s390/bpf: Add s390x eBPF JIT compiler backend

Replace 32 bit BPF JIT backend with new 64 bit eBPF backend.

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

4 files changed, 1234 insertions, 803 deletions

diff --git a/arch/s390/Kconfig b/arch/s390/Kconfig
index a5ced5c3c1e0..53a91425d92d 100644
--- a/arch/s390/Kconfig
+++ b/arch/s390/Kconfig
@@ -115,7 +115,7 @@ config S390
 	select HAVE_ARCH_SECCOMP_FILTER
 	select HAVE_ARCH_TRACEHOOK
 	select HAVE_ARCH_TRANSPARENT_HUGEPAGE
-	select HAVE_BPF_JIT if PACK_STACK
+	select HAVE_BPF_JIT if PACK_STACK && HAVE_MARCH_Z9_109_FEATURES
 	select HAVE_CMPXCHG_DOUBLE
 	select HAVE_CMPXCHG_LOCAL
 	select HAVE_DEBUG_KMEMLEAK
diff --git a/arch/s390/net/bpf_jit.S b/arch/s390/net/bpf_jit.S
index ba44c9f55346..a1c917d881ec 100644
--- a/arch/s390/net/bpf_jit.S
+++ b/arch/s390/net/bpf_jit.S
@@ -1,134 +1,115 @@
 /*
  * BPF Jit compiler for s390, help functions.
  *
- * Copyright IBM Corp. 2012
+ * Copyright IBM Corp. 2012,2015
  *
  * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
+ *	      Michael Holzheu <holzheu@linux.vnet.ibm.com>
  */
+
 #include <linux/linkage.h>
+#include "bpf_jit.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
+ * registers %r7-%r10, %r11,%r13, and %r15 are call saved
+ *
+ * Input (64 bit):
+ *   %r3 (%b2) = offset into skb data
+ *   %r6 (%b5) = return address
+ *   %r7 (%b6) = skb pointer
+ *   %r12      = skb data pointer
+ *
+ * Output:
+ *   %r14= %b0 = return value (read skb value)
+ *
+ * Work registers: %r2,%r4,%r5,%r14
  *
  * skb_copy_bits takes 4 parameters:
  *   %r2 = skb pointer
  *   %r3 = offset into skb data
  *   %r4 = pointer to temp buffer
  *   %r5 = length to copy
+ *   Return value in %r2: 0 = ok
+ *
+ * bpf_internal_load_pointer_neg_helper takes 3 parameters:
+ *   %r2 = skb pointer
+ *   %r3 = offset into data
+ *   %r4 = length to copy
+ *   Return value in %r2: Pointer to data
  */
-#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
-	lgr	%r3,%r1			# offset
-	la	%r4,160(%r15)		# pointer to temp buffer
-	lghi	%r5,4			# 4 bytes
-	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
+#define SKF_MAX_NEG_OFF	-0x200000	/* SKF_LL_OFF from filter.h */
 
-	/* 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
-	lgr	%r3,%r1			# offset
-	la	%r4,162(%r15)		# pointer to temp buffer
-	lghi	%r5,2			# 2 bytes
-	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
+/*
+ * Load SIZE bytes from SKB
+ */
+#define sk_load_common(NAME, SIZE, LOAD)				\
+ENTRY(sk_load_##NAME);							\
+	ltgr	%r3,%r3;		/* Is offset negative? */	\
+	jl	sk_load_##NAME##_slow_neg;				\
+ENTRY(sk_load_##NAME##_pos);						\
+	aghi	%r3,SIZE;		/* Offset + SIZE */		\
+	clg	%r3,STK_OFF_HLEN(%r15);	/* Offset + SIZE > hlen? */	\
+	jh	sk_load_##NAME##_slow;					\
+	LOAD	%r14,-SIZE(%r3,%r12);	/* Get data from skb */		\
+	b	OFF_OK(%r6);		/* Return */			\
+									\
+sk_load_##NAME##_slow:;							\
+	lgr	%r2,%r7;		/* Arg1 = skb pointer */	\
+	aghi	%r3,-SIZE;		/* Arg2 = offset */		\
+	la	%r4,STK_OFF_TMP(%r15);	/* Arg3 = temp bufffer */	\
+	lghi	%r5,SIZE;		/* Arg4 = size */		\
+	brasl	%r14,skb_copy_bits;	/* Get data from skb */		\
+	LOAD	%r14,STK_OFF_TMP(%r15);	/* Load from temp bufffer */	\
+	ltgr	%r2,%r2;		/* Set cc to (%r2 != 0) */	\
+	br	%r6;			/* Return */
 
-	/* A = *(u8 *) (skb->data+K+X) */
-ENTRY(sk_load_byte_ind)
-	ar	%r3,%r12		# offset += X
-	bmr	%r8			# < 0 -> return with cc
+sk_load_common(word, 4, llgf)	/* r14 = *(u32 *) (skb->data+offset) */
+sk_load_common(half, 2, llgh)	/* r14 = *(u16 *) (skb->data+offset) */
 
