powerpc/ebpf/jit: Implement JIT compiler for extended BPF

PPC64 eBPF JIT compiler.

Enable with:
  echo 1 > /proc/sys/net/core/bpf_jit_enable
or
  echo 2 > /proc/sys/net/core/bpf_jit_enable

... to see the generated JIT code. This can further be processed with
tools/net/bpf_jit_disasm.

With CONFIG_TEST_BPF=m and 'modprobe test_bpf':

 test_bpf: Summary: 305 PASSED, 0 FAILED, [297/297 JIT'ed]

... on both ppc64 BE and LE.

The details of the approach are documented through various comments in
the code.
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

arch/powerpc/Kconfig

@@ -128,7 +128,8 @@ config PPC
 	select IRQ_FORCED_THREADING
 	select HAVE_RCU_TABLE_FREE if SMP
 	select HAVE_SYSCALL_TRACEPOINTS
-	select HAVE_CBPF_JIT
+	select HAVE_CBPF_JIT if !PPC64
+	select HAVE_EBPF_JIT if PPC64
 	select HAVE_ARCH_JUMP_LABEL
 	select ARCH_HAVE_NMI_SAFE_CMPXCHG
 	select ARCH_HAS_GCOV_PROFILE_ALL

arch/powerpc/include/asm/asm-compat.h

@@ -36,11 +36,13 @@
 #define PPC_MIN_STKFRM	112
 
 #ifdef __BIG_ENDIAN__
+#define LHZX_BE	stringify_in_c(lhzx)
 #define LWZX_BE	stringify_in_c(lwzx)
 #define LDX_BE	stringify_in_c(ldx)
 #define STWX_BE	stringify_in_c(stwx)
 #define STDX_BE	stringify_in_c(stdx)
 #else
+#define LHZX_BE	stringify_in_c(lhbrx)
 #define LWZX_BE	stringify_in_c(lwbrx)
 #define LDX_BE	stringify_in_c(ldbrx)
 #define STWX_BE	stringify_in_c(stwbrx)

arch/powerpc/include/asm/ppc-opcode.h

@@ -142,9 +142,11 @@
 #define PPC_INST_ISEL			0x7c00001e
 #define PPC_INST_ISEL_MASK		0xfc00003e
 #define PPC_INST_LDARX			0x7c0000a8
+#define PPC_INST_STDCX			0x7c0001ad
 #define PPC_INST_LSWI			0x7c0004aa
 #define PPC_INST_LSWX			0x7c00042a
 #define PPC_INST_LWARX			0x7c000028
+#define PPC_INST_STWCX			0x7c00012d
 #define PPC_INST_LWSYNC			0x7c2004ac
 #define PPC_INST_SYNC			0x7c0004ac
 #define PPC_INST_SYNC_MASK		0xfc0007fe
@@ -211,8 +213,11 @@
 #define PPC_INST_LBZ			0x88000000
 #define PPC_INST_LD			0xe8000000
 #define PPC_INST_LHZ			0xa0000000
-#define PPC_INST_LHBRX			0x7c00062c
 #define PPC_INST_LWZ			0x80000000
+#define PPC_INST_LHBRX			0x7c00062c
+#define PPC_INST_LDBRX			0x7c000428
+#define PPC_INST_STB			0x98000000
+#define PPC_INST_STH			0xb0000000
 #define PPC_INST_STD			0xf8000000
 #define PPC_INST_STDU			0xf8000001
 #define PPC_INST_STW			0x90000000
@@ -221,22 +226,34 @@
 #define PPC_INST_MTLR			0x7c0803a6
 #define PPC_INST_CMPWI			0x2c000000
 #define PPC_INST_CMPDI			0x2c200000
+#define PPC_INST_CMPW			0x7c000000
+#define PPC_INST_CMPD			0x7c200000
 #define PPC_INST_CMPLW			0x7c000040
+#define PPC_INST_CMPLD			0x7c200040
 #define PPC_INST_CMPLWI			0x28000000
+#define PPC_INST_CMPLDI			0x28200000
 #define PPC_INST_ADDI			0x38000000
 #define PPC_INST_ADDIS			0x3c000000
 #define PPC_INST_ADD			0x7c000214
 #define PPC_INST_SUB			0x7c000050
 #define PPC_INST_BLR			0x4e800020
 #define PPC_INST_BLRL			0x4e800021
+#define PPC_INST_MULLD			0x7c0001d2
 #define PPC_INST_MULLW			0x7c0001d6
 #define PPC_INST_MULHWU			0x7c000016
 #define PPC_INST_MULLI			0x1c000000
 #define PPC_INST_DIVWU			0x7c000396
+#define PPC_INST_DIVD			0x7c0003d2
 #define PPC_INST_RLWINM			0x54000000
+#define PPC_INST_RLWIMI			0x50000000
+#define PPC_INST_RLDICL			0x78000000
 #define PPC_INST_RLDICR			0x78000004
 #define PPC_INST_SLW			0x7c000030
+#define PPC_INST_SLD			0x7c000036
 #define PPC_INST_SRW			0x7c000430
+#define PPC_INST_SRD			0x7c000436
+#define PPC_INST_SRAD			0x7c000634
+#define PPC_INST_SRADI			0x7c000674
 #define PPC_INST_AND			0x7c000038
 #define PPC_INST_ANDDOT			0x7c000039
 #define PPC_INST_OR			0x7c000378
@@ -247,6 +264,7 @@
 #define PPC_INST_XORI			0x68000000
 #define PPC_INST_XORIS			0x6c000000
 #define PPC_INST_NEG			0x7c0000d0
+#define PPC_INST_EXTSW			0x7c0007b4
 #define PPC_INST_BRANCH			0x48000000
 #define PPC_INST_BRANCH_COND		0x40800000
 #define PPC_INST_LBZCIX			0x7c0006aa

arch/powerpc/net/Makefile

 #
 # Arch-specific network modules
 #
+ifeq ($(CONFIG_PPC64),y)
+obj-$(CONFIG_BPF_JIT) += bpf_jit_asm64.o bpf_jit_comp64.o
+else
 obj-$(CONFIG_BPF_JIT) += bpf_jit_asm.o bpf_jit_comp.o
+endif

arch/powerpc/net/bpf_jit.h

@@ -2,6 +2,7 @@
  * bpf_jit.h: BPF JIT compiler for PPC
  *
  * Copyright 2011 Matt Evans <matt@ozlabs.org>, IBM Corporation
+ * 	     2016 Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
  *
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License
@@ -13,7 +14,9 @@
 
 #ifndef __ASSEMBLY__
 
-#ifdef CONFIG_PPC64
+#include <asm/types.h>
+
+#ifdef PPC64_ELF_ABI_v1
 #define FUNCTION_DESCR_SIZE	24
 #else
 #define FUNCTION_DESCR_SIZE	0
@@ -52,6 +55,10 @@
 				     ___PPC_RA(base) | IMM_L(i))
 #define PPC_STWU(r, base, i)	EMIT(PPC_INST_STWU | ___PPC_RS(r) |	      \
 				     ___PPC_RA(base) | IMM_L(i))
+#define PPC_STH(r, base, i)	EMIT(PPC_INST_STH | ___PPC_RS(r) |	      \
+				     ___PPC_RA(base) | IMM_L(i))
+#define PPC_STB(r, base, i)	EMIT(PPC_INST_STB | ___PPC_RS(r) |	      \
+				     ___PPC_RA(base) | IMM_L(i))
 
 #define PPC_LBZ(r, base, i)	EMIT(PPC_INST_LBZ | ___PPC_RT(r) |	      \
 				     ___PPC_RA(base) | IMM_L(i))
@@ -63,6 +70,19 @@
 				     ___PPC_RA(base) | IMM_L(i))
 #define PPC_LHBRX(r, base, b)	EMIT(PPC_INST_LHBRX | ___PPC_RT(r) |	      \
 				     ___PPC_RA(base) | ___PPC_RB(b))
+#define PPC_LDBRX(r, base, b)	EMIT(PPC_INST_LDBRX | ___PPC_RT(r) |	      \
+				     ___PPC_RA(base) | ___PPC_RB(b))
+
+#define PPC_BPF_LDARX(t, a, b, eh) EMIT(PPC_INST_LDARX | ___PPC_RT(t) |	      \
+					___PPC_RA(a) | ___PPC_RB(b) |	      \
+					__PPC_EH(eh))
+#define PPC_BPF_LWARX(t, a, b, eh) EMIT(PPC_INST_LWARX | ___PPC_RT(t) |	      \
+					___PPC_RA(a) | ___PPC_RB(b) |	      \
+					__PPC_EH(eh))
+#define PPC_BPF_STWCX(s, a, b)	EMIT(PPC_INST_STWCX | ___PPC_RS(s) |	      \
+					___PPC_RA(a) | ___PPC_RB(b))
+#define PPC_BPF_STDCX(s, a, b)	EMIT(PPC_INST_STDCX | ___PPC_RS(s) |	      \
+					___PPC_RA(a) | ___PPC_RB(b))
 
