Re: [Qemu-devel] [PATCH] vl: Adjust the place of processing '-mon'

[Markus Armbruster <armbru@redhat.com>]

Peter Maydell <peter.maydell@linaro.org> writes:

> On 5 March 2015 at 02:02, Markus Armbruster <armbru@redhat.com> wrote:
>> zhanghailiang <zhang.zhanghailiang@huawei.com> writes:
>>
>>> From: Xiangyou Xie <xiexiangyou@huawei.com>
>>>
>>> If VM is configured with large size of hugepage, when startup,
>>> it will consume lots of time to zero the hugepage memory in the function
>>> 'os_mem_prealloc'.
>>> Libvirtd will wait 30 seconds for qemu to establish the monitor,
>>> If the timeout triggers, libvirtd will send TERM signal to kill qemu.
>>>
>>> To solve the problem, adjust the processing of '-mon' to the ahead of '-object'.
>>>
>>> Signed-off-by: Xiangyou Xie <xiexiangyou@huawei.com>
>>> Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
>>> ---
>>>  vl.c | 8 ++++----
>>>  1 file changed, 4 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/vl.c b/vl.c
>>> index 86bdce0..d0c03fe 100644
>>> --- a/vl.c
>>> +++ b/vl.c
>>> @@ -4000,6 +4000,10 @@ int main(int argc, char **argv, char **envp)
>>>          exit(0);
>>>      }
>>>
>>> +    if (qemu_opts_foreach(qemu_find_opts("mon"), mon_init_func, NULL, 1) != 0) {
>>> +        exit(1);
>>> +    }
>>> +
>>>      if (qemu_opts_foreach(qemu_find_opts("object"),
>>>                            object_create, NULL, 0) != 0) {
>>>          exit(1);
>>> @@ -4154,10 +4158,6 @@ int main(int argc, char **argv, char **envp)
>>>
>>>      parse_numa_opts();
>>>
>>> -    if (qemu_opts_foreach(qemu_find_opts("mon"), mon_init_func, NULL, 1) != 0) {
>>> -        exit(1);
>>> -    }
>>> -
>>>      if (foreach_device_config(DEV_SERIAL, serial_parse) < 0)
>>>          exit(1);
>>>      if (foreach_device_config(DEV_PARALLEL, parallel_parse) < 0)
>>
>>
>> Errors after monitor initialization look ugly when a monitor is on
>> stdio:
>>
>>     $ qemu -nodefaults -monitor stdio -vga xxx
>>     QEMU 2.2.50 monitor - type 'help' for more information
>>     (qemu) Unknown vga type: xxx
>>     $
>>
>> This patch initializes monitors earlier, thus makes more errors look
>> ugly.  Do we care?
>
> Yeah, this doesn't seem great. Surely the actual problem here
> is that we're doing something that takes 30 seconds to initialise
> as part of our command line option parsing phase ?

Yep.  Quoting myself:

    Our startup is a big happy ad hoc mess.  A more organizes [sic]
    program would read and check configuration first (quick, can fail),
    then allocate resources (quick, can fail), then initialize (somewhat
    slow, failure unlikely).

Reorganizing everything to work like that is a huge task.

Until we get that (if ever!), all we can do is delay the most
egregiously slow initializations until after configuration checking and
monitor creation.