16e6d8ac6SEric Auger /* 26e6d8ac6SEric Auger * iommufd container backend 36e6d8ac6SEric Auger * 46e6d8ac6SEric Auger * Copyright (C) 2023 Intel Corporation. 56e6d8ac6SEric Auger * Copyright Red Hat, Inc. 2023 66e6d8ac6SEric Auger * 76e6d8ac6SEric Auger * Authors: Yi Liu <yi.l.liu@intel.com> 86e6d8ac6SEric Auger * Eric Auger <eric.auger@redhat.com> 96e6d8ac6SEric Auger * 106e6d8ac6SEric Auger * SPDX-License-Identifier: GPL-2.0-or-later 116e6d8ac6SEric Auger */ 126e6d8ac6SEric Auger 136e6d8ac6SEric Auger #include "qemu/osdep.h" 146e6d8ac6SEric Auger #include "sysemu/iommufd.h" 156e6d8ac6SEric Auger #include "qapi/error.h" 166e6d8ac6SEric Auger #include "qemu/module.h" 176e6d8ac6SEric Auger #include "qom/object_interfaces.h" 186e6d8ac6SEric Auger #include "qemu/error-report.h" 196e6d8ac6SEric Auger #include "monitor/monitor.h" 206e6d8ac6SEric Auger #include "trace.h" 216e6d8ac6SEric Auger #include <sys/ioctl.h> 226e6d8ac6SEric Auger #include <linux/iommufd.h> 236e6d8ac6SEric Auger 246e6d8ac6SEric Auger static void iommufd_backend_init(Object *obj) 256e6d8ac6SEric Auger { 266e6d8ac6SEric Auger IOMMUFDBackend *be = IOMMUFD_BACKEND(obj); 276e6d8ac6SEric Auger 286e6d8ac6SEric Auger be->fd = -1; 296e6d8ac6SEric Auger be->users = 0; 306e6d8ac6SEric Auger be->owned = true; 316e6d8ac6SEric Auger } 326e6d8ac6SEric Auger 336e6d8ac6SEric Auger static void iommufd_backend_finalize(Object *obj) 346e6d8ac6SEric Auger { 356e6d8ac6SEric Auger IOMMUFDBackend *be = IOMMUFD_BACKEND(obj); 366e6d8ac6SEric Auger 376e6d8ac6SEric Auger if (be->owned) { 386e6d8ac6SEric Auger close(be->fd); 396e6d8ac6SEric Auger be->fd = -1; 406e6d8ac6SEric Auger } 416e6d8ac6SEric Auger } 426e6d8ac6SEric Auger 436e6d8ac6SEric Auger static void iommufd_backend_set_fd(Object *obj, const char *str, Error **errp) 446e6d8ac6SEric Auger { 45c1cccad8SZhao Liu ERRP_GUARD(); 466e6d8ac6SEric Auger IOMMUFDBackend *be = IOMMUFD_BACKEND(obj); 476e6d8ac6SEric Auger int fd = -1; 486e6d8ac6SEric Auger 496e6d8ac6SEric Auger fd = monitor_fd_param(monitor_cur(), str, errp); 506e6d8ac6SEric Auger if (fd == -1) { 516e6d8ac6SEric Auger error_prepend(errp, "Could not parse remote object fd %s:", str); 526e6d8ac6SEric Auger return; 536e6d8ac6SEric Auger } 546e6d8ac6SEric Auger be->fd = fd; 556e6d8ac6SEric Auger be->owned = false; 566e6d8ac6SEric Auger trace_iommu_backend_set_fd(be->fd); 576e6d8ac6SEric Auger } 586e6d8ac6SEric Auger 596e6d8ac6SEric Auger static bool iommufd_backend_can_be_deleted(UserCreatable *uc) 606e6d8ac6SEric Auger { 616e6d8ac6SEric Auger IOMMUFDBackend *be = IOMMUFD_BACKEND(uc); 626e6d8ac6SEric Auger 636e6d8ac6SEric Auger return !be->users; 646e6d8ac6SEric Auger } 656e6d8ac6SEric Auger 666e6d8ac6SEric Auger static void iommufd_backend_class_init(ObjectClass *oc, void *data) 676e6d8ac6SEric Auger { 686e6d8ac6SEric