diff mbox series

[RFC,bpf-next,v1,11/14] MIPS: eBPF: add core support for 32/64-bit systems

Message ID 9a6e1b11fdc1f6a4c1b81cfb1b361d4628cc14c6.1625970384.git.Tony.Ambardar@gmail.com (mailing list archive)
State Superseded
Delegated to: BPF
Headers show
Series MIPS: eBPF: refactor code, add MIPS32 JIT | expand

Checks

Context Check Description
netdev/cover_letter success Link
netdev/fixes_present success Link
netdev/patch_count success Link
netdev/tree_selection success Clearly marked for bpf-next
netdev/subject_prefix success Link
netdev/cc_maintainers success CCed 13 of 13 maintainers
netdev/source_inline fail Was 0 now: 5
netdev/verify_signedoff success Link
netdev/module_param success Was 0 now: 0
netdev/build_32bit success Errors and warnings before: 0 this patch: 0
netdev/kdoc fail Errors and warnings before: 4 this patch: 5
netdev/verify_fixes success Link
netdev/checkpatch warning CHECK: Alignment should match open parenthesis CHECK: Please use a blank line after function/struct/union/enum declarations CHECK: Unnecessary parentheses around 'insn->src_reg != BPF_PSEUDO_CALL' CHECK: Unnecessary parentheses around 'insn->src_reg == BPF_PSEUDO_CALL' CHECK: spaces preferred around that '|' (ctx:VxV) WARNING: 'regster' may be misspelled - perhaps 'register'? WARNING: From:/Signed-off-by: email address mismatch: 'From: Tony Ambardar <tony.ambardar@gmail.com>' != 'Signed-off-by: Tony Ambardar <Tony.Ambardar@gmail.com>' WARNING: line length of 82 exceeds 80 columns WARNING: line length of 84 exceeds 80 columns
netdev/build_allmodconfig_warn success Errors and warnings before: 0 this patch: 0
netdev/header_inline success Link

Commit Message

Tony Ambardar July 12, 2021, 12:34 a.m. UTC
Update register definitions and flags for both 32/64-bit operation. Add a
common register lookup table and modify ebpf_to_mips_reg() to use this,
and add is64bit() and isbigend() common helper functions.

On MIPS32, BPF registers are held in register pairs defined by the base
register. Word-size and endian-aware helper macros select 32-bit registers
from a pair and generate 32-bit word memory storage offsets. The BPF TCC
is stored to the stack due to register pressure.

Modify build_int_prologue() and build_int_epilogue() to handle MIPS32
registers and any adjustments needed during program entry/exit/transfer
if transitioning between the native N64/O32 ABI and the BPF 64-bit ABI.
Also ensure ABI-consistent stack alignment and use verifier-provided stack
depth during setup.

Update emit_const_to_reg() to work across MIPS64 and MIPS32 systems and
optimize gen_imm_to_reg() to only set the lower halfword if needed.

Rework emit_bpf_tail_call() to also support MIPS32 usage and add common
helpers, emit_bpf_call() and emit_push_args(), handling TCC and ABI
variations on MIPS32/MIPS64. Add tail_call_present() and update tailcall
handling to support mixing BPF2BPF subprograms and tailcalls.

Add sign and zero-extension helpers usable with verifier zext insertion,
gen_zext_insn() and gen_sext_insn().

Add common functions emit_caller_save() and emit_caller_restore(), which
push and pop all caller-saved BPF registers to the stack, for use with
JIT-internal kernel calls such as those needed for BPF insns unsupported
by native system ISA opcodes. Since these calls would be hidden from any
BPF C compiler, which would normally spill needed registers during a call,
the JIT must handle save/restore itself.

