Explicitly zero cr4 in prepare_64() instead of "zeroing" it in the
common enter_long_mode(). Clobbering cr4 in enter_long_mode() breaks
switch_to_5level(), which sets cr4.LA57 before calling enter_long_mode()
and obviously expects cr4 to be preserved.
Fixes: d86ef58 ("cstart: do not assume CR4 starts as zero")
Cc: Nadav Amit <namit@xxxxxxxxxx>
Cc: Paolo Bonzini <pbonzini@xxxxxxxxxx>
Signed-off-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
---
Two lines of code, two bugs. I'm pretty sure Paolo should win some kind
of award. :-D
x86/cstart64.S | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/x86/cstart64.S b/x86/cstart64.S
index 3ae98d3..2d16688 100644
--- a/x86/cstart64.S
+++ b/x86/cstart64.S
@@ -175,8 +175,12 @@ prepare_64:
lgdt gdt64_desc
setup_segments
+ xor %eax, %eax
+ mov %eax, %cr4
+
enter_long_mode:
- mov $(1 << 5), %eax // pae
+ mov %cr4, %eax
+ bts $5, %eax // pae
mov %eax, %cr4
mov pt_root, %eax
--
2.26.0
