Emmanuel Dreyfus a écrit:

>port-mips/16154 is still open, despite the fact we went through a
>relative in-depth anlysis of the problem. Any mips expert to peek at
>this one?
>  
>
I've just read PR/16154, and I have some thinking about it:

Firstly I have doubts about the correctness of the way the case is 
handled: it seems a little odd to me to restart a faulting instruction, 
especially if the fault was explicitly ignored. Shouldn't the kernel 
skip that instruction before restarting the process, since the process 
itself asked the kernel to ignore the fault ?

As for the problem reported, interrupts are really the key: the kernel 
should give soft interrupts a chance to trigger before returning to 
userland.