-	/* A = *(u8 *) (skb->data+K) */
+/*
+ * Load 1 byte from SKB (optimized version)
+ */
+	/* r14 = *(u8 *) (skb->data+offset) */
 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
+	ltgr	%r3,%r3			# Is offset negative?
+	jl	sk_load_byte_slow_neg
+ENTRY(sk_load_byte_pos)
+	clg	%r3,STK_OFF_HLEN(%r15)	# Offset >= hlen?
+	jnl	sk_load_byte_slow
+	llgc	%r14,0(%r3,%r12)	# Get byte from skb
+	b	OFF_OK(%r6)		# Return OK
 
 sk_load_byte_slow:
-	lgr	%r9,%r2			# save %r2
-	lgr	%r3,%r1			# offset
-	la	%r4,163(%r15)		# pointer to temp buffer
-	lghi	%r5,1			# 1 byte
-	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
+	lgr	%r2,%r7			# Arg1 = skb pointer
+					# Arg2 = offset
+	la	%r4,STK_OFF_TMP(%r15)	# Arg3 = pointer to temp buffer
+	lghi	%r5,1			# Arg4 = size (1 byte)
+	brasl	%r14,skb_copy_bits	# Get data from skb
+	llgc	%r14,STK_OFF_TMP(%r15)	# Load result from temp buffer
+	ltgr	%r2,%r2			# Set cc to (%r2 != 0)
+	br	%r6			# Return cc
+
+#define sk_negative_common(NAME, SIZE, LOAD)				\
+sk_load_##NAME##_slow_neg:;						\
+	cgfi	%r3,SKF_MAX_NEG_OFF;					\
+	jl	bpf_error;						\
+	lgr	%r2,%r7;		/* Arg1 = skb pointer */	\
+					/* Arg2 = offset */		\
+	lghi	%r4,SIZE;		/* Arg3 = size */		\
+	brasl	%r14,bpf_internal_load_pointer_neg_helper;		\
+	ltgr	%r2,%r2;						\
+	jz	bpf_error;						\
+	LOAD	%r14,0(%r2);		/* Get data from pointer */	\
+	xr	%r3,%r3;		/* Set cc to zero */		\
+	br	%r6;			/* Return cc */
 
-	/* X = (*(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_msh_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_negative_common(word, 4, llgf)
+sk_negative_common(half, 2, llgh)
+sk_negative_common(byte, 1, llgc)
 
-sk_load_byte_msh_slow:
-	lgr	%r9,%r2			# save %r2
-	lgr	%r3,%r1			# offset
-	la	%r4,163(%r15)		# pointer to temp buffer
-	lghi	%r5,1			# 1 byte
-	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
+bpf_error:
+# force a return 0 from jit handler
+	ltgr	%r15,%r15	# Set condition code
+	br	%r6
diff --git a/arch/s390/net/bpf_jit.h b/arch/s390/net/bpf_jit.h
new file mode 100644
index 000000000000..ba8593a515ba
--- /dev/null
+++ b/arch/s390/net/bpf_jit.h
@@ -0,0 +1,58 @@
+/*
+ * BPF Jit compiler defines
+ *
+ * Copyright IBM Corp. 2012,2015
+ *
+ * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
+ *	      Michael Holzheu <holzheu@linux.vnet.ibm.com>
+ */
+
+#ifndef __ARCH_S390_NET_BPF_JIT_H
+#define __ARCH_S390_NET_BPF_JIT_H
+
+#ifndef __ASSEMBLY__
+
+#include <linux/filter.h>
+#include <linux/types.h>
+
+extern u8 sk_load_word_pos[], sk_load_half_pos[], sk_load_byte_pos[];
+extern u8 sk_load_word[], sk_load_half[], sk_load_byte[];
+
+#endif /* __ASSEMBLY__ */
+
+/*
+ * Stackframe layout (packed stack):
+ *
+ *				    ^ high
+ *	      +---------------+     |
+ *	      | old backchain |     |
+ *	      +---------------+     |
+ *	      |   r15 - r6    |     |
+ * BFP	   -> +===============+     |
+ *	      |		      |     |
+ *	      |   BPF stack   |     |
+ *	      |		      |     |
+ *	      +---------------+     |
+ *	      | 8 byte hlen   |     |
+ * R15+168 -> +---------------+     |
+ *	      | 4 byte align  |     |
+ *	      +---------------+     |
+ *	      | 4 byte temp   |     |
+ *	      | for bpf_jit.S |     |
+ * R15+160 -> +---------------+     |
+ *	      | new backchain |     |
+ * R15+152 -> +---------------+     |
+ *	      | + 152 byte SA |     |
+ * R15	   -> +---------------+     + low
+ *
+ * We get 160 bytes stack space from calling function, but only use
+ * 11 * 8 byte (old backchain + r15 - r6) for storing registers.
+ */
+#define STK_OFF (MAX_BPF_STACK + 8 + 4 + 4 + (160 - 11 * 8))
+#define STK_OFF_TMP	160	/* Offset of tmp buffer on stack */
+#define STK_OFF_HLEN	168	/* Offset of SKB header length on stack */
+
+/* Offset to skip condition code check */
+#define OFF_OK		4
+
+#endif /* __ARCH_S390_NET_BPF_JIT_H */
diff --git a/arch/s390/net/bpf_jit_comp.c b/arch/s390/net/bpf_jit_comp.c
index bbd1981cc150..7690dc8e1ab5 100644
--- a/arch/s390/net/bpf_jit_comp.c
+++ b/arch/s390/net/bpf_jit_comp.c
@@ -1,817 +1,1209 @@
 /*
  * BPF Jit compiler for s390.
  *
- * Copyright IBM Corp. 2012
+ * Minimum build requirements:
+ *
+ *  - HAVE_MARCH_Z196_FEATURES: laal, laalg
+ *  - HAVE_MARCH_Z10_FEATURES: msfi, cgrj, clgrj
+ *  - HAVE_MARCH_Z9_109_FEATURES: alfi, llilf, clfi, oilf, nilf
+ *  - PACK_STACK
+ *  - 64BIT
+ *
+ * Copyright IBM Corp. 2012,2015
  *
  * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>
+ *	      Michael Holzheu <holzheu@linux.vnet.ibm.com>
  */
+
+#define KMSG_COMPONENT "bpf_jit"
+#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
+
 #include <linux/netdevice.h>
-#include <linux/if_vlan.h>
 #include <linux/filter.h>
 #include <linux/init.h>
 #include <asm/cacheflush.h>
-#include <asm/facility.h>
 #include <asm/dis.h>
+#include "bpf_jit.h"
 
