lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite for Android: free password hash cracker in your pocket
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <1183984629.5961.68.camel@localhost.localdomain>
Date:	Mon, 09 Jul 2007 22:37:08 +1000
From:	Benjamin Herrenschmidt <benh@...nel.crashing.org>
To:	Nick Piggin <nickpiggin@...oo.com.au>
Cc:	linux-mm@...ck.org,
	Linux Kernel list <linux-kernel@...r.kernel.org>
Subject: Re: [RFC/PATCH] Use mmu_gather for fork() instead of flush_tlb_mm()


> To elaborate on this one... I realise for this one that in the kernel
> where this is currently used everything is non-preemptible anyway
> because of the ptl. And I also realise that -rt kernel issues don't
> really have a bearing on mainline kernel.. but the generic
> implementation of this API is fundamentally used to operate on a
> per-cpu data structure that is only required when tearing down page
> tables. That makes this necessarily non-preemptible.
> 
> Which shows that it adds more restrictions that may not otherwise be
> required.

Yes, it's a bit annoying but not necessarily that bad. In fact, we don't
have to make it non-preemptible, we did it because it was easier that way
I strongly suspect. In fact, the batch could actually be attached to the
mm rather than the CPU for that matter, no ? Or is there a fudamental
reason I'm not seeing why it -has- to be per-cpu ?

> > has to look in there and touch the cacheline. You're also having to
> > do more work when unlocking/relocking the ptl etc.
> > 
> > 
> >> I really think it's the right API
> 
> OK, the *form* of the API is fine, I have no arguments. I just don't
> know why you have to reuse the same thing. If you provided a new set of
> names then you can trivially do a generic implementation which compiles
> to exactly the same code for all architectures right now. That seems to
> me like the right way to go...

But that means two different APIs for almost the same thing. I'm trying
to clean up the mess, not add more :-) Beside, that "other" API would
have overall much of the same issues no ? Or do you want to have that
"other" API not actually provide a percpu "mmu_gather" type structure at
all in asm-generic (but basically just boil down to an empty inline for
creating the "other" batch and flush_tlb_mm() for finishing it with an
empty inline for "adding" a PTE to the list of invalidation targets ?)

Ben.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