ppc64/cell: local TLB flush with active SPEs

[Arnd Bergmann]

On Dunnersdag 13 Oktober 2005 00:09, Mark Nutter wrote:
> As long as we are thinking about a proper solution, the whole 
> mm->cpu_vm_mask thing is broken, at least as a selector for local -vs- 
> global TLBIE.  The problem, as I see it, is that memory regions can shared 
> among processes (via mmap/shmat), with each task bound to different 
> processors.  If we are to continue using a cpumask as selector for TLBIE, 
> then we really need a vma->cpu_vma_mask. 
>  
No, because different tasks mapping the same address_space result in distinct
virtual (though not necessarily effective) addresses, so the TLB entries
are never shared across processes. A TLB entry on ppc64 always maps between
the (mm_struct,effective address) tuple and the real address and is local to
one CPU or SPU.
If we want to be clever, we could optimize the case where an mm_struct has
been used on exactly on CPU (the currently running one) and at least one
SPU. In that case, doing a TLBIEL plus a separate flush of each of the
MFCs that were used (by writing to their TLB_Invalidate_Entry registers)
is probably better than a global TLBIE.

	Arnd <><