s->msi) {
729         xen_pt_msi_disable(s);
730     }
731     if (s->msix) {
732         xen_pt_msix_disable(s);
733     }
734 
735     if (machine_irq) {
736         xen_pt_mapped_machine_irq[machine_irq]--;
737 
738         if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
739             rc = xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq);
740 
741             if (rc < 0) {
742                 XEN_PT_ERR(d, "unmapping of interrupt %i failed. (err: %d)"
743                            " But bravely continuing on..\n",
744                            machine_irq, errno);
745             }
746         }
747         s->machine_irq = 0;
748     }
749 
750     /* delete all emulated config registers */
751     xen_pt_config_delete(s);
752 
753     xen_pt_unregister_vga_regions(host_dev);
754 
755     if (s->listener_set) {
756         memory_listener_unregister(&s->memory_listener);
757         memory_listener_unregister(&s->io_listener);
758         s->listener_set = false;
759     }
760     if (!xen_host_pci_device_closed(&s->real_device)) {
761         xen_host_pci_device_put(&s->real_device);
762     }
763 }
764 /* init */
765 
766 static void xen_pt_realize(PCIDevice *d, Error **errp)
767 {
768     XenPCIPassthroughState *s = XEN_PT_DEVICE(d);
769     int i, rc = 0;
770     uint8_t machine_irq = 0, scratch;
771     uint16_t cmd = 0;
772     int pirq = XEN_PT_UNASSIGNED_PIRQ;
773     Error *err = NULL;
774 
775     /* register real device */
776     XEN_PT_LOG(d, "Assigning real physical device %02x:%02x.%d"
777                " to devfn %#x\n",
778                s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function,
779                s->dev.devfn);
780 
781     xen_host_pci_device_get(&s->real_device,
782                             s->hostaddr.domain, s->hostaddr.bus,
783                             s->hostaddr.slot, s->hostaddr.function,
784                             &err);
785     if (err) {
786         error_append_hint(&err, "Failed to \"open\" the real pci device");
787         error_propagate(errp, err);
788         return;
789     }
790 
791     s->is_virtfn = s->real_device.is_virtfn;
792     if (s->is_virtfn) {
793         XEN_PT_LOG(d, "%04x:%02x:%02x.%d is a SR-IOV Virtual Function\n",
794                    s->real_device.domain, s->real_device.bus,
795                    s->real_device.dev, s->real_device.func);
796     }
797 
798     /* Initialize virtualized PCI configuration (Extended 256 Bytes) */
799     memset(d->config, 0, PCI_CONFIG_SPACE_SIZE);
800 
801     s->memory_listener = xen_pt_memory_listener;
802     s->io_listener = xen_pt_io_listener;
803 
804     /* Setup VGA bios for passthrough GFX */
805     if ((s->real_device.domain == 0) && (s->real_device.bus == 0) &&
806         (s->real_device.dev == 2) && (s->real_device.func == 0)) {
807         if (!is_igd_vga_passthrough(&s->real_device)) {
808             error_setg(errp, "Need to enable igd-passthru if you're trying"
809                     " to passthrough IGD GFX");
810             xen_host_pci_device_put(&s->real_device);
811             return;
812         }
813 
814         xen_pt_setup_vga(s, &s->real_device, &err);
815         if (err) {
816             error_append_hint(&err, "Setup VGA BIOS of passthrough"
817                     " GFX failed");
818             error_propagate(errp, err);
819             xen_host_pci_device_put(&s->real_device);
820             return;
821         }
822 
823         /* Register ISA bridge for passthrough GFX. */
824         xen_igd_passthrough_isa_bridge_create(s, &s->real_device);
825     }
826 
827     /* Handle real device's MMIO/PIO BARs */
828     xen_pt_register_regions(s, &cmd);
829 
830     /* reinitialize each config register to be emulated */
831     xen_pt_config_init(s, &err);
832     if (err) {
833         error_append_hint(&err, "PCI Config space initialisation failed");
834         error_propagate(errp, err);
835         rc = -1;
836         goto err_out;
837     }
838 
839     /* Bind interrupt */
840     rc = xen_host_pci_get_byte(&s->real_device, PCI_INTERRUPT_PIN, &scratch);
841     if (rc) {
842         error_setg_errno(errp, errno, "Failed to read PCI_INTERRUPT_PIN");
843         goto err_out;
844     }
845     if (!scratch) {
846         XEN_PT_LOG(d, "no pin interrupt\n");
847         goto out;
848     }
849 
850     machine_irq = s->real_device.irq;
851     if (machine_irq == 0) {
852         XEN_PT_LOG(d, "machine irq is 0\n");
853         cmd |= PCI_COMMAND_INTX_DISABLE;
854         goto out;
855     }
856 
857     rc = xc_physdev_map_pirq(xen_xc, xen_domid, machine_irq, &pirq);
858     if (rc < 0) {
859         error_setg_errno(errp, errno, "Mapping machine irq %u to"
860                          " pirq %i failed", machine_irq, pirq);
861 
