Re: [patch] threading fix, tid-2.5.47-A3

[Luca Barbieri <ldb@ldb.ods.org>]

[-- Attachment #1: Type: text/plain, Size: 1206 bytes --]

> It's all encapsulated in the libpthread which is used.  No apps need to
> be recompiled so it is OK to make this incompatible change.
My point was that aesthetical reasons do not justify breaking anything
(aka forcing people to figure out what happened and spend time
recompiling).
Anyway, I'll need to recompile anyway, so I don't really care.

> I cannot see any reasonable way out if any of the put_user calls fail?
> Do you want the clone() call to fail if the parent's receiving address
> is wrong?  You'd have to go and kill the child again since it's already
> created.
> 
> Instead let the user initialize the memory location the clone call is
> supposed to write in with zero.  if the value didn't change the
> user-level code can detect the error and handle appropriately.  So,
> ignore the put_user errors.  Maybe say so explicitly and add (void) in
> front.
I proposed to fail in put_user for the parent tid simply because the old
code did it.
I'm not completely sure whether we should fail or not, but if put_user
fails something bad is happening, so it may be better to signal the
error rather than just continuing since it can be done easily (for the
parent tid).


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]