Re: determining the stack bounds

[Bruno Haible]

Eric Blake wrote:
> On Solaris 8, uc_stack is (IMHO properly) pointing to the primary stack,
> even though the handler is executing on the alternate stack.

That's accidental, IMO.

The only documented use of uc_stack in POSIX:2001 is in the description of
'makecontext', as an input parameter. Its meaning is not documented in the
use of 'getcontext' or when a signal handler is invoked. Therefore, if the
next version of POSIX does not have 'makecontext', IMO they should also
remove the uc_stack member.

Also, you know that the main context has a growing stack; it's managed by
the kernel as something dynamic (taking into account the RLIMITs that will
be in effect), not as something static. Such a dynamic piece of stack cannot
be adequately represented by a stack_t. Solaris escapes the problem by
returning to you a snapshot of the stack VMA bounds; Linux escapes the
problem by returning the alternate stack bounds (which _is_ a stack_t).

> Again, anyone know how to report something like that?

First, gather enough arguments that it is really a bug in Linux (which I
dispute). Then, write a kernel patch that fixes the bug, relative to the newest
development kernels. Third, submit it to LKML, and be prepared to either a
heated discussion or complete silence :-)

Bruno