Re: [PATCH] monitor: Fix slow reading

[Markus Armbruster]

Yury Kotov <address@hidden> writes:

> The monitor_can_read (as a callback of qemu_chr_fe_set_handlers)
> should return size of buffer which monitor_qmp_read or monitor_read
> can process.
> Currently, monitor_can_read returns 1 as a result of logical not.
> Thus, for each QMP command, len(QMD) iterations of the main loop
> are required to handle a command.
> In fact, these both functions can process any buffer size.
> So, return 1024 as a reasonable size which is enough to process
> the most QMP commands, but not too big to block the main loop for
> a long time.
>
> Signed-off-by: Yury Kotov <address@hidden>
> ---
>  monitor/monitor.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/monitor/monitor.c b/monitor/monitor.c
> index 12898b6448..cac3f39727 100644
> --- a/monitor/monitor.c
> +++ b/monitor/monitor.c
> @@ -50,6 +50,13 @@ typedef struct {
>      int64_t rate;       /* Minimum time (in ns) between two events */
>  } MonitorQAPIEventConf;
>  
> +/*
> + * The maximum buffer size which the monitor can process in one iteration
> + * of the main loop. We don't want to block the loop for a long time
> + * because of JSON parser, so use a reasonable value.
> + */
> +#define MONITOR_READ_LEN_MAX 1024
> +
>  /* Shared monitor I/O thread */
>  IOThread *mon_iothread;
>  
> @@ -498,7 +505,7 @@ int monitor_can_read(void *opaque)
>  {
>      Monitor *mon = opaque;
>  
> -    return !atomic_mb_read(&mon->suspend_cnt);
> +    return atomic_mb_read(&mon->suspend_cnt) ? 0 : MONITOR_READ_LEN_MAX;
>  }
>  
>  void monitor_list_append(Monitor *mon)

Prior attempt:
[PATCH 1/1] monitor: increase amount of data for monitor to read
Message-Id: <address@hidden>
https://lists.nongnu.org/archive/html/qemu-devel/2017-05/msg00206.html

Review concluded that it breaks HMP command migrate without -d.  QMP is
probably okay.  Sadly, no v2.

Next one:
Subject: [PATCH] monitor: increase amount of data for monitor to read
Message-Id: <address@hidden>
https://lists.nongnu.org/archive/html/qemu-devel/2019-06/msg01912.html

Same patch, with a second, suspicious-looking hunk thrown in.  I didn't
make the connection to the prior attempt back then.  I wrote "I think I
need to (re-)examine how QMP reads input, with special consideration to
its OOB feature."

This patch is a cleaner variation on the same theme.  Its ramifications
are as unobvious as ever.

I figure the HMP situation is unchanged: not safe, although we could
probably make it safe if we wanted to (Daniel sketched how).  My simpler
suggestion stands: separate f_can_read() callbacks for HMP and QMP
[PATCH 1], then change only the one for QMP [PATCH 2].

The QMP situation is also unchanged: we still need to think through how
this affects reading of QMP input, in particular OOB.