 #ifdef CONFIG_PPC64
 #define PPC_BPF_LL(r, base, i) do { PPC_LD(r, base, i); } while(0)
@@ -76,14 +96,23 @@
 
 #define PPC_CMPWI(a, i)		EMIT(PPC_INST_CMPWI | ___PPC_RA(a) | IMM_L(i))
 #define PPC_CMPDI(a, i)		EMIT(PPC_INST_CMPDI | ___PPC_RA(a) | IMM_L(i))
+#define PPC_CMPW(a, b)		EMIT(PPC_INST_CMPW | ___PPC_RA(a) |	      \
+					___PPC_RB(b))
+#define PPC_CMPD(a, b)		EMIT(PPC_INST_CMPD | ___PPC_RA(a) |	      \
+					___PPC_RB(b))
 #define PPC_CMPLWI(a, i)	EMIT(PPC_INST_CMPLWI | ___PPC_RA(a) | IMM_L(i))
+#define PPC_CMPLDI(a, i)	EMIT(PPC_INST_CMPLDI | ___PPC_RA(a) | IMM_L(i))
 #define PPC_CMPLW(a, b)		EMIT(PPC_INST_CMPLW | ___PPC_RA(a) |	      \
 					___PPC_RB(b))
+#define PPC_CMPLD(a, b)		EMIT(PPC_INST_CMPLD | ___PPC_RA(a) |	      \
+					___PPC_RB(b))
 
 #define PPC_SUB(d, a, b)	EMIT(PPC_INST_SUB | ___PPC_RT(d) |	      \
 				     ___PPC_RB(a) | ___PPC_RA(b))
 #define PPC_ADD(d, a, b)	EMIT(PPC_INST_ADD | ___PPC_RT(d) |	      \
 				     ___PPC_RA(a) | ___PPC_RB(b))
+#define PPC_MULD(d, a, b)	EMIT(PPC_INST_MULLD | ___PPC_RT(d) |	      \
+				     ___PPC_RA(a) | ___PPC_RB(b))
 #define PPC_MULW(d, a, b)	EMIT(PPC_INST_MULLW | ___PPC_RT(d) |	      \
 				     ___PPC_RA(a) | ___PPC_RB(b))
 #define PPC_MULHWU(d, a, b)	EMIT(PPC_INST_MULHWU | ___PPC_RT(d) |	      \
@@ -92,6 +121,8 @@
 				     ___PPC_RA(a) | IMM_L(i))
 #define PPC_DIVWU(d, a, b)	EMIT(PPC_INST_DIVWU | ___PPC_RT(d) |	      \
 				     ___PPC_RA(a) | ___PPC_RB(b))
+#define PPC_DIVD(d, a, b)	EMIT(PPC_INST_DIVD | ___PPC_RT(d) |	      \
+				     ___PPC_RA(a) | ___PPC_RB(b))
 #define PPC_AND(d, a, b)	EMIT(PPC_INST_AND | ___PPC_RA(d) |	      \
 				     ___PPC_RS(a) | ___PPC_RB(b))
 #define PPC_ANDI(d, a, i)	EMIT(PPC_INST_ANDI | ___PPC_RA(d) |	      \
@@ -100,6 +131,7 @@
 				     ___PPC_RS(a) | ___PPC_RB(b))
 #define PPC_OR(d, a, b)		EMIT(PPC_INST_OR | ___PPC_RA(d) |	      \
 				     ___PPC_RS(a) | ___PPC_RB(b))
+#define PPC_MR(d, a)		PPC_OR(d, a, a)
 #define PPC_ORI(d, a, i)	EMIT(PPC_INST_ORI | ___PPC_RA(d) |	      \
 				     ___PPC_RS(a) | IMM_L(i))
 #define PPC_ORIS(d, a, i)	EMIT(PPC_INST_ORIS | ___PPC_RA(d) |	      \
@@ -110,13 +142,30 @@
 				     ___PPC_RS(a) | IMM_L(i))
 #define PPC_XORIS(d, a, i)	EMIT(PPC_INST_XORIS | ___PPC_RA(d) |	      \
 				     ___PPC_RS(a) | IMM_L(i))
+#define PPC_EXTSW(d, a)		EMIT(PPC_INST_EXTSW | ___PPC_RA(d) |	      \
+				     ___PPC_RS(a))
 #define PPC_SLW(d, a, s)	EMIT(PPC_INST_SLW | ___PPC_RA(d) |	      \
 				     ___PPC_RS(a) | ___PPC_RB(s))
+#define PPC_SLD(d, a, s)	EMIT(PPC_INST_SLD | ___PPC_RA(d) |	      \
+				     ___PPC_RS(a) | ___PPC_RB(s))
 #define PPC_SRW(d, a, s)	EMIT(PPC_INST_SRW | ___PPC_RA(d) |	      \
 				     ___PPC_RS(a) | ___PPC_RB(s))
+#define PPC_SRD(d, a, s)	EMIT(PPC_INST_SRD | ___PPC_RA(d) |	      \
+				     ___PPC_RS(a) | ___PPC_RB(s))
+#define PPC_SRAD(d, a, s)	EMIT(PPC_INST_SRAD | ___PPC_RA(d) |	      \
+				     ___PPC_RS(a) | ___PPC_RB(s))
+#define PPC_SRADI(d, a, i)	EMIT(PPC_INST_SRADI | ___PPC_RA(d) |	      \
+				     ___PPC_RS(a) | __PPC_SH(i) |             \
+				     (((i) & 0x20) >> 4))
 #define PPC_RLWINM(d, a, i, mb, me)	EMIT(PPC_INST_RLWINM | ___PPC_RA(d) | \
 					___PPC_RS(a) | __PPC_SH(i) |	      \
 					__PPC_MB(mb) | __PPC_ME(me))
+#define PPC_RLWIMI(d, a, i, mb, me)	EMIT(PPC_INST_RLWIMI | ___PPC_RA(d) | \
+					___PPC_RS(a) | __PPC_SH(i) |	      \
+					__PPC_MB(mb) | __PPC_ME(me))
+#define PPC_RLDICL(d, a, i, mb)		EMIT(PPC_INST_RLDICL | ___PPC_RA(d) | \
+					___PPC_RS(a) | __PPC_SH(i) |	      \
+					__PPC_MB64(mb) | (((i) & 0x20) >> 4))
 #define PPC_RLDICR(d, a, i, me)		EMIT(PPC_INST_RLDICR | ___PPC_RA(d) | \
 					___PPC_RS(a) | __PPC_SH(i) |	      \
 					__PPC_ME64(me) | (((i) & 0x20) >> 4))
@@ -127,6 +176,8 @@
 #define PPC_SRWI(d, a, i)	PPC_RLWINM(d, a, 32-(i), i, 31)
 /* sldi = rldicr Rx, Ry, n, 63-n */
 #define PPC_SLDI(d, a, i)	PPC_RLDICR(d, a, i, 63-(i))
+/* sldi = rldicl Rx, Ry, 64-n, n */
+#define PPC_SRDI(d, a, i)	PPC_RLDICL(d, a, 64-(i), i)
 
