[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH v2 2/6] accel: accel_available() function

From: Claudio Fontana
Subject: Re: [PATCH v2 2/6] accel: accel_available() function
Date: Fri, 27 Nov 2020 15:58:13 +0100
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.12.0

On 11/27/20 3:45 PM, Markus Armbruster wrote:
> Claudio Fontana <cfontana@suse.de> writes:
>> On 11/26/20 10:48 PM, Eduardo Habkost wrote:
>>> On Thu, Nov 26, 2020 at 10:06:03PM +0100, Claudio Fontana wrote:
>>>> On 11/26/20 3:25 PM, Paolo Bonzini wrote:
>>>>> On 26/11/20 15:13, Claudio Fontana wrote:
>>>>>> One option I see is simply to document the behavior where
>>>>>> accel_available() is declared in accel.h (ie do not use in fast
>>>>>> path), as well as in accel_find() actually, so that both accel_find()
>>>>>> and accel_available() are avoided in fast path and avoid being called
>>>>>> frequently at runtime.
>>>>>> Another option could be to remove the allocation completely, and use
>>>>>> for example accel_find(ACCEL_CLASS_NAME("tcg")), or another option
>>>>>> again would be to remove the allocation and use either a fixed buffer
>>>>>> + snprintf, or alloca -like builtin code to use the stack, ...
>>>>>> Not a big deal, but with a general utility and short name like
>>>>>> accel_available(name) it might be tempting to use this more in the
>>>>>> future?
>>>>> I think it's just that the usecase is not that common.  "Is this 
>>>>> accelerator compiled in the binary" is not something you need after 
>>>>> startup (or if querying the monitor).
>>>>> Paolo
>>>> A script that repeatedly uses the QMP interface to query for
>>>> the status could generate fragmentation this way I think.
>>> Is this a problem?  Today, execution of a "query-kvm" command
>>> calls g_malloc() 37 times.
>> Not ideal in my view, but not the end of the world either.
> QMP's appetite for malloc is roughly comparable to a pig's for truffles.


Btw, do we have limits on the maximum size of these objects? I mean, a single 
QMP command,
a single QEMU object type name, etc?

In this case we could do some overall improvement there, and might even avoid 
some problems down the road..



reply via email to

[Prev in Thread] Current Thread [Next in Thread]