Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <7e859e72-4624-73c7-8195-c252ace0a49d@intel.com>
Date:   Thu, 12 Jan 2023 10:17:11 -0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.6.1
Subject: Re: [PATCH 5/7] x86/cpu: Enable LASS (Linear Address Space
 Separation)
Content-Language: en-US
To:     Yian Chen <yian.chen@intel.com>, linux-kernel@vger.kernel.org,
        x86@kernel.org, Andy Lutomirski <luto@kernel.org>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        Ravi Shankar <ravi.v.shankar@intel.com>,
        Tony Luck <tony.luck@intel.com>,
        Sohil Mehta <sohil.mehta@intel.com>,
        Paul Lai <paul.c.lai@intel.com>
References: <20230110055204.3227669-1-yian.chen@intel.com>
 <20230110055204.3227669-6-yian.chen@intel.com>
From:   Dave Hansen <dave.hansen@intel.com>
In-Reply-To: <20230110055204.3227669-6-yian.chen@intel.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 1/9/23 21:52, Yian Chen wrote:
> +static __always_inline void setup_lass(struct cpuinfo_x86 *c)
> +{
> +	if (cpu_feature_enabled(X86_FEATURE_LASS)) {
> +		cr4_set_bits(X86_CR4_LASS);
> +	} else {
> +		/*
> +		 * only clear the feature and cr4 bits when hardware
> +		 * supports LASS, in case it was enabled in a previous
> +		 * boot (e.g., via kexec)
> +		 */
> +		if (cpu_has(c, X86_FEATURE_LASS)) {
> +			cr4_clear_bits(X86_CR4_LASS);
> +			clear_cpu_cap(c, X86_FEATURE_LASS);
> +		}
> +	}
> +}

Could you try testing this, please?

Please remember that there are three things in play here:
 1. disabled-features.h.  Makes cpu_feature_enabled(X86_FEATURE_LASS)==0
    when CONFIG_X86_LASS=n.
 2. The X86_FEATURE_LASS software feature bit itself.  clearcpuid=lass
    would clear it.
 3. The actual CPU enumeration of X86_FEATURE_LASS

The else{} is handling the case where X86_FEATURE_LASS is compiled out
but where the CPU supports LASS.  It doesn't do anything when the CPU
lacks LASS support *OR* when clearcpuid=lass is used.

In the end, want X86_CR4_LASS set when the kernel wants LASS and clear
in *ALL* other cases.  That would be simply:

	if (cpu_feature_enabled(X86_FEATURE_LASS)) {
		cr4_set_bits(X86_CR4_LASS);
	} else {
		cr4_clear_bits(X86_CR4_LASS);
	}

The cr4_clear_bits(X86_CR4_LASS) should be safe regardless of CPU or
kernel support for LASS.

As for the:

	clear_cpu_cap(c, X86_FEATURE_LASS);

It really only matters for kernels where CONFIG_X86_LASS=n but where the
CPU supports it.  I'm not clear on what specifically you expect to gain
from it, though.

I'm also wondering if we even want a Kconfig option.  Is anyone
realistically going to be compiling this support out?