Fix a bug introduced by the UEFI support where setup_tss() can race with enable_x2apic() and crash the test due to attempting to read an x2APIC MSR prior to enabling x2APIC (failure manifests in smptest3). In an attempt to avoid similar bugs in the future, clean up the per-cpu stuff and convert apic_ops into a per-cpu pointer. 32-bit KUT has a chick-and-egg problem due to using the APIC ID to choose the selector for GS (the per-cpu segment), so the original bug "has" to say on a dedicated helper. Sean Christopherson (8): x86: Always use legacy xAPIC to get APIC ID during TSS setup x86: nVMX: Load actual GS.base for both guest and host x86: smp: Replace spaces with tabs x86: desc: Replace spaces with tabs x86: Add proper helpers for per-cpu reads/writes x86: apic: Replace spaces with tabs x86: apic: Track APIC ops on a per-cpu basis x86: apic: Make xAPIC and I/O APIC pointers static lib/x86/apic-defs.h | 3 +- lib/x86/apic.c | 157 ++++++++++++++++++++++++-------------------- lib/x86/apic.h | 5 +- lib/x86/desc.c | 120 ++++++++++++++++----------------- lib/x86/desc.h | 68 +++++++++---------- lib/x86/setup.c | 4 +- lib/x86/smp.c | 128 +++++++++++++++++------------------- lib/x86/smp.h | 67 +++++++++++++++++++ x86/vmx.c | 4 +- 9 files changed, 311 insertions(+), 245 deletions(-) base-commit: 3df301615cead4142fe28629d86142de32fc6768 -- 2.35.0.rc0.227.g00780c9af4-goog
