Re: Fw: [PATCH] ia64: race flushing icache in do_no_page path
From: Nick Piggin
Date: Fri Apr 27 2007 - 22:01:02 EST
Rohit Seth wrote:
On Fri, 2007-04-27 at 21:55 +1000, Nick Piggin wrote:
That's the theory. However, I'd still like to know how the arch code can
make the assertion that icache is known to be at all times other than at
the time of a fault?
Kernel needs to only worry about the updates that it does. So, if
kernel is writing into a page that is getting marked with execute
permission then it will need to make sure that caches are coherent.
ia64 Kernel keeps track of whether it has done any write operation on a
page or not using PG_arch_1. And accordingly flushes icaches.
It flushes icache at fault time, I know. What I don't know is why we
leave them to drift out of sync afterwards.
Ie. what if an operation which causes incoherency is carried out _after_
an executable mapping is installed for that page.
You mean by user space? If so, then it is user space responsibility to
do the appropriate operations (like flush icache in this case).
No, I mean places that set PG_arch_1. flush_dcache_page. This can
happen for mapped pages in write, splice, install_arg_page looks
questionable, direct IO...
Actually there are various windows where mapped pages can be !uptodate,
so there is technically most of the filesystem code as well, but I'm
trying to stamp those out, so let's ignore that for now.
What if you were to say remove all the PG_arch_1 code, and do something
really simple like flush icache in flush_dcache_page? Would performance
SUSE Labs, Novell Inc.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/