Re: [draft patch] <unavailable> unwinder for btrace [Re: [rfc 3/5] record: make it build again]

[Jan Kratochvil <jan dot kratochvil at redhat dot com>]

On Thu, 28 Mar 2013 08:44:12 +0100, Metzger, Markus T wrote:
> Thanks, I'll try that.  Although, in the end, it's really the same as if we made
> other frames prologue cache visible.

Yes.  You can make a separate patch to make it visible but one htab_t may be
easier.  With public cache pointer someone could misues it inappropriately
etc.


> > > > Also what's the lifetime of a frame_info and frame_id object?
> > > > When the branch trace is cleared, any pointers to it will become
> > > > stale.
> > 
> > There is frame_unwind->dealloc_cache, any reinit_frame_cache() call inside GDB
> > will clear the prologue cache which is very common.
> > 
> > I see now btrace_thread_info->btrace may change more often - such as during
> > each "info record" command.  So call reinit_frame_cache() in the cases btrace
> > cache may get rebuilt.
> 
> The data structure should only change when there is new trace, which requires
> the target to continue.  "info record" should, like any other record-btrace command,
> fetch the new trace once and then operate on the cached trace data.

In non-stop mode I belive there will be new btrace info on each "info record"
command, won't be?  I have not tried it but it seems so to me.


> Is there a guarantee that frame_info and frame_id objects are destroyed
> when the target resumes?  Or could I trigger their destruction from within
> btrace_clear?

"trigger frame_info and frame_id destruction" == reinit_frame_cache().

Accessing frame_info after reinit_frame_cache() is always a crash.

Accessing frame_id after reinit_frame_cache() is safe but one needs to be
prepared frame_find_by_id may return NULL if it is no longer available.

When you introduce new reinit_frame_cache() call one just needs to be careful
no caller holds that time a frame_info * pointer in a local variable. 
It would be a bug in such caller to call some non-trivial caller while holding
frame_info * but there were many such bugs in GDB.

I would not rely on any reinit_frame_cache() calls, calling
reinit_frame_cache() more times is zero-cost, I think you should call
reinit_frame_cache() from btrace_clear as you ask above.


Thanks,
Jan