xref: /qemu/hw/vfio/common.c (revision 33848cee)
1 /*
2  * generic functions used by VFIO devices
3  *
4  * Copyright Red Hat, Inc. 2012
5  *
6  * Authors:
7  *  Alex Williamson <alex.williamson@redhat.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  * Based on qemu-kvm device-assignment:
13  *  Adapted for KVM by Qumranet.
14  *  Copyright (c) 2007, Neocleus, Alex Novik (alex@neocleus.com)
15  *  Copyright (c) 2007, Neocleus, Guy Zana (guy@neocleus.com)
16  *  Copyright (C) 2008, Qumranet, Amit Shah (amit.shah@qumranet.com)
17  *  Copyright (C) 2008, Red Hat, Amit Shah (amit.shah@redhat.com)
18  *  Copyright (C) 2008, IBM, Muli Ben-Yehuda (muli@il.ibm.com)
19  */
20 
21 #include "qemu/osdep.h"
22 #include <sys/ioctl.h>
23 #ifdef CONFIG_KVM
24 #include <linux/kvm.h>
25 #endif
26 #include <linux/vfio.h>
27 
28 #include "hw/vfio/vfio-common.h"
29 #include "hw/vfio/vfio.h"
30 #include "exec/address-spaces.h"
31 #include "exec/memory.h"
32 #include "hw/hw.h"
33 #include "qemu/error-report.h"
34 #include "qemu/range.h"
35 #include "sysemu/kvm.h"
36 #include "trace.h"
37 #include "qapi/error.h"
38 
39 struct vfio_group_head vfio_group_list =
40     QLIST_HEAD_INITIALIZER(vfio_group_list);
41 struct vfio_as_head vfio_address_spaces =
42     QLIST_HEAD_INITIALIZER(vfio_address_spaces);
43 
44 #ifdef CONFIG_KVM
45 /*
46  * We have a single VFIO pseudo device per KVM VM.  Once created it lives
47  * for the life of the VM.  Closing the file descriptor only drops our
48  * reference to it and the device's reference to kvm.  Therefore once
49  * initialized, this file descriptor is only released on QEMU exit and
50  * we'll re-use it should another vfio device be attached before then.
51  */
52 static int vfio_kvm_device_fd = -1;
53 #endif
54 
55 /*
56  * Common VFIO interrupt disable
57  */
58 void vfio_disable_irqindex(VFIODevice *vbasedev, int index)
59 {
60     struct vfio_irq_set irq_set = {
61         .argsz = sizeof(irq_set),
62         .flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER,
63         .index = index,
64         .start = 0,
65         .count = 0,
66     };
67 
68     ioctl(vbasedev->fd, VFIO_DEVICE_SET_IRQS, &irq_set);
69 }
70 
71 void vfio_unmask_single_irqindex(VFIODevice *vbasedev, int index)
72 {
73     struct vfio_irq_set irq_set = {
74         .argsz = sizeof(irq_set),
75         .flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_UNMASK,
76         .index = index,
77         .start = 0,
78         .count = 1,
79     };
80 
81     ioctl(vbasedev->fd, VFIO_DEVICE_SET_IRQS, &irq_set);
82 }
83 
84 void vfio_mask_single_irqindex(VFIODevice *vbasedev, int index)
85 {
86     struct vfio_irq_set irq_set = {
87         .argsz = sizeof(irq_set),
88         .flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_MASK,
89         .index = index,
90         .start = 0,
91         .count = 1,
92     };
93 
94     ioctl(vbasedev->fd, VFIO_DEVICE_SET_IRQS, &irq_set);
95 }
96 
97 /*
98  * IO Port/MMIO - Beware of the endians, VFIO is always little endian
99  */
100 void vfio_region_write(void *opaque, hwaddr addr,
101                        uint64_t data, unsigned size)
102 {
103     VFIORegion *region = opaque;
104     VFIODevice *vbasedev = region->vbasedev;
105     union {
106         uint8_t byte;
107         uint16_t word;
108         uint32_t dword;
109         uint64_t qword;
110     } buf;
111 
112     switch (size) {
113     case 1:
114         buf.byte = data;
115         break;
116     case 2:
117         buf.word = cpu_to_le16(data);
118         break;
119     case 4:
120         buf.dword = cpu_to_le32(data);
121         break;
122     default:
123         hw_error("vfio: unsupported write size, %d bytes", size);
124         break;
125     }
126 
127     if (pwrite(vbasedev->fd, &buf, size, region->fd_offset + addr) != size) {
128         error_report("%s(%s:region%d+0x%"HWADDR_PRIx", 0x%"PRIx64
129                      ",%d) failed: %m",
130                      __func__, vbasedev->name, region->nr,
131                      addr, data, size);
132     }
133 
134     trace_vfio_region_write(vbasedev->name, region->nr, addr, data, size);
135 
136     /*
137      * A read or write to a BAR always signals an INTx EOI.  This will
138      * do nothing if not pending (including not in INTx mode).  We assume
139      * that a BAR access is in response to an interrupt and that BAR
140      * accesses will service the interrupt.  Unfortunately, we don't know
141      * which access will service the interrupt, so we're potentially
142      * getting quite a few host interrupts per guest interrupt.
