39 MODULE_DESCRIPTION("Microsoft Hyper-V root partition VMM interface /dev/mshv");
75 static int mshv_init_async_handler(struct mshv_partition *partition);
119  * Only allow hypercalls that have a u64 partition id as the first member of
150 static int mshv_ioctl_passthru_hvcall(struct mshv_partition *partition,  in mshv_ioctl_passthru_hvcall()  argument
182 		/* async hypercalls can only be called from partition fd */  in mshv_ioctl_passthru_hvcall()
185 		ret = mshv_init_async_handler(partition);  in mshv_ioctl_passthru_hvcall()
211 	*(u64 *)input_pg = partition->pt_id;  in mshv_ioctl_passthru_hvcall()
221 			mshv_async_hvcall_handler(partition, &status);  in mshv_ioctl_passthru_hvcall()
229 		ret = hv_call_deposit_pages(NUMA_NO_NODE, partition->pt_id, 1);  in mshv_ioctl_passthru_hvcall()
888 mshv_partition_ioctl_create_vp(struct mshv_partition *partition,  in mshv_partition_ioctl_create_vp()  argument
903 	if (partition->pt_vp_array[args.vp_index])  in mshv_partition_ioctl_create_vp()
906 	ret = hv_call_create_vp(NUMA_NO_NODE, partition->pt_id, args.vp_index,  in mshv_partition_ioctl_create_vp()
907 				0 /* Only valid for root partition VPs */);  in mshv_partition_ioctl_create_vp()
911 	ret = hv_call_map_vp_state_page(partition->pt_id, args.vp_index,  in mshv_partition_ioctl_create_vp()
918 	if (!mshv_partition_encrypted(partition)) {  in mshv_partition_ioctl_create_vp()
919 		ret = hv_call_map_vp_state_page(partition->pt_id, args.vp_index,  in mshv_partition_ioctl_create_vp()
927 	if (mshv_partition_encrypted(partition) &&  in mshv_partition_ioctl_create_vp()
929 		ret = hv_call_map_vp_state_page(partition->pt_id, args.vp_index,  in mshv_partition_ioctl_create_vp()
938 		ret = mshv_vp_stats_map(partition->pt_id, args.vp_index,  in mshv_partition_ioctl_create_vp()
948 	vp->vp_partition = mshv_partition_get(partition);  in mshv_partition_ioctl_create_vp()
960 	if (!mshv_partition_encrypted(partition))  in mshv_partition_ioctl_create_vp()
963 	if (mshv_partition_encrypted(partition) && is_ghcb_mapping_available())  in mshv_partition_ioctl_create_vp()
978 	/* already exclusive with the partition mutex for all ioctls */  in mshv_partition_ioctl_create_vp()
979 	partition->pt_vp_count++;  in mshv_partition_ioctl_create_vp()
980 	partition->pt_vp_array[args.vp_index] = vp;  in mshv_partition_ioctl_create_vp()
985 	mshv_partition_put(partition);  in mshv_partition_ioctl_create_vp()
990 		mshv_vp_stats_unmap(partition->pt_id, args.vp_index);  in mshv_partition_ioctl_create_vp()
992 	if (mshv_partition_encrypted(partition) && is_ghcb_mapping_available()) {  in mshv_partition_ioctl_create_vp()
993 		hv_call_unmap_vp_state_page(partition->pt_id, args.vp_index,  in mshv_partition_ioctl_create_vp()
998 	if (!mshv_partition_encrypted(partition)) {  in mshv_partition_ioctl_create_vp()
999 		hv_call_unmap_vp_state_page(partition->pt_id, args.vp_index,  in mshv_partition_ioctl_create_vp()
1004 	hv_call_unmap_vp_state_page(partition->pt_id, args.vp_index,  in mshv_partition_ioctl_create_vp()
1008 	hv_call_delete_vp(partition->pt_id, args.vp_index);  in mshv_partition_ioctl_create_vp()
1012 static int mshv_init_async_handler(struct mshv_partition *partition)  in mshv_init_async_handler()  argument
