[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20200324180207.GD5998@linux.intel.com>
Date: Tue, 24 Mar 2020 11:02:07 -0700
From: Sean Christopherson <sean.j.christopherson@...el.com>
To: Thomas Gleixner <tglx@...utronix.de>
Cc: Xiaoyao Li <xiaoyao.li@...el.com>, Ingo Molnar <mingo@...hat.com>,
Borislav Petkov <bp@...en8.de>, hpa@...or.com,
Paolo Bonzini <pbonzini@...hat.com>, kvm@...r.kernel.org,
x86@...nel.org, linux-kernel@...r.kernel.org,
Andy Lutomirski <luto@...nel.org>,
Peter Zijlstra <peterz@...radead.org>,
Arvind Sankar <nivedita@...m.mit.edu>,
Fenghua Yu <fenghua.yu@...el.com>,
Tony Luck <tony.luck@...el.com>,
Vitaly Kuznetsov <vkuznets@...hat.com>,
Jim Mattson <jmattson@...gle.com>
Subject: Re: [PATCH v5 3/9] x86/split_lock: Re-define the kernel param option
for split_lock_detect
On Tue, Mar 24, 2020 at 11:40:18AM +0100, Thomas Gleixner wrote:
> Xiaoyao Li <xiaoyao.li@...el.com> writes:
> > On 3/24/2020 1:10 AM, Thomas Gleixner wrote:
> >> Xiaoyao Li <xiaoyao.li@...el.com> writes:
> >>
> >>> Change sld_off to sld_disable, which means disabling feature split lock
> >>> detection and it cannot be used in kernel nor can kvm expose it guest.
> >>> Of course, the X86_FEATURE_SPLIT_LOCK_DETECT is not set.
> >>>
> >>> Add a new optioin sld_kvm_only, which means kernel turns split lock
> >>> detection off, but kvm can expose it to guest.
> >>
> >> What's the point of this? If the host is not clean, then you better fix
> >> the host first before trying to expose it to guests.
> >
> > It's not about whether or not host is clean. It's for the cases that
> > users just don't want it enabled on host, to not break the applications
> > or drivers that do have split lock issue.
>
> It's very much about whether the host is split lock clean.
>
> If your host kernel is not, then this wants to be fixed first. If your
> host application is broken, then either fix it or use "warn".
The "kvm only" option was my suggestion. The thought was to provide a way
for users to leverage KVM to debug/test kernels without having to have a
known good kernel and/or to minimize the risk of crashing their physical
system. E.g. debug a misbehaving driver by assigning its associated device
to a guest.
Powered by blists - more mailing lists