[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <4d172613-481b-4b45-aa59-165ef1c7dc9b@intel.com>
Date: Sun, 15 Sep 2024 05:58:01 -0700
From: Dave Hansen <dave.hansen@...el.com>
To: John <therealgraysky@...ton.me>
Cc: "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Thomas Gleixner <tglx@...utronix.de>, Ingo Molnar <mingo@...hat.com>,
Borislav Petkov <bp@...en8.de>, Dave Hansen <dave.hansen@...ux.intel.com>,
Unknown <x86@...nel.org>, "H. Peter Anvin" <hpa@...or.com>
Subject: Re: [PATCH] x86: add more x86-64 micro-architecture levels
On 9/15/24 05:25, John wrote:
>> Why is this copied and pasted six times?
>>
>> + depends on (CC_IS_GCC && GCC_VERSION > 110000)...
> I believe the version requirement is needed for each of these new
> options. Please correct me if I am mistaken.
The requirement is fine.
But copying and pasting the same string without refactoring it is not.
You should refactor it:
bool SUPPORT_MARCH_CODEVERS
depends on (CC_IS_GCC && GCC_VERSION > 110000)...
depends on X86_64
and then have each site do this:
+config MINTEL_CPU_V4
+ bool "Intel x86-64-v4"
+ depends on SUPPORT_MARCH_CODEVERS
+ help
...
>> Why are there v4's for both AMD and Intel that do the exact same
>> thing?
>
> I did it this way to selectively include the AMD-specific and
> Intel-specific membership in the config options below. For example,
> the AMD options should be included in the X86_INTEL_USERCOPY config.
I think you mean "the AMD options should *not* be included..."
...
>> Alternatively, anyone wanting to do this could just hack their makefile
>> or (I assume) pass CFLAGS= into the build command-line. Why is
>> something like that insufficient.
>
> I believe this would work:
> export KCFLAGS=' -march=x86-64-v3'
> export KCPPFLAGS=' -march=x86-64-v3'
So why not just have users do that?
>> So, taking a step back: Please convince us that this is something we
>> want to expose to end users in the first place, as opposed to having
>> them hack makefiles or just allowing users a string instead of using the
>> existing CONFIG_M* Kconfig options.
>
> This was just the logical extension of the already included and now
> antiquated options, for example pentium-mmx, k6, etc.
It's probably best not to extend that beast. It really is a relic of
the past and, practically, all of our 64-bit builds are GENERIC_CPU=y
and have been for a long time. We've moved away from the old days where
you could easily compile a kernel that didn't boot.
We're basically handing our users a big long piece of rope with which to
hang themselves here. This patch makes it easy and doesn't do a great
job of explaining why they'd take the risk or what the benefit is.
I don't think we should do this.
Powered by blists - more mailing lists