linux-kernel - Re: [PATCH] arm64: arm_pmuv3: Correctly extract and check the PMUVer

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: <20240409160915.GA24004@willie-the-truck>
Date: Tue, 9 Apr 2024 17:09:16 +0100
From: Will Deacon <will@...nel.org>
To: Yicong Yang <yangyicong@...wei.com>
Cc: mark.rutland@....com, catalin.marinas@....com, broonie@...nel.org,
	linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
	jonathan.cameron@...wei.com, prime.zeng@...ilicon.com,
	linuxarm@...wei.com, yangyicong@...ilicon.com
Subject: Re: [PATCH] arm64: arm_pmuv3: Correctly extract and check the PMUVer

On Mon, Apr 08, 2024 at 04:11:58PM +0800, Yicong Yang wrote:
> From: Yicong Yang <yangyicong@...ilicon.com>
> 
> Currently we're using "sbfx" to extract the PMUVer from ID_AA64DFR0_EL1
> and skip the init/reset if no PMU present when the extracted PMUVer is
> negative or is zero. However for PMUv3p8 the PMUVer will be 0b1000 and
> PMUVer extracted by "sbfx" will always be negative and we'll skip the
> init/reset in __init_el2_debug/reset_pmuserenr_el0 unexpectedly.
> 
> So this patch use "ubfx" instead of "sbfx" to extract the PMUVer. If
> the PMUVer is implementation defined (0b1111) then reset it to zero
> and skip the reset/init. Previously we'll also skip the init/reset
> if the PMUVer is higher than the version we known (currently PMUv3p9),
> with this patch we'll only skip if the PMU is not implemented or
> implementation defined. This keeps consistence with how we probe
> the PMU in the driver with pmuv3_implemented().
> 
> Signed-off-by: Yicong Yang <yangyicong@...ilicon.com>
> ---
>  arch/arm64/include/asm/assembler.h | 5 ++++-
>  arch/arm64/include/asm/el2_setup.h | 5 ++++-
>  2 files changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/assembler.h b/arch/arm64/include/asm/assembler.h
> index ab8b396428da..3b7373d6c565 100644
> --- a/arch/arm64/include/asm/assembler.h
> +++ b/arch/arm64/include/asm/assembler.h
> @@ -480,7 +480,10 @@ alternative_endif
>   */
>  	.macro	reset_pmuserenr_el0, tmpreg
>  	mrs	\tmpreg, id_aa64dfr0_el1
> -	sbfx	\tmpreg, \tmpreg, #ID_AA64DFR0_EL1_PMUVer_SHIFT, #4
> +	ubfx	\tmpreg, \tmpreg, #ID_AA64DFR0_EL1_PMUVer_SHIFT, #4
> +	cmp	\tmpreg, #ID_AA64DFR0_EL1_PMUVer_IMP_DEF
> +	csel	\tmpreg, xzr, \tmpreg, eq	// If PMU's IMP_DEF, regard it
> +						// as not implemented and skip
>  	cmp	\tmpreg, #1			// Skip if no PMU present
>  	b.lt	9000f
>  	msr	pmuserenr_el0, xzr		// Disable PMU access from EL0

I think the cmp/csel/cmp/b.lt sequence might be a little tidier if you
reworked it to use ccmp. For example, something like (totally untested):

	cmp	\tmpreg, ID_AA64DFR0_EL1_PMUVer_NI
	ccmp	\tmpreg, ID_AA64DFR0_EL1_PMUVer_IMP_DEF, #4, ne

would then, I think, mean we could just b.eq 9000f. But please check
this because encoding nzcv as an immediate always catches me out.

> diff --git a/arch/arm64/include/asm/el2_setup.h b/arch/arm64/include/asm/el2_setup.h
> index b7afaa026842..2438e12b60c5 100644
> --- a/arch/arm64/include/asm/el2_setup.h
> +++ b/arch/arm64/include/asm/el2_setup.h
> @@ -59,7 +59,10 @@
>  
>  .macro __init_el2_debug
>  	mrs	x1, id_aa64dfr0_el1
> -	sbfx	x0, x1, #ID_AA64DFR0_EL1_PMUVer_SHIFT, #4
> +	ubfx	x0, x1, #ID_AA64DFR0_EL1_PMUVer_SHIFT, #4
> +	cmp	x0, #ID_AA64DFR0_EL1_PMUVer_IMP_DEF
> +	csel	x0, xzr, x0, eq			// If PMU's IMP_DEF, regard it
> +						// as not implemented and skip
>  	cmp	x0, #1
>  	b.lt	.Lskip_pmu_\@			// Skip if no PMU present
>  	mrs	x0, pmcr_el0			// Disable debug access traps

Similar sort of thing here.

Will