862         /* Disable PCI intx assertion (turn on bit10 of devctl) */
863         cmd |= PCI_COMMAND_INTX_DISABLE;
864         machine_irq = 0;
865         s->machine_irq = 0;
866     } else {
867         machine_irq = pirq;
868         s->machine_irq = pirq;
869         xen_pt_mapped_machine_irq[machine_irq]++;
870     }
871 
872     /* bind machine_irq to device */
873     if (machine_irq != 0) {
874         uint8_t e_intx = xen_pt_pci_intx(s);
875 
876         rc = xc_domain_bind_pt_pci_irq(xen_xc, xen_domid, machine_irq,
877                                        pci_dev_bus_num(d),
878                                        PCI_SLOT(d->devfn),
879                                        e_intx);
880         if (rc < 0) {
881             error_setg_errno(errp, errno, "Binding of interrupt %u failed",
882                              e_intx);
883 
884             /* Disable PCI intx assertion (turn on bit10 of devctl) */
885             cmd |= PCI_COMMAND_INTX_DISABLE;
886             xen_pt_mapped_machine_irq[machine_irq]--;
887 
888             if (xen_pt_mapped_machine_irq[machine_irq] == 0) {
889                 if (xc_physdev_unmap_pirq(xen_xc, xen_domid, machine_irq)) {
890                     error_setg_errno(errp, errno, "Unmapping of machine"
891                             " interrupt %u failed", machine_irq);
892                 }
893             }
894             s->machine_irq = 0;
895         }
896     }
897 
898 out:
899     if (cmd) {
900         uint16_t val;
901 
902         rc = xen_host_pci_get_word(&s->real_device, PCI_COMMAND, &val);
903         if (rc) {
904             error_setg_errno(errp, errno, "Failed to read PCI_COMMAND");
905             goto err_out;
906         } else {
907             val |= cmd;
908             rc = xen_host_pci_set_word(&s->real_device, PCI_COMMAND, val);
909             if (rc) {
910                 error_setg_errno(errp, errno, "Failed to write PCI_COMMAND"
911                                  " val = 0x%x", val);
912                 goto err_out;
913             }
914         }
915     }
916 
917     memory_listener_register(&s->memory_listener, &address_space_memory);
918     memory_listener_register(&s->io_listener, &address_space_io);
919     s->listener_set = true;
920     XEN_PT_LOG(d,
921                "Real physical device %02x:%02x.%d registered successfully\n",
922                s->hostaddr.bus, s->hostaddr.slot, s->hostaddr.function);
923 
924     return;
925 
926 err_out:
927     for (i = 0; i < PCI_ROM_SLOT; i++) {
928         object_unparent(OBJECT(&s->bar[i]));
929     }
930     object_unparent(OBJECT(&s->rom));
931 
932     xen_pt_destroy(d);
933     assert(rc);
934 }
935 
936 static void xen_pt_unregister_device(PCIDevice *d)
937 {
938     xen_pt_destroy(d);
939 }
940 
941 static Property xen_pci_passthrough_properties[] = {
942     DEFINE_PROP_PCI_HOST_DEVADDR("hostaddr", XenPCIPassthroughState, hostaddr),
943     DEFINE_PROP_BOOL("permissive", XenPCIPassthroughState, permissive, false),
944     DEFINE_PROP_END_OF_LIST(),
945 };
946 
947 static void xen_pci_passthrough_instance_init(Object *obj)
948 {
949     /* QEMU_PCI_CAP_EXPRESS initialization does not depend on QEMU command
950      * line, therefore, no need to wait to realize like other devices */
951     PCI_DEVICE(obj)->cap_present |= QEMU_PCI_CAP_EXPRESS;
952 }
953 
954 static void xen_pci_passthrough_class_init(ObjectClass *klass, void *data)
955 {
956     DeviceClass *dc = DEVICE_CLASS(klass);
957     PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
958 
959     k->realize = xen_pt_realize;
960     k->exit = xen_pt_unregister_device;
961     k->config_read = xen_pt_pci_read_config;
962     k->config_write = xen_pt_pci_write_config;
963     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
964     dc->desc = "Assign an host PCI device with Xen";
965     dc->props = xen_pci_passthrough_properties;
966 };
967 
968 static void xen_pci_passthrough_finalize(Object *obj)
969 {
970     XenPCIPassthroughState *s = XEN_PT_DEVICE(obj);
971 
972     xen_pt_msix_delete(s);
973 }
974 
975 static const TypeInfo xen_pci_passthrough_info = {
976     .name = TYPE_XEN_PT_DEVICE,
977     .parent = TYPE_PCI_DEVICE,
978     .instance_size = sizeof(XenPCIPassthroughState),
979     .instance_finalize = xen_pci_passthrough_finalize,
980     .class_init = xen_pci_passthrough_class_init,
981     .instance_init = xen_pci_passthrough_instance_init,
982     .interfaces = (InterfaceInfo[]) {
983         { INTERFACE_CONVENTIONAL_PCI_DEVICE },
984         { INTERFACE_PCIE_DEVICE },
985         { },
986     },
987 };
988 
989 static void xen_pci_passthrough_register_types(void)
990 {
991     type_register_static(&xen_pci_passthrough_info);
992 }
993 
994 type_init(xen_pci_passthrough_register_types)
995