linux-kernel - Re: [PATCH 3/7] x86/cpu: Disable kernel LASS when patching kernel alternatives

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: <243f684d-93a3-3aa4-cfc7-cfc0f9fd53cd@intel.com>
Date:   Tue, 31 Jan 2023 18:10:10 -0800
From:   Sohil Mehta <sohil.mehta@...el.com>
To:     Dave Hansen <dave.hansen@...el.com>,
        "Chen, Yian" <yian.chen@...el.com>, <linux-kernel@...r.kernel.org>,
        <x86@...nel.org>, Andy Lutomirski <luto@...nel.org>,
        Dave Hansen <dave.hansen@...ux.intel.com>,
        Ravi Shankar <ravi.v.shankar@...el.com>,
        Tony Luck <tony.luck@...el.com>,
        Paul Lai <paul.c.lai@...el.com>
Subject: Re: [PATCH 3/7] x86/cpu: Disable kernel LASS when patching kernel
 alternatives

On 1/11/2023 4:37 PM, Dave Hansen wrote:
> clearcpuid=smap means that the kernel should be running as if
> CPUID.(EAX=07H,ECX=0H):EBX.SMAP[bit 20]==0.  STAC/CLAC should #UD in
> that case.
> 
> The only reason that it happens to work is that STAC/CLAC apparently
> actually continue to work even if CR4.SMAP==0.
> 

It seems this is by design. I was trying to experiment with
clearcpuid=smap on an older platform (KBL). I noticed that even if
CR4.SMAP==0 the STAC/CLAC instructions do not fault.

The STAC instruction operation in the SDM also suggests that it may be
intentional. It does *not* list CR4.SMAP=0 as a reason for #UD.

#UD	If the LOCK prefix is used.
	If the CPL > 0.
	If CPUID.(EAX=07H, ECX=0H):EBX.SMAP[bit 20] = 0.

> I'm actually a _bit_ surprised by this, but I bet there's a good reason
> for it.
> 

I would love to find out the actual reasoning behind this. I'll try to
poke some of the architects internally.

> In any case, please just make LASS dependent on SMAP.  It's the right
> thing to do on several levels.

Anyway, the end result still remains the same. We should still make LASS
dependent on SMAP since:

1) The spec says that LASS enforcement only happens when SMAP is enabled.
"A supervisor-mode data access causes a LASS violation only if
supervisor-mode access protection is enabled (because CR4.SMAP = 1)"

2) In the extremely improbably case that LASS is available but SMAP is
not available on the hardware, the STAC and CLAC instructions will fault
due to the missing SMAP CPUID bit.

-Sohil