 #define PPC_NEG(d, a)		EMIT(PPC_INST_NEG | ___PPC_RT(d) | ___PPC_RA(a))
 

arch/powerpc/net/bpf_jit64.h

+/*
+ * bpf_jit64.h: BPF JIT compiler for PPC64
+ *
+ * Copyright 2016 Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
+ *		  IBM Corporation
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; version 2
+ * of the License.
+ */
+#ifndef _BPF_JIT64_H
+#define _BPF_JIT64_H
+
+#include "bpf_jit.h"
+
+/*
+ * Stack layout:
+ *
+ *		[	prev sp		] <-------------
+ *		[   nv gpr save area	] 8*8		|
+ * fp (r31) -->	[   ebpf stack space	] 512		|
+ *		[  local/tmp var space	] 16		|
+ *		[     frame header	] 32/112	|
+ * sp (r1) --->	[    stack pointer	] --------------
+ */
+
+/* for bpf JIT code internal usage */
+#define BPF_PPC_STACK_LOCALS	16
+/* for gpr non volatile registers BPG_REG_6 to 10, plus skb cache registers */
+#define BPF_PPC_STACK_SAVE	(8*8)
+/* Ensure this is quadword aligned */
+#define BPF_PPC_STACKFRAME	(STACK_FRAME_MIN_SIZE + BPF_PPC_STACK_LOCALS + \
+				 MAX_BPF_STACK + BPF_PPC_STACK_SAVE)
+
+#ifndef __ASSEMBLY__
+
+/* BPF register usage */
+#define SKB_HLEN_REG	(MAX_BPF_REG + 0)
+#define SKB_DATA_REG	(MAX_BPF_REG + 1)
+#define TMP_REG_1	(MAX_BPF_REG + 2)
+#define TMP_REG_2	(MAX_BPF_REG + 3)
+
+/* BPF to ppc register mappings */
+static const int b2p[] = {
+	/* function return value */
+	[BPF_REG_0] = 8,
+	/* function arguments */
+	[BPF_REG_1] = 3,
+	[BPF_REG_2] = 4,
+	[BPF_REG_3] = 5,
+	[BPF_REG_4] = 6,
+	[BPF_REG_5] = 7,
+	/* non volatile registers */
+	[BPF_REG_6] = 27,
+	[BPF_REG_7] = 28,
+	[BPF_REG_8] = 29,
+	[BPF_REG_9] = 30,
+	/* frame pointer aka BPF_REG_10 */
+	[BPF_REG_FP] = 31,
+	/* eBPF jit internal registers */
+	[SKB_HLEN_REG] = 25,
+	[SKB_DATA_REG] = 26,
+	[TMP_REG_1] = 9,
+	[TMP_REG_2] = 10
+};
+
+/* Assembly helpers */
+#define DECLARE_LOAD_FUNC(func)	u64 func(u64 r3, u64 r4);			\
+				u64 func##_negative_offset(u64 r3, u64 r4);	\
+				u64 func##_positive_offset(u64 r3, u64 r4);
+
+DECLARE_LOAD_FUNC(sk_load_word);
+DECLARE_LOAD_FUNC(sk_load_half);
+DECLARE_LOAD_FUNC(sk_load_byte);
+
+#define CHOOSE_LOAD_FUNC(imm, func)						\
+			(imm < 0 ?						\
+			(imm >= SKF_LL_OFF ? func##_negative_offset : func) :	\
+			func##_positive_offset)
+
+#define SEEN_FUNC	0x1000 /* might call external helpers */
+#define SEEN_STACK	0x2000 /* uses BPF stack */
+#define SEEN_SKB	0x4000 /* uses sk_buff */
+
+struct codegen_context {
+	/*
+	 * This is used to track register usage as well
+	 * as calls to external helpers.
+	 * - register usage is tracked with corresponding
+	 *   bits (r3-r10 and r25-r31)
+	 * - rest of the bits can be used to track other
+	 *   things -- for now, we use bits 16 to 23
+	 *   encoded in SEEN_* macros above
+	 */
+	unsigned int seen;
+	unsigned int idx;
+};
+
+#endif /* !__ASSEMBLY__ */
+
+#endif

arch/powerpc/net/bpf_jit_asm64.S

+/*
+ * bpf_jit_asm64.S: Packet/header access helper functions
+ * for PPC64 BPF compiler.
+ *
+ * Copyright 2016, Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
+ * 		   IBM Corporation
+ *
+ * Based on bpf_jit_asm.S by Matt Evans
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; version 2
+ * of the License.
+ */
+
+#include <asm/ppc_asm.h>
+#include <asm/ptrace.h>
+#include "bpf_jit64.h"
+
+/*
+ * All of these routines are called directly from generated code,
+ * with the below register usage:
+ * r27		skb pointer (ctx)
+ * r25		skb header length
+ * r26		skb->data pointer
+ * r4		offset
+ *
+ * Result is passed back in:
+ * r8		data read in host endian format (accumulator)
+ *
+ * r9 is used as a temporary register
+ */
+
+#define r_skb	r27
+#define r_hlen	r25
+#define r_data	r26
+#define r_off	r4
+#define r_val	r8
+#define r_tmp	r9
+
+_GLOBAL_TOC(sk_load_word)
+	cmpdi	r_off, 0
+	blt	bpf_slow_path_word_neg
+	b	sk_load_word_positive_offset
+
+_GLOBAL_TOC(sk_load_word_positive_offset)
+	/* Are we accessing past headlen? */
+	subi	r_tmp, r_hlen, 4
+	cmpd	r_tmp, r_off
+	blt	bpf_slow_path_word
+	/* Nope, just hitting the header.  cr0 here is eq or gt! */
+	LWZX_BE	r_val, r_data, r_off
+	blr	/* Return success, cr0 != LT */
+
+_GLOBAL_TOC(sk_load_half)
+	cmpdi	r_off, 0
+	blt	bpf_slow_path_half_neg
+	b	sk_load_half_positive_offset
+
+_GLOBAL_TOC(sk_load_half_positive_offset)
+	subi	r_tmp, r_hlen, 2
+	cmpd	r_tmp, r_off
+	blt	bpf_slow_path_half
+	LHZX_BE	r_val, r_data, r_off
+	blr
+
+_GLOBAL_TOC(sk_load_byte)
+	cmpdi	r_off, 0
+	blt	bpf_slow_path_byte_neg
+	b	sk_load_byte_positive_offset
+
+_GLOBAL_TOC(sk_load_byte_positive_offset)
+	cmpd	r_hlen, r_off
+	ble	bpf_slow_path_byte
+	lbzx	r_val, r_data, r_off
+	blr
+
+/*
+ * Call out to skb_copy_bits:
+ * Allocate a new stack frame here to remain ABI-compliant in
+ * stashing LR.
+ */
+#define bpf_slow_path_common(SIZE)					\
+	mflr	r0;							\
+	std	r0, PPC_LR_STKOFF(r1);					\
+	stdu	r1, -(STACK_FRAME_MIN_SIZE + BPF_PPC_STACK_LOCALS)(r1);	\
+	mr	r3, r_skb;						\
+	/* r4 = r_off as passed */					\
+	addi	r5, r1, STACK_FRAME_MIN_SIZE;				\
+	li	r6, SIZE;						\
+	bl	skb_copy_bits;						\
+	nop;								\
+	/* save r5 */							\
+	addi	r5, r1, STACK_FRAME_MIN_SIZE;				\
+	/* r3 = 0 on success */						\
+	addi	r1, r1, STACK_FRAME_MIN_SIZE + BPF_PPC_STACK_LOCALS;	\
+	ld	r0, PPC_LR_STKOFF(r1);					\
+	mtlr	r0;							\
+	cmpdi	r3, 0;							\
+	blt	bpf_error;	/* cr0 = LT */
+
+bpf_slow_path_word:
+	bpf_slow_path_common(4)
+	/* Data value is on stack, and cr0 != LT */
+	LWZX_BE	r_val, 0, r5
+	blr
+
+bpf_slow_path_half:
+	bpf_slow_path_common(2)
+	LHZX_BE	r_val, 0, r5
+	blr
+
+bpf_slow_path_byte:
+	bpf_slow_path_common(1)
+	lbzx	r_val, 0, r5
+	blr
+
+/*
+ * Call out to bpf_internal_load_pointer_neg_helper
+ */
+#define sk_negative_common(SIZE)				\
+	mflr	r0;						\
+	std	r0, PPC_LR_STKOFF(r1);				\
+	stdu	r1, -STACK_FRAME_MIN_SIZE(r1);			\
+	mr	r3, r_skb;					\
+	/* r4 = r_off, as passed */				\
+	li	r5, SIZE;					\
+	bl	bpf_internal_load_pointer_neg_helper;		\
+	nop;							\
+	addi	r1, r1, STACK_FRAME_MIN_SIZE;			\
+	ld	r0, PPC_LR_STKOFF(r1);				\
+	mtlr	r0;						\
+	/* R3 != 0 on success */				\
+	cmpldi	r3, 0;						\
+	beq	bpf_error_slow;	/* cr0 = EQ */
+
+bpf_slow_path_word_neg:
+	lis     r_tmp, -32	/* SKF_LL_OFF */
+	cmpd	r_off, r_tmp	/* addr < SKF_* */
+	blt	bpf_error	/* cr0 = LT */
+	b	sk_load_word_negative_offset
+
+_GLOBAL_TOC(sk_load_word_negative_offset)
+	sk_negative_common(4)
+	LWZX_BE	r_val, 0, r3
+	blr
+
+bpf_slow_path_half_neg:
+	lis     r_tmp, -32	/* SKF_LL_OFF */
+	cmpd	r_off, r_tmp	/* addr < SKF_* */
+	blt	bpf_error	/* cr0 = LT */
+	b	sk_load_half_negative_offset
+
+_GLOBAL_TOC(sk_load_half_negative_offset)
+	sk_negative_common(2)
+	LHZX_BE	r_val, 0, r3
+	blr
+
+bpf_slow_path_byte_neg:
+	lis     r_tmp, -32	/* SKF_LL_OFF */
+	cmpd	r_off, r_tmp	/* addr < SKF_* */
+	blt	bpf_error	/* cr0 = LT */
+	b	sk_load_byte_negative_offset
+
+_GLOBAL_TOC(sk_load_byte_negative_offset)
+	sk_negative_common(1)
+	lbzx	r_val, 0, r3
+	blr
+
+bpf_error_slow:
+	/* fabricate a cr0 = lt */
+	li	r_tmp, -1
+	cmpdi	r_tmp, 0
+bpf_error:
+	/*
+	 * Entered with cr0 = lt
+	 * Generated code will 'blt epilogue', returning 0.
+	 */
+	li	r_val, 0
+	blr