-/*
- * Conventions:
- *   %r2 = skb pointer
- *   %r3 = offset parameter
- *   %r4 = scratch register / length 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
- *   %r13 = literal pool pointer
- *   0(%r15) - 63(%r15) scratch memory array with BPF_MEMWORDS
- */
 int bpf_jit_enable __read_mostly;
 
+struct bpf_jit {
+	u32 seen;		/* Flags to remember seen eBPF instructions */
+	u32 seen_reg[16];	/* Array to remember which registers are used */
+	u32 *addrs;		/* Array with relative instruction addresses */
+	u8 *prg_buf;		/* Start of program */
+	int size;		/* Size of program and literal pool */
+	int size_prg;		/* Size of program */
+	int prg;		/* Current position in program */
+	int lit_start;		/* Start of literal pool */
+	int lit;		/* Current position in literal pool */
+	int base_ip;		/* Base address for literal pool */
+	int ret0_ip;		/* Address of return 0 */
+	int exit_ip;		/* Address of exit */
+};
+
+#define BPF_SIZE_MAX	4096	/* Max size for program */
+
+#define SEEN_SKB	1	/* skb access */
+#define SEEN_MEM	2	/* use mem[] for temporary storage */
+#define SEEN_RET0	4	/* ret0_ip points to a valid return 0 */
+#define SEEN_LITERAL	8	/* code uses literals */
+#define SEEN_FUNC	16	/* calls C functions */
+#define SEEN_STACK	(SEEN_FUNC | SEEN_MEM | SEEN_SKB)
+
 /*
- * assembly code in arch/x86/net/bpf_jit.S
+ * s390 registers
  */
-extern u8 sk_load_word[], sk_load_half[], sk_load_byte[], sk_load_byte_msh[];
-extern u8 sk_load_word_ind[], sk_load_half_ind[], sk_load_byte_ind[];
+#define REG_W0		(__MAX_BPF_REG+0)	/* Work register 1 (even) */
+#define REG_W1		(__MAX_BPF_REG+1)	/* Work register 2 (odd) */
+#define REG_SKB_DATA	(__MAX_BPF_REG+2)	/* SKB data register */
+#define REG_L		(__MAX_BPF_REG+3)	/* Literal pool register */
+#define REG_15		(__MAX_BPF_REG+4)	/* Register 15 */
+#define REG_0		REG_W0			/* Register 0 */
+#define REG_2		BPF_REG_1		/* Register 2 */
+#define REG_14		BPF_REG_0		/* Register 14 */
 
-struct bpf_jit {
-	unsigned int seen;
-	u8 *start;
-	u8 *prg;
-	u8 *mid;
-	u8 *lit;
-	u8 *end;
-	u8 *base_ip;
-	u8 *ret0_ip;
-	u8 *exit_ip;
-	unsigned int off_load_word;
-	unsigned int off_load_half;
-	unsigned int off_load_byte;
-	unsigned int off_load_bmsh;
-	unsigned int off_load_iword;
-	unsigned int off_load_ihalf;
-	unsigned int off_load_ibyte;
+/*
+ * Mapping of BPF registers to s390 registers
+ */
+static const int reg2hex[] = {
+	/* Return code */
+	[BPF_REG_0]	= 14,
+	/* Function parameters */
+	[BPF_REG_1]	= 2,
+	[BPF_REG_2]	= 3,
+	[BPF_REG_3]	= 4,
+	[BPF_REG_4]	= 5,
+	[BPF_REG_5]	= 6,
+	/* Call saved registers */
+	[BPF_REG_6]	= 7,
+	[BPF_REG_7]	= 8,
+	[BPF_REG_8]	= 9,
+	[BPF_REG_9]	= 10,
+	/* BPF stack pointer */
+	[BPF_REG_FP]	= 13,
+	/* SKB data pointer */
+	[REG_SKB_DATA]	= 12,
+	/* Work registers for s390x backend */
+	[REG_W0]	= 0,
+	[REG_W1]	= 1,
+	[REG_L]		= 11,
+	[REG_15]	= 15,
 };
 
-#define BPF_SIZE_MAX	4096	/* Max size for program */
+static inline u32 reg(u32 dst_reg, u32 src_reg)
+{
+	return reg2hex[dst_reg] << 4 | reg2hex[src_reg];
+}
+
+static inline u32 reg_high(u32 reg)
+{
+	return reg2hex[reg] << 4;
+}
+
+static inline void reg_set_seen(struct bpf_jit *jit, u32 b1)
+{
+	u32 r1 = reg2hex[b1];
+
+	if (!jit->seen_reg[r1] && r1 >= 6 && r1 <= 15)
+		jit->seen_reg[r1] = 1;
+}
+
+#define REG_SET_SEEN(b1)					\
+({								\
+	reg_set_seen(jit, b1);					\
+})
+
+#define REG_SEEN(b1) jit->seen_reg[reg2hex[(b1)]]
+
+/*
+ * EMIT macros for code generation
+ */
+
+#define _EMIT2(op)						\
+({								\
+	if (jit->prg_buf)					\
+		*(u16 *) (jit->prg_buf + jit->prg) = op;	\
+	jit->prg += 2;						\
+})
 
