Re: [PATCH 28/44] Add translate_signal() helper

[Richard Weinberger <richard@nod.at>]

Am 04.03.2014 19:41, schrieb Oleg Nesterov:
> On 03/03, Richard Weinberger wrote:
>>
>> +static inline int translate_signal(int sig)
>> +{
>> +	struct thread_info *info = current_thread_info();
>> +
>> +	if (info->exec_domain
> 
> Afaics it can't be NULL.

Good point. Will change.

>> && info->exec_domain->signal_invmap && sig < 32)
>> +		return info->exec_domain->signal_invmap[sig];
>> +
> 
> And why default_exec_domain has .signal_invmap = ident_map ?
> translate_signal-like code should check ->signal_invmap != NULL
> anyway.
> 
> Off topic, but can't we kill exec_domain->signal_map ? it seems that
> it is unused.
> 
> Hmm. And who actually uses exec_domain->handler() ?? asm-offsets.c,
> but I do not see any usage of TI_EXEC_DOMAIN.
> 
> In short: I agree with this change but I am totally confused. To the
> point, I am not sure I even understand ->signal_invmap. I mean, it
> seems that only arthur_exec_domain has a non-ident map. Why, say,
> x86 does signr_convert() ? and only if CONFIG_X86_32? Is there any
> external module which does register_exec_domain(strange_invmap) ?

I remember a discussion with Al where he noted that register_exec_domain()
is currently only used by out of tree horrors. B-)

I'm sure we can rip it out of most archs. They seem to have it only because
they've copy&pasted it from i386.

Thanks,
//richard