143      */
144     vbasedev->ops->vfio_eoi(vbasedev);
145 }
146 
147 uint64_t vfio_region_read(void *opaque,
148                           hwaddr addr, unsigned size)
149 {
150     VFIORegion *region = opaque;
151     VFIODevice *vbasedev = region->vbasedev;
152     union {
153         uint8_t byte;
154         uint16_t word;
155         uint32_t dword;
156         uint64_t qword;
157     } buf;
158     uint64_t data = 0;
159 
160     if (pread(vbasedev->fd, &buf, size, region->fd_offset + addr) != size) {
161         error_report("%s(%s:region%d+0x%"HWADDR_PRIx", %d) failed: %m",
162                      __func__, vbasedev->name, region->nr,
163                      addr, size);
164         return (uint64_t)-1;
165     }
166     switch (size) {
167     case 1:
168         data = buf.byte;
169         break;
170     case 2:
171         data = le16_to_cpu(buf.word);
172         break;
173     case 4:
174         data = le32_to_cpu(buf.dword);
175         break;
176     default:
177         hw_error("vfio: unsupported read size, %d bytes", size);
178         break;
179     }
180 
181     trace_vfio_region_read(vbasedev->name, region->nr, addr, size, data);
182 
183     /* Same as write above */
184     vbasedev->ops->vfio_eoi(vbasedev);
185 
186     return data;
187 }
188 
189 const MemoryRegionOps vfio_region_ops = {
190     .read = vfio_region_read,
191     .write = vfio_region_write,
192     .endianness = DEVICE_LITTLE_ENDIAN,
193 };
194 
195 /*
196  * DMA - Mapping and unmapping for the "type1" IOMMU interface used on x86
197  */
198 static int vfio_dma_unmap(VFIOContainer *container,
199                           hwaddr iova, ram_addr_t size)
200 {
201     struct vfio_iommu_type1_dma_unmap unmap = {
202         .argsz = sizeof(unmap),
203         .flags = 0,
204         .iova = iova,
205         .size = size,
206     };
207 
208     if (ioctl(container->fd, VFIO_IOMMU_UNMAP_DMA, &unmap)) {
209         error_report("VFIO_UNMAP_DMA: %d", -errno);
210         return -errno;
211     }
212 
213     return 0;
214 }
215 
216 static int vfio_dma_map(VFIOContainer *container, hwaddr iova,
217                         ram_addr_t size, void *vaddr, bool readonly)
218 {
219     struct vfio_iommu_type1_dma_map map = {
220         .argsz = sizeof(map),
221         .flags = VFIO_DMA_MAP_FLAG_READ,
222         .vaddr = (__u64)(uintptr_t)vaddr,
223         .iova = iova,
224         .size = size,
225     };
226 
227     if (!readonly) {
228         map.flags |= VFIO_DMA_MAP_FLAG_WRITE;
229     }
230 
231     /*
232      * Try the mapping, if it fails with EBUSY, unmap the region and try
233      * again.  This shouldn't be necessary, but we sometimes see it in
234      * the VGA ROM space.
235      */
236     if (ioctl(container->fd, VFIO_IOMMU_MAP_DMA, &map) == 0 ||
237         (errno == EBUSY && vfio_dma_unmap(container, iova, size) == 0 &&
238          ioctl(container->fd, VFIO_IOMMU_MAP_DMA, &map) == 0)) {
239         return 0;
240     }
241 
242     error_report("VFIO_MAP_DMA: %d", -errno);
243     return -errno;
244 }
245 
246 static void vfio_host_win_add(VFIOContainer *container,
247                               hwaddr min_iova, hwaddr max_iova,
248                               uint64_t iova_pgsizes)
249 {
250     VFIOHostDMAWindow *hostwin;
251 
252     QLIST_FOREACH(hostwin, &container->hostwin_list, hostwin_next) {
253         if (ranges_overlap(hostwin->min_iova,
254                            hostwin->max_iova - hostwin->min_iova + 1,
255                            min_iova,
256                            max_iova - min_iova + 1)) {
257             hw_error("%s: Overlapped IOMMU are not enabled", __func__);
258         }
259     }
260 
261     hostwin = g_malloc0(sizeof(*hostwin));
262 
263     hostwin->min_iova = min_iova;
264     hostwin->max_iova = max_iova;
265     hostwin->iova_pgsizes = iova_pgsizes;
266     QLIST_INSERT_HEAD(&container->hostwin_list, hostwin, hostwin_next);
267 }
268 
269 static int vfio_host_win_del(VFIOContainer *container, hwaddr min_iova,
270                              hwaddr max_iova)
271 {
272     VFIOHostDMAWindow *hostwin;
273 
274     QLIST_FOREACH(hostwin, &container->hostwin_list, hostwin_next) {
275         if (hostwin->min_iova == min_iova && hostwin->max_iova == max_iova) {
276             QLIST_REMOVE(hostwin, hostwin_next);
277             return 0;
278         }
279     }
280 
281     return -1;
282 }
283 
284 static bool vfio_listener_skipped_section(MemoryRegionSection *section)
285 {
286     return (!memory_region_is_ram(section->mr) &&
287             !memory_region_is_iommu(section->mr)) ||
288            /*
289             * Sizing an enabled 64-bit BAR can cause spurious mappings to
290             * addresses in the upper part of the 64-bit address space.  These
291             * are never accessed by the CPU and beyond the address width of
292             * some IOMMU hardware.  TODO: VFIO should tell us the IOMMU width.
293             */
294            section->offset_within_address_space & (1ULL << 63);
295 }
296 
297 static void vfio_iommu_map_notify(IOMMUNotifier *n, IOMMUTLBEntry *iotlb)
298 {
299     VFIOGuestIOMMU *giommu = container_of(n, VFIOGuestIOMMU, n);
300     VFIOContainer *container = giommu->container;
301     hwaddr iova = iotlb->iova + giommu->iommu_offset;
302     MemoryRegion *mr;
303     hwaddr xlat;
304     hwaddr len = iotlb->addr_mask + 1;
305     void *vaddr;
306     int ret;
307 
308     trace_vfio_iommu_map_notify(iova, iova + iotlb->addr_mask);
309 
310     if (iotlb->target_as != &address_space_memory) {
311         error_report("Wrong target AS \"%s\", only system memory is allowed",
312                      iotlb->target_as->name ? iotlb->target_as->name : "none");
313         return;
314     }
315 
316     /*
317      * The IOMMU TLB entry we have just covers translation through
318      * this IOMMU to its immediate target.  We need to translate
319      * it the rest of the way through to memory.
320      */
321     rcu_read_lock();
322     mr = address_space_translate(&address_space_memory,
323                                  iotlb->translated_addr,
324                                  &xlat, &len, iotlb->perm & IOMMU_WO);
325     if (!memory_region_is_ram(mr)) {
326         error_report("iommu map to non memory area %"HWADDR_PRIx"",
327                      xlat);
328         goto out;
329     }
330     /*
331      * Translation truncates length to the IOMMU page size,
332      * check that it did not truncate too much.