-#define SEEN_DATAREF	1	/* might call external helpers */
-#define SEEN_XREG	2	/* ebx is used */
-#define SEEN_MEM	4	/* use mem[] for temporary storage */
-#define SEEN_RET0	8	/* pc_ret0 points to a valid return 0 */
-#define SEEN_LITERAL	16	/* code uses literals */
-#define SEEN_LOAD_WORD	32	/* code uses sk_load_word */
-#define SEEN_LOAD_HALF	64	/* code uses sk_load_half */
-#define SEEN_LOAD_BYTE	128	/* code uses sk_load_byte */
-#define SEEN_LOAD_BMSH	256	/* code uses sk_load_byte_msh */
-#define SEEN_LOAD_IWORD	512	/* code uses sk_load_word_ind */
-#define SEEN_LOAD_IHALF	1024	/* code uses sk_load_half_ind */
-#define SEEN_LOAD_IBYTE	2048	/* code uses sk_load_byte_ind */
-
-#define EMIT2(op)					\
-({							\
-	if (jit->prg + 2 <= jit->mid)			\
-		*(u16 *) jit->prg = op;			\
-	jit->prg += 2;					\
+#define EMIT2(op, b1, b2)					\
+({								\
+	_EMIT2(op | reg(b1, b2));				\
+	REG_SET_SEEN(b1);					\
+	REG_SET_SEEN(b2);					\
 })
 
-#define EMIT4(op)					\
-({							\
-	if (jit->prg + 4 <= jit->mid)			\
-		*(u32 *) jit->prg = op;			\
-	jit->prg += 4;					\
+#define _EMIT4(op)						\
+({								\
+	if (jit->prg_buf)					\
+		*(u32 *) (jit->prg_buf + jit->prg) = op;	\
+	jit->prg += 4;						\
 })
 
-#define EMIT4_DISP(op, disp)				\
-({							\
-	unsigned int __disp = (disp) & 0xfff;		\
-	EMIT4(op | __disp);				\
+#define EMIT4(op, b1, b2)					\
+({								\
+	_EMIT4(op | reg(b1, b2));				\
+	REG_SET_SEEN(b1);					\
+	REG_SET_SEEN(b2);					\
 })
 
-#define EMIT4_IMM(op, imm)				\
-({							\
-	unsigned int __imm = (imm) & 0xffff;		\
-	EMIT4(op | __imm);				\
+#define EMIT4_RRF(op, b1, b2, b3)				\
+({								\
+	_EMIT4(op | reg_high(b3) << 8 | reg(b1, b2));		\
+	REG_SET_SEEN(b1);					\
+	REG_SET_SEEN(b2);					\
+	REG_SET_SEEN(b3);					\
 })
 
-#define EMIT4_PCREL(op, pcrel)				\
-({							\
-	long __pcrel = ((pcrel) >> 1) & 0xffff;		\
-	EMIT4(op | __pcrel);				\
+#define _EMIT4_DISP(op, disp)					\
+({								\
+	unsigned int __disp = (disp) & 0xfff;			\
+	_EMIT4(op | __disp);					\
 })
 
-#define EMIT6(op1, op2)					\
-({							\
-	if (jit->prg + 6 <= jit->mid) {			\
-		*(u32 *) jit->prg = op1;		\
-		*(u16 *) (jit->prg + 4) = op2;		\
-	}						\
-	jit->prg += 6;					\
+#define EMIT4_DISP(op, b1, b2, disp)				\
+({								\
+	_EMIT4_DISP(op | reg_high(b1) << 16 |			\
+		    reg_high(b2) << 8, disp);			\
+	REG_SET_SEEN(b1);					\
+	REG_SET_SEEN(b2);					\
 })
 
-#define EMIT6_DISP(op1, op2, disp)			\
-({							\
-	unsigned int __disp = (disp) & 0xfff;		\
-	EMIT6(op1 | __disp, op2);			\
+#define EMIT4_IMM(op, b1, imm)					\
+({								\
+	unsigned int __imm = (imm) & 0xffff;			\
+	_EMIT4(op | reg_high(b1) << 16 | __imm);		\
+	REG_SET_SEEN(b1);					\
 })
 
-#define EMIT6_IMM(op, imm)				\
-({							\
-	unsigned int __imm = (imm);			\
-	EMIT6(op | (__imm >> 16), __imm & 0xffff);	\
+#define EMIT4_PCREL(op, pcrel)					\
+({								\
+	long __pcrel = ((pcrel) >> 1) & 0xffff;			\
+	_EMIT4(op | __pcrel);					\
 })
 
-#define EMIT_CONST(val)					\
-({							\
-	unsigned int ret;				\
-	ret = (unsigned int) (jit->lit - jit->base_ip);	\
-	jit->seen |= SEEN_LITERAL;			\
-	if (jit->lit + 4 <= jit->end)			\
-		*(u32 *) jit->lit = val;		\
-	jit->lit += 4;					\
-	ret;						\
+#define _EMIT6(op1, op2)					\
+({								\
+	if (jit->prg_buf) {					\
+		*(u32 *) (jit->prg_buf + jit->prg) = op1;	\
+		*(u16 *) (jit->prg_buf + jit->prg + 4) = op2;	\
+	}							\
+	jit->prg += 6;						\
 })
 
-#define EMIT_FN_CONST(bit, fn)				\
-({							\
-	unsigned int ret;				\
-	ret = (unsigned int) (jit->lit - jit->base_ip);	\
-	if (jit->seen & bit) {				\
-		jit->seen |= SEEN_LITERAL;		\
-		if (jit->lit + 8 <= jit->end)		\
-			*(void **) jit->lit = fn;	\
-		jit->lit += 8;				\
-	}						\
-	ret;						\
+#define _EMIT6_DISP(op1, op2, disp)				\
+({								\
+	unsigned int __disp = (disp) & 0xfff;			\
+	_EMIT6(op1 | __disp, op2);				\
 })
 
