Yoshihiro YUNOMAE | 108fc82 | 2012-08-09 21:31:30 +0900 | [diff] [blame] | 1 | /* |
| 2 | * Read/write thread of a guest agent for virtio-trace |
| 3 | * |
| 4 | * Copyright (C) 2012 Hitachi, Ltd. |
| 5 | * Created by Yoshihiro Yunomae <yoshihiro.yunomae.ez@hitachi.com> |
| 6 | * Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> |
| 7 | * |
| 8 | * Licensed under GPL version 2 only. |
| 9 | * |
| 10 | */ |
| 11 | |
| 12 | #define _GNU_SOURCE |
| 13 | #include <fcntl.h> |
| 14 | #include <stdio.h> |
| 15 | #include <stdlib.h> |
| 16 | #include <unistd.h> |
| 17 | #include <sys/syscall.h> |
| 18 | #include "trace-agent.h" |
| 19 | |
| 20 | #define READ_WAIT_USEC 100000 |
| 21 | |
| 22 | void *rw_thread_info_new(void) |
| 23 | { |
| 24 | struct rw_thread_info *rw_ti; |
| 25 | |
| 26 | rw_ti = zalloc(sizeof(struct rw_thread_info)); |
| 27 | if (rw_ti == NULL) { |
| 28 | pr_err("rw_thread_info zalloc error\n"); |
| 29 | exit(EXIT_FAILURE); |
| 30 | } |
| 31 | |
| 32 | rw_ti->cpu_num = -1; |
| 33 | rw_ti->in_fd = -1; |
| 34 | rw_ti->out_fd = -1; |
| 35 | rw_ti->read_pipe = -1; |
| 36 | rw_ti->write_pipe = -1; |
| 37 | rw_ti->pipe_size = PIPE_INIT; |
| 38 | |
| 39 | return rw_ti; |
| 40 | } |
| 41 | |
| 42 | void *rw_thread_init(int cpu, const char *in_path, const char *out_path, |
| 43 | bool stdout_flag, unsigned long pipe_size, |
| 44 | struct rw_thread_info *rw_ti) |
| 45 | { |
| 46 | int data_pipe[2]; |
| 47 | |
| 48 | rw_ti->cpu_num = cpu; |
| 49 | |
| 50 | /* set read(input) fd */ |
| 51 | rw_ti->in_fd = open(in_path, O_RDONLY); |
| 52 | if (rw_ti->in_fd == -1) { |
| 53 | pr_err("Could not open in_fd (CPU:%d)\n", cpu); |
| 54 | goto error; |
| 55 | } |
| 56 | |
| 57 | /* set write(output) fd */ |
| 58 | if (!stdout_flag) { |
| 59 | /* virtio-serial output mode */ |
| 60 | rw_ti->out_fd = open(out_path, O_WRONLY); |
| 61 | if (rw_ti->out_fd == -1) { |
| 62 | pr_err("Could not open out_fd (CPU:%d)\n", cpu); |
| 63 | goto error; |
| 64 | } |
| 65 | } else |
| 66 | /* stdout mode */ |
| 67 | rw_ti->out_fd = STDOUT_FILENO; |
| 68 | |
| 69 | if (pipe2(data_pipe, O_NONBLOCK) < 0) { |
| 70 | pr_err("Could not create pipe in rw-thread(%d)\n", cpu); |
| 71 | goto error; |
| 72 | } |
| 73 | |
| 74 | /* |
| 75 | * Size of pipe is 64kB in default based on fs/pipe.c. |
| 76 | * To read/write trace data speedy, pipe size is changed. |
| 77 | */ |
| 78 | if (fcntl(*data_pipe, F_SETPIPE_SZ, pipe_size) < 0) { |
| 79 | pr_err("Could not change pipe size in rw-thread(%d)\n", cpu); |
| 80 | goto error; |
| 81 | } |
| 82 | |
| 83 | rw_ti->read_pipe = data_pipe[1]; |
| 84 | rw_ti->write_pipe = data_pipe[0]; |
| 85 | rw_ti->pipe_size = pipe_size; |
| 86 | |
| 87 | return NULL; |
| 88 | |
| 89 | error: |
| 90 | exit(EXIT_FAILURE); |
| 91 | } |
| 92 | |
| 93 | /* Bind a thread to a cpu */ |
| 94 | static void bind_cpu(int cpu_num) |
| 95 | { |
| 96 | cpu_set_t mask; |
| 97 | |
| 98 | CPU_ZERO(&mask); |
| 99 | CPU_SET(cpu_num, &mask); |
| 100 | |