333      */
334     if (len & iotlb->addr_mask) {
335         error_report("iommu has granularity incompatible with target AS");
336         goto out;
337     }
338 
339     if ((iotlb->perm & IOMMU_RW) != IOMMU_NONE) {
340         vaddr = memory_region_get_ram_ptr(mr) + xlat;
341         ret = vfio_dma_map(container, iova,
342                            iotlb->addr_mask + 1, vaddr,
343                            !(iotlb->perm & IOMMU_WO) || mr->readonly);
344         if (ret) {
345             error_report("vfio_dma_map(%p, 0x%"HWADDR_PRIx", "
346                          "0x%"HWADDR_PRIx", %p) = %d (%m)",
347                          container, iova,
348                          iotlb->addr_mask + 1, vaddr, ret);
349         }
350     } else {
351         ret = vfio_dma_unmap(container, iova, iotlb->addr_mask + 1);
352         if (ret) {
353             error_report("vfio_dma_unmap(%p, 0x%"HWADDR_PRIx", "
354                          "0x%"HWADDR_PRIx") = %d (%m)",
355                          container, iova,
356                          iotlb->addr_mask + 1, ret);
357         }
358     }
359 out:
360     rcu_read_unlock();
361 }
362 
363 static void vfio_listener_region_add(MemoryListener *listener,
364                                      MemoryRegionSection *section)
365 {
366     VFIOContainer *container = container_of(listener, VFIOContainer, listener);
367     hwaddr iova, end;
368     Int128 llend, llsize;
369     void *vaddr;
370     int ret;
371     VFIOHostDMAWindow *hostwin;
372     bool hostwin_found;
373 
374     if (vfio_listener_skipped_section(section)) {
375         trace_vfio_listener_region_add_skip(
376                 section->offset_within_address_space,
377                 section->offset_within_address_space +
378                 int128_get64(int128_sub(section->size, int128_one())));
379         return;
380     }
381 
382     if (unlikely((section->offset_within_address_space & ~TARGET_PAGE_MASK) !=
383                  (section->offset_within_region & ~TARGET_PAGE_MASK))) {
384         error_report("%s received unaligned region", __func__);
385         return;
386     }
387 
388     iova = TARGET_PAGE_ALIGN(section->offset_within_address_space);
389     llend = int128_make64(section->offset_within_address_space);
390     llend = int128_add(llend, section->size);
391     llend = int128_and(llend, int128_exts64(TARGET_PAGE_MASK));
392 
393     if (int128_ge(int128_make64(iova), llend)) {
394         return;
395     }
396     end = int128_get64(int128_sub(llend, int128_one()));
397 
398     if (container->iommu_type == VFIO_SPAPR_TCE_v2_IOMMU) {
399         VFIOHostDMAWindow *hostwin;
400         hwaddr pgsize = 0;
401 
402         /* For now intersections are not allowed, we may relax this later */
403         QLIST_FOREACH(hostwin, &container->hostwin_list, hostwin_next) {
404             if (ranges_overlap(hostwin->min_iova,
405                                hostwin->max_iova - hostwin->min_iova + 1,
406                                section->offset_within_address_space,
407                                int128_get64(section->size))) {
408                 ret = -1;
409                 goto fail;
410             }
411         }
412 
413         ret = vfio_spapr_create_window(container, section, &pgsize);
414         if (ret) {
415             goto fail;
416         }
417 
418         vfio_host_win_add(container, section->offset_within_address_space,
419                           section->offset_within_address_space +
420                           int128_get64(section->size) - 1, pgsize);
421     }
422 
423     hostwin_found = false;
424     QLIST_FOREACH(hostwin, &container->hostwin_list, hostwin_next) {
425         if (hostwin->min_iova <= iova && end <= hostwin->max_iova) {
426             hostwin_found = true;
427             break;
428         }
429     }
430 
431     if (!hostwin_found) {
432         error_report("vfio: IOMMU container %p can't map guest IOVA region"
433                      " 0x%"HWADDR_PRIx"..0x%"HWADDR_PRIx,
434                      container, iova, end);
435         ret = -EFAULT;
436         goto fail;
437     }
438 
439     memory_region_ref(section->mr);
440 
441     if (memory_region_is_iommu(section->mr)) {
442         VFIOGuestIOMMU *giommu;
443 
444         trace_vfio_listener_region_add_iommu(iova, end);
445         /*
446          * FIXME: For VFIO iommu types which have KVM acceleration to
447          * avoid bouncing all map/unmaps through qemu this way, this
448          * would be the right place to wire that up (tell the KVM
449          * device emulation the VFIO iommu handles to use).
450          */
451         giommu = g_malloc0(sizeof(*giommu));
452         giommu->iommu = section->mr;
453         giommu->iommu_offset = section->offset_within_address_space -
454                                section->offset_within_region;
455         giommu->container = container;
456         giommu->n.notify = vfio_iommu_map_notify;
457         giommu->n.notifier_flags = IOMMU_NOTIFIER_ALL;
458         QLIST_INSERT_HEAD(&container->giommu_list, giommu, giommu_next);
459 
460         memory_region_register_iommu_notifier(giommu->iommu, &giommu->n);
461         memory_region_iommu_replay(giommu->iommu, &giommu->n, false);
462 
463         return;
464     }
465 
466     /* Here we assume that memory_region_is_ram(section->mr)==true */
467 
468     vaddr = memory_region_get_ram_ptr(section->mr) +
469             section->offset_within_region +
470             (iova - section->offset_within_address_space);
471 
472     trace_vfio_listener_region_add_ram(iova, end, vaddr);
473 
474     llsize = int128_sub(llend, int128_make64(iova));
475 
476     ret = vfio_dma_map(container, iova, int128_get64(llsize),
477                        vaddr, section->readonly);
478     if (ret) {
479         error_report("vfio_dma_map(%p, 0x%"HWADDR_PRIx", "
480                      "0x%"HWADDR_PRIx", %p) = %d (%m)",
481                      container, iova, int128_get64(llsize), vaddr, ret);
482         goto fail;
483     }
484 
485     return;
486 
487 fail:
488     /*
489      * On the initfn path, store the first error in the container so we
490      * can gracefully fail.  Runtime, there's not much we can do other
491      * than throw a hardware error.
492      */
493     if (!container->initialized) {
494         if (!container->error) {
495             container->error = ret;
496         }
497     } else {
498         hw_error("vfio: DMA mapping failed, unable to continue");
499     }
500 }
501 
502 static void vfio_listener_region_del(MemoryListener *listener,
503                                      MemoryRegionSection *section)
504 {
505     VFIOContainer *container = container_of(listener, VFIOContainer, listener);
506     hwaddr iova, end;
507     Int128 llend, llsize;
508     int ret;
509 
510     if (vfio_listener_skipped_section(section)) {
511         trace_vfio_listener_region_del_skip(
512                 section->offset_within_address_space,
513                 section->offset_within_address_space +
514                 int128_get64(int128_sub(section->size, int128_one())));
515         return;
516     }
517 
518     if (unlikely((section->offset_within_address_space & ~TARGET_PAGE_MASK) !=
519                  (section->offset_within_region & ~TARGET_PAGE_MASK))) {
520         error_report("%s received unaligned region", __func__);
521         return;
522     }
523 
524     if (memory_region_is_iommu(section->mr)) {
525         VFIOGuestIOMMU *giommu;
526 
527         QLIST_FOREACH(giommu, &container->giommu_list, giommu_next) {
528             if (giommu->iommu == section->mr) {
529                 memory_region_unregister_iommu_notifier(giommu->iommu,
530                                                         &giommu->n);
531                 QLIST_REMOVE(giommu, giommu_next);
532                 g_free(giommu);
533                 break;
534             }
535         }
536 
537         /*
538          * FIXME: We assume the one big unmap below is adequate to
539          * remove any individual page mappings in the IOMMU which
540          * might have been copied into VFIO. This works for a page table
541          * based IOMMU where a big unmap flattens a large range of IO-PTEs.
