106164d2bSGeorge Zhang /* 206164d2bSGeorge Zhang * VMware VMCI Driver 306164d2bSGeorge Zhang * 406164d2bSGeorge Zhang * Copyright (C) 2012 VMware, Inc. All rights reserved. 506164d2bSGeorge Zhang * 606164d2bSGeorge Zhang * This program is free software; you can redistribute it and/or modify it 706164d2bSGeorge Zhang * under the terms of the GNU General Public License as published by the 806164d2bSGeorge Zhang * Free Software Foundation version 2 and no later version. 906164d2bSGeorge Zhang * 1006164d2bSGeorge Zhang * This program is distributed in the hope that it will be useful, but 1106164d2bSGeorge Zhang * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY 1206164d2bSGeorge Zhang * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 1306164d2bSGeorge Zhang * for more details. 1406164d2bSGeorge Zhang */ 1506164d2bSGeorge Zhang 1606164d2bSGeorge Zhang #include <linux/device-mapper.h> 1706164d2bSGeorge Zhang #include <linux/vmw_vmci_defs.h> 1806164d2bSGeorge Zhang #include <linux/vmw_vmci_api.h> 1906164d2bSGeorge Zhang #include <linux/kernel.h> 2006164d2bSGeorge Zhang #include <linux/module.h> 2106164d2bSGeorge Zhang #include <linux/mutex.h> 2206164d2bSGeorge Zhang #include <linux/socket.h> 2306164d2bSGeorge Zhang #include <linux/wait.h> 2406164d2bSGeorge Zhang 2506164d2bSGeorge Zhang #include "vmci_handle_array.h" 2606164d2bSGeorge Zhang #include "vmci_queue_pair.h" 2706164d2bSGeorge Zhang #include "vmci_datagram.h" 2806164d2bSGeorge Zhang #include "vmci_resource.h" 2906164d2bSGeorge Zhang #include "vmci_context.h" 3006164d2bSGeorge Zhang #include "vmci_driver.h" 3106164d2bSGeorge Zhang #include "vmci_event.h" 3206164d2bSGeorge Zhang #include "vmci_route.h" 3306164d2bSGeorge Zhang 3406164d2bSGeorge Zhang /* 3506164d2bSGeorge Zhang * In the following, we will distinguish between two kinds of VMX processes - 3606164d2bSGeorge Zhang * the ones with versions lower than VMCI_VERSION_NOVMVM that use specialized 3706164d2bSGeorge Zhang * VMCI page files in the VMX and supporting VM to VM communication and the 3806164d2bSGeorge Zhang * newer ones that use the guest memory directly. We will in the following 3906164d2bSGeorge Zhang * refer to the older VMX versions as old-style VMX'en, and the newer ones as 4006164d2bSGeorge Zhang * new-style VMX'en. 4106164d2bSGeorge Zhang * 4206164d2bSGeorge Zhang * The state transition datagram is as follows (the VMCIQPB_ prefix has been 4306164d2bSGeorge Zhang * removed for readability) - see below for more details on the transtions: 4406164d2bSGeorge Zhang * 4506164d2bSGeorge Zhang * -------------- NEW ------------- 4606164d2bSGeorge Zhang * | | 4706164d2bSGeorge Zhang * \_/ \_/ 4806164d2bSGeorge Zhang * CREATED_NO_MEM <-----------------> CREATED_MEM 4906164d2bSGeorge Zhang * | | | 5006164d2bSGeorge Zhang * | o-----------------------o | 5106164d2bSGeorge Zhang * | | | 5206164d2bSGeorge Zhang * \_/ \_/ \_/ 5306164d2bSGeorge Zhang * ATTACHED_NO_MEM <----------------> ATTACHED_MEM 5406164d2bSGeorge Zhang * | | | 5506164d2bSGeorge Zhang * | o----------------------o | 5606164d2bSGeorge Zhang * | | | 5706164d2bSGeorge Zhang * \_/ \_/ \_/ 5806164d2bSGeorge Zhang * SHUTDOWN_NO_MEM <----------------> SHUTDOWN_MEM 5906164d2bSGeorge Zhang * | | 6006164d2bSGeorge Zhang * | | 6106164d2bSGeorge Zhang * -------------> gone <------------- 6206164d2bSGeorge Zhang * 6306164d2bSGeorge Zhang * In more detail. When a VMCI queue pair is first created, it will be in the 6406164d2bSGeorge Zhang * VMCIQPB_NEW state. It will then move into one of the following states: 6506164d2bSGeorge Zhang * 6606164d2bSGeorge Zhang * - VMCIQPB_CREATED_NO_MEM: this state indicates that either: 6706164d2bSGeorge Zhang * 6806164d2bSGeorge Zhang * - the created was performed by a host endpoint, in which case there is 6906164d2bSGeorge Zhang * no backing memory yet. 7006164d2bSGeorge Zhang * 7106164d2bSGeorge Zhang * - the create was initiated by an old-style VMX, that uses 7206164d2bSGeorge Zhang * vmci_qp_broker_set_page_store to specify the UVAs of the queue pair at 7306164d2bSGeorge Zhang * a later point in time. This state can be distinguished from the one 7406164d2bSGeorge Zhang * above by the context ID of the creator. A host side is not allowed to 7506164d2bSGeorge Zhang * attach until the page store has been set. 7606164d2bSGeorge Zhang * 7706164d2bSGeorge Zhang * - VMCIQPB_CREATED_MEM: this state is the result when the queue pair 7806164d2bSGeorge Zhang * is created by a VMX using the queue pair device backend that 7906164d2bSGeorge Zhang * sets the UVAs of the queue pair immediately and stores the 8006164d2bSGeorge Zhang * information for later attachers. At this point, it is ready for 8106164d2bSGeorge Zhang * the host side to attach to it. 8206164d2bSGeorge Zhang * 8306164d2bSGeorge Zhang * Once the queue pair is in one of the created states (with the exception of 8406164d2bSGeorge Zhang * the case mentioned for older VMX'en above), it is possible to attach to the 8506164d2bSGeorge Zhang * queue pair. Again we have two new states possible: 8606164d2bSGeorge Zhang * 8706164d2bSGeorge Zhang * - VMCIQPB_ATTACHED_MEM: this state can be reached through the following 8806164d2bSGeorge Zhang * paths: 8906164d2bSGeorge Zhang * 9006164d2bSGeorge Zhang * - from VMCIQPB_CREATED_NO_MEM when a new-style VMX allocates a queue 9106164d2bSGeorge Zhang * pair, and attaches to a queue pair previously created by the host side. 9206164d2bSGeorge Zhang * 9306164d2bSGeorge Zhang * - from VMCIQPB_CREATED_MEM when the host side attaches to a queue pair 9406164d2bSGeorge Zhang * already created by a guest. 9506164d2bSGeorge Zhang * 9606164d2bSGeorge Zhang * - from VMCIQPB_ATTACHED_NO_MEM, when an old-style VMX calls 9706164d2bSGeorge Zhang * vmci_qp_broker_set_page_store (see below). 9806164d2bSGeorge Zhang * 9906164d2bSGeorge Zhang * - VMCIQPB_ATTACHED_NO_MEM: If the queue pair already was in the 10006164d2bSGeorge Zhang * VMCIQPB_CREATED_NO_MEM due to a host side create, an old-style VMX will 10106164d2bSGeorge Zhang * bring the queue pair into this state. Once vmci_qp_broker_set_page_store 10206164d2bSGeorge Zhang * is called to register the user memory, the VMCIQPB_ATTACH_MEM state 10306164d2bSGeorge Zhang * will be entered. 10406164d2bSGeorge Zhang * 10506164d2bSGeorge Zhang * From the attached queue pair, the queue pair can enter the shutdown states 10606164d2bSGeorge Zhang * when either side of the queue pair detaches. If the guest side detaches 10706164d2bSGeorge Zhang * first, the queue pair will enter the VMCIQPB_SHUTDOWN_NO_MEM state, where 10806164d2bSGeorge Zhang * the content of the queue pair will no longer be available. If the host 10906164d2bSGeorge Zhang * side detaches first, the queue pair will either enter the 11006164d2bSGeorge Zhang * VMCIQPB_SHUTDOWN_MEM, if the guest memory is currently mapped, or 11106164d2bSGeorge Zhang * VMCIQPB_SHUTDOWN_NO_MEM, if the guest memory is not mapped 11206164d2bSGeorge Zhang * (e.g., the host detaches while a guest is stunned). 11306164d2bSGeorge Zhang * 11406164d2bSGeorge Zhang * New-style VMX'en will also unmap guest memory, if the guest is 11506164d2bSGeorge Zhang * quiesced, e.g., during a snapshot operation. In that case, the guest 11606164d2bSGeorge Zhang * memory will no longer be available, and the queue pair will transition from 11706164d2bSGeorge Zhang * *_MEM state to a *_NO_MEM state. The VMX may later map the memory once more, 11806164d2bSGeorge Zhang * in which case the queue pair will transition from the *_NO_MEM state at that 11906164d2bSGeorge Zhang * point back to the *_MEM state. Note that the *_NO_MEM state may have changed, 12006164d2bSGeorge Zhang * since the peer may have either attached or detached in the meantime. The 12106164d2bSGeorge Zhang * values are laid out such that ++ on a state will move from a *_NO_MEM to a 12206164d2bSGeorge Zhang * *_MEM state, and vice versa. 12306164d2bSGeorge Zhang */ 12406164d2bSGeorge Zhang 12506164d2bSGeorge Zhang /* 12606164d2bSGeorge Zhang * VMCIMemcpy{To,From}QueueFunc() prototypes. Functions of these 12706164d2bSGeorge Zhang * types are passed around to enqueue and dequeue routines. Note that 12806164d2bSGeorge Zhang * often the functions passed are simply wrappers around memcpy 12906164d2bSGeorge Zhang * itself. 13006164d2bSGeorge Zhang * 13106164d2bSGeorge Zhang * Note: In order for the memcpy typedefs to be compatible with the VMKernel, 13206164d2bSGeorge Zhang * there's an unused last parameter for the hosted side. In 13306164d2bSGeorge Zhang * ESX, that parameter holds a buffer type. 13406164d2bSGeorge Zhang */ 13506164d2bSGeorge Zhang typedef int vmci_memcpy_to_queue_func(struct vmci_queue *queue, 13606164d2bSGeorge Zhang u64 queue_offset, const void *src, 13706164d2bSGeorge Zhang size_t src_offset, size_t size); 13806164d2bSGeorge Zhang typedef int vmci_memcpy_from_queue_func(void *dest, size_t dest_offset, 13906164d2bSGeorge Zhang const struct vmci_queue *queue, 14006164d2bSGeorge Zhang u64 queue_offset, size_t size); 14106164d2bSGeorge Zhang 14206164d2bSGeorge Zhang /* The Kernel specific component of the struct vmci_queue structure. */ 14306164d2bSGeorge Zhang struct vmci_queue_kern_if { 14406164d2bSGeorge Zhang struct page **page; 14506164d2bSGeorge Zhang struct page **header_page; 14606164d2bSGeorge Zhang void *va; 14706164d2bSGeorge Zhang struct mutex __mutex; /* Protects the queue. */ 14806164d2bSGeorge Zhang struct mutex *mutex; /* Shared by producer and consumer queues. */ 14906164d2bSGeorge Zhang bool host; 15006164d2bSGeorge Zhang size_t num_pages; 15106164d2bSGeorge Zhang bool mapped; 15206164d2bSGeorge Zhang }; 15306164d2bSGeorge Zhang 15406164d2bSGeorge Zhang /* 15506164d2bSGeorge Zhang * This structure is opaque to the clients. 15606164d2bSGeorge Zhang */ 15706164d2bSGeorge Zhang struct vmci_qp { 15806164d2bSGeorge Zhang struct vmci_handle handle; 15906164d2bSGeorge Zhang struct vmci_queue *produce_q; 16006164d2bSGeorge Zhang struct vmci_queue *consume_q; 16106164d2bSGeorge Zhang u64 produce_q_size; 16206164d2bSGeorge Zhang u64 consume_q_size; 16306164d2bSGeorge Zhang u32 peer; 16406164d2bSGeorge Zhang u32 flags; 16506164d2bSGeorge Zhang u32 priv_flags; 16606164d2bSGeorge Zhang bool guest_endpoint; 16706164d2bSGeorge Zhang unsigned int blocked; 16806164d2bSGeorge Zhang unsigned int generation; 16906164d2bSGeorge Zhang wait_queue_head_t event; 17006164d2bSGeorge Zhang }; 17106164d2bSGeorge Zhang 17206164d2bSGeorge Zhang enum qp_broker_state { 17306164d2bSGeorge Zhang VMCIQPB_NEW, 17406164d2bSGeorge Zhang VMCIQPB_CREATED_NO_MEM, 17506164d2bSGeorge Zhang VMCIQPB_CREATED_MEM, 17606164d2bSGeorge Zhang VMCIQPB_ATTACHED_NO_MEM, 17706164d2bSGeorge Zhang VMCIQPB_ATTACHED_MEM, 17806164d2bSGeorge Zhang VMCIQPB_SHUTDOWN_NO_MEM, 17906164d2bSGeorge Zhang VMCIQPB_SHUTDOWN_MEM, 18006164d2bSGeorge Zhang VMCIQPB_GONE 18106164d2bSGeorge Zhang }; 18206164d2bSGeorge Zhang 18306164d2bSGeorge Zhang #define QPBROKERSTATE_HAS_MEM(_qpb) (_qpb->state == VMCIQPB_CREATED_MEM || \ 18406164d2bSGeorge Zhang _qpb->state == VMCIQPB_ATTACHED_MEM || \ 18506164d2bSGeorge Zhang _qpb->state == VMCIQPB_SHUTDOWN_MEM) 18606164d2bSGeorge Zhang 18706164d2bSGeorge Zhang /* 18806164d2bSGeorge Zhang * In the queue pair broker, we always use the guest point of view for 18906164d2bSGeorge Zhang * the produce and consume queue values and references, e.g., the 19006164d2bSGeorge Zhang * produce queue size stored is the guests produce queue size. The 19106164d2bSGeorge Zhang * host endpoint will need to swap these around. The only exception is 19206164d2bSGeorge Zhang * the local queue pairs on the host, in which case the host endpoint 19306164d2bSGeorge Zhang * that creates the queue pair will have the right orientation, and 19406164d2bSGeorge Zhang * the attaching host endpoint will need to swap. 19506164d2bSGeorge Zhang */ 19606164d2bSGeorge Zhang struct qp_entry { 19706164d2bSGeorge Zhang struct list_head list_item; 19806164d2bSGeorge Zhang struct vmci_handle handle; 19906164d2bSGeorge Zhang u32 peer; 20006164d2bSGeorge Zhang u32 flags; 20106164d2bSGeorge Zhang u64 produce_size; 20206164d2bSGeorge Zhang u64 consume_size; 20306164d2bSGeorge Zhang u32 ref_count; 20406164d2bSGeorge Zhang }; 20506164d2bSGeorge Zhang 20606164d2bSGeorge Zhang struct qp_broker_entry { 20706164d2bSGeorge Zhang struct vmci_resource resource; 20806164d2bSGeorge Zhang struct qp_entry qp; 20906164d2bSGeorge Zhang u32 create_id; 21006164d2bSGeorge Zhang u32 attach_id; 21106164d2bSGeorge Zhang enum qp_broker_state state; 21206164d2bSGeorge Zhang bool require_trusted_attach; 21306164d2bSGeorge Zhang bool created_by_trusted; 21406164d2bSGeorge Zhang bool vmci_page_files; /* Created by VMX using VMCI page files */ 21506164d2bSGeorge Zhang struct vmci_queue *produce_q; 21606164d2bSGeorge Zhang struct vmci_queue *consume_q; 21706164d2bSGeorge Zhang struct vmci_queue_header saved_produce_q; 21806164d2bSGeorge Zhang struct vmci_queue_header saved_consume_q; 21906164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb; 22006164d2bSGeorge Zhang void *client_data; 22106164d2bSGeorge Zhang void *local_mem; /* Kernel memory for local queue pair */ 22206164d2bSGeorge Zhang }; 22306164d2bSGeorge Zhang 22406164d2bSGeorge Zhang struct qp_guest_endpoint { 22506164d2bSGeorge Zhang struct vmci_resource resource; 22606164d2bSGeorge Zhang struct qp_entry qp; 22706164d2bSGeorge Zhang u64 num_ppns; 22806164d2bSGeorge Zhang void *produce_q; 22906164d2bSGeorge Zhang void *consume_q; 23006164d2bSGeorge Zhang struct PPNSet ppn_set; 23106164d2bSGeorge Zhang }; 23206164d2bSGeorge Zhang 23306164d2bSGeorge Zhang struct qp_list { 23406164d2bSGeorge Zhang struct list_head head; 23506164d2bSGeorge Zhang struct mutex mutex; /* Protect queue list. */ 23606164d2bSGeorge Zhang }; 23706164d2bSGeorge Zhang 23806164d2bSGeorge Zhang static struct qp_list qp_broker_list = { 23906164d2bSGeorge Zhang .head = LIST_HEAD_INIT(qp_broker_list.head), 24006164d2bSGeorge Zhang .mutex = __MUTEX_INITIALIZER(qp_broker_list.mutex), 24106164d2bSGeorge Zhang }; 24206164d2bSGeorge Zhang 24306164d2bSGeorge Zhang static struct qp_list qp_guest_endpoints = { 24406164d2bSGeorge Zhang .head = LIST_HEAD_INIT(qp_guest_endpoints.head), 24506164d2bSGeorge Zhang .mutex = __MUTEX_INITIALIZER(qp_guest_endpoints.mutex), 24606164d2bSGeorge Zhang }; 24706164d2bSGeorge Zhang 24806164d2bSGeorge Zhang #define INVALID_VMCI_GUEST_MEM_ID 0 24906164d2bSGeorge Zhang #define QPE_NUM_PAGES(_QPE) ((u32) \ 25006164d2bSGeorge Zhang (dm_div_up(_QPE.produce_size, PAGE_SIZE) + \ 25106164d2bSGeorge Zhang dm_div_up(_QPE.consume_size, PAGE_SIZE) + 2)) 25206164d2bSGeorge Zhang 25306164d2bSGeorge Zhang 25406164d2bSGeorge Zhang /* 25506164d2bSGeorge Zhang * Frees kernel VA space for a given queue and its queue header, and 25606164d2bSGeorge Zhang * frees physical data pages. 25706164d2bSGeorge Zhang */ 25806164d2bSGeorge Zhang static void qp_free_queue(void *q, u64 size) 25906164d2bSGeorge Zhang { 26006164d2bSGeorge Zhang struct vmci_queue *queue = q; 26106164d2bSGeorge Zhang 26206164d2bSGeorge Zhang if (queue) { 26306164d2bSGeorge Zhang u64 i = dm_div_up(size, PAGE_SIZE); 26406164d2bSGeorge Zhang 26506164d2bSGeorge Zhang if (queue->kernel_if->mapped) { 26606164d2bSGeorge Zhang vunmap(queue->kernel_if->va); 26706164d2bSGeorge Zhang queue->kernel_if->va = NULL; 26806164d2bSGeorge Zhang } 26906164d2bSGeorge Zhang 27006164d2bSGeorge Zhang while (i) 27106164d2bSGeorge Zhang __free_page(queue->kernel_if->page[--i]); 27206164d2bSGeorge Zhang 27306164d2bSGeorge Zhang vfree(queue->q_header); 27406164d2bSGeorge Zhang } 27506164d2bSGeorge Zhang } 27606164d2bSGeorge Zhang 27706164d2bSGeorge Zhang /* 27806164d2bSGeorge Zhang * Allocates kernel VA space of specified size, plus space for the 27906164d2bSGeorge Zhang * queue structure/kernel interface and the queue header. Allocates 28006164d2bSGeorge Zhang * physical pages for the queue data pages. 28106164d2bSGeorge Zhang * 28206164d2bSGeorge Zhang * PAGE m: struct vmci_queue_header (struct vmci_queue->q_header) 28306164d2bSGeorge Zhang * PAGE m+1: struct vmci_queue 28406164d2bSGeorge Zhang * PAGE m+1+q: struct vmci_queue_kern_if (struct vmci_queue->kernel_if) 28506164d2bSGeorge Zhang * PAGE n-size: Data pages (struct vmci_queue->kernel_if->page[]) 28606164d2bSGeorge Zhang */ 28706164d2bSGeorge Zhang static void *qp_alloc_queue(u64 size, u32 flags) 28806164d2bSGeorge Zhang { 28906164d2bSGeorge Zhang u64 i; 29006164d2bSGeorge Zhang struct vmci_queue *queue; 29106164d2bSGeorge Zhang struct vmci_queue_header *q_header; 29206164d2bSGeorge Zhang const u64 num_data_pages = dm_div_up(size, PAGE_SIZE); 29306164d2bSGeorge Zhang const uint queue_size = 29406164d2bSGeorge Zhang PAGE_SIZE + 29506164d2bSGeorge Zhang sizeof(*queue) + sizeof(*(queue->kernel_if)) + 29606164d2bSGeorge Zhang num_data_pages * sizeof(*(queue->kernel_if->page)); 29706164d2bSGeorge Zhang 29806164d2bSGeorge Zhang q_header = vmalloc(queue_size); 29906164d2bSGeorge Zhang if (!q_header) 30006164d2bSGeorge Zhang return NULL; 30106164d2bSGeorge Zhang 30206164d2bSGeorge Zhang queue = (void *)q_header + PAGE_SIZE; 30306164d2bSGeorge Zhang queue->q_header = q_header; 30406164d2bSGeorge Zhang queue->saved_header = NULL; 30506164d2bSGeorge Zhang queue->kernel_if = (struct vmci_queue_kern_if *)(queue + 1); 30606164d2bSGeorge Zhang queue->kernel_if->header_page = NULL; /* Unused in guest. */ 30706164d2bSGeorge Zhang queue->kernel_if->page = (struct page **)(queue->kernel_if + 1); 30806164d2bSGeorge Zhang queue->kernel_if->host = false; 30906164d2bSGeorge Zhang queue->kernel_if->va = NULL; 31006164d2bSGeorge Zhang queue->kernel_if->mapped = false; 31106164d2bSGeorge Zhang 31206164d2bSGeorge Zhang for (i = 0; i < num_data_pages; i++) { 31306164d2bSGeorge Zhang queue->kernel_if->page[i] = alloc_pages(GFP_KERNEL, 0); 31406164d2bSGeorge Zhang if (!queue->kernel_if->page[i]) 31506164d2bSGeorge Zhang goto fail; 31606164d2bSGeorge Zhang } 31706164d2bSGeorge Zhang 31806164d2bSGeorge Zhang if (vmci_qp_pinned(flags)) { 31906164d2bSGeorge Zhang queue->kernel_if->va = 32006164d2bSGeorge Zhang vmap(queue->kernel_if->page, num_data_pages, VM_MAP, 32106164d2bSGeorge Zhang PAGE_KERNEL); 32206164d2bSGeorge Zhang if (!queue->kernel_if->va) 32306164d2bSGeorge Zhang goto fail; 32406164d2bSGeorge Zhang 32506164d2bSGeorge Zhang queue->kernel_if->mapped = true; 32606164d2bSGeorge Zhang } 32706164d2bSGeorge Zhang 32806164d2bSGeorge Zhang return (void *)queue; 32906164d2bSGeorge Zhang 33006164d2bSGeorge Zhang fail: 33106164d2bSGeorge Zhang qp_free_queue(queue, i * PAGE_SIZE); 33206164d2bSGeorge Zhang return NULL; 33306164d2bSGeorge Zhang } 33406164d2bSGeorge Zhang 33506164d2bSGeorge Zhang /* 33606164d2bSGeorge Zhang * Copies from a given buffer or iovector to a VMCI Queue. Uses 33706164d2bSGeorge Zhang * kmap()/kunmap() to dynamically map/unmap required portions of the queue 33806164d2bSGeorge Zhang * by traversing the offset -> page translation structure for the queue. 33906164d2bSGeorge Zhang * Assumes that offset + size does not wrap around in the queue. 34006164d2bSGeorge Zhang */ 34106164d2bSGeorge Zhang static int __qp_memcpy_to_queue(struct vmci_queue *queue, 34206164d2bSGeorge Zhang u64 queue_offset, 34306164d2bSGeorge Zhang const void *src, 34406164d2bSGeorge Zhang size_t size, 34506164d2bSGeorge Zhang bool is_iovec) 34606164d2bSGeorge Zhang { 34706164d2bSGeorge Zhang struct vmci_queue_kern_if *kernel_if = queue->kernel_if; 34806164d2bSGeorge Zhang size_t bytes_copied = 0; 34906164d2bSGeorge Zhang 35006164d2bSGeorge Zhang while (bytes_copied < size) { 35106164d2bSGeorge Zhang u64 page_index = (queue_offset + bytes_copied) / PAGE_SIZE; 35206164d2bSGeorge Zhang size_t page_offset = 35306164d2bSGeorge Zhang (queue_offset + bytes_copied) & (PAGE_SIZE - 1); 35406164d2bSGeorge Zhang void *va; 35506164d2bSGeorge Zhang size_t to_copy; 35606164d2bSGeorge Zhang 35706164d2bSGeorge Zhang if (!kernel_if->mapped) 35806164d2bSGeorge Zhang va = kmap(kernel_if->page[page_index]); 35906164d2bSGeorge Zhang else 36006164d2bSGeorge Zhang va = (void *)((u8 *)kernel_if->va + 36106164d2bSGeorge Zhang (page_index * PAGE_SIZE)); 36206164d2bSGeorge Zhang 36306164d2bSGeorge Zhang if (size - bytes_copied > PAGE_SIZE - page_offset) 36406164d2bSGeorge Zhang /* Enough payload to fill up from this page. */ 36506164d2bSGeorge Zhang to_copy = PAGE_SIZE - page_offset; 36606164d2bSGeorge Zhang else 36706164d2bSGeorge Zhang to_copy = size - bytes_copied; 36806164d2bSGeorge Zhang 36906164d2bSGeorge Zhang if (is_iovec) { 37006164d2bSGeorge Zhang struct iovec *iov = (struct iovec *)src; 37106164d2bSGeorge Zhang int err; 37206164d2bSGeorge Zhang 37306164d2bSGeorge Zhang /* The iovec will track bytes_copied internally. */ 37406164d2bSGeorge Zhang err = memcpy_fromiovec((u8 *)va + page_offset, 37506164d2bSGeorge Zhang iov, to_copy); 37606164d2bSGeorge Zhang if (err != 0) { 37706164d2bSGeorge Zhang kunmap(kernel_if->page[page_index]); 37806164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 37906164d2bSGeorge Zhang } 38006164d2bSGeorge Zhang } else { 38106164d2bSGeorge Zhang memcpy((u8 *)va + page_offset, 38206164d2bSGeorge Zhang (u8 *)src + bytes_copied, to_copy); 38306164d2bSGeorge Zhang } 38406164d2bSGeorge Zhang 38506164d2bSGeorge Zhang bytes_copied += to_copy; 38606164d2bSGeorge Zhang if (!kernel_if->mapped) 38706164d2bSGeorge Zhang kunmap(kernel_if->page[page_index]); 38806164d2bSGeorge Zhang } 38906164d2bSGeorge Zhang 39006164d2bSGeorge Zhang return VMCI_SUCCESS; 39106164d2bSGeorge Zhang } 39206164d2bSGeorge Zhang 39306164d2bSGeorge Zhang /* 39406164d2bSGeorge Zhang * Copies to a given buffer or iovector from a VMCI Queue. Uses 39506164d2bSGeorge Zhang * kmap()/kunmap() to dynamically map/unmap required portions of the queue 39606164d2bSGeorge Zhang * by traversing the offset -> page translation structure for the queue. 39706164d2bSGeorge Zhang * Assumes that offset + size does not wrap around in the queue. 