Re: [PATCH] x86/resctrl: Preserve CDP enable over cpuhp

[James Morse]

Hi Reinette,

On 13/02/2020 19:45, Reinette Chatre wrote:
> On 2/13/2020 9:42 AM, James Morse wrote:
>> On 12/02/2020 22:53, Reinette Chatre wrote:
>>> On 2/12/2020 10:53 AM, James Morse wrote:
>>>> mounted, and that cpus remember their CDP-enabled state over cpu
>>>> hotplug.
>>>>
>>>> This goes wrong when resctrl's CDP-enabled state changes while all
>>>> the cpus in a domain are offline.
>>>>
>>>> When a domain comes online, enable (or disable!) CDP to match resctrl's
>>>> current setting.

>> ... I think you're describing adding:

[...]

>> to rdtgroup.c and using that from core.c?
> 
> If I understand this correctly the CDP configuration will be done twice
> for each CDP resource, and four times for each CDP resource on a system
> supporting both L2 and L3 CDP. I think it is possible to do
> configuration once for each. Also take care on systems that support MBA
> that would not be caught by the first if statement. A system supporting
> MBA and CDP may thus attempt the configuration even more. It should be
> possible to use the resource parameter for a positive test and then just
> let the other resources fall through? Considering this, what do you
> think of something like below?
> 
> void rdt_domain_reconfigure_cdp(struct rdt_resource *r)
> {
> 	if (!r->alloc_capable)
> 		return;
> 
> 	if (r == &rdt_resources_all[RDT_RESOURCE_L2DATA])
> 		l2_qos_cfg_update(&r->alloc_enabled);
> 
> 	if (r == &rdt_resources_all[RDT_RESOURCE_L3DATA])
> 		l3_qos_cfg_update(&r->alloc_enabled);
> }

Sold!

(the !r->alloc_capable are already filtered out by the caller, but checking is the
least-surprise option)

I'll send a v2 shortly with your suggested-by. I'd like to keep the lockdep annotations as
the MPAM tree tries to stop the arch code taking the rdtgroup_mutex. Those patches
changing these annotations makes it nice and clear what is going on.



Thanks,

James