Adopt a dedicated BPF FP (in MIPS_R_S8), and relax FP usage within insns.
This reduces ad-hoc code doing $sp manipulation with temp registers, and
allows wider usage of BPF FP for comparison and arithmetic. For example,
the following tests from test_verifier are now jited but not previously:

  939/p store PTR_TO_STACK in R10 to array map using BPF_B
  981/p unpriv: cmp pointer with pointer
  984/p unpriv: indirectly pass pointer on stack to helper function
  985/p unpriv: mangle pointer on stack 1
  986/p unpriv: mangle pointer on stack 2
  1001/p unpriv: partial copy of pointer
  1097/p xadd/w check whether src/dst got mangled, 1
  1098/p xadd/w check whether src/dst got mangled, 2

Signed-off-by: Tony Ambardar <Tony.Ambardar@gmail.com>
---
 arch/mips/net/ebpf_jit.c | 750 +++++++++++++++++++++++++++++----------
 1 file changed, 563 insertions(+), 187 deletions(-)
diff mbox series

Patch

diff --git a/arch/mips/net/ebpf_jit.c b/arch/mips/net/ebpf_jit.c
index 61d7894051aa..525f1df8db21 100644
--- a/arch/mips/net/ebpf_jit.c
+++ b/arch/mips/net/ebpf_jit.c
@@ -1,11 +1,13 @@ 
 // SPDX-License-Identifier: GPL-2.0-only
 /*
- * Just-In-Time compiler for eBPF filters on MIPS
- *
- * Copyright (c) 2017 Cavium, Inc.
+ * Just-In-Time compiler for eBPF filters on MIPS32/MIPS64
+ * Copyright (c) 2021 Tony Ambardar <Tony.Ambardar@gmail.com>
  *
  * Based on code from:
  *
+ * Copyright (c) 2017 Cavium, Inc.
+ * Author: David Daney <david.daney@cavium.com>
+ *
  * Copyright (c) 2014 Imagination Technologies Ltd.
  * Author: Markos Chandras <markos.chandras@imgtec.com>
  */
@@ -22,31 +24,42 @@ 
 #include <asm/isa-rev.h>
 #include <asm/uasm.h>
 
-/* Registers used by JIT */
+/* Registers used by JIT:	  (MIPS32)	(MIPS64) */
 #define MIPS_R_ZERO	0
 #define MIPS_R_AT	1
-#define MIPS_R_V0	2	/* BPF_R0 */
-#define MIPS_R_V1	3
-#define MIPS_R_A0	4	/* BPF_R1 */
-#define MIPS_R_A1	5	/* BPF_R2 */
-#define MIPS_R_A2	6	/* BPF_R3 */
-#define MIPS_R_A3	7	/* BPF_R4 */
-#define MIPS_R_A4	8	/* BPF_R5 */
-#define MIPS_R_T4	12	/* BPF_AX */
-#define MIPS_R_T5	13
-#define MIPS_R_T6	14
-#define MIPS_R_T7	15
-#define MIPS_R_S0	16	/* BPF_R6 */
-#define MIPS_R_S1	17	/* BPF_R7 */
-#define MIPS_R_S2	18	/* BPF_R8 */
-#define MIPS_R_S3	19	/* BPF_R9 */
-#define MIPS_R_S4	20	/* BPF_TCC */
-#define MIPS_R_S5	21
-#define MIPS_R_S6	22
-#define MIPS_R_S7	23
-#define MIPS_R_T8	24
-#define MIPS_R_T9	25
+#define MIPS_R_V0	2	/* BPF_R0	BPF_R0 */
+#define MIPS_R_V1	3	/* BPF_R0	BPF_TCC */
+#define MIPS_R_A0	4	/* BPF_R1	BPF_R1 */
+#define MIPS_R_A1	5	/* BPF_R1	BPF_R2 */
+#define MIPS_R_A2	6	/* BPF_R2	BPF_R3 */
+#define MIPS_R_A3	7	/* BPF_R2	BPF_R4 */
+
+/* MIPS64 replaces T0-T3 scratch regs with extra arguments A4-A7. */
+#ifdef CONFIG_64BIT
+#  define MIPS_R_A4	8	/* (n/a)	BPF_R5 */
+#else
+#  define MIPS_R_T0	8	/* BPF_R3	(n/a)  */
+#  define MIPS_R_T1	9	/* BPF_R3	(n/a)  */
+#  define MIPS_R_T2	10	/* BPF_R4	(n/a)  */
+#  define MIPS_R_T3	11	/* BPF_R4	(n/a)  */
+#endif
+
+#define MIPS_R_T4	12	/* BPF_R5	BPF_AX */
+#define MIPS_R_T5	13	/* BPF_R5	(free) */
+#define MIPS_R_T6	14	/* BPF_AX	(used) */
+#define MIPS_R_T7	15	/* BPF_AX	(free) */
+#define MIPS_R_S0	16	/* BPF_R6	BPF_R6 */
+#define MIPS_R_S1	17	/* BPF_R6	BPF_R7 */
+#define MIPS_R_S2	18	/* BPF_R7	BPF_R8 */
+#define MIPS_R_S3	19	/* BPF_R7	BPF_R9 */
+#define MIPS_R_S4	20	/* BPF_R8	BPF_TCC */
+#define MIPS_R_S5	21	/* BPF_R8	(free) */
+#define MIPS_R_S6	22	/* BPF_R9	(free) */
+#define MIPS_R_S7	23	/* BPF_R9	(free) */
+#define MIPS_R_T8	24	/* (used)	(used) */
+#define MIPS_R_T9	25	/* (used)	(used) */
 #define MIPS_R_SP	29
+#define MIPS_R_S8	30	/* BPF_R10	BPF_R10 */
 #define MIPS_R_RA	31
 
