On 12/23/2015 12:15 PM, Russell King - ARM Linux wrote:

On Wed, Dec 02, 2015 at 12:27:25PM -0800, Kees Cook wrote:

quoted

The use of CONFIG_DEBUG_RODATA is generally seen as an essential part of
kernel self-protection:
http://www.openwall.com/lists/kernel-hardening/2015/11/30/13
Additionally, its name has grown to mean things beyond just rodata. To
get ARM closer to this, we ought to rearrange the names of the configs
that control how the kernel protects its memory. What was called
CONFIG_ARM_KERNMEM_PERMS is really doing the work that other architectures
call CONFIG_DEBUG_RODATA.

Kees,

There is a subtle problem with the kernel memory permissions and the
DMA debugging.

DMA debugging checks whether we're trying to do DMA from the kernel
mappings (text, rodata, data etc).  It checks _text.._etext.  However,
when RODATA is enabled, we have about one section between _text and
_stext which are freed into the kernel's page pool, and then become
available for allocation and use for DMA.

This then causes the DMA debugging sanity check to fire.

So, I think I'll revert this change for the time being as it seems to
be causing many people problems, and having this enabled is creating
extra warnings when kernel debug options are enabled along with it.

Sorry.

in include/asm-generic/sections.h:

/*
  * Usage guidelines:
  * _text, _data: architecture specific, don't use them in arch-independent code
  * [_stext, _etext]: contains .text.* sections, may also contain .rodata.*
  *                   and/or .init.* sections


So based on that comment it seems like the dma-debug should be checking for
_stext not _text since only _stext is guaranteed across all architectures.
I'll submit a patch to dma-debug.c if this seems appropriate or if you
haven't done so already.

Thanks,
Laura