39806164d2bSGeorge Zhang */ 39906164d2bSGeorge Zhang static int __qp_memcpy_from_queue(void *dest, 40006164d2bSGeorge Zhang const struct vmci_queue *queue, 40106164d2bSGeorge Zhang u64 queue_offset, 40206164d2bSGeorge Zhang size_t size, 40306164d2bSGeorge Zhang bool is_iovec) 40406164d2bSGeorge Zhang { 40506164d2bSGeorge Zhang struct vmci_queue_kern_if *kernel_if = queue->kernel_if; 40606164d2bSGeorge Zhang size_t bytes_copied = 0; 40706164d2bSGeorge Zhang 40806164d2bSGeorge Zhang while (bytes_copied < size) { 40906164d2bSGeorge Zhang u64 page_index = (queue_offset + bytes_copied) / PAGE_SIZE; 41006164d2bSGeorge Zhang size_t page_offset = 41106164d2bSGeorge Zhang (queue_offset + bytes_copied) & (PAGE_SIZE - 1); 41206164d2bSGeorge Zhang void *va; 41306164d2bSGeorge Zhang size_t to_copy; 41406164d2bSGeorge Zhang 41506164d2bSGeorge Zhang if (!kernel_if->mapped) 41606164d2bSGeorge Zhang va = kmap(kernel_if->page[page_index]); 41706164d2bSGeorge Zhang else 41806164d2bSGeorge Zhang va = (void *)((u8 *)kernel_if->va + 41906164d2bSGeorge Zhang (page_index * PAGE_SIZE)); 42006164d2bSGeorge Zhang 42106164d2bSGeorge Zhang if (size - bytes_copied > PAGE_SIZE - page_offset) 42206164d2bSGeorge Zhang /* Enough payload to fill up this page. */ 42306164d2bSGeorge Zhang to_copy = PAGE_SIZE - page_offset; 42406164d2bSGeorge Zhang else 42506164d2bSGeorge Zhang to_copy = size - bytes_copied; 42606164d2bSGeorge Zhang 42706164d2bSGeorge Zhang if (is_iovec) { 42806164d2bSGeorge Zhang struct iovec *iov = (struct iovec *)dest; 42906164d2bSGeorge Zhang int err; 43006164d2bSGeorge Zhang 43106164d2bSGeorge Zhang /* The iovec will track bytes_copied internally. */ 43206164d2bSGeorge Zhang err = memcpy_toiovec(iov, (u8 *)va + page_offset, 43306164d2bSGeorge Zhang to_copy); 43406164d2bSGeorge Zhang if (err != 0) { 43506164d2bSGeorge Zhang kunmap(kernel_if->page[page_index]); 43606164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 43706164d2bSGeorge Zhang } 43806164d2bSGeorge Zhang } else { 43906164d2bSGeorge Zhang memcpy((u8 *)dest + bytes_copied, 44006164d2bSGeorge Zhang (u8 *)va + page_offset, to_copy); 44106164d2bSGeorge Zhang } 44206164d2bSGeorge Zhang 44306164d2bSGeorge Zhang bytes_copied += to_copy; 44406164d2bSGeorge Zhang if (!kernel_if->mapped) 44506164d2bSGeorge Zhang kunmap(kernel_if->page[page_index]); 44606164d2bSGeorge Zhang } 44706164d2bSGeorge Zhang 44806164d2bSGeorge Zhang return VMCI_SUCCESS; 44906164d2bSGeorge Zhang } 45006164d2bSGeorge Zhang 45106164d2bSGeorge Zhang /* 45206164d2bSGeorge Zhang * Allocates two list of PPNs --- one for the pages in the produce queue, 45306164d2bSGeorge Zhang * and the other for the pages in the consume queue. Intializes the list 45406164d2bSGeorge Zhang * of PPNs with the page frame numbers of the KVA for the two queues (and 45506164d2bSGeorge Zhang * the queue headers). 45606164d2bSGeorge Zhang */ 45706164d2bSGeorge Zhang static int qp_alloc_ppn_set(void *prod_q, 45806164d2bSGeorge Zhang u64 num_produce_pages, 45906164d2bSGeorge Zhang void *cons_q, 46006164d2bSGeorge Zhang u64 num_consume_pages, struct PPNSet *ppn_set) 46106164d2bSGeorge Zhang { 46206164d2bSGeorge Zhang u32 *produce_ppns; 46306164d2bSGeorge Zhang u32 *consume_ppns; 46406164d2bSGeorge Zhang struct vmci_queue *produce_q = prod_q; 46506164d2bSGeorge Zhang struct vmci_queue *consume_q = cons_q; 46606164d2bSGeorge Zhang u64 i; 46706164d2bSGeorge Zhang 46806164d2bSGeorge Zhang if (!produce_q || !num_produce_pages || !consume_q || 46906164d2bSGeorge Zhang !num_consume_pages || !ppn_set) 47006164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 47106164d2bSGeorge Zhang 47206164d2bSGeorge Zhang if (ppn_set->initialized) 47306164d2bSGeorge Zhang return VMCI_ERROR_ALREADY_EXISTS; 47406164d2bSGeorge Zhang 47506164d2bSGeorge Zhang produce_ppns = 47606164d2bSGeorge Zhang kmalloc(num_produce_pages * sizeof(*produce_ppns), GFP_KERNEL); 47706164d2bSGeorge Zhang if (!produce_ppns) 47806164d2bSGeorge Zhang return VMCI_ERROR_NO_MEM; 47906164d2bSGeorge Zhang 48006164d2bSGeorge Zhang consume_ppns = 48106164d2bSGeorge Zhang kmalloc(num_consume_pages * sizeof(*consume_ppns), GFP_KERNEL); 48206164d2bSGeorge Zhang if (!consume_ppns) { 48306164d2bSGeorge Zhang kfree(produce_ppns); 48406164d2bSGeorge Zhang return VMCI_ERROR_NO_MEM; 48506164d2bSGeorge Zhang } 48606164d2bSGeorge Zhang 48706164d2bSGeorge Zhang produce_ppns[0] = page_to_pfn(vmalloc_to_page(produce_q->q_header)); 48806164d2bSGeorge Zhang for (i = 1; i < num_produce_pages; i++) { 48906164d2bSGeorge Zhang unsigned long pfn; 49006164d2bSGeorge Zhang 49106164d2bSGeorge Zhang produce_ppns[i] = 49206164d2bSGeorge Zhang page_to_pfn(produce_q->kernel_if->page[i - 1]); 49306164d2bSGeorge Zhang pfn = produce_ppns[i]; 49406164d2bSGeorge Zhang 49506164d2bSGeorge Zhang /* Fail allocation if PFN isn't supported by hypervisor. */ 49606164d2bSGeorge Zhang if (sizeof(pfn) > sizeof(*produce_ppns) 49706164d2bSGeorge Zhang && pfn != produce_ppns[i]) 49806164d2bSGeorge Zhang goto ppn_error; 49906164d2bSGeorge Zhang } 50006164d2bSGeorge Zhang 50106164d2bSGeorge Zhang consume_ppns[0] = page_to_pfn(vmalloc_to_page(consume_q->q_header)); 50206164d2bSGeorge Zhang for (i = 1; i < num_consume_pages; i++) { 50306164d2bSGeorge Zhang unsigned long pfn; 50406164d2bSGeorge Zhang 50506164d2bSGeorge Zhang consume_ppns[i] = 50606164d2bSGeorge Zhang page_to_pfn(consume_q->kernel_if->page[i - 1]); 50706164d2bSGeorge Zhang pfn = consume_ppns[i]; 50806164d2bSGeorge Zhang 50906164d2bSGeorge Zhang /* Fail allocation if PFN isn't supported by hypervisor. */ 51006164d2bSGeorge Zhang if (sizeof(pfn) > sizeof(*consume_ppns) 51106164d2bSGeorge Zhang && pfn != consume_ppns[i]) 51206164d2bSGeorge Zhang goto ppn_error; 51306164d2bSGeorge Zhang } 51406164d2bSGeorge Zhang 51506164d2bSGeorge Zhang ppn_set->num_produce_pages = num_produce_pages; 51606164d2bSGeorge Zhang ppn_set->num_consume_pages = num_consume_pages; 51706164d2bSGeorge Zhang ppn_set->produce_ppns = produce_ppns; 51806164d2bSGeorge Zhang ppn_set->consume_ppns = consume_ppns; 51906164d2bSGeorge Zhang ppn_set->initialized = true; 52006164d2bSGeorge Zhang return VMCI_SUCCESS; 52106164d2bSGeorge Zhang 52206164d2bSGeorge Zhang ppn_error: 52306164d2bSGeorge Zhang kfree(produce_ppns); 52406164d2bSGeorge Zhang kfree(consume_ppns); 52506164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 52606164d2bSGeorge Zhang } 52706164d2bSGeorge Zhang 52806164d2bSGeorge Zhang /* 52906164d2bSGeorge Zhang * Frees the two list of PPNs for a queue pair. 53006164d2bSGeorge Zhang */ 53106164d2bSGeorge Zhang static void qp_free_ppn_set(struct PPNSet *ppn_set) 53206164d2bSGeorge Zhang { 53306164d2bSGeorge Zhang if (ppn_set->initialized) { 53406164d2bSGeorge Zhang /* Do not call these functions on NULL inputs. */ 53506164d2bSGeorge Zhang kfree(ppn_set->produce_ppns); 53606164d2bSGeorge Zhang kfree(ppn_set->consume_ppns); 53706164d2bSGeorge Zhang } 53806164d2bSGeorge Zhang memset(ppn_set, 0, sizeof(*ppn_set)); 53906164d2bSGeorge Zhang } 54006164d2bSGeorge Zhang 54106164d2bSGeorge Zhang /* 54206164d2bSGeorge Zhang * Populates the list of PPNs in the hypercall structure with the PPNS 54306164d2bSGeorge Zhang * of the produce queue and the consume queue. 54406164d2bSGeorge Zhang */ 54506164d2bSGeorge Zhang static int qp_populate_ppn_set(u8 *call_buf, const struct PPNSet *ppn_set) 54606164d2bSGeorge Zhang { 54706164d2bSGeorge Zhang memcpy(call_buf, ppn_set->produce_ppns, 54806164d2bSGeorge Zhang ppn_set->num_produce_pages * sizeof(*ppn_set->produce_ppns)); 54906164d2bSGeorge Zhang memcpy(call_buf + 55006164d2bSGeorge Zhang ppn_set->num_produce_pages * sizeof(*ppn_set->produce_ppns), 55106164d2bSGeorge Zhang ppn_set->consume_ppns, 55206164d2bSGeorge Zhang ppn_set->num_consume_pages * sizeof(*ppn_set->consume_ppns)); 55306164d2bSGeorge Zhang 55406164d2bSGeorge Zhang return VMCI_SUCCESS; 55506164d2bSGeorge Zhang } 55606164d2bSGeorge Zhang 55706164d2bSGeorge Zhang static int qp_memcpy_to_queue(struct vmci_queue *queue, 55806164d2bSGeorge Zhang u64 queue_offset, 55906164d2bSGeorge Zhang const void *src, size_t src_offset, size_t size) 56006164d2bSGeorge Zhang { 56106164d2bSGeorge Zhang return __qp_memcpy_to_queue(queue, queue_offset, 56206164d2bSGeorge Zhang (u8 *)src + src_offset, size, false); 56306164d2bSGeorge Zhang } 56406164d2bSGeorge Zhang 56506164d2bSGeorge Zhang static int qp_memcpy_from_queue(void *dest, 56606164d2bSGeorge Zhang size_t dest_offset, 56706164d2bSGeorge Zhang const struct vmci_queue *queue, 56806164d2bSGeorge Zhang u64 queue_offset, size_t size) 56906164d2bSGeorge Zhang { 57006164d2bSGeorge Zhang return __qp_memcpy_from_queue((u8 *)dest + dest_offset, 57106164d2bSGeorge Zhang queue, queue_offset, size, false); 57206164d2bSGeorge Zhang } 57306164d2bSGeorge Zhang 57406164d2bSGeorge Zhang /* 57506164d2bSGeorge Zhang * Copies from a given iovec from a VMCI Queue. 57606164d2bSGeorge Zhang */ 57706164d2bSGeorge Zhang static int qp_memcpy_to_queue_iov(struct vmci_queue *queue, 57806164d2bSGeorge Zhang u64 queue_offset, 57906164d2bSGeorge Zhang const void *src, 58006164d2bSGeorge Zhang size_t src_offset, size_t size) 58106164d2bSGeorge Zhang { 58206164d2bSGeorge Zhang 58306164d2bSGeorge Zhang /* 58406164d2bSGeorge Zhang * We ignore src_offset because src is really a struct iovec * and will 58506164d2bSGeorge Zhang * maintain offset internally. 58606164d2bSGeorge Zhang */ 58706164d2bSGeorge Zhang return __qp_memcpy_to_queue(queue, queue_offset, src, size, true); 58806164d2bSGeorge Zhang } 58906164d2bSGeorge Zhang 59006164d2bSGeorge Zhang /* 59106164d2bSGeorge Zhang * Copies to a given iovec from a VMCI Queue. 59206164d2bSGeorge Zhang */ 59306164d2bSGeorge Zhang static int qp_memcpy_from_queue_iov(void *dest, 59406164d2bSGeorge Zhang size_t dest_offset, 59506164d2bSGeorge Zhang const struct vmci_queue *queue, 59606164d2bSGeorge Zhang u64 queue_offset, size_t size) 59706164d2bSGeorge Zhang { 59806164d2bSGeorge Zhang /* 59906164d2bSGeorge Zhang * We ignore dest_offset because dest is really a struct iovec * and 60006164d2bSGeorge Zhang * will maintain offset internally. 60106164d2bSGeorge Zhang */ 60206164d2bSGeorge Zhang return __qp_memcpy_from_queue(dest, queue, queue_offset, size, true); 60306164d2bSGeorge Zhang } 60406164d2bSGeorge Zhang 60506164d2bSGeorge Zhang /* 60606164d2bSGeorge Zhang * Allocates kernel VA space of specified size plus space for the queue 60706164d2bSGeorge Zhang * and kernel interface. This is different from the guest queue allocator, 60806164d2bSGeorge Zhang * because we do not allocate our own queue header/data pages here but 60906164d2bSGeorge Zhang * share those of the guest. 61006164d2bSGeorge Zhang */ 61106164d2bSGeorge Zhang static struct vmci_queue *qp_host_alloc_queue(u64 size) 61206164d2bSGeorge Zhang { 61306164d2bSGeorge Zhang struct vmci_queue *queue; 61406164d2bSGeorge Zhang const size_t num_pages = dm_div_up(size, PAGE_SIZE) + 1; 61506164d2bSGeorge Zhang const size_t queue_size = sizeof(*queue) + sizeof(*(queue->kernel_if)); 61606164d2bSGeorge Zhang const size_t queue_page_size = 61706164d2bSGeorge Zhang num_pages * sizeof(*queue->kernel_if->page); 61806164d2bSGeorge Zhang 61906164d2bSGeorge Zhang queue = kzalloc(queue_size + queue_page_size, GFP_KERNEL); 62006164d2bSGeorge Zhang if (queue) { 62106164d2bSGeorge Zhang queue->q_header = NULL; 62206164d2bSGeorge Zhang queue->saved_header = NULL; 62306164d2bSGeorge Zhang queue->kernel_if = 62406164d2bSGeorge Zhang (struct vmci_queue_kern_if *)((u8 *)queue + 62506164d2bSGeorge Zhang sizeof(*queue)); 62606164d2bSGeorge Zhang queue->kernel_if->host = true; 62706164d2bSGeorge Zhang queue->kernel_if->mutex = NULL; 62806164d2bSGeorge Zhang queue->kernel_if->num_pages = num_pages; 62906164d2bSGeorge Zhang queue->kernel_if->header_page = 63006164d2bSGeorge Zhang (struct page **)((u8 *)queue + queue_size); 63106164d2bSGeorge Zhang queue->kernel_if->page = &queue->kernel_if->header_page[1]; 63206164d2bSGeorge Zhang queue->kernel_if->va = NULL; 63306164d2bSGeorge Zhang queue->kernel_if->mapped = false; 63406164d2bSGeorge Zhang } 63506164d2bSGeorge Zhang 63606164d2bSGeorge Zhang return queue; 63706164d2bSGeorge Zhang } 63806164d2bSGeorge Zhang 63906164d2bSGeorge Zhang /* 64006164d2bSGeorge Zhang * Frees kernel memory for a given queue (header plus translation 64106164d2bSGeorge Zhang * structure). 64206164d2bSGeorge Zhang */ 64306164d2bSGeorge Zhang static void qp_host_free_queue(struct vmci_queue *queue, u64 queue_size) 64406164d2bSGeorge Zhang { 64506164d2bSGeorge Zhang kfree(queue); 64606164d2bSGeorge Zhang } 64706164d2bSGeorge Zhang 64806164d2bSGeorge Zhang /* 64906164d2bSGeorge Zhang * Initialize the mutex for the pair of queues. This mutex is used to 65006164d2bSGeorge Zhang * protect the q_header and the buffer from changing out from under any 65106164d2bSGeorge Zhang * users of either queue. Of course, it's only any good if the mutexes 65206164d2bSGeorge Zhang * are actually acquired. Queue structure must lie on non-paged memory 65306164d2bSGeorge Zhang * or we cannot guarantee access to the mutex. 65406164d2bSGeorge Zhang */ 65506164d2bSGeorge Zhang static void qp_init_queue_mutex(struct vmci_queue *produce_q, 65606164d2bSGeorge Zhang struct vmci_queue *consume_q) 65706164d2bSGeorge Zhang { 65806164d2bSGeorge Zhang /* 65906164d2bSGeorge Zhang * Only the host queue has shared state - the guest queues do not 66006164d2bSGeorge Zhang * need to synchronize access using a queue mutex. 66106164d2bSGeorge Zhang */ 66206164d2bSGeorge Zhang 66306164d2bSGeorge Zhang if (produce_q->kernel_if->host) { 66406164d2bSGeorge Zhang produce_q->kernel_if->mutex = &produce_q->kernel_if->__mutex; 66506164d2bSGeorge Zhang consume_q->kernel_if->mutex = &produce_q->kernel_if->__mutex; 66606164d2bSGeorge Zhang mutex_init(produce_q->kernel_if->mutex); 66706164d2bSGeorge Zhang } 66806164d2bSGeorge Zhang } 66906164d2bSGeorge Zhang 67006164d2bSGeorge Zhang /* 67106164d2bSGeorge Zhang * Cleans up the mutex for the pair of queues. 67206164d2bSGeorge Zhang */ 67306164d2bSGeorge Zhang static void qp_cleanup_queue_mutex(struct vmci_queue *produce_q, 67406164d2bSGeorge Zhang struct vmci_queue *consume_q) 67506164d2bSGeorge Zhang { 67606164d2bSGeorge Zhang if (produce_q->kernel_if->host) { 67706164d2bSGeorge Zhang produce_q->kernel_if->mutex = NULL; 67806164d2bSGeorge Zhang consume_q->kernel_if->mutex = NULL; 67906164d2bSGeorge Zhang } 68006164d2bSGeorge Zhang } 68106164d2bSGeorge Zhang 68206164d2bSGeorge Zhang /* 68306164d2bSGeorge Zhang * Acquire the mutex for the queue. Note that the produce_q and 68406164d2bSGeorge Zhang * the consume_q share a mutex. So, only one of the two need to 68506164d2bSGeorge Zhang * be passed in to this routine. Either will work just fine. 68606164d2bSGeorge Zhang */ 68706164d2bSGeorge Zhang static void qp_acquire_queue_mutex(struct vmci_queue *queue) 68806164d2bSGeorge Zhang { 68906164d2bSGeorge Zhang if (queue->kernel_if->host) 69006164d2bSGeorge Zhang mutex_lock(queue->kernel_if->mutex); 69106164d2bSGeorge Zhang } 69206164d2bSGeorge Zhang 69306164d2bSGeorge Zhang /* 69406164d2bSGeorge Zhang * Release the mutex for the queue. Note that the produce_q and 69506164d2bSGeorge Zhang * the consume_q share a mutex. So, only one of the two need to 69606164d2bSGeorge Zhang * be passed in to this routine. Either will work just fine. 69706164d2bSGeorge Zhang */ 69806164d2bSGeorge Zhang static void qp_release_queue_mutex(struct vmci_queue *queue) 69906164d2bSGeorge Zhang { 70006164d2bSGeorge Zhang if (queue->kernel_if->host) 70106164d2bSGeorge Zhang mutex_unlock(queue->kernel_if->mutex); 70206164d2bSGeorge Zhang } 70306164d2bSGeorge Zhang 70406164d2bSGeorge Zhang /* 70506164d2bSGeorge Zhang * Helper function to release pages in the PageStoreAttachInfo 70606164d2bSGeorge Zhang * previously obtained using get_user_pages. 70706164d2bSGeorge Zhang */ 70806164d2bSGeorge Zhang static void qp_release_pages(struct page **pages, 70906164d2bSGeorge Zhang u64 num_pages, bool dirty) 71006164d2bSGeorge Zhang { 71106164d2bSGeorge Zhang int i; 71206164d2bSGeorge Zhang 71306164d2bSGeorge Zhang for (i = 0; i < num_pages; i++) { 71406164d2bSGeorge Zhang if (dirty) 71506164d2bSGeorge Zhang set_page_dirty(pages[i]); 71606164d2bSGeorge Zhang 71706164d2bSGeorge Zhang page_cache_release(pages[i]); 71806164d2bSGeorge Zhang pages[i] = NULL; 71906164d2bSGeorge Zhang } 72006164d2bSGeorge Zhang } 72106164d2bSGeorge Zhang 72206164d2bSGeorge Zhang /* 72306164d2bSGeorge Zhang * Lock the user pages referenced by the {produce,consume}Buffer 72406164d2bSGeorge Zhang * struct into memory and populate the {produce,consume}Pages 72506164d2bSGeorge Zhang * arrays in the attach structure with them. 72606164d2bSGeorge Zhang */ 72706164d2bSGeorge Zhang static int qp_host_get_user_memory(u64 produce_uva, 72806164d2bSGeorge Zhang u64 consume_uva, 72906164d2bSGeorge Zhang struct vmci_queue *produce_q, 73006164d2bSGeorge Zhang struct vmci_queue *consume_q) 73106164d2bSGeorge Zhang { 73206164d2bSGeorge Zhang int retval; 73306164d2bSGeorge Zhang int err = VMCI_SUCCESS; 73406164d2bSGeorge Zhang 73506164d2bSGeorge Zhang down_write(¤t->mm->mmap_sem); 73606164d2bSGeorge Zhang retval = get_user_pages(current, 73706164d2bSGeorge Zhang current->mm, 73806164d2bSGeorge Zhang (uintptr_t) produce_uva, 73906164d2bSGeorge Zhang produce_q->kernel_if->num_pages, 74006164d2bSGeorge Zhang 1, 0, produce_q->kernel_if->header_page, NULL); 74106164d2bSGeorge Zhang if (retval < produce_q->kernel_if->num_pages) { 74206164d2bSGeorge Zhang pr_warn("get_user_pages(produce) failed (retval=%d)", retval); 74306164d2bSGeorge Zhang qp_release_pages(produce_q->kernel_if->header_page, retval, 74406164d2bSGeorge Zhang false); 74506164d2bSGeorge Zhang err = VMCI_ERROR_NO_MEM; 74606164d2bSGeorge Zhang goto out; 74706164d2bSGeorge Zhang } 74806164d2bSGeorge Zhang 74906164d2bSGeorge Zhang retval = get_user_pages(current, 75006164d2bSGeorge Zhang current->mm, 75106164d2bSGeorge Zhang (uintptr_t) consume_uva, 75206164d2bSGeorge Zhang consume_q->kernel_if->num_pages, 75306164d2bSGeorge Zhang 1, 0, consume_q->kernel_if->header_page, NULL); 75406164d2bSGeorge Zhang if (retval < consume_q->kernel_if->num_pages) { 75506164d2bSGeorge Zhang pr_warn("get_user_pages(consume) failed (retval=%d)", retval); 75606164d2bSGeorge Zhang qp_release_pages(consume_q->kernel_if->header_page, retval, 75706164d2bSGeorge Zhang false); 75806164d2bSGeorge Zhang qp_release_pages(produce_q->kernel_if->header_page, 75906164d2bSGeorge Zhang produce_q->kernel_if->num_pages, false); 76006164d2bSGeorge Zhang err = VMCI_ERROR_NO_MEM; 76106164d2bSGeorge Zhang } 76206164d2bSGeorge Zhang 76306164d2bSGeorge Zhang out: 76406164d2bSGeorge Zhang up_write(¤t->mm->mmap_sem); 76506164d2bSGeorge Zhang 76606164d2bSGeorge Zhang return err; 76706164d2bSGeorge Zhang } 76806164d2bSGeorge Zhang 76906164d2bSGeorge Zhang /* 77006164d2bSGeorge Zhang * Registers the specification of the user pages used for backing a queue 77106164d2bSGeorge Zhang * pair. Enough information to map in pages is stored in the OS specific 77206164d2bSGeorge Zhang * part of the struct vmci_queue structure. 77306164d2bSGeorge Zhang */ 77406164d2bSGeorge Zhang static int qp_host_register_user_memory(struct vmci_qp_page_store *page_store, 77506164d2bSGeorge Zhang struct vmci_queue *produce_q, 77606164d2bSGeorge Zhang struct vmci_queue *consume_q) 77706164d2bSGeorge Zhang { 77806164d2bSGeorge Zhang u64 produce_uva; 77906164d2bSGeorge Zhang u64 consume_uva; 78006164d2bSGeorge Zhang 78106164d2bSGeorge Zhang /* 78206164d2bSGeorge Zhang * The new style and the old style mapping only differs in 78306164d2bSGeorge Zhang * that we either get a single or two UVAs, so we split the 78406164d2bSGeorge Zhang * single UVA range at the appropriate spot. 78506164d2bSGeorge Zhang */ 78606164d2bSGeorge Zhang produce_uva = page_store->pages; 78706164d2bSGeorge Zhang consume_uva = page_store->pages + 78806164d2bSGeorge Zhang produce_q->kernel_if->num_pages * PAGE_SIZE; 78906164d2bSGeorge Zhang return qp_host_get_user_memory(produce_uva, consume_uva, produce_q, 79006164d2bSGeorge Zhang consume_q); 79106164d2bSGeorge Zhang } 79206164d2bSGeorge Zhang 79306164d2bSGeorge Zhang /* 79406164d2bSGeorge Zhang * Releases and removes the references to user pages stored in the attach 79506164d2bSGeorge Zhang * struct. Pages are released from the page cache and may become 79606164d2bSGeorge Zhang * swappable again. 79706164d2bSGeorge Zhang */ 79806164d2bSGeorge Zhang static void qp_host_unregister_user_memory(struct vmci_queue *produce_q, 79906164d2bSGeorge Zhang struct vmci_queue *consume_q) 80006164d2bSGeorge Zhang { 80106164d2bSGeorge Zhang qp_release_pages(produce_q->kernel_if->header_page, 80206164d2bSGeorge Zhang produce_q->kernel_if->num_pages, true); 80306164d2bSGeorge Zhang memset(produce_q->kernel_if->header_page, 0, 80406164d2bSGeorge Zhang sizeof(*produce_q->kernel_if->header_page) * 80506164d2bSGeorge Zhang produce_q->kernel_if->num_pages); 80606164d2bSGeorge Zhang qp_release_pages(consume_q->kernel_if->header_page, 80706164d2bSGeorge Zhang consume_q->kernel_if->num_pages, true); 80806164d2bSGeorge Zhang memset(consume_q->kernel_if->header_page, 0, 80906164d2bSGeorge Zhang sizeof(*consume_q->kernel_if->header_page) * 81006164d2bSGeorge Zhang consume_q->kernel_if->num_pages); 81106164d2bSGeorge Zhang } 81206164d2bSGeorge Zhang 81306164d2bSGeorge Zhang /* 81406164d2bSGeorge Zhang * Once qp_host_register_user_memory has been performed on a 81506164d2bSGeorge Zhang * queue, the queue pair headers can be mapped into the 81606164d2bSGeorge Zhang * kernel. Once mapped, they must be unmapped with 81706164d2bSGeorge Zhang * qp_host_unmap_queues prior to calling 81806164d2bSGeorge Zhang * qp_host_unregister_user_memory. 81906164d2bSGeorge Zhang * Pages are pinned. 82006164d2bSGeorge Zhang */ 82106164d2bSGeorge Zhang static int qp_host_map_queues(struct vmci_queue *produce_q, 82206164d2bSGeorge Zhang struct vmci_queue *consume_q) 82306164d2bSGeorge Zhang { 82406164d2bSGeorge Zhang int result; 82506164d2bSGeorge Zhang 82606164d2bSGeorge Zhang if (!produce_q->q_header || !consume_q->q_header) { 82706164d2bSGeorge Zhang struct page *headers[2]; 82806164d2bSGeorge Zhang 82906164d2bSGeorge Zhang if (produce_q->q_header != consume_q->q_header) 83006164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_MISMATCH; 83106164d2bSGeorge Zhang 83206164d2bSGeorge Zhang if (produce_q->kernel_if->header_page == NULL || 83306164d2bSGeorge Zhang *produce_q->kernel_if->header_page == NULL) 83406164d2bSGeorge Zhang return VMCI_ERROR_UNAVAILABLE; 83506164d2bSGeorge Zhang 83606164d2bSGeorge Zhang headers[0] = *produce_q->kernel_if->header_page; 83706164d2bSGeorge Zhang headers[1] = *consume_q->kernel_if->header_page; 83806164d2bSGeorge Zhang 83906164d2bSGeorge Zhang produce_q->q_header = vmap(headers, 2, VM_MAP, PAGE_KERNEL); 84006164d2bSGeorge Zhang if (produce_q->q_header != NULL) { 84106164d2bSGeorge Zhang consume_q->q_header = 84206164d2bSGeorge Zhang (struct vmci_queue_header *)((u8 *) 84306164d2bSGeorge Zhang produce_q->q_header + 84406164d2bSGeorge Zhang PAGE_SIZE); 84506164d2bSGeorge Zhang result = VMCI_SUCCESS; 84606164d2bSGeorge Zhang } else { 84706164d2bSGeorge Zhang pr_warn("vmap failed\n"); 84806164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 84906164d2bSGeorge Zhang } 85006164d2bSGeorge Zhang } else { 85106164d2bSGeorge Zhang result = VMCI_SUCCESS; 85206164d2bSGeorge Zhang } 85306164d2bSGeorge Zhang 85406164d2bSGeorge Zhang return result; 85506164d2bSGeorge Zhang } 85606164d2bSGeorge Zhang 85706164d2bSGeorge Zhang /* 85806164d2bSGeorge Zhang * Unmaps previously mapped queue pair headers from the kernel. 