542          * That may not be true for all IOMMU types.
543          */
544     }
545 
546     iova = TARGET_PAGE_ALIGN(section->offset_within_address_space);
547     llend = int128_make64(section->offset_within_address_space);
548     llend = int128_add(llend, section->size);
549     llend = int128_and(llend, int128_exts64(TARGET_PAGE_MASK));
550 
551     if (int128_ge(int128_make64(iova), llend)) {
552         return;
553     }
554     end = int128_get64(int128_sub(llend, int128_one()));
555 
556     llsize = int128_sub(llend, int128_make64(iova));
557 
558     trace_vfio_listener_region_del(iova, end);
559 
560     ret = vfio_dma_unmap(container, iova, int128_get64(llsize));
561     memory_region_unref(section->mr);
562     if (ret) {
563         error_report("vfio_dma_unmap(%p, 0x%"HWADDR_PRIx", "
564                      "0x%"HWADDR_PRIx") = %d (%m)",
565                      container, iova, int128_get64(llsize), ret);
566     }
567 
568     if (container->iommu_type == VFIO_SPAPR_TCE_v2_IOMMU) {
569         vfio_spapr_remove_window(container,
570                                  section->offset_within_address_space);
571         if (vfio_host_win_del(container,
572                               section->offset_within_address_space,
573                               section->offset_within_address_space +
574                               int128_get64(section->size) - 1) < 0) {
575             hw_error("%s: Cannot delete missing window at %"HWADDR_PRIx,
576                      __func__, section->offset_within_address_space);
577         }
578     }
579 }
580 
581 static const MemoryListener vfio_memory_listener = {
582     .region_add = vfio_listener_region_add,
583     .region_del = vfio_listener_region_del,
584 };
585 
586 static void vfio_listener_release(VFIOContainer *container)
587 {
588     memory_listener_unregister(&container->listener);
589     if (container->iommu_type == VFIO_SPAPR_TCE_v2_IOMMU) {
590         memory_listener_unregister(&container->prereg_listener);
591     }
592 }
593 
594 static struct vfio_info_cap_header *
595 vfio_get_region_info_cap(struct vfio_region_info *info, uint16_t id)
596 {
597     struct vfio_info_cap_header *hdr;
598     void *ptr = info;
599 
600     if (!(info->flags & VFIO_REGION_INFO_FLAG_CAPS)) {
601         return NULL;
602     }
603 
604     for (hdr = ptr + info->cap_offset; hdr != ptr; hdr = ptr + hdr->next) {
605         if (hdr->id == id) {
606             return hdr;
607         }
608     }
609 
610     return NULL;
611 }
612 
613 static int vfio_setup_region_sparse_mmaps(VFIORegion *region,
614                                           struct vfio_region_info *info)
615 {
616     struct vfio_info_cap_header *hdr;
617     struct vfio_region_info_cap_sparse_mmap *sparse;
618     int i, j;
619 
620     hdr = vfio_get_region_info_cap(info, VFIO_REGION_INFO_CAP_SPARSE_MMAP);
621     if (!hdr) {
622         return -ENODEV;
623     }
624 
625     sparse = container_of(hdr, struct vfio_region_info_cap_sparse_mmap, header);
626 
627     trace_vfio_region_sparse_mmap_header(region->vbasedev->name,
628                                          region->nr, sparse->nr_areas);
629 
630     region->mmaps = g_new0(VFIOMmap, sparse->nr_areas);
631 
632     for (i = 0, j = 0; i < sparse->nr_areas; i++) {
633         trace_vfio_region_sparse_mmap_entry(i, sparse->areas[i].offset,
634                                             sparse->areas[i].offset +
635                                             sparse->areas[i].size);
636 
637         if (sparse->areas[i].size) {
638             region->mmaps[j].offset = sparse->areas[i].offset;
639             region->mmaps[j].size = sparse->areas[i].size;
640             j++;
641         }
642     }
643 
644     region->nr_mmaps = j;
645     region->mmaps = g_realloc(region->mmaps, j * sizeof(VFIOMmap));
646 
647     return 0;
648 }
649 
650 int vfio_region_setup(Object *obj, VFIODevice *vbasedev, VFIORegion *region,
651                       int index, const char *name)
652 {
653     struct vfio_region_info *info;
654     int ret;
655 
656     ret = vfio_get_region_info(vbasedev, index, &info);
657     if (ret) {
658         return ret;
659     }
660 
661     region->vbasedev = vbasedev;
662     region->flags = info->flags;
663     region->size = info->size;
664     region->fd_offset = info->offset;
665     region->nr = index;
666 
667     if (region->size) {
668         region->mem = g_new0(MemoryRegion, 1);
669         memory_region_init_io(region->mem, obj, &vfio_region_ops,
670                               region, name, region->size);
671 
672         if (!vbasedev->no_mmap &&
673             region->flags & VFIO_REGION_INFO_FLAG_MMAP) {
674 
675             ret = vfio_setup_region_sparse_mmaps(region, info);
676 
677             if (ret) {
678                 region->nr_mmaps = 1;
679                 region->mmaps = g_new0(VFIOMmap, region->nr_mmaps);
680                 region->mmaps[0].offset = 0;
681                 region->mmaps[0].size = region->size;
682             }
683         }
684     }
685 
686     g_free(info);
687 
688     trace_vfio_region_setup(vbasedev->name, index, name,
689                             region->flags, region->fd_offset, region->size);
690     return 0;
691 }
692 
693 int vfio_region_mmap(VFIORegion *region)
694 {
695     int i, prot = 0;
696     char *name;
697 
698     if (!region->mem) {
699         return 0;
700     }
701 
702     prot |= region->flags & VFIO_REGION_INFO_FLAG_READ ? PROT_READ : 0;
703     prot |= region->flags & VFIO_REGION_INFO_FLAG_WRITE ? PROT_WRITE : 0;
704 
705     for (i = 0; i < region->nr_mmaps; i++) {
706         region->mmaps[i].mmap = mmap(NULL, region->mmaps[i].size, prot,
707                                      MAP_SHARED, region->vbasedev->fd,
708                                      region->fd_offset +
709                                      region->mmaps[i].offset);
710         if (region->mmaps[i].mmap == MAP_FAILED) {
