Bits 19:16 of LAR's result are undefined, and some upcoming
improvements to the test case seem to trigger this. Mask off those
bits to avoid spurious failures.
commit 5b781c7e317f ("x86/tls: Forcibly set the accessed bit in TLS
segments") adds a valid case in which LAR's output doesn't quite
agree with set_thread_area()'s input. This isn't triggered in the
test as is, but it will be if we start calling set_thread_area()
with the accessed bit clear. Work around this discrepency.
I've added a Fixes tag so that -stable can pick this up if neccesary.
Fixes: 5b781c7e317f ("x86/tls: Forcibly set the accessed bit in TLS segments")
Signed-off-by: Andy Lutomirski <[email protected]>
---
tools/testing/selftests/x86/ldt_gdt.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/tools/testing/selftests/x86/ldt_gdt.c
b/tools/testing/selftests/x86/ldt_gdt.c
index b9a22f18566a..b2c54f4673f2 100644
--- a/tools/testing/selftests/x86/ldt_gdt.c
+++ b/tools/testing/selftests/x86/ldt_gdt.c
@@ -114,7 +114,15 @@ static void check_valid_segment(uint16_t index, int ldt,
return;
}
- if (ar != expected_ar) {
+ /* The SDM says "bits 19:16 are undefined". Thanks. */
+ ar &= ~0xF0000;
+
+ /*
+ * NB: Different Linux versions do different things with the
+ * accessed bit in set_thread_area().
+ */
+ if (ar != expected_ar &&
+ (ldt || ar != (expected_ar | AR_ACCESSED))) {
printf("[FAIL]\t%s entry %hu has AR 0x%08X but expected
0x%08X\n",
(ldt ? "LDT" : "GDT"), index, ar, expected_ar);
nerrs++;
--
2.13.6
