Re: [RFC] TIF_NOTIFY_RESUME, arch/*/*/*signal*.c and all such

[Martin Schwidefsky <schwidefsky@de.ibm.com>]

On Thu, 26 Apr 2012 15:52:55 +0200
Oleg Nesterov <oleg@redhat.com> wrote:

> On 04/26, Martin Schwidefsky wrote:
> > You need to be careful with inferior calls there. gdb likes to play games
> > with the registers inside the get_signal_to_deliver call, it wants to be
> > able to jump out of an interrupted system call, do its inferior call in
> > the debugee and then return to the interrupted system call.
> 
> Ah.
> 
> > You would have to to read, modify & restore the NEED_RESTART flag in gdb
> > over an inferior call.
> 
> I am not sure, but perhaps this is not really needed...
> 
> But at least this means that "if (we have any business doing restarts)"
> above is meaningless before get_signal_to_deliver().
> 
> 
> And I am confused, off-topic question... How it is possible to
> "then return to the interrupted system call" if that system call
> returned -ERESTART_RESTARTBLOCK but the inferior call in turn
> does the system call which changes restart_block->fn/etc ?

Returning from an inferior gdb call to a system call that needs to be
restarted with -ERESTART_RESTARTBLOCK is broken on all architectures
as far as I can tell. And it would be hard to fix it, basically you
have to save and restore the restart block.

-- 
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.