 /* eBPF flags */
@@ -55,10 +68,117 @@ 
 #define EBPF_SAVE_S2	BIT(2)
 #define EBPF_SAVE_S3	BIT(3)
 #define EBPF_SAVE_S4	BIT(4)
-#define EBPF_SAVE_RA	BIT(5)
-#define EBPF_SEEN_FP	BIT(6)
-#define EBPF_SEEN_TC	BIT(7)
-#define EBPF_TCC_IN_V1	BIT(8)
+#define EBPF_SAVE_S5	BIT(5)
+#define EBPF_SAVE_S6	BIT(6)
+#define EBPF_SAVE_S7	BIT(7)
+#define EBPF_SAVE_S8	BIT(8)
+#define EBPF_SAVE_RA	BIT(9)
+#define EBPF_SEEN_FP	BIT(10)
+#define EBPF_SEEN_TC	BIT(11)
+#define EBPF_TCC_IN_RUN	BIT(12)
+
+/*
+ * Extra JIT registers dedicated to holding TCC during runtime or saving
+ * across calls.
+ */
+enum {
+	JIT_RUN_TCC = MAX_BPF_JIT_REG,
+	JIT_SAV_TCC
+};
+/* Temporary register for passing TCC if nothing dedicated. */
+#define TEMP_PASS_TCC MIPS_R_T8
+
+/*
+ * Word-size and endianness-aware helpers for building MIPS32 vs MIPS64
+ * tables and selecting 32-bit subregisters from a register pair base.
+ * Simplify use by emulating MIPS_R_SP and MIPS_R_ZERO as register pairs
+ * and adding HI/LO word memory offsets.
+ */
+#ifdef CONFIG_64BIT
+#  define HI(reg) (reg)
+#  define LO(reg) (reg)
+#  define OFFHI(mem) (mem)
+#  define OFFLO(mem) (mem)
+#else	/* CONFIG_32BIT */
+#  ifdef __BIG_ENDIAN
+#    define HI(reg) ((reg) == MIPS_R_SP ? MIPS_R_ZERO : \
+		     (reg) == MIPS_R_S8 ? MIPS_R_ZERO : \
+		     (reg))
+#    define LO(reg) ((reg) == MIPS_R_ZERO ? (reg) : \
+		     (reg) == MIPS_R_SP ? (reg) : \
+		     (reg) == MIPS_R_S8 ? (reg) : \
+		     (reg) + 1)
+#    define OFFHI(mem) (mem)
+#    define OFFLO(mem) ((mem) + sizeof(long))
+#  else	/* __LITTLE_ENDIAN */
+#    define HI(reg) ((reg) == MIPS_R_ZERO ? (reg) : \
+		     (reg) == MIPS_R_SP ? MIPS_R_ZERO : \
+		     (reg) == MIPS_R_S8 ? MIPS_R_ZERO : \
+		     (reg) + 1)
+#    define LO(reg) (reg)
+#    define OFFHI(mem) ((mem) + sizeof(long))
+#    define OFFLO(mem) (mem)
+#  endif
+#endif
+
+#ifdef CONFIG_64BIT
+#  define M(expr32, expr64) (expr64)
+#else
+#  define M(expr32, expr64) (expr32)
