一，前言
之前Qemu2.8虚拟机源码分析—Apple的学习笔记中我已经大概了的了解了glib在qemu中的应用，就是和事件触发相关。那么本次我在我新建立的虚拟机ubuntu18.04中学习下glib的API，首先自己编译了个2.67版本的glib，用了meson和nija编译的。对应着官网的API手册来学习glib源码test文件夹中的example code。
二，glib event应用源码分析
image.png
三，源码
#undef G_DISABLE_ASSERT
#undef G_LOG_DOMAIN
#include
#include
#ifdef G_OS_UNIX
#include
#endif
#include
#include
#ifdef G_OS_WIN32
#include /* For _O_BINARY used by pipe() macro */
#include /* for _pipe() */
#define pipe(fds) _pipe(fds, 4096, _O_BINARY)
#endif
#define VERBOSE
#define ITERS 50
#define INCREMENT 10
#define NTHREADS 2//4
#define NCRAWLERS 4
#define CRAWLER_TIMEOUT_RANGE 40
#define RECURSER_TIMEOUT 50
/* The partial ordering between the context array mutex and
* crawler array mutex is that the crawler array mutex cannot
* be locked while the context array mutex is locked
*/
GPtrArray *context_array;
GMutex context_array_mutex;
GCond context_array_cond;
GMainLoop *main_loop;
G_LOCK_DEFINE_STATIC (crawler_array_lock);
GPtrArray *crawler_array;
typedef struct _AddrData AddrData;
typedef struct _TestData TestData;
struct _AddrData
{
GMainLoop *loop;
GIOChannel *dest;
gint count;
};
struct _TestData
{
gint current_val;
gint iters;
GIOChannel *in;
};
static void cleanup_crawlers (GMainContext *context);
static gboolean
read_all (GIOChannel *channel, char *buf, gsize len)
{
gsize bytes_read = 0;
gsize count;
GIOError err;
while (bytes_read < len)
{
err = g_io_channel_read (channel, buf + bytes_read, len - bytes_read, &count);
if (err)
{
printf("\nerr1\n");
if (err != G_IO_ERROR_AGAIN)
return FALSE;
}
else if (count == 0)
{
printf("\ncount=0\n");
return FALSE;
}
bytes_read += count;
}
return TRUE;
}
static gboolean
write_all (GIOChannel *channel, char *buf, gsize len)
{
gsize bytes_written = 0;
gsize count;
GIOError err;
while (bytes_written < len)
{
err = g_io_channel_write (channel, buf + bytes_written, len - bytes_written, &count);
if (err && err != G_IO_ERROR_AGAIN)
return FALSE;
bytes_written += count;
}
return TRUE;
}
static gboolean
adder_callback (GIOChannel *source,
GIOCondition condition,
gpointer data)
{
char buf1[32];
char buf2[32];
printf("\nadder_callback\n");
char result[32] = { 0, };
AddrData *addr_data = data;
if (!read_all (source, buf1, 32) ||
!read_all (source, buf2, 32))
{
g_main_loop_quit (addr_data->loop);
printf("\n gohere\n");
return FALSE;
}
sprintf (result, "%d", atoi(buf1) + atoi(buf2));
printf("all:%s,buf1=%s,buf2=%s ",result,buf1,buf2);
write_all (addr_data->dest, result, 32);
return TRUE;
}
static gboolean
timeout_callback (gpointer data)
{
AddrData *addr_data = data;
addr_data->count++;
return TRUE;
}
static gpointer
adder_thread (gpointer data)
{
GMainContext *context;
GSource *adder_source;
GSource *timeout_source;
GIOChannel **channels = data;
AddrData addr_data;
context = g_main_context_new ();
g_mutex_lock (&context_array_mutex);
g_ptr_array_add (context_array, context);
if (context_array->len == NTHREADS)
g_cond_broadcast (&context_array_cond);
g_mutex_unlock (&context_array_mutex);
addr_data.dest = channels[1];//use outchannel[1]
addr_data.loop = g_main_loop_new (context, FALSE);
addr_data.count = 0;
adder_source = g_io_create_watch (channels[0], G_IO_IN | G_IO_HUP);//use inchannel[0]
//adder_source = g_io_create_watch (channels[0], G_IO_IN );//use inchannel[0]
g_source_set_name (adder_source, "Adder I/O");
