Call for testing: spi-mem driver for Aspeed SMC controllers

[Cédric Le Goater]

On 3/9/22 02:58, Lei Yu wrote:
>>>>>>
>>>>>> I don't see why. Is that the same kernel version and config ?
>>>>>>
>>>>>> May be check the allocations in
>>>>>>
>>>>>>       /proc/vmallocinfo
>>>>>
>>>>> It shows a large area as below:
>>>>>
>>>>>     0xc1d30472-0x406b4a0c 268439552 __devm_ioremap_resource+0x178/0x1f0
>>>>> phys=0x20000000 ioremap
>>>>
>>>> Is CONFIG_VMSPLIT_2G set ?
>>>
>>> Yes, it's enabled in aspeed-g6/defconfig
>>
>> The OpenBMC tree only has a aspeed_g5_defconfig.
> 
> It does have aspeed-g6/defconfig and CONFIG_VMSPLIT_2G is enabled:
> https://github.com/openbmc/openbmc/blob/master/meta-aspeed/recipes-kernel/linux/linux-aspeed/aspeed-g6/defconfig#L33

ok. I was referring to the default configs in the kernel tree.
That's fine.

>>
>> How much RAM do you have ? Could check the totals in meminfo ?
> 
> Our system (and QEMU model) has 2GiB.

ah. you changed the default. Here is why we didn't merge the AST2600
with 2GB :

   https://lists.nongnu.org/archive/html/qemu-devel/2019-10/msg03903.html

> 
> Boot with `vmalloc=500M`:
> # cat /proc/meminfo
> MemTotal:        1984880 kB
> MemFree:         1860700 kB
> MemAvailable:    1882116 kB

That's the reason. The vmalloc area size is in direct relation with
the available RAM. See the ARM memory layout :

   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/arm/memory.rst

May be we should consider 64-bit kernel one day.

Thanks,

C.