[v2,0/8] x86: Expose consistent topology to guests

Message

Alejandro Vallejo May 8, 2024, 12:39 p.m. UTC

v1 -> v2:

  * v1/patch 4 replaced by a different strategy (See patches 4 and 5 in v2):
      * Have hvmloader populate MADT with the real APIC IDs as read by the APs
        themselves rather than giving it knowledge on how to derive them.
  * Removed patches 2 and 3 in v1, as no longer relevant.
  * Split v1/patch6 in two parts ((a) creating the generator and (b) plugging it
    in) and use the generator in the unit tests of the vcpuid->apicid mapping
    function. Becomes patches 6 and 8 in v2.

Patch 1: Same as v1/patch1.
Patch 2: Header dependency cleanup in preparation for patch3.
Patch 3: Adds vlapic_hidden check for the newly introduced reserved area.
Patch 4: [hvmloader] Replaces INIT+SIPI+SIPI sequences with hypercalls.
Patch 5: [hvmloader] Retrieve the per-CPU APIC IDs from the APs themselves.
Patch 6: Split from v1/patch6.
Patch 7: Logically matching v1/patch5, but using v2/patch6 for testing.
Patch 8: Split from v1/patch6.

=== Original cover letter ===

Current topology handling is close to non-existent. As things stand, APIC
IDs are allocated through the apic_id=vcpu_id*2 relation without giving any
hints to the OS on how to parse the x2APIC ID of a given CPU and assuming
the guest will assume 2 threads per core.

This series involves bringing x2APIC IDs into the migration stream, so
older guests keep operating as they used to and enhancing Xen+toolstack so
new guests get topology information consistent with their x2APIC IDs. As a
side effect of this, x2APIC IDs are now packed and don't have (unless under
a pathological case) gaps.

Further work ought to allow combining this topology configurations with
gang-scheduling of guest hyperthreads into affine physical hyperthreads.
For the time being it purposefully keeps the configuration of "1 socket" +
"1 thread per core" + "1 core per vCPU".

Patch 1: Includes x2APIC IDs in the migration stream. This allows Xen to
         reconstruct the right x2APIC IDs on migrated-in guests, and
         future-proofs itself in the face of x2APIC ID derivation changes.
Patch 2: Minor refactor to expose xc_cpu_policy in libxl
Patch 3: Refactors xen/lib/x86 to work on non-Xen freestanding environments
         (e.g: hvmloader)
Patch 4: Remove old assumptions about vcpu_id<->apic_id relationship in hvmloader
Patch 5: Add logic to derive x2APIC IDs given a CPU policy and vCPU IDs
Patch 6: Includes a simple topology generator for toolstack so new guests
         have topologically consistent information in CPUID

Alejandro Vallejo (8):
  xen/x86: Add initial x2APIC ID to the per-vLAPIC save area
  xen/x86: Simplify header dependencies in x86/hvm
  x86/vlapic: Move lapic_load_hidden migration checks to the check hook
  tools/hvmloader: Wake APs with hypercalls and not with INIT+SIPI+SIPI
  tools/hvmloader: Retrieve (x2)APIC IDs from the APs themselves
  xen/lib: Add topology generator for x86
  xen/x86: Derive topologically correct x2APIC IDs from the policy
  xen/x86: Synthesise domain topologies

 tools/firmware/hvmloader/config.h        |   6 +-
 tools/firmware/hvmloader/hvmloader.c     |   4 +-
 tools/firmware/hvmloader/smp.c           | 151 +++++++++---------
 tools/firmware/hvmloader/util.h          |   5 +
 tools/libs/guest/xg_cpuid_x86.c          |  62 ++------
 tools/tests/cpu-policy/test-cpu-policy.c | 191 +++++++++++++++++++++++
 xen/arch/x86/cpu-policy.c                |   9 +-
 xen/arch/x86/cpuid.c                     |  15 +-
 xen/arch/x86/hvm/irq.c                   |   6 +-
 xen/arch/x86/hvm/vlapic.c                |  75 +++++++--
 xen/arch/x86/include/asm/hvm/hvm.h       |   8 +
 xen/arch/x86/include/asm/hvm/vlapic.h    |   8 +-
 xen/arch/x86/include/asm/hvm/vpt.h       |   1 -
 xen/include/public/arch-x86/hvm/save.h   |   2 +
 xen/include/xen/lib/x86/cpu-policy.h     |  27 ++++
 xen/lib/x86/policy.c                     | 160 +++++++++++++++++++
 16 files changed, 562 insertions(+), 168 deletions(-)