arch/powerpc/net/bpf_jit_comp64.c

+/*
+ * bpf_jit_comp64.c: eBPF JIT compiler
+ *
+ * Copyright 2016 Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
+ *		  IBM Corporation
+ *
+ * Based on the powerpc classic BPF JIT compiler by Matt Evans
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; version 2
+ * of the License.
+ */
+#include <linux/moduleloader.h>
+#include <asm/cacheflush.h>
+#include <linux/netdevice.h>
+#include <linux/filter.h>
+#include <linux/if_vlan.h>
+#include <asm/kprobes.h>
+
+#include "bpf_jit64.h"
+
+int bpf_jit_enable __read_mostly;
+
+static void bpf_jit_fill_ill_insns(void *area, unsigned int size)
+{
+	int *p = area;
+
+	/* Fill whole space with trap instructions */
+	while (p < (int *)((char *)area + size))
+		*p++ = BREAKPOINT_INSTRUCTION;
+}
+
+static inline void bpf_flush_icache(void *start, void *end)
+{
+	smp_wmb();
+	flush_icache_range((unsigned long)start, (unsigned long)end);
+}
+
+static inline bool bpf_is_seen_register(struct codegen_context *ctx, int i)
+{
+	return (ctx->seen & (1 << (31 - b2p[i])));
+}
+
+static inline void bpf_set_seen_register(struct codegen_context *ctx, int i)
+{
+	ctx->seen |= (1 << (31 - b2p[i]));
+}
+
+static inline bool bpf_has_stack_frame(struct codegen_context *ctx)
+{
+	/*
+	 * We only need a stack frame if:
+	 * - we call other functions (kernel helpers), or
+	 * - the bpf program uses its stack area
+	 * The latter condition is deduced from the usage of BPF_REG_FP
+	 */
+	return ctx->seen & SEEN_FUNC || bpf_is_seen_register(ctx, BPF_REG_FP);
+}
+
+static void bpf_jit_emit_skb_loads(u32 *image, struct codegen_context *ctx)
+{
+	/*
+	 * Load skb->len and skb->data_len
+	 * r3 points to skb
+	 */
+	PPC_LWZ(b2p[SKB_HLEN_REG], 3, offsetof(struct sk_buff, len));
+	PPC_LWZ(b2p[TMP_REG_1], 3, offsetof(struct sk_buff, data_len));
+	/* header_len = len - data_len */
+	PPC_SUB(b2p[SKB_HLEN_REG], b2p[SKB_HLEN_REG], b2p[TMP_REG_1]);
+
+	/* skb->data pointer */
+	PPC_BPF_LL(b2p[SKB_DATA_REG], 3, offsetof(struct sk_buff, data));
+}
+
+static void bpf_jit_emit_func_call(u32 *image, struct codegen_context *ctx, u64 func)
+{
+#ifdef PPC64_ELF_ABI_v1
+	/* func points to the function descriptor */
+	PPC_LI64(b2p[TMP_REG_2], func);
+	/* Load actual entry point from function descriptor */
+	PPC_BPF_LL(b2p[TMP_REG_1], b2p[TMP_REG_2], 0);
+	/* ... and move it to LR */
+	PPC_MTLR(b2p[TMP_REG_1]);
+	/*
+	 * Load TOC from function descriptor at offset 8.
+	 * We can clobber r2 since we get called through a
+	 * function pointer (so caller will save/restore r2)
+	 * and since we don't use a TOC ourself.
+	 */
+	PPC_BPF_LL(2, b2p[TMP_REG_2], 8);
+#else
+	/* We can clobber r12 */
+	PPC_FUNC_ADDR(12, func);
+	PPC_MTLR(12);
+#endif
+	PPC_BLRL();
+}
+
+static void bpf_jit_build_prologue(u32 *image, struct codegen_context *ctx)
+{
+	int i;
+	bool new_stack_frame = bpf_has_stack_frame(ctx);
+
+	if (new_stack_frame) {
+		/*
+		 * We need a stack frame, but we don't necessarily need to
+		 * save/restore LR unless we call other functions
+		 */
+		if (ctx->seen & SEEN_FUNC) {
+			EMIT(PPC_INST_MFLR | __PPC_RT(R0));
+			PPC_BPF_STL(0, 1, PPC_LR_STKOFF);
+		}
+
+		PPC_BPF_STLU(1, 1, -BPF_PPC_STACKFRAME);
+	}
+
+	/*
+	 * Back up non-volatile regs -- BPF registers 6-10
+	 * If we haven't created our own stack frame, we save these
+	 * in the protected zone below the previous stack frame
+	 */
+	for (i = BPF_REG_6; i <= BPF_REG_10; i++)
+		if (bpf_is_seen_register(ctx, i))
+			PPC_BPF_STL(b2p[i], 1,
+				(new_stack_frame ? BPF_PPC_STACKFRAME : 0) -
+					(8 * (32 - b2p[i])));
+
+	/*
+	 * Save additional non-volatile regs if we cache skb
+	 * Also, setup skb data
+	 */
+	if (ctx->seen & SEEN_SKB) {
+		PPC_BPF_STL(b2p[SKB_HLEN_REG], 1,
+			BPF_PPC_STACKFRAME - (8 * (32 - b2p[SKB_HLEN_REG])));
+		PPC_BPF_STL(b2p[SKB_DATA_REG], 1,
+			BPF_PPC_STACKFRAME - (8 * (32 - b2p[SKB_DATA_REG])));
+		bpf_jit_emit_skb_loads(image, ctx);
+	}
+
+	/* Setup frame pointer to point to the bpf stack area */
+	if (bpf_is_seen_register(ctx, BPF_REG_FP))
+		PPC_ADDI(b2p[BPF_REG_FP], 1,
+				BPF_PPC_STACKFRAME - BPF_PPC_STACK_SAVE);
+}
+
+static void bpf_jit_build_epilogue(u32 *image, struct codegen_context *ctx)
+{
+	int i;
+	bool new_stack_frame = bpf_has_stack_frame(ctx);
+
+	/* Move result to r3 */
+	PPC_MR(3, b2p[BPF_REG_0]);
+
+	/* Restore NVRs */
+	for (i = BPF_REG_6; i <= BPF_REG_10; i++)
+		if (bpf_is_seen_register(ctx, i))
+			PPC_BPF_LL(b2p[i], 1,
+				(new_stack_frame ? BPF_PPC_STACKFRAME : 0) -
+					(8 * (32 - b2p[i])));
+
+	/* Restore non-volatile registers used for skb cache */
+	if (ctx->seen & SEEN_SKB) {
+		PPC_BPF_LL(b2p[SKB_HLEN_REG], 1,
+			BPF_PPC_STACKFRAME - (8 * (32 - b2p[SKB_HLEN_REG])));
+		PPC_BPF_LL(b2p[SKB_DATA_REG], 1,
+			BPF_PPC_STACKFRAME - (8 * (32 - b2p[SKB_DATA_REG])));
+	}
+
+	/* Tear down our stack frame */
+	if (new_stack_frame) {
+		PPC_ADDI(1, 1, BPF_PPC_STACKFRAME);
+		if (ctx->seen & SEEN_FUNC) {
+			PPC_BPF_LL(0, 1, PPC_LR_STKOFF);
+			PPC_MTLR(0);
+		}
+	}
+
+	PPC_BLR();
+}
+
+/* Assemble the body code between the prologue & epilogue */
+static int bpf_jit_build_body(struct bpf_prog *fp, u32 *image,
+			      struct codegen_context *ctx,
+			      u32 *addrs)
+{
+	const struct bpf_insn *insn = fp->insnsi;
+	int flen = fp->len;
+	int i;
+
+	/* Start of epilogue code - will only be valid 2nd pass onwards */
+	u32 exit_addr = addrs[flen];
+
+	for (i = 0; i < flen; i++) {
+		u32 code = insn[i].code;
+		u32 dst_reg = b2p[insn[i].dst_reg];
+		u32 src_reg = b2p[insn[i].src_reg];
+		s16 off = insn[i].off;
+		s32 imm = insn[i].imm;
+		u64 imm64;
+		u8 *func;
+		u32 true_cond;