-static void bpf_jit_fill_hole(void *area, unsigned int size)
+#define EMIT6_DISP(op1, op2, b1, b2, b3, disp)			\
+({								\
+	_EMIT6_DISP(op1 | reg(b1, b2) << 16 |			\
+		    reg_high(b3) << 8, op2, disp);		\
+	REG_SET_SEEN(b1);					\
+	REG_SET_SEEN(b2);					\
+	REG_SET_SEEN(b3);					\
+})
+
+#define _EMIT6_DISP_LH(op1, op2, disp)				\
+({								\
+	unsigned int __disp_h = ((u32)disp) & 0xff000;		\
+	unsigned int __disp_l = ((u32)disp) & 0x00fff;		\
+	_EMIT6(op1 | __disp_l, op2 | __disp_h >> 4);		\
+})
+
+#define EMIT6_DISP_LH(op1, op2, b1, b2, b3, disp)		\
+({								\
+	_EMIT6_DISP_LH(op1 | reg(b1, b2) << 16 |		\
+		       reg_high(b3) << 8, op2, disp);		\
+	REG_SET_SEEN(b1);					\
+	REG_SET_SEEN(b2);					\
+	REG_SET_SEEN(b3);					\
+})
+
+#define EMIT6_PCREL(op1, op2, b1, b2, i, off, mask)		\
+({								\
+	/* Branch instruction needs 6 bytes */			\
+	int rel = (addrs[i + off + 1] - (addrs[i + 1] - 6)) / 2;\
+	_EMIT6(op1 | reg(b1, b2) << 16 | rel, op2 | mask);	\
+	REG_SET_SEEN(b1);					\
+	REG_SET_SEEN(b2);					\
+})
+
+#define _EMIT6_IMM(op, imm)					\
+({								\
+	unsigned int __imm = (imm);				\
+	_EMIT6(op | (__imm >> 16), __imm & 0xffff);		\
+})
+
+#define EMIT6_IMM(op, b1, imm)					\
+({								\
+	_EMIT6_IMM(op | reg_high(b1) << 16, imm);		\
+	REG_SET_SEEN(b1);					\
+})
+
+#define EMIT_CONST_U32(val)					\
+({								\
+	unsigned int ret;					\
+	ret = jit->lit - jit->base_ip;				\
+	jit->seen |= SEEN_LITERAL;				\
+	if (jit->prg_buf)					\
+		*(u32 *) (jit->prg_buf + jit->lit) = (u32) val;	\
+	jit->lit += 4;						\
+	ret;							\
+})
+
+#define EMIT_CONST_U64(val)					\
+({								\
+	unsigned int ret;					\
+	ret = jit->lit - jit->base_ip;				\
+	jit->seen |= SEEN_LITERAL;				\
+	if (jit->prg_buf)					\
+		*(u64 *) (jit->prg_buf + jit->lit) = (u64) val;	\
+	jit->lit += 8;						\
+	ret;							\
+})
+
+#define EMIT_ZERO(b1)						\
+({								\
+	/* llgfr %dst,%dst (zero extend to 64 bit) */		\
+	EMIT4(0xb9160000, b1, b1);				\
+	REG_SET_SEEN(b1);					\
+})
+
+/*
+ * Fill whole space with illegal instructions
+ */
+static void jit_fill_hole(void *area, unsigned int size)
 {
-	/* Fill whole space with illegal instructions */
 	memset(area, 0, size);
 }
 
-static void bpf_jit_prologue(struct bpf_jit *jit)
+/*
+ * Save registers from "rs" (register start) to "re" (register end) on stack
+ */
+static void save_regs(struct bpf_jit *jit, u32 rs, u32 re)
+{
+	u32 off = 72 + (rs - 6) * 8;
+
+	if (rs == re)
+		/* stg %rs,off(%r15) */
+		_EMIT6(0xe300f000 | rs << 20 | off, 0x0024);
+	else
+		/* stmg %rs,%re,off(%r15) */
+		_EMIT6_DISP(0xeb00f000 | rs << 20 | re << 16, 0x0024, off);
+}
+
+/*
+ * Restore registers from "rs" (register start) to "re" (register end) on stack
+ */
+static void restore_regs(struct bpf_jit *jit, u32 rs, u32 re)
 {
-	/* Save registers and create stack frame if necessary */
-	if (jit->seen & SEEN_DATAREF) {
-		/* stmg %r8,%r15,88(%r15) */
-		EMIT6(0xeb8ff058, 0x0024);
-		/* lgr %r14,%r15 */
-		EMIT4(0xb90400ef);
-		/* aghi %r15,<offset> */
-		EMIT4_IMM(0xa7fb0000, (jit->seen & SEEN_MEM) ? -112 : -80);
-		/* stg %r14,152(%r15) */
-		EMIT6(0xe3e0f098, 0x0024);
-	} else if ((jit->seen & SEEN_XREG) && (jit->seen & SEEN_LITERAL))
-		/* stmg %r12,%r13,120(%r15) */
-		EMIT6(0xebcdf078, 0x0024);
-	else if (jit->seen & SEEN_XREG)
-		/* stg %r12,120(%r15) */
-		EMIT6(0xe3c0f078, 0x0024);
-	else if (jit->seen & SEEN_LITERAL)
-		/* stg %r13,128(%r15) */
-		EMIT6(0xe3d0f080, 0x0024);
+	u32 off = 72 + (rs - 6) * 8;
+
+	if (jit->seen & SEEN_STACK)
+		off += STK_OFF;
+
+	if (rs == re)
+		/* lg %rs,off(%r15) */
+		_EMIT6(0xe300f000 | rs << 20 | off, 0x0004);
+	else
+		/* lmg %rs,%re,off(%r15) */
+		_EMIT6_DISP(0xeb00f000 | rs << 20 | re << 16, 0x0004, off);
+}
 
