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