Hi Lukas

...

From: Auer, Lukas [mailto:lukas.auer@aisec.fraunhofer.de] Sent: Thursday, August 08, 2019 10:04 PM To: bmeng.cn@gmail.com Cc: Rick Jian-Zhi Chen(陳建志); u-boot@lists.denx.de Subject: Re: [PATCH] riscv: cpu: Skip unavailable hart in the get_count() op

Hi Bin,

On Thu, 2019-08-08 at 21:25 +0800, Bin Meng wrote:

...

Hi Lukas,

On Thu, Aug 8, 2019 at 7:22 PM Auer, Lukas lukas.auer@aisec.fraunhofer.de wrote:

...

Hi Bin,

On Thu, 2019-08-08 at 00:52 -0700, Bin Meng wrote:

...

We should not count in hart that is marked as not available in the device tree in riscv_cpu_get_count().

I think it might make sense to also exclude harts that are not listed as available in the available_harts mask. So the same logic as in arch/riscv/lib/smp.c. In this case, the bind function should probably check the mask as well.

Yes, the check to available_harts mask can be added in riscv_cpu_get_count(). I will do that.

However, I doubt we could do available_harts mask check in the bind function. If it returns error for non-available harts, the whole initialization process fails.

You are right, I did not think about that. Perhaps it's best to ignore the available_harts mask in the count function as well then.

It just checked where cpu_get_count() is used. The Andes PLIC driver (lib/andes_plic.c) uses it in a way that the patch might break the driver. It iterates over all harts, assuming the first hart is 0 and the last hart the result of cpu_get_count(). If any of the harts is set to disabled, this will no longer be true. It's probably best to update the driver to use the ofnode_for_each_subnode() macro to iterate over all harts. Rick, can you update the driver?

Yes, I will prepare a patch to update it. Thanks for your catch and reminding.

Regards, Rick

...

Thanks, Lukas

...

...

Looks good otherwise!

Reviewed-by: Lukas Auer lukas.auer@aisec.fraunhofer.de

Regards, Bin