+#endif
+const struct {
+	/* Register or pair base */
+	int reg;
+	/* Register flags */
+	u32 flags;
+	/* Usage table:   (MIPS32)			 (MIPS64) */
+} bpf2mips[] = {
+	/* Return value from in-kernel function, and exit value from eBPF. */
+	[BPF_REG_0] =  {M(MIPS_R_V0,			MIPS_R_V0)},
+	/* Arguments from eBPF program to in-kernel/BPF functions. */
+	[BPF_REG_1] =  {M(MIPS_R_A0,			MIPS_R_A0)},
+	[BPF_REG_2] =  {M(MIPS_R_A2,			MIPS_R_A1)},
+	[BPF_REG_3] =  {M(MIPS_R_T0,			MIPS_R_A2)},
+	[BPF_REG_4] =  {M(MIPS_R_T2,			MIPS_R_A3)},
+	[BPF_REG_5] =  {M(MIPS_R_T4,			MIPS_R_A4)},
+	/* Callee-saved registers preserved by in-kernel/BPF functions. */
+	[BPF_REG_6] =  {M(MIPS_R_S0,			MIPS_R_S0),
+			M(EBPF_SAVE_S0|EBPF_SAVE_S1,	EBPF_SAVE_S0)},
+	[BPF_REG_7] =  {M(MIPS_R_S2,			MIPS_R_S1),
+			M(EBPF_SAVE_S2|EBPF_SAVE_S3,	EBPF_SAVE_S1)},
+	[BPF_REG_8] =  {M(MIPS_R_S4,			MIPS_R_S2),
+			M(EBPF_SAVE_S4|EBPF_SAVE_S5,	EBPF_SAVE_S2)},
+	[BPF_REG_9] =  {M(MIPS_R_S6,			MIPS_R_S3),
+			M(EBPF_SAVE_S6|EBPF_SAVE_S7,	EBPF_SAVE_S3)},
+	[BPF_REG_10] = {M(MIPS_R_S8,			MIPS_R_S8),
+			M(EBPF_SAVE_S8|EBPF_SEEN_FP,	EBPF_SAVE_S8|EBPF_SEEN_FP)},
+	/* Internal register for rewriting insns during JIT blinding. */
+	[BPF_REG_AX] = {M(MIPS_R_T6,			MIPS_R_T4)},
+	/*
+	 * Internal registers for TCC runtime holding and saving during
+	 * calls. A zero save register indicates using scratch space on
+	 * the stack for storage during calls. A zero hold register means
+	 * no dedicated register holds TCC during runtime (but a temp reg
+	 * still passes TCC to tailcall or bpf2bpf call).
+	 */
+	[JIT_RUN_TCC] =	{M(0,				MIPS_R_V1)},
+	[JIT_SAV_TCC] =	{M(0,				MIPS_R_S4),
+			 M(0,				EBPF_SAVE_S4)}
+};
+#undef M
+
+static inline bool is64bit(void)
+{
+	return IS_ENABLED(CONFIG_64BIT);
+}
+
+static inline bool isbigend(void)
+{
+	return IS_ENABLED(CONFIG_CPU_BIG_ENDIAN);
+}
+
+/* Stack region alignment under N64 and O32 ABIs */
+#define STACK_ALIGN (2 * sizeof(long))
 
