An early(ish) pull request for 6.9 (not 6.8!) to clean up KVM's async #PF guest/host ABI. Due to a goof many years ago, the structure shared between the guest and host was expanded to 68 bytes, not the intended 64 bytes (to fit in a cache line). Rather than document the goof, just drop the problematic 4 bytes from the ABI as KVM-the-host never actually used them. The following changes since commit 41bccc98fb7931d63d03f326a746ac4d429c1dd3: Linux 6.8-rc2 (2024-01-28 17:01:12 -0800) are available in the Git repository at: https://github.com/kvm-x86/linux.git tags/kvm-x86-asyncpf_abi-6.9 for you to fetch changes up to df01f0a1165c35e95b5f52c7ba25c19020352ff9: KVM: x86: Improve documentation of MSR_KVM_ASYNC_PF_EN (2024-02-06 10:59:52 -0800) ---------------------------------------------------------------- Guest-side KVM async #PF ABI cleanup for 6.9 Delete kvm_vcpu_pv_apf_data.enabled to fix a goof in KVM's async #PF ABI where the enabled field pushes the size of "struct kvm_vcpu_pv_apf_data" from 64 to 68 bytes, i.e. beyond a single cache line. The enabled field is purely a guest-side flag that Linux-as-a-guest uses to track whether or not the guest has enabled async #PF support. The actual flag that is passed to the host, i.e. to KVM proper, is a single bit in a synthetic MSR, MSR_KVM_ASYNC_PF_EN, i.e. is in a location completely unrelated to the shared kvm_vcpu_pv_apf_data structure. Simply drop the the field and use a dedicated guest-side per-CPU variable to fix the ABI, as opposed to fixing the documentation to match reality. KVM has never consumed kvm_vcpu_pv_apf_data.enabled, so the odds of the ABI change breaking anything are extremely low. ---------------------------------------------------------------- Xiaoyao Li (2): x86/kvm: Use separate percpu variable to track the enabling of asyncpf KVM: x86: Improve documentation of MSR_KVM_ASYNC_PF_EN Documentation/virt/kvm/x86/msr.rst | 19 +++++++++---------- arch/x86/include/uapi/asm/kvm_para.h | 1 - arch/x86/kernel/kvm.c | 11 ++++++----- 3 files changed, 15 insertions(+), 16 deletions(-)
