public inbox for kvm@vger.kernel.org
 help / color / mirror / Atom feed
From: Sean Christopherson <seanjc@google.com>
To: David Woodhouse <dwmw2@infradead.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>,
	"kvm@vger.kernel.org" <kvm@vger.kernel.org>
Subject: Re: [PATCH] KVM: x86: Reinstate userspace hypercall support
Date: Mon, 30 Nov 2020 20:46:51 +0000	[thread overview]
Message-ID: <X8VaO9DxaaKP7PFl@google.com> (raw)
In-Reply-To: <1bde4a992be29581e559f7a57818e206e11f84f5.camel@infradead.org>

On Sat, Nov 28, 2020, David Woodhouse wrote:

...

> +static int complete_userspace_hypercall(struct kvm_vcpu *vcpu)
> +{
> +	kvm_rax_write(vcpu, vcpu->run->hypercall.ret);
> +
> +	return kvm_skip_emulated_instruction(vcpu);

This should probably verify the linear RIP is unchanged before modifying guest
state, e.g. to let userspace reset the vCPU in response to a hypercall.  See
complete_fast_pio_out().

> +}
> +
> +int kvm_userspace_hypercall(struct kvm_vcpu *vcpu)
> +{
> +	struct kvm_run *run = vcpu->run;
> +
> +	if (is_long_mode(vcpu)) {
> +		run->hypercall.longmode = 1;

Should this also capture the CPL?  I assume the motivation for grabbing regs
and longmode is to avoid having to call back into KVM on every hypercall, and I
also assume there are (or will be) hypercalls that are CPL0 only.

> +		run->hypercall.nr = kvm_rax_read(vcpu);
> +		run->hypercall.args[0] = kvm_rdi_read(vcpu);
> +		run->hypercall.args[1] = kvm_rsi_read(vcpu);
> +		run->hypercall.args[2] = kvm_rdx_read(vcpu);
> +		run->hypercall.args[3] = kvm_r10_read(vcpu);
> +		run->hypercall.args[4] = kvm_r8_read(vcpu);
> +		run->hypercall.args[5] = kvm_r9_read(vcpu);
> +		run->hypercall.ret = -KVM_ENOSYS;
> +	} else {
> +		run->hypercall.longmode = 0;
> +		run->hypercall.nr = (u32)kvm_rbx_read(vcpu);
> +		run->hypercall.args[0] = (u32)kvm_rbx_read(vcpu);
> +		run->hypercall.args[1] = (u32)kvm_rcx_read(vcpu);
> +		run->hypercall.args[2] = (u32)kvm_rdx_read(vcpu);
> +		run->hypercall.args[3] = (u32)kvm_rsi_read(vcpu);
> +		run->hypercall.args[4] = (u32)kvm_rdi_read(vcpu);
> +		run->hypercall.args[5] = (u32)kvm_rbp_read(vcpu);
> +		run->hypercall.ret = (u32)-KVM_ENOSYS;
> +	}

Why not get/set all GPRs (except maybe RIP and RSP) as opposed to implementing 
what I presume is Xen's ABI in a generic KVM user exit?  Copying 10 more GPRs
to/from memory is likely lost in the noise relative to the cost of the userspace
roundtrip.

> +	run->exit_reason = KVM_EXIT_HYPERCALL;
> +	vcpu->arch.complete_userspace_io = complete_userspace_hypercall;
> +
> +	return 0;
> +}
> +
>  int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
>  {
>  	unsigned long nr, a0, a1, a2, a3, ret;
>  	int op_64_bit;
>  
> +	if (vcpu->kvm->arch.user_space_hypercall)
> +		return kvm_userspace_hypercall(vcpu);
> +
>  	if (kvm_hv_hypercall_enabled(vcpu->kvm))
>  		return kvm_hv_hypercall(vcpu);
>  

  parent reply	other threads:[~2020-11-30 20:47 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-11-28 14:20 [PATCH] KVM: x86: Reinstate userspace hypercall support David Woodhouse
2020-11-28 17:50 ` David Woodhouse
2020-11-28 23:16 ` Joao Martins
2020-11-29  2:05   ` David Woodhouse
2020-11-30 18:36 ` Paolo Bonzini
2020-11-30 18:59   ` David Woodhouse
2020-11-30 20:46 ` Sean Christopherson [this message]
2020-11-30 22:42   ` David Woodhouse
2020-12-02  9:20     ` Paolo Bonzini
2020-12-02 14:23       ` Joao Martins

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=X8VaO9DxaaKP7PFl@google.com \
    --to=seanjc@google.com \
    --cc=dwmw2@infradead.org \
    --cc=kvm@vger.kernel.org \
    --cc=pbonzini@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox