linux-kernel - Re: [PATCH v4 0/4] x86/cpufeatures: Automatically generate required and disabled feature masks

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: <308cddef-9efa-47ab-b499-34c52480dd09@zytor.com>
Date: Wed, 4 Dec 2024 09:56:02 -0800
From: Xin Li <xin@...or.com>
To: linux-kernel@...r.kernel.org, linux-perf-users@...r.kernel.org
Cc: tglx@...utronix.de, mingo@...hat.com, bp@...en8.de,
        dave.hansen@...ux.intel.com, x86@...nel.org, hpa@...or.com,
        will@...nel.org, peterz@...radead.org, akpm@...ux-foundation.org,
        acme@...nel.org, namhyung@...nel.org, brgerst@...il.com,
        andrew.cooper3@...rix.com, nik.borisov@...e.com
Subject: Re: [PATCH v4 0/4] x86/cpufeatures: Automatically generate required
 and disabled feature masks

On 8/19/2024 11:25 PM, Xin Li wrote:
> On 6/28/2024 10:45 AM, Xin Li (Intel) wrote:
>> The x86 build process first generates required and disabled feature
>> masks based on current build config, and then uses these generated
>> masks to compile the source code. When a CPU feature is not enabled
>> in a build config, e.g., when CONFIG_X86_FRED=n, its feature disable
>> flag, i.e., DISABLE_FRED, needs to be properly defined and added to
>> a specific disabled CPU features mask in <asm/disabled-features.h>,
>> as the following patch does:
>> https://lore.kernel.org/all/20231205105030.8698-8-xin3.li@intel.com/.
>> As a result, the FRED feature bit is surely cleared in the generated
>> kernel binary when CONFIG_X86_FRED=n.
>>
>> Recently there is another case to repeat the same exercise for the
>> AMD SEV-SNP CPU feature:
>> https://lore.kernel.org/all/20240126041126.1927228-2- 
>> michael.roth@....com/.
>> https://lore.kernel.org/all/20240126041126.1927228-23- 
>> michael.roth@....com/.
>>
>> It was one thing when there were four of CPU feature masks, but with
>> over 20 it is going to cause mistakes, e.g.,
>> https://lore.kernel.org/lkml/aaed79d5-d683-d1bc-7ba1- 
>> b33c8d6db618@...e.com/.
>>
>> We want to eliminate the stupidly repeated exercise to manually assign
>> features to CPU feature words through introducing an AWK script to
>> automatically generate a header with required and disabled CPU feature
>> masks based on current build config, and this patch set does that.
>>
>> Link to v3:
>> https://lore.kernel.org/lkml/20240622171435.3725548-1-xin@zytor.com/
>>
>> Changes since v3:
>> * Add a few high-level comments to the AWK script (Nikolay Borisov).
>> * Use '1U' instead of '1' in feature mask shifting (Andrew Cooper).
>> * Checking NCAPINTS isn't necessary anymore.  It was needed when these
>>    macros had to be manually updated, but now if cpufeatures.h changes
>>    this header will be regenerated (Brian Gerst).
>> * Enforce CPU feature mask values to be unsigned.
>>
>> Changes since v2:
>> * Remove AWK code that generates extra debugging comments (Brian Gerst).
>> * Move SSE_MASK to verify_cpu.S, the only place it is used (Brian Gerst).
>> * Add a patch to generate macros {REQUIRED|DISABLED}_MASK_BIT_SET in the
>>    new AWK script (Brian Gerst).
>>
>> Changes since v1:
>> * Keep the X86_{REQUIRED,DISABLED}_FEATURE_ prefixes solely in
>>    arch/x86/Kconfig.cpufeatures (Borislav Petkov).
>> * Explain how config option names X86_{REQUIRED,DISABLED}_FEATURE_<name>
>>    are formed (Borislav Petkov).
>> * Remove code generating unused macros {REQUIRED,DISABLED}_FEATURE(x)
>>    to tell if a CPU feature, e.g., X86_FEATURE_FRED, is a required or
>>    disabled feature for this particular compile-time configuration.
>>
> 
> A gentle ping...
> 

Another ping!