[PATCH v5 26/41] arm_mpam: resctrl: Add monitor initialisation and domain boilerplate
Zeng Heng
zengheng4 at huawei.com
Thu Feb 26 19:01:54 PST 2026
Hi Ben,
On 2026/2/26 18:26, Ben Horgan wrote:
> Hi Zeng,
>
> On 2/26/26 03:47, Zeng Heng wrote:
>> Hi Ben,
>>
>> On 2026/2/25 1:57, Ben Horgan wrote:
>>> Add the boilerplate that tells resctrl about the mpam monitors that are
>>> available. resctrl expects all (non-telemetry) monitors to be on the
>>> L3 and
>>> so advertise them there and invent an L3 resctrl resource if required.
>>> The
>>> L3 cache itself has to exist as the cache ids are used as the domain
>>> ids.
>>>
>>> Bring the resctrl monitor domains online and offline based on the cpus
>>> they contain.
>>>
>>> Support for specific monitor types is left to later.
>>>
>>> Signed-off-by: Ben Horgan <ben.horgan at arm.com>
>>> ---
>>> New patch but mostly moved from the existing patches to
>>> separate the monitors from the controls and the boilerplate
>>> from the specific counters.
>>> Use l3->mon_capable in resctrl_arch_mon_capable() as
>>> resctrl_enable_mon_event() now returns a bool.
>>> ---
>>> drivers/resctrl/mpam_internal.h | 7 ++
>>> drivers/resctrl/mpam_resctrl.c | 142 +++++++++++++++++++++++++++++---
>>> 2 files changed, 139 insertions(+), 10 deletions(-)
>>>
>>
>> [...]
>>
>>> @@ -922,6 +1000,20 @@ mpam_resctrl_alloc_domain(unsigned int cpu,
>>> struct mpam_resctrl_res *res)
>>> } else {
>>> pr_debug("Skipped control domain online - no controls\n");
>>> }
>>> +
>>> + if (resctrl_arch_mon_capable()) {
>>> + mon_d = &dom->resctrl_mon_dom;
>>> + mpam_resctrl_domain_hdr_init(cpu, any_mon_comp, r->rid,
>>> &mon_d->hdr);
>>> + mon_d->hdr.type = RESCTRL_MON_DOMAIN;
>>> + err = resctrl_online_mon_domain(r, &mon_d->hdr);
>>> + if (err)
>>> + goto offline_ctrl_domain;
>>> +
>>> + mpam_resctrl_domain_insert(&r->mon_domains, &mon_d->hdr);
>>> + } else {
>>> + pr_debug("Skipped monitor domain online - no monitors\n");
>>> + }
>>> +
>>> return dom;
>>>
>>
>> I noticed that resctrl_arch_mon_capable() only performs checks for L3
>> monitoring functionality. This leads to an issue on platforms that
>> include L2 monitoring capabilities, where the code incorrectly enters
>> this branch and triggers the following warning by
>> mpam_resctrl_domain_insert():
>>
>> [ 22.867070] ------------[ cut here ]------------
>> [ 22.867073] WARNING: drivers/resctrl/mpam_resctrl.c:1495 at
>> mpam_resctrl_domain_insert+0x74/0x80, CPU#2: cpuhp/2/25
>> [ 29.376035] Modules linked in:
>> [ 29.379080] CPU: 2 UID: 0 PID: 25 Comm: cpuhp/2 Not tainted 7.0.0-
>> rc1-g4288ec146462 #30 PREEMPT
>> [ 29.387853] Hardware name: To Be Filled By O.E.M. 183.0/To Be Filled
>> By O.E.M., BIOS 183.0 02/12/2026
>> [ 29.397058] pstate: 61400009 (nZCv daif +PAN -UAO -TCO +DIT -SSBS
>> BTYPE=--)
>> [ 29.404007] pc : mpam_resctrl_domain_insert+0x74/0x80
>> [ 29.409048] lr : mpam_resctrl_domain_insert+0x34/0x80
>> [ 29.414088] sp : ffff8000876abc60
>> ...
>> [ 29.488625] Call trace:
>> [ 29.491060] mpam_resctrl_domain_insert+0x74/0x80 (P)
>> [ 29.496100] mpam_resctrl_online_cpu+0x2b4/0x428
>> [ 29.500706] mpam_cpu_online+0x274/0x298
>> [ 29.504618] cpuhp_invoke_callback+0x104/0x20c
>> [ 29.509052] cpuhp_thread_fun+0xa4/0x17c
>> [ 29.512963] smpboot_thread_fn+0x220/0x24c
>> [ 29.517048] kthread+0x120/0x12c
>> [ 29.520265] ret_from_fork+0x10/0x20
>> [ 29.523830] ---[ end trace 0000000000000000 ]---
>
> Thanks for reporting this bug. It looks to be because resctrl_arch_mon_capable() is telling us if
> there is any mon capable resource when really what we want to know is if this resource is mon capable.
> The pattern occurs in a few places. Does this diff help?
>
I've adapted to the changes and local verification passes also.
