linux-kernel - Re: [PATCH v3 1/9] drivers: base: Add hardware prefetch control core driver

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]

Date:   Wed, 20 Apr 2022 23:40:48 +0200
From:   Thomas Gleixner <tglx@...utronix.de>
To:     Kohei Tarumizu <tarumizu.kohei@...itsu.com>,
        catalin.marinas@....com, will@...nel.org, mingo@...hat.com,
        bp@...en8.de, dave.hansen@...ux.intel.com, x86@...nel.org,
        hpa@...or.com, linux-arm-kernel@...ts.infradead.org,
        linux-kernel@...r.kernel.org, fenghua.yu@...el.com,
        reinette.chatre@...el.com
Cc:     tarumizu.kohei@...itsu.com,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        "Rafael J. Wysocki" <rafael@...nel.org>
Subject: Re: [PATCH v3 1/9] drivers: base: Add hardware prefetch control
 core driver

Kohei,

Cc+: driver core maintainers

On Wed, Apr 20 2022 at 12:02, Kohei Tarumizu wrote:
> +static const char dist_auto_string[] = "auto";
> +static const char strength_strong_string[] = "strong";
> +static const char strength_weak_string[] = "weak";

This is A64FX specific.

> +pfctl_enable_show(hardware_prefetcher, HWPF_ENABLE);
> +pfctl_enable_show(ip_prefetcher, IPPF_ENABLE);
> +pfctl_enable_show(adjacent_cache_line_prefetcher, ACLPF_ENABLE);

This is x86 specific.

> +pfctl_enable_show(stream_detect_prefetcher, SDPF_ENABLE);

This is A64FX specific.

So why is this in generic code and why needs x86 to populate the A64FX
bits and make them invisible? Same the other way round.

Now imagine a few other [sub]architectures come around and add their
specific prefetcher control knobs, strings and whatever. That's going to
be unmaintainable in no time.

This is not comparable to the cache attributes where the architectures
share a significant amount of subsets. You just demonstrated that X86
and A64FX share not even a single entry.

The core code should provide infrastructure to manage the
[sub]architecture specific control files at different cache levels.

Not more not less.

Thanks,

        tglx