Auger UserCreatableClass *ucc = USER_CREATABLE_CLASS(oc); 696e6d8ac6SEric Auger 706e6d8ac6SEric Auger ucc->can_be_deleted = iommufd_backend_can_be_deleted; 716e6d8ac6SEric Auger 726e6d8ac6SEric Auger object_class_property_add_str(oc, "fd", NULL, iommufd_backend_set_fd); 736e6d8ac6SEric Auger } 746e6d8ac6SEric Auger 759067d50dSZhenzhong Duan bool iommufd_backend_connect(IOMMUFDBackend *be, Error **errp) 766e6d8ac6SEric Auger { 779067d50dSZhenzhong Duan int fd; 786e6d8ac6SEric Auger 796e6d8ac6SEric Auger if (be->owned && !be->users) { 806e6d8ac6SEric Auger fd = qemu_open_old("/dev/iommu", O_RDWR); 816e6d8ac6SEric Auger if (fd < 0) { 826e6d8ac6SEric Auger error_setg_errno(errp, errno, "/dev/iommu opening failed"); 839067d50dSZhenzhong Duan return false; 846e6d8ac6SEric Auger } 856e6d8ac6SEric Auger be->fd = fd; 866e6d8ac6SEric Auger } 876e6d8ac6SEric Auger be->users++; 889067d50dSZhenzhong Duan 899067d50dSZhenzhong Duan trace_iommufd_backend_connect(be->fd, be->owned, be->users); 909067d50dSZhenzhong Duan return true; 916e6d8ac6SEric Auger } 926e6d8ac6SEric Auger 936e6d8ac6SEric Auger void iommufd_backend_disconnect(IOMMUFDBackend *be) 946e6d8ac6SEric Auger { 956e6d8ac6SEric Auger if (!be->users) { 966e6d8ac6SEric Auger goto out; 976e6d8ac6SEric Auger } 986e6d8ac6SEric Auger be->users--; 996e6d8ac6SEric Auger if (!be->users && be->owned) { 1006e6d8ac6SEric Auger close(be->fd); 1016e6d8ac6SEric Auger be->fd = -1; 1026e6d8ac6SEric Auger } 1036e6d8ac6SEric Auger out: 1046e6d8ac6SEric Auger trace_iommufd_backend_disconnect(be->fd, be->users); 1056e6d8ac6SEric Auger } 1066e6d8ac6SEric Auger 1079067d50dSZhenzhong Duan bool iommufd_backend_alloc_ioas(IOMMUFDBackend *be, uint32_t *ioas_id, 1086e6d8ac6SEric Auger Error **errp) 1096e6d8ac6SEric Auger { 1109067d50dSZhenzhong Duan int fd = be->fd; 1116e6d8ac6SEric Auger struct iommu_ioas_alloc alloc_data = { 1126e6d8ac6SEric Auger .size = sizeof(alloc_data), 1136e6d8ac6SEric Auger .flags = 0, 1146e6d8ac6SEric Auger }; 1156e6d8ac6SEric Auger 1169067d50dSZhenzhong Duan if (ioctl(fd, IOMMU_IOAS_ALLOC, &alloc_data)) { 1176e6d8ac6SEric Auger error_setg_errno(errp, errno, "Failed to allocate ioas"); 1189067d50dSZhenzhong Duan return false; 1196e6d8ac6SEric Auger } 1206e6d8ac6SEric Auger 1216e6d8ac6SEric Auger *ioas_id = alloc_data.out_ioas_id; 1229067d50dSZhenzhong Duan trace_iommufd_backend_alloc_ioas(fd, *ioas_id); 1236e6d8ac6SEric Auger 1249067d50dSZhenzhong Duan return true; 1256e6d8ac6SEric Auger } 1266e6d8ac6SEric Auger 1276e6d8ac6SEric Auger void iommufd_backend_free_id(IOMMUFDBackend *be, uint32_t id) 1286e6d8ac6SEric Auger { 1296e6d8ac6SEric Auger int ret, fd = be->fd; 1306e6d8ac6SEric Auger struct iommu_destroy des = { 1316e6d8ac6SEric Auger .size = sizeof(des), 1326e6d8ac6SEric Auger .id = id, 1336e6d8ac6SEric Auger }; 