1014 	if (completion_done(&partition->async_hypercall)) {  in mshv_init_async_handler()
1015 		pt_err(partition,  in mshv_init_async_handler()
1020 	reinit_completion(&partition->async_hypercall);  in mshv_init_async_handler()
1026 	struct mshv_partition *partition = data;  in mshv_async_hvcall_handler()  local
1028 	wait_for_completion(&partition->async_hypercall);  in mshv_async_hvcall_handler()
1029 	pt_dbg(partition, "Async hypercall completed!\n");  in mshv_async_hvcall_handler()
1031 	*status = partition->async_hypercall_status;  in mshv_async_hvcall_handler()
1042 	return hv_call_modify_spa_host_access(region->partition->pt_id,  in mshv_partition_region_share()
1056 	return hv_call_modify_spa_host_access(region->partition->pt_id,  in mshv_partition_region_unshare()
1073 	return hv_call_map_gpa_pages(region->partition->pt_id,  in mshv_region_remap_pages()
1162 mshv_partition_region_by_gfn(struct mshv_partition *partition, u64 gfn)  in mshv_partition_region_by_gfn()  argument
1166 	hlist_for_each_entry(region, &partition->pt_mem_regions, hnode) {  in mshv_partition_region_by_gfn()
1176 mshv_partition_region_by_uaddr(struct mshv_partition *partition, u64 uaddr)  in mshv_partition_region_by_uaddr()  argument
1180 	hlist_for_each_entry(region, &partition->pt_mem_regions, hnode) {  in mshv_partition_region_by_uaddr()
1194 static int mshv_partition_create_region(struct mshv_partition *partition,  in mshv_partition_create_region()  argument
1203 	if (mshv_partition_region_by_gfn(partition, mem->guest_pfn) ||  in mshv_partition_create_region()
1204 	    mshv_partition_region_by_gfn(partition, mem->guest_pfn + nr_pages - 1) ||  in mshv_partition_create_region()
1205 	    mshv_partition_region_by_uaddr(partition, mem->userspace_addr) ||  in mshv_partition_create_region()
1206 	    mshv_partition_region_by_uaddr(partition, mem->userspace_addr + mem->size - 1))  in mshv_partition_create_region()
1226 	region->partition = partition;  in mshv_partition_create_region()
1240 	struct mshv_partition *partition = region->partition;  in mshv_partition_mem_region_map()  local
1245 		pt_err(partition, "Failed to populate memory region: %d\n",  in mshv_partition_mem_region_map()
1251 	 * For an SNP partition it is a requirement that for every memory region  in mshv_partition_mem_region_map()
1252 	 * that we are going to map for this partition we should make sure that  in mshv_partition_mem_region_map()
1257 	if (mshv_partition_encrypted(partition)) {  in mshv_partition_mem_region_map()
1260 			pt_err(partition,  in mshv_partition_mem_region_map()
1268 	if (ret && mshv_partition_encrypted(partition)) {  in mshv_partition_mem_region_map()
1275 		pt_err(partition,  in mshv_partition_mem_region_map()
1306 mshv_map_user_memory(struct mshv_partition *partition,  in mshv_map_user_memory()  argument
1328 	ret = mshv_partition_create_region(partition, &mem, &region,  in mshv_map_user_memory()
1334 		ret = hv_call_map_mmio_pages(partition->pt_id, mem.guest_pfn,  in mshv_map_user_memory()
1343 	hlist_add_head(&region->hnode, &partition->pt_mem_regions);  in mshv_map_user_memory()
1354 mshv_unmap_user_memory(struct mshv_partition *partition,  in mshv_unmap_user_memory()  argument
1363 	region = mshv_partition_region_by_gfn(partition, mem.guest_pfn);  in mshv_unmap_user_memory()
