On 5 December 2014 at 17:27, Russell King - ARM Linux
[off-list ref] wrote:

On Fri, Dec 05, 2014 at 05:07:45PM +0000, Catalin Marinas wrote:

quoted

On Fri, Dec 05, 2014 at 12:05:06PM +0000, Will Deacon wrote:

quoted

Care to submit this as a proper patch? We should at least fix Peter's issue
before doing things like extending headers, which won't work for older
kernels anyway.

Quick fix is the revert of the whole patch, together with removing
PAGE_ALIGN(end) in poison_init_mem() on arm32. If Russell is ok with
this patch, we can take it via the arm64 tree, otherwise I'll send you a
partial revert only for the arm64 part.

Not really.  Let's look at the history.

For years, we've been poisoning memory, page aligning the end pointer.
This has never been an issue.

Depends what you mean by "never been an issue". I had to change
QEMU (commit 98ed805c, January 2013) for 32-bit ARM back when the
kernel started trashing the tail end of the page after the initrd
with the poisoning, to 4K-align the dtb so it didn't share a page
with the initrd-tail. That nobody else complained suggests that most
bootloaders don't in practice overlap the two, though (ie that
QEMU is an outlier in how it chooses to arrange things in memory).

I should probably have reported the breakage at the time, but
I took the pragmatic (lazy?) approach of just changing our
bootloader code.

thanks
-- PMM