711             int ret = -errno;
712 
713             trace_vfio_region_mmap_fault(memory_region_name(region->mem), i,
714                                          region->fd_offset +
715                                          region->mmaps[i].offset,
716                                          region->fd_offset +
717                                          region->mmaps[i].offset +
718                                          region->mmaps[i].size - 1, ret);
719 
720             region->mmaps[i].mmap = NULL;
721 
722             for (i--; i >= 0; i--) {
723                 memory_region_del_subregion(region->mem, &region->mmaps[i].mem);
724                 munmap(region->mmaps[i].mmap, region->mmaps[i].size);
725                 object_unparent(OBJECT(&region->mmaps[i].mem));
726                 region->mmaps[i].mmap = NULL;
727             }
728 
729             return ret;
730         }
731 
732         name = g_strdup_printf("%s mmaps[%d]",
733                                memory_region_name(region->mem), i);
734         memory_region_init_ram_device_ptr(&region->mmaps[i].mem,
735                                           memory_region_owner(region->mem),
736                                           name, region->mmaps[i].size,
737                                           region->mmaps[i].mmap);
738         g_free(name);
739         memory_region_add_subregion(region->mem, region->mmaps[i].offset,
740                                     &region->mmaps[i].mem);
741 
742         trace_vfio_region_mmap(memory_region_name(&region->mmaps[i].mem),
743                                region->mmaps[i].offset,
744                                region->mmaps[i].offset +
745                                region->mmaps[i].size - 1);
746     }
747 
748     return 0;
749 }
750 
751 void vfio_region_exit(VFIORegion *region)
752 {
753     int i;
754 
755     if (!region->mem) {
756         return;
757     }
758 
759     for (i = 0; i < region->nr_mmaps; i++) {
760         if (region->mmaps[i].mmap) {
761             memory_region_del_subregion(region->mem, &region->mmaps[i].mem);
762         }
763     }
764 
765     trace_vfio_region_exit(region->vbasedev->name, region->nr);
766 }
767 
768 void vfio_region_finalize(VFIORegion *region)
769 {
770     int i;
771 
772     if (!region->mem) {
773         return;
774     }
775 
776     for (i = 0; i < region->nr_mmaps; i++) {
777         if (region->mmaps[i].mmap) {
778             munmap(region->mmaps[i].mmap, region->mmaps[i].size);
779             object_unparent(OBJECT(&region->mmaps[i].mem));
780         }
781     }
782 
783     object_unparent(OBJECT(region->mem));
784 
785     g_free(region->mem);
786     g_free(region->mmaps);
787 
788     trace_vfio_region_finalize(region->vbasedev->name, region->nr);
789 }
790 
791 void vfio_region_mmaps_set_enabled(VFIORegion *region, bool enabled)
792 {
793     int i;
794 
795     if (!region->mem) {
796         return;
797     }
798 
799     for (i = 0; i < region->nr_mmaps; i++) {
800         if (region->mmaps[i].mmap) {
801             memory_region_set_enabled(&region->mmaps[i].mem, enabled);
802         }
803     }
804 
805     trace_vfio_region_mmaps_set_enabled(memory_region_name(region->mem),
806                                         enabled);
807 }
808 
809 void vfio_reset_handler(void *opaque)
810 {
811     VFIOGroup *group;
812     VFIODevice *vbasedev;
813 
814     QLIST_FOREACH(group, &vfio_group_list, next) {
815         QLIST_FOREACH(vbasedev, &group->device_list, next) {
816             vbasedev->ops->vfio_compute_needs_reset(vbasedev);
817         }
818     }
819 
820     QLIST_FOREACH(group, &vfio_group_list, next) {
821         QLIST_FOREACH(vbasedev, &group->device_list, next) {
822             if (vbasedev->needs_reset) {
823                 vbasedev->ops->vfio_hot_reset_multi(vbasedev);
824             }
825         }
826     }
827 }
828 
829 static void vfio_kvm_device_add_group(VFIOGroup *group)
830 {
831 #ifdef CONFIG_KVM
832     struct kvm_device_attr attr = {
833         .group = KVM_DEV_VFIO_GROUP,
834         .attr = KVM_DEV_VFIO_GROUP_ADD,
835         .addr = (uint64_t)(unsigned long)&group->fd,
836     };
837 
838     if (!kvm_enabled()) {
839         return;
840     }
841 
842     if (vfio_kvm_device_fd < 0) {
843         struct kvm_create_device cd = {
844             .type = KVM_DEV_TYPE_VFIO,
845         };
846 
847         if (kvm_vm_ioctl(kvm_state, KVM_CREATE_DEVICE, &cd)) {
848             error_report("Failed to create KVM VFIO device: %m");
849             return;
850         }
851 
852         vfio_kvm_device_fd = cd.fd;
853     }
854 
855     if (ioctl(vfio_kvm_device_fd, KVM_SET_DEVICE_ATTR, &attr)) {
856         error_report("Failed to add group %d to KVM VFIO device: %m",
857                      group->groupid);
858     }
859 #endif
860 }
861 
862 static void vfio_kvm_device_del_group(VFIOGroup *group)
863 {
864 #ifdef CONFIG_KVM
865     struct kvm_device_attr attr = {
866         .group = KVM_DEV_VFIO_GROUP,
867         .attr = KVM_DEV_VFIO_GROUP_DEL,
868         .addr = (uint64_t)(unsigned long)&group->fd,
869     };
870 
871     if (vfio_kvm_device_fd < 0) {
872         return;
873     }
874 
875     if (ioctl(vfio_kvm_device_fd, KVM_SET_DEVICE_ATTR, &attr)) {
876         error_report("Failed to remove group %d from KVM VFIO device: %m",
877                      group->groupid);
878     }
879 #endif
880 }
881 
882 static VFIOAddressSpace *vfio_get_address_space(AddressSpace *as)
883 {
884     VFIOAddressSpace *space;
885 
886     QLIST_FOREACH(space, &vfio_address_spaces, list) {
887         if (space->as == as) {
888             return space;
889         }
890     }
891 
892     /* No suitable VFIOAddressSpace, create a new one */
893     space = g_malloc0(sizeof(*space));
894     space->as = as;
895     QLIST_INIT(&space->containers);
896 
897     QLIST_INSERT_HEAD(&vfio_address_spaces, space, list);
898 
899     return space;
