linux-kernel - Re: [PATCH 11/20] perf/x86/intel: Setup PEBS constraints base on counter & pdist map

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <e123a3c7-098b-4caf-9f69-601a435bda70@linux.intel.com>
Date: Thu, 6 Feb 2025 10:01:47 -0500
From: "Liang, Kan" <kan.liang@...ux.intel.com>
To: "Mi, Dapeng" <dapeng1.mi@...ux.intel.com>,
 Peter Zijlstra <peterz@...radead.org>, Ingo Molnar <mingo@...hat.com>,
 Arnaldo Carvalho de Melo <acme@...nel.org>,
 Namhyung Kim <namhyung@...nel.org>, Ian Rogers <irogers@...gle.com>,
 Adrian Hunter <adrian.hunter@...el.com>,
 Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
 Andi Kleen <ak@...ux.intel.com>, Eranian Stephane <eranian@...gle.com>
Cc: linux-kernel@...r.kernel.org, linux-perf-users@...r.kernel.org,
 Dapeng Mi <dapeng1.mi@...el.com>
Subject: Re: [PATCH 11/20] perf/x86/intel: Setup PEBS constraints base on
 counter & pdist map



On 2025-02-05 9:47 p.m., Mi, Dapeng wrote:
> 
> On 1/28/2025 12:07 AM, Liang, Kan wrote:
>>
>> On 2025-01-23 9:07 a.m., Dapeng Mi wrote:
>>> arch-PEBS provides CPUIDs to enumerate which counters support PEBS
>>> sampling and precise distribution PEBS sampling. Thus PEBS constraints
>>> can be dynamically configured base on these counter and precise
>>> distribution bitmap instead of defining them statically.
>>>
>>> Signed-off-by: Dapeng Mi <dapeng1.mi@...ux.intel.com>
>>> ---
>>>  arch/x86/events/intel/core.c | 20 ++++++++++++++++++++
>>>  arch/x86/events/intel/ds.c   |  1 +
>>>  2 files changed, 21 insertions(+)
>>>
>>> diff --git a/arch/x86/events/intel/core.c b/arch/x86/events/intel/core.c
>>> index 7775e1e1c1e9..0f1be36113fa 100644
>>> --- a/arch/x86/events/intel/core.c
>>> +++ b/arch/x86/events/intel/core.c
>>> @@ -3728,6 +3728,7 @@ intel_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
>>>  			    struct perf_event *event)
>>>  {
>>>  	struct event_constraint *c1, *c2;
>>> +	struct pmu *pmu = event->pmu;
>>>  
>>>  	c1 = cpuc->event_constraint[idx];
>>>  
>>> @@ -3754,6 +3755,25 @@ intel_get_event_constraints(struct cpu_hw_events *cpuc, int idx,
>>>  		c2->weight = hweight64(c2->idxmsk64);
>>>  	}
>>>  
>>> +	if (x86_pmu.arch_pebs && event->attr.precise_ip) {
>>> +		u64 pebs_cntrs_mask;
>>> +		u64 cntrs_mask;
>>> +
>>> +		if (event->attr.precise_ip >= 3)
>>> +			pebs_cntrs_mask = hybrid(pmu, arch_pebs_cap).pdists;
>>> +		else
>>> +			pebs_cntrs_mask = hybrid(pmu, arch_pebs_cap).counters;
>>> +
>>> +		cntrs_mask = hybrid(pmu, fixed_cntr_mask64) << INTEL_PMC_IDX_FIXED |
>>> +			     hybrid(pmu, cntr_mask64);
>>> +
>>> +		if (pebs_cntrs_mask != cntrs_mask) {
>>> +			c2 = dyn_constraint(cpuc, c2, idx);
>>> +			c2->idxmsk64 &= pebs_cntrs_mask;
>>> +			c2->weight = hweight64(c2->idxmsk64);
>>> +		}
>>> +	}
>> The pebs_cntrs_mask and cntrs_mask wouldn't be changed since the machine
>> boot. I don't think it's efficient to calculate them every time.
>>
>> Maybe adding a local pebs_event_constraints_pdist[] and update both
>> pebs_event_constraints[] and pebs_event_constraints_pdist[] with the
>> enumerated mask at initialization time.
>>
>> Update the intel_pebs_constraints() to utilize the corresponding array
>> according to the precise_ip.
>>
>> The above may be avoided.
> 
> Even we have these two arrays, we still need the dynamic constraint, right?
> We can't predict what the event is, the event may be mapped to a quite
> specific event constraint and we can know it in advance.

The dynamic constraint is not necessary, but two arrays seems not
enough. Because a PEBS event may fall back to the event_constraints as
well. Sigh.
Four arrays should be required. pebs_event_constraints[],
pebs_event_constraints_pdist[], event_constraints_for_pebs[],
event_constraints_for_pdist_pebs[].
But it seems too complicated. It may not be implemented now.

But, at least the pebs_cntrs_mask and cntrs_mask can be calculated in
the hw_config(), or even intel_pmu_init() once. It should not be
calculated every time in the critical path.

Thanks,
Kan
> 
> 
>>
>> Thanks,
>> Kan
>>
>>> +
>>>  	return c2;
>>>  }
>>>  
>>> diff --git a/arch/x86/events/intel/ds.c b/arch/x86/events/intel/ds.c
>>> index 2f2c6b7c801b..a573ce0e576a 100644
>>> --- a/arch/x86/events/intel/ds.c
>>> +++ b/arch/x86/events/intel/ds.c
>>> @@ -2941,6 +2941,7 @@ static void __init intel_arch_pebs_init(void)
>>>  	x86_pmu.pebs_buffer_size = PEBS_BUFFER_SIZE;
>>>  	x86_pmu.drain_pebs = intel_pmu_drain_arch_pebs;
>>>  	x86_pmu.pebs_capable = ~0ULL;
>>> +	x86_pmu.flags |= PMU_FL_PEBS_ALL;
>>>  }
>>>  
>>>  /*
>