[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20191015161427.GC15015@linux.intel.com>
Date: Tue, 15 Oct 2019 09:14:28 -0700
From: Sean Christopherson <sean.j.christopherson@...el.com>
To: Vitaly Kuznetsov <vkuznets@...hat.com>
Cc: Paolo Bonzini <pbonzini@...hat.com>,
Xiaoyao Li <xiaoyao.li@...el.com>, kvm@...r.kernel.org,
linux-kernel@...r.kernel.org, Jim Mattson <jmattson@...gle.com>
Subject: Re: [PATCH] KVM: X86: Make fpu allocation a common function
On Tue, Oct 15, 2019 at 04:36:57PM +0200, Vitaly Kuznetsov wrote:
> Paolo Bonzini <pbonzini@...hat.com> writes:
>
> > On 15/10/19 12:53, Vitaly Kuznetsov wrote:
> >> A very theoretical question: why do we have 'struct vcpu' embedded in
> >> vcpu_vmx/vcpu_svm and not the other way around (e.g. in a union)? That
> >> would've allowed us to allocate memory in common code and then fill in
> >> vendor-specific details in .create_vcpu().
A union would waste a non-trivial amount of memory on SVM.
SVM: struct size = 14560
VMX: struct size = 16192
There are ways around that, but...
> >
> > Probably "because it's always been like that" is the most accurate answer.
> >
>
> OK, so let me make my question a bit less theoretical: would you be in
> favor of changing the status quo? :-)
... we don't need to invert the strut embedding to re-order the create
flow. 'struct kvm_vcpu' must be at offset zero and the size of the vcpu
is vendor defined, so kvm_arch_vcpu_create() can allocate the struct and
directly cast it to a 'struct kvm_vcpu *'.
Powered by blists - more mailing lists