Re: Tests requiring MD hooks

[jean-Yves Migeon <jym%NetBSD.org@localhost>]

On Mon, 11 Jul 2011 19:46:33 +0400, Valeriy E. Ushakov wrote:
> But you do cast your mapped address to a function pointer.  Does that
> do the right thing for architectures with fat function pointers?

Untested; will see this evening what the resulting code does.

> You also don't seem to mention instruction/data-cache issues.  I 
> don't
> remember if mprotecting something +x makes any guarantees about 
> caches
> (cf. arm_sync_icache(2) for example).

OG does say a thing about this. I assume that mprotect(2) should flush 
them (especially icache upon removal of --x right), and count this as a 
bug if it does not.

Best way to see is through testing, although I see a couple of 
icache_sync in ARM pmap each time execute mappings are modified.

> > I can't do that simply through rtld, as I need to memcpy(2) the 
> > payload
> > and therefore have access to the beginning (which is not necessarily
> > trivial given certain architectures), and the end (never found a way 
> > to
> > properly label those without using assembly routines).
>
> You can always parse ELF yourself (unfortunately libelf is still not
> imported, it seems).

That's a possibility, yes. Although this would restrict me to ELF, and 
rule out all other binary formats.

--
Jean-Yves Migeon
jym%NetBSD.org@localhost