Re: [Qemu-devel] debug logging (was: Re: [PATCHv2 3/4] qxl: add debug_cs and cmdlog_cs)

[Alon Levy <alevy@redhat.com>]

On Mon, May 02, 2011 at 11:02:22AM +0200, Gerd Hoffmann wrote:
>   Hi,
> 
> >The bigger hack called "qxl_terse" I didn't even send :) ok, so now I get
> >to either keep this in my closet or think of how to do a minimal acceptable
> >qemu logging infrastructure that would let me register a logging handle and
> >use that to redirect to a chardev (they would all default to being muxed over
> >stdio?)
> >
> >QemuLogger *qemu_create_logger(const char *logger_id);
> >  - logger_id is used to match to the chardev given on the command line
> >  - need to prevent collision, so probably easier to have a logger_id be an int
> >  and have that looked up to a string in an automatically generated table?
> 
> What I have in mind is a simple dbg_print() function, integrated
> with qdev and simliar to what the linux kernel has, i.e. something
> like:
> 
> int dbg_print(DeviceState *qdev, int loglevel, const char *fmt, ...);
> 
> This can basically be used that as drop-in replacement for the
> fprintf(stderr, fmt, ...) style found in many drivers.
> 
> dbg_print would get the driver name via qdev->info->name, get the
> instance via qdev->parent_bus->info->get_dev_path(), get a
> timestamp, then create a standard prefix for all messages for easy
> grepping.  It would also check qdev->loglevel to figure whenever the
> message should be printed or not.  All devices get a property to set
> qdev->loglevel, so all logging can be configured at runtime per
> device.
> 
> Messages go to stderr by default.  A -debug switch could be added to
> route them another way.
> 
> Comments in the idea?

The thing with grepping is that it conflicts with using the monitor on stdout,
which is why I wanted to use a chardev. And of course a chardev solves the
need for grep (at least to get a coarse level - of course you still need to
grep later).

dbg_print takes care of making it standard to have a loglevel and prefix, sounds
good, but I'd still like to know if it is acceptable to also redirect with -debug,
I guess reusing the DeviceState then, instead of my added struct, so just letting
DeviceState.debug_chardev == NULL by default, and settable with
-debug <devname>,id=<chardev_id>

> 
> cheers,
>   Gerd
> 
>