ARM Cortex-A72 cores running in 32-bit mode are affected by a silicon
erratum (1655431: ELR recorded incorrectly on interrupt taken between
cryptographic instructions in a sequence [0]) where the second instruction
of a AES instruction pair may execute twice if an interrupt is taken right
after the first instruction consumes an input register of which a single
32-bit lane has been updated the last time it was modified.
This is not such a rare occurrence as it may seem: in counter mode, only
the least significant 32-bit word is incremented in the absence of a
carry, which makes our counter mode implementation susceptible to the
erratum.
So let's shuffle the counter assignments around a bit so that the most
recent updates when the AES instruction pair executes are 128-bit wide.
[0] ARM-EPM-012079 v11.0 Cortex-A72 MPCore Software Developers Errata Notice
Cc: <stable@xxxxxxxxxxxxxxx> # v5.4+
Signed-off-by: Ard Biesheuvel <ardb@xxxxxxxxxx>
---
arch/arm/crypto/aes-ce-core.S | 20 ++++++++++----------
1 file changed, 10 insertions(+), 10 deletions(-)
diff --git a/arch/arm/crypto/aes-ce-core.S b/arch/arm/crypto/aes-ce-core.S
index 4d1707388d94..c0ef9680d90b 100644
--- a/arch/arm/crypto/aes-ce-core.S
+++ b/arch/arm/crypto/aes-ce-core.S
@@ -386,20 +386,20 @@ ENTRY(ce_aes_ctr_encrypt)
.Lctrloop4x:
subs r4, r4, #4
bmi .Lctr1x
- add r6, r6, #1
+ add ip, r6, #1
vmov q0, q7
+ rev ip, ip
+ add lr, r6, #2
+ vmov s31, ip
+ add ip, r6, #3
+ rev lr, lr
vmov q1, q7
- rev ip, r6
- add r6, r6, #1
+ vmov s31, lr
+ rev ip, ip
vmov q2, q7
- vmov s7, ip
- rev ip, r6
- add r6, r6, #1
+ vmov s31, ip
+ add r6, r6, #4
vmov q3, q7
- vmov s11, ip
- rev ip, r6
- add r6, r6, #1
- vmov s15, ip
vld1.8 {q4-q5}, [r1]!
vld1.8 {q6}, [r1]!
vld1.8 {q15}, [r1]!
--
2.17.1
