[PATCH v4 9/9] x86/cpuid: Advertise XEN_HVM_CPUID_EXT_DEST_ID when device model opts in

[Julian Vetter <julian.vetter@xxxxxxxxxx>]

Set the XEN_HVM_CPUID_EXT_DEST_ID bit in the HVM hypervisor CPUID leaf
based on the domain-level ext_dest_id flag, which is locked at domain
creation time by taking the AND across all registered ioreq servers that
set XEN_DMOP_IOREQ_SERVER_EXT_DEST_ID. This guarantees that the bit is
only advertised when every active device model will use
XEN_DMOP_bind_pt_msi_irq for passthrough MSIs, so Xen can decode the
extended destination bits from the raw MSI address internally.

After creation_finished the locked d->arch.hvm.ext_dest_id is used
directly, providing a stable and migration-safe value independent of
whether ioreq servers have been re-registered yet. Before
creation_finished the dynamic per-server check is used so toolstack
queries during domain setup reflect the current state.

An old device model that never sets XEN_DMOP_IOREQ_SERVER_EXT_DEST_ID
will keep the bit clear, preserving backwards compatibility with guests
that only use 8-bit APIC destination IDs. A device model that opts in
enables support for x2APIC destination IDs above 255.

Signed-off-by: Julian Vetter <julian.vetter@xxxxxxxxxx>
---
Changes in v4:
- Corresponds to v3 patch 7, updated for the new negotiation mechanism
  introduced in patch 8. The bit now reflects the levelled result
  computed at arch_domain_creation_finished()
    - Before creation_finished(): the dynamic hvm_ext_dest_id_enabled()
      check is used (so toolstack CPUID queries during domain setup see
      the current state)
    - After creation_finished(): the locked d->arch.hvm.ext_dest_id is used
      directly for migration-safe value that does not depend
      on whether ioreq servers have been re-registered on the
      destination host yet
---
 xen/arch/x86/cpuid.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/xen/arch/x86/cpuid.c b/xen/arch/x86/cpuid.c
index 6e9b15c9c3..828aaa9f5b 100644
--- a/xen/arch/x86/cpuid.c
+++ b/xen/arch/x86/cpuid.c
@@ -1,3 +1,4 @@
+#include <xen/ioreq.h>
 #include <xen/sched.h>
 #include <xen/types.h>
 #include <xen/version.h>
@@ -148,6 +149,18 @@ static void cpuid_hypervisor_leaves(const struct vcpu *v, 
uint32_t leaf,
         res->a |= XEN_HVM_CPUID_DOMID_PRESENT;
         res->c = d->domain_id;
 
+        /*
+         * Advertise extended destination ID support. Once domain creation has
+         * finished the locked domain flag is authoritative (ensuring a
+         * consistent view across migration, before the device model
+         * re-registers its ioreq servers on the destination). Before
+         * creation_finished the dynamic per-server check is used so that
+         * toolstack queries during domain setup reflect the current state.
+         */
+        if ( d->creation_finished ? d->arch.hvm.ext_dest_id
+                                  : hvm_ext_dest_id_enabled(d) )
+            res->a |= XEN_HVM_CPUID_EXT_DEST_ID;
+
         /*
          * Per-vCPU event channel upcalls are implemented and work
          * correctly with PIRQs routed over event channels.
-- 
2.53.0



--
Julian Vetter | Vates Hypervisor & Kernel Developer

XCP-ng & Xen Orchestra - Vates solutions

web: https://vates.tech