+		int stack_local_off;
+
+		/*
+		 * addrs[] maps a BPF bytecode address into a real offset from
+		 * the start of the body code.
+		 */
+		addrs[i] = ctx->idx * 4;
+
+		/*
+		 * As an optimization, we note down which non-volatile registers
+		 * are used so that we can only save/restore those in our
+		 * prologue and epilogue. We do this here regardless of whether
+		 * the actual BPF instruction uses src/dst registers or not
+		 * (for instance, BPF_CALL does not use them). The expectation
+		 * is that those instructions will have src_reg/dst_reg set to
+		 * 0. Even otherwise, we just lose some prologue/epilogue
+		 * optimization but everything else should work without
+		 * any issues.
+		 */
+		if (dst_reg >= 24 && dst_reg <= 31)
+			bpf_set_seen_register(ctx, insn[i].dst_reg);
+		if (src_reg >= 24 && src_reg <= 31)
+			bpf_set_seen_register(ctx, insn[i].src_reg);
+
+		switch (code) {
+		/*
+		 * Arithmetic operations: ADD/SUB/MUL/DIV/MOD/NEG
+		 */
+		case BPF_ALU | BPF_ADD | BPF_X: /* (u32) dst += (u32) src */
+		case BPF_ALU64 | BPF_ADD | BPF_X: /* dst += src */
+			PPC_ADD(dst_reg, dst_reg, src_reg);
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_SUB | BPF_X: /* (u32) dst -= (u32) src */
+		case BPF_ALU64 | BPF_SUB | BPF_X: /* dst -= src */
+			PPC_SUB(dst_reg, dst_reg, src_reg);
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_ADD | BPF_K: /* (u32) dst += (u32) imm */
+		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)
+					PPC_ADDI(dst_reg, dst_reg, IMM_L(imm));
+				else {
+					PPC_LI32(b2p[TMP_REG_1], imm);
+					PPC_ADD(dst_reg, dst_reg, b2p[TMP_REG_1]);
+				}
+			}
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_MUL | BPF_X: /* (u32) dst *= (u32) src */
+		case BPF_ALU64 | BPF_MUL | BPF_X: /* dst *= src */
+			if (BPF_CLASS(code) == BPF_ALU)
+				PPC_MULW(dst_reg, dst_reg, src_reg);
+			else
+				PPC_MULD(dst_reg, dst_reg, src_reg);
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_MUL | BPF_K: /* (u32) dst *= (u32) imm */
+		case BPF_ALU64 | BPF_MUL | BPF_K: /* dst *= imm */
+			if (imm >= -32768 && imm < 32768)
+				PPC_MULI(dst_reg, dst_reg, IMM_L(imm));
+			else {
+				PPC_LI32(b2p[TMP_REG_1], imm);
+				if (BPF_CLASS(code) == BPF_ALU)
+					PPC_MULW(dst_reg, dst_reg,
+							b2p[TMP_REG_1]);
+				else
+					PPC_MULD(dst_reg, dst_reg,
+							b2p[TMP_REG_1]);
+			}
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_DIV | BPF_X: /* (u32) dst /= (u32) src */
+		case BPF_ALU | BPF_MOD | BPF_X: /* (u32) dst %= (u32) src */
+			PPC_CMPWI(src_reg, 0);
+			PPC_BCC_SHORT(COND_NE, (ctx->idx * 4) + 12);
+			PPC_LI(b2p[BPF_REG_0], 0);
+			PPC_JMP(exit_addr);
+			if (BPF_OP(code) == BPF_MOD) {
+				PPC_DIVWU(b2p[TMP_REG_1], dst_reg, src_reg);
+				PPC_MULW(b2p[TMP_REG_1], src_reg,
+						b2p[TMP_REG_1]);
+				PPC_SUB(dst_reg, dst_reg, b2p[TMP_REG_1]);
+			} else
+				PPC_DIVWU(dst_reg, dst_reg, src_reg);
+			goto bpf_alu32_trunc;
+		case BPF_ALU64 | BPF_DIV | BPF_X: /* dst /= src */
+		case BPF_ALU64 | BPF_MOD | BPF_X: /* dst %= src */
+			PPC_CMPDI(src_reg, 0);
+			PPC_BCC_SHORT(COND_NE, (ctx->idx * 4) + 12);
+			PPC_LI(b2p[BPF_REG_0], 0);
+			PPC_JMP(exit_addr);
+			if (BPF_OP(code) == BPF_MOD) {
+				PPC_DIVD(b2p[TMP_REG_1], dst_reg, src_reg);
+				PPC_MULD(b2p[TMP_REG_1], src_reg,
+						b2p[TMP_REG_1]);
+				PPC_SUB(dst_reg, dst_reg, b2p[TMP_REG_1]);
+			} else
+				PPC_DIVD(dst_reg, dst_reg, src_reg);
+			break;
+		case BPF_ALU | BPF_MOD | BPF_K: /* (u32) dst %= (u32) imm */
+		case BPF_ALU | BPF_DIV | BPF_K: /* (u32) dst /= (u32) imm */
+		case BPF_ALU64 | BPF_MOD | BPF_K: /* dst %= imm */
+		case BPF_ALU64 | BPF_DIV | BPF_K: /* dst /= imm */
+			if (imm == 0)
+				return -EINVAL;
+			else if (imm == 1)
+				goto bpf_alu32_trunc;
+
+			PPC_LI32(b2p[TMP_REG_1], imm);
+			switch (BPF_CLASS(code)) {
+			case BPF_ALU:
+				if (BPF_OP(code) == BPF_MOD) {
+					PPC_DIVWU(b2p[TMP_REG_2], dst_reg,
+							b2p[TMP_REG_1]);
+					PPC_MULW(b2p[TMP_REG_1],
+							b2p[TMP_REG_1],
+							b2p[TMP_REG_2]);
+					PPC_SUB(dst_reg, dst_reg,
+							b2p[TMP_REG_1]);
+				} else
+					PPC_DIVWU(dst_reg, dst_reg,
+							b2p[TMP_REG_1]);
+				break;
+			case BPF_ALU64:
+				if (BPF_OP(code) == BPF_MOD) {
+					PPC_DIVD(b2p[TMP_REG_2], dst_reg,
+							b2p[TMP_REG_1]);
+					PPC_MULD(b2p[TMP_REG_1],
+							b2p[TMP_REG_1],
+							b2p[TMP_REG_2]);
+					PPC_SUB(dst_reg, dst_reg,
+							b2p[TMP_REG_1]);
+				} else
+					PPC_DIVD(dst_reg, dst_reg,
+							b2p[TMP_REG_1]);
+				break;
+			}
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_NEG: /* (u32) dst = -dst */
+		case BPF_ALU64 | BPF_NEG: /* dst = -dst */
+			PPC_NEG(dst_reg, dst_reg);
+			goto bpf_alu32_trunc;
+
+		/*
+		 * Logical operations: AND/OR/XOR/[A]LSH/[A]RSH
+		 */
+		case BPF_ALU | BPF_AND | BPF_X: /* (u32) dst = dst & src */
+		case BPF_ALU64 | BPF_AND | BPF_X: /* dst = dst & src */
+			PPC_AND(dst_reg, dst_reg, src_reg);
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_AND | BPF_K: /* (u32) dst = dst & imm */
+		case BPF_ALU64 | BPF_AND | BPF_K: /* dst = dst & imm */
+			if (!IMM_H(imm))
+				PPC_ANDI(dst_reg, dst_reg, IMM_L(imm));
+			else {
+				/* Sign-extended */
+				PPC_LI32(b2p[TMP_REG_1], imm);
+				PPC_AND(dst_reg, dst_reg, b2p[TMP_REG_1]);
+			}
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_OR | BPF_X: /* dst = (u32) dst | (u32) src */
+		case BPF_ALU64 | BPF_OR | BPF_X: /* dst = dst | src */
+			PPC_OR(dst_reg, dst_reg, src_reg);
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_OR | BPF_K:/* dst = (u32) dst | (u32) imm */
+		case BPF_ALU64 | BPF_OR | BPF_K:/* dst = dst | imm */
