Re: [Qemu-devel] What's the intended use of log.h logging?

[Markus Armbruster]

Peter Maydell <address@hidden> writes:

> On 16 October 2015 at 15:17, Paolo Bonzini <address@hidden> wrote:
>> On 16/10/2015 15:36, Peter Maydell wrote:
>>>> We could add "-d trace:help" to list valid tracepoint names.
>>>
>>> I think people are mostly going to care about categories
>>> (eg "enable tracing for device X", or "enable tracing for
>>> all unimplemented features", or "enable a medium level of
>>> debug tracing for device Y".
>>
>> I hadn't thought about levels, but honestly when I use KVM tracing (which
>> is often measured in tens of megabytes per run) I don't care.
>
> In a lot of cases, especially with the TCG logging, not enabling
> voluminous tracing is really important because if you turn it all
> on then the system is way too slow (and can behave differently
> as a result), and generates gigabytes of trace output. (-d exec
> and -d cpu will do this, for instance.)

This is at least as much an argument for use of tracing as against it.
Tracing is a lot more flexible than log.h, and with the right backend,
it's much more efficient, too.

If the appropriate trace patterns turn out to be too hard to remember,
we can provide canned trace patterns with names that are easy to
remember.

-d could become sugar for a suitable trace patterns.