Message-Id: <062f7269-580e-4008-904a-919ca0bda482@app.fastmail.com>
Date: Fri, 13 Dec 2024 10:22:28 +0100
From: "Arnd Bergmann" <arnd@...db.de>
To: "Paolo Bonzini" <pbonzini@...hat.com>, "Arnd Bergmann" <arnd@...nel.org>,
 kvm@...r.kernel.org
Cc: "Thomas Bogendoerfer" <tsbogend@...ha.franken.de>,
 "Huacai Chen" <chenhuacai@...nel.org>,
 "Jiaxun Yang" <jiaxun.yang@...goat.com>,
 "Michael Ellerman" <mpe@...erman.id.au>,
 "Nicholas Piggin" <npiggin@...il.com>,
 "Christophe Leroy" <christophe.leroy@...roup.eu>,
 "Naveen N Rao" <naveen@...nel.org>,
 "Madhavan Srinivasan" <maddy@...ux.ibm.com>,
 "Alexander Graf" <graf@...zon.com>, "Crystal Wood" <crwood@...hat.com>,
 "Anup Patel" <anup@...infault.org>,
 "Atish Patra" <atishp@...shpatra.org>,
 "Paul Walmsley" <paul.walmsley@...ive.com>,
 "Palmer Dabbelt" <palmer@...belt.com>,
 "Albert Ou" <aou@...s.berkeley.edu>,
 "Sean Christopherson" <seanjc@...gle.com>,
 "Thomas Gleixner" <tglx@...utronix.de>, "Ingo Molnar" <mingo@...hat.com>,
 "Borislav Petkov" <bp@...en8.de>,
 "Dave Hansen" <dave.hansen@...ux.intel.com>, x86@...nel.org,
 "H. Peter Anvin" <hpa@...or.com>,
 "Vitaly Kuznetsov" <vkuznets@...hat.com>,
 "David Woodhouse" <dwmw2@...radead.org>, "Paul Durrant" <paul@....org>,
 "Marc Zyngier" <maz@...nel.org>, linux-kernel@...r.kernel.org,
 linux-mips@...r.kernel.org, linuxppc-dev@...ts.ozlabs.org,
 kvm-riscv@...ts.infradead.org, linux-riscv@...ts.infradead.org
Subject: Re: [RFC 5/5] x86: kvm drop 32-bit host support

>n Thu, Dec 12, 2024, at 17:27, Paolo Bonzini wrote:
> On 12/12/24 13:55, Arnd Bergmann wrote:
>> From: Arnd Bergmann <arnd@...db.de>
>> 
>> There are very few 32-bit machines that support KVM, the main exceptions
>> are the "Yonah" Generation Xeon-LV and Core Duo from 2006 and the Atom
>> Z5xx "Silverthorne" from 2008 that were all released just before their
>> 64-bit counterparts.
>
> Unlike other architectures where you can't run a "short bitness" kernel 
> at all, or 32-bit systems require hardware enablement that simply does 
> not exist, the x86 situation is a bit different: 32-bit KVM would not be 
> used on 32-bit processors, but on 64-bit kernels running 32-bit kernels; 
> presumably on a machine with 4 or 8 GB of memory, above which you're 
> hurting yourself even more, and for smaller guests where the limitations 
> in userspace address space size don't matter.
>
> Apart from a bunch of CONFIG_X86_64 conditionals, the main issue that 
> KVM has with 32-bit x86 is that they cannot read/write a PTE atomically 
> (i.e. without tearing) and therefore they can't use the newer and more 
> scalable page table management code.  So no objections from me for 
> removing this support, but the justification should be the truth, i.e. 
> developers don't care enough.

Right, I should have updated the description based on the comments
for the first version, especially after separating it from the patches
that make it harder to run 32-bit kernels on 64-bit hardware.

I've updated the changelog now to

    x86: kvm drop 32-bit host support
    
    There are very few 32-bit machines that support KVM, the main exceptions
    are the "Yonah" Generation Xeon-LV and Core Duo from 2006 and the Atom
    Z5xx "Silverthorne" from 2008 that were all released just before their
    64-bit counterparts.
    
    The main usecase for KVM in x86-32 kernels these days is to verify
    that 32-bit KVM is still working, by running it on 64-bit hardware.
    With KVM support on other 32-bit architectures going away, and x86-32
    kernels on 64-bit hardware becoming more limited in available RAM,
    this usecase becomes much less interesting.
    
    Remove this support to make KVM exclusive to 64-bit hosts on all
    architectures, and stop testing 32-bit host mode.
    
    Link: https://lore.kernel.org/all/Z1B1phcpbiYWLgCD@google.com/
    Signed-off-by: Arnd Bergmann <arnd@...db.de>

which assumes that we end up going ahead with the powerpc
patches. Does that work for you?

      Arnd

Powered by blists - more mailing lists