Re: [Sbcl-devel] [PATCH] Improved os_context_t / void_context handling

[Thiemo S. <th...@ne...>]

Christophe Rhodes wrote:
[snip]
> So, to summarize, I think the right fix to all this is
> 
> * forget about getting a void * back from an os_context *; it should
>   never be necessary;
> 
> * make the various pseudo-handlers called from sigsegv_handler
>   explicitly take os_context_t * third arguments rather than void *;
> 
> * divide the thread-specific handler functions into two parts: one, a
>   kernel-facing handler function such as sig_stop_for_gc_handler,
>   which is responsible for getting a context from a void_context, and
>   a lisp-facing function sig_stop_for_gc which is responsible for
>   performing the action given an os_context_t, and as such can
>   therefore be used as a deferred handler.
> 
> Then, when we hit the trap at the end of a pseudo-atomic section, the
> kernel-facing handler gets us an os_context_t from the handler
> arguments, and calls the handler function that we've deferred to with
> that os_context_t.
> 
> Does that make sense?  Sorry for not reviewing your work in a terribly
> timely fashion. :-(

Yes, it does make sense, but with the current patches pending and Gabor's
changes around I'll postpone that.


Thiemo