> diff --git a/drivers/resctrl/mpam_resctrl.c b/drivers/resctrl/mpam_resctrl.c
> index 694ea8548a05..19b306017845 100644
> --- a/drivers/resctrl/mpam_resctrl.c
> +++ b/drivers/resctrl/mpam_resctrl.c
> @@ -1543,7 +1543,7 @@ mpam_resctrl_alloc_domain(unsigned int cpu, struct mpam_resctrl_res *res)
> if (!dom)
> return ERR_PTR(-ENOMEM);
>
> - if (resctrl_arch_alloc_capable()) {
> + if (r->alloc_capable) {
Yes, using r->alloc_capable and r->mon_capable here is indeed more
accurate and appropriate. I should have noticed this when reviewing
resctrl_arch_alloc_capable() and resctrl_arch_mon_capable().
Reviewed-by: Zeng Heng <zengheng4 at huawei.com>
Thanks,
Zeng Heng
> dom->ctrl_comp = ctrl_comp;
>
> ctrl_d = &dom->resctrl_ctrl_dom;
> @@ -1558,7 +1558,7 @@ mpam_resctrl_alloc_domain(unsigned int cpu, struct mpam_resctrl_res *res)
> pr_debug("Skipped control domain online - no controls\n");
> }
>
> - if (resctrl_arch_mon_capable()) {
> + if (r->mon_capable) {
> struct mpam_component *any_mon_comp;
> struct mpam_resctrl_mon *mon;
> enum resctrl_event_id eventid;
> @@ -1603,7 +1603,7 @@ mpam_resctrl_alloc_domain(unsigned int cpu, struct mpam_resctrl_res *res)
> return dom;
>
> offline_ctrl_domain:
> - if (resctrl_arch_alloc_capable()) {
> + if (r->alloc_capable) {
> mpam_resctrl_offline_domain_hdr(cpu, &ctrl_d->hdr);
> resctrl_offline_ctrl_domain(r, ctrl_d);
> }
> @@ -1671,6 +1671,7 @@ int mpam_resctrl_online_cpu(unsigned int cpu)
> guard(mutex)(&domain_list_lock);
> for_each_mpam_resctrl_control(res, rid) {
> struct mpam_resctrl_dom *dom;
> + struct rdt_resource *r = &res->resctrl_res;
>
> if (!res->class)
> continue; // dummy_resource;
> @@ -1679,12 +1680,12 @@ int mpam_resctrl_online_cpu(unsigned int cpu)
> if (!dom) {
> dom = mpam_resctrl_alloc_domain(cpu, res);
> } else {
> - if (resctrl_arch_alloc_capable()) {
> + if (r->alloc_capable) {
> struct rdt_ctrl_domain *ctrl_d = &dom->resctrl_ctrl_dom;
>
> mpam_resctrl_online_domain_hdr(cpu, &ctrl_d->hdr);
> }
> - if (resctrl_arch_mon_capable()) {
> + if (r->mon_capable) {
> struct rdt_l3_mon_domain *mon_d = &dom->resctrl_mon_dom;
>
> mpam_resctrl_online_domain_hdr(cpu, &mon_d->hdr);
> @@ -1712,6 +1713,7 @@ void mpam_resctrl_offline_cpu(unsigned int cpu)
> struct rdt_l3_mon_domain *mon_d;
> struct rdt_ctrl_domain *ctrl_d;
> bool ctrl_dom_empty, mon_dom_empty;
> + struct rdt_resource *r = &res->resctrl_res;
>
> if (!res->class)
> continue; // dummy resource
> @@ -1720,7 +1722,7 @@ void mpam_resctrl_offline_cpu(unsigned int cpu)
> if (WARN_ON_ONCE(!dom))
> continue;
>
> - if (resctrl_arch_alloc_capable()) {
> + if (r->alloc_capable) {
> ctrl_d = &dom->resctrl_ctrl_dom;
> ctrl_dom_empty = mpam_resctrl_offline_domain_hdr(cpu, &ctrl_d->hdr);
> if (ctrl_dom_empty)
> @@ -1729,7 +1731,7 @@ void mpam_resctrl_offline_cpu(unsigned int cpu)
> ctrl_dom_empty = true;
> }
>
> - if (resctrl_arch_mon_capable()) {
> + if (r->mon_capable) {
> mon_d = &dom->resctrl_mon_dom;
> mon_dom_empty = mpam_resctrl_offline_domain_hdr(cpu, &mon_d->hdr);
> if (mon_dom_empty)
>
>
>>
>>
>> To preserve the existing public interface of resctrl_arch_mon_capable(),
>> please consider the following approach:
>>
>> diff --git a/drivers/resctrl/mpam_resctrl.c b/drivers/resctrl/
>> mpam_resctrl.c
>> index 694ea8548a05..b06a89494ff0 100644
>> --- a/drivers/resctrl/mpam_resctrl.c
>> +++ b/drivers/resctrl/mpam_resctrl.c
>> @@ -1563,6 +1563,10 @@ mpam_resctrl_alloc_domain(unsigned int cpu,
>> struct mpam_resctrl_res *res)
>> if (resctrl_arch_mon_capable()) {
>> struct mpam_component *any_mon_comp;
>> struct mpam_resctrl_mon *mon;
>> enum resctrl_event_id eventid;
>>
>> + /* TODO: Only supports L3 monitor type currently. */
>> + if (r->rid != RDT_RESOURCE_L3)
>> + return dom;
>>
>>
>>
>> Best regards,
>> Zeng Heng
>
>
> Thanks,
>
> Ben
>
>
More information about the linux-arm-kernel
mailing list