David S. Miller wrote:

>On Fri, 19 Mar 2004 16:42:03 -0600
>Corey Minyard <minyard@acm.org> wrote:
>
>  
>
>>BTW, here is a new patch that covers that problem.
>>    
>>
>
>This still doesn't cut it, next the code in handle_signal() performs operations
>on ka->sa.sa_mask if SA_{NOMASK,NODEFER} is not set,
>then a recalc_sigpending() is made.
>
>I think we really need to pull the locking up a level or something like that.
>  
>
I saw it using the sa_sigmask, but not modifying it.

But you still may be right, pulling up the lock a level may be more 
optimal.  It just may mess with latency.

Here's another patch with the SEGV problems handled and the ka's renamed 
to ka_copy.

-Corey