85906164d2bSGeorge Zhang * Pages are unpinned. 86006164d2bSGeorge Zhang */ 86106164d2bSGeorge Zhang static int qp_host_unmap_queues(u32 gid, 86206164d2bSGeorge Zhang struct vmci_queue *produce_q, 86306164d2bSGeorge Zhang struct vmci_queue *consume_q) 86406164d2bSGeorge Zhang { 86506164d2bSGeorge Zhang if (produce_q->q_header) { 86606164d2bSGeorge Zhang if (produce_q->q_header < consume_q->q_header) 86706164d2bSGeorge Zhang vunmap(produce_q->q_header); 86806164d2bSGeorge Zhang else 86906164d2bSGeorge Zhang vunmap(consume_q->q_header); 87006164d2bSGeorge Zhang 87106164d2bSGeorge Zhang produce_q->q_header = NULL; 87206164d2bSGeorge Zhang consume_q->q_header = NULL; 87306164d2bSGeorge Zhang } 87406164d2bSGeorge Zhang 87506164d2bSGeorge Zhang return VMCI_SUCCESS; 87606164d2bSGeorge Zhang } 87706164d2bSGeorge Zhang 87806164d2bSGeorge Zhang /* 87906164d2bSGeorge Zhang * Finds the entry in the list corresponding to a given handle. Assumes 88006164d2bSGeorge Zhang * that the list is locked. 88106164d2bSGeorge Zhang */ 88206164d2bSGeorge Zhang static struct qp_entry *qp_list_find(struct qp_list *qp_list, 88306164d2bSGeorge Zhang struct vmci_handle handle) 88406164d2bSGeorge Zhang { 88506164d2bSGeorge Zhang struct qp_entry *entry; 88606164d2bSGeorge Zhang 88706164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle)) 88806164d2bSGeorge Zhang return NULL; 88906164d2bSGeorge Zhang 89006164d2bSGeorge Zhang list_for_each_entry(entry, &qp_list->head, list_item) { 89106164d2bSGeorge Zhang if (vmci_handle_is_equal(entry->handle, handle)) 89206164d2bSGeorge Zhang return entry; 89306164d2bSGeorge Zhang } 89406164d2bSGeorge Zhang 89506164d2bSGeorge Zhang return NULL; 89606164d2bSGeorge Zhang } 89706164d2bSGeorge Zhang 89806164d2bSGeorge Zhang /* 89906164d2bSGeorge Zhang * Finds the entry in the list corresponding to a given handle. 90006164d2bSGeorge Zhang */ 90106164d2bSGeorge Zhang static struct qp_guest_endpoint * 90206164d2bSGeorge Zhang qp_guest_handle_to_entry(struct vmci_handle handle) 90306164d2bSGeorge Zhang { 90406164d2bSGeorge Zhang struct qp_guest_endpoint *entry; 90506164d2bSGeorge Zhang struct qp_entry *qp = qp_list_find(&qp_guest_endpoints, handle); 90606164d2bSGeorge Zhang 90706164d2bSGeorge Zhang entry = qp ? container_of( 90806164d2bSGeorge Zhang qp, struct qp_guest_endpoint, qp) : NULL; 90906164d2bSGeorge Zhang return entry; 91006164d2bSGeorge Zhang } 91106164d2bSGeorge Zhang 91206164d2bSGeorge Zhang /* 91306164d2bSGeorge Zhang * Finds the entry in the list corresponding to a given handle. 91406164d2bSGeorge Zhang */ 91506164d2bSGeorge Zhang static struct qp_broker_entry * 91606164d2bSGeorge Zhang qp_broker_handle_to_entry(struct vmci_handle handle) 91706164d2bSGeorge Zhang { 91806164d2bSGeorge Zhang struct qp_broker_entry *entry; 91906164d2bSGeorge Zhang struct qp_entry *qp = qp_list_find(&qp_broker_list, handle); 92006164d2bSGeorge Zhang 92106164d2bSGeorge Zhang entry = qp ? container_of( 92206164d2bSGeorge Zhang qp, struct qp_broker_entry, qp) : NULL; 92306164d2bSGeorge Zhang return entry; 92406164d2bSGeorge Zhang } 92506164d2bSGeorge Zhang 92606164d2bSGeorge Zhang /* 92706164d2bSGeorge Zhang * Dispatches a queue pair event message directly into the local event 92806164d2bSGeorge Zhang * queue. 92906164d2bSGeorge Zhang */ 93006164d2bSGeorge Zhang static int qp_notify_peer_local(bool attach, struct vmci_handle handle) 93106164d2bSGeorge Zhang { 93206164d2bSGeorge Zhang u32 context_id = vmci_get_context_id(); 93306164d2bSGeorge Zhang struct vmci_event_qp ev; 93406164d2bSGeorge Zhang 93506164d2bSGeorge Zhang ev.msg.hdr.dst = vmci_make_handle(context_id, VMCI_EVENT_HANDLER); 93606164d2bSGeorge Zhang ev.msg.hdr.src = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, 93706164d2bSGeorge Zhang VMCI_CONTEXT_RESOURCE_ID); 93806164d2bSGeorge Zhang ev.msg.hdr.payload_size = sizeof(ev) - sizeof(ev.msg.hdr); 93906164d2bSGeorge Zhang ev.msg.event_data.event = 94006164d2bSGeorge Zhang attach ? VMCI_EVENT_QP_PEER_ATTACH : VMCI_EVENT_QP_PEER_DETACH; 94106164d2bSGeorge Zhang ev.payload.peer_id = context_id; 94206164d2bSGeorge Zhang ev.payload.handle = handle; 94306164d2bSGeorge Zhang 94406164d2bSGeorge Zhang return vmci_event_dispatch(&ev.msg.hdr); 94506164d2bSGeorge Zhang } 94606164d2bSGeorge Zhang 94706164d2bSGeorge Zhang /* 94806164d2bSGeorge Zhang * Allocates and initializes a qp_guest_endpoint structure. 94906164d2bSGeorge Zhang * Allocates a queue_pair rid (and handle) iff the given entry has 95006164d2bSGeorge Zhang * an invalid handle. 0 through VMCI_RESERVED_RESOURCE_ID_MAX 95106164d2bSGeorge Zhang * are reserved handles. Assumes that the QP list mutex is held 95206164d2bSGeorge Zhang * by the caller. 95306164d2bSGeorge Zhang */ 95406164d2bSGeorge Zhang static struct qp_guest_endpoint * 95506164d2bSGeorge Zhang qp_guest_endpoint_create(struct vmci_handle handle, 95606164d2bSGeorge Zhang u32 peer, 95706164d2bSGeorge Zhang u32 flags, 95806164d2bSGeorge Zhang u64 produce_size, 95906164d2bSGeorge Zhang u64 consume_size, 96006164d2bSGeorge Zhang void *produce_q, 96106164d2bSGeorge Zhang void *consume_q) 96206164d2bSGeorge Zhang { 96306164d2bSGeorge Zhang int result; 96406164d2bSGeorge Zhang struct qp_guest_endpoint *entry; 96506164d2bSGeorge Zhang /* One page each for the queue headers. */ 96606164d2bSGeorge Zhang const u64 num_ppns = dm_div_up(produce_size, PAGE_SIZE) + 96706164d2bSGeorge Zhang dm_div_up(consume_size, PAGE_SIZE) + 2; 96806164d2bSGeorge Zhang 96906164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle)) { 97006164d2bSGeorge Zhang u32 context_id = vmci_get_context_id(); 97106164d2bSGeorge Zhang 97206164d2bSGeorge Zhang handle = vmci_make_handle(context_id, VMCI_INVALID_ID); 97306164d2bSGeorge Zhang } 97406164d2bSGeorge Zhang 97506164d2bSGeorge Zhang entry = kzalloc(sizeof(*entry), GFP_KERNEL); 97606164d2bSGeorge Zhang if (entry) { 97706164d2bSGeorge Zhang entry->qp.peer = peer; 97806164d2bSGeorge Zhang entry->qp.flags = flags; 97906164d2bSGeorge Zhang entry->qp.produce_size = produce_size; 98006164d2bSGeorge Zhang entry->qp.consume_size = consume_size; 98106164d2bSGeorge Zhang entry->qp.ref_count = 0; 98206164d2bSGeorge Zhang entry->num_ppns = num_ppns; 98306164d2bSGeorge Zhang entry->produce_q = produce_q; 98406164d2bSGeorge Zhang entry->consume_q = consume_q; 98506164d2bSGeorge Zhang INIT_LIST_HEAD(&entry->qp.list_item); 98606164d2bSGeorge Zhang 98706164d2bSGeorge Zhang /* Add resource obj */ 98806164d2bSGeorge Zhang result = vmci_resource_add(&entry->resource, 98906164d2bSGeorge Zhang VMCI_RESOURCE_TYPE_QPAIR_GUEST, 99006164d2bSGeorge Zhang handle); 99106164d2bSGeorge Zhang entry->qp.handle = vmci_resource_handle(&entry->resource); 99206164d2bSGeorge Zhang if ((result != VMCI_SUCCESS) || 99306164d2bSGeorge Zhang qp_list_find(&qp_guest_endpoints, entry->qp.handle)) { 99406164d2bSGeorge Zhang pr_warn("Failed to add new resource (handle=0x%x:0x%x), error: %d", 99506164d2bSGeorge Zhang handle.context, handle.resource, result); 99606164d2bSGeorge Zhang kfree(entry); 99706164d2bSGeorge Zhang entry = NULL; 99806164d2bSGeorge Zhang } 99906164d2bSGeorge Zhang } 100006164d2bSGeorge Zhang return entry; 100106164d2bSGeorge Zhang } 100206164d2bSGeorge Zhang 100306164d2bSGeorge Zhang /* 100406164d2bSGeorge Zhang * Frees a qp_guest_endpoint structure. 100506164d2bSGeorge Zhang */ 100606164d2bSGeorge Zhang static void qp_guest_endpoint_destroy(struct qp_guest_endpoint *entry) 100706164d2bSGeorge Zhang { 100806164d2bSGeorge Zhang qp_free_ppn_set(&entry->ppn_set); 100906164d2bSGeorge Zhang qp_cleanup_queue_mutex(entry->produce_q, entry->consume_q); 101006164d2bSGeorge Zhang qp_free_queue(entry->produce_q, entry->qp.produce_size); 101106164d2bSGeorge Zhang qp_free_queue(entry->consume_q, entry->qp.consume_size); 101206164d2bSGeorge Zhang /* Unlink from resource hash table and free callback */ 101306164d2bSGeorge Zhang vmci_resource_remove(&entry->resource); 101406164d2bSGeorge Zhang 101506164d2bSGeorge Zhang kfree(entry); 101606164d2bSGeorge Zhang } 101706164d2bSGeorge Zhang 101806164d2bSGeorge Zhang /* 101906164d2bSGeorge Zhang * Helper to make a queue_pairAlloc hypercall when the driver is 102006164d2bSGeorge Zhang * supporting a guest device. 102106164d2bSGeorge Zhang */ 102206164d2bSGeorge Zhang static int qp_alloc_hypercall(const struct qp_guest_endpoint *entry) 102306164d2bSGeorge Zhang { 102406164d2bSGeorge Zhang struct vmci_qp_alloc_msg *alloc_msg; 102506164d2bSGeorge Zhang size_t msg_size; 102606164d2bSGeorge Zhang int result; 102706164d2bSGeorge Zhang 102806164d2bSGeorge Zhang if (!entry || entry->num_ppns <= 2) 102906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 103006164d2bSGeorge Zhang 103106164d2bSGeorge Zhang msg_size = sizeof(*alloc_msg) + 103206164d2bSGeorge Zhang (size_t) entry->num_ppns * sizeof(u32); 103306164d2bSGeorge Zhang alloc_msg = kmalloc(msg_size, GFP_KERNEL); 103406164d2bSGeorge Zhang if (!alloc_msg) 103506164d2bSGeorge Zhang return VMCI_ERROR_NO_MEM; 103606164d2bSGeorge Zhang 103706164d2bSGeorge Zhang alloc_msg->hdr.dst = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, 103806164d2bSGeorge Zhang VMCI_QUEUEPAIR_ALLOC); 103906164d2bSGeorge Zhang alloc_msg->hdr.src = VMCI_ANON_SRC_HANDLE; 104006164d2bSGeorge Zhang alloc_msg->hdr.payload_size = msg_size - VMCI_DG_HEADERSIZE; 104106164d2bSGeorge Zhang alloc_msg->handle = entry->qp.handle; 104206164d2bSGeorge Zhang alloc_msg->peer = entry->qp.peer; 104306164d2bSGeorge Zhang alloc_msg->flags = entry->qp.flags; 104406164d2bSGeorge Zhang alloc_msg->produce_size = entry->qp.produce_size; 104506164d2bSGeorge Zhang alloc_msg->consume_size = entry->qp.consume_size; 104606164d2bSGeorge Zhang alloc_msg->num_ppns = entry->num_ppns; 104706164d2bSGeorge Zhang 104806164d2bSGeorge Zhang result = qp_populate_ppn_set((u8 *)alloc_msg + sizeof(*alloc_msg), 104906164d2bSGeorge Zhang &entry->ppn_set); 105006164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 105106164d2bSGeorge Zhang result = vmci_send_datagram(&alloc_msg->hdr); 105206164d2bSGeorge Zhang 105306164d2bSGeorge Zhang kfree(alloc_msg); 105406164d2bSGeorge Zhang 105506164d2bSGeorge Zhang return result; 105606164d2bSGeorge Zhang } 105706164d2bSGeorge Zhang 105806164d2bSGeorge Zhang /* 105906164d2bSGeorge Zhang * Helper to make a queue_pairDetach hypercall when the driver is 106006164d2bSGeorge Zhang * supporting a guest device. 106106164d2bSGeorge Zhang */ 106206164d2bSGeorge Zhang static int qp_detatch_hypercall(struct vmci_handle handle) 106306164d2bSGeorge Zhang { 106406164d2bSGeorge Zhang struct vmci_qp_detach_msg detach_msg; 106506164d2bSGeorge Zhang 106606164d2bSGeorge Zhang detach_msg.hdr.dst = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, 106706164d2bSGeorge Zhang VMCI_QUEUEPAIR_DETACH); 106806164d2bSGeorge Zhang detach_msg.hdr.src = VMCI_ANON_SRC_HANDLE; 106906164d2bSGeorge Zhang detach_msg.hdr.payload_size = sizeof(handle); 107006164d2bSGeorge Zhang detach_msg.handle = handle; 107106164d2bSGeorge Zhang 107206164d2bSGeorge Zhang return vmci_send_datagram(&detach_msg.hdr); 107306164d2bSGeorge Zhang } 107406164d2bSGeorge Zhang 107506164d2bSGeorge Zhang /* 107606164d2bSGeorge Zhang * Adds the given entry to the list. Assumes that the list is locked. 107706164d2bSGeorge Zhang */ 107806164d2bSGeorge Zhang static void qp_list_add_entry(struct qp_list *qp_list, struct qp_entry *entry) 107906164d2bSGeorge Zhang { 108006164d2bSGeorge Zhang if (entry) 108106164d2bSGeorge Zhang list_add(&entry->list_item, &qp_list->head); 108206164d2bSGeorge Zhang } 108306164d2bSGeorge Zhang 108406164d2bSGeorge Zhang /* 108506164d2bSGeorge Zhang * Removes the given entry from the list. Assumes that the list is locked. 108606164d2bSGeorge Zhang */ 108706164d2bSGeorge Zhang static void qp_list_remove_entry(struct qp_list *qp_list, 108806164d2bSGeorge Zhang struct qp_entry *entry) 108906164d2bSGeorge Zhang { 109006164d2bSGeorge Zhang if (entry) 109106164d2bSGeorge Zhang list_del(&entry->list_item); 109206164d2bSGeorge Zhang } 109306164d2bSGeorge Zhang 109406164d2bSGeorge Zhang /* 109506164d2bSGeorge Zhang * Helper for VMCI queue_pair detach interface. Frees the physical 109606164d2bSGeorge Zhang * pages for the queue pair. 109706164d2bSGeorge Zhang */ 109806164d2bSGeorge Zhang static int qp_detatch_guest_work(struct vmci_handle handle) 109906164d2bSGeorge Zhang { 110006164d2bSGeorge Zhang int result; 110106164d2bSGeorge Zhang struct qp_guest_endpoint *entry; 110206164d2bSGeorge Zhang u32 ref_count = ~0; /* To avoid compiler warning below */ 110306164d2bSGeorge Zhang 110406164d2bSGeorge Zhang mutex_lock(&qp_guest_endpoints.mutex); 110506164d2bSGeorge Zhang 110606164d2bSGeorge Zhang entry = qp_guest_handle_to_entry(handle); 110706164d2bSGeorge Zhang if (!entry) { 110806164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 110906164d2bSGeorge Zhang return VMCI_ERROR_NOT_FOUND; 111006164d2bSGeorge Zhang } 111106164d2bSGeorge Zhang 111206164d2bSGeorge Zhang if (entry->qp.flags & VMCI_QPFLAG_LOCAL) { 111306164d2bSGeorge Zhang result = VMCI_SUCCESS; 111406164d2bSGeorge Zhang 111506164d2bSGeorge Zhang if (entry->qp.ref_count > 1) { 111606164d2bSGeorge Zhang result = qp_notify_peer_local(false, handle); 111706164d2bSGeorge Zhang /* 111806164d2bSGeorge Zhang * We can fail to notify a local queuepair 111906164d2bSGeorge Zhang * because we can't allocate. We still want 112006164d2bSGeorge Zhang * to release the entry if that happens, so 112106164d2bSGeorge Zhang * don't bail out yet. 112206164d2bSGeorge Zhang */ 112306164d2bSGeorge Zhang } 112406164d2bSGeorge Zhang } else { 112506164d2bSGeorge Zhang result = qp_detatch_hypercall(handle); 112606164d2bSGeorge Zhang if (result < VMCI_SUCCESS) { 112706164d2bSGeorge Zhang /* 112806164d2bSGeorge Zhang * We failed to notify a non-local queuepair. 112906164d2bSGeorge Zhang * That other queuepair might still be 113006164d2bSGeorge Zhang * accessing the shared memory, so don't 113106164d2bSGeorge Zhang * release the entry yet. It will get cleaned 113206164d2bSGeorge Zhang * up by VMCIqueue_pair_Exit() if necessary 113306164d2bSGeorge Zhang * (assuming we are going away, otherwise why 113406164d2bSGeorge Zhang * did this fail?). 113506164d2bSGeorge Zhang */ 113606164d2bSGeorge Zhang 113706164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 113806164d2bSGeorge Zhang return result; 113906164d2bSGeorge Zhang } 114006164d2bSGeorge Zhang } 114106164d2bSGeorge Zhang 114206164d2bSGeorge Zhang /* 114306164d2bSGeorge Zhang * If we get here then we either failed to notify a local queuepair, or 114406164d2bSGeorge Zhang * we succeeded in all cases. Release the entry if required. 114506164d2bSGeorge Zhang */ 114606164d2bSGeorge Zhang 114706164d2bSGeorge Zhang entry->qp.ref_count--; 114806164d2bSGeorge Zhang if (entry->qp.ref_count == 0) 114906164d2bSGeorge Zhang qp_list_remove_entry(&qp_guest_endpoints, &entry->qp); 115006164d2bSGeorge Zhang 115106164d2bSGeorge Zhang /* If we didn't remove the entry, this could change once we unlock. */ 115206164d2bSGeorge Zhang if (entry) 115306164d2bSGeorge Zhang ref_count = entry->qp.ref_count; 115406164d2bSGeorge Zhang 115506164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 115606164d2bSGeorge Zhang 115706164d2bSGeorge Zhang if (ref_count == 0) 115806164d2bSGeorge Zhang qp_guest_endpoint_destroy(entry); 115906164d2bSGeorge Zhang 116006164d2bSGeorge Zhang return result; 116106164d2bSGeorge Zhang } 116206164d2bSGeorge Zhang 116306164d2bSGeorge Zhang /* 116406164d2bSGeorge Zhang * This functions handles the actual allocation of a VMCI queue 116506164d2bSGeorge Zhang * pair guest endpoint. Allocates physical pages for the queue 116606164d2bSGeorge Zhang * pair. It makes OS dependent calls through generic wrappers. 116706164d2bSGeorge Zhang */ 116806164d2bSGeorge Zhang static int qp_alloc_guest_work(struct vmci_handle *handle, 116906164d2bSGeorge Zhang struct vmci_queue **produce_q, 117006164d2bSGeorge Zhang u64 produce_size, 117106164d2bSGeorge Zhang struct vmci_queue **consume_q, 117206164d2bSGeorge Zhang u64 consume_size, 117306164d2bSGeorge Zhang u32 peer, 117406164d2bSGeorge Zhang u32 flags, 117506164d2bSGeorge Zhang u32 priv_flags) 117606164d2bSGeorge Zhang { 117706164d2bSGeorge Zhang const u64 num_produce_pages = 117806164d2bSGeorge Zhang dm_div_up(produce_size, PAGE_SIZE) + 1; 117906164d2bSGeorge Zhang const u64 num_consume_pages = 118006164d2bSGeorge Zhang dm_div_up(consume_size, PAGE_SIZE) + 1; 118106164d2bSGeorge Zhang void *my_produce_q = NULL; 118206164d2bSGeorge Zhang void *my_consume_q = NULL; 118306164d2bSGeorge Zhang int result; 118406164d2bSGeorge Zhang struct qp_guest_endpoint *queue_pair_entry = NULL; 118506164d2bSGeorge Zhang 118606164d2bSGeorge Zhang if (priv_flags != VMCI_NO_PRIVILEGE_FLAGS) 118706164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 118806164d2bSGeorge Zhang 118906164d2bSGeorge Zhang mutex_lock(&qp_guest_endpoints.mutex); 119006164d2bSGeorge Zhang 119106164d2bSGeorge Zhang queue_pair_entry = qp_guest_handle_to_entry(*handle); 119206164d2bSGeorge Zhang if (queue_pair_entry) { 119306164d2bSGeorge Zhang if (queue_pair_entry->qp.flags & VMCI_QPFLAG_LOCAL) { 119406164d2bSGeorge Zhang /* Local attach case. */ 119506164d2bSGeorge Zhang if (queue_pair_entry->qp.ref_count > 1) { 119606164d2bSGeorge Zhang pr_devel("Error attempting to attach more than once\n"); 119706164d2bSGeorge Zhang result = VMCI_ERROR_UNAVAILABLE; 119806164d2bSGeorge Zhang goto error_keep_entry; 119906164d2bSGeorge Zhang } 120006164d2bSGeorge Zhang 120106164d2bSGeorge Zhang if (queue_pair_entry->qp.produce_size != consume_size || 120206164d2bSGeorge Zhang queue_pair_entry->qp.consume_size != 120306164d2bSGeorge Zhang produce_size || 120406164d2bSGeorge Zhang queue_pair_entry->qp.flags != 120506164d2bSGeorge Zhang (flags & ~VMCI_QPFLAG_ATTACH_ONLY)) { 120606164d2bSGeorge Zhang pr_devel("Error mismatched queue pair in local attach\n"); 120706164d2bSGeorge Zhang result = VMCI_ERROR_QUEUEPAIR_MISMATCH; 120806164d2bSGeorge Zhang goto error_keep_entry; 120906164d2bSGeorge Zhang } 121006164d2bSGeorge Zhang 121106164d2bSGeorge Zhang /* 121206164d2bSGeorge Zhang * Do a local attach. We swap the consume and 121306164d2bSGeorge Zhang * produce queues for the attacher and deliver 121406164d2bSGeorge Zhang * an attach event. 121506164d2bSGeorge Zhang */ 121606164d2bSGeorge Zhang result = qp_notify_peer_local(true, *handle); 121706164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 121806164d2bSGeorge Zhang goto error_keep_entry; 121906164d2bSGeorge Zhang 122006164d2bSGeorge Zhang my_produce_q = queue_pair_entry->consume_q; 122106164d2bSGeorge Zhang my_consume_q = queue_pair_entry->produce_q; 122206164d2bSGeorge Zhang goto out; 122306164d2bSGeorge Zhang } 122406164d2bSGeorge Zhang 122506164d2bSGeorge Zhang result = VMCI_ERROR_ALREADY_EXISTS; 122606164d2bSGeorge Zhang goto error_keep_entry; 122706164d2bSGeorge Zhang } 122806164d2bSGeorge Zhang 122906164d2bSGeorge Zhang my_produce_q = qp_alloc_queue(produce_size, flags); 123006164d2bSGeorge Zhang if (!my_produce_q) { 123106164d2bSGeorge Zhang pr_warn("Error allocating pages for produce queue\n"); 123206164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 123306164d2bSGeorge Zhang goto error; 123406164d2bSGeorge Zhang } 123506164d2bSGeorge Zhang 123606164d2bSGeorge Zhang my_consume_q = qp_alloc_queue(consume_size, flags); 123706164d2bSGeorge Zhang if (!my_consume_q) { 123806164d2bSGeorge Zhang pr_warn("Error allocating pages for consume queue\n"); 123906164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 124006164d2bSGeorge Zhang goto error; 124106164d2bSGeorge Zhang } 124206164d2bSGeorge Zhang 124306164d2bSGeorge Zhang queue_pair_entry = qp_guest_endpoint_create(*handle, peer, flags, 124406164d2bSGeorge Zhang produce_size, consume_size, 124506164d2bSGeorge Zhang my_produce_q, my_consume_q); 124606164d2bSGeorge Zhang if (!queue_pair_entry) { 124706164d2bSGeorge Zhang pr_warn("Error allocating memory in %s\n", __func__); 124806164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 124906164d2bSGeorge Zhang goto error; 125006164d2bSGeorge Zhang } 125106164d2bSGeorge Zhang 125206164d2bSGeorge Zhang result = qp_alloc_ppn_set(my_produce_q, num_produce_pages, my_consume_q, 125306164d2bSGeorge Zhang num_consume_pages, 125406164d2bSGeorge Zhang &queue_pair_entry->ppn_set); 125506164d2bSGeorge Zhang if (result < VMCI_SUCCESS) { 125606164d2bSGeorge Zhang pr_warn("qp_alloc_ppn_set failed\n"); 125706164d2bSGeorge Zhang goto error; 125806164d2bSGeorge Zhang } 125906164d2bSGeorge Zhang 126006164d2bSGeorge Zhang /* 126106164d2bSGeorge Zhang * It's only necessary to notify the host if this queue pair will be 126206164d2bSGeorge Zhang * attached to from another context. 