[Arm-netbook] Mele A2000 as a Server
Emilio López
turl at linux-sunxi.org
Tue Jul 17 06:47:43 BST 2012
El 16/07/12 07:02, Alejandro Mery escribió:
> On 15 July 2012 19:43, Henrik Nordström <henrik at henriknordstrom.net> wrote:
>> sön 2012-07-15 klockan 16:51 +0600 skrev Roman Mamedov:
>>
>>> It might be possible to free up 64 MB more, but I believe this requires
>>> reconfiguring the uboot or changing the script.bin. I did not investigate
>>> much, 443 MB is enough to me for now.
>>
>> MALI needs to be disabled in kernel build, and some ugly code ripped out
>> as well, see arch/arm/mach-sun4i/core.c sw_core_fixup. just kill it
>> entirely, it deserves no right to live there.
>
> there is currently a branch awaiting for testers called
> `wip/linux-sunxi-3.0/mem` which:
> * honors atag memory info passed by the bootloader (unless told the
> oposite by enabling `CONFIG_SUNXI_IGNORE_ATAG_MEM`)
> * allocated the 64M for mali only if it's enabled (will probably gain
> an option for off-tree mali to allocate them even when mali is
> disabled)
>
>
> $ free
> total used free shared buffers cached
> Mem: 511324 62240 449084 0 4204 39040
> -/+ buffers/cache: 18996 492328
> Swap: 0 0 0
>
> tests from people actually using mali are needed before merging this branch.
>
That branch causes a kernel panic if you actually use the GPU
See here for the actual error, I got this using ramconsole
http://paste.debian.net/179422/
Something like this on the other hand works just fine, as the code
pretty much remains unchanged when you do use the GPU, and should cause
no memory loss if mali is disabled. The if logic could be moved into the
code as an #if to keep it totally dynamic I suppose.
http://paste.ubuntu.com/1095929/
Also, for the record, I tried to move all the kernelspace mali
allocation to the end of my 1GB RAM, and although kernel wise it seems
to work fine, the userspace doesn't like it. At all. The kernel region
address and size is hardcoded on the mali userspace libs :/
More information about the arm-netbook
mailing list