1379 	hv_call_unmap_gpa_pages(partition->pt_id, region->start_gfn,  in mshv_unmap_user_memory()
1389 mshv_partition_ioctl_set_memory(struct mshv_partition *partition,  in mshv_partition_ioctl_set_memory()  argument
1405 		return mshv_unmap_user_memory(partition, mem);  in mshv_partition_ioctl_set_memory()
1407 	return mshv_map_user_memory(partition, mem);  in mshv_partition_ioctl_set_memory()
1411 mshv_partition_ioctl_ioeventfd(struct mshv_partition *partition,  in mshv_partition_ioctl_ioeventfd()  argument
1419 	return mshv_set_unset_ioeventfd(partition, &args);  in mshv_partition_ioctl_ioeventfd()
1423 mshv_partition_ioctl_irqfd(struct mshv_partition *partition,  in mshv_partition_ioctl_irqfd()  argument
1431 	return mshv_set_unset_irqfd(partition, &args);  in mshv_partition_ioctl_irqfd()
1435 mshv_partition_ioctl_get_gpap_access_bitmap(struct mshv_partition *partition,  in mshv_partition_ioctl_get_gpap_access_bitmap()  argument
1492 	ret = hv_call_get_gpa_access_states(partition->pt_id, args.page_count,  in mshv_partition_ioctl_get_gpap_access_bitmap()
1519 mshv_partition_ioctl_set_msi_routing(struct mshv_partition *partition,  in mshv_partition_ioctl_set_msi_routing()  argument
1542 	ret = mshv_update_routing_table(partition, entries, args.nr);  in mshv_partition_ioctl_set_msi_routing()
1549 mshv_partition_ioctl_initialize(struct mshv_partition *partition)  in mshv_partition_ioctl_initialize()  argument
1553 	if (partition->pt_initialized)  in mshv_partition_ioctl_initialize()
1556 	ret = hv_call_initialize_partition(partition->pt_id);  in mshv_partition_ioctl_initialize()
1560 	partition->pt_initialized = true;  in mshv_partition_ioctl_initialize()
1565 	hv_call_withdraw_memory(U64_MAX, NUMA_NO_NODE, partition->pt_id);  in mshv_partition_ioctl_initialize()
1573 	struct mshv_partition *partition = filp->private_data;  in mshv_partition_ioctl()  local
1577 	if (mutex_lock_killable(&partition->pt_mutex))  in mshv_partition_ioctl()
1582 		ret = mshv_partition_ioctl_initialize(partition);  in mshv_partition_ioctl()
1585 		ret = mshv_partition_ioctl_set_memory(partition, uarg);  in mshv_partition_ioctl()
1588 		ret = mshv_partition_ioctl_create_vp(partition, uarg);  in mshv_partition_ioctl()
1591 		ret = mshv_partition_ioctl_irqfd(partition, uarg);  in mshv_partition_ioctl()
1594 		ret = mshv_partition_ioctl_ioeventfd(partition, uarg);  in mshv_partition_ioctl()
1597 		ret = mshv_partition_ioctl_set_msi_routing(partition, uarg);  in mshv_partition_ioctl()
1600 		ret = mshv_partition_ioctl_get_gpap_access_bitmap(partition,  in mshv_partition_ioctl()
1604 		ret = mshv_ioctl_passthru_hvcall(partition, true, uarg);  in mshv_partition_ioctl()
1610 	mutex_unlock(&partition->pt_mutex);  in mshv_partition_ioctl()
1678 static void drain_all_vps(const struct mshv_partition *partition)  in drain_all_vps()  argument
1684 	 * VPs are reachable from ISR. It is safe to not take the partition  in drain_all_vps()
1686 	 * partition from the list.  in drain_all_vps()
1689 		vp = partition->pt_vp_array[i];  in drain_all_vps()
1703 remove_partition(struct mshv_partition *partition)  in remove_partition()  argument
1706 	hlist_del_rcu(&partition->pt_hnode);  in remove_partition()
1713  * Tear down a partition and remove it from the list.