g_source_set_callback (adder_source, (GSourceFunc)adder_callback, &addr_data, NULL);
g_source_attach (adder_source, context);
g_source_unref (adder_source);
timeout_source = g_timeout_source_new (10);
g_source_set_name (timeout_source, "Adder timeout");
g_source_set_callback (timeout_source, (GSourceFunc)timeout_callback, &addr_data, NULL);
g_source_set_priority (timeout_source, G_PRIORITY_HIGH);
g_source_attach (timeout_source, context);
g_source_unref (timeout_source);
g_main_loop_run (addr_data.loop);
g_io_channel_unref (channels[0]);
g_io_channel_unref (channels[1]);
g_free (channels);
g_main_loop_unref (addr_data.loop);
#ifdef VERBOSE
g_print ("Timeout run %d times\n", addr_data.count);
#endif
g_mutex_lock (&context_array_mutex);
g_ptr_array_remove (context_array, context);
if (context_array->len == 0)
g_main_loop_quit (main_loop);
g_mutex_unlock (&context_array_mutex);
cleanup_crawlers (context);
g_main_context_unref (context);
return NULL;
}
static void
io_pipe (GIOChannel **channels)
{
gint fds[2];
if (pipe(fds) < 0)
{
int errsv = errno;
g_warning ("Cannot create pipe %s", g_strerror (errsv));
exit (1);
}
channels[0] = g_io_channel_unix_new (fds[0]);
channels[1] = g_io_channel_unix_new (fds[1]);
g_io_channel_set_close_on_unref (channels[0], TRUE);
g_io_channel_set_close_on_unref (channels[1], TRUE);
}
static void
do_add (GIOChannel *in, gint a, gint b)
{
char buf1[32] = { 0, };
char buf2[32] = { 0, };
sprintf (buf1, "%d", a);
sprintf (buf2, "%d", b);
write_all (in, buf1, 32);
write_all (in, buf2, 32);
}
static gboolean
adder_response (GIOChannel *source,
GIOCondition condition,
gpointer data)
{
char result[32];
TestData *test_data = data;
if (!read_all (source, result, 32))
return FALSE;
printf("adder_response\n");
test_data->current_val = atoi (result);
test_data->iters--;
if (test_data->iters == 0)
{
if (test_data->current_val != ITERS * INCREMENT)
{
g_print ("Addition failed: %d != %d\n",
test_data->current_val, ITERS * INCREMENT);
exit (1);
}
g_io_channel_unref (source);
g_io_channel_unref (test_data->in); // 这句很重要，是配合G_IO_HUP的，否则thread的loop关闭不了。等于删除了inchannel[1]写，那么inchannel[0]读就会挂起，不通了。
g_free (test_data);
return FALSE;
}
printf("thread do_add\n");
do_add (test_data->in, test_data->current_val, INCREMENT);
return TRUE;
}
static GThread *
create_adder_thread (void)
{
GThread *thread;
TestData *test_data;
GIOChannel *in_channels[2];
GIOChannel *out_channels[2];
GIOChannel **sub_channels;
sub_channels = g_new (GIOChannel *, 2);
io_pipe (in_channels);
io_pipe (out_channels);
sub_channels[0] = in_channels[0];
sub_channels[1] = out_channels[1];
thread = g_thread_new ("adder", adder_thread, sub_channels);
test_data = g_new (TestData, 1);
test_data->in = in_channels[1];
test_data->current_val = 0;
test_data->iters = ITERS;
//g_io_add_watch (out_channels[0], G_IO_IN | G_IO_HUP,adder_response, test_data);
g_io_add_watch (out_channels[0], G_IO_IN ,adder_response, test_data);
printf("do_add\n");
do_add (test_data->in, test_data->current_val, INCREMENT);//从0开始，每次加10
return thread;
}
static void create_crawler (void);
static void
remove_crawler (void)
{
GSource *other_source;
if (crawler_array->len > 0)
{
other_source = crawler_array->pdata[g_random_int_range (0, crawler_array->len)];
g_source_destroy (other_source);
g_assert (g_ptr_array_remove_fast (crawler_array, other_source));
