linux-kernel - Re: [PATCH] ARM: convert max_pfn and max_low

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: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:	Thu, 13 Jun 2013 11:30:20 -0700
From:	Colin Cross <ccross@...roid.com>
To:	Russell King - ARM Linux <linux@....linux.org.uk>
Cc:	"linux-arm-kernel@...ts.infradead.org" 
	<linux-arm-kernel@...ts.infradead.org>,
	open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] ARM: convert max_pfn and max_low_pfn to be relative to PFN0

On Thu, Jun 13, 2013 at 7:37 AM, Russell King - ARM Linux
<linux@....linux.org.uk> wrote:
> On Wed, Jun 12, 2013 at 07:13:23PM -0700, Colin Cross wrote:
>> >From code inspection, I believe this will also improve block device
>> performance where the bounce limit was set to BLK_BOUNCE_HIGH, which
>> was bouncing unnecessarily for the top PHYS_PFN_OFFSET pages of low
>> memory.
>
> This has the potential to break platforms.  The problem is the duality
> of the dma_mask - is it a mask of the bits which the device can drive,
> or a PFN limit.  The block layer interprets it as a PFN limit, because
> of course everywhere starts their memory at physical address zero.

I've never come across a device with dma_mask set to anything but
0xFFFFFFFF, and a quick search didn't find me any good examples.
dma_mask set to the mask of bits the device can drive seems logical,
and it doesn't seem hard to fix the block layer (and the few other
users of max_pfn) to use min_low_pfn << PAGE_SHIFT + dma_mask.

> This gets into a world of pain if you have any of these conditions:
> (a) RAM not starting at physical address zero
The device I tested on has RAM at 0x40000000, which is what caused my problem.

> (b) Any translation between physical addresses and bus addresses
Is that just footbridge, integrator, and ks8695?  Those are the only
machines that define __virt_to_bus

> What we know is that the existing stuff works.  What we don't know is
> whether changing it will break anything which falls into the above
> two categories.

The existing stuff breaks a userspace API, /proc/kpagecount and
/proc/kpageflags.  It is currently impossible to read the page at
max_pfn on ARM.  It is possible to read the pages after max_pfn
because of an underflow in the kpagecount bounds check, which doesn't
cause problems because it checks pfn_valid on every pfn.  Just taking
out the bounds check against max_pfn in kpagecount and kpageflags
would also fix my problem, but it seems correct on everything but ARM.
--
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/