linux-kernel - Re: [PATCH RFC 1/3] KVM + perf: Rename *_intel_pt

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: Windows password security audit tool. GUI, reports in PDF.

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

Date:   Mon, 3 Oct 2022 19:39:09 +0000
From:   Sean Christopherson <seanjc@...gle.com>
To:     Like Xu <like.xu.linux@...il.com>
Cc:     Peter Zijlstra <peterz@...radead.org>,
        Kan Liang <kan.liang@...ux.intel.com>,
        Adrian Hunter <adrian.hunter@...el.com>,
        Andi Kleen <ak@...ux.intel.com>,
        Jim Mattson <jmattson@...gle.com>,
        Paolo Bonzini <pbonzini@...hat.com>,
        linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org,
        x86@...nel.org, kvm@...r.kernel.org
Subject: Re: [PATCH RFC 1/3] KVM + perf: Rename *_intel_pt_intr() for generic
 usage

On Mon, Sep 26, 2022, Like Xu wrote:
> From: Like Xu <likexu@...cent.com>
> 
> The perf_guest_info_callbacks is common to KVM, while intel_pt is not,
> not even common to x86. In the VMX context, it makes sense to hook
> up the intel_pt specific hook, and given the uniqueness of this usage,
> calling the  generic callback in the explicit location of the perf context
> is not functionally broken.

But it's extremely misleading.  If I were a developer writing the perf hooks for
a different architecture, I would expect perf_handle_guest_intr() to be called on
_every_ perf interrupt that occurred in the guest.

Genericizing the hook also complicates wiring up the hook and consuming the interrupt
type.  E.g. patch 3 is buggy; it wires up the VMX handler if and only if PT is in
PT_MODE_HOST_GUEST, and then takes a dependency on that buggy behavior by not
checking if Intel PT is supported in the now-generic vmx_handle_guest_intr().

This also doesn't really clean up the API from a non-x86 perspective, it just doesn't
make it any worse, i.e. other architectures are still exposed to an x86-specific hook.

Unless we anticipate ARM or RISC-V (which IIRC is gaining PMU support "soon") needing
to hook into "special" perf interrupts, it might be better to figure out a way to make
the hooks themselves more extensible for per-arch behavior.  E.g similar to
kvm_vcpu and kvm_vcpu_arch, add an embedded arch (or vice versa) struct in
perf_guest_info_callbacks plus a perf-internal arch hook to update static calls,
and use that to wire up handle_intel_pt_int for x86.  It'll require more work up
front, but in theory it will require less maintenance in the long run.

> Rename a bunch of intel_pt_intr() functions to the generic guest_intr().
> No functional change intended.

This changelog never says _why_.  Looking forward, the reason for the rename is
to piggyback the hook for BTS.