Re: [PATCH -next] arch_topology: Fix cache attributes detection in the CPU hotplug path

From: Sudeep Holla
Date: Fri Jul 15 2022 - 11:41:30 EST

On Fri, Jul 15, 2022 at 02:04:41PM +0000, Conor.Dooley@xxxxxxxxxxxxx wrote:
> On 15/07/2022 10:16, Conor Dooley wrote:
> > On 15/07/2022 10:11, Sudeep Holla wrote:
> >> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
> >>
> >> On Thu, Jul 14, 2022 at 04:10:36PM +0000, Conor.Dooley@xxxxxxxxxxxxx wrote:
> >>> With the GFP_ATOMIC, behaviour is the same as before for me.
> >>>
> >>
> >> So you still get -ENOMEM failure on your platform. It is fine, just that
> >> I am bit curious to know why as it succeeds at device_initcall later.
> >> I was hoping this might fix your memory allocation failure.
> >
> > Correct.
> I take that back. On today's next with patch 2 applied, I don't see a
> problem with no memory, so this does appear to have sorted the memory
> allocation failure. Sorry for misleading you & thanks!

No worries. Glad to hear this fixed the memory allocation issue you had
on your platform, I was hopeful based on some analysis I did. I must have
done this from first, I think I had seen the bug in my initial testing and
moved the call to detect_cache_attributes() into init_cpu_topology() instead
which fixed it but broke hotplug which I didn't notice on few platforms
I tested until Ionela tested it on a particular platform.

> With my patches for store_cpu_topology on RISC-V I do see it though,
> when called on the boot cpu. I must have mixed up which I had tested.
> I have a fix for that though & will update my patches later.