Machine-independant handling of registers set with ptrace is tough.

[Etienne Robillard]

Dear list,

I'm actively into the process of hatching at the
Fiasco-UX source-code, and have found the ptrace(2)
interface to be quite entertaining.

Firstly, the naming convention (ptrace macros) differs
from Linux to FreeBSD. However, its relatively trivial
to fix (thx to Alexander Julliard), by adding
something like:

#ifndef PTRACE_PEEKTEXT
#define PTRACE_PEEKTEXT PT_READ_D 
#endif 

Secondly, a lot of things is done in `Usermode.cpp' 
in order to manipulate user-visible register set.
In particular, I have to trust the effectiveness and
the reliability of the following constructor:

// save it to <machine/reg.h>
struct user_regs_struct regs;
regs.xfs = context->uc_mcontext.gregs[REG_FS] 

In comparison, the FreeBSD kernel implements 
the ptrace(2) proc_[read/write]_register(proc,regs)   
for doing this kind of stuff, but its not even 
close to be a member of uc_mcontext.

All in all, it would be nice to know how (and why)
this is done on different arch, and finally if the
uc_mcontext.gregs constructor is the sole
alternative to registers translation.

Cheers,
Etienne
  

______________________________________________________________________ 
Post your free ad now! http://personals.yahoo.ca