+			if (imm < 0 && BPF_CLASS(code) == BPF_ALU64) {
+				/* Sign-extended */
+				PPC_LI32(b2p[TMP_REG_1], imm);
+				PPC_OR(dst_reg, dst_reg, b2p[TMP_REG_1]);
+			} else {
+				if (IMM_L(imm))
+					PPC_ORI(dst_reg, dst_reg, IMM_L(imm));
+				if (IMM_H(imm))
+					PPC_ORIS(dst_reg, dst_reg, IMM_H(imm));
+			}
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_XOR | BPF_X: /* (u32) dst ^= src */
+		case BPF_ALU64 | BPF_XOR | BPF_X: /* dst ^= src */
+			PPC_XOR(dst_reg, dst_reg, src_reg);
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_XOR | BPF_K: /* (u32) dst ^= (u32) imm */
+		case BPF_ALU64 | BPF_XOR | BPF_K: /* dst ^= imm */
+			if (imm < 0 && BPF_CLASS(code) == BPF_ALU64) {
+				/* Sign-extended */
+				PPC_LI32(b2p[TMP_REG_1], imm);
+				PPC_XOR(dst_reg, dst_reg, b2p[TMP_REG_1]);
+			} else {
+				if (IMM_L(imm))
+					PPC_XORI(dst_reg, dst_reg, IMM_L(imm));
+				if (IMM_H(imm))
+					PPC_XORIS(dst_reg, dst_reg, IMM_H(imm));
+			}
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_LSH | BPF_X: /* (u32) dst <<= (u32) src */
+			/* slw clears top 32 bits */
+			PPC_SLW(dst_reg, dst_reg, src_reg);
+			break;
+		case BPF_ALU64 | BPF_LSH | BPF_X: /* dst <<= src; */
+			PPC_SLD(dst_reg, dst_reg, src_reg);
+			break;
+		case BPF_ALU | BPF_LSH | BPF_K: /* (u32) dst <<== (u32) imm */
+			/* with imm 0, we still need to clear top 32 bits */
+			PPC_SLWI(dst_reg, dst_reg, imm);
+			break;
+		case BPF_ALU64 | BPF_LSH | BPF_K: /* dst <<== imm */
+			if (imm != 0)
+				PPC_SLDI(dst_reg, dst_reg, imm);
+			break;
+		case BPF_ALU | BPF_RSH | BPF_X: /* (u32) dst >>= (u32) src */
+			PPC_SRW(dst_reg, dst_reg, src_reg);
+			break;
+		case BPF_ALU64 | BPF_RSH | BPF_X: /* dst >>= src */
+			PPC_SRD(dst_reg, dst_reg, src_reg);
+			break;
+		case BPF_ALU | BPF_RSH | BPF_K: /* (u32) dst >>= (u32) imm */
+			PPC_SRWI(dst_reg, dst_reg, imm);
+			break;
+		case BPF_ALU64 | BPF_RSH | BPF_K: /* dst >>= imm */
+			if (imm != 0)
+				PPC_SRDI(dst_reg, dst_reg, imm);
+			break;
+		case BPF_ALU64 | BPF_ARSH | BPF_X: /* (s64) dst >>= src */
+			PPC_SRAD(dst_reg, dst_reg, src_reg);
+			break;
+		case BPF_ALU64 | BPF_ARSH | BPF_K: /* (s64) dst >>= imm */
+			if (imm != 0)
+				PPC_SRADI(dst_reg, dst_reg, imm);
+			break;
+
+		/*
+		 * MOV
+		 */
+		case BPF_ALU | BPF_MOV | BPF_X: /* (u32) dst = src */
+		case BPF_ALU64 | BPF_MOV | BPF_X: /* dst = src */
+			PPC_MR(dst_reg, src_reg);
+			goto bpf_alu32_trunc;
+		case BPF_ALU | BPF_MOV | BPF_K: /* (u32) dst = imm */
+		case BPF_ALU64 | BPF_MOV | BPF_K: /* dst = (s64) imm */
+			PPC_LI32(dst_reg, imm);
+			if (imm < 0)
+				goto bpf_alu32_trunc;
+			break;
+
+bpf_alu32_trunc:
+		/* Truncate to 32-bits */
+		if (BPF_CLASS(code) == BPF_ALU)
+			PPC_RLWINM(dst_reg, dst_reg, 0, 0, 31);
+		break;
+
+		/*
+		 * BPF_FROM_BE/LE
+		 */
+		case BPF_ALU | BPF_END | BPF_FROM_LE:
+		case BPF_ALU | BPF_END | BPF_FROM_BE:
+#ifdef __BIG_ENDIAN__
+			if (BPF_SRC(code) == BPF_FROM_BE)
+				goto emit_clear;
+#else /* !__BIG_ENDIAN__ */
+			if (BPF_SRC(code) == BPF_FROM_LE)
+				goto emit_clear;
+#endif
+			switch (imm) {
+			case 16:
+				/* Rotate 8 bits left & mask with 0x0000ff00 */
+				PPC_RLWINM(b2p[TMP_REG_1], dst_reg, 8, 16, 23);
+				/* Rotate 8 bits right & insert LSB to reg */
+				PPC_RLWIMI(b2p[TMP_REG_1], dst_reg, 24, 24, 31);
+				/* Move result back to dst_reg */
+				PPC_MR(dst_reg, b2p[TMP_REG_1]);
+				break;
+			case 32:
+				/*
+				 * Rotate word left by 8 bits:
+				 * 2 bytes are already in their final position
+				 * -- byte 2 and 4 (of bytes 1, 2, 3 and 4)
+				 */
+				PPC_RLWINM(b2p[TMP_REG_1], dst_reg, 8, 0, 31);
+				/* Rotate 24 bits and insert byte 1 */
+				PPC_RLWIMI(b2p[TMP_REG_1], dst_reg, 24, 0, 7);
+				/* Rotate 24 bits and insert byte 3 */
+				PPC_RLWIMI(b2p[TMP_REG_1], dst_reg, 24, 16, 23);
+				PPC_MR(dst_reg, b2p[TMP_REG_1]);
+				break;
+			case 64:
+				/*
+				 * Way easier and faster(?) to store the value
+				 * into stack and then use ldbrx
+				 *
+				 * First, determine where in stack we can store
+				 * this:
+				 * - if we have allotted a stack frame, then we
+				 *   will utilize the area set aside by
+				 *   BPF_PPC_STACK_LOCALS
+				 * - else, we use the area beneath the NV GPR
+				 *   save area
+				 *
+				 * ctx->seen will be reliable in pass2, but
+				 * the instructions generated will remain the
+				 * same across all passes
+				 */
+				if (bpf_has_stack_frame(ctx))
+					stack_local_off = STACK_FRAME_MIN_SIZE;
+				else
+					stack_local_off = -(BPF_PPC_STACK_SAVE + 8);
+
+				PPC_STD(dst_reg, 1, stack_local_off);
+				PPC_ADDI(b2p[TMP_REG_1], 1, stack_local_off);
+				PPC_LDBRX(dst_reg, 0, b2p[TMP_REG_1]);
+				break;
+			}
+			break;
+
+emit_clear:
+			switch (imm) {
+			case 16:
+				/* zero-extend 16 bits into 64 bits */
+				PPC_RLDICL(dst_reg, dst_reg, 0, 48);
+				break;
+			case 32:
+				/* zero-extend 32 bits into 64 bits */
+				PPC_RLDICL(dst_reg, dst_reg, 0, 32);
+				break;
+			case 64:
+				/* nop */
+				break;
+			}
+			break;
+
+		/*
+		 * BPF_ST(X)
+		 */
+		case BPF_STX | BPF_MEM | BPF_B: /* *(u8 *)(dst + off) = src */
+		case BPF_ST | BPF_MEM | BPF_B: /* *(u8 *)(dst + off) = imm */
+			if (BPF_CLASS(code) == BPF_ST) {
+				PPC_LI(b2p[TMP_REG_1], imm);
+				src_reg = b2p[TMP_REG_1];
+			}
+			PPC_STB(src_reg, dst_reg, off);
+			break;
+		case BPF_STX | BPF_MEM | BPF_H: /* (u16 *)(dst + off) = src */
+		case BPF_ST | BPF_MEM | BPF_H: /* (u16 *)(dst + off) = imm */
+			if (BPF_CLASS(code) == BPF_ST) {
+				PPC_LI(b2p[TMP_REG_1], imm);
+				src_reg = b2p[TMP_REG_1];
+			}
+			PPC_STH(src_reg, dst_reg, off);
+			break;