 /*
  * For the mips64 ISA, we need to track the value range or type for
@@ -100,6 +220,7 @@  enum reg_val_type {
 struct jit_ctx {
 	const struct bpf_prog *skf;
 	int stack_size;
+	int bpf_stack_off;
 	u32 idx;
 	u32 flags;
 	u32 *offsets;
@@ -177,6 +298,34 @@  static u32 b_imm(unsigned int tgt, struct jit_ctx *ctx)
 		(ctx->idx * 4) - 4;
 }
 
+/* Sign-extend dst register or HI 32-bit reg of pair. */
+static inline void gen_sext_insn(int dst, struct jit_ctx *ctx)
+{
+	if (is64bit())
+		emit_instr(ctx, sll, dst, dst, 0);
+	else
+		emit_instr(ctx, sra, HI(dst), LO(dst), 31);
+}
+
+/*
+ * Zero-extend dst register or HI 32-bit reg of pair, if either forced
+ * or the BPF verifier does not insert its own zext insns.
+ */
+static inline void gen_zext_insn(int dst, bool force, struct jit_ctx *ctx)
+{
+	if (!ctx->skf->aux->verifier_zext || force) {
+		if (is64bit())
+			emit_instr(ctx, dinsu, dst, MIPS_R_ZERO, 32, 32);
+		else
+			emit_instr(ctx, and, HI(dst), MIPS_R_ZERO, MIPS_R_ZERO);
+	}
+}
+
+static inline bool tail_call_present(struct jit_ctx *ctx)
+{
+	return ctx->flags & EBPF_SEEN_TC || ctx->skf->aux->tail_call_reachable;
+}
+
 enum reg_usage {
 	REG_SRC_FP_OK,
 	REG_SRC_NO_FP,
@@ -186,9 +335,8 @@  enum reg_usage {
 
 /*
  * For eBPF, the register mapping naturally falls out of the
- * requirements of eBPF and the MIPS n64 ABI.  We don't maintain a
- * separate frame pointer, so BPF_REG_10 relative accesses are
- * adjusted to be $sp relative.
+ * requirements of eBPF and the MIPS N64/O32 ABIs. We also maintain
+ * a separate frame pointer, setting BPF_REG_10 relative to $sp.
  */
 static int ebpf_to_mips_reg(struct jit_ctx *ctx,
 			    const struct bpf_insn *insn,
@@ -199,110 +347,139 @@  static int ebpf_to_mips_reg(struct jit_ctx *ctx,
 
