[PATCH libdrm 02/24] radeon: remove empty function declarations

Jerome Glisse j.glisse at gmail.com
Wed Apr 1 14:24:45 PDT 2015


On Wed, Apr 01, 2015 at 09:34:05PM +0100, Emil Velikov wrote:
> On 1 April 2015 at 18:30, Jerome Glisse <j.glisse at gmail.com> wrote:
> > On Wed, Apr 01, 2015 at 05:15:13PM +0100, Emil Velikov wrote:
> >> Missing definition and unused since their introduction.
> >>
> >> Cc: Jerome Glisse <jglisse at redhat.com>
> >> Signed-off-by: Emil Velikov <emil.l.velikov at gmail.com>
> >
> > NAK
> >
> > I use all this in tools to debug lockup. Best course of action is to
> > exclude bof.h from being distributed. My tools static link and i just
> > point them to libdrm git tree.
> >
> Did not notice any mention of such out-of-tree tools in the commit
> that introduced these functions, so I've naively assumed that they are
> unused. Sorry about that. Do you mind if I add a note about it, or
> alternatively will you be ok with pushing your tool to libdrm ? The
> Intel team already have a test_decode tool in, which is similar in
> nature.

It would need cleanup before this can happen, saddly my schedule is kind
of full for foreseeable future. So i do not want to commit to do such thing.
But i definitly use the bof feature, last time was a month or so ago to
debug something. It have been very usefull to me in the past and i expect
for as long as the radeon ddx stays releavant it will be in the future.

But with the advance of the modesetting ddx and glamor, the tracing that
does exist in mesa will be as easy as the bof tracing. If not easier. So
i am not sure of the value there is into putting effort into this.

This kind of feature is really usefull when debugging lockup, at least
this allow me to bisect offend command stream to pin point the last
dword before lockup and thus get a clue about what kind of cmd is the
root cause. Dunno how others dev do such thing.

> I'm not sure that your suggestion will work - one cannot exclude bof.h
> (and bof.c) from the distribution as it's used by radeon_cs_gem.c.
> Annotating the symbols as hidden/private should work for everyone. How
> does that sound ?

I do not see how this is an issue, symbol needed by radeon_cs_gem can
be hidden from other and thus there is no point into shipping bof.h
Really no symbol need to be exported, iirc i tend to ln -s the bof
files in my tools or simply cp the lastest version from libdrm into
a local copy but i still need bof.h to have all symbol listed.

Cheers,
Jérôme

> 
> ...
> >> -extern int bof_file_flush(bof_t *root);
> >> -extern bof_t *bof_file_new(const char *filename);
> >> -extern int bof_object_dump(bof_t *object, const char *filename);
> >> -
> Can you please elaborate how you are using these three, do you have
> them implemented outside of libdrm as well ?
> 
> Cheers,
> Emil


More information about the dri-devel mailing list