126306164d2bSGeorge Zhang */ 126406164d2bSGeorge Zhang if (queue_pair_entry->qp.flags & VMCI_QPFLAG_LOCAL) { 126506164d2bSGeorge Zhang /* Local create case. */ 126606164d2bSGeorge Zhang u32 context_id = vmci_get_context_id(); 126706164d2bSGeorge Zhang 126806164d2bSGeorge Zhang /* 126906164d2bSGeorge Zhang * Enforce similar checks on local queue pairs as we 127006164d2bSGeorge Zhang * do for regular ones. The handle's context must 127106164d2bSGeorge Zhang * match the creator or attacher context id (here they 127206164d2bSGeorge Zhang * are both the current context id) and the 127306164d2bSGeorge Zhang * attach-only flag cannot exist during create. We 127406164d2bSGeorge Zhang * also ensure specified peer is this context or an 127506164d2bSGeorge Zhang * invalid one. 127606164d2bSGeorge Zhang */ 127706164d2bSGeorge Zhang if (queue_pair_entry->qp.handle.context != context_id || 127806164d2bSGeorge Zhang (queue_pair_entry->qp.peer != VMCI_INVALID_ID && 127906164d2bSGeorge Zhang queue_pair_entry->qp.peer != context_id)) { 128006164d2bSGeorge Zhang result = VMCI_ERROR_NO_ACCESS; 128106164d2bSGeorge Zhang goto error; 128206164d2bSGeorge Zhang } 128306164d2bSGeorge Zhang 128406164d2bSGeorge Zhang if (queue_pair_entry->qp.flags & VMCI_QPFLAG_ATTACH_ONLY) { 128506164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 128606164d2bSGeorge Zhang goto error; 128706164d2bSGeorge Zhang } 128806164d2bSGeorge Zhang } else { 128906164d2bSGeorge Zhang result = qp_alloc_hypercall(queue_pair_entry); 129006164d2bSGeorge Zhang if (result < VMCI_SUCCESS) { 129106164d2bSGeorge Zhang pr_warn("qp_alloc_hypercall result = %d\n", result); 129206164d2bSGeorge Zhang goto error; 129306164d2bSGeorge Zhang } 129406164d2bSGeorge Zhang } 129506164d2bSGeorge Zhang 129606164d2bSGeorge Zhang qp_init_queue_mutex((struct vmci_queue *)my_produce_q, 129706164d2bSGeorge Zhang (struct vmci_queue *)my_consume_q); 129806164d2bSGeorge Zhang 129906164d2bSGeorge Zhang qp_list_add_entry(&qp_guest_endpoints, &queue_pair_entry->qp); 130006164d2bSGeorge Zhang 130106164d2bSGeorge Zhang out: 130206164d2bSGeorge Zhang queue_pair_entry->qp.ref_count++; 130306164d2bSGeorge Zhang *handle = queue_pair_entry->qp.handle; 130406164d2bSGeorge Zhang *produce_q = (struct vmci_queue *)my_produce_q; 130506164d2bSGeorge Zhang *consume_q = (struct vmci_queue *)my_consume_q; 130606164d2bSGeorge Zhang 130706164d2bSGeorge Zhang /* 130806164d2bSGeorge Zhang * We should initialize the queue pair header pages on a local 130906164d2bSGeorge Zhang * queue pair create. For non-local queue pairs, the 131006164d2bSGeorge Zhang * hypervisor initializes the header pages in the create step. 131106164d2bSGeorge Zhang */ 131206164d2bSGeorge Zhang if ((queue_pair_entry->qp.flags & VMCI_QPFLAG_LOCAL) && 131306164d2bSGeorge Zhang queue_pair_entry->qp.ref_count == 1) { 131406164d2bSGeorge Zhang vmci_q_header_init((*produce_q)->q_header, *handle); 131506164d2bSGeorge Zhang vmci_q_header_init((*consume_q)->q_header, *handle); 131606164d2bSGeorge Zhang } 131706164d2bSGeorge Zhang 131806164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 131906164d2bSGeorge Zhang 132006164d2bSGeorge Zhang return VMCI_SUCCESS; 132106164d2bSGeorge Zhang 132206164d2bSGeorge Zhang error: 132306164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 132406164d2bSGeorge Zhang if (queue_pair_entry) { 132506164d2bSGeorge Zhang /* The queues will be freed inside the destroy routine. */ 132606164d2bSGeorge Zhang qp_guest_endpoint_destroy(queue_pair_entry); 132706164d2bSGeorge Zhang } else { 132806164d2bSGeorge Zhang qp_free_queue(my_produce_q, produce_size); 132906164d2bSGeorge Zhang qp_free_queue(my_consume_q, consume_size); 133006164d2bSGeorge Zhang } 133106164d2bSGeorge Zhang return result; 133206164d2bSGeorge Zhang 133306164d2bSGeorge Zhang error_keep_entry: 133406164d2bSGeorge Zhang /* This path should only be used when an existing entry was found. */ 133506164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 133606164d2bSGeorge Zhang return result; 133706164d2bSGeorge Zhang } 133806164d2bSGeorge Zhang 133906164d2bSGeorge Zhang /* 134006164d2bSGeorge Zhang * The first endpoint issuing a queue pair allocation will create the state 134106164d2bSGeorge Zhang * of the queue pair in the queue pair broker. 134206164d2bSGeorge Zhang * 134306164d2bSGeorge Zhang * If the creator is a guest, it will associate a VMX virtual address range 134406164d2bSGeorge Zhang * with the queue pair as specified by the page_store. For compatibility with 134506164d2bSGeorge Zhang * older VMX'en, that would use a separate step to set the VMX virtual 134606164d2bSGeorge Zhang * address range, the virtual address range can be registered later using 134706164d2bSGeorge Zhang * vmci_qp_broker_set_page_store. In that case, a page_store of NULL should be 134806164d2bSGeorge Zhang * used. 134906164d2bSGeorge Zhang * 135006164d2bSGeorge Zhang * If the creator is the host, a page_store of NULL should be used as well, 135106164d2bSGeorge Zhang * since the host is not able to supply a page store for the queue pair. 135206164d2bSGeorge Zhang * 135306164d2bSGeorge Zhang * For older VMX and host callers, the queue pair will be created in the 135406164d2bSGeorge Zhang * VMCIQPB_CREATED_NO_MEM state, and for current VMX callers, it will be 135506164d2bSGeorge Zhang * created in VMCOQPB_CREATED_MEM state. 135606164d2bSGeorge Zhang */ 135706164d2bSGeorge Zhang static int qp_broker_create(struct vmci_handle handle, 135806164d2bSGeorge Zhang u32 peer, 135906164d2bSGeorge Zhang u32 flags, 136006164d2bSGeorge Zhang u32 priv_flags, 136106164d2bSGeorge Zhang u64 produce_size, 136206164d2bSGeorge Zhang u64 consume_size, 136306164d2bSGeorge Zhang struct vmci_qp_page_store *page_store, 136406164d2bSGeorge Zhang struct vmci_ctx *context, 136506164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb, 136606164d2bSGeorge Zhang void *client_data, struct qp_broker_entry **ent) 136706164d2bSGeorge Zhang { 136806164d2bSGeorge Zhang struct qp_broker_entry *entry = NULL; 136906164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 137006164d2bSGeorge Zhang bool is_local = flags & VMCI_QPFLAG_LOCAL; 137106164d2bSGeorge Zhang int result; 137206164d2bSGeorge Zhang u64 guest_produce_size; 137306164d2bSGeorge Zhang u64 guest_consume_size; 137406164d2bSGeorge Zhang 137506164d2bSGeorge Zhang /* Do not create if the caller asked not to. */ 137606164d2bSGeorge Zhang if (flags & VMCI_QPFLAG_ATTACH_ONLY) 137706164d2bSGeorge Zhang return VMCI_ERROR_NOT_FOUND; 137806164d2bSGeorge Zhang 137906164d2bSGeorge Zhang /* 138006164d2bSGeorge Zhang * Creator's context ID should match handle's context ID or the creator 138106164d2bSGeorge Zhang * must allow the context in handle's context ID as the "peer". 138206164d2bSGeorge Zhang */ 138306164d2bSGeorge Zhang if (handle.context != context_id && handle.context != peer) 138406164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 138506164d2bSGeorge Zhang 138606164d2bSGeorge Zhang if (VMCI_CONTEXT_IS_VM(context_id) && VMCI_CONTEXT_IS_VM(peer)) 138706164d2bSGeorge Zhang return VMCI_ERROR_DST_UNREACHABLE; 138806164d2bSGeorge Zhang 138906164d2bSGeorge Zhang /* 139006164d2bSGeorge Zhang * Creator's context ID for local queue pairs should match the 139106164d2bSGeorge Zhang * peer, if a peer is specified. 139206164d2bSGeorge Zhang */ 139306164d2bSGeorge Zhang if (is_local && peer != VMCI_INVALID_ID && context_id != peer) 139406164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 139506164d2bSGeorge Zhang 139606164d2bSGeorge Zhang entry = kzalloc(sizeof(*entry), GFP_ATOMIC); 139706164d2bSGeorge Zhang if (!entry) 139806164d2bSGeorge Zhang return VMCI_ERROR_NO_MEM; 139906164d2bSGeorge Zhang 140006164d2bSGeorge Zhang if (vmci_ctx_get_id(context) == VMCI_HOST_CONTEXT_ID && !is_local) { 140106164d2bSGeorge Zhang /* 140206164d2bSGeorge Zhang * The queue pair broker entry stores values from the guest 140306164d2bSGeorge Zhang * point of view, so a creating host side endpoint should swap 140406164d2bSGeorge Zhang * produce and consume values -- unless it is a local queue 140506164d2bSGeorge Zhang * pair, in which case no swapping is necessary, since the local 140606164d2bSGeorge Zhang * attacher will swap queues. 140706164d2bSGeorge Zhang */ 140806164d2bSGeorge Zhang 140906164d2bSGeorge Zhang guest_produce_size = consume_size; 141006164d2bSGeorge Zhang guest_consume_size = produce_size; 141106164d2bSGeorge Zhang } else { 141206164d2bSGeorge Zhang guest_produce_size = produce_size; 141306164d2bSGeorge Zhang guest_consume_size = consume_size; 141406164d2bSGeorge Zhang } 141506164d2bSGeorge Zhang 141606164d2bSGeorge Zhang entry->qp.handle = handle; 141706164d2bSGeorge Zhang entry->qp.peer = peer; 141806164d2bSGeorge Zhang entry->qp.flags = flags; 141906164d2bSGeorge Zhang entry->qp.produce_size = guest_produce_size; 142006164d2bSGeorge Zhang entry->qp.consume_size = guest_consume_size; 142106164d2bSGeorge Zhang entry->qp.ref_count = 1; 142206164d2bSGeorge Zhang entry->create_id = context_id; 142306164d2bSGeorge Zhang entry->attach_id = VMCI_INVALID_ID; 142406164d2bSGeorge Zhang entry->state = VMCIQPB_NEW; 142506164d2bSGeorge Zhang entry->require_trusted_attach = 142606164d2bSGeorge Zhang !!(context->priv_flags & VMCI_PRIVILEGE_FLAG_RESTRICTED); 142706164d2bSGeorge Zhang entry->created_by_trusted = 142806164d2bSGeorge Zhang !!(priv_flags & VMCI_PRIVILEGE_FLAG_TRUSTED); 142906164d2bSGeorge Zhang entry->vmci_page_files = false; 143006164d2bSGeorge Zhang entry->wakeup_cb = wakeup_cb; 143106164d2bSGeorge Zhang entry->client_data = client_data; 143206164d2bSGeorge Zhang entry->produce_q = qp_host_alloc_queue(guest_produce_size); 143306164d2bSGeorge Zhang if (entry->produce_q == NULL) { 143406164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 143506164d2bSGeorge Zhang goto error; 143606164d2bSGeorge Zhang } 143706164d2bSGeorge Zhang entry->consume_q = qp_host_alloc_queue(guest_consume_size); 143806164d2bSGeorge Zhang if (entry->consume_q == NULL) { 143906164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 144006164d2bSGeorge Zhang goto error; 144106164d2bSGeorge Zhang } 144206164d2bSGeorge Zhang 144306164d2bSGeorge Zhang qp_init_queue_mutex(entry->produce_q, entry->consume_q); 144406164d2bSGeorge Zhang 144506164d2bSGeorge Zhang INIT_LIST_HEAD(&entry->qp.list_item); 144606164d2bSGeorge Zhang 144706164d2bSGeorge Zhang if (is_local) { 144806164d2bSGeorge Zhang u8 *tmp; 144906164d2bSGeorge Zhang 145006164d2bSGeorge Zhang entry->local_mem = kcalloc(QPE_NUM_PAGES(entry->qp), 145106164d2bSGeorge Zhang PAGE_SIZE, GFP_KERNEL); 145206164d2bSGeorge Zhang if (entry->local_mem == NULL) { 145306164d2bSGeorge Zhang result = VMCI_ERROR_NO_MEM; 145406164d2bSGeorge Zhang goto error; 145506164d2bSGeorge Zhang } 145606164d2bSGeorge Zhang entry->state = VMCIQPB_CREATED_MEM; 145706164d2bSGeorge Zhang entry->produce_q->q_header = entry->local_mem; 145806164d2bSGeorge Zhang tmp = (u8 *)entry->local_mem + PAGE_SIZE * 145906164d2bSGeorge Zhang (dm_div_up(entry->qp.produce_size, PAGE_SIZE) + 1); 146006164d2bSGeorge Zhang entry->consume_q->q_header = (struct vmci_queue_header *)tmp; 146106164d2bSGeorge Zhang } else if (page_store) { 146206164d2bSGeorge Zhang /* 146306164d2bSGeorge Zhang * The VMX already initialized the queue pair headers, so no 146406164d2bSGeorge Zhang * need for the kernel side to do that. 146506164d2bSGeorge Zhang */ 146606164d2bSGeorge Zhang result = qp_host_register_user_memory(page_store, 146706164d2bSGeorge Zhang entry->produce_q, 146806164d2bSGeorge Zhang entry->consume_q); 146906164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 147006164d2bSGeorge Zhang goto error; 147106164d2bSGeorge Zhang 147206164d2bSGeorge Zhang entry->state = VMCIQPB_CREATED_MEM; 147306164d2bSGeorge Zhang } else { 147406164d2bSGeorge Zhang /* 147506164d2bSGeorge Zhang * A create without a page_store may be either a host 147606164d2bSGeorge Zhang * side create (in which case we are waiting for the 147706164d2bSGeorge Zhang * guest side to supply the memory) or an old style 147806164d2bSGeorge Zhang * queue pair create (in which case we will expect a 147906164d2bSGeorge Zhang * set page store call as the next step). 148006164d2bSGeorge Zhang */ 148106164d2bSGeorge Zhang entry->state = VMCIQPB_CREATED_NO_MEM; 148206164d2bSGeorge Zhang } 148306164d2bSGeorge Zhang 148406164d2bSGeorge Zhang qp_list_add_entry(&qp_broker_list, &entry->qp); 148506164d2bSGeorge Zhang if (ent != NULL) 148606164d2bSGeorge Zhang *ent = entry; 148706164d2bSGeorge Zhang 148806164d2bSGeorge Zhang /* Add to resource obj */ 148906164d2bSGeorge Zhang result = vmci_resource_add(&entry->resource, 149006164d2bSGeorge Zhang VMCI_RESOURCE_TYPE_QPAIR_HOST, 149106164d2bSGeorge Zhang handle); 149206164d2bSGeorge Zhang if (result != VMCI_SUCCESS) { 149306164d2bSGeorge Zhang pr_warn("Failed to add new resource (handle=0x%x:0x%x), error: %d", 149406164d2bSGeorge Zhang handle.context, handle.resource, result); 149506164d2bSGeorge Zhang goto error; 149606164d2bSGeorge Zhang } 149706164d2bSGeorge Zhang 149806164d2bSGeorge Zhang entry->qp.handle = vmci_resource_handle(&entry->resource); 149906164d2bSGeorge Zhang if (is_local) { 150006164d2bSGeorge Zhang vmci_q_header_init(entry->produce_q->q_header, 150106164d2bSGeorge Zhang entry->qp.handle); 150206164d2bSGeorge Zhang vmci_q_header_init(entry->consume_q->q_header, 150306164d2bSGeorge Zhang entry->qp.handle); 150406164d2bSGeorge Zhang } 150506164d2bSGeorge Zhang 150606164d2bSGeorge Zhang vmci_ctx_qp_create(context, entry->qp.handle); 150706164d2bSGeorge Zhang 150806164d2bSGeorge Zhang return VMCI_SUCCESS; 150906164d2bSGeorge Zhang 151006164d2bSGeorge Zhang error: 151106164d2bSGeorge Zhang if (entry != NULL) { 151206164d2bSGeorge Zhang qp_host_free_queue(entry->produce_q, guest_produce_size); 151306164d2bSGeorge Zhang qp_host_free_queue(entry->consume_q, guest_consume_size); 151406164d2bSGeorge Zhang kfree(entry); 151506164d2bSGeorge Zhang } 151606164d2bSGeorge Zhang 151706164d2bSGeorge Zhang return result; 151806164d2bSGeorge Zhang } 151906164d2bSGeorge Zhang 152006164d2bSGeorge Zhang /* 152106164d2bSGeorge Zhang * Enqueues an event datagram to notify the peer VM attached to 152206164d2bSGeorge Zhang * the given queue pair handle about attach/detach event by the 152306164d2bSGeorge Zhang * given VM. Returns Payload size of datagram enqueued on 152406164d2bSGeorge Zhang * success, error code otherwise. 152506164d2bSGeorge Zhang */ 152606164d2bSGeorge Zhang static int qp_notify_peer(bool attach, 152706164d2bSGeorge Zhang struct vmci_handle handle, 152806164d2bSGeorge Zhang u32 my_id, 152906164d2bSGeorge Zhang u32 peer_id) 153006164d2bSGeorge Zhang { 153106164d2bSGeorge Zhang int rv; 153206164d2bSGeorge Zhang struct vmci_event_qp ev; 153306164d2bSGeorge Zhang 153406164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || my_id == VMCI_INVALID_ID || 153506164d2bSGeorge Zhang peer_id == VMCI_INVALID_ID) 153606164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 153706164d2bSGeorge Zhang 153806164d2bSGeorge Zhang /* 153906164d2bSGeorge Zhang * In vmci_ctx_enqueue_datagram() we enforce the upper limit on 154006164d2bSGeorge Zhang * number of pending events from the hypervisor to a given VM 154106164d2bSGeorge Zhang * otherwise a rogue VM could do an arbitrary number of attach 154206164d2bSGeorge Zhang * and detach operations causing memory pressure in the host 154306164d2bSGeorge Zhang * kernel. 154406164d2bSGeorge Zhang */ 154506164d2bSGeorge Zhang 154606164d2bSGeorge Zhang ev.msg.hdr.dst = vmci_make_handle(peer_id, VMCI_EVENT_HANDLER); 154706164d2bSGeorge Zhang ev.msg.hdr.src = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, 154806164d2bSGeorge Zhang VMCI_CONTEXT_RESOURCE_ID); 154906164d2bSGeorge Zhang ev.msg.hdr.payload_size = sizeof(ev) - sizeof(ev.msg.hdr); 155006164d2bSGeorge Zhang ev.msg.event_data.event = attach ? 155106164d2bSGeorge Zhang VMCI_EVENT_QP_PEER_ATTACH : VMCI_EVENT_QP_PEER_DETACH; 155206164d2bSGeorge Zhang ev.payload.handle = handle; 155306164d2bSGeorge Zhang ev.payload.peer_id = my_id; 155406164d2bSGeorge Zhang 155506164d2bSGeorge Zhang rv = vmci_datagram_dispatch(VMCI_HYPERVISOR_CONTEXT_ID, 155606164d2bSGeorge Zhang &ev.msg.hdr, false); 155706164d2bSGeorge Zhang if (rv < VMCI_SUCCESS) 155806164d2bSGeorge Zhang pr_warn("Failed to enqueue queue_pair %s event datagram for context (ID=0x%x)\n", 155906164d2bSGeorge Zhang attach ? "ATTACH" : "DETACH", peer_id); 156006164d2bSGeorge Zhang 156106164d2bSGeorge Zhang return rv; 156206164d2bSGeorge Zhang } 156306164d2bSGeorge Zhang 156406164d2bSGeorge Zhang /* 156506164d2bSGeorge Zhang * The second endpoint issuing a queue pair allocation will attach to 156606164d2bSGeorge Zhang * the queue pair registered with the queue pair broker. 156706164d2bSGeorge Zhang * 156806164d2bSGeorge Zhang * If the attacher is a guest, it will associate a VMX virtual address 156906164d2bSGeorge Zhang * range with the queue pair as specified by the page_store. At this 157006164d2bSGeorge Zhang * point, the already attach host endpoint may start using the queue 157106164d2bSGeorge Zhang * pair, and an attach event is sent to it. For compatibility with 157206164d2bSGeorge Zhang * older VMX'en, that used a separate step to set the VMX virtual 157306164d2bSGeorge Zhang * address range, the virtual address range can be registered later 157406164d2bSGeorge Zhang * using vmci_qp_broker_set_page_store. In that case, a page_store of 157506164d2bSGeorge Zhang * NULL should be used, and the attach event will be generated once 157606164d2bSGeorge Zhang * the actual page store has been set. 157706164d2bSGeorge Zhang * 157806164d2bSGeorge Zhang * If the attacher is the host, a page_store of NULL should be used as 157906164d2bSGeorge Zhang * well, since the page store information is already set by the guest. 158006164d2bSGeorge Zhang * 158106164d2bSGeorge Zhang * For new VMX and host callers, the queue pair will be moved to the 158206164d2bSGeorge Zhang * VMCIQPB_ATTACHED_MEM state, and for older VMX callers, it will be 158306164d2bSGeorge Zhang * moved to the VMCOQPB_ATTACHED_NO_MEM state. 158406164d2bSGeorge Zhang */ 158506164d2bSGeorge Zhang static int qp_broker_attach(struct qp_broker_entry *entry, 158606164d2bSGeorge Zhang u32 peer, 158706164d2bSGeorge Zhang u32 flags, 158806164d2bSGeorge Zhang u32 priv_flags, 158906164d2bSGeorge Zhang u64 produce_size, 159006164d2bSGeorge Zhang u64 consume_size, 159106164d2bSGeorge Zhang struct vmci_qp_page_store *page_store, 159206164d2bSGeorge Zhang struct vmci_ctx *context, 159306164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb, 159406164d2bSGeorge Zhang void *client_data, 159506164d2bSGeorge Zhang struct qp_broker_entry **ent) 159606164d2bSGeorge Zhang { 159706164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 159806164d2bSGeorge Zhang bool is_local = flags & VMCI_QPFLAG_LOCAL; 159906164d2bSGeorge Zhang int result; 160006164d2bSGeorge Zhang 160106164d2bSGeorge Zhang if (entry->state != VMCIQPB_CREATED_NO_MEM && 160206164d2bSGeorge Zhang entry->state != VMCIQPB_CREATED_MEM) 160306164d2bSGeorge Zhang return VMCI_ERROR_UNAVAILABLE; 160406164d2bSGeorge Zhang 160506164d2bSGeorge Zhang if (is_local) { 160606164d2bSGeorge Zhang if (!(entry->qp.flags & VMCI_QPFLAG_LOCAL) || 160706164d2bSGeorge Zhang context_id != entry->create_id) { 160806164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 160906164d2bSGeorge Zhang } 161006164d2bSGeorge Zhang } else if (context_id == entry->create_id || 161106164d2bSGeorge Zhang context_id == entry->attach_id) { 161206164d2bSGeorge Zhang return VMCI_ERROR_ALREADY_EXISTS; 161306164d2bSGeorge Zhang } 161406164d2bSGeorge Zhang 161506164d2bSGeorge Zhang if (VMCI_CONTEXT_IS_VM(context_id) && 161606164d2bSGeorge Zhang VMCI_CONTEXT_IS_VM(entry->create_id)) 161706164d2bSGeorge Zhang return VMCI_ERROR_DST_UNREACHABLE; 161806164d2bSGeorge Zhang 161906164d2bSGeorge Zhang /* 162006164d2bSGeorge Zhang * If we are attaching from a restricted context then the queuepair 162106164d2bSGeorge Zhang * must have been created by a trusted endpoint. 162206164d2bSGeorge Zhang */ 162306164d2bSGeorge Zhang if ((context->priv_flags & VMCI_PRIVILEGE_FLAG_RESTRICTED) && 162406164d2bSGeorge Zhang !entry->created_by_trusted) 162506164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 162606164d2bSGeorge Zhang 162706164d2bSGeorge Zhang /* 162806164d2bSGeorge Zhang * If we are attaching to a queuepair that was created by a restricted 162906164d2bSGeorge Zhang * context then we must be trusted. 163006164d2bSGeorge Zhang */ 163106164d2bSGeorge Zhang if (entry->require_trusted_attach && 163206164d2bSGeorge Zhang (!(priv_flags & VMCI_PRIVILEGE_FLAG_TRUSTED))) 163306164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 163406164d2bSGeorge Zhang 163506164d2bSGeorge Zhang /* 163606164d2bSGeorge Zhang * If the creator specifies VMCI_INVALID_ID in "peer" field, access 163706164d2bSGeorge Zhang * control check is not performed. 163806164d2bSGeorge Zhang */ 163906164d2bSGeorge Zhang if (entry->qp.peer != VMCI_INVALID_ID && entry->qp.peer != context_id) 164006164d2bSGeorge Zhang return VMCI_ERROR_NO_ACCESS; 164106164d2bSGeorge Zhang 164206164d2bSGeorge Zhang if (entry->create_id == VMCI_HOST_CONTEXT_ID) { 164306164d2bSGeorge Zhang /* 164406164d2bSGeorge Zhang * Do not attach if the caller doesn't support Host Queue Pairs 164506164d2bSGeorge Zhang * and a host created this queue pair. 164606164d2bSGeorge Zhang */ 164706164d2bSGeorge Zhang 164806164d2bSGeorge Zhang if (!vmci_ctx_supports_host_qp(context)) 164906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_RESOURCE; 165006164d2bSGeorge Zhang 165106164d2bSGeorge Zhang } else if (context_id == VMCI_HOST_CONTEXT_ID) { 165206164d2bSGeorge Zhang struct vmci_ctx *create_context; 165306164d2bSGeorge Zhang bool supports_host_qp; 165406164d2bSGeorge Zhang 165506164d2bSGeorge Zhang /* 165606164d2bSGeorge Zhang * Do not attach a host to a user created queue pair if that 165706164d2bSGeorge Zhang * user doesn't support host queue pair end points. 165806164d2bSGeorge Zhang */ 165906164d2bSGeorge Zhang 166006164d2bSGeorge Zhang create_context = vmci_ctx_get(entry->create_id); 166106164d2bSGeorge Zhang supports_host_qp = vmci_ctx_supports_host_qp(create_context); 166206164d2bSGeorge Zhang vmci_ctx_put(create_context); 166306164d2bSGeorge Zhang 166406164d2bSGeorge Zhang if (!supports_host_qp) 166506164d2bSGeorge Zhang return VMCI_ERROR_INVALID_RESOURCE; 166606164d2bSGeorge Zhang } 166706164d2bSGeorge Zhang 166806164d2bSGeorge Zhang if ((entry->qp.flags & ~VMCI_QP_ASYMM) != (flags & ~VMCI_QP_ASYMM_PEER)) 166906164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_MISMATCH; 167006164d2bSGeorge Zhang 167106164d2bSGeorge Zhang if (context_id != VMCI_HOST_CONTEXT_ID) { 167206164d2bSGeorge Zhang /* 167306164d2bSGeorge Zhang * The queue pair broker entry stores values from the guest 167406164d2bSGeorge Zhang * point of view, so an attaching guest should match the values 167506164d2bSGeorge Zhang * stored in the entry. 167606164d2bSGeorge Zhang */ 167706164d2bSGeorge Zhang 167806164d2bSGeorge Zhang if (entry->qp.produce_size != produce_size || 167906164d2bSGeorge Zhang entry->qp.consume_size != consume_size) { 168006164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_MISMATCH; 168106164d2bSGeorge Zhang } 168206164d2bSGeorge Zhang } else if (entry->qp.produce_size != consume_size || 168306164d2bSGeorge Zhang entry->qp.consume_size != produce_size) { 168406164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_MISMATCH; 168506164d2bSGeorge Zhang } 168606164d2bSGeorge Zhang 168706164d2bSGeorge Zhang if (context_id != VMCI_HOST_CONTEXT_ID) { 168806164d2bSGeorge Zhang /* 168906164d2bSGeorge Zhang * If a guest attached to a queue pair, it will supply 169006164d2bSGeorge Zhang * the backing memory. If this is a pre NOVMVM vmx, 169106164d2bSGeorge Zhang * the backing memory will be supplied by calling 169206164d2bSGeorge Zhang * vmci_qp_broker_set_page_store() following the 169306164d2bSGeorge Zhang * return of the vmci_qp_broker_alloc() call. If it is 169406164d2bSGeorge Zhang * a vmx of version NOVMVM or later, the page store 169506164d2bSGeorge Zhang * must be supplied as part of the 169606164d2bSGeorge Zhang * vmci_qp_broker_alloc call. Under all circumstances 169706164d2bSGeorge Zhang * must the initially created queue pair not have any 169806164d2bSGeorge Zhang * memory associated with it already. 