Take an 'unsigned long' instead of 'hpa_t' in the recently added vmsave()
helper, as loading a 64-bit GPR isn't possible in 32-bit mode. This is
properly reflected in the SVM ISA, which explicitly states that VMSAVE,
VMLOAD, VMRUN, etc... consume rAX based on the effective address size.
Don't bother with a WARN to detect breakage on 32-bit KVM, the VMCB PA is
stored as an 'unsigned long', i.e. the bad address is long since gone.
Not to mention that a 32-bit kernel is completely hosed if alloc_page()
hands out pages in high memory.
Reported-by: kernel test robot <lkp@xxxxxxxxx>
Cc: Robert Hu <robert.hu@xxxxxxxxx>
Cc: Farrah Chen <farrah.chen@xxxxxxxxx>
Cc: Danmei Wei <danmei.wei@xxxxxxxxx>
Cc: Tom Lendacky <Thomas.Lendacky@xxxxxxx>
Signed-off-by: Sean Christopherson <seanjc@xxxxxxxxxx>
---
arch/x86/kvm/svm/svm_ops.h | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/arch/x86/kvm/svm/svm_ops.h b/arch/x86/kvm/svm/svm_ops.h
index 0c8377aee52c..9f007bc8409a 100644
--- a/arch/x86/kvm/svm/svm_ops.h
+++ b/arch/x86/kvm/svm/svm_ops.h
@@ -51,7 +51,12 @@ static inline void invlpga(unsigned long addr, u32 asid)
svm_asm2(invlpga, "c"(asid), "a"(addr));
}
-static inline void vmsave(hpa_t pa)
+/*
+ * Despite being a physical address, the portion of rAX that is consumed by
+ * VMSAVE, VMLOAD, etc... is still controlled by the effective address size,
+ * hence 'unsigned long' instead of 'hpa_t'.
+ */
+static inline void vmsave(unsigned long pa)
{
svm_asm1(vmsave, "a" (pa), "memory");
}
--
2.30.0.365.g02bc693789-goog