| 101 | /* bind my thread to cpu_num by assigning zero to the first argument */ |
| 102 | if (sched_setaffinity(0, sizeof(mask), &mask) == -1) |
| 103 | pr_err("Could not set CPU#%d affinity\n", (int)cpu_num); |
| 104 | } |
| 105 | |
| 106 | static void *rw_thread_main(void *thread_info) |
| 107 | { |
| 108 | ssize_t rlen, wlen; |
| 109 | ssize_t ret; |
| 110 | struct rw_thread_info *ts = (struct rw_thread_info *)thread_info; |
| 111 | |
| 112 | bind_cpu(ts->cpu_num); |
| 113 | |
| 114 | while (1) { |
| 115 | /* Wait for a read order of trace data by Host OS */ |
| 116 | if (!global_run_operation) { |
| 117 | pthread_mutex_lock(&mutex_notify); |
| 118 | pthread_cond_wait(&cond_wakeup, &mutex_notify); |
| 119 | pthread_mutex_unlock(&mutex_notify); |
| 120 | } |
| 121 | |
| 122 | if (global_sig_receive) |
| 123 | break; |
| 124 | |
| 125 | /* |
| 126 | * Each thread read trace_pipe_raw of each cpu bounding the |
| 127 | * thread, so contention of multi-threads does not occur. |
| 128 | */ |
| 129 | rlen = splice(ts->in_fd, NULL, ts->read_pipe, NULL, |
| 130 | ts->pipe_size, SPLICE_F_MOVE | SPLICE_F_MORE); |
| 131 | |
| 132 | if (rlen < 0) { |
| 133 | pr_err("Splice_read in rw-thread(%d)\n", ts->cpu_num); |
| 134 | goto error; |
| 135 | } else if (rlen == 0) { |
| 136 | /* |
| 137 | * If trace data do not exist or are unreadable not |
| 138 | * for exceeding the page size, splice_read returns |
| 139 | * NULL. Then, this waits for being filled the data in a |
| 140 | * ring-buffer. |
| 141 | */ |
| 142 | usleep(READ_WAIT_USEC); |
| 143 | pr_debug("Read retry(cpu:%d)\n", ts->cpu_num); |
| 144 | continue; |
| 145 | } |
| 146 | |
| 147 | wlen = 0; |
| 148 | |
| 149 | do { |
| 150 | ret = splice(ts->write_pipe, NULL, ts->out_fd, NULL, |
| 151 | rlen - wlen, |
| 152 | SPLICE_F_MOVE | SPLICE_F_MORE); |
| 153 | |
| 154 | if (ret < 0) { |
| 155 | pr_err("Splice_write in rw-thread(%d)\n", |
| 156 | ts->cpu_num); |
| 157 | goto error; |
| 158 | } else if (ret == 0) |
| 159 | /* |
| 160 | * When host reader is not in time for reading |
| 161 | * trace data, guest will be stopped. This is |
| 162 | * because char dev in QEMU is not supported |
| 163 | * non-blocking mode. Then, writer might be |
| 164 | * sleep in that case. |
| 165 | * This sleep will be removed by supporting |
| 166 | * non-blocking mode. |
| 167 | */ |
| 168 | sleep(1); |
| 169 | wlen += ret; |
| 170 | } while (wlen < rlen); |
| 171 | } |
| 172 | |
| 173 | return NULL; |
| 174 | |
| 175 | error: |
| 176 | exit(EXIT_FAILURE); |
| 177 | } |
| 178 | |
| 179 | |
| 180 | pthread_t rw_thread_run(struct rw_thread_info *rw_ti) |
| 181 | { |
| 182 | int ret; |
| 183 | pthread_t rw_thread_per_cpu; |
| 184 | |
| 185 | ret = pthread_create(&rw_thread_per_cpu, NULL, rw_thread_main, rw_ti); |
| 186 | if (ret != 0) { |
| 187 | pr_err("Could not create a rw thread(%d)\n", rw_ti->cpu_num); |
| 188 | exit(EXIT_FAILURE); |
| 189 | } |
| 190 | |
| 191 | return rw_thread_per_cpu; |
| 192 | } |