+		case BPF_STX | BPF_MEM | BPF_W: /* *(u32 *)(dst + off) = src */
+		case BPF_ST | BPF_MEM | BPF_W: /* *(u32 *)(dst + off) = imm */
+			if (BPF_CLASS(code) == BPF_ST) {
+				PPC_LI32(b2p[TMP_REG_1], imm);
+				src_reg = b2p[TMP_REG_1];
+			}
+			PPC_STW(src_reg, dst_reg, off);
+			break;
+		case BPF_STX | BPF_MEM | BPF_DW: /* (u64 *)(dst + off) = src */
+		case BPF_ST | BPF_MEM | BPF_DW: /* *(u64 *)(dst + off) = imm */
+			if (BPF_CLASS(code) == BPF_ST) {
+				PPC_LI32(b2p[TMP_REG_1], imm);
+				src_reg = b2p[TMP_REG_1];
+			}
+			PPC_STD(src_reg, dst_reg, off);
+			break;
+
+		/*
+		 * BPF_STX XADD (atomic_add)
+		 */
+		/* *(u32 *)(dst + off) += src */
+		case BPF_STX | BPF_XADD | BPF_W:
+			/* Get EA into TMP_REG_1 */
+			PPC_ADDI(b2p[TMP_REG_1], dst_reg, off);
+			/* error if EA is not word-aligned */
+			PPC_ANDI(b2p[TMP_REG_2], b2p[TMP_REG_1], 0x03);
+			PPC_BCC_SHORT(COND_EQ, (ctx->idx * 4) + 12);
+			PPC_LI(b2p[BPF_REG_0], 0);
+			PPC_JMP(exit_addr);
+			/* load value from memory into TMP_REG_2 */
+			PPC_BPF_LWARX(b2p[TMP_REG_2], 0, b2p[TMP_REG_1], 0);
+			/* add value from src_reg into this */
+			PPC_ADD(b2p[TMP_REG_2], b2p[TMP_REG_2], src_reg);
+			/* store result back */
+			PPC_BPF_STWCX(b2p[TMP_REG_2], 0, b2p[TMP_REG_1]);
+			/* we're done if this succeeded */
+			PPC_BCC_SHORT(COND_EQ, (ctx->idx * 4) + (7*4));
+			/* otherwise, let's try once more */
+			PPC_BPF_LWARX(b2p[TMP_REG_2], 0, b2p[TMP_REG_1], 0);
+			PPC_ADD(b2p[TMP_REG_2], b2p[TMP_REG_2], src_reg);
+			PPC_BPF_STWCX(b2p[TMP_REG_2], 0, b2p[TMP_REG_1]);
+			/* exit if the store was not successful */
+			PPC_LI(b2p[BPF_REG_0], 0);
+			PPC_BCC(COND_NE, exit_addr);
+			break;
+		/* *(u64 *)(dst + off) += src */
+		case BPF_STX | BPF_XADD | BPF_DW:
+			PPC_ADDI(b2p[TMP_REG_1], dst_reg, off);
+			/* error if EA is not doubleword-aligned */
+			PPC_ANDI(b2p[TMP_REG_2], b2p[TMP_REG_1], 0x07);
+			PPC_BCC_SHORT(COND_EQ, (ctx->idx * 4) + (3*4));
+			PPC_LI(b2p[BPF_REG_0], 0);
+			PPC_JMP(exit_addr);
+			PPC_BPF_LDARX(b2p[TMP_REG_2], 0, b2p[TMP_REG_1], 0);
+			PPC_ADD(b2p[TMP_REG_2], b2p[TMP_REG_2], src_reg);
+			PPC_BPF_STDCX(b2p[TMP_REG_2], 0, b2p[TMP_REG_1]);
+			PPC_BCC_SHORT(COND_EQ, (ctx->idx * 4) + (7*4));
+			PPC_BPF_LDARX(b2p[TMP_REG_2], 0, b2p[TMP_REG_1], 0);
+			PPC_ADD(b2p[TMP_REG_2], b2p[TMP_REG_2], src_reg);
+			PPC_BPF_STDCX(b2p[TMP_REG_2], 0, b2p[TMP_REG_1]);
+			PPC_LI(b2p[BPF_REG_0], 0);
+			PPC_BCC(COND_NE, exit_addr);
+			break;
+
+		/*
+		 * BPF_LDX
+		 */
+		/* dst = *(u8 *)(ul) (src + off) */
+		case BPF_LDX | BPF_MEM | BPF_B:
+			PPC_LBZ(dst_reg, src_reg, off);
+			break;
+		/* dst = *(u16 *)(ul) (src + off) */
+		case BPF_LDX | BPF_MEM | BPF_H:
+			PPC_LHZ(dst_reg, src_reg, off);
+			break;
+		/* dst = *(u32 *)(ul) (src + off) */
+		case BPF_LDX | BPF_MEM | BPF_W:
+			PPC_LWZ(dst_reg, src_reg, off);
+			break;
+		/* dst = *(u64 *)(ul) (src + off) */
+		case BPF_LDX | BPF_MEM | BPF_DW:
+			PPC_LD(dst_reg, src_reg, off);
+			break;
+
+		/*
+		 * Doubleword load
+		 * 16 byte instruction that uses two 'struct bpf_insn'
+		 */
+		case BPF_LD | BPF_IMM | BPF_DW: /* dst = (u64) imm */
+			imm64 = ((u64)(u32) insn[i].imm) |
+				    (((u64)(u32) insn[i+1].imm) << 32);
+			/* Adjust for two bpf instructions */
+			addrs[++i] = ctx->idx * 4;
+			PPC_LI64(dst_reg, imm64);
+			break;
+
+		/*
+		 * Return/Exit
+		 */
+		case BPF_JMP | BPF_EXIT:
+			/*
+			 * If this isn't the very last instruction, branch to
+			 * the epilogue. If we _are_ the last instruction,
+			 * we'll just fall through to the epilogue.
+			 */
+			if (i != flen - 1)
+				PPC_JMP(exit_addr);
+			/* else fall through to the epilogue */
+			break;
+
+		/*
+		 * Call kernel helper
+		 */
+		case BPF_JMP | BPF_CALL:
+			ctx->seen |= SEEN_FUNC;
+			func = (u8 *) __bpf_call_base + imm;
+
+			/* Save skb pointer if we need to re-cache skb data */
+			if (bpf_helper_changes_skb_data(func))
+				PPC_BPF_STL(3, 1, STACK_FRAME_MIN_SIZE);
+
+			bpf_jit_emit_func_call(image, ctx, (u64)func);
+
+			/* move return value from r3 to BPF_REG_0 */
+			PPC_MR(b2p[BPF_REG_0], 3);
+
+			/* refresh skb cache */
+			if (bpf_helper_changes_skb_data(func)) {
+				/* reload skb pointer to r3 */
+				PPC_BPF_LL(3, 1, STACK_FRAME_MIN_SIZE);
+				bpf_jit_emit_skb_loads(image, ctx);
+			}
+			break;
+
+		/*
+		 * Jumps and branches
+		 */
+		case BPF_JMP | BPF_JA:
+			PPC_JMP(addrs[i + 1 + off]);
+			break;
+
+		case BPF_JMP | BPF_JGT | BPF_K:
+		case BPF_JMP | BPF_JGT | BPF_X:
+		case BPF_JMP | BPF_JSGT | BPF_K:
+		case BPF_JMP | BPF_JSGT | BPF_X:
+			true_cond = COND_GT;
+			goto cond_branch;
+		case BPF_JMP | BPF_JGE | BPF_K:
+		case BPF_JMP | BPF_JGE | BPF_X:
+		case BPF_JMP | BPF_JSGE | BPF_K:
+		case BPF_JMP | BPF_JSGE | BPF_X:
+			true_cond = COND_GE;
+			goto cond_branch;
+		case BPF_JMP | BPF_JEQ | BPF_K:
+		case BPF_JMP | BPF_JEQ | BPF_X:
+			true_cond = COND_EQ;
+			goto cond_branch;
+		case BPF_JMP | BPF_JNE | BPF_K:
+		case BPF_JMP | BPF_JNE | BPF_X:
+			true_cond = COND_NE;
+			goto cond_branch;
+		case BPF_JMP | BPF_JSET | BPF_K:
+		case BPF_JMP | BPF_JSET | BPF_X:
+			true_cond = COND_NE;
+			/* Fall through */
+
+cond_branch:
+			switch (code) {
+			case BPF_JMP | BPF_JGT | BPF_X:
+			case BPF_JMP | BPF_JGE | BPF_X:
+			case BPF_JMP | BPF_JEQ | BPF_X:
+			case BPF_JMP | BPF_JNE | BPF_X:
+				/* unsigned comparison */
+				PPC_CMPLD(dst_reg, src_reg);
+				break;
+			case BPF_JMP | BPF_JSGT | BPF_X:
+			case BPF_JMP | BPF_JSGE | BPF_X:
+				/* signed comparison */
+				PPC_CMPD(dst_reg, src_reg);
+				break;
+			case BPF_JMP | BPF_JSET | BPF_X:
+				PPC_AND_DOT(b2p[TMP_REG_1], dst_reg, src_reg);
