Subject: [PATCH] Do not use signals in posix-aio

The signal-select race forces us to use a file descriptor for posix-aio
notification.  Currently, we use a pipe and write to it from the signal handler
because normal posix-aio does not provide file descriptor notification support.

Since we now emulation posix-aio on our own, we can clean things up and just
use a file descriptor for notification.

N.B. This is almost certainly going to cause regressions until we fix the
select loop to run in a separate thread since there will no longer be a signal
that breaks us out of the TCG run loop.  We need to hold off on applying this
patch until that is fixed properly.

Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

diff --git a/block-raw-posix.c b/block-raw-posix.c
index 4404eb1..cae4408 100644
--- a/block-raw-posix.c
+++ b/block-raw-posix.c
@@ -511,20 +511,11 @@ static int posix_aio_flush(void *opaque)
 
 static PosixAioState *posix_aio_state;
 
-static void aio_signal_handler(int signum)
-{
-    if (posix_aio_state) {
-        char byte = 0;
-
-        write(posix_aio_state->wfd, &byte, sizeof(byte));
-    }
-
-    qemu_service_io();
-}
+/* FIXME need to fix the select() loop issue b/c this IO will not
+ * necessarily break out us out the TCG loop */
 
 static int posix_aio_init(void)
 {
-    struct sigaction act;
     PosixAioState *s;
     int fds[2];
     struct qemu_paioinit ai;
@@ -536,11 +527,6 @@ static int posix_aio_init(void)
     if (s == NULL)
         return -ENOMEM;
 
-    sigfillset(&act.sa_mask);
-    act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
-    act.sa_handler = aio_signal_handler;
-    sigaction(SIGUSR2, &act, NULL);
-
     s->first_aio = NULL;
     if (pipe(fds) == -1) {
         fprintf(stderr, "failed to create pipe\n");
@@ -579,7 +565,7 @@ static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
     if (!acb)
         return NULL;
     acb->aiocb.aio_fildes = s->fd;
-    acb->aiocb.ev_signo = SIGUSR2;
+    acb->aiocb.ev_fd = posix_aio_state->wfd;
     acb->aiocb.aio_buf = buf;
     if (nb_sectors < 0)
         acb->aiocb.aio_nbytes = -nb_sectors;
diff --git a/posix-aio-compat.c b/posix-aio-compat.c
index f141cd9..4409899 100644
--- a/posix-aio-compat.c
+++ b/posix-aio-compat.c
@@ -27,6 +27,15 @@ static int cur_threads = 0;
 static int idle_threads = 0;
 static TAILQ_HEAD(, qemu_paiocb) request_list;
 
+static void signal_fd(int fd)
+{
+    ssize_t len;
+    do {
+        char value = 1;
+        len = write(fd, &value, 1);
+    } while (len == -1 && errno == EINTR);
+}
+
 static void *aio_thread(void *unused)
 {
     sigset_t set;
@@ -92,9 +101,8 @@ static void *aio_thread(void *unused)
         pthread_mutex_lock(&lock);
         aiocb->ret = offset;
         idle_threads++;
+        signal_fd(aiocb->ev_fd);
         pthread_mutex_unlock(&lock);
-
-        kill(getpid(), aiocb->ev_signo);
     }
 
     idle_threads--;
diff --git a/posix-aio-compat.h b/posix-aio-compat.h
index 0bc10f5..ec17bb6 100644
--- a/posix-aio-compat.h
+++ b/posix-aio-compat.h
@@ -29,7 +29,7 @@ struct qemu_paiocb
     int aio_fildes;
     void *aio_buf;
     size_t aio_nbytes;
-    int ev_signo;
+    int ev_fd;
     off_t aio_offset;
 
     /* private */