+/*
+ * Return first seen register (from start)
+ */
+static int get_start(struct bpf_jit *jit, int start)
+{
+	int i;
+
+	for (i = start; i <= 15; i++) {
+		if (jit->seen_reg[i])
+			return i;
+	}
+	return 0;
+}
+
+/*
+ * Return last seen register (from start) (gap >= 2)
+ */
+static int get_end(struct bpf_jit *jit, int start)
+{
+	int i;
+
+	for (i = start; i < 15; i++) {
+		if (!jit->seen_reg[i] && !jit->seen_reg[i + 1])
+			return i - 1;
+	}
+	return jit->seen_reg[15] ? 15 : 14;
+}
+
+#define REGS_SAVE	1
+#define REGS_RESTORE	0
+/*
+ * Save and restore clobbered registers (6-15) on stack.
+ * We save/restore registers in chunks with gap >= 2 registers.
+ */
+static void save_restore_regs(struct bpf_jit *jit, int op)
+{
+
+	int re = 6, rs;
+
+	do {
+		rs = get_start(jit, re);
+		if (!rs)
+			break;
+		re = get_end(jit, rs + 1);
+		if (op == REGS_SAVE)
+			save_regs(jit, rs, re);
+		else
+			restore_regs(jit, rs, re);
+		re++;
+	} while (re <= 15);
+}
+
+/*
+ * Emit function prologue
+ *
+ * Save registers and create stack frame if necessary.
+ * See stack frame layout desription in "bpf_jit.h"!
+ */
+static void bpf_jit_prologue(struct bpf_jit *jit)
+{
+	/* Save registers */
+	save_restore_regs(jit, REGS_SAVE);
 	/* Setup literal pool */
 	if (jit->seen & SEEN_LITERAL) {
 		/* basr %r13,0 */
-		EMIT2(0x0dd0);
+		EMIT2(0x0d00, REG_L, REG_0);
 		jit->base_ip = jit->prg;
 	}
-	jit->off_load_word = EMIT_FN_CONST(SEEN_LOAD_WORD, sk_load_word);
-	jit->off_load_half = EMIT_FN_CONST(SEEN_LOAD_HALF, sk_load_half);
-	jit->off_load_byte = EMIT_FN_CONST(SEEN_LOAD_BYTE, sk_load_byte);
-	jit->off_load_bmsh = EMIT_FN_CONST(SEEN_LOAD_BMSH, sk_load_byte_msh);
-	jit->off_load_iword = EMIT_FN_CONST(SEEN_LOAD_IWORD, sk_load_word_ind);
-	jit->off_load_ihalf = EMIT_FN_CONST(SEEN_LOAD_IHALF, sk_load_half_ind);
-	jit->off_load_ibyte = EMIT_FN_CONST(SEEN_LOAD_IBYTE, sk_load_byte_ind);
-
-	/* Filter needs to access skb data */
-	if (jit->seen & SEEN_DATAREF) {
-		/* l %r11,<len>(%r2) */
-		EMIT4_DISP(0x58b02000, offsetof(struct sk_buff, len));
-		/* s %r11,<data_len>(%r2) */
-		EMIT4_DISP(0x5bb02000, offsetof(struct sk_buff, data_len));
-		/* lg %r10,<data>(%r2) */
-		EMIT6_DISP(0xe3a02000, 0x0004,
-			   offsetof(struct sk_buff, data));
+	/* Setup stack and backchain */
+	if (jit->seen & SEEN_STACK) {
+		/* lgr %bfp,%r15 (BPF frame pointer) */
+		EMIT4(0xb9040000, BPF_REG_FP, REG_15);
+		/* aghi %r15,-STK_OFF */
+		EMIT4_IMM(0xa70b0000, REG_15, -STK_OFF);
+		if (jit->seen & SEEN_FUNC)
+			/* stg %bfp,152(%r15) (backchain) */
+			EMIT6_DISP_LH(0xe3000000, 0x0024, BPF_REG_FP, REG_0,
+				      REG_15, 152);
+	}
+	/*
+	 * For SKB access %b1 contains the SKB pointer. For "bpf_jit.S"
+	 * we store the SKB header length on the stack and the SKB data
+	 * pointer in REG_SKB_DATA.
+	 */
+	if (jit->seen & SEEN_SKB) {
+		/* Header length: llgf %w1,<len>(%b1) */
+		EMIT6_DISP_LH(0xe3000000, 0x0016, REG_W1, REG_0, BPF_REG_1,
+			      offsetof(struct sk_buff, len));
+		/* s %w1,<data_len>(%b1) */
+		EMIT4_DISP(0x5b000000, REG_W1, BPF_REG_1,
+			   offsetof(struct sk_buff, data_len));
+		/* stg %w1,ST_OFF_HLEN(%r0,%r15) */
+		EMIT6_DISP_LH(0xe3000000, 0x0024, REG_W1, REG_0, REG_15,
+			      STK_OFF_HLEN);
+		/* lg %skb_data,data_off(%b1) */
+		EMIT6_DISP_LH(0xe3000000, 0x0004, REG_SKB_DATA, REG_0,
+			      BPF_REG_1, offsetof(struct sk_buff, data));
 	}
+	/* BPF compatibility: clear A (%b7) and X (%b8) registers */
+	if (REG_SEEN(BPF_REG_7))
+		/* lghi %b7,0 */
+		EMIT4_IMM(0xa7090000, BPF_REG_7, 0);
+	if (REG_SEEN(BPF_REG_8))
+		/* lghi %b8,0 */
+		EMIT4_IMM(0xa7090000, BPF_REG_8, 0);
 }
 