1714  * Partition's refcount must be 0
1716 static void destroy_partition(struct mshv_partition *partition)  in destroy_partition()  argument
1723 	if (refcount_read(&partition->pt_ref_count)) {  in destroy_partition()
1724 		pt_err(partition,  in destroy_partition()
1725 		       "Attempt to destroy partition but refcount > 0\n");  in destroy_partition()
1729 	if (partition->pt_initialized) {  in destroy_partition()
1732 		 * done before removing the partition from the partition list.  in destroy_partition()
1735 			drain_all_vps(partition);  in destroy_partition()
1739 			vp = partition->pt_vp_array[i];  in destroy_partition()
1744 				mshv_vp_stats_unmap(partition->pt_id, vp->vp_index);  in destroy_partition()
1747 				(void)hv_call_unmap_vp_state_page(partition->pt_id,  in destroy_partition()
1754 			(void)hv_call_unmap_vp_state_page(partition->pt_id,  in destroy_partition()
1761 				(void)hv_call_unmap_vp_state_page(partition->pt_id,  in destroy_partition()
1770 			partition->pt_vp_array[i] = NULL;  in destroy_partition()
1774 		hv_call_finalize_partition(partition->pt_id);  in destroy_partition()
1776 		partition->pt_initialized = false;  in destroy_partition()
1779 	remove_partition(partition);  in destroy_partition()
1782 	hlist_for_each_entry_safe(region, n, &partition->pt_mem_regions,  in destroy_partition()
1786 		if (mshv_partition_encrypted(partition)) {  in destroy_partition()
1789 				pt_err(partition,  in destroy_partition()
1802 	hv_call_withdraw_memory(U64_MAX, NUMA_NO_NODE, partition->pt_id);  in destroy_partition()
1803 	hv_call_delete_partition(partition->pt_id);  in destroy_partition()
1805 	mshv_free_routing_table(partition);  in destroy_partition()
1806 	kfree(partition);  in destroy_partition()
1810 mshv_partition *mshv_partition_get(struct mshv_partition *partition)  in mshv_partition_get()  argument
1812 	if (refcount_inc_not_zero(&partition->pt_ref_count))  in mshv_partition_get()
1813 		return partition;  in mshv_partition_get()
1832 mshv_partition_put(struct mshv_partition *partition)  in mshv_partition_put()  argument
1834 	if (refcount_dec_and_test(&partition->pt_ref_count))  in mshv_partition_put()
1835 		destroy_partition(partition);  in mshv_partition_put()
1841 	struct mshv_partition *partition = filp->private_data;  in mshv_partition_release()  local
1843 	mshv_eventfd_release(partition);  in mshv_partition_release()
1845 	cleanup_srcu_struct(&partition->pt_irq_srcu);  in mshv_partition_release()
1847 	mshv_partition_put(partition);  in mshv_partition_release()
1853 add_partition(struct mshv_partition *partition)  in add_partition()  argument
1857 	hash_add_rcu(mshv_root.pt_htable, &partition->pt_hnode,  in add_partition()
1858 		     partition->pt_id);  in add_partition()
1872 	struct mshv_partition *partition;  in mshv_ioctl_create_partition()  local
1902 	partition = kzalloc(sizeof(*partition), GFP_KERNEL);  in mshv_ioctl_create_partition()
1903 	if (!partition)  in mshv_ioctl_create_partition()
1906 	partition->pt_module_dev = module_dev;  in mshv_ioctl_create_partition()
1907 	partition->isolation_type = isolation_properties.isolation_type;  in mshv_ioctl_create_partition()
1909 	refcount_set(&partition->pt_ref_count, 1);  in mshv_ioctl_create_partition()
1911 	mutex_init(&partition->pt_mutex);  in mshv_ioctl_create_partition()
1913 	mutex_init(&partition->pt_irq_lock);  in mshv_ioctl_create_partition()
1915 	init_completion(&partition->async_hypercall);  in mshv_ioctl_create_partition()
1917 	INIT_HLIST_HEAD(&partition->irq_ack_notifier_list);  in mshv_ioctl_create_partition()
1919 	INIT_HLIST_HEAD(&partition->pt_devices);  in mshv_ioctl_create_partition()
1921 	INIT_HLIST_HEAD(&partition->pt_mem_regions);  in mshv_ioctl_create_partition()
1923 	mshv_eventfd_init(partition);  in mshv_ioctl_create_partition()
1925 	ret = init_srcu_struct(&partition->pt_irq_srcu);  in mshv_ioctl_create_partition()
1932 				       &partition->pt_id);  in mshv_ioctl_create_partition()
1936 	ret = add_partition(partition);  in mshv_ioctl_create_partition()
1940 	ret = mshv_init_async_handler(partition);  in mshv_ioctl_create_partition()
1951 				  partition, O_RDWR);  in mshv_ioctl_create_partition()
1964 	remove_partition(partition);  in mshv_ioctl_create_partition()
1966 	hv_call_delete_partition(partition->pt_id);  in mshv_ioctl_create_partition()
1968 	cleanup_srcu_struct(&partition->pt_irq_srcu);  in mshv_ioctl_create_partition()
1970 	kfree(partition);  in mshv_ioctl_create_partition()