Subject: Re: cleanup of uvm_fault*() interfaces
To: Matt Thomas <firstname.lastname@example.org>
From: Matthias Drochner <M.Drochner@fz-juelich.de>
Date: 03/02/2006 18:36:07
> The changes just bloat the kernel a bit
Yep, the tiny wrapper -- if this is issue it could
be easily replaced by a cpp macro.
> without significantly simplifying things
The idea is to keep interfaces private which are not really
needed outside uvm.
There is no useful "fault type" information to be represented
outside uvm, so there is no need to expose a type for it. Its
current use, just for flags to deal with "simulated faults",
is a misnomer anyway.
This gives the freedom to rearrange things in uvm without
the need to care about use in md code.
I'm not seriously working on such a rearrangement yet but I see
eg room for improvement in the interface of uvm_fault_wire().
(Wiring page by page, throwing away the vm_map_entry information,
looks like a waste for me, in particular for userland buffers
where submaps are not used. We are doing some data acquisition
which does big DMA to userland, so this matters.)