[PATCH] ARM: shmobile: Include all 2 GiB of memory on APE6EVM

[Magnus Damm]

Hi Simon,

On Wed, Nov 6, 2013 at 10:48 AM, Simon Horman <horms at verge.net.au> wrote:
> On Thu, Oct 31, 2013 at 05:02:13PM +0900, Magnus Damm wrote:
>> Hi Simon,
>>
>> On Thu, Oct 31, 2013 at 1:20 PM, Simon Horman <horms at verge.net.au> wrote:
>> > On Thu, Oct 31, 2013 at 12:15:49PM +0900, Magnus Damm wrote:
>> >> From: Takashi Yoshii <takasi-y at ops.dti.ne.jp>
>> >>
>> >> Add 1GiB of DRAM at 0x2_0000_0000 to support the full 2GiB
>> >> of APE6EVM system memory.
>> >>
>> >> Signed-off-by: Takashi Yoshii <takasi-y at ops.dti.ne.jp>
>> >> Signed-off-by: Magnus Damm <damm at opensource.se>
>> >> ---
>> >>
>> >>  For correct run time operation regardless of kernel configuration
>> >>  ARM patches 7863/1 and 7864/1 are needed.
>> >
>> > I take it that I should wait for those patches before applying this one.
>>
>> That's probably a good idea. RMK has applied them to "git-curr", so I
>> suppose they are queued up for something at least:
>>
>> http://www.arm.linux.org.uk/developer/patches/viewpatch.php?id=7863/1
>
> I have also sighted them in linux-next.
> So I expect they will show up in v3.13-rc1 or thereabouts.
>
>> > With those two patches and this one applied on top of
>> > renesas-devel-v3.12-rc7-20131030 I was able to successfully
>> > boot an ape6evm board. But /proc/meminfo still indicates only 1Gb of
>> > memory. What am I missing.
>>
>> You probably need to enable HIGHMEM and LPAE to be able to access to
>> all memory. Same thing on Lager, to get the full 4GiB you need to
>> select those in your kernel configuration too.
>
> Thanks, I have been able to successfully test all 4 patches in this
> series with those settings enabled.
>
> Tested-by: Simon Horman <horms+renesas at verge.net.au>
>
>> Please note that having memory located on 64-bit addresses may break
>> some drivers doing bus mastering or using the DMAC. These drivers need
>> to be fixed both with short term DMA zone workarounds and proper
>> 64-bit address support whenever possible.
>>
>> So if I could suggest a policy then having LPAE=n in defconfig for the
>> C board code for default stable operation. This together with LPAE=y
>> in the case of DT reference and/or Multiplatform where we can be a bit
>> more experimental. I don't care enough to send defconfig patches, but
>> if some issue comes up then perhaps adjusting the defconfig to disable
>> LPAE may work well as a workaround.
>
> Thanks.
>
> As it stands the defconfig for both APE6EVM and Lager lead to a
> configuration with LPAE disabled, although neither explicitly disable LPAE.
> I think it is best to leave things as-is in this regards and explicitly
> disable LPAE in the defconfigs if the need arises.
>
> While looking over this I noticed that the APE6EVM defconfig selects
> HIGHMEM while the Lager defconfig does not and it is not selected in the
> resulting config. Would it be worth updating the Lager defconfig to enable
> HIGHMEM? If so, I'll make a patch.

Thanks. I think it mainly makes sense to enable LPAE and HIGHMEM
together, so easiest is probably just to wait with both Lager and
APE6EVM.

Cheers,

/ magnus