linux-kernel - Re: [RFC PATCH v2 2/4] perf: arm_pmuv3: Introduce module param to partition the PMU

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

Message-ID: <Z6mp4NeklzORJg5y@linux.dev>
Date: Sun, 9 Feb 2025 23:25:20 -0800
From: Oliver Upton <oliver.upton@...ux.dev>
To: Colton Lewis <coltonlewis@...gle.com>
Cc: kvm@...r.kernel.org, Russell King <linux@...linux.org.uk>,
	Catalin Marinas <catalin.marinas@....com>,
	Will Deacon <will@...nel.org>, Marc Zyngier <maz@...nel.org>,
	Joey Gouly <joey.gouly@....com>,
	Suzuki K Poulose <suzuki.poulose@....com>,
	Zenghui Yu <yuzenghui@...wei.com>,
	Mark Rutland <mark.rutland@....com>,
	Paolo Bonzini <pbonzini@...hat.com>, Shuah Khan <shuah@...nel.org>,
	linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
	kvmarm@...ts.linux.dev, linux-perf-users@...r.kernel.org,
	linux-kselftest@...r.kernel.org
Subject: Re: [RFC PATCH v2 2/4] perf: arm_pmuv3: Introduce module param to
 partition the PMU

Hi Colton,

On Sat, Feb 08, 2025 at 02:01:09AM +0000, Colton Lewis wrote:
> For PMUv3, the register MDCR_EL2.HPMN partitiones the PMU counters
> into two ranges where counters 0..HPMN-1 are accessible by EL1 and, if
> allowed, EL0 while counters HPMN..N are only accessible by EL2.
> 
> Introduce a module parameter in the PMUv3 driver to set this
> register. The name reserved_host_counters reflects the intent to
> reserve some counters for the host so the guest may eventually be
> allowed direct access to a subset of PMU functionality for increased
> performance.
> 
> Track HPMN and whether the pmu is partitioned in struct arm_pmu
> because KVM will need to know that to handle guests correctly.
> 
> While FEAT_HPMN0 does allow HPMN to be set to 0, this patch
> specifically disallows that case because it's not useful given the
> intention to allow guests access to their own counters.

Quite the contrary.

FEAT_HPMN0 is useful if userspace wants to provide a vPMU that has a
fixed cycle counter w/o event counters. Certain OSes refuse to boot
without it...

> static inline u32 read_mdcr(void)
> {
> 	return read_sysreg(HDCR);
> }
> 
> static inline void write_mdcr(u32 val)
> {
> 	write_sysreg(val, HDCR);
> }
> 

Hmm... While this fixes the 32bit compilation issues, it opens a new can
of worms.

VHE is a 64bit only feature, so you're *guaranteed* that these accessors
will undef (running at EL1).

> +static void armv8pmu_partition(u8 hpmn)
> +{
> +	u64 mdcr = armv8pmu_mdcr_read();
> +
> +	mdcr &= ~ARMV8_PMU_MDCR_HPMN;
> +	mdcr |= FIELD_PREP(ARMV8_PMU_MDCR_HPMN, hpmn);
> +	/* Prevent guest counters counting at EL2 */
> +	mdcr |= ARMV8_PMU_MDCR_HPMD;
> +
> +	armv8pmu_mdcr_write(mdcr);
> +}
> +

After giving this a read, I don't think the host PMU driver should care
about MDCR_EL2 at all. The only time that 'guest' events are loaded into
the PMU is between vcpu_load() / vcpu_put(), at which point *KVM* has
reconfigured MDCR_EL2.

I'm not sure if there's much involvement with the host PMU driver beyond
it pinky-swearing to only use the specified counters. KVM is what
actually will program HPMN.

That'd alleviate the PMU driver from having VHE awareness or caring
about MDCR_EL2.

-- 
Thanks,
Oliver