A more extreme approach to this sort of thing would be to ship the
OLPC platform, as much as possible, as one massive dynamic library.
This would (done properly) save on startup time/power as internal
relocations within the libplatform.so could be precomputed without the
need for things like prelink, and of course yes you save on VMA
regions too.

For even more savings a dynamic/static hybrid could be used ...
basically an ELF DSO that the isn't compileas a PIC, so the linker
knows will it be at a certain range of addresses and direct jumps can
be emitted rather than indirect jumps via the global offset table.
This is getting into games console levels of efficiency though ....
obviously the danger is that the platform gets bigger and bigger with
each release until it no longer fits in the allocated address space :)

thanks -mike

--
olpc-software mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/olpc-software

Reply via email to