Re: [Qemu-devel] [PATCH] qemu-char: Fixed the bug lead to dead lock

[Li, LiangX Z]

Yes, my patch will break the thread safe implementation, just avoid the 
deadlock.
This is the call trace when the dead lock happened.

#0  0x00007ffff27dbf79 in __GI_raise (address@hidden) at 
../nptl/sysdeps/unix/sysv/linux/raise.c:56
#1  0x00007ffff27df388 in __GI_abort () at abort.c:89
#2  0x00005555555ef489 in error_exit (err=<optimized out>, address@hidden 
<__func__.5959> "qemu_mutex_lock") at util/qemu-thread-posix.c:47
#3  0x00005555558f9080 in qemu_mutex_lock (address@hidden) at 
util/qemu-thread-posix.c:78
#4  0x0000555555713936 in qemu_chr_fe_write (s=0x555556248a30, address@hidden 
"QEMU 2.0.90 monitor - type 'help' for more information\r\n", len=56) at 
qemu-char.c:127
#5  0x00005555556217fd in monitor_flush_locked (address@hidden) at 
/home/liliang/vt-sync/qemu/monitor.c:310
#6  0x0000555555621a12 in monitor_flush_locked (mon=0x555556251fd0) at 
/home/liliang/vt-sync/qemu/monitor.c:302
#7  monitor_puts (address@hidden, str=0x55555634bfa7 "", address@hidden "QEMU 
2.0.90 monitor - type 'help' for more information\n") at 
/home/liliang/vt-sync/qemu/monitor.c:352
#8  0x0000555555624359 in monitor_vprintf (mon=0x555556251fd0, fmt=<optimized 
out>, ap=<optimized out>) at /home/liliang/vt-sync/qemu/monitor.c:370
#9  0x0000555555624414 in monitor_printf (mon=<optimized out>, address@hidden 
"QEMU %s monitor - type 'help' for more information\n") at 
/home/liliang/vt-sync/qemu/monitor.c:378
#10 0x0000555555629806 in monitor_event (opaque=0x555556251fd0, 
event=<optimized out>) at /home/liliang/vt-sync/qemu/monitor.c:5243
#11 0x000055555571343c in qemu_chr_be_generic_open (s=0x555556248a30) at 
qemu-char.c:120
#12 pty_chr_state (address@hidden, address@hidden) at qemu-char.c:1216
#13 0x00005555557134ee in pty_chr_update_read_handler_locked (address@hidden) 
at qemu-char.c:1133
#14 0x00005555557135ab in pty_chr_timer (opaque=0x555556248a30) at 
qemu-char.c:1097
#15 0x00007ffff629d703 in ?? () from /lib/x86_64-linux-gnu/libglib-2.0.so.0
#16 0x00007ffff629cce5 in g_main_context_dispatch () from 
/lib/x86_64-linux-gnu/libglib-2.0.so.0
#17 0x00005555558ac888 in glib_pollfds_poll () at main-loop.c:190
#18 os_host_main_loop_wait (timeout=<optimized out>) at main-loop.c:235
#19 main_loop_wait (nonblocking=<optimized out>) at main-loop.c:484
#20 0x00005555555f153e in main_loop () at vl.c:2010
#21 main (argc=<optimized out>, argv=<optimized out>, envp=<optimized out>) at 
vl.c:4530


It can be showed simply like bellow:

        ptr_char_timer ()
                {
                qemu_mutex_lock(&chr->chr_write_lock);
            ...
            monitor_flush_locked()
                        {        
                        ...
                        qemu_chr_fe_write()
                                {
                                qemu_mutex_lock(&chr->chr_write_lock);
                                ...
                                qemu_mutex_unlock(&chr->chr_write_lock);
                                }
                                …
                        }
            qemu_mutex_unlock(&chr->chr_write_lock);

                }

Acquire mutex in this way cause deadlock.


-----Original Message-----
From: Peter Maydell [mailto:address@hidden 
Sent: Friday, July 11, 2014 5:08 PM
To: Li, LiangX Z
Cc: QEMU Developers; Anthony Liguori
Subject: Re: [Qemu-devel] [PATCH] qemu-char: Fixed the bug lead to dead lock

On 11 July 2014 05:58, Liliang <address@hidden> wrote:
> From: Li Liang <address@hidden>
>
> This bug was introduced in the commit 
> 9005b2a7589540a3733b3abdcfbccfe7746cd1a1,
> it will cause deadlock when create a vm with the parameter "-monitor 
> pty" and then try to read from /dev/pts/x.
>
> Signed-off-by: Li Liang <address@hidden>
> ---
>  qemu-char.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/qemu-char.c b/qemu-char.c index 55e372c..55cdded 100644
> --- a/qemu-char.c
> +++ b/qemu-char.c
> @@ -1103,9 +1103,11 @@ static gboolean pty_chr_timer(gpointer opaque)
>      s->timer_tag = 0;
>      if (!s->connected) {
>          /* Next poll ... */
> +        qemu_mutex_unlock(&chr->chr_write_lock);
>          pty_chr_update_read_handler_locked(chr);
> +    } else {
> +        qemu_mutex_unlock(&chr->chr_write_lock);
>      }
> -    qemu_mutex_unlock(&chr->chr_write_lock);
>      return FALSE;
>  }

This is clearly not the correct fix -- it is now calling
pty_chr_update_read_handler_locked() without holding the chr_write_lock mutex, 
which is not permitted.

thanks
-- PMM