1346e6d8ac6SEric Auger 1356e6d8ac6SEric Auger ret = ioctl(fd, IOMMU_DESTROY, &des); 1366e6d8ac6SEric Auger trace_iommufd_backend_free_id(fd, id, ret); 1376e6d8ac6SEric Auger if (ret) { 1386e6d8ac6SEric Auger error_report("Failed to free id: %u %m", id); 1396e6d8ac6SEric Auger } 1406e6d8ac6SEric Auger } 1416e6d8ac6SEric Auger 1426e6d8ac6SEric Auger int iommufd_backend_map_dma(IOMMUFDBackend *be, uint32_t ioas_id, hwaddr iova, 1436e6d8ac6SEric Auger ram_addr_t size, void *vaddr, bool readonly) 1446e6d8ac6SEric Auger { 1456e6d8ac6SEric Auger int ret, fd = be->fd; 1466e6d8ac6SEric Auger struct iommu_ioas_map map = { 1476e6d8ac6SEric Auger .size = sizeof(map), 1486e6d8ac6SEric Auger .flags = IOMMU_IOAS_MAP_READABLE | 1496e6d8ac6SEric Auger IOMMU_IOAS_MAP_FIXED_IOVA, 1506e6d8ac6SEric Auger .ioas_id = ioas_id, 1516e6d8ac6SEric Auger .__reserved = 0, 1526e6d8ac6SEric Auger .user_va = (uintptr_t)vaddr, 1536e6d8ac6SEric Auger .iova = iova, 1546e6d8ac6SEric Auger .length = size, 1556e6d8ac6SEric Auger }; 1566e6d8ac6SEric Auger 1576e6d8ac6SEric Auger if (!readonly) { 1586e6d8ac6SEric Auger map.flags |= IOMMU_IOAS_MAP_WRITEABLE; 1596e6d8ac6SEric Auger } 1606e6d8ac6SEric Auger 1616e6d8ac6SEric Auger ret = ioctl(fd, IOMMU_IOAS_MAP, &map); 1626e6d8ac6SEric Auger trace_iommufd_backend_map_dma(fd, ioas_id, iova, size, 1636e6d8ac6SEric Auger vaddr, readonly, ret); 1646e6d8ac6SEric Auger if (ret) { 1656e6d8ac6SEric Auger ret = -errno; 1666e6d8ac6SEric Auger 1676e6d8ac6SEric Auger /* TODO: Not support mapping hardware PCI BAR region for now. */ 1686e6d8ac6SEric Auger if (errno == EFAULT) { 1696e6d8ac6SEric Auger warn_report("IOMMU_IOAS_MAP failed: %m, PCI BAR?"); 1706e6d8ac6SEric Auger } else { 1716e6d8ac6SEric Auger error_report("IOMMU_IOAS_MAP failed: %m"); 1726e6d8ac6SEric Auger } 1736e6d8ac6SEric Auger } 1746e6d8ac6SEric Auger return ret; 1756e6d8ac6SEric Auger } 1766e6d8ac6SEric Auger 1776e6d8ac6SEric Auger int iommufd_backend_unmap_dma(IOMMUFDBackend *be, uint32_t ioas_id, 1786e6d8ac6SEric Auger hwaddr iova, ram_addr_t size) 1796e6d8ac6SEric Auger { 1806e6d8ac6SEric Auger int ret, fd = be->fd; 1816e6d8ac6SEric Auger struct iommu_ioas_unmap unmap = { 1826e6d8ac6SEric Auger .size = sizeof(unmap), 1836e6d8ac6SEric Auger .ioas_id = ioas_id, 1846e6d8ac6SEric Auger .iova = iova, 1856e6d8ac6SEric Auger .length = size, 1866e6d8ac6SEric Auger }; 1876e6d8ac6SEric Auger 1886e6d8ac6SEric Auger ret = ioctl(fd, IOMMU_IOAS_UNMAP, &unmap); 1896e6d8ac6SEric Auger /* 1906e6d8ac6SEric Auger * IOMMUFD takes mapping as some kind of object, unmapping 1916e6d8ac6SEric Auger * nonexistent mapping is treated as deleting a nonexistent 1926e6d8ac6SEric Auger * object and return ENOENT. This is different from legacy 1936e6d8ac6SEric Auger * backend which allows it. vIOMMU may trigger a lot of 1946e6d8ac6SEric Auger * redundant unmapping, to avoid flush the log, treat them 1956e6d8ac6SEric Auger * as succeess for IOMMUFD just like legacy backend. 