Christophe Leroy wrote:

Le 01/10/2021 à 23:14, Naveen N. Rao a écrit :

quoted

We aren't handling subtraction involving an immediate value of
0x80000000 properly. Fix the same.

Fixes: 156d0e290e969c ("powerpc/ebpf/jit: Implement JIT compiler for extended BPF")
Signed-off-by: Naveen N. Rao <redacted>
---
  arch/powerpc/net/bpf_jit_comp64.c | 16 ++++++++--------
  1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/arch/powerpc/net/bpf_jit_comp64.c b/arch/powerpc/net/bpf_jit_comp64.c
index ffb7a2877a8469..4641a50e82d50d 100644
--- a/arch/powerpc/net/bpf_jit_comp64.c
+++ b/arch/powerpc/net/bpf_jit_comp64.c

@@ -333,15 +333,15 @@ int bpf_jit_build_body(struct bpf_prog *fp, u32 *image, struct codegen_context *
  		case BPF_ALU | BPF_SUB | BPF_K: /* (u32) dst -= (u32) imm */
  		case BPF_ALU64 | BPF_ADD | BPF_K: /* dst += imm */
  		case BPF_ALU64 | BPF_SUB | BPF_K: /* dst -= imm */
-			if (BPF_OP(code) == BPF_SUB)
-				imm = -imm;
-			if (imm) {
-				if (imm >= -32768 && imm < 32768)
-					EMIT(PPC_RAW_ADDI(dst_reg, dst_reg, IMM_L(imm)));
-				else {
-					PPC_LI32(b2p[TMP_REG_1], imm);
+			if (imm > -32768 && imm < 32768) {
+				EMIT(PPC_RAW_ADDI(dst_reg, dst_reg,
+					BPF_OP(code) == BPF_SUB ? IMM_L(-imm) : IMM_L(imm)));
+			} else {
+				PPC_LI32(b2p[TMP_REG_1], imm);
+				if (BPF_OP(code) == BPF_SUB)
+					EMIT(PPC_RAW_SUB(dst_reg, dst_reg, b2p[TMP_REG_1]));
+				else
  					EMIT(PPC_RAW_ADD(dst_reg, dst_reg, b2p[TMP_REG_1]));
-				}
  			}
  			goto bpf_alu32_trunc;

There is now so few code common to both BPF_ADD and BPF_SUB that you 
should make them different cases.

While at it, why not also use ADDIS if imm is 32 bits ? That would be an 
ADDIS/ADDI instead of LIS/ORI/ADD

Sure. I wanted to limit the change for this fix. We can do a separate 
patch to optimize code generation for BPF_ADD.


- Naveen