}
}
static gint
crawler_callback (gpointer data)
{
GSource *source = data;
G_LOCK (crawler_array_lock);
if (!g_ptr_array_remove_fast (crawler_array, source))
remove_crawler();
remove_crawler();
G_UNLOCK (crawler_array_lock);
create_crawler();
create_crawler();
return FALSE;
}
static void
create_crawler (void)
{
GSource *source = g_timeout_source_new (g_random_int_range (0, CRAWLER_TIMEOUT_RANGE));
g_source_set_name (source, "Crawler timeout");
g_source_set_callback (source, (GSourceFunc)crawler_callback, source, NULL);
G_LOCK (crawler_array_lock);
g_ptr_array_add (crawler_array, source);
g_mutex_lock (&context_array_mutex);
g_source_attach (source, context_array->pdata[g_random_int_range (0, context_array->len)]);
g_source_unref (source);
g_mutex_unlock (&context_array_mutex);
G_UNLOCK (crawler_array_lock);
}
static void
cleanup_crawlers (GMainContext *context)
{
gint i;
G_LOCK (crawler_array_lock);
for (i=0; i < crawler_array->len; i++)
{
if (g_source_get_context (crawler_array->pdata[i]) == context)
{
g_source_destroy (g_ptr_array_remove_index (crawler_array, i));
i--;
}
}
G_UNLOCK (crawler_array_lock);
}
static gboolean
recurser_idle (gpointer data)
{
GMainContext *context = data;
gint i;
for (i = 0; i < 10; i++)
g_main_context_iteration (context, FALSE);
return FALSE;
}
static gboolean
recurser_start (gpointer data)
{
GMainContext *context;
GSource *source;
g_mutex_lock (&context_array_mutex);
if (context_array->len > 0)
{
context = context_array->pdata[g_random_int_range (0, context_array->len)];
source = g_idle_source_new ();
g_source_set_name (source, "Recursing idle source");
g_source_set_callback (source, recurser_idle, context, NULL);
g_source_attach (source, context);
g_source_unref (source);
}
g_mutex_unlock (&context_array_mutex);
return TRUE;
}
int
main (int argc,
char *argv[])
{
gint i;
GThread *threads[NTHREADS];
context_array = g_ptr_array_new ();
crawler_array = g_ptr_array_new ();
main_loop = g_main_loop_new (NULL, FALSE);
for (i = 0; i < NTHREADS; i++)
threads[i] = create_adder_thread ();
/* Wait for all threads to start
*/
g_mutex_lock (&context_array_mutex);
while (context_array->len < NTHREADS)
g_cond_wait (&context_array_cond, &context_array_mutex);
g_mutex_unlock (&context_array_mutex);
//for (i = 0; i < NCRAWLERS; i++)
// create_crawler ();
g_timeout_add (RECURSER_TIMEOUT, recurser_start, NULL);
g_main_loop_run (main_loop);
g_main_loop_unref (main_loop);
for (i = 0; i < NTHREADS; i++)
g_thread_join (threads[i]);
g_ptr_array_unref (crawler_array);
g_ptr_array_unref (context_array);
return 0;
}
四，qemu中的glib应用
qemu中没有看到g_main_loop_run，它是用do while循环4个query，check，dispatch等来代替main_loop_run的。另外用的是ppoll，传入的是一个fd的数组和数组元素，因此它能够同时poll多个fd，poll的时候是休眠的，一旦其中一个fd准备好，poll就会被唤醒，同时GPollFD.revent会被内核设置，表明fd的状态。
qemu_init_main_loop
->aio_context_new
-->event_notifier_init(创建2条nonblock pipe)
-->aio_set_event_notifier->aio_set_fd_handler（find_aio_handler就是找到node）,g_source_add_poll后面是将fd绑定到node中，将node加入链表。将source添加到线程池。最后还是设置了event为true。
然后就是创建了带source的handler，qemu_aio_context和iohandler_ctx，并且命名"aio-context"和"io-handler"都将source绑定到了default context上。qemu中关于glib的polling调用为如下：
main_loop->main_loop_wait->os_host_main_loop_wait->glib_pollfds_poll
五，小结
此次其实就是单独的学习glib库，我的的另外一个目的是glib是gtk的基础库，所以将来学习图像相关的gtk，由于我现在对glib的学习经验，将来学习gtk源码会更顺利些。