 	switch (ebpf_reg) {
 	case BPF_REG_0:
-		return MIPS_R_V0;
 	case BPF_REG_1:
-		return MIPS_R_A0;
 	case BPF_REG_2:
-		return MIPS_R_A1;
 	case BPF_REG_3:
-		return MIPS_R_A2;
 	case BPF_REG_4:
-		return MIPS_R_A3;
 	case BPF_REG_5:
-		return MIPS_R_A4;
 	case BPF_REG_6:
-		ctx->flags |= EBPF_SAVE_S0;
-		return MIPS_R_S0;
 	case BPF_REG_7:
-		ctx->flags |= EBPF_SAVE_S1;
-		return MIPS_R_S1;
 	case BPF_REG_8:
-		ctx->flags |= EBPF_SAVE_S2;
-		return MIPS_R_S2;
 	case BPF_REG_9:
-		ctx->flags |= EBPF_SAVE_S3;
-		return MIPS_R_S3;
+	case BPF_REG_AX:
+		ctx->flags |= bpf2mips[ebpf_reg].flags;
+		return bpf2mips[ebpf_reg].reg;
 	case BPF_REG_10:
 		if (u == REG_DST_NO_FP || u == REG_SRC_NO_FP)
 			goto bad_reg;
-		ctx->flags |= EBPF_SEEN_FP;
-		/*
-		 * Needs special handling, return something that
-		 * cannot be clobbered just in case.
-		 */
-		return MIPS_R_ZERO;
-	case BPF_REG_AX:
-		return MIPS_R_T4;
+		ctx->flags |= bpf2mips[ebpf_reg].flags;
+		return bpf2mips[ebpf_reg].reg;
 	default:
 bad_reg:
 		WARN(1, "Illegal bpf reg: %d\n", ebpf_reg);
 		return -EINVAL;
 	}
 }
+
 /*
  * eBPF stack frame will be something like:
  *
  *  Entry $sp ------>   +--------------------------------+
  *                      |   $ra  (optional)              |
  *                      +--------------------------------+
- *                      |   $s0  (optional)              |
+ *                      |   $s8  (optional)              |
  *                      +--------------------------------+
- *                      |   $s1  (optional)              |
+ *                      |   $s7  (optional)              |
  *                      +--------------------------------+
- *                      |   $s2  (optional)              |
+ *                      |   $s6  (optional)              |
  *                      +--------------------------------+
- *                      |   $s3  (optional)              |
+ *                      |   $s5  (optional)              |
  *                      +--------------------------------+
  *                      |   $s4  (optional)              |
  *                      +--------------------------------+
- *                      |   tmp-storage  (if $ra saved)  |
- * $sp + tmp_offset --> +--------------------------------+ <--BPF_REG_10
+ *                      |   $s3  (optional)              |
+ *                      +--------------------------------+
+ *                      |   $s2  (optional)              |
+ *                      +--------------------------------+
+ *                      |   $s1  (optional)              |
+ *                      +--------------------------------+
+ *                      |   $s0  (optional)              |
+ *                      +--------------------------------+
+ *                      |   tmp-storage  (optional)      |
+ * $sp + bpf_stack_off->+--------------------------------+ <--BPF_REG_10
  *                      |   BPF_REG_10 relative storage  |
  *                      |    MAX_BPF_STACK (optional)    |
  *                      |      .                         |
  *                      |      .                         |
  *                      |      .                         |
- *     $sp -------->    +--------------------------------+
+ *        $sp ------>   +--------------------------------+
  *
  * If BPF_REG_10 is never referenced, then the MAX_BPF_STACK sized
  * area is not allocated.
  */
-static int gen_int_prologue(struct jit_ctx *ctx)
+static int build_int_prologue(struct jit_ctx *ctx)
 {
+	int tcc_run = bpf2mips[JIT_RUN_TCC].reg ?
+		      bpf2mips[JIT_RUN_TCC].reg :
+		      TEMP_PASS_TCC;
+	int tcc_sav = bpf2mips[JIT_SAV_TCC].reg;
+	const struct bpf_prog *prog = ctx->skf;
+	int r10 = bpf2mips[BPF_REG_10].reg;
+	int r1 = bpf2mips[BPF_REG_1].reg;
 	int stack_adjust = 0;
 	int store_offset;
 	int locals_size;
 
 	if (ctx->flags & EBPF_SAVE_RA)
-		/*
-		 * If RA we are doing a function call and may need
-		 * extra 8-byte tmp area.
-		 */
-		stack_adjust += 2 * sizeof(long);
-	if (ctx->flags & EBPF_SAVE_S0)
 		stack_adjust += sizeof(long);
-	if (ctx->flags & EBPF_SAVE_S1)
+	if (ctx->flags & EBPF_SAVE_S8)
 		stack_adjust += sizeof(long);
-	if (ctx->flags & EBPF_SAVE_S2)
+	if (ctx->flags & EBPF_SAVE_S7)
 		stack_adjust += sizeof(long);
-	if (ctx->flags & EBPF_SAVE_S3)
+	if (ctx->flags & EBPF_SAVE_S6)
+		stack_adjust += sizeof(long);
+	if (ctx->flags & EBPF_SAVE_S5)
 		stack_adjust += sizeof(long);
 	if (