On Wed, Oct 21, 2015 at 01:22:22PM +0100, Peter Maydell wrote:
> On 21 October 2015 at 11:41, Stefan Hajnoczi <stefa...@redhat.com> wrote:
> > DPRINTF() falls under both #1 and #2. Care needs to be taken to
> > separate the logging from the tracing when converting DPRINTF().
> >
> > Side note: we don't have effect "once only" or rate-limiting built into
> > error_report(). Guest-triggerable errors or warnings should be limited
> > to prevent log flooding.
> >
> > It would be nice to unify TCG's "qemu log" and error_report() into a
> > single logging mechanism that is thread-safe and has rate-limiting.
>
> As long as you can disable the rate-limiting: for almost all
> of the use cases for the qemu-log code you really don't want
> to have it drop logging messages.
In the Linux kernel there are a few primitives:
printk() - plain old printf
printk_once() - uses a static bool for once-only semantics,
modulo races between CPUs
printk_ratelimited() - time-based rate-limiting
(And if you need to configure the ratelimit threshold, you can use the
underlying __ratelimit() function.)
The same could work for QEMU. Existing "qemu log" stuff is mostly
printk(). Any messages that are not errors or warnings don't need
rate-limiting because they shouldn't be enabled by default (and when you
do want them for troubleshooting, then rate-limiting gets in the way).
Stefan