169906164d2bSGeorge Zhang */ 170006164d2bSGeorge Zhang 170106164d2bSGeorge Zhang if (entry->state != VMCIQPB_CREATED_NO_MEM) 170206164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 170306164d2bSGeorge Zhang 170406164d2bSGeorge Zhang if (page_store != NULL) { 170506164d2bSGeorge Zhang /* 170606164d2bSGeorge Zhang * Patch up host state to point to guest 170706164d2bSGeorge Zhang * supplied memory. The VMX already 170806164d2bSGeorge Zhang * initialized the queue pair headers, so no 170906164d2bSGeorge Zhang * need for the kernel side to do that. 171006164d2bSGeorge Zhang */ 171106164d2bSGeorge Zhang 171206164d2bSGeorge Zhang result = qp_host_register_user_memory(page_store, 171306164d2bSGeorge Zhang entry->produce_q, 171406164d2bSGeorge Zhang entry->consume_q); 171506164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 171606164d2bSGeorge Zhang return result; 171706164d2bSGeorge Zhang 171806164d2bSGeorge Zhang /* 171906164d2bSGeorge Zhang * Preemptively load in the headers if non-blocking to 172006164d2bSGeorge Zhang * prevent blocking later. 172106164d2bSGeorge Zhang */ 172206164d2bSGeorge Zhang if (entry->qp.flags & VMCI_QPFLAG_NONBLOCK) { 172306164d2bSGeorge Zhang result = qp_host_map_queues(entry->produce_q, 172406164d2bSGeorge Zhang entry->consume_q); 172506164d2bSGeorge Zhang if (result < VMCI_SUCCESS) { 172606164d2bSGeorge Zhang qp_host_unregister_user_memory( 172706164d2bSGeorge Zhang entry->produce_q, 172806164d2bSGeorge Zhang entry->consume_q); 172906164d2bSGeorge Zhang return result; 173006164d2bSGeorge Zhang } 173106164d2bSGeorge Zhang } 173206164d2bSGeorge Zhang 173306164d2bSGeorge Zhang entry->state = VMCIQPB_ATTACHED_MEM; 173406164d2bSGeorge Zhang } else { 173506164d2bSGeorge Zhang entry->state = VMCIQPB_ATTACHED_NO_MEM; 173606164d2bSGeorge Zhang } 173706164d2bSGeorge Zhang } else if (entry->state == VMCIQPB_CREATED_NO_MEM) { 173806164d2bSGeorge Zhang /* 173906164d2bSGeorge Zhang * The host side is attempting to attach to a queue 174006164d2bSGeorge Zhang * pair that doesn't have any memory associated with 174106164d2bSGeorge Zhang * it. This must be a pre NOVMVM vmx that hasn't set 174206164d2bSGeorge Zhang * the page store information yet, or a quiesced VM. 174306164d2bSGeorge Zhang */ 174406164d2bSGeorge Zhang 174506164d2bSGeorge Zhang return VMCI_ERROR_UNAVAILABLE; 174606164d2bSGeorge Zhang } else { 174706164d2bSGeorge Zhang /* 174806164d2bSGeorge Zhang * For non-blocking queue pairs, we cannot rely on 174906164d2bSGeorge Zhang * enqueue/dequeue to map in the pages on the 175006164d2bSGeorge Zhang * host-side, since it may block, so we make an 175106164d2bSGeorge Zhang * attempt here. 175206164d2bSGeorge Zhang */ 175306164d2bSGeorge Zhang 175406164d2bSGeorge Zhang if (flags & VMCI_QPFLAG_NONBLOCK) { 175506164d2bSGeorge Zhang result = 175606164d2bSGeorge Zhang qp_host_map_queues(entry->produce_q, 175706164d2bSGeorge Zhang entry->consume_q); 175806164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 175906164d2bSGeorge Zhang return result; 176006164d2bSGeorge Zhang 176106164d2bSGeorge Zhang entry->qp.flags |= flags & 176206164d2bSGeorge Zhang (VMCI_QPFLAG_NONBLOCK | VMCI_QPFLAG_PINNED); 176306164d2bSGeorge Zhang } 176406164d2bSGeorge Zhang 176506164d2bSGeorge Zhang /* The host side has successfully attached to a queue pair. */ 176606164d2bSGeorge Zhang entry->state = VMCIQPB_ATTACHED_MEM; 176706164d2bSGeorge Zhang } 176806164d2bSGeorge Zhang 176906164d2bSGeorge Zhang if (entry->state == VMCIQPB_ATTACHED_MEM) { 177006164d2bSGeorge Zhang result = 177106164d2bSGeorge Zhang qp_notify_peer(true, entry->qp.handle, context_id, 177206164d2bSGeorge Zhang entry->create_id); 177306164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 177406164d2bSGeorge Zhang pr_warn("Failed to notify peer (ID=0x%x) of attach to queue pair (handle=0x%x:0x%x)\n", 177506164d2bSGeorge Zhang entry->create_id, entry->qp.handle.context, 177606164d2bSGeorge Zhang entry->qp.handle.resource); 177706164d2bSGeorge Zhang } 177806164d2bSGeorge Zhang 177906164d2bSGeorge Zhang entry->attach_id = context_id; 178006164d2bSGeorge Zhang entry->qp.ref_count++; 178106164d2bSGeorge Zhang if (wakeup_cb) { 178206164d2bSGeorge Zhang entry->wakeup_cb = wakeup_cb; 178306164d2bSGeorge Zhang entry->client_data = client_data; 178406164d2bSGeorge Zhang } 178506164d2bSGeorge Zhang 178606164d2bSGeorge Zhang /* 178706164d2bSGeorge Zhang * When attaching to local queue pairs, the context already has 178806164d2bSGeorge Zhang * an entry tracking the queue pair, so don't add another one. 178906164d2bSGeorge Zhang */ 179006164d2bSGeorge Zhang if (!is_local) 179106164d2bSGeorge Zhang vmci_ctx_qp_create(context, entry->qp.handle); 179206164d2bSGeorge Zhang 179306164d2bSGeorge Zhang if (ent != NULL) 179406164d2bSGeorge Zhang *ent = entry; 179506164d2bSGeorge Zhang 179606164d2bSGeorge Zhang return VMCI_SUCCESS; 179706164d2bSGeorge Zhang } 179806164d2bSGeorge Zhang 179906164d2bSGeorge Zhang /* 180006164d2bSGeorge Zhang * queue_pair_Alloc for use when setting up queue pair endpoints 180106164d2bSGeorge Zhang * on the host. 180206164d2bSGeorge Zhang */ 180306164d2bSGeorge Zhang static int qp_broker_alloc(struct vmci_handle handle, 180406164d2bSGeorge Zhang u32 peer, 180506164d2bSGeorge Zhang u32 flags, 180606164d2bSGeorge Zhang u32 priv_flags, 180706164d2bSGeorge Zhang u64 produce_size, 180806164d2bSGeorge Zhang u64 consume_size, 180906164d2bSGeorge Zhang struct vmci_qp_page_store *page_store, 181006164d2bSGeorge Zhang struct vmci_ctx *context, 181106164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb, 181206164d2bSGeorge Zhang void *client_data, 181306164d2bSGeorge Zhang struct qp_broker_entry **ent, 181406164d2bSGeorge Zhang bool *swap) 181506164d2bSGeorge Zhang { 181606164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 181706164d2bSGeorge Zhang bool create; 181806164d2bSGeorge Zhang struct qp_broker_entry *entry = NULL; 181906164d2bSGeorge Zhang bool is_local = flags & VMCI_QPFLAG_LOCAL; 182006164d2bSGeorge Zhang int result; 182106164d2bSGeorge Zhang 182206164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || 182306164d2bSGeorge Zhang (flags & ~VMCI_QP_ALL_FLAGS) || is_local || 182406164d2bSGeorge Zhang !(produce_size || consume_size) || 182506164d2bSGeorge Zhang !context || context_id == VMCI_INVALID_ID || 182606164d2bSGeorge Zhang handle.context == VMCI_INVALID_ID) { 182706164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 182806164d2bSGeorge Zhang } 182906164d2bSGeorge Zhang 183006164d2bSGeorge Zhang if (page_store && !VMCI_QP_PAGESTORE_IS_WELLFORMED(page_store)) 183106164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 183206164d2bSGeorge Zhang 183306164d2bSGeorge Zhang /* 183406164d2bSGeorge Zhang * In the initial argument check, we ensure that non-vmkernel hosts 183506164d2bSGeorge Zhang * are not allowed to create local queue pairs. 183606164d2bSGeorge Zhang */ 183706164d2bSGeorge Zhang 183806164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 183906164d2bSGeorge Zhang 184006164d2bSGeorge Zhang if (!is_local && vmci_ctx_qp_exists(context, handle)) { 184106164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) already attached to queue pair (handle=0x%x:0x%x)\n", 184206164d2bSGeorge Zhang context_id, handle.context, handle.resource); 184306164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 184406164d2bSGeorge Zhang return VMCI_ERROR_ALREADY_EXISTS; 184506164d2bSGeorge Zhang } 184606164d2bSGeorge Zhang 184706164d2bSGeorge Zhang if (handle.resource != VMCI_INVALID_ID) 184806164d2bSGeorge Zhang entry = qp_broker_handle_to_entry(handle); 184906164d2bSGeorge Zhang 185006164d2bSGeorge Zhang if (!entry) { 185106164d2bSGeorge Zhang create = true; 185206164d2bSGeorge Zhang result = 185306164d2bSGeorge Zhang qp_broker_create(handle, peer, flags, priv_flags, 185406164d2bSGeorge Zhang produce_size, consume_size, page_store, 185506164d2bSGeorge Zhang context, wakeup_cb, client_data, ent); 185606164d2bSGeorge Zhang } else { 185706164d2bSGeorge Zhang create = false; 185806164d2bSGeorge Zhang result = 185906164d2bSGeorge Zhang qp_broker_attach(entry, peer, flags, priv_flags, 186006164d2bSGeorge Zhang produce_size, consume_size, page_store, 186106164d2bSGeorge Zhang context, wakeup_cb, client_data, ent); 186206164d2bSGeorge Zhang } 186306164d2bSGeorge Zhang 186406164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 186506164d2bSGeorge Zhang 186606164d2bSGeorge Zhang if (swap) 186706164d2bSGeorge Zhang *swap = (context_id == VMCI_HOST_CONTEXT_ID) && 186806164d2bSGeorge Zhang !(create && is_local); 186906164d2bSGeorge Zhang 187006164d2bSGeorge Zhang return result; 187106164d2bSGeorge Zhang } 187206164d2bSGeorge Zhang 187306164d2bSGeorge Zhang /* 187406164d2bSGeorge Zhang * This function implements the kernel API for allocating a queue 187506164d2bSGeorge Zhang * pair. 187606164d2bSGeorge Zhang */ 187706164d2bSGeorge Zhang static int qp_alloc_host_work(struct vmci_handle *handle, 187806164d2bSGeorge Zhang struct vmci_queue **produce_q, 187906164d2bSGeorge Zhang u64 produce_size, 188006164d2bSGeorge Zhang struct vmci_queue **consume_q, 188106164d2bSGeorge Zhang u64 consume_size, 188206164d2bSGeorge Zhang u32 peer, 188306164d2bSGeorge Zhang u32 flags, 188406164d2bSGeorge Zhang u32 priv_flags, 188506164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb, 188606164d2bSGeorge Zhang void *client_data) 188706164d2bSGeorge Zhang { 188806164d2bSGeorge Zhang struct vmci_handle new_handle; 188906164d2bSGeorge Zhang struct vmci_ctx *context; 189006164d2bSGeorge Zhang struct qp_broker_entry *entry; 189106164d2bSGeorge Zhang int result; 189206164d2bSGeorge Zhang bool swap; 189306164d2bSGeorge Zhang 189406164d2bSGeorge Zhang if (vmci_handle_is_invalid(*handle)) { 189506164d2bSGeorge Zhang new_handle = vmci_make_handle( 189606164d2bSGeorge Zhang VMCI_HOST_CONTEXT_ID, VMCI_INVALID_ID); 189706164d2bSGeorge Zhang } else 189806164d2bSGeorge Zhang new_handle = *handle; 189906164d2bSGeorge Zhang 190006164d2bSGeorge Zhang context = vmci_ctx_get(VMCI_HOST_CONTEXT_ID); 190106164d2bSGeorge Zhang entry = NULL; 190206164d2bSGeorge Zhang result = 190306164d2bSGeorge Zhang qp_broker_alloc(new_handle, peer, flags, priv_flags, 190406164d2bSGeorge Zhang produce_size, consume_size, NULL, context, 190506164d2bSGeorge Zhang wakeup_cb, client_data, &entry, &swap); 190606164d2bSGeorge Zhang if (result == VMCI_SUCCESS) { 190706164d2bSGeorge Zhang if (swap) { 190806164d2bSGeorge Zhang /* 190906164d2bSGeorge Zhang * If this is a local queue pair, the attacher 191006164d2bSGeorge Zhang * will swap around produce and consume 191106164d2bSGeorge Zhang * queues. 191206164d2bSGeorge Zhang */ 191306164d2bSGeorge Zhang 191406164d2bSGeorge Zhang *produce_q = entry->consume_q; 191506164d2bSGeorge Zhang *consume_q = entry->produce_q; 191606164d2bSGeorge Zhang } else { 191706164d2bSGeorge Zhang *produce_q = entry->produce_q; 191806164d2bSGeorge Zhang *consume_q = entry->consume_q; 191906164d2bSGeorge Zhang } 192006164d2bSGeorge Zhang 192106164d2bSGeorge Zhang *handle = vmci_resource_handle(&entry->resource); 192206164d2bSGeorge Zhang } else { 192306164d2bSGeorge Zhang *handle = VMCI_INVALID_HANDLE; 192406164d2bSGeorge Zhang pr_devel("queue pair broker failed to alloc (result=%d)\n", 192506164d2bSGeorge Zhang result); 192606164d2bSGeorge Zhang } 192706164d2bSGeorge Zhang vmci_ctx_put(context); 192806164d2bSGeorge Zhang return result; 192906164d2bSGeorge Zhang } 193006164d2bSGeorge Zhang 193106164d2bSGeorge Zhang /* 193206164d2bSGeorge Zhang * Allocates a VMCI queue_pair. Only checks validity of input 193306164d2bSGeorge Zhang * arguments. The real work is done in the host or guest 193406164d2bSGeorge Zhang * specific function. 193506164d2bSGeorge Zhang */ 193606164d2bSGeorge Zhang int vmci_qp_alloc(struct vmci_handle *handle, 193706164d2bSGeorge Zhang struct vmci_queue **produce_q, 193806164d2bSGeorge Zhang u64 produce_size, 193906164d2bSGeorge Zhang struct vmci_queue **consume_q, 194006164d2bSGeorge Zhang u64 consume_size, 194106164d2bSGeorge Zhang u32 peer, 194206164d2bSGeorge Zhang u32 flags, 194306164d2bSGeorge Zhang u32 priv_flags, 194406164d2bSGeorge Zhang bool guest_endpoint, 194506164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb, 194606164d2bSGeorge Zhang void *client_data) 194706164d2bSGeorge Zhang { 194806164d2bSGeorge Zhang if (!handle || !produce_q || !consume_q || 194906164d2bSGeorge Zhang (!produce_size && !consume_size) || (flags & ~VMCI_QP_ALL_FLAGS)) 195006164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 195106164d2bSGeorge Zhang 195206164d2bSGeorge Zhang if (guest_endpoint) { 195306164d2bSGeorge Zhang return qp_alloc_guest_work(handle, produce_q, 195406164d2bSGeorge Zhang produce_size, consume_q, 195506164d2bSGeorge Zhang consume_size, peer, 195606164d2bSGeorge Zhang flags, priv_flags); 195706164d2bSGeorge Zhang } else { 195806164d2bSGeorge Zhang return qp_alloc_host_work(handle, produce_q, 195906164d2bSGeorge Zhang produce_size, consume_q, 196006164d2bSGeorge Zhang consume_size, peer, flags, 196106164d2bSGeorge Zhang priv_flags, wakeup_cb, client_data); 196206164d2bSGeorge Zhang } 196306164d2bSGeorge Zhang } 196406164d2bSGeorge Zhang 196506164d2bSGeorge Zhang /* 196606164d2bSGeorge Zhang * This function implements the host kernel API for detaching from 196706164d2bSGeorge Zhang * a queue pair. 196806164d2bSGeorge Zhang */ 196906164d2bSGeorge Zhang static int qp_detatch_host_work(struct vmci_handle handle) 197006164d2bSGeorge Zhang { 197106164d2bSGeorge Zhang int result; 197206164d2bSGeorge Zhang struct vmci_ctx *context; 197306164d2bSGeorge Zhang 197406164d2bSGeorge Zhang context = vmci_ctx_get(VMCI_HOST_CONTEXT_ID); 197506164d2bSGeorge Zhang 197606164d2bSGeorge Zhang result = vmci_qp_broker_detach(handle, context); 197706164d2bSGeorge Zhang 197806164d2bSGeorge Zhang vmci_ctx_put(context); 197906164d2bSGeorge Zhang return result; 198006164d2bSGeorge Zhang } 198106164d2bSGeorge Zhang 198206164d2bSGeorge Zhang /* 198306164d2bSGeorge Zhang * Detaches from a VMCI queue_pair. Only checks validity of input argument. 198406164d2bSGeorge Zhang * Real work is done in the host or guest specific function. 198506164d2bSGeorge Zhang */ 198606164d2bSGeorge Zhang static int qp_detatch(struct vmci_handle handle, bool guest_endpoint) 198706164d2bSGeorge Zhang { 198806164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle)) 198906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 199006164d2bSGeorge Zhang 199106164d2bSGeorge Zhang if (guest_endpoint) 199206164d2bSGeorge Zhang return qp_detatch_guest_work(handle); 199306164d2bSGeorge Zhang else 199406164d2bSGeorge Zhang return qp_detatch_host_work(handle); 199506164d2bSGeorge Zhang } 199606164d2bSGeorge Zhang 199706164d2bSGeorge Zhang /* 199806164d2bSGeorge Zhang * Returns the entry from the head of the list. Assumes that the list is 199906164d2bSGeorge Zhang * locked. 200006164d2bSGeorge Zhang */ 200106164d2bSGeorge Zhang static struct qp_entry *qp_list_get_head(struct qp_list *qp_list) 200206164d2bSGeorge Zhang { 200306164d2bSGeorge Zhang if (!list_empty(&qp_list->head)) { 200406164d2bSGeorge Zhang struct qp_entry *entry = 200506164d2bSGeorge Zhang list_first_entry(&qp_list->head, struct qp_entry, 200606164d2bSGeorge Zhang list_item); 200706164d2bSGeorge Zhang return entry; 200806164d2bSGeorge Zhang } 200906164d2bSGeorge Zhang 201006164d2bSGeorge Zhang return NULL; 201106164d2bSGeorge Zhang } 201206164d2bSGeorge Zhang 201306164d2bSGeorge Zhang void vmci_qp_broker_exit(void) 201406164d2bSGeorge Zhang { 201506164d2bSGeorge Zhang struct qp_entry *entry; 201606164d2bSGeorge Zhang struct qp_broker_entry *be; 201706164d2bSGeorge Zhang 201806164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 201906164d2bSGeorge Zhang 202006164d2bSGeorge Zhang while ((entry = qp_list_get_head(&qp_broker_list))) { 202106164d2bSGeorge Zhang be = (struct qp_broker_entry *)entry; 202206164d2bSGeorge Zhang 202306164d2bSGeorge Zhang qp_list_remove_entry(&qp_broker_list, entry); 202406164d2bSGeorge Zhang kfree(be); 202506164d2bSGeorge Zhang } 202606164d2bSGeorge Zhang 202706164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 202806164d2bSGeorge Zhang } 202906164d2bSGeorge Zhang 203006164d2bSGeorge Zhang /* 203106164d2bSGeorge Zhang * Requests that a queue pair be allocated with the VMCI queue 203206164d2bSGeorge Zhang * pair broker. Allocates a queue pair entry if one does not 203306164d2bSGeorge Zhang * exist. Attaches to one if it exists, and retrieves the page 203406164d2bSGeorge Zhang * files backing that queue_pair. Assumes that the queue pair 203506164d2bSGeorge Zhang * broker lock is held. 203606164d2bSGeorge Zhang */ 203706164d2bSGeorge Zhang int vmci_qp_broker_alloc(struct vmci_handle handle, 203806164d2bSGeorge Zhang u32 peer, 203906164d2bSGeorge Zhang u32 flags, 204006164d2bSGeorge Zhang u32 priv_flags, 204106164d2bSGeorge Zhang u64 produce_size, 204206164d2bSGeorge Zhang u64 consume_size, 204306164d2bSGeorge Zhang struct vmci_qp_page_store *page_store, 204406164d2bSGeorge Zhang struct vmci_ctx *context) 204506164d2bSGeorge Zhang { 204606164d2bSGeorge Zhang return qp_broker_alloc(handle, peer, flags, priv_flags, 204706164d2bSGeorge Zhang produce_size, consume_size, 204806164d2bSGeorge Zhang page_store, context, NULL, NULL, NULL, NULL); 204906164d2bSGeorge Zhang } 205006164d2bSGeorge Zhang 205106164d2bSGeorge Zhang /* 205206164d2bSGeorge Zhang * VMX'en with versions lower than VMCI_VERSION_NOVMVM use a separate 205306164d2bSGeorge Zhang * step to add the UVAs of the VMX mapping of the queue pair. This function 205406164d2bSGeorge Zhang * provides backwards compatibility with such VMX'en, and takes care of 205506164d2bSGeorge Zhang * registering the page store for a queue pair previously allocated by the 205606164d2bSGeorge Zhang * VMX during create or attach. This function will move the queue pair state 205706164d2bSGeorge Zhang * to either from VMCIQBP_CREATED_NO_MEM to VMCIQBP_CREATED_MEM or 205806164d2bSGeorge Zhang * VMCIQBP_ATTACHED_NO_MEM to VMCIQBP_ATTACHED_MEM. If moving to the 205906164d2bSGeorge Zhang * attached state with memory, the queue pair is ready to be used by the 206006164d2bSGeorge Zhang * host peer, and an attached event will be generated. 206106164d2bSGeorge Zhang * 206206164d2bSGeorge Zhang * Assumes that the queue pair broker lock is held. 206306164d2bSGeorge Zhang * 206406164d2bSGeorge Zhang * This function is only used by the hosted platform, since there is no 206506164d2bSGeorge Zhang * issue with backwards compatibility for vmkernel. 206606164d2bSGeorge Zhang */ 206706164d2bSGeorge Zhang int vmci_qp_broker_set_page_store(struct vmci_handle handle, 206806164d2bSGeorge Zhang u64 produce_uva, 206906164d2bSGeorge Zhang u64 consume_uva, 207006164d2bSGeorge Zhang struct vmci_ctx *context) 207106164d2bSGeorge Zhang { 207206164d2bSGeorge Zhang struct qp_broker_entry *entry; 207306164d2bSGeorge Zhang int result; 207406164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 207506164d2bSGeorge Zhang 207606164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || !context || 207706164d2bSGeorge Zhang context_id == VMCI_INVALID_ID) 207806164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 207906164d2bSGeorge Zhang 208006164d2bSGeorge Zhang /* 208106164d2bSGeorge Zhang * We only support guest to host queue pairs, so the VMX must 208206164d2bSGeorge Zhang * supply UVAs for the mapped page files. 208306164d2bSGeorge Zhang */ 208406164d2bSGeorge Zhang 208506164d2bSGeorge Zhang if (produce_uva == 0 || consume_uva == 0) 208606164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 208706164d2bSGeorge Zhang 208806164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 208906164d2bSGeorge Zhang 209006164d2bSGeorge Zhang if (!vmci_ctx_qp_exists(context, handle)) { 209106164d2bSGeorge Zhang pr_warn("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n", 209206164d2bSGeorge Zhang context_id, handle.context, handle.resource); 209306164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 209406164d2bSGeorge Zhang goto out; 209506164d2bSGeorge Zhang } 209606164d2bSGeorge Zhang 209706164d2bSGeorge Zhang entry = qp_broker_handle_to_entry(handle); 209806164d2bSGeorge Zhang if (!entry) { 209906164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 210006164d2bSGeorge Zhang goto out; 210106164d2bSGeorge Zhang } 210206164d2bSGeorge Zhang 210306164d2bSGeorge Zhang /* 210406164d2bSGeorge Zhang * If I'm the owner then I can set the page store. 210506164d2bSGeorge Zhang * 210606164d2bSGeorge Zhang * Or, if a host created the queue_pair and I'm the attached peer 210706164d2bSGeorge Zhang * then I can set the page store. 210806164d2bSGeorge Zhang */ 210906164d2bSGeorge Zhang if (entry->create_id != context_id && 211006164d2bSGeorge Zhang (entry->create_id != VMCI_HOST_CONTEXT_ID || 211106164d2bSGeorge Zhang entry->attach_id != context_id)) { 211206164d2bSGeorge Zhang result = VMCI_ERROR_QUEUEPAIR_NOTOWNER; 211306164d2bSGeorge Zhang goto out; 211406164d2bSGeorge Zhang } 211506164d2bSGeorge Zhang 211606164d2bSGeorge Zhang if (entry->state != VMCIQPB_CREATED_NO_MEM && 211706164d2bSGeorge Zhang entry->state != VMCIQPB_ATTACHED_NO_MEM) { 211806164d2bSGeorge Zhang result = VMCI_ERROR_UNAVAILABLE; 211906164d2bSGeorge Zhang goto out; 212006164d2bSGeorge Zhang } 212106164d2bSGeorge Zhang 212206164d2bSGeorge Zhang result = qp_host_get_user_memory(produce_uva, consume_uva, 212306164d2bSGeorge Zhang entry->produce_q, entry->consume_q); 212406164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 212506164d2bSGeorge Zhang goto out; 212606164d2bSGeorge Zhang 212706164d2bSGeorge Zhang result = qp_host_map_queues(entry->produce_q, entry->consume_q); 212806164d2bSGeorge Zhang if (result < VMCI_SUCCESS) { 212906164d2bSGeorge Zhang qp_host_unregister_user_memory(entry->produce_q, 213006164d2bSGeorge Zhang entry->consume_q); 213106164d2bSGeorge Zhang goto out; 213206164d2bSGeorge Zhang } 213306164d2bSGeorge Zhang 213406164d2bSGeorge Zhang if (entry->state == VMCIQPB_CREATED_NO_MEM) 213506164d2bSGeorge Zhang entry->state = VMCIQPB_CREATED_MEM; 213606164d2bSGeorge Zhang else 213706164d2bSGeorge Zhang entry->state = VMCIQPB_ATTACHED_MEM; 213806164d2bSGeorge Zhang 213906164d2bSGeorge Zhang entry->vmci_page_files = true; 214006164d2bSGeorge Zhang 214106164d2bSGeorge Zhang if (entry->state == VMCIQPB_ATTACHED_MEM) { 214206164d2bSGeorge Zhang result = 214306164d2bSGeorge Zhang qp_notify_peer(true, handle, context_id, entry->create_id); 214406164d2bSGeorge Zhang if (result < VMCI_SUCCESS) { 214506164d2bSGeorge Zhang pr_warn("Failed to notify peer (ID=0x%x) of attach to queue pair (handle=0x%x:0x%x)\n", 214606164d2bSGeorge Zhang entry->create_id, entry->qp.handle.context, 214706164d2bSGeorge Zhang entry->qp.handle.resource); 214806164d2bSGeorge Zhang } 214906164d2bSGeorge Zhang } 215006164d2bSGeorge Zhang 215106164d2bSGeorge Zhang result = VMCI_SUCCESS; 215206164d2bSGeorge Zhang out: 215306164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 215406164d2bSGeorge Zhang return result; 215506164d2bSGeorge Zhang } 215606164d2bSGeorge Zhang 215706164d2bSGeorge Zhang /* 215806164d2bSGeorge Zhang * Resets saved queue headers for the given QP broker 215906164d2bSGeorge Zhang * entry. Should be used when guest memory becomes available 216006164d2bSGeorge Zhang * again, or the guest detaches. 