linux-kernel - Re: [PATCH v2] LoongArch: Fix PMU counter allocation for mixed-type event groups

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: <42bcaa57-bab3-457f-83cc-d908303090f2@linux.intel.com>
Date: Thu, 8 Jan 2026 16:07:33 +0800
From: "Mi, Dapeng" <dapeng1.mi@...ux.intel.com>
To: Lisa Robinson <lisa@...efly.space>, Peter Zijlstra
 <peterz@...radead.org>, Ingo Molnar <mingo@...hat.com>,
 Arnaldo Carvalho de Melo <acme@...nel.org>,
 Namhyung Kim <namhyung@...nel.org>, Huacai Chen <chenhuacai@...nel.org>
Cc: Mark Rutland <mark.rutland@....com>,
 Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
 Jiri Olsa <jolsa@...nel.org>, Ian Rogers <irogers@...gle.com>,
 Adrian Hunter <adrian.hunter@...el.com>, James Clark
 <james.clark@...aro.org>, WANG Xuerui <kernel@...0n.name>,
 linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org,
 loongarch@...ts.linux.dev
Subject: Re: [PATCH v2] LoongArch: Fix PMU counter allocation for mixed-type
 event groups


On 1/5/2026 12:23 AM, Lisa Robinson wrote:
> When validating a perf event group, validate_group() unconditionally
> attempts to allocate hardware PMU counters for the leader, sibling
> events and the new event being added.
>
> This is incorrect for mixed-type groups. If a PERF_TYPE_SOFTWARE event
> ispart of the group, the current code still tries to allocate a hardware

ispart -> "is part"


> PMU counter for it, which can wrongly consume hardware PMU resources and
> cause spurious allocation failures.
>
> Fix this by only allocating PMU counters for hardware events during group
> validation, and skipping software events.
>
> A trimmed down reproducer is as simple as this:
>
>   #include <stdio.h>
>   #include <assert.h>
>   #include <unistd.h>
>   #include <string.h>
>   #include <sys/syscall.h>
>   #include <linux/perf_event.h>
>
>   int
>   main (int argc, char *argv[])
>   {
>   	struct perf_event_attr attr = { 0 };
>   	int fds[5];
>
>   	attr.disabled = 1;
>   	attr.exclude_kernel = 1;
>   	attr.exclude_hv = 1;
>   	attr.read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
>   		PERF_FORMAT_TOTAL_TIME_RUNNING | PERF_FORMAT_ID | PERF_FORMAT_GROUP;
>   	attr.size = sizeof (attr);
>
>   	attr.type = PERF_TYPE_SOFTWARE;
>   	attr.config = PERF_COUNT_SW_DUMMY;
>   	fds[0] = syscall (SYS_perf_event_open, &attr, 0, -1, -1, 0);
>   	assert (fds[0] >= 0);
>
>   	attr.type = PERF_TYPE_HARDWARE;
>   	attr.config = PERF_COUNT_HW_CPU_CYCLES;
>   	fds[1] = syscall (SYS_perf_event_open, &attr, 0, -1, fds[0], 0);
>   	assert (fds[1] >= 0);
>
>   	attr.type = PERF_TYPE_HARDWARE;
>   	attr.config = PERF_COUNT_HW_INSTRUCTIONS;
>   	fds[2] = syscall (SYS_perf_event_open, &attr, 0, -1, fds[0], 0);
>   	assert (fds[2] >= 0);
>
>   	attr.type = PERF_TYPE_HARDWARE;
>   	attr.config = PERF_COUNT_HW_BRANCH_MISSES;
>   	fds[3] = syscall (SYS_perf_event_open, &attr, 0, -1, fds[0], 0);
>   	assert (fds[3] >= 0);
>
>   	attr.type = PERF_TYPE_HARDWARE;
>   	attr.config = PERF_COUNT_HW_CACHE_REFERENCES;
>   	fds[4] = syscall (SYS_perf_event_open, &attr, 0, -1, fds[0], 0);
>   	assert (fds[4] >= 0);
>
>   	printf ("PASSED\n");
>
>   	return 0;
>   }
>
> Fixes: b37042b2bb7c ("LoongArch: Add perf events support")
> Signed-off-by: Lisa Robinson <lisa@...efly.space>
> ---
> Changes in v2:
> - Factor out duplicated perf event type checks into an inline helper.
> ---
>  arch/loongarch/kernel/perf_event.c | 21 ++++++++++++++++++---
>  1 file changed, 18 insertions(+), 3 deletions(-)
>
> diff --git a/arch/loongarch/kernel/perf_event.c b/arch/loongarch/kernel/perf_event.c
> index 9d257c8519c9..e34a6fb33e11 100644
> --- a/arch/loongarch/kernel/perf_event.c
> +++ b/arch/loongarch/kernel/perf_event.c
> @@ -626,6 +626,18 @@ static const struct loongarch_perf_event *loongarch_pmu_map_cache_event(u64 conf
>  	return pev;
>  }
>  
> +static inline bool loongarch_pmu_event_requires_counter(const struct perf_event *event)
> +{
> +	switch (event->attr.type) {
> +	case PERF_TYPE_HARDWARE:
> +	case PERF_TYPE_HW_CACHE:
> +	case PERF_TYPE_RAW:
> +		return true;
> +	default:
> +		return false;
> +	}
> +}
> +
>  static int validate_group(struct perf_event *event)
>  {
>  	struct cpu_hw_events fake_cpuc;
> @@ -633,15 +645,18 @@ static int validate_group(struct perf_event *event)
>  
>  	memset(&fake_cpuc, 0, sizeof(fake_cpuc));
>  
> -	if (loongarch_pmu_alloc_counter(&fake_cpuc, &leader->hw) < 0)
> +	if (loongarch_pmu_event_requires_counter(leader) &&
> +	    loongarch_pmu_alloc_counter(&fake_cpuc, &leader->hw) < 0)
>  		return -EINVAL;
>  
>  	for_each_sibling_event(sibling, leader) {
> -		if (loongarch_pmu_alloc_counter(&fake_cpuc, &sibling->hw) < 0)
> +		if (loongarch_pmu_event_requires_counter(sibling) &&
> +		    loongarch_pmu_alloc_counter(&fake_cpuc, &sibling->hw) < 0)
>  			return -EINVAL;
>  	}
>  
> -	if (loongarch_pmu_alloc_counter(&fake_cpuc, &event->hw) < 0)
> +	if (loongarch_pmu_event_requires_counter(event) &&
> +	    loongarch_pmu_alloc_counter(&fake_cpuc, &event->hw) < 0)
>  		return -EINVAL;
>  
>  	return 0;

The code looks good to me, but I'm not quite familiar the loongarch perf
code, then I won't give a reviewed-by and leave it to loongarch perf
experts. Thanks.