+/*
+ * Function epilogue
+ */
 static void bpf_jit_epilogue(struct bpf_jit *jit)
 {
 	/* Return 0 */
 	if (jit->seen & SEEN_RET0) {
 		jit->ret0_ip = jit->prg;
-		/* lghi %r2,0 */
-		EMIT4(0xa7290000);
+		/* lghi %b0,0 */
+		EMIT4_IMM(0xa7090000, BPF_REG_0, 0);
 	}
 	jit->exit_ip = jit->prg;
+	/* Load exit code: lgr %r2,%b0 */
+	EMIT4(0xb9040000, REG_2, BPF_REG_0);
 	/* Restore registers */
-	if (jit->seen & SEEN_DATAREF)
-		/* lmg %r8,%r15,<offset>(%r15) */
-		EMIT6_DISP(0xeb8ff000, 0x0004,
-			   (jit->seen & SEEN_MEM) ? 200 : 168);
-	else if ((jit->seen & SEEN_XREG) && (jit->seen & SEEN_LITERAL))
-		/* lmg %r12,%r13,120(%r15) */
-		EMIT6(0xebcdf078, 0x0004);
-	else if (jit->seen & SEEN_XREG)
-		/* lg %r12,120(%r15) */
-		EMIT6(0xe3c0f078, 0x0004);
-	else if (jit->seen & SEEN_LITERAL)
-		/* lg %r13,128(%r15) */
-		EMIT6(0xe3d0f080, 0x0004);
+	save_restore_regs(jit, REGS_RESTORE);
 	/* br %r14 */
-	EMIT2(0x07fe);
+	_EMIT2(0x07fe);
 }
 
 /*
- * make sure we dont leak kernel information to user
+ * Compile one eBPF instruction into s390x code
  */
-static void bpf_jit_noleaks(struct bpf_jit *jit, struct sock_filter *filter)
+static int bpf_jit_insn(struct bpf_jit *jit, struct bpf_prog *fp, int i)
 {
-	/* Clear temporary memory if (seen & SEEN_MEM) */
-	if (jit->seen & SEEN_MEM)
-		/* xc 0(64,%r15),0(%r15) */
-		EMIT6(0xd73ff000, 0xf000);
-	/* Clear X if (seen & SEEN_XREG) */
-	if (jit->seen & SEEN_XREG)
-		/* lhi %r12,0 */
-		EMIT4(0xa7c80000);
-	/* Clear A if the first register does not set it. */
-	switch (filter[0].code) {
-	case BPF_LD | BPF_W | BPF_ABS:
-	case BPF_LD | BPF_H | BPF_ABS:
-	case BPF_LD | BPF_B | BPF_ABS:
-	case BPF_LD | BPF_W | BPF_LEN:
-	case BPF_LD | BPF_W | BPF_IND:
-	case BPF_LD | BPF_H | BPF_IND:
-	case BPF_LD | BPF_B | BPF_IND:
-	case BPF_LD | BPF_IMM:
-	case BPF_LD | BPF_MEM:
-	case BPF_MISC | BPF_TXA:
-	case BPF_RET | BPF_K:
-		/* first instruction sets A register */
-		break;
-	default: /* A = 0 */
-		/* lhi %r5,0 */
-		EMIT4(0xa7580000);
-	}
-}
+	struct bpf_insn *insn = &fp->insnsi[i];
+	int jmp_off, last, insn_count = 1;
+	unsigned int func_addr, mask;
+	u32 dst_reg = insn->dst_reg;
+	u32 src_reg = insn->src_reg;
+	u32 *addrs = jit->addrs;
+	s32 imm = insn->imm;
+	s16 off = insn->off;
 