216106164d2bSGeorge Zhang */ 216206164d2bSGeorge Zhang static void qp_reset_saved_headers(struct qp_broker_entry *entry) 216306164d2bSGeorge Zhang { 216406164d2bSGeorge Zhang entry->produce_q->saved_header = NULL; 216506164d2bSGeorge Zhang entry->consume_q->saved_header = NULL; 216606164d2bSGeorge Zhang } 216706164d2bSGeorge Zhang 216806164d2bSGeorge Zhang /* 216906164d2bSGeorge Zhang * The main entry point for detaching from a queue pair registered with the 217006164d2bSGeorge Zhang * queue pair broker. If more than one endpoint is attached to the queue 217106164d2bSGeorge Zhang * pair, the first endpoint will mainly decrement a reference count and 217206164d2bSGeorge Zhang * generate a notification to its peer. The last endpoint will clean up 217306164d2bSGeorge Zhang * the queue pair state registered with the broker. 217406164d2bSGeorge Zhang * 217506164d2bSGeorge Zhang * When a guest endpoint detaches, it will unmap and unregister the guest 217606164d2bSGeorge Zhang * memory backing the queue pair. If the host is still attached, it will 217706164d2bSGeorge Zhang * no longer be able to access the queue pair content. 217806164d2bSGeorge Zhang * 217906164d2bSGeorge Zhang * If the queue pair is already in a state where there is no memory 218006164d2bSGeorge Zhang * registered for the queue pair (any *_NO_MEM state), it will transition to 218106164d2bSGeorge Zhang * the VMCIQPB_SHUTDOWN_NO_MEM state. This will also happen, if a guest 218206164d2bSGeorge Zhang * endpoint is the first of two endpoints to detach. If the host endpoint is 218306164d2bSGeorge Zhang * the first out of two to detach, the queue pair will move to the 218406164d2bSGeorge Zhang * VMCIQPB_SHUTDOWN_MEM state. 218506164d2bSGeorge Zhang */ 218606164d2bSGeorge Zhang int vmci_qp_broker_detach(struct vmci_handle handle, struct vmci_ctx *context) 218706164d2bSGeorge Zhang { 218806164d2bSGeorge Zhang struct qp_broker_entry *entry; 218906164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 219006164d2bSGeorge Zhang u32 peer_id; 219106164d2bSGeorge Zhang bool is_local = false; 219206164d2bSGeorge Zhang int result; 219306164d2bSGeorge Zhang 219406164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || !context || 219506164d2bSGeorge Zhang context_id == VMCI_INVALID_ID) { 219606164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 219706164d2bSGeorge Zhang } 219806164d2bSGeorge Zhang 219906164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 220006164d2bSGeorge Zhang 220106164d2bSGeorge Zhang if (!vmci_ctx_qp_exists(context, handle)) { 220206164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n", 220306164d2bSGeorge Zhang context_id, handle.context, handle.resource); 220406164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 220506164d2bSGeorge Zhang goto out; 220606164d2bSGeorge Zhang } 220706164d2bSGeorge Zhang 220806164d2bSGeorge Zhang entry = qp_broker_handle_to_entry(handle); 220906164d2bSGeorge Zhang if (!entry) { 221006164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) reports being attached to queue pair(handle=0x%x:0x%x) that isn't present in broker\n", 221106164d2bSGeorge Zhang context_id, handle.context, handle.resource); 221206164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 221306164d2bSGeorge Zhang goto out; 221406164d2bSGeorge Zhang } 221506164d2bSGeorge Zhang 221606164d2bSGeorge Zhang if (context_id != entry->create_id && context_id != entry->attach_id) { 221706164d2bSGeorge Zhang result = VMCI_ERROR_QUEUEPAIR_NOTATTACHED; 221806164d2bSGeorge Zhang goto out; 221906164d2bSGeorge Zhang } 222006164d2bSGeorge Zhang 222106164d2bSGeorge Zhang if (context_id == entry->create_id) { 222206164d2bSGeorge Zhang peer_id = entry->attach_id; 222306164d2bSGeorge Zhang entry->create_id = VMCI_INVALID_ID; 222406164d2bSGeorge Zhang } else { 222506164d2bSGeorge Zhang peer_id = entry->create_id; 222606164d2bSGeorge Zhang entry->attach_id = VMCI_INVALID_ID; 222706164d2bSGeorge Zhang } 222806164d2bSGeorge Zhang entry->qp.ref_count--; 222906164d2bSGeorge Zhang 223006164d2bSGeorge Zhang is_local = entry->qp.flags & VMCI_QPFLAG_LOCAL; 223106164d2bSGeorge Zhang 223206164d2bSGeorge Zhang if (context_id != VMCI_HOST_CONTEXT_ID) { 223306164d2bSGeorge Zhang bool headers_mapped; 223406164d2bSGeorge Zhang 223506164d2bSGeorge Zhang /* 223606164d2bSGeorge Zhang * Pre NOVMVM vmx'en may detach from a queue pair 223706164d2bSGeorge Zhang * before setting the page store, and in that case 223806164d2bSGeorge Zhang * there is no user memory to detach from. Also, more 223906164d2bSGeorge Zhang * recent VMX'en may detach from a queue pair in the 224006164d2bSGeorge Zhang * quiesced state. 224106164d2bSGeorge Zhang */ 224206164d2bSGeorge Zhang 224306164d2bSGeorge Zhang qp_acquire_queue_mutex(entry->produce_q); 224406164d2bSGeorge Zhang headers_mapped = entry->produce_q->q_header || 224506164d2bSGeorge Zhang entry->consume_q->q_header; 224606164d2bSGeorge Zhang if (QPBROKERSTATE_HAS_MEM(entry)) { 224706164d2bSGeorge Zhang result = 224806164d2bSGeorge Zhang qp_host_unmap_queues(INVALID_VMCI_GUEST_MEM_ID, 224906164d2bSGeorge Zhang entry->produce_q, 225006164d2bSGeorge Zhang entry->consume_q); 225106164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 225206164d2bSGeorge Zhang pr_warn("Failed to unmap queue headers for queue pair (handle=0x%x:0x%x,result=%d)\n", 225306164d2bSGeorge Zhang handle.context, handle.resource, 225406164d2bSGeorge Zhang result); 225506164d2bSGeorge Zhang 225606164d2bSGeorge Zhang if (entry->vmci_page_files) 225706164d2bSGeorge Zhang qp_host_unregister_user_memory(entry->produce_q, 225806164d2bSGeorge Zhang entry-> 225906164d2bSGeorge Zhang consume_q); 226006164d2bSGeorge Zhang else 226106164d2bSGeorge Zhang qp_host_unregister_user_memory(entry->produce_q, 226206164d2bSGeorge Zhang entry-> 226306164d2bSGeorge Zhang consume_q); 226406164d2bSGeorge Zhang 226506164d2bSGeorge Zhang } 226606164d2bSGeorge Zhang 226706164d2bSGeorge Zhang if (!headers_mapped) 226806164d2bSGeorge Zhang qp_reset_saved_headers(entry); 226906164d2bSGeorge Zhang 227006164d2bSGeorge Zhang qp_release_queue_mutex(entry->produce_q); 227106164d2bSGeorge Zhang 227206164d2bSGeorge Zhang if (!headers_mapped && entry->wakeup_cb) 227306164d2bSGeorge Zhang entry->wakeup_cb(entry->client_data); 227406164d2bSGeorge Zhang 227506164d2bSGeorge Zhang } else { 227606164d2bSGeorge Zhang if (entry->wakeup_cb) { 227706164d2bSGeorge Zhang entry->wakeup_cb = NULL; 227806164d2bSGeorge Zhang entry->client_data = NULL; 227906164d2bSGeorge Zhang } 228006164d2bSGeorge Zhang } 228106164d2bSGeorge Zhang 228206164d2bSGeorge Zhang if (entry->qp.ref_count == 0) { 228306164d2bSGeorge Zhang qp_list_remove_entry(&qp_broker_list, &entry->qp); 228406164d2bSGeorge Zhang 228506164d2bSGeorge Zhang if (is_local) 228606164d2bSGeorge Zhang kfree(entry->local_mem); 228706164d2bSGeorge Zhang 228806164d2bSGeorge Zhang qp_cleanup_queue_mutex(entry->produce_q, entry->consume_q); 228906164d2bSGeorge Zhang qp_host_free_queue(entry->produce_q, entry->qp.produce_size); 229006164d2bSGeorge Zhang qp_host_free_queue(entry->consume_q, entry->qp.consume_size); 229106164d2bSGeorge Zhang /* Unlink from resource hash table and free callback */ 229206164d2bSGeorge Zhang vmci_resource_remove(&entry->resource); 229306164d2bSGeorge Zhang 229406164d2bSGeorge Zhang kfree(entry); 229506164d2bSGeorge Zhang 229606164d2bSGeorge Zhang vmci_ctx_qp_destroy(context, handle); 229706164d2bSGeorge Zhang } else { 229806164d2bSGeorge Zhang qp_notify_peer(false, handle, context_id, peer_id); 229906164d2bSGeorge Zhang if (context_id == VMCI_HOST_CONTEXT_ID && 230006164d2bSGeorge Zhang QPBROKERSTATE_HAS_MEM(entry)) { 230106164d2bSGeorge Zhang entry->state = VMCIQPB_SHUTDOWN_MEM; 230206164d2bSGeorge Zhang } else { 230306164d2bSGeorge Zhang entry->state = VMCIQPB_SHUTDOWN_NO_MEM; 230406164d2bSGeorge Zhang } 230506164d2bSGeorge Zhang 230606164d2bSGeorge Zhang if (!is_local) 230706164d2bSGeorge Zhang vmci_ctx_qp_destroy(context, handle); 230806164d2bSGeorge Zhang 230906164d2bSGeorge Zhang } 231006164d2bSGeorge Zhang result = VMCI_SUCCESS; 231106164d2bSGeorge Zhang out: 231206164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 231306164d2bSGeorge Zhang return result; 231406164d2bSGeorge Zhang } 231506164d2bSGeorge Zhang 231606164d2bSGeorge Zhang /* 231706164d2bSGeorge Zhang * Establishes the necessary mappings for a queue pair given a 231806164d2bSGeorge Zhang * reference to the queue pair guest memory. This is usually 231906164d2bSGeorge Zhang * called when a guest is unquiesced and the VMX is allowed to 232006164d2bSGeorge Zhang * map guest memory once again. 232106164d2bSGeorge Zhang */ 232206164d2bSGeorge Zhang int vmci_qp_broker_map(struct vmci_handle handle, 232306164d2bSGeorge Zhang struct vmci_ctx *context, 232406164d2bSGeorge Zhang u64 guest_mem) 232506164d2bSGeorge Zhang { 232606164d2bSGeorge Zhang struct qp_broker_entry *entry; 232706164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 232806164d2bSGeorge Zhang bool is_local = false; 232906164d2bSGeorge Zhang int result; 233006164d2bSGeorge Zhang 233106164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || !context || 233206164d2bSGeorge Zhang context_id == VMCI_INVALID_ID) 233306164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 233406164d2bSGeorge Zhang 233506164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 233606164d2bSGeorge Zhang 233706164d2bSGeorge Zhang if (!vmci_ctx_qp_exists(context, handle)) { 233806164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n", 233906164d2bSGeorge Zhang context_id, handle.context, handle.resource); 234006164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 234106164d2bSGeorge Zhang goto out; 234206164d2bSGeorge Zhang } 234306164d2bSGeorge Zhang 234406164d2bSGeorge Zhang entry = qp_broker_handle_to_entry(handle); 234506164d2bSGeorge Zhang if (!entry) { 234606164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) reports being attached to queue pair (handle=0x%x:0x%x) that isn't present in broker\n", 234706164d2bSGeorge Zhang context_id, handle.context, handle.resource); 234806164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 234906164d2bSGeorge Zhang goto out; 235006164d2bSGeorge Zhang } 235106164d2bSGeorge Zhang 235206164d2bSGeorge Zhang if (context_id != entry->create_id && context_id != entry->attach_id) { 235306164d2bSGeorge Zhang result = VMCI_ERROR_QUEUEPAIR_NOTATTACHED; 235406164d2bSGeorge Zhang goto out; 235506164d2bSGeorge Zhang } 235606164d2bSGeorge Zhang 235706164d2bSGeorge Zhang is_local = entry->qp.flags & VMCI_QPFLAG_LOCAL; 235806164d2bSGeorge Zhang result = VMCI_SUCCESS; 235906164d2bSGeorge Zhang 236006164d2bSGeorge Zhang if (context_id != VMCI_HOST_CONTEXT_ID) { 236106164d2bSGeorge Zhang struct vmci_qp_page_store page_store; 236206164d2bSGeorge Zhang 236306164d2bSGeorge Zhang page_store.pages = guest_mem; 236406164d2bSGeorge Zhang page_store.len = QPE_NUM_PAGES(entry->qp); 236506164d2bSGeorge Zhang 236606164d2bSGeorge Zhang qp_acquire_queue_mutex(entry->produce_q); 236706164d2bSGeorge Zhang qp_reset_saved_headers(entry); 236806164d2bSGeorge Zhang result = 236906164d2bSGeorge Zhang qp_host_register_user_memory(&page_store, 237006164d2bSGeorge Zhang entry->produce_q, 237106164d2bSGeorge Zhang entry->consume_q); 237206164d2bSGeorge Zhang qp_release_queue_mutex(entry->produce_q); 237306164d2bSGeorge Zhang if (result == VMCI_SUCCESS) { 237406164d2bSGeorge Zhang /* Move state from *_NO_MEM to *_MEM */ 237506164d2bSGeorge Zhang 237606164d2bSGeorge Zhang entry->state++; 237706164d2bSGeorge Zhang 237806164d2bSGeorge Zhang if (entry->wakeup_cb) 237906164d2bSGeorge Zhang entry->wakeup_cb(entry->client_data); 238006164d2bSGeorge Zhang } 238106164d2bSGeorge Zhang } 238206164d2bSGeorge Zhang 238306164d2bSGeorge Zhang out: 238406164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 238506164d2bSGeorge Zhang return result; 238606164d2bSGeorge Zhang } 238706164d2bSGeorge Zhang 238806164d2bSGeorge Zhang /* 238906164d2bSGeorge Zhang * Saves a snapshot of the queue headers for the given QP broker 239006164d2bSGeorge Zhang * entry. Should be used when guest memory is unmapped. 239106164d2bSGeorge Zhang * Results: 239206164d2bSGeorge Zhang * VMCI_SUCCESS on success, appropriate error code if guest memory 239306164d2bSGeorge Zhang * can't be accessed.. 239406164d2bSGeorge Zhang */ 239506164d2bSGeorge Zhang static int qp_save_headers(struct qp_broker_entry *entry) 239606164d2bSGeorge Zhang { 239706164d2bSGeorge Zhang int result; 239806164d2bSGeorge Zhang 239906164d2bSGeorge Zhang if (entry->produce_q->saved_header != NULL && 240006164d2bSGeorge Zhang entry->consume_q->saved_header != NULL) { 240106164d2bSGeorge Zhang /* 240206164d2bSGeorge Zhang * If the headers have already been saved, we don't need to do 240306164d2bSGeorge Zhang * it again, and we don't want to map in the headers 240406164d2bSGeorge Zhang * unnecessarily. 240506164d2bSGeorge Zhang */ 240606164d2bSGeorge Zhang 240706164d2bSGeorge Zhang return VMCI_SUCCESS; 240806164d2bSGeorge Zhang } 240906164d2bSGeorge Zhang 241006164d2bSGeorge Zhang if (NULL == entry->produce_q->q_header || 241106164d2bSGeorge Zhang NULL == entry->consume_q->q_header) { 241206164d2bSGeorge Zhang result = qp_host_map_queues(entry->produce_q, entry->consume_q); 241306164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 241406164d2bSGeorge Zhang return result; 241506164d2bSGeorge Zhang } 241606164d2bSGeorge Zhang 241706164d2bSGeorge Zhang memcpy(&entry->saved_produce_q, entry->produce_q->q_header, 241806164d2bSGeorge Zhang sizeof(entry->saved_produce_q)); 241906164d2bSGeorge Zhang entry->produce_q->saved_header = &entry->saved_produce_q; 242006164d2bSGeorge Zhang memcpy(&entry->saved_consume_q, entry->consume_q->q_header, 242106164d2bSGeorge Zhang sizeof(entry->saved_consume_q)); 242206164d2bSGeorge Zhang entry->consume_q->saved_header = &entry->saved_consume_q; 242306164d2bSGeorge Zhang 242406164d2bSGeorge Zhang return VMCI_SUCCESS; 242506164d2bSGeorge Zhang } 242606164d2bSGeorge Zhang 242706164d2bSGeorge Zhang /* 242806164d2bSGeorge Zhang * Removes all references to the guest memory of a given queue pair, and 242906164d2bSGeorge Zhang * will move the queue pair from state *_MEM to *_NO_MEM. It is usually 243006164d2bSGeorge Zhang * called when a VM is being quiesced where access to guest memory should 243106164d2bSGeorge Zhang * avoided. 243206164d2bSGeorge Zhang */ 243306164d2bSGeorge Zhang int vmci_qp_broker_unmap(struct vmci_handle handle, 243406164d2bSGeorge Zhang struct vmci_ctx *context, 243506164d2bSGeorge Zhang u32 gid) 243606164d2bSGeorge Zhang { 243706164d2bSGeorge Zhang struct qp_broker_entry *entry; 243806164d2bSGeorge Zhang const u32 context_id = vmci_ctx_get_id(context); 243906164d2bSGeorge Zhang bool is_local = false; 244006164d2bSGeorge Zhang int result; 244106164d2bSGeorge Zhang 244206164d2bSGeorge Zhang if (vmci_handle_is_invalid(handle) || !context || 244306164d2bSGeorge Zhang context_id == VMCI_INVALID_ID) 244406164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 244506164d2bSGeorge Zhang 244606164d2bSGeorge Zhang mutex_lock(&qp_broker_list.mutex); 244706164d2bSGeorge Zhang 244806164d2bSGeorge Zhang if (!vmci_ctx_qp_exists(context, handle)) { 244906164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) not attached to queue pair (handle=0x%x:0x%x)\n", 245006164d2bSGeorge Zhang context_id, handle.context, handle.resource); 245106164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 245206164d2bSGeorge Zhang goto out; 245306164d2bSGeorge Zhang } 245406164d2bSGeorge Zhang 245506164d2bSGeorge Zhang entry = qp_broker_handle_to_entry(handle); 245606164d2bSGeorge Zhang if (!entry) { 245706164d2bSGeorge Zhang pr_devel("Context (ID=0x%x) reports being attached to queue pair (handle=0x%x:0x%x) that isn't present in broker\n", 245806164d2bSGeorge Zhang context_id, handle.context, handle.resource); 245906164d2bSGeorge Zhang result = VMCI_ERROR_NOT_FOUND; 246006164d2bSGeorge Zhang goto out; 246106164d2bSGeorge Zhang } 246206164d2bSGeorge Zhang 246306164d2bSGeorge Zhang if (context_id != entry->create_id && context_id != entry->attach_id) { 246406164d2bSGeorge Zhang result = VMCI_ERROR_QUEUEPAIR_NOTATTACHED; 246506164d2bSGeorge Zhang goto out; 246606164d2bSGeorge Zhang } 246706164d2bSGeorge Zhang 246806164d2bSGeorge Zhang is_local = entry->qp.flags & VMCI_QPFLAG_LOCAL; 246906164d2bSGeorge Zhang 247006164d2bSGeorge Zhang if (context_id != VMCI_HOST_CONTEXT_ID) { 247106164d2bSGeorge Zhang qp_acquire_queue_mutex(entry->produce_q); 247206164d2bSGeorge Zhang result = qp_save_headers(entry); 247306164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 247406164d2bSGeorge Zhang pr_warn("Failed to save queue headers for queue pair (handle=0x%x:0x%x,result=%d)\n", 247506164d2bSGeorge Zhang handle.context, handle.resource, result); 247606164d2bSGeorge Zhang 247706164d2bSGeorge Zhang qp_host_unmap_queues(gid, entry->produce_q, entry->consume_q); 247806164d2bSGeorge Zhang 247906164d2bSGeorge Zhang /* 248006164d2bSGeorge Zhang * On hosted, when we unmap queue pairs, the VMX will also 248106164d2bSGeorge Zhang * unmap the guest memory, so we invalidate the previously 248206164d2bSGeorge Zhang * registered memory. If the queue pair is mapped again at a 248306164d2bSGeorge Zhang * later point in time, we will need to reregister the user 248406164d2bSGeorge Zhang * memory with a possibly new user VA. 248506164d2bSGeorge Zhang */ 248606164d2bSGeorge Zhang qp_host_unregister_user_memory(entry->produce_q, 248706164d2bSGeorge Zhang entry->consume_q); 248806164d2bSGeorge Zhang 248906164d2bSGeorge Zhang /* 249006164d2bSGeorge Zhang * Move state from *_MEM to *_NO_MEM. 249106164d2bSGeorge Zhang */ 249206164d2bSGeorge Zhang entry->state--; 249306164d2bSGeorge Zhang 249406164d2bSGeorge Zhang qp_release_queue_mutex(entry->produce_q); 249506164d2bSGeorge Zhang } 249606164d2bSGeorge Zhang 249706164d2bSGeorge Zhang result = VMCI_SUCCESS; 249806164d2bSGeorge Zhang 249906164d2bSGeorge Zhang out: 250006164d2bSGeorge Zhang mutex_unlock(&qp_broker_list.mutex); 250106164d2bSGeorge Zhang return result; 250206164d2bSGeorge Zhang } 250306164d2bSGeorge Zhang 250406164d2bSGeorge Zhang /* 250506164d2bSGeorge Zhang * Destroys all guest queue pair endpoints. If active guest queue 250606164d2bSGeorge Zhang * pairs still exist, hypercalls to attempt detach from these 250706164d2bSGeorge Zhang * queue pairs will be made. Any failure to detach is silently 250806164d2bSGeorge Zhang * ignored. 250906164d2bSGeorge Zhang */ 251006164d2bSGeorge Zhang void vmci_qp_guest_endpoints_exit(void) 251106164d2bSGeorge Zhang { 251206164d2bSGeorge Zhang struct qp_entry *entry; 251306164d2bSGeorge Zhang struct qp_guest_endpoint *ep; 251406164d2bSGeorge Zhang 251506164d2bSGeorge Zhang mutex_lock(&qp_guest_endpoints.mutex); 251606164d2bSGeorge Zhang 251706164d2bSGeorge Zhang while ((entry = qp_list_get_head(&qp_guest_endpoints))) { 251806164d2bSGeorge Zhang ep = (struct qp_guest_endpoint *)entry; 251906164d2bSGeorge Zhang 252006164d2bSGeorge Zhang /* Don't make a hypercall for local queue_pairs. */ 252106164d2bSGeorge Zhang if (!(entry->flags & VMCI_QPFLAG_LOCAL)) 252206164d2bSGeorge Zhang qp_detatch_hypercall(entry->handle); 252306164d2bSGeorge Zhang 252406164d2bSGeorge Zhang /* We cannot fail the exit, so let's reset ref_count. */ 252506164d2bSGeorge Zhang entry->ref_count = 0; 252606164d2bSGeorge Zhang qp_list_remove_entry(&qp_guest_endpoints, entry); 252706164d2bSGeorge Zhang 252806164d2bSGeorge Zhang qp_guest_endpoint_destroy(ep); 252906164d2bSGeorge Zhang } 253006164d2bSGeorge Zhang 253106164d2bSGeorge Zhang mutex_unlock(&qp_guest_endpoints.mutex); 253206164d2bSGeorge Zhang } 253306164d2bSGeorge Zhang 253406164d2bSGeorge Zhang /* 253506164d2bSGeorge Zhang * Helper routine that will lock the queue pair before subsequent 253606164d2bSGeorge Zhang * operations. 253706164d2bSGeorge Zhang * Note: Non-blocking on the host side is currently only implemented in ESX. 253806164d2bSGeorge Zhang * Since non-blocking isn't yet implemented on the host personality we 253906164d2bSGeorge Zhang * have no reason to acquire a spin lock. So to avoid the use of an 254006164d2bSGeorge Zhang * unnecessary lock only acquire the mutex if we can block. 254106164d2bSGeorge Zhang * Note: It is assumed that QPFLAG_PINNED implies QPFLAG_NONBLOCK. Therefore 254206164d2bSGeorge Zhang * we can use the same locking function for access to both the queue 254306164d2bSGeorge Zhang * and the queue headers as it is the same logic. Assert this behvior. 254406164d2bSGeorge Zhang */ 254506164d2bSGeorge Zhang static void qp_lock(const struct vmci_qp *qpair) 254606164d2bSGeorge Zhang { 254706164d2bSGeorge Zhang if (vmci_can_block(qpair->flags)) 254806164d2bSGeorge Zhang qp_acquire_queue_mutex(qpair->produce_q); 254906164d2bSGeorge Zhang } 255006164d2bSGeorge Zhang 255106164d2bSGeorge Zhang /* 255206164d2bSGeorge Zhang * Helper routine that unlocks the queue pair after calling 255306164d2bSGeorge Zhang * qp_lock. Respects non-blocking and pinning flags. 