900 }
901 
902 static void vfio_put_address_space(VFIOAddressSpace *space)
903 {
904     if (QLIST_EMPTY(&space->containers)) {
905         QLIST_REMOVE(space, list);
906         g_free(space);
907     }
908 }
909 
910 static int vfio_connect_container(VFIOGroup *group, AddressSpace *as,
911                                   Error **errp)
912 {
913     VFIOContainer *container;
914     int ret, fd;
915     VFIOAddressSpace *space;
916 
917     space = vfio_get_address_space(as);
918 
919     QLIST_FOREACH(container, &space->containers, next) {
920         if (!ioctl(group->fd, VFIO_GROUP_SET_CONTAINER, &container->fd)) {
921             group->container = container;
922             QLIST_INSERT_HEAD(&container->group_list, group, container_next);
923             return 0;
924         }
925     }
926 
927     fd = qemu_open("/dev/vfio/vfio", O_RDWR);
928     if (fd < 0) {
929         error_setg_errno(errp, errno, "failed to open /dev/vfio/vfio");
930         ret = -errno;
931         goto put_space_exit;
932     }
933 
934     ret = ioctl(fd, VFIO_GET_API_VERSION);
935     if (ret != VFIO_API_VERSION) {
936         error_setg(errp, "supported vfio version: %d, "
937                    "reported version: %d", VFIO_API_VERSION, ret);
938         ret = -EINVAL;
939         goto close_fd_exit;
940     }
941 
942     container = g_malloc0(sizeof(*container));
943     container->space = space;
944     container->fd = fd;
945     if (ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_TYPE1_IOMMU) ||
946         ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_TYPE1v2_IOMMU)) {
947         bool v2 = !!ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_TYPE1v2_IOMMU);
948         struct vfio_iommu_type1_info info;
949 
950         ret = ioctl(group->fd, VFIO_GROUP_SET_CONTAINER, &fd);
951         if (ret) {
952             error_setg_errno(errp, errno, "failed to set group container");
953             ret = -errno;
954             goto free_container_exit;
955         }
956 
957         container->iommu_type = v2 ? VFIO_TYPE1v2_IOMMU : VFIO_TYPE1_IOMMU;
958         ret = ioctl(fd, VFIO_SET_IOMMU, container->iommu_type);
959         if (ret) {
960             error_setg_errno(errp, errno, "failed to set iommu for container");
961             ret = -errno;
962             goto free_container_exit;
963         }
964 
965         /*
966          * FIXME: This assumes that a Type1 IOMMU can map any 64-bit
967          * IOVA whatsoever.  That's not actually true, but the current
968          * kernel interface doesn't tell us what it can map, and the
969          * existing Type1 IOMMUs generally support any IOVA we're
970          * going to actually try in practice.
971          */
972         info.argsz = sizeof(info);
973         ret = ioctl(fd, VFIO_IOMMU_GET_INFO, &info);
974         /* Ignore errors */
975         if (ret || !(info.flags & VFIO_IOMMU_INFO_PGSIZES)) {
976             /* Assume 4k IOVA page size */
977             info.iova_pgsizes = 4096;
978         }
979         vfio_host_win_add(container, 0, (hwaddr)-1, info.iova_pgsizes);
980     } else if (ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_SPAPR_TCE_IOMMU) ||
981                ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_SPAPR_TCE_v2_IOMMU)) {
982         struct vfio_iommu_spapr_tce_info info;
983         bool v2 = !!ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_SPAPR_TCE_v2_IOMMU);
984 
985         ret = ioctl(group->fd, VFIO_GROUP_SET_CONTAINER, &fd);
986         if (ret) {
987             error_setg_errno(errp, errno, "failed to set group container");
988             ret = -errno;
989             goto free_container_exit;
990         }
991         container->iommu_type =
992             v2 ? VFIO_SPAPR_TCE_v2_IOMMU : VFIO_SPAPR_TCE_IOMMU;
993         ret = ioctl(fd, VFIO_SET_IOMMU, container->iommu_type);
994         if (ret) {
995             error_setg_errno(errp, errno, "failed to set iommu for container");
996             ret = -errno;
997             goto free_container_exit;
998         }
999 
1000         /*
1001          * The host kernel code implementing VFIO_IOMMU_DISABLE is called
1002          * when container fd is closed so we do not call it explicitly
1003          * in this file.
1004          */
1005         if (!v2) {
1006             ret = ioctl(fd, VFIO_IOMMU_ENABLE);
1007             if (ret) {
1008                 error_setg_errno(errp, errno, "failed to enable container");
1009                 ret = -errno;
1010                 goto free_container_exit;
1011             }
1012         } else {
1013             container->prereg_listener = vfio_prereg_listener;
1014 
1015             memory_listener_register(&container->prereg_listener,
1016                                      &address_space_memory);
1017             if (container->error) {
1018                 memory_listener_unregister(&container->prereg_listener);
1019                 ret = container->error;
1020                 error_setg(errp,
1021                     "RAM memory listener initialization failed for container");
1022                 goto free_container_exit;
1023             }
1024         }
1025 
1026         info.argsz = sizeof(info);
1027         ret = ioctl(fd, VFIO_IOMMU_SPAPR_TCE_GET_INFO, &info);
1028         if (ret) {
1029             error_setg_errno(errp, errno,
1030                              "VFIO_IOMMU_SPAPR_TCE_GET_INFO failed");
1031             ret = -errno;
1032             if (v2) {
1033                 memory_listener_unregister(&container->prereg_listener);
1034             }
1035             goto free_container_exit;
1036         }
1037 
1038         if (v2) {
1039             /*
1040              * There is a default window in just created container.
1041              * To make region_add/del simpler, we better remove this
1042              * window now and let those iommu_listener callbacks
1043              * create/remove them when needed.
1044              */
1045             ret = vfio_spapr_remove_window(container, info.dma32_window_start);
1046             if (ret) {
1047                 error_setg_errno(errp, -ret,
1048                                  "failed to remove existing window");
1049                 goto free_container_exit;
1050             }
1051         } else {
1052             /* The default table uses 4K pages */
1053             vfio_host_win_add(container, info.dma32_window_start,
1054                               info.dma32_window_start +
1055                               info.dma32_window_size - 1,
1056                               0x1000);
1057         }
1058     } else {
1059         error_setg(errp, "No available IOMMU models");
1060         ret = -EINVAL;
1061         goto free_container_exit;
1062     }
1063 
1064     container->listener = vfio_memory_listener;
1065 
1066     memory_listener_register(&container->listener, container->space->as);
1067 
1068     if (container->error) {
1069         ret = container->error;
1070         error_setg_errno(errp, -ret,
1071                          "memory listener initialization failed for container");
1072         goto listener_release_exit;
1073     }
1074 
1075     container->initialized = true;
1076 
1077     QLIST_INIT(&container->group_list);
1078     QLIST_INSERT_HEAD(&space->containers, container, next);
1079 
1080     group->container = container;
1081     QLIST_INSERT_HEAD(&container->group_list, group, container_next);
1082 
1083     return 0;
1084 listener_release_exit:
1085     vfio_listener_release(container);
1086 
1087 free_container_exit:
1088     g_free(container);
1089 
1090 close_fd_exit:
1091     close(fd);
1092 
1093 put_space_exit:
1094     vfio_put_address_space(space);
1095 
1096     return ret;
1097 }
1098 
1099 static void vfio_disconnect_container(VFIOGroup *group)
1100 {
1101     VFIOContainer *container = group->container;
1102 
1103     if (ioctl(group->fd, VFIO_GROUP_UNSET_CONTAINER, &container->fd)) {
1104         error_report("vfio: error disconnecting group %d from container",
1105                      group->groupid);
1106     }
1107 
1108     QLIST_REMOVE(group, container_next);
1109     group->container = NULL;
1110 
1111     if (QLIST_EMPTY(&container->group_list)) {
1112         VFIOAddressSpace *space = container->space;
1113         VFIOGuestIOMMU *giommu, *tmp;
1114 
1115         vfio_listener_release(container);
1116         QLIST_REMOVE(container, next);
1117 
1118         QLIST_FOREACH_SAFE(giommu, &container->giommu_list, giommu_next, tmp) {
1119             memory_region_unregister_iommu_notifier(giommu->iommu, &giommu->n);
1120             QLIST_REMOVE(giommu, giommu_next);
1121             g_free(giommu);
1122         }
1123 
1124         trace_vfio_disconnect_container(container->fd);
1125         close(container->fd);
1126         g_free(container);
1127 
1128         vfio_put_address_space(space);
1129     }
1130 }
1131 
1132 VFIOGroup *vfio_get_group(int groupid, AddressSpace *as, Error **errp)
1133 {
1134     VFIOGroup *group;
1135     char path[32];
1136     struct vfio_group_status status = { .argsz = sizeof(status) };
1137 
1138     QLIST_FOREACH(group, &vfio_group_list, next) {
1139         if (group->groupid == groupid) {
1140             /* Found it.  Now is it already in the right context? */
1141             if (group->container->space->as == as) {
1142                 return group;
1143             } else {
1144                 error_setg(errp, "group %d used in multiple address spaces",
1145                            group->groupid);
1146                 return NULL;
1147             }
1148         }
1149     }
1150 
1151     group = g_malloc0(sizeof(*group));
1152 
1153     snprintf(path, sizeof(path), "/dev/vfio/%d", groupid);
1154     group->fd = qemu_open(path, O_RDWR);
1155     if (group->fd < 0) {
1156         error_setg_errno(errp, errno, "failed to open %s", path);
1157         goto free_group_exit;
1158     }
1159 
1160     if (ioctl(group->fd, VFIO_GROUP_GET_STATUS, &status)) {
1161         error_setg_errno(errp, errno, "failed to get group %d status", groupid);
1162         goto close_fd_exit;
1163     }
1164 
1165     if (!(status.flags & VFIO_GROUP_FLAGS_VIABLE)) {
1166         error_setg(errp, "group %d is not viable", groupid);
1167         error_append_hint(errp,
1168                           "Please ensure all devices within the iommu_group "
1169                           "are bound to their vfio bus driver.\n");
1170         goto close_fd_exit;
1171     }
1172 
1173     group->groupid = groupid;
1174     QLIST_INIT(&group->device_list);
1175 
1176     if (vfio_connect_container(group, as, errp)) {
1177         error_prepend(errp, "failed to setup container for group %d: ",
1178                       groupid);
1179         goto close_fd_exit;
1180     }
1181 
1182     if (QLIST_EMPTY(&vfio_group_list)) {
1183         qemu_register_reset(vfio_reset_handler, NULL);
1184     }
1185 
1186     QLIST_INSERT_HEAD(&vfio_group_list, group, next);
1187 
1188     vfio_kvm_device_add_group(group);
1189 
1190     return group;
1191 
1192 close_fd_exit:
1193     close(group->fd);
1194 
1195 free_group_exit:
1196     g_free(group);
1197 
1198     return NULL;
1199 }
1200 
1201 void vfio_put_group(VFIOGroup *group)
1202 {
1203     if (!group || !QLIST_EMPTY(&group->device_list)) {
1204         return;
1205     }
1206 
1207     vfio_kvm_device_del_group(group);
1208     vfio_disconnect_container(group);
1209     QLIST_REMOVE(group, next);
1210     trace_vfio_put_group(group->fd);
1211     close(group->fd);
1212     g_free(group);
1213 
1214     if (QLIST_EMPTY(&vfio_group_list)) {
1215         qemu_unregister_reset(vfio_reset_handler, NULL);
1216     }
1217 }
1218 
1219 int vfio_get_device(VFIOGroup *group, const char *name,
1220                     VFIODevice *vbasedev, Error **errp)
1221 {
1222     struct vfio_device_info dev_info = { .argsz = sizeof(dev_info) };
1223     int ret, fd;
1224 
1225     fd = ioctl(group->fd, VFIO_GROUP_GET_DEVICE_FD, name);
1226     if (fd < 0) {