-static int bpf_jit_insn(struct bpf_jit *jit, struct sock_filter *filter,
-			unsigned int *addrs, int i, int last)
-{
-	unsigned int K;
-	int offset;
-	unsigned int mask;
-	u16 code;
-
-	K = filter->k;
-	code = bpf_anc_helper(filter);
-
-	switch (code) {
-	case BPF_ALU | BPF_ADD | BPF_X: /* A += X */
-		jit->seen |= SEEN_XREG;
-		/* ar %r5,%r12 */
-		EMIT2(0x1a5c);
-		break;
-	case BPF_ALU | BPF_ADD | BPF_K: /* A += K */
-		if (!K)
+	switch (insn->code) {
+	/*
+	 * BPF_MOV
+	 */
+	case BPF_ALU | BPF_MOV | BPF_X: /* dst = (u32) src */
+		/* llgfr %dst,%src */
+		EMIT4(0xb9160000, dst_reg, src_reg);
+		break;
+	case BPF_ALU64 | BPF_MOV | BPF_X: /* dst = src */
+		/* lgr %dst,%src */
+		EMIT4(0xb9040000, dst_reg, src_reg);
+		break;
+	case BPF_ALU | BPF_MOV | BPF_K: /* dst = (u32) imm */
+		/* llilf %dst,imm */
+		EMIT6_IMM(0xc00f0000, dst_reg, imm);
+		break;
+	case BPF_ALU64 | BPF_MOV | BPF_K: /* dst = imm */
+		/* lgfi %dst,imm */
+		EMIT6_IMM(0xc0010000, dst_reg, imm);
+		break;
+	/*
+	 * BPF_LD 64
+	 */
+	case BPF_LD | BPF_IMM | BPF_DW: /* dst = (u64) imm */
+	{
+		/* 16 byte instruction that uses two 'struct bpf_insn' */
+		u64 imm64;
+
+		imm64 = (u64)(u32) insn[0].imm | ((u64)(u32) insn[1].imm) << 32;
+		/* lg %dst,<d(imm)>(%l) */
+		EMIT6_DISP_LH(0xe3000000, 0x0004, dst_reg, REG_0, REG_L,
+			      EMIT_CONST_U64(imm64));
+		insn_count = 2;
+		break;
+	}
+	/*
+	 * BPF_ADD
+	 */
+	case BPF_ALU | BPF_ADD | BPF_X: /* dst = (u32) dst + (u32) src */
+		/* ar %dst,%src */
+		EMIT2(0x1a00, dst_reg, src_reg);
+		EMIT_ZERO(dst_reg);
+		break;
+	case BPF_ALU64 | BPF_ADD | BPF_X: /* dst = dst + src */
+		/* agr %dst,%src */
+		EMIT4(0xb9080000, dst_reg, src_reg);
+		break;
+	case BPF_ALU | BPF_ADD | BPF_K: /* dst = (u32) dst + (u32) imm */
+		if (!imm)
 			break;
-		if (K <= 16383)
-			/* ahi %r5,<K> */
-			EMIT4_IMM(0xa75a0000, K);
-		else if (test_facility(21))
-			/* alfi %r5,<K> */
-			EMIT6_IMM(0xc25b0000, K);
-		else
-			/* a %r5,<d(K)>(%r13) */
-			EMIT4_DISP(0x5a50d000, EMIT_CONST(K));
+		/* alfi %dst,imm */
+		EMIT6_IMM(0xc20b0000, dst_reg, imm);
+		EMIT_ZERO(dst_reg);
+		break;
+	case BPF_ALU64 | BPF_ADD | BPF_K: /* dst = dst + imm */
+		if (!imm)
+			break;
+		/* agfi %dst,imm */
+		EMIT6_IMM(0xc2080000, dst_reg, imm);
+		break;
+	/*
+	 * BPF_SUB
+	 */
+	case BPF_ALU | BPF_SUB | BPF_X: /* dst = (u32) dst - (u32) src */
+		/* sr %dst,%src */
+		EMIT2(0x1b00, dst_reg, src_reg);
+		EMIT_ZERO(dst_reg);
 		break;
-	case BPF_ALU | BPF_SUB | BPF_X: /* A -= X */
-		jit->seen |= SEEN_XREG;
-		/* sr %r5,%r12 */
-		EMIT2(0x1b5c);
+	case BPF_ALU64 | BPF_SUB | BPF_X: /* dst = dst - src */
+		/* sgr %dst,%src */
+		EMIT4(0xb9090000, dst_reg, src_reg);
 		break;
-	case BPF_ALU | BPF_SUB | BPF_K: /* A -= K */
-		if (!K)
+	case BPF_ALU | BPF_SUB | BPF_K: /* dst = (u32) dst - (u32) imm */
+		if (!imm)
 			break;
-		if (K <= 16384)
-			/* ahi %r5,-K */
-			EMIT4_IMM(0xa75a0000, -K);
-		else if (test_facility(21))
-			/* alfi %r5,-K */
-			EMIT6_IMM(0xc25b0000, -K);
-		else
-			/* s %r5,<d(K)>(%r13) */
-			EMIT4_DISP(0x5b50d000, EMIT_CONST(K));
-		break;
-	case BPF_ALU | BPF_MUL | BPF_X: /* A *= X */
-		jit->seen |= SEEN_XREG;
-		/* msr %r5,%r12 */
-		EMIT4(0xb252005c);
-		break;
-	case BPF_ALU | BPF_MUL | BPF_K: /* A *= K */
-		if (K <= 16383)
-			/* mhi %r5,K */
-			EMIT4_IMM(0xa75c0000, K);
-		else if (test_facility(34))
-			/* msfi %r5,<K> */
-			EMIT6_IMM(0xc2510000, K);
-		else
-			/* ms %r5,<d(K)>(%r13) */
-			EMIT4_DISP(0x7150d000, EMIT_CONST(K));
+		/* alfi %dst,-imm */
+		EMIT6_IMM(0xc20b0000, dst_reg, -imm);
+		EMIT_ZERO(dst_reg);
 		break;
-	case BPF_ALU | BPF_DIV | BPF_X: /* A /= X */
-		jit->seen |= SEEN_XREG | SEEN_RET0;
-		/* ltr %r12,%r12 */
-		EMIT2(0x12cc);
-		/* jz <ret0> */
-		EMIT4_PCREL(0xa7840000, (jit->ret0_ip - jit->prg));
-		/* lhi %r4,0 */
-		EMIT4(0xa7480000);
-		/* dlr %r4,%r12 */
-		EMIT4(0xb997004c);
-		break;
-	case BPF_ALU | BPF_DIV | BPF_K: /* A /= K */
-		if (K == 1)
+	case BPF_ALU64 | BPF_SUB | BPF_K: /* dst = dst - imm */
+		if (!imm)
+			break;
+		/* agfi %dst,-imm */
+		EMIT6_IMM(0xc2080000, dst_reg, -imm);
+		break;
+	/*
+	 * BPF_MUL
+	 */
+	case BPF_ALU | BPF_MUL | BPF_X: /* dst = (u32) dst * (u32) src */
+		/* msr %dst,%src */
+		EMIT4(0xb2520000, dst_reg, src_reg);
+		EMIT_ZERO(dst_reg);
+		break;
+	case BPF_ALU64 | BPF_MUL | BPF_X: /* dst = dst * src */
+		/* msgr %dst,%src */
+		EMIT4(0xb90c0000, dst_reg, src_reg);
+		break;
+	case BPF_ALU | BPF_MUL | BPF_K: /* dst = (u32) dst * (u32) imm */
+		if (imm == 1)
+			break;
+		/* msfi %r5,imm */
+		EMIT6_IMM(0xc2010000, dst_reg, imm);
+		EMIT_ZERO(dst_reg);
+		break;
+	case BPF_ALU64 | BPF_MUL | BPF_K: /* dst = dst * imm *