linux-kernel - Re: [PATCH] Ability to read the MKTME status from userspace

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 for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Fri, 19 Jun 2020 07:23:40 -0700
From:   Dave Hansen <dave.hansen@...el.com>
To:     Richard Hughes <hughsient@...il.com>
Cc:     Daniel Gutson <daniel@...ypsium.com>,
        Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>,
        x86@...nel.org, "H. Peter Anvin" <hpa@...or.com>,
        Arnd Bergmann <arnd@...db.de>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Peter Zijlstra <peterz@...radead.org>,
        "David S. Miller" <davem@...emloft.net>,
        Rob Herring <robh@...nel.org>, Tony Luck <tony.luck@...el.com>,
        Rahul Tanwar <rahul.tanwar@...ux.intel.com>,
        Xiaoyao Li <xiaoyao.li@...el.com>,
        Sean Christopherson <sean.j.christopherson@...el.com>,
        Dave Hansen <dave.hansen@...ux.intel.com>,
        linux-kernel <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] Ability to read the MKTME status from userspace

On 6/19/20 7:09 AM, Richard Hughes wrote:
> On Fri, 19 Jun 2020 at 14:58, Dave Hansen <dave.hansen@...el.com> wrote:
>>> Right, but for the most part you'd agree that a machine with
>>> functioning TME and encrypted swap partition is more secure than a
>>> machine without TME?
>>
>> Nope.  There might be zero memory connected to the memory controller
>> that supports TME.
> 
> So you're saying that a machine with TME available and enabled is not
> considered more secure than a machine without TME?

Yes, it is not necessarily more secure.

> What I want to do is have a sliding scale of TME not available < TME
> available but disabled < TME available and enabled < TME available,
> enabled and being used. The extra nugget of data gets me from state 2
> to state 3.

I'd assert that availability tells you nothing if you don't pair it with
use.

Last night, I asked my kids if they brushed their teeth.  They said:
"Dad, my toothbrush was available."  They argued that mere availability
was a better situation than not *having* a toothbrush.  They were
logically right, of course, but they still got cavities.

>>> Can I use TME if the CPU supports it, but the platform has disabled
>>> it? How do I know that my system is actually *using* the benefits the
>>> TME feature provides?
>>
>> You must have a system with UEFI 2.8, ensure TME is enabled, then make
>> sure the OS parses EFI_MEMORY_CPU_CRYPTO, then ensure you request that
>> you data be placed in a region marked with EFI_MEMORY_CPU_CRYPTO, and
>> that it be *kept* there (hint: NUMA APIs don't do this).
> 
> So my take-away from that is that it's currently impossible to
> actually say if your system is *actually* using TME.

Not in a generic way, and it can't be derived from cpuid or MSRs alone.

Let's say I'm buying a fleet of servers.  I know I'm buying some fancy
Xeon with TME, I know I'm only using DRAM for storing user data, and I
don't have any accelerators around.  I control and enforce my BIOS
settings.  I'm pretty sure I'm using TME, but I didn't become sure from
poking at sysfs.