+				break;
+			case BPF_JMP | BPF_JNE | BPF_K:
+			case BPF_JMP | BPF_JEQ | BPF_K:
+			case BPF_JMP | BPF_JGT | BPF_K:
+			case BPF_JMP | BPF_JGE | BPF_K:
+				/*
+				 * Need sign-extended load, so only positive
+				 * values can be used as imm in cmpldi
+				 */
+				if (imm >= 0 && imm < 32768)
+					PPC_CMPLDI(dst_reg, imm);
+				else {
+					/* sign-extending load */
+					PPC_LI32(b2p[TMP_REG_1], imm);
+					/* ... but unsigned comparison */
+					PPC_CMPLD(dst_reg, b2p[TMP_REG_1]);
+				}
+				break;
+			case BPF_JMP | BPF_JSGT | BPF_K:
+			case BPF_JMP | BPF_JSGE | BPF_K:
+				/*
+				 * signed comparison, so any 16-bit value
+				 * can be used in cmpdi
+				 */
+				if (imm >= -32768 && imm < 32768)
+					PPC_CMPDI(dst_reg, imm);
+				else {
+					PPC_LI32(b2p[TMP_REG_1], imm);
+					PPC_CMPD(dst_reg, b2p[TMP_REG_1]);
+				}
+				break;
+			case BPF_JMP | BPF_JSET | BPF_K:
+				/* andi does not sign-extend the immediate */
+				if (imm >= 0 && imm < 32768)
+					/* PPC_ANDI is _only/always_ dot-form */
+					PPC_ANDI(b2p[TMP_REG_1], dst_reg, imm);
+				else {
+					PPC_LI32(b2p[TMP_REG_1], imm);
+					PPC_AND_DOT(b2p[TMP_REG_1], dst_reg,
+						    b2p[TMP_REG_1]);
+				}
+				break;
+			}
+			PPC_BCC(true_cond, addrs[i + 1 + off]);
+			break;
+
+		/*
+		 * Loads from packet header/data
+		 * Assume 32-bit input value in imm and X (src_reg)
+		 */
+
+		/* Absolute loads */
+		case BPF_LD | BPF_W | BPF_ABS:
+			func = (u8 *)CHOOSE_LOAD_FUNC(imm, sk_load_word);
+			goto common_load_abs;
+		case BPF_LD | BPF_H | BPF_ABS:
+			func = (u8 *)CHOOSE_LOAD_FUNC(imm, sk_load_half);
+			goto common_load_abs;
+		case BPF_LD | BPF_B | BPF_ABS:
+			func = (u8 *)CHOOSE_LOAD_FUNC(imm, sk_load_byte);
+common_load_abs:
+			/*
+			 * Load from [imm]
+			 * Load into r4, which can just be passed onto
+			 *  skb load helpers as the second parameter
+			 */
+			PPC_LI32(4, imm);
+			goto common_load;
+
+		/* Indirect loads */
+		case BPF_LD | BPF_W | BPF_IND:
+			func = (u8 *)sk_load_word;
+			goto common_load_ind;
+		case BPF_LD | BPF_H | BPF_IND:
+			func = (u8 *)sk_load_half;
+			goto common_load_ind;
+		case BPF_LD | BPF_B | BPF_IND:
+			func = (u8 *)sk_load_byte;
+common_load_ind:
+			/*
+			 * Load from [src_reg + imm]
+			 * Treat src_reg as a 32-bit value
+			 */
+			PPC_EXTSW(4, src_reg);
+			if (imm) {
+				if (imm >= -32768 && imm < 32768)
+					PPC_ADDI(4, 4, IMM_L(imm));
+				else {
+					PPC_LI32(b2p[TMP_REG_1], imm);
+					PPC_ADD(4, 4, b2p[TMP_REG_1]);
+				}
+			}
+
+common_load:
+			ctx->seen |= SEEN_SKB;
+			ctx->seen |= SEEN_FUNC;
+			bpf_jit_emit_func_call(image, ctx, (u64)func);
+
+			/*
+			 * Helper returns 'lt' condition on error, and an
+			 * appropriate return value in BPF_REG_0
+			 */
+			PPC_BCC(COND_LT, exit_addr);
+			break;
+
+		/*
+		 * TODO: Tail call
+		 */
+		case BPF_JMP | BPF_CALL | BPF_X:
+
+		default:
+			/*
+			 * The filter contains something cruel & unusual.
+			 * We don't handle it, but also there shouldn't be
+			 * anything missing from our list.
+			 */
+			pr_err_ratelimited("eBPF filter opcode %04x (@%d) unsupported\n",
+					code, i);
+			return -ENOTSUPP;
+		}
+	}
+
+	/* Set end-of-body-code address for exit. */
+	addrs[i] = ctx->idx * 4;
+
+	return 0;
+}
+
+void bpf_jit_compile(struct bpf_prog *fp) { }
+
+struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
+{
+	u32 proglen;
+	u32 alloclen;
+	u8 *image = NULL;
+	u32 *code_base;
+	u32 *addrs;
+	struct codegen_context cgctx;
+	int pass;
+	int flen;
+	struct bpf_binary_header *bpf_hdr;
+
+	if (!bpf_jit_enable)
+		return fp;
+
+	flen = fp->len;
+	addrs = kzalloc((flen+1) * sizeof(*addrs), GFP_KERNEL);
+	if (addrs == NULL)
+		return fp;
+
+	cgctx.idx = 0;
+	cgctx.seen = 0;
+	/* Scouting faux-generate pass 0 */
+	if (bpf_jit_build_body(fp, 0, &cgctx, addrs))
+		/* We hit something illegal or unsupported. */
+		goto out;
+
+	/*
+	 * Pretend to build prologue, given the features we've seen.  This will
+	 * update ctgtx.idx as it pretends to output instructions, then we can
+	 * calculate total size from idx.
+	 */
+	bpf_jit_build_prologue(0, &cgctx);
+	bpf_jit_build_epilogue(0, &cgctx);
+
+	proglen = cgctx.idx * 4;
+	alloclen = proglen + FUNCTION_DESCR_SIZE;
+
+	bpf_hdr = bpf_jit_binary_alloc(alloclen, &image, 4,
+			bpf_jit_fill_ill_insns);
+	if (!bpf_hdr)
+		goto out;
+
+	code_base = (u32 *)(image + FUNCTION_DESCR_SIZE);
+
+	/* Code generation passes 1-2 */
+	for (pass = 1; pass < 3; pass++) {
+		/* Now build the prologue, body code & epilogue for real. */
+		cgctx.idx = 0;
+		bpf_jit_build_prologue(code_base, &cgctx);
+		bpf_jit_build_body(fp, code_base, &cgctx, addrs);
+		bpf_jit_build_epilogue(code_base, &cgctx);
+
+		if (bpf_jit_enable > 1)
+			pr_info("Pass %d: shrink = %d, seen = 0x%x\n", pass,
+				proglen - (cgctx.idx * 4), cgctx.seen);
+	}
+
+	if (bpf_jit_enable > 1)
+		/*
+		 * Note that we output the base address of the code_base
+		 * rather than image, since opcodes are in code_base.
+		 */
+		bpf_jit_dump(flen, proglen, pass, code_base);
+
+	if (image) {
+		bpf_flush_icache(bpf_hdr, image + alloclen);
+#ifdef PPC64_ELF_ABI_v1
+		/* Function descriptor nastiness: Address + TOC */
+		((u64 *)image)[0] = (u64)code_base;
+		((u64 *)image)[1] = local_paca->kernel_toc;
+#endif
+		fp->bpf_func = (void *)image;
+		fp->jited = 1;
+	}
+
+out:
+	kfree(addrs);
+	return fp;
+}
+
+void bpf_jit_free(struct bpf_prog *fp)
+{
+	unsigned long addr = (unsigned long)fp->bpf_func & PAGE_MASK;
+	struct bpf_binary_header *bpf_hdr = (void *)addr;
+
+	if (fp->jited)
+		bpf_jit_binary_free(bpf_hdr);
+
+	bpf_prog_unlock_free(fp);
+}