Hi Babu,
On 6/26/25 10:41 AM, Moger, Babu wrote:
> On 6/24/25 22:03, Reinette Chatre wrote:
>> On 6/13/25 2:05 PM, Babu Moger wrote:
..
>>> ---
>>> arch/x86/kernel/cpu/resctrl/monitor.c | 38 +++++++++++++++++++++++++++
>>> include/linux/resctrl.h | 19 ++++++++++++++
>>> 2 files changed, 57 insertions(+)
>>>
>>> diff --git a/arch/x86/kernel/cpu/resctrl/monitor.c b/arch/x86/kernel/cpu/resctrl/monitor.c
>>> index 0ad9c731c13e..6b0ea4b17c7a 100644
>>> --- a/arch/x86/kernel/cpu/resctrl/monitor.c
>>> +++ b/arch/x86/kernel/cpu/resctrl/monitor.c
>>> @@ -444,3 +444,41 @@ bool resctrl_arch_mbm_cntr_assign_enabled(struct rdt_resource *r)
>>> {
>>> return resctrl_to_arch_res(r)->mbm_cntr_assign_enabled;
>>> }
>>> +
>>> +static void resctrl_abmc_config_one_amd(void *info)
>>> +{
>>> + union l3_qos_abmc_cfg *abmc_cfg = info;
>>> +
>>> + wrmsrl(MSR_IA32_L3_QOS_ABMC_CFG, abmc_cfg->full);
>>> +}
>>> +
>>> +/*
>>> + * Send an IPI to the domain to assign the counter to RMID, event pair.
>>> + */
>>> +void resctrl_arch_config_cntr(struct rdt_resource *r, struct rdt_mon_domain *d,
>>> + enum resctrl_event_id evtid, u32 rmid, u32 closid,
>>> + u32 cntr_id, bool assign)
>>> +{
>>> + struct rdt_hw_mon_domain *hw_dom = resctrl_to_arch_mon_dom(d);
>>> + union l3_qos_abmc_cfg abmc_cfg = { 0 };
>>> + struct arch_mbm_state *am;
>>> +
>>> + abmc_cfg.split.cfg_en = 1;
>>> + abmc_cfg.split.cntr_en = assign ? 1 : 0;
>>> + abmc_cfg.split.cntr_id = cntr_id;
>>> + abmc_cfg.split.bw_src = rmid;
>>> + if (assign)
>>> + abmc_cfg.split.bw_type = resctrl_get_mon_evt_cfg(evtid);
>>> +
>>> + smp_call_function_any(&d->hdr.cpu_mask, resctrl_abmc_config_one_amd, &abmc_cfg, 1);
>>
>> An optimization to consider is to direct the IPI to a housekeeping CPU.
>> If one exist, a further optimization could be to queue it on that CPU
>> instead of IPI. Your call since I am not familiar with the use cases here.
>> Looks like all paths leading to this is triggered by a user space action
>> (mount, mkdir, or write to update event config). This would require exposing
>> the housekeeping CPU code to arch.
>
> Do you mean something like this?
>
> cpu = cpumask_any_housekeeping(&d->hdr.cpu_mask, RESCTRL_PICK_ANY_CPU);
>
> smp_call_on_cpu(cpu, resctrl_abmc_config_one_amd, &abmc_cfg, false);
Please note the returned "cpu" may be nohz_full and if it is it would need
an IPI. Similar to mon_event_read().
>
>
> You want to do these changes now or later? It requires few other changes
> to move around the code.
I'll leave this up to you. I think it would be ideal if cpumask_any_housekeeping()
can be hosted in include/linux/cpumask.h instead of moving it around within
resctrl.
>
>>
>>> +
>>> + /*
>>> + * The hardware counter is reset (because cfg_en == 1) so there is no
>>> + * need to record initial non-zero counts.
>>> + */
>>> + if (assign) {
>>> + am = get_arch_mbm_state(hw_dom, rmid, evtid);
>>> + if (am)
>>> + memset(am, 0, sizeof(*am));
>>> + }
>>
>> I am not able to recognize how the struct rdt_resource parameter is used. What am I missing?
>
> No. It is not used here. It is kept as other arch's can use it. I think
> James commented about it earlier.
I see, thank you.
Reinette
