s390/bpf,jit: BPF Just In Time compiler for s390
The s390 implementation of the JIT compiler for packet filter speedup.
Reviewed-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
diff --git a/arch/s390/net/bpf_jit.S b/arch/s390/net/bpf_jit.S
new file mode 100644
index 0000000..7e45d13
--- /dev/null
+++ b/arch/s390/net/bpf_jit.S
@@ -0,0 +1,130 @@
+/*
+ * BPF Jit compiler for s390, help functions.
+ *
+ * Copyright IBM Corp. 2012
+ *
+ * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
+ */
+#include <linux/linkage.h>
+
+/*
+ * Calling convention:
+ * registers %r2, %r6-%r8, %r10-%r11, %r13, %r15 are call saved
+ * %r2: skb pointer
+ * %r3: offset parameter
+ * %r5: BPF A accumulator
+ * %r8: return address
+ * %r9: save register for skb pointer
+ * %r10: skb->data
+ * %r11: skb->len - skb->data_len (headlen)
+ * %r12: BPF X accumulator
+ *
+ * skb_copy_bits takes 4 parameters:
+ * %r2 = skb pointer
+ * %r3 = offset into skb data
+ * %r4 = length to copy
+ * %r5 = pointer to temp buffer
+ */
+#define SKBDATA %r8
+
+ /* A = *(u32 *) (skb->data+K+X) */
+ENTRY(sk_load_word_ind)
+ ar %r3,%r12 # offset += X
+ bmr %r8 # < 0 -> return with cc
+
+ /* A = *(u32 *) (skb->data+K) */
+ENTRY(sk_load_word)
+ llgfr %r1,%r3 # extend offset
+ ahi %r3,4 # offset + 4
+ clr %r11,%r3 # hlen <= offset + 4 ?
+ jl sk_load_word_slow
+ l %r5,0(%r1,%r10) # get word from skb
+ xr %r1,%r1 # set cc to zero
+ br %r8
+
+sk_load_word_slow:
+ lgr %r9,%r2 # save %r2
+ lhi %r4,4 # 4 bytes
+ la %r5,160(%r15) # pointer to temp buffer
+ brasl %r14,skb_copy_bits # get data from skb
+ l %r5,160(%r15) # load result from temp buffer
+ ltgr %r2,%r2 # set cc to (%r2 != 0)
+ lgr %r2,%r9 # restore %r2
+ br %r8
+
+ /* A = *(u16 *) (skb->data+K+X) */
+ENTRY(sk_load_half_ind)
+ ar %r3,%r12 # offset += X
+ bmr %r8 # < 0 -> return with cc
+
+ /* A = *(u16 *) (skb->data+K) */
+ENTRY(sk_load_half)
+ llgfr %r1,%r3 # extend offset
+ ahi %r3,2 # offset + 2
+ clr %r11,%r3 # hlen <= offset + 2 ?
+ jl sk_load_half_slow
+ llgh %r5,0(%r1,%r10) # get half from skb
+ xr %r1,%r1 # set cc to zero
+ br %r8
+
+sk_load_half_slow:
+ lgr %r9,%r2 # save %r2
+ lhi %r4,2 # 2 bytes
+ la %r5,162(%r15) # pointer to temp buffer
+ brasl %r14,skb_copy_bits # get data from skb
+ xc 160(2,%r15),160(%r15)
+ l %r5,160(%r15) # load result from temp buffer
+ ltgr %r2,%r2 # set cc to (%r2 != 0)
+ lgr %r2,%r9 # restore %r2
+ br %r8
+
+ /* A = *(u8 *) (skb->data+K+X) */
+ENTRY(sk_load_byte_ind)
+ ar %r3,%r12 # offset += X
+ bmr %r8 # < 0 -> return with cc
+
+ /* A = *(u8 *) (skb->data+K) */
+ENTRY(sk_load_byte)
+ llgfr %r1,%r3 # extend offset
+ clr %r11,%r3 # hlen < offset ?
+ jle sk_load_byte_slow
+ lhi %r5,0
+ ic %r5,0(%r1,%r10) # get byte from skb
+ xr %r1,%r1 # set cc to zero
+ br %r8
+
+sk_load_byte_slow:
+ lgr %r9,%r2 # save %r2
+ lhi %r4,1 # 1 bytes
+ la %r5,163(%r15) # pointer to temp buffer
+ brasl %r14,skb_copy_bits # get data from skb
+ xc 160(3,%r15),160(%r15)
+ l %r5,160(%r15) # load result from temp buffer
+ ltgr %r2,%r2 # set cc to (%r2 != 0)
+ lgr %r2,%r9 # restore %r2
+ br %r8
+
+ /* A = (*(u8 *)(skb->data+K) & 0xf) << 2 */
+ENTRY(sk_load_byte_msh)
+ llgfr %r1,%r3 # extend offset
+ clr %r11,%r3 # hlen < offset ?
+ jle sk_load_byte_slow
+ lhi %r12,0
+ ic %r12,0(%r1,%r10) # get byte from skb
+ nill %r12,0x0f
+ sll %r12,2
+ xr %r1,%r1 # set cc to zero
+ br %r8
+
+sk_load_byte_msh_slow:
+ lgr %r9,%r2 # save %r2
+ lhi %r4,2 # 2 bytes
+ la %r5,162(%r15) # pointer to temp buffer
+ brasl %r14,skb_copy_bits # get data from skb
+ xc 160(3,%r15),160(%r15)
+ l %r12,160(%r15) # load result from temp buffer
+ nill %r12,0x0f
+ sll %r12,2
+ ltgr %r2,%r2 # set cc to (%r2 != 0)
+ lgr %r2,%r9 # restore %r2
+ br %r8