[PATCH v2 3/4] ARM: mm: truncate memory banks to fit in 4GB space for classic MMU

[David Brown]

On Wed, Apr 11, 2012 at 05:27:30PM +0100, Russell King - ARM Linux wrote:
> On Wed, Apr 11, 2012 at 04:52:32PM +0100, Russell King - ARM Linux wrote:
> > On Wed, Apr 11, 2012 at 10:44:22AM -0400, Nicolas Pitre wrote:
> > > On Wed, 11 Apr 2012, Will Deacon wrote:
> > > 
> > > > If a bank of memory spanning the 4GB boundary is added on a !CONFIG_LPAE
> > > > kernel then we will hang early during boot since the memory bank will
> > > > have wrapped around to zero.
> > > > 
> > > > This patch truncates memory banks for !LPAE configurations when the end
> > > > address is not representable in 32 bits.
> > > > 
> > > > Cc: Nicolas Pitre <nico at fluxnic.net>
> > > > Signed-off-by: Will Deacon <will.deacon at arm.com>
> > > 
> > > Acked-by: Nicolas Pitre <nico at linaro.org>
> > > 
> > > Now what if start = 1G and size = 5G? The size variable is an unsigned 
> > > long, meaning that right now the size might be truncated to 1G.
> > 
> > There's a solution to that which is quite easy to do: convert the bank
> > information to PFNs instead of addresses.  That will probably eliminate
> > some corner cases with partial pages which would be desirable too.
> 
> Of course, what prevents us doing that conversion sanely is all the
> shite platform code doing crap stuff like this:
> 
> arch/arm/mach-msm/board-halibut.c:      mi->bank[0].start = PHYS_OFFSET;
> arch/arm/mach-msm/board-halibut.c:      mi->bank[0].size = (101*1024*1024);
> 
> which I went through everything a few years ago and eliminated all this
> crap.  It's back now.  Sod it, we'll stick with the current 4GiB limited
> way as long as we have platform maintainers who do this kind of crappy
> hack.

I'm not sure there are even any working "halibut" targets (MSM7201
SURF).  It wasn't a generally available target.  I think the only
in-use fish target is trout (HTC Dream).  Although, it seems that
every time I think this, someone will speak up about one of these
targets.

> While here, I propose to delete these:
> 
> arch/arm/mach-msm/board-mahimahi.c:     mi->bank[0].start = PHYS_OFFSET;
> arch/arm/mach-msm/board-mahimahi.c:     mi->bank[0].node = PHYS_TO_NID(PHYS_OFFSET);
> arch/arm/mach-msm/board-mahimahi.c:     mi->bank[0].size = (219*1024*1024);
> arch/arm/mach-msm/board-mahimahi.c:     mi->bank[1].start = MSM_HIGHMEM_BASE;
> arch/arm/mach-msm/board-mahimahi.c:     mi->bank[1].node = PHYS_TO_NID(MSM_HIGHMEM_BASE);
> arch/arm/mach-msm/board-mahimahi.c:     mi->bank[1].size = MSM_HIGHMEM_SIZE;
> arch/arm/mach-msm/board-sapphire.c:     mi->bank[0].start = PHYS_OFFSET;
> arch/arm/mach-msm/board-sapphire.c:     mi->bank[0].node = PHYS_TO_NID(PHYS_OFFSET);
> arch/arm/mach-msm/board-sapphire.c:             mi->bank[0].size = (84*1024*1024);
> arch/arm/mach-msm/board-sapphire.c:             mi->bank[0].size = (101*1024*1024);
> arch/arm/mach-msm/board-sapphire.c:             mi->bank[0].size = (101*1024*1024);
> 
> because they haven't been buildable since 7th May 2010 (that's 23 months
> ago), and no one has reported any build errors with them.  They're only
> receiving updates from other sweeps and nothing more.  This all means no
> one is even attempting to build this code.  It's pointless having
> unbuildable code in the kernel, and it's nothing more than a useless
> maintanence burden.

I don't think these were ever built.  They were never added to the
Makefiles.

Patch to follow.

David

-- 
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.