255406164d2bSGeorge Zhang */ 255506164d2bSGeorge Zhang static void qp_unlock(const struct vmci_qp *qpair) 255606164d2bSGeorge Zhang { 255706164d2bSGeorge Zhang if (vmci_can_block(qpair->flags)) 255806164d2bSGeorge Zhang qp_release_queue_mutex(qpair->produce_q); 255906164d2bSGeorge Zhang } 256006164d2bSGeorge Zhang 256106164d2bSGeorge Zhang /* 256206164d2bSGeorge Zhang * The queue headers may not be mapped at all times. If a queue is 256306164d2bSGeorge Zhang * currently not mapped, it will be attempted to do so. 256406164d2bSGeorge Zhang */ 256506164d2bSGeorge Zhang static int qp_map_queue_headers(struct vmci_queue *produce_q, 256606164d2bSGeorge Zhang struct vmci_queue *consume_q, 256706164d2bSGeorge Zhang bool can_block) 256806164d2bSGeorge Zhang { 256906164d2bSGeorge Zhang int result; 257006164d2bSGeorge Zhang 257106164d2bSGeorge Zhang if (NULL == produce_q->q_header || NULL == consume_q->q_header) { 257206164d2bSGeorge Zhang if (can_block) 257306164d2bSGeorge Zhang result = qp_host_map_queues(produce_q, consume_q); 257406164d2bSGeorge Zhang else 257506164d2bSGeorge Zhang result = VMCI_ERROR_QUEUEPAIR_NOT_READY; 257606164d2bSGeorge Zhang 257706164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 257806164d2bSGeorge Zhang return (produce_q->saved_header && 257906164d2bSGeorge Zhang consume_q->saved_header) ? 258006164d2bSGeorge Zhang VMCI_ERROR_QUEUEPAIR_NOT_READY : 258106164d2bSGeorge Zhang VMCI_ERROR_QUEUEPAIR_NOTATTACHED; 258206164d2bSGeorge Zhang } 258306164d2bSGeorge Zhang 258406164d2bSGeorge Zhang return VMCI_SUCCESS; 258506164d2bSGeorge Zhang } 258606164d2bSGeorge Zhang 258706164d2bSGeorge Zhang /* 258806164d2bSGeorge Zhang * Helper routine that will retrieve the produce and consume 258906164d2bSGeorge Zhang * headers of a given queue pair. If the guest memory of the 259006164d2bSGeorge Zhang * queue pair is currently not available, the saved queue headers 259106164d2bSGeorge Zhang * will be returned, if these are available. 259206164d2bSGeorge Zhang */ 259306164d2bSGeorge Zhang static int qp_get_queue_headers(const struct vmci_qp *qpair, 259406164d2bSGeorge Zhang struct vmci_queue_header **produce_q_header, 259506164d2bSGeorge Zhang struct vmci_queue_header **consume_q_header) 259606164d2bSGeorge Zhang { 259706164d2bSGeorge Zhang int result; 259806164d2bSGeorge Zhang 259906164d2bSGeorge Zhang result = qp_map_queue_headers(qpair->produce_q, qpair->consume_q, 260006164d2bSGeorge Zhang vmci_can_block(qpair->flags)); 260106164d2bSGeorge Zhang if (result == VMCI_SUCCESS) { 260206164d2bSGeorge Zhang *produce_q_header = qpair->produce_q->q_header; 260306164d2bSGeorge Zhang *consume_q_header = qpair->consume_q->q_header; 260406164d2bSGeorge Zhang } else if (qpair->produce_q->saved_header && 260506164d2bSGeorge Zhang qpair->consume_q->saved_header) { 260606164d2bSGeorge Zhang *produce_q_header = qpair->produce_q->saved_header; 260706164d2bSGeorge Zhang *consume_q_header = qpair->consume_q->saved_header; 260806164d2bSGeorge Zhang result = VMCI_SUCCESS; 260906164d2bSGeorge Zhang } 261006164d2bSGeorge Zhang 261106164d2bSGeorge Zhang return result; 261206164d2bSGeorge Zhang } 261306164d2bSGeorge Zhang 261406164d2bSGeorge Zhang /* 261506164d2bSGeorge Zhang * Callback from VMCI queue pair broker indicating that a queue 261606164d2bSGeorge Zhang * pair that was previously not ready, now either is ready or 261706164d2bSGeorge Zhang * gone forever. 261806164d2bSGeorge Zhang */ 261906164d2bSGeorge Zhang static int qp_wakeup_cb(void *client_data) 262006164d2bSGeorge Zhang { 262106164d2bSGeorge Zhang struct vmci_qp *qpair = (struct vmci_qp *)client_data; 262206164d2bSGeorge Zhang 262306164d2bSGeorge Zhang qp_lock(qpair); 262406164d2bSGeorge Zhang while (qpair->blocked > 0) { 262506164d2bSGeorge Zhang qpair->blocked--; 262606164d2bSGeorge Zhang qpair->generation++; 262706164d2bSGeorge Zhang wake_up(&qpair->event); 262806164d2bSGeorge Zhang } 262906164d2bSGeorge Zhang qp_unlock(qpair); 263006164d2bSGeorge Zhang 263106164d2bSGeorge Zhang return VMCI_SUCCESS; 263206164d2bSGeorge Zhang } 263306164d2bSGeorge Zhang 263406164d2bSGeorge Zhang /* 263506164d2bSGeorge Zhang * Makes the calling thread wait for the queue pair to become 263606164d2bSGeorge Zhang * ready for host side access. Returns true when thread is 263706164d2bSGeorge Zhang * woken up after queue pair state change, false otherwise. 263806164d2bSGeorge Zhang */ 263906164d2bSGeorge Zhang static bool qp_wait_for_ready_queue(struct vmci_qp *qpair) 264006164d2bSGeorge Zhang { 264106164d2bSGeorge Zhang unsigned int generation; 264206164d2bSGeorge Zhang 264306164d2bSGeorge Zhang if (qpair->flags & VMCI_QPFLAG_NONBLOCK) 264406164d2bSGeorge Zhang return false; 264506164d2bSGeorge Zhang 264606164d2bSGeorge Zhang qpair->blocked++; 264706164d2bSGeorge Zhang generation = qpair->generation; 264806164d2bSGeorge Zhang qp_unlock(qpair); 264906164d2bSGeorge Zhang wait_event(qpair->event, generation != qpair->generation); 265006164d2bSGeorge Zhang qp_lock(qpair); 265106164d2bSGeorge Zhang 265206164d2bSGeorge Zhang return true; 265306164d2bSGeorge Zhang } 265406164d2bSGeorge Zhang 265506164d2bSGeorge Zhang /* 265606164d2bSGeorge Zhang * Enqueues a given buffer to the produce queue using the provided 265706164d2bSGeorge Zhang * function. As many bytes as possible (space available in the queue) 265806164d2bSGeorge Zhang * are enqueued. Assumes the queue->mutex has been acquired. Returns 265906164d2bSGeorge Zhang * VMCI_ERROR_QUEUEPAIR_NOSPACE if no space was available to enqueue 266006164d2bSGeorge Zhang * data, VMCI_ERROR_INVALID_SIZE, if any queue pointer is outside the 266106164d2bSGeorge Zhang * queue (as defined by the queue size), VMCI_ERROR_INVALID_ARGS, if 266206164d2bSGeorge Zhang * an error occured when accessing the buffer, 266306164d2bSGeorge Zhang * VMCI_ERROR_QUEUEPAIR_NOTATTACHED, if the queue pair pages aren't 266406164d2bSGeorge Zhang * available. Otherwise, the number of bytes written to the queue is 266506164d2bSGeorge Zhang * returned. Updates the tail pointer of the produce queue. 266606164d2bSGeorge Zhang */ 266706164d2bSGeorge Zhang static ssize_t qp_enqueue_locked(struct vmci_queue *produce_q, 266806164d2bSGeorge Zhang struct vmci_queue *consume_q, 266906164d2bSGeorge Zhang const u64 produce_q_size, 267006164d2bSGeorge Zhang const void *buf, 267106164d2bSGeorge Zhang size_t buf_size, 267206164d2bSGeorge Zhang vmci_memcpy_to_queue_func memcpy_to_queue, 267306164d2bSGeorge Zhang bool can_block) 267406164d2bSGeorge Zhang { 267506164d2bSGeorge Zhang s64 free_space; 267606164d2bSGeorge Zhang u64 tail; 267706164d2bSGeorge Zhang size_t written; 267806164d2bSGeorge Zhang ssize_t result; 267906164d2bSGeorge Zhang 268006164d2bSGeorge Zhang result = qp_map_queue_headers(produce_q, consume_q, can_block); 268106164d2bSGeorge Zhang if (unlikely(result != VMCI_SUCCESS)) 268206164d2bSGeorge Zhang return result; 268306164d2bSGeorge Zhang 268406164d2bSGeorge Zhang free_space = vmci_q_header_free_space(produce_q->q_header, 268506164d2bSGeorge Zhang consume_q->q_header, 268606164d2bSGeorge Zhang produce_q_size); 268706164d2bSGeorge Zhang if (free_space == 0) 268806164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_NOSPACE; 268906164d2bSGeorge Zhang 269006164d2bSGeorge Zhang if (free_space < VMCI_SUCCESS) 269106164d2bSGeorge Zhang return (ssize_t) free_space; 269206164d2bSGeorge Zhang 269306164d2bSGeorge Zhang written = (size_t) (free_space > buf_size ? buf_size : free_space); 269406164d2bSGeorge Zhang tail = vmci_q_header_producer_tail(produce_q->q_header); 269506164d2bSGeorge Zhang if (likely(tail + written < produce_q_size)) { 269606164d2bSGeorge Zhang result = memcpy_to_queue(produce_q, tail, buf, 0, written); 269706164d2bSGeorge Zhang } else { 269806164d2bSGeorge Zhang /* Tail pointer wraps around. */ 269906164d2bSGeorge Zhang 270006164d2bSGeorge Zhang const size_t tmp = (size_t) (produce_q_size - tail); 270106164d2bSGeorge Zhang 270206164d2bSGeorge Zhang result = memcpy_to_queue(produce_q, tail, buf, 0, tmp); 270306164d2bSGeorge Zhang if (result >= VMCI_SUCCESS) 270406164d2bSGeorge Zhang result = memcpy_to_queue(produce_q, 0, buf, tmp, 270506164d2bSGeorge Zhang written - tmp); 270606164d2bSGeorge Zhang } 270706164d2bSGeorge Zhang 270806164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 270906164d2bSGeorge Zhang return result; 271006164d2bSGeorge Zhang 271106164d2bSGeorge Zhang vmci_q_header_add_producer_tail(produce_q->q_header, written, 271206164d2bSGeorge Zhang produce_q_size); 271306164d2bSGeorge Zhang return written; 271406164d2bSGeorge Zhang } 271506164d2bSGeorge Zhang 271606164d2bSGeorge Zhang /* 271706164d2bSGeorge Zhang * Dequeues data (if available) from the given consume queue. Writes data 271806164d2bSGeorge Zhang * to the user provided buffer using the provided function. 271906164d2bSGeorge Zhang * Assumes the queue->mutex has been acquired. 272006164d2bSGeorge Zhang * Results: 272106164d2bSGeorge Zhang * VMCI_ERROR_QUEUEPAIR_NODATA if no data was available to dequeue. 272206164d2bSGeorge Zhang * VMCI_ERROR_INVALID_SIZE, if any queue pointer is outside the queue 272306164d2bSGeorge Zhang * (as defined by the queue size). 272406164d2bSGeorge Zhang * VMCI_ERROR_INVALID_ARGS, if an error occured when accessing the buffer. 272506164d2bSGeorge Zhang * Otherwise the number of bytes dequeued is returned. 272606164d2bSGeorge Zhang * Side effects: 272706164d2bSGeorge Zhang * Updates the head pointer of the consume queue. 272806164d2bSGeorge Zhang */ 272906164d2bSGeorge Zhang static ssize_t qp_dequeue_locked(struct vmci_queue *produce_q, 273006164d2bSGeorge Zhang struct vmci_queue *consume_q, 273106164d2bSGeorge Zhang const u64 consume_q_size, 273206164d2bSGeorge Zhang void *buf, 273306164d2bSGeorge Zhang size_t buf_size, 273406164d2bSGeorge Zhang vmci_memcpy_from_queue_func memcpy_from_queue, 273506164d2bSGeorge Zhang bool update_consumer, 273606164d2bSGeorge Zhang bool can_block) 273706164d2bSGeorge Zhang { 273806164d2bSGeorge Zhang s64 buf_ready; 273906164d2bSGeorge Zhang u64 head; 274006164d2bSGeorge Zhang size_t read; 274106164d2bSGeorge Zhang ssize_t result; 274206164d2bSGeorge Zhang 274306164d2bSGeorge Zhang result = qp_map_queue_headers(produce_q, consume_q, can_block); 274406164d2bSGeorge Zhang if (unlikely(result != VMCI_SUCCESS)) 274506164d2bSGeorge Zhang return result; 274606164d2bSGeorge Zhang 274706164d2bSGeorge Zhang buf_ready = vmci_q_header_buf_ready(consume_q->q_header, 274806164d2bSGeorge Zhang produce_q->q_header, 274906164d2bSGeorge Zhang consume_q_size); 275006164d2bSGeorge Zhang if (buf_ready == 0) 275106164d2bSGeorge Zhang return VMCI_ERROR_QUEUEPAIR_NODATA; 275206164d2bSGeorge Zhang 275306164d2bSGeorge Zhang if (buf_ready < VMCI_SUCCESS) 275406164d2bSGeorge Zhang return (ssize_t) buf_ready; 275506164d2bSGeorge Zhang 275606164d2bSGeorge Zhang read = (size_t) (buf_ready > buf_size ? buf_size : buf_ready); 275706164d2bSGeorge Zhang head = vmci_q_header_consumer_head(produce_q->q_header); 275806164d2bSGeorge Zhang if (likely(head + read < consume_q_size)) { 275906164d2bSGeorge Zhang result = memcpy_from_queue(buf, 0, consume_q, head, read); 276006164d2bSGeorge Zhang } else { 276106164d2bSGeorge Zhang /* Head pointer wraps around. */ 276206164d2bSGeorge Zhang 276306164d2bSGeorge Zhang const size_t tmp = (size_t) (consume_q_size - head); 276406164d2bSGeorge Zhang 276506164d2bSGeorge Zhang result = memcpy_from_queue(buf, 0, consume_q, head, tmp); 276606164d2bSGeorge Zhang if (result >= VMCI_SUCCESS) 276706164d2bSGeorge Zhang result = memcpy_from_queue(buf, tmp, consume_q, 0, 276806164d2bSGeorge Zhang read - tmp); 276906164d2bSGeorge Zhang 277006164d2bSGeorge Zhang } 277106164d2bSGeorge Zhang 277206164d2bSGeorge Zhang if (result < VMCI_SUCCESS) 277306164d2bSGeorge Zhang return result; 277406164d2bSGeorge Zhang 277506164d2bSGeorge Zhang if (update_consumer) 277606164d2bSGeorge Zhang vmci_q_header_add_consumer_head(produce_q->q_header, 277706164d2bSGeorge Zhang read, consume_q_size); 277806164d2bSGeorge Zhang 277906164d2bSGeorge Zhang return read; 278006164d2bSGeorge Zhang } 278106164d2bSGeorge Zhang 278206164d2bSGeorge Zhang /* 278306164d2bSGeorge Zhang * vmci_qpair_alloc() - Allocates a queue pair. 278406164d2bSGeorge Zhang * @qpair: Pointer for the new vmci_qp struct. 278506164d2bSGeorge Zhang * @handle: Handle to track the resource. 278606164d2bSGeorge Zhang * @produce_qsize: Desired size of the producer queue. 278706164d2bSGeorge Zhang * @consume_qsize: Desired size of the consumer queue. 278806164d2bSGeorge Zhang * @peer: ContextID of the peer. 278906164d2bSGeorge Zhang * @flags: VMCI flags. 279006164d2bSGeorge Zhang * @priv_flags: VMCI priviledge flags. 279106164d2bSGeorge Zhang * 279206164d2bSGeorge Zhang * This is the client interface for allocating the memory for a 279306164d2bSGeorge Zhang * vmci_qp structure and then attaching to the underlying 279406164d2bSGeorge Zhang * queue. If an error occurs allocating the memory for the 279506164d2bSGeorge Zhang * vmci_qp structure no attempt is made to attach. If an 279606164d2bSGeorge Zhang * error occurs attaching, then the structure is freed. 279706164d2bSGeorge Zhang */ 279806164d2bSGeorge Zhang int vmci_qpair_alloc(struct vmci_qp **qpair, 279906164d2bSGeorge Zhang struct vmci_handle *handle, 280006164d2bSGeorge Zhang u64 produce_qsize, 280106164d2bSGeorge Zhang u64 consume_qsize, 280206164d2bSGeorge Zhang u32 peer, 280306164d2bSGeorge Zhang u32 flags, 280406164d2bSGeorge Zhang u32 priv_flags) 280506164d2bSGeorge Zhang { 280606164d2bSGeorge Zhang struct vmci_qp *my_qpair; 280706164d2bSGeorge Zhang int retval; 280806164d2bSGeorge Zhang struct vmci_handle src = VMCI_INVALID_HANDLE; 280906164d2bSGeorge Zhang struct vmci_handle dst = vmci_make_handle(peer, VMCI_INVALID_ID); 281006164d2bSGeorge Zhang enum vmci_route route; 281106164d2bSGeorge Zhang vmci_event_release_cb wakeup_cb; 281206164d2bSGeorge Zhang void *client_data; 281306164d2bSGeorge Zhang 281406164d2bSGeorge Zhang /* 281506164d2bSGeorge Zhang * Restrict the size of a queuepair. The device already 281606164d2bSGeorge Zhang * enforces a limit on the total amount of memory that can be 281706164d2bSGeorge Zhang * allocated to queuepairs for a guest. However, we try to 281806164d2bSGeorge Zhang * allocate this memory before we make the queuepair 281906164d2bSGeorge Zhang * allocation hypercall. On Linux, we allocate each page 282006164d2bSGeorge Zhang * separately, which means rather than fail, the guest will 282106164d2bSGeorge Zhang * thrash while it tries to allocate, and will become 282206164d2bSGeorge Zhang * increasingly unresponsive to the point where it appears to 282306164d2bSGeorge Zhang * be hung. So we place a limit on the size of an individual 282406164d2bSGeorge Zhang * queuepair here, and leave the device to enforce the 282506164d2bSGeorge Zhang * restriction on total queuepair memory. (Note that this 282606164d2bSGeorge Zhang * doesn't prevent all cases; a user with only this much 282706164d2bSGeorge Zhang * physical memory could still get into trouble.) The error 282806164d2bSGeorge Zhang * used by the device is NO_RESOURCES, so use that here too. 282906164d2bSGeorge Zhang */ 283006164d2bSGeorge Zhang 283106164d2bSGeorge Zhang if (produce_qsize + consume_qsize < max(produce_qsize, consume_qsize) || 283206164d2bSGeorge Zhang produce_qsize + consume_qsize > VMCI_MAX_GUEST_QP_MEMORY) 283306164d2bSGeorge Zhang return VMCI_ERROR_NO_RESOURCES; 283406164d2bSGeorge Zhang 283506164d2bSGeorge Zhang retval = vmci_route(&src, &dst, false, &route); 283606164d2bSGeorge Zhang if (retval < VMCI_SUCCESS) 283706164d2bSGeorge Zhang route = vmci_guest_code_active() ? 283806164d2bSGeorge Zhang VMCI_ROUTE_AS_GUEST : VMCI_ROUTE_AS_HOST; 283906164d2bSGeorge Zhang 284006164d2bSGeorge Zhang /* If NONBLOCK or PINNED is set, we better be the guest personality. */ 284106164d2bSGeorge Zhang if ((!vmci_can_block(flags) || vmci_qp_pinned(flags)) && 284206164d2bSGeorge Zhang VMCI_ROUTE_AS_GUEST != route) { 284306164d2bSGeorge Zhang pr_devel("Not guest personality w/ NONBLOCK OR PINNED set"); 284406164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 284506164d2bSGeorge Zhang } 284606164d2bSGeorge Zhang 284706164d2bSGeorge Zhang /* 284806164d2bSGeorge Zhang * Limit the size of pinned QPs and check sanity. 284906164d2bSGeorge Zhang * 285006164d2bSGeorge Zhang * Pinned pages implies non-blocking mode. Mutexes aren't acquired 285106164d2bSGeorge Zhang * when the NONBLOCK flag is set in qpair code; and also should not be 285206164d2bSGeorge Zhang * acquired when the PINNED flagged is set. Since pinning pages 285306164d2bSGeorge Zhang * implies we want speed, it makes no sense not to have NONBLOCK 285406164d2bSGeorge Zhang * set if PINNED is set. Hence enforce this implication. 285506164d2bSGeorge Zhang */ 285606164d2bSGeorge Zhang if (vmci_qp_pinned(flags)) { 285706164d2bSGeorge Zhang if (vmci_can_block(flags)) { 285806164d2bSGeorge Zhang pr_err("Attempted to enable pinning w/o non-blocking"); 285906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 286006164d2bSGeorge Zhang } 286106164d2bSGeorge Zhang 286206164d2bSGeorge Zhang if (produce_qsize + consume_qsize > VMCI_MAX_PINNED_QP_MEMORY) 286306164d2bSGeorge Zhang return VMCI_ERROR_NO_RESOURCES; 286406164d2bSGeorge Zhang } 286506164d2bSGeorge Zhang 286606164d2bSGeorge Zhang my_qpair = kzalloc(sizeof(*my_qpair), GFP_KERNEL); 286706164d2bSGeorge Zhang if (!my_qpair) 286806164d2bSGeorge Zhang return VMCI_ERROR_NO_MEM; 286906164d2bSGeorge Zhang 287006164d2bSGeorge Zhang my_qpair->produce_q_size = produce_qsize; 287106164d2bSGeorge Zhang my_qpair->consume_q_size = consume_qsize; 287206164d2bSGeorge Zhang my_qpair->peer = peer; 287306164d2bSGeorge Zhang my_qpair->flags = flags; 287406164d2bSGeorge Zhang my_qpair->priv_flags = priv_flags; 287506164d2bSGeorge Zhang 287606164d2bSGeorge Zhang wakeup_cb = NULL; 287706164d2bSGeorge Zhang client_data = NULL; 287806164d2bSGeorge Zhang 287906164d2bSGeorge Zhang if (VMCI_ROUTE_AS_HOST == route) { 288006164d2bSGeorge Zhang my_qpair->guest_endpoint = false; 288106164d2bSGeorge Zhang if (!(flags & VMCI_QPFLAG_LOCAL)) { 288206164d2bSGeorge Zhang my_qpair->blocked = 0; 288306164d2bSGeorge Zhang my_qpair->generation = 0; 288406164d2bSGeorge Zhang init_waitqueue_head(&my_qpair->event); 288506164d2bSGeorge Zhang wakeup_cb = qp_wakeup_cb; 288606164d2bSGeorge Zhang client_data = (void *)my_qpair; 288706164d2bSGeorge Zhang } 288806164d2bSGeorge Zhang } else { 288906164d2bSGeorge Zhang my_qpair->guest_endpoint = true; 289006164d2bSGeorge Zhang } 289106164d2bSGeorge Zhang 289206164d2bSGeorge Zhang retval = vmci_qp_alloc(handle, 289306164d2bSGeorge Zhang &my_qpair->produce_q, 289406164d2bSGeorge Zhang my_qpair->produce_q_size, 289506164d2bSGeorge Zhang &my_qpair->consume_q, 289606164d2bSGeorge Zhang my_qpair->consume_q_size, 289706164d2bSGeorge Zhang my_qpair->peer, 289806164d2bSGeorge Zhang my_qpair->flags, 289906164d2bSGeorge Zhang my_qpair->priv_flags, 290006164d2bSGeorge Zhang my_qpair->guest_endpoint, 290106164d2bSGeorge Zhang wakeup_cb, client_data); 290206164d2bSGeorge Zhang 290306164d2bSGeorge Zhang if (retval < VMCI_SUCCESS) { 290406164d2bSGeorge Zhang kfree(my_qpair); 290506164d2bSGeorge Zhang return retval; 290606164d2bSGeorge Zhang } 290706164d2bSGeorge Zhang 290806164d2bSGeorge Zhang *qpair = my_qpair; 290906164d2bSGeorge Zhang my_qpair->handle = *handle; 291006164d2bSGeorge Zhang 291106164d2bSGeorge Zhang return retval; 291206164d2bSGeorge Zhang } 291306164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_alloc); 291406164d2bSGeorge Zhang 291506164d2bSGeorge Zhang /* 291606164d2bSGeorge Zhang * vmci_qpair_detach() - Detatches the client from a queue pair. 291706164d2bSGeorge Zhang * @qpair: Reference of a pointer to the qpair struct. 291806164d2bSGeorge Zhang * 291906164d2bSGeorge Zhang * This is the client interface for detaching from a VMCIQPair. 292006164d2bSGeorge Zhang * Note that this routine will free the memory allocated for the 292106164d2bSGeorge Zhang * vmci_qp structure too. 292206164d2bSGeorge Zhang */ 292306164d2bSGeorge Zhang int vmci_qpair_detach(struct vmci_qp **qpair) 292406164d2bSGeorge Zhang { 292506164d2bSGeorge Zhang int result; 292606164d2bSGeorge Zhang struct vmci_qp *old_qpair; 292706164d2bSGeorge Zhang 292806164d2bSGeorge Zhang if (!qpair || !(*qpair)) 292906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 293006164d2bSGeorge Zhang 293106164d2bSGeorge Zhang old_qpair = *qpair; 293206164d2bSGeorge Zhang result = qp_detatch(old_qpair->handle, old_qpair->guest_endpoint); 293306164d2bSGeorge Zhang 293406164d2bSGeorge Zhang /* 293506164d2bSGeorge Zhang * The guest can fail to detach for a number of reasons, and 293606164d2bSGeorge Zhang * if it does so, it will cleanup the entry (if there is one). 293706164d2bSGeorge Zhang * The host can fail too, but it won't cleanup the entry 293806164d2bSGeorge Zhang * immediately, it will do that later when the context is 293906164d2bSGeorge Zhang * freed. Either way, we need to release the qpair struct 294006164d2bSGeorge Zhang * here; there isn't much the caller can do, and we don't want 294106164d2bSGeorge Zhang * to leak. 294206164d2bSGeorge Zhang */ 294306164d2bSGeorge Zhang 294406164d2bSGeorge Zhang memset(old_qpair, 0, sizeof(*old_qpair)); 294506164d2bSGeorge Zhang old_qpair->handle = VMCI_INVALID_HANDLE; 294606164d2bSGeorge Zhang old_qpair->peer = VMCI_INVALID_ID; 294706164d2bSGeorge Zhang kfree(old_qpair); 294806164d2bSGeorge Zhang *qpair = NULL; 294906164d2bSGeorge Zhang 295006164d2bSGeorge Zhang return result; 295106164d2bSGeorge Zhang } 295206164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_detach); 295306164d2bSGeorge Zhang 295406164d2bSGeorge Zhang /* 295506164d2bSGeorge Zhang * vmci_qpair_get_produce_indexes() - Retrieves the indexes of the producer. 295606164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 295706164d2bSGeorge Zhang * @producer_tail: Reference used for storing producer tail index. 295806164d2bSGeorge Zhang * @consumer_head: Reference used for storing the consumer head index. 295906164d2bSGeorge Zhang * 296006164d2bSGeorge Zhang * This is the client interface for getting the current indexes of the 296106164d2bSGeorge Zhang * QPair from the point of the view of the caller as the producer. 