On Mon, Nov 02, 2026 at 11:27:37AM -0800, Kameron Carr wrote:
On 11/2/2026 4:12 AM, Catalin Marinas wrote:
quoted
On Fri, Oct 30, 2026 at 05:14:08PM -0700, Kameron Carr wrote:
quoted
When booting without ACPI, hyperv_detect_via_smccc() detects Hyper-V by
querying the SMCCC vendor hypervisor UID. In an Arm CCA Realm the SMCCC
call is handled by the RMM rather than by the hypervisor, so the
query never reaches Hyper-V and detection fails.
When running in a Realm, forward ARM_SMCCC_VENDOR_HYP_CALL_UID_FUNC_ID
to the host with RSI_HOST_CALL instead, which is how Hyper-V hypercalls
are issued from a Realm.
This is done in the Hyper-V code rather than in the generic
arm_smccc_hypervisor_has_uuid() because the other users expect different
behavior.
hyperv_init() is an early_initcall and runs before the secondary CPUs
are brought up, so a single __initdata host call structure is used
without locking.
ACPI-based detection and non-Realm guests are unaffected.
Signed-off-by: Kameron Carr <redacted>
This works for me, thanks. I replied to Dexuan clarifying my objection
to v5 do the UUID check generically.
Thank you for your review and suggestions.
Your clarification makes sense.
Is this a "Reviewed-by" or an "Acked-by"?
I don't usually ack the hyperv patches as I don't know enough about it
but given that I commented on this particular patch in v5:
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
--
Catalin