Re: [PATCH] libdwfl: Use process_vm_readv when available.

[Mark Wielaard <mark at klomp dot org>]

On Wed, Mar 21, 2018 at 02:28:48AM +0300, Dmitry V. Levin wrote:
> On Sun, Mar 18, 2018 at 01:43:23AM +0100, Mark Wielaard wrote:
> > +  /* Let the ptrace fallback deal with the corner case of the address
> > +     possibly crossing a page boundery.  */
> > +  if ((addr & ((Dwarf_Addr)__LIBDWFL_REMOTE_MEM_CACHE_SIZE - 1))
> > +      > (Dwarf_Addr)__LIBDWFL_REMOTE_MEM_CACHE_SIZE - sizeof (unsigned long))
> 
> It looks odd that the variable that is going to be assigned has type
> Dwarf_Word, while the size being checked has type unsigned long.
> Shouldn't it be sizeof(*result) instead?
> 
> > +      d = &mem_cache->buf[addr - mem_cache->addr];
> > +      *result = *(unsigned long *) d;
> 
> Likewise, shouldn't it be memcpy(result, d, sizeof(*result)) instead?

That is indeed not immediately clear. I'll add some documentation.
Although the functions do use Dwarf_Word (which is always 64bits)
they actually return the result of an unsigned long/address. This
is true for both the pid based and core based memory read functions.
I am not completely sure if this was originally deliberate, or if
this was the accidental result of the ptrace interface returning a
long (target word) for PTRACE_PEEKDATA.

Thanks for reviewing.

Cheers,

Mark