This is the mail archive of the libc-ports@sources.redhat.com mailing list for the libc-ports project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] powerp-cpu add-on for --with-cpu


You cannot add hp-timing in this way and do what you want.  While it is not
part of the public ABI, hp_timing_t is part of the internal ABI between
ld.so, libc.so, libpthread.so, and librt.so.  You can build variants with
different definitions all you like, but you must use the same definition in
ld.so that you use in libc.so et al.  Your plan is to build variant
libraries and select one at runtime via hwcap.  In this plan there is only
one ld.so, and all your variant builds of libc.so et al must exactly match
in all internal data structures the build that ld.so comes from.  So, at
the very least you must define hp_timing_t to the same size across the
board for powerpc32.

If you cannot safely attempt a time sample in your least common denominator
powerpc32 configuration, then there's really not much point.  It's pretty
much only used for the low-overhead timings in ld.so startup.  Those are
very early, before it can do much of anything.  If it cannot take a sample
there, then the variables won't be set up for the CPU clocks to make
worthwhile use of it later.  And for that it's only used on old kernels
that don't support the real CPU clocks that actually have correct semantics.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]