1966e6d8ac6SEric Auger */ 1976e6d8ac6SEric Auger if (ret && errno == ENOENT) { 1986e6d8ac6SEric Auger trace_iommufd_backend_unmap_dma_non_exist(fd, ioas_id, iova, size, ret); 1996e6d8ac6SEric Auger ret = 0; 2006e6d8ac6SEric Auger } else { 2016e6d8ac6SEric Auger trace_iommufd_backend_unmap_dma(fd, ioas_id, iova, size, ret); 2026e6d8ac6SEric Auger } 2036e6d8ac6SEric Auger 2046e6d8ac6SEric Auger if (ret) { 2056e6d8ac6SEric Auger ret = -errno; 2066e6d8ac6SEric Auger error_report("IOMMU_IOAS_UNMAP failed: %m"); 2076e6d8ac6SEric Auger } 2086e6d8ac6SEric Auger return ret; 2096e6d8ac6SEric Auger } 2106e6d8ac6SEric Auger 211*42965386SZhenzhong Duan bool iommufd_backend_get_device_info(IOMMUFDBackend *be, uint32_t devid, 212*42965386SZhenzhong Duan uint32_t *type, void *data, uint32_t len, 213*42965386SZhenzhong Duan Error **errp) 214*42965386SZhenzhong Duan { 215*42965386SZhenzhong Duan struct iommu_hw_info info = { 216*42965386SZhenzhong Duan .size = sizeof(info), 217*42965386SZhenzhong Duan .dev_id = devid, 218*42965386SZhenzhong Duan .data_len = len, 219*42965386SZhenzhong Duan .data_uptr = (uintptr_t)data, 220*42965386SZhenzhong Duan }; 221*42965386SZhenzhong Duan 222*42965386SZhenzhong Duan if (ioctl(be->fd, IOMMU_GET_HW_INFO, &info)) { 223*42965386SZhenzhong Duan error_setg_errno(errp, errno, "Failed to get hardware info"); 224*42965386SZhenzhong Duan return false; 225*42965386SZhenzhong Duan } 226*42965386SZhenzhong Duan 227*42965386SZhenzhong Duan g_assert(type); 228*42965386SZhenzhong Duan *type = info.out_data_type; 229*42965386SZhenzhong Duan 230*42965386SZhenzhong Duan return true; 231*42965386SZhenzhong Duan } 232*42965386SZhenzhong Duan 2339005f928SZhenzhong Duan static const TypeInfo types[] = { 2349005f928SZhenzhong Duan { 2356e6d8ac6SEric Auger .name = TYPE_IOMMUFD_BACKEND, 2366e6d8ac6SEric Auger .parent = TYPE_OBJECT, 2376e6d8ac6SEric Auger .instance_size = sizeof(IOMMUFDBackend), 2386e6d8ac6SEric Auger .instance_init = iommufd_backend_init, 2396e6d8ac6SEric Auger .instance_finalize = iommufd_backend_finalize, 2406e6d8ac6SEric Auger .class_size = sizeof(IOMMUFDBackendClass), 2416e6d8ac6SEric Auger .class_init = iommufd_backend_class_init, 2426e6d8ac6SEric Auger .interfaces = (InterfaceInfo[]) { 2436e6d8ac6SEric Auger { TYPE_USER_CREATABLE }, 2446e6d8ac6SEric Auger { } 2456e6d8ac6SEric Auger } 2469005f928SZhenzhong Duan }, { 2479005f928SZhenzhong Duan .name = TYPE_HOST_IOMMU_DEVICE_IOMMUFD, 2489005f928SZhenzhong Duan .parent = TYPE_HOST_IOMMU_DEVICE, 2499005f928SZhenzhong Duan .abstract = true, 2509005f928SZhenzhong Duan } 2516e6d8ac6SEric Auger }; 2526e6d8ac6SEric Auger 2539005f928SZhenzhong Duan DEFINE_TYPES(types) 254