296206164d2bSGeorge Zhang */ 296306164d2bSGeorge Zhang int vmci_qpair_get_produce_indexes(const struct vmci_qp *qpair, 296406164d2bSGeorge Zhang u64 *producer_tail, 296506164d2bSGeorge Zhang u64 *consumer_head) 296606164d2bSGeorge Zhang { 296706164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 296806164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 296906164d2bSGeorge Zhang int result; 297006164d2bSGeorge Zhang 297106164d2bSGeorge Zhang if (!qpair) 297206164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 297306164d2bSGeorge Zhang 297406164d2bSGeorge Zhang qp_lock(qpair); 297506164d2bSGeorge Zhang result = 297606164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 297706164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 297806164d2bSGeorge Zhang vmci_q_header_get_pointers(produce_q_header, consume_q_header, 297906164d2bSGeorge Zhang producer_tail, consumer_head); 298006164d2bSGeorge Zhang qp_unlock(qpair); 298106164d2bSGeorge Zhang 298206164d2bSGeorge Zhang if (result == VMCI_SUCCESS && 298306164d2bSGeorge Zhang ((producer_tail && *producer_tail >= qpair->produce_q_size) || 298406164d2bSGeorge Zhang (consumer_head && *consumer_head >= qpair->produce_q_size))) 298506164d2bSGeorge Zhang return VMCI_ERROR_INVALID_SIZE; 298606164d2bSGeorge Zhang 298706164d2bSGeorge Zhang return result; 298806164d2bSGeorge Zhang } 298906164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_get_produce_indexes); 299006164d2bSGeorge Zhang 299106164d2bSGeorge Zhang /* 299206164d2bSGeorge Zhang * vmci_qpair_get_consume_indexes() - Retrieves the indexes of the comsumer. 299306164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 299406164d2bSGeorge Zhang * @consumer_tail: Reference used for storing consumer tail index. 299506164d2bSGeorge Zhang * @producer_head: Reference used for storing the producer head index. 299606164d2bSGeorge Zhang * 299706164d2bSGeorge Zhang * This is the client interface for getting the current indexes of the 299806164d2bSGeorge Zhang * QPair from the point of the view of the caller as the consumer. 299906164d2bSGeorge Zhang */ 300006164d2bSGeorge Zhang int vmci_qpair_get_consume_indexes(const struct vmci_qp *qpair, 300106164d2bSGeorge Zhang u64 *consumer_tail, 300206164d2bSGeorge Zhang u64 *producer_head) 300306164d2bSGeorge Zhang { 300406164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 300506164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 300606164d2bSGeorge Zhang int result; 300706164d2bSGeorge Zhang 300806164d2bSGeorge Zhang if (!qpair) 300906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 301006164d2bSGeorge Zhang 301106164d2bSGeorge Zhang qp_lock(qpair); 301206164d2bSGeorge Zhang result = 301306164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 301406164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 301506164d2bSGeorge Zhang vmci_q_header_get_pointers(consume_q_header, produce_q_header, 301606164d2bSGeorge Zhang consumer_tail, producer_head); 301706164d2bSGeorge Zhang qp_unlock(qpair); 301806164d2bSGeorge Zhang 301906164d2bSGeorge Zhang if (result == VMCI_SUCCESS && 302006164d2bSGeorge Zhang ((consumer_tail && *consumer_tail >= qpair->consume_q_size) || 302106164d2bSGeorge Zhang (producer_head && *producer_head >= qpair->consume_q_size))) 302206164d2bSGeorge Zhang return VMCI_ERROR_INVALID_SIZE; 302306164d2bSGeorge Zhang 302406164d2bSGeorge Zhang return result; 302506164d2bSGeorge Zhang } 302606164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_get_consume_indexes); 302706164d2bSGeorge Zhang 302806164d2bSGeorge Zhang /* 302906164d2bSGeorge Zhang * vmci_qpair_produce_free_space() - Retrieves free space in producer queue. 303006164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 303106164d2bSGeorge Zhang * 303206164d2bSGeorge Zhang * This is the client interface for getting the amount of free 303306164d2bSGeorge Zhang * space in the QPair from the point of the view of the caller as 303406164d2bSGeorge Zhang * the producer which is the common case. Returns < 0 if err, else 303506164d2bSGeorge Zhang * available bytes into which data can be enqueued if > 0. 303606164d2bSGeorge Zhang */ 303706164d2bSGeorge Zhang s64 vmci_qpair_produce_free_space(const struct vmci_qp *qpair) 303806164d2bSGeorge Zhang { 303906164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 304006164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 304106164d2bSGeorge Zhang s64 result; 304206164d2bSGeorge Zhang 304306164d2bSGeorge Zhang if (!qpair) 304406164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 304506164d2bSGeorge Zhang 304606164d2bSGeorge Zhang qp_lock(qpair); 304706164d2bSGeorge Zhang result = 304806164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 304906164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 305006164d2bSGeorge Zhang result = vmci_q_header_free_space(produce_q_header, 305106164d2bSGeorge Zhang consume_q_header, 305206164d2bSGeorge Zhang qpair->produce_q_size); 305306164d2bSGeorge Zhang else 305406164d2bSGeorge Zhang result = 0; 305506164d2bSGeorge Zhang 305606164d2bSGeorge Zhang qp_unlock(qpair); 305706164d2bSGeorge Zhang 305806164d2bSGeorge Zhang return result; 305906164d2bSGeorge Zhang } 306006164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_produce_free_space); 306106164d2bSGeorge Zhang 306206164d2bSGeorge Zhang /* 306306164d2bSGeorge Zhang * vmci_qpair_consume_free_space() - Retrieves free space in consumer queue. 306406164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 306506164d2bSGeorge Zhang * 306606164d2bSGeorge Zhang * This is the client interface for getting the amount of free 306706164d2bSGeorge Zhang * space in the QPair from the point of the view of the caller as 306806164d2bSGeorge Zhang * the consumer which is not the common case. Returns < 0 if err, else 306906164d2bSGeorge Zhang * available bytes into which data can be enqueued if > 0. 307006164d2bSGeorge Zhang */ 307106164d2bSGeorge Zhang s64 vmci_qpair_consume_free_space(const struct vmci_qp *qpair) 307206164d2bSGeorge Zhang { 307306164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 307406164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 307506164d2bSGeorge Zhang s64 result; 307606164d2bSGeorge Zhang 307706164d2bSGeorge Zhang if (!qpair) 307806164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 307906164d2bSGeorge Zhang 308006164d2bSGeorge Zhang qp_lock(qpair); 308106164d2bSGeorge Zhang result = 308206164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 308306164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 308406164d2bSGeorge Zhang result = vmci_q_header_free_space(consume_q_header, 308506164d2bSGeorge Zhang produce_q_header, 308606164d2bSGeorge Zhang qpair->consume_q_size); 308706164d2bSGeorge Zhang else 308806164d2bSGeorge Zhang result = 0; 308906164d2bSGeorge Zhang 309006164d2bSGeorge Zhang qp_unlock(qpair); 309106164d2bSGeorge Zhang 309206164d2bSGeorge Zhang return result; 309306164d2bSGeorge Zhang } 309406164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_consume_free_space); 309506164d2bSGeorge Zhang 309606164d2bSGeorge Zhang /* 309706164d2bSGeorge Zhang * vmci_qpair_produce_buf_ready() - Gets bytes ready to read from 309806164d2bSGeorge Zhang * producer queue. 309906164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 310006164d2bSGeorge Zhang * 310106164d2bSGeorge Zhang * This is the client interface for getting the amount of 310206164d2bSGeorge Zhang * enqueued data in the QPair from the point of the view of the 310306164d2bSGeorge Zhang * caller as the producer which is not the common case. Returns < 0 if err, 310406164d2bSGeorge Zhang * else available bytes that may be read. 310506164d2bSGeorge Zhang */ 310606164d2bSGeorge Zhang s64 vmci_qpair_produce_buf_ready(const struct vmci_qp *qpair) 310706164d2bSGeorge Zhang { 310806164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 310906164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 311006164d2bSGeorge Zhang s64 result; 311106164d2bSGeorge Zhang 311206164d2bSGeorge Zhang if (!qpair) 311306164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 311406164d2bSGeorge Zhang 311506164d2bSGeorge Zhang qp_lock(qpair); 311606164d2bSGeorge Zhang result = 311706164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 311806164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 311906164d2bSGeorge Zhang result = vmci_q_header_buf_ready(produce_q_header, 312006164d2bSGeorge Zhang consume_q_header, 312106164d2bSGeorge Zhang qpair->produce_q_size); 312206164d2bSGeorge Zhang else 312306164d2bSGeorge Zhang result = 0; 312406164d2bSGeorge Zhang 312506164d2bSGeorge Zhang qp_unlock(qpair); 312606164d2bSGeorge Zhang 312706164d2bSGeorge Zhang return result; 312806164d2bSGeorge Zhang } 312906164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_produce_buf_ready); 313006164d2bSGeorge Zhang 313106164d2bSGeorge Zhang /* 313206164d2bSGeorge Zhang * vmci_qpair_consume_buf_ready() - Gets bytes ready to read from 313306164d2bSGeorge Zhang * consumer queue. 313406164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 313506164d2bSGeorge Zhang * 313606164d2bSGeorge Zhang * This is the client interface for getting the amount of 313706164d2bSGeorge Zhang * enqueued data in the QPair from the point of the view of the 313806164d2bSGeorge Zhang * caller as the consumer which is the normal case. Returns < 0 if err, 313906164d2bSGeorge Zhang * else available bytes that may be read. 314006164d2bSGeorge Zhang */ 314106164d2bSGeorge Zhang s64 vmci_qpair_consume_buf_ready(const struct vmci_qp *qpair) 314206164d2bSGeorge Zhang { 314306164d2bSGeorge Zhang struct vmci_queue_header *produce_q_header; 314406164d2bSGeorge Zhang struct vmci_queue_header *consume_q_header; 314506164d2bSGeorge Zhang s64 result; 314606164d2bSGeorge Zhang 314706164d2bSGeorge Zhang if (!qpair) 314806164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 314906164d2bSGeorge Zhang 315006164d2bSGeorge Zhang qp_lock(qpair); 315106164d2bSGeorge Zhang result = 315206164d2bSGeorge Zhang qp_get_queue_headers(qpair, &produce_q_header, &consume_q_header); 315306164d2bSGeorge Zhang if (result == VMCI_SUCCESS) 315406164d2bSGeorge Zhang result = vmci_q_header_buf_ready(consume_q_header, 315506164d2bSGeorge Zhang produce_q_header, 315606164d2bSGeorge Zhang qpair->consume_q_size); 315706164d2bSGeorge Zhang else 315806164d2bSGeorge Zhang result = 0; 315906164d2bSGeorge Zhang 316006164d2bSGeorge Zhang qp_unlock(qpair); 316106164d2bSGeorge Zhang 316206164d2bSGeorge Zhang return result; 316306164d2bSGeorge Zhang } 316406164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_consume_buf_ready); 316506164d2bSGeorge Zhang 316606164d2bSGeorge Zhang /* 316706164d2bSGeorge Zhang * vmci_qpair_enqueue() - Throw data on the queue. 316806164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 316906164d2bSGeorge Zhang * @buf: Pointer to buffer containing data 317006164d2bSGeorge Zhang * @buf_size: Length of buffer. 317106164d2bSGeorge Zhang * @buf_type: Buffer type (Unused). 317206164d2bSGeorge Zhang * 317306164d2bSGeorge Zhang * This is the client interface for enqueueing data into the queue. 317406164d2bSGeorge Zhang * Returns number of bytes enqueued or < 0 on error. 317506164d2bSGeorge Zhang */ 317606164d2bSGeorge Zhang ssize_t vmci_qpair_enqueue(struct vmci_qp *qpair, 317706164d2bSGeorge Zhang const void *buf, 317806164d2bSGeorge Zhang size_t buf_size, 317906164d2bSGeorge Zhang int buf_type) 318006164d2bSGeorge Zhang { 318106164d2bSGeorge Zhang ssize_t result; 318206164d2bSGeorge Zhang 318306164d2bSGeorge Zhang if (!qpair || !buf) 318406164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 318506164d2bSGeorge Zhang 318606164d2bSGeorge Zhang qp_lock(qpair); 318706164d2bSGeorge Zhang 318806164d2bSGeorge Zhang do { 318906164d2bSGeorge Zhang result = qp_enqueue_locked(qpair->produce_q, 319006164d2bSGeorge Zhang qpair->consume_q, 319106164d2bSGeorge Zhang qpair->produce_q_size, 319206164d2bSGeorge Zhang buf, buf_size, 319306164d2bSGeorge Zhang qp_memcpy_to_queue, 319406164d2bSGeorge Zhang vmci_can_block(qpair->flags)); 319506164d2bSGeorge Zhang 319606164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 319706164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 319806164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 319906164d2bSGeorge Zhang 320006164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 320106164d2bSGeorge Zhang 320206164d2bSGeorge Zhang qp_unlock(qpair); 320306164d2bSGeorge Zhang 320406164d2bSGeorge Zhang return result; 320506164d2bSGeorge Zhang } 320606164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_enqueue); 320706164d2bSGeorge Zhang 320806164d2bSGeorge Zhang /* 320906164d2bSGeorge Zhang * vmci_qpair_dequeue() - Get data from the queue. 321006164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 321106164d2bSGeorge Zhang * @buf: Pointer to buffer for the data 321206164d2bSGeorge Zhang * @buf_size: Length of buffer. 321306164d2bSGeorge Zhang * @buf_type: Buffer type (Unused). 321406164d2bSGeorge Zhang * 321506164d2bSGeorge Zhang * This is the client interface for dequeueing data from the queue. 321606164d2bSGeorge Zhang * Returns number of bytes dequeued or < 0 on error. 321706164d2bSGeorge Zhang */ 321806164d2bSGeorge Zhang ssize_t vmci_qpair_dequeue(struct vmci_qp *qpair, 321906164d2bSGeorge Zhang void *buf, 322006164d2bSGeorge Zhang size_t buf_size, 322106164d2bSGeorge Zhang int buf_type) 322206164d2bSGeorge Zhang { 322306164d2bSGeorge Zhang ssize_t result; 322406164d2bSGeorge Zhang 322506164d2bSGeorge Zhang if (!qpair || !buf) 322606164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 322706164d2bSGeorge Zhang 322806164d2bSGeorge Zhang qp_lock(qpair); 322906164d2bSGeorge Zhang 323006164d2bSGeorge Zhang do { 323106164d2bSGeorge Zhang result = qp_dequeue_locked(qpair->produce_q, 323206164d2bSGeorge Zhang qpair->consume_q, 323306164d2bSGeorge Zhang qpair->consume_q_size, 323406164d2bSGeorge Zhang buf, buf_size, 323506164d2bSGeorge Zhang qp_memcpy_from_queue, true, 323606164d2bSGeorge Zhang vmci_can_block(qpair->flags)); 323706164d2bSGeorge Zhang 323806164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 323906164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 324006164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 324106164d2bSGeorge Zhang 324206164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 324306164d2bSGeorge Zhang 324406164d2bSGeorge Zhang qp_unlock(qpair); 324506164d2bSGeorge Zhang 324606164d2bSGeorge Zhang return result; 324706164d2bSGeorge Zhang } 324806164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_dequeue); 324906164d2bSGeorge Zhang 325006164d2bSGeorge Zhang /* 325106164d2bSGeorge Zhang * vmci_qpair_peek() - Peek at the data in the queue. 325206164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 325306164d2bSGeorge Zhang * @buf: Pointer to buffer for the data 325406164d2bSGeorge Zhang * @buf_size: Length of buffer. 325506164d2bSGeorge Zhang * @buf_type: Buffer type (Unused on Linux). 325606164d2bSGeorge Zhang * 325706164d2bSGeorge Zhang * This is the client interface for peeking into a queue. (I.e., 325806164d2bSGeorge Zhang * copy data from the queue without updating the head pointer.) 325906164d2bSGeorge Zhang * Returns number of bytes dequeued or < 0 on error. 326006164d2bSGeorge Zhang */ 326106164d2bSGeorge Zhang ssize_t vmci_qpair_peek(struct vmci_qp *qpair, 326206164d2bSGeorge Zhang void *buf, 326306164d2bSGeorge Zhang size_t buf_size, 326406164d2bSGeorge Zhang int buf_type) 326506164d2bSGeorge Zhang { 326606164d2bSGeorge Zhang ssize_t result; 326706164d2bSGeorge Zhang 326806164d2bSGeorge Zhang if (!qpair || !buf) 326906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 327006164d2bSGeorge Zhang 327106164d2bSGeorge Zhang qp_lock(qpair); 327206164d2bSGeorge Zhang 327306164d2bSGeorge Zhang do { 327406164d2bSGeorge Zhang result = qp_dequeue_locked(qpair->produce_q, 327506164d2bSGeorge Zhang qpair->consume_q, 327606164d2bSGeorge Zhang qpair->consume_q_size, 327706164d2bSGeorge Zhang buf, buf_size, 327806164d2bSGeorge Zhang qp_memcpy_from_queue, false, 327906164d2bSGeorge Zhang vmci_can_block(qpair->flags)); 328006164d2bSGeorge Zhang 328106164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 328206164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 328306164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 328406164d2bSGeorge Zhang 328506164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 328606164d2bSGeorge Zhang 328706164d2bSGeorge Zhang qp_unlock(qpair); 328806164d2bSGeorge Zhang 328906164d2bSGeorge Zhang return result; 329006164d2bSGeorge Zhang } 329106164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_peek); 329206164d2bSGeorge Zhang 329306164d2bSGeorge Zhang /* 329406164d2bSGeorge Zhang * vmci_qpair_enquev() - Throw data on the queue using iov. 329506164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 329606164d2bSGeorge Zhang * @iov: Pointer to buffer containing data 329706164d2bSGeorge Zhang * @iov_size: Length of buffer. 329806164d2bSGeorge Zhang * @buf_type: Buffer type (Unused). 329906164d2bSGeorge Zhang * 330006164d2bSGeorge Zhang * This is the client interface for enqueueing data into the queue. 330106164d2bSGeorge Zhang * This function uses IO vectors to handle the work. Returns number 330206164d2bSGeorge Zhang * of bytes enqueued or < 0 on error. 330306164d2bSGeorge Zhang */ 330406164d2bSGeorge Zhang ssize_t vmci_qpair_enquev(struct vmci_qp *qpair, 330506164d2bSGeorge Zhang void *iov, 330606164d2bSGeorge Zhang size_t iov_size, 330706164d2bSGeorge Zhang int buf_type) 330806164d2bSGeorge Zhang { 330906164d2bSGeorge Zhang ssize_t result; 331006164d2bSGeorge Zhang 331106164d2bSGeorge Zhang if (!qpair || !iov) 331206164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 331306164d2bSGeorge Zhang 331406164d2bSGeorge Zhang qp_lock(qpair); 331506164d2bSGeorge Zhang 331606164d2bSGeorge Zhang do { 331706164d2bSGeorge Zhang result = qp_enqueue_locked(qpair->produce_q, 331806164d2bSGeorge Zhang qpair->consume_q, 331906164d2bSGeorge Zhang qpair->produce_q_size, 332006164d2bSGeorge Zhang iov, iov_size, 332106164d2bSGeorge Zhang qp_memcpy_to_queue_iov, 332206164d2bSGeorge Zhang vmci_can_block(qpair->flags)); 332306164d2bSGeorge Zhang 332406164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 332506164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 332606164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 332706164d2bSGeorge Zhang 332806164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 332906164d2bSGeorge Zhang 333006164d2bSGeorge Zhang qp_unlock(qpair); 333106164d2bSGeorge Zhang 333206164d2bSGeorge Zhang return result; 333306164d2bSGeorge Zhang } 333406164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_enquev); 333506164d2bSGeorge Zhang 333606164d2bSGeorge Zhang /* 333706164d2bSGeorge Zhang * vmci_qpair_dequev() - Get data from the queue using iov. 333806164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 333906164d2bSGeorge Zhang * @iov: Pointer to buffer for the data 334006164d2bSGeorge Zhang * @iov_size: Length of buffer. 334106164d2bSGeorge Zhang * @buf_type: Buffer type (Unused). 334206164d2bSGeorge Zhang * 334306164d2bSGeorge Zhang * This is the client interface for dequeueing data from the queue. 334406164d2bSGeorge Zhang * This function uses IO vectors to handle the work. Returns number 334506164d2bSGeorge Zhang * of bytes dequeued or < 0 on error. 334606164d2bSGeorge Zhang */ 334706164d2bSGeorge Zhang ssize_t vmci_qpair_dequev(struct vmci_qp *qpair, 334806164d2bSGeorge Zhang void *iov, 334906164d2bSGeorge Zhang size_t iov_size, 335006164d2bSGeorge Zhang int buf_type) 335106164d2bSGeorge Zhang { 335206164d2bSGeorge Zhang ssize_t result; 335306164d2bSGeorge Zhang 335406164d2bSGeorge Zhang qp_lock(qpair); 335506164d2bSGeorge Zhang 335606164d2bSGeorge Zhang if (!qpair || !iov) 335706164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 335806164d2bSGeorge Zhang 335906164d2bSGeorge Zhang do { 336006164d2bSGeorge Zhang result = qp_dequeue_locked(qpair->produce_q, 336106164d2bSGeorge Zhang qpair->consume_q, 336206164d2bSGeorge Zhang qpair->consume_q_size, 336306164d2bSGeorge Zhang iov, iov_size, 336406164d2bSGeorge Zhang qp_memcpy_from_queue_iov, 336506164d2bSGeorge Zhang true, vmci_can_block(qpair->flags)); 336606164d2bSGeorge Zhang 336706164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 336806164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 336906164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 337006164d2bSGeorge Zhang 337106164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 337206164d2bSGeorge Zhang 337306164d2bSGeorge Zhang qp_unlock(qpair); 337406164d2bSGeorge Zhang 337506164d2bSGeorge Zhang return result; 337606164d2bSGeorge Zhang } 337706164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_dequev); 337806164d2bSGeorge Zhang 337906164d2bSGeorge Zhang /* 338006164d2bSGeorge Zhang * vmci_qpair_peekv() - Peek at the data in the queue using iov. 338106164d2bSGeorge Zhang * @qpair: Pointer to the queue pair struct. 338206164d2bSGeorge Zhang * @iov: Pointer to buffer for the data 338306164d2bSGeorge Zhang * @iov_size: Length of buffer. 338406164d2bSGeorge Zhang * @buf_type: Buffer type (Unused on Linux). 338506164d2bSGeorge Zhang * 338606164d2bSGeorge Zhang * This is the client interface for peeking into a queue. (I.e., 338706164d2bSGeorge Zhang * copy data from the queue without updating the head pointer.) 338806164d2bSGeorge Zhang * This function uses IO vectors to handle the work. Returns number 338906164d2bSGeorge Zhang * of bytes peeked or < 0 on error. 339006164d2bSGeorge Zhang */ 339106164d2bSGeorge Zhang ssize_t vmci_qpair_peekv(struct vmci_qp *qpair, 339206164d2bSGeorge Zhang void *iov, 339306164d2bSGeorge Zhang size_t iov_size, 339406164d2bSGeorge Zhang int buf_type) 339506164d2bSGeorge Zhang { 339606164d2bSGeorge Zhang ssize_t result; 339706164d2bSGeorge Zhang 339806164d2bSGeorge Zhang if (!qpair || !iov) 339906164d2bSGeorge Zhang return VMCI_ERROR_INVALID_ARGS; 340006164d2bSGeorge Zhang 340106164d2bSGeorge Zhang qp_lock(qpair); 340206164d2bSGeorge Zhang 340306164d2bSGeorge Zhang do { 340406164d2bSGeorge Zhang result = qp_dequeue_locked(qpair->produce_q, 340506164d2bSGeorge Zhang qpair->consume_q, 340606164d2bSGeorge Zhang qpair->consume_q_size, 340706164d2bSGeorge Zhang iov, iov_size, 340806164d2bSGeorge Zhang qp_memcpy_from_queue_iov, 340906164d2bSGeorge Zhang false, vmci_can_block(qpair->flags)); 341006164d2bSGeorge Zhang 341106164d2bSGeorge Zhang if (result == VMCI_ERROR_QUEUEPAIR_NOT_READY && 341206164d2bSGeorge Zhang !qp_wait_for_ready_queue(qpair)) 341306164d2bSGeorge Zhang result = VMCI_ERROR_WOULD_BLOCK; 341406164d2bSGeorge Zhang 341506164d2bSGeorge Zhang } while (result == VMCI_ERROR_QUEUEPAIR_NOT_READY); 341606164d2bSGeorge Zhang 341706164d2bSGeorge Zhang qp_unlock(qpair); 341806164d2bSGeorge Zhang return result; 341906164d2bSGeorge Zhang } 342006164d2bSGeorge Zhang EXPORT_SYMBOL_GPL(vmci_qpair_peekv); 3421