Re: [Xen-devel] [PATCH] gcov: Support gcc 4.7

[George Dunlap <george.dunlap@xxxxxxxxxxxxx>]

On 06/18/2013 03:32 PM, Frediano Ziglio wrote:
> On Mon, 2013-06-17 at 11:49 +0100, Ian Campbell wrote:
> > On Mon, 2013-06-17 at 11:46 +0100, Frediano Ziglio wrote:
> > > Perhaps a simple easy solution would be to allow privileged domains to
> > > read/write arbitrary memory from Xen (write to reset counters).
> > "Arbitrary memory" would be a very hard sell.
> >
> > If all these counters could be constrained to pages which contains these
> > counters and nothing else then that might be a possibility to consider.
> >
> > Ian.
> It depends on how this would be implemented. A future option of gcc
> could help at least having everything in contiguous memory regions.
>
> The fact that these information are generated by the compiler does not
> help that much. Every version could change the format "slightly" but
> enough to make Xen core when such information are requested!
>
> Actually they changed format for these version of gcc:
> - 3.3
> - 3.4
> - 4.7
>
> And honestly I still don't understand clearly why they changed for gcc
> 4.7! Probably not for a C feature but to support C++ templates and
> comdat. A "smart" preprocessor sit in the middle of gcc and gas that
> process assembly generated, check for structures and put everything in
> different sections could really help... but is not that easy project!
> Another similar way would be to parse the xen-syms output file and
> generate needed blob/code. Quite interesting challenge (I did more
> terrible things then this) but I don't thing that Citrix would agree.
> Possibly a safe thing to do would be in the code to check that compiler
> is really gcc and that we can support the version that is compiling,
> just to make sure. But is also true that this feature is a debug one
> disabled on production so I don't know how worth would be.
>
> Not speaking of clang support, currently quite hard (the structures are
> less documented, the code hooks destructors path which is not safe to
> call from Xen code which is not expected to terminate).
Doesn't Linux have gcov support via lcov?  Do they have to deal with 
this stuff as well, or is there a generic way to deal with it?
 -George

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel