The same bug that gcc hit in the past is apparently now showing
up with clang, which decides to inline __serpent_setkey_sbox:
crypto/serpent_generic.c:268:5: error: stack frame size of 2112 bytes in function '__serpent_setkey' [-Werror,-Wframe-larger-than=]
Marking it 'noinline' reduces the stack usage from 2112 bytes to
192 and 96 bytes, respectively, and seems to generate more
useful object code.
Fixes: c871c10e4ea7 ("crypto: serpent - improve __serpent_setkey with UBSAN")
Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
---
v2: style improvements suggested by Eric Biggers
---
crypto/serpent_generic.c | 8 +++++++-
1 file changed, 7 insertions(+), 1 deletion(-)
diff --git a/crypto/serpent_generic.c b/crypto/serpent_generic.c
index e57757904677..56fa665a4f01 100644
--- a/crypto/serpent_generic.c
+++ b/crypto/serpent_generic.c
@@ -225,7 +225,13 @@
x4 ^= x2; \
})
-static void __serpent_setkey_sbox(u32 r0, u32 r1, u32 r2, u32 r3, u32 r4, u32 *k)
+/*
+ * both gcc and clang have misoptimized this function in the past,
+ * producing horrible object code from spilling temporary variables
+ * on the stack. Forcing this part out of line avoids that.
+ */
+static noinline void __serpent_setkey_sbox(u32 r0, u32 r1, u32 r2,
+ u32 r3, u32 r4, u32 *k)
{
k += 100;
S3(r3, r4, r0, r1, r2); store_and_load_keys(r1, r2, r4, r3, 28, 24);
--
2.20.0