1227         error_setg_errno(errp, errno, "error getting device from group %d",
1228                          group->groupid);
1229         error_append_hint(errp,
1230                       "Verify all devices in group %d are bound to vfio-<bus> "
1231                       "or pci-stub and not already in use\n", group->groupid);
1232         return fd;
1233     }
1234 
1235     ret = ioctl(fd, VFIO_DEVICE_GET_INFO, &dev_info);
1236     if (ret) {
1237         error_setg_errno(errp, errno, "error getting device info");
1238         close(fd);
1239         return ret;
1240     }
1241 
1242     vbasedev->fd = fd;
1243     vbasedev->group = group;
1244     QLIST_INSERT_HEAD(&group->device_list, vbasedev, next);
1245 
1246     vbasedev->num_irqs = dev_info.num_irqs;
1247     vbasedev->num_regions = dev_info.num_regions;
1248     vbasedev->flags = dev_info.flags;
1249 
1250     trace_vfio_get_device(name, dev_info.flags, dev_info.num_regions,
1251                           dev_info.num_irqs);
1252 
1253     vbasedev->reset_works = !!(dev_info.flags & VFIO_DEVICE_FLAGS_RESET);
1254     return 0;
1255 }
1256 
1257 void vfio_put_base_device(VFIODevice *vbasedev)
1258 {
1259     if (!vbasedev->group) {
1260         return;
1261     }
1262     QLIST_REMOVE(vbasedev, next);
1263     vbasedev->group = NULL;
1264     trace_vfio_put_base_device(vbasedev->fd);
1265     close(vbasedev->fd);
1266 }
1267 
1268 int vfio_get_region_info(VFIODevice *vbasedev, int index,
1269                          struct vfio_region_info **info)
1270 {
1271     size_t argsz = sizeof(struct vfio_region_info);
1272 
1273     *info = g_malloc0(argsz);
1274 
1275     (*info)->index = index;
1276 retry:
1277     (*info)->argsz = argsz;
1278 
1279     if (ioctl(vbasedev->fd, VFIO_DEVICE_GET_REGION_INFO, *info)) {
1280         g_free(*info);
1281         *info = NULL;
1282         return -errno;
1283     }
1284 
1285     if ((*info)->argsz > argsz) {
1286         argsz = (*info)->argsz;
1287         *info = g_realloc(*info, argsz);
1288 
1289         goto retry;
1290     }
1291 
1292     return 0;
1293 }
1294 
1295 int vfio_get_dev_region_info(VFIODevice *vbasedev, uint32_t type,
1296                              uint32_t subtype, struct vfio_region_info **info)
1297 {
1298     int i;
1299 
1300     for (i = 0; i < vbasedev->num_regions; i++) {
1301         struct vfio_info_cap_header *hdr;
1302         struct vfio_region_info_cap_type *cap_type;
1303 
1304         if (vfio_get_region_info(vbasedev, i, info)) {
1305             continue;
1306         }
1307 
1308         hdr = vfio_get_region_info_cap(*info, VFIO_REGION_INFO_CAP_TYPE);
1309         if (!hdr) {
1310             g_free(*info);
1311             continue;
1312         }
1313 
1314         cap_type = container_of(hdr, struct vfio_region_info_cap_type, header);
1315 
1316         trace_vfio_get_dev_region(vbasedev->name, i,
1317                                   cap_type->type, cap_type->subtype);
1318 
1319         if (cap_type->type == type && cap_type->subtype == subtype) {
1320             return 0;
1321         }
1322 
1323         g_free(*info);
1324     }
1325 
1326     *info = NULL;
1327     return -ENODEV;
1328 }
1329 
1330 /*
1331  * Interfaces for IBM EEH (Enhanced Error Handling)
1332  */
1333 static bool vfio_eeh_container_ok(VFIOContainer *container)
1334 {
1335     /*
1336      * As of 2016-03-04 (linux-4.5) the host kernel EEH/VFIO
1337      * implementation is broken if there are multiple groups in a
1338      * container.  The hardware works in units of Partitionable
1339      * Endpoints (== IOMMU groups) and the EEH operations naively
1340      * iterate across all groups in the container, without any logic
1341      * to make sure the groups have their state synchronized.  For
1342      * certain operations (ENABLE) that might be ok, until an error
1343      * occurs, but for others (GET_STATE) it's clearly broken.
1344      */
1345 
1346     /*
1347      * XXX Once fixed kernels exist, test for them here
1348      */
1349 
1350     if (QLIST_EMPTY(&container->group_list)) {
1351         return false;
1352     }
1353 
1354     if (QLIST_NEXT(QLIST_FIRST(&container->group_list), container_next)) {
1355         return false;
1356     }
1357 
1358     return true;
1359 }
1360 
1361 static int vfio_eeh_container_op(VFIOContainer *container, uint32_t op)
1362 {
1363     struct vfio_eeh_pe_op pe_op = {
1364         .argsz = sizeof(pe_op),
1365         .op = op,
1366     };
1367     int ret;
1368 
1369     if (!vfio_eeh_container_ok(container)) {
1370         error_report("vfio/eeh: EEH_PE_OP 0x%x: "
1371                      "kernel requires a container with exactly one group", op);
1372         return -EPERM;
1373     }
1374 
1375     ret = ioctl(container->fd, VFIO_EEH_PE_OP, &pe_op);
1376     if (ret < 0) {
1377         error_report("vfio/eeh: EEH_PE_OP 0x%x failed: %m", op);
1378         return -errno;
1379     }
1380 
1381     return ret;
1382 }
1383 
1384 static VFIOContainer *vfio_eeh_as_container(AddressSpace *as)
1385 {
1386     VFIOAddressSpace *space = vfio_get_address_space(as);
1387     VFIOContainer *container = NULL;
1388 
1389     if (QLIST_EMPTY(&space->containers)) {
1390         /* No containers to act on */
1391         goto out;
1392     }
1393 
1394     container = QLIST_FIRST(&space->containers);
1395 
1396     if (QLIST_NEXT(container, next)) {
1397         /* We don't yet have logic to synchronize EEH state across
1398          * multiple containers */
1399         container = NULL;
1400         goto out;
1401     }
1402 
1403 out:
1404     vfio_put_address_space(space);
1405     return container;
1406 }
1407 
1408 bool vfio_eeh_as_ok(AddressSpace *as)
1409 {
1410     VFIOContainer *container = vfio_eeh_as_container(as);
1411 
1412     return (container != NULL) && vfio_eeh_container_ok(container);
1413 }
1414 
1415 int vfio_eeh_as_op(AddressSpace *as, uint32_t op)
1416 {
1417     VFIOContainer *container = vfio_eeh_as_container(as);
1418 
1419     if (!container) {
1420         return -ENODEV;
1421     }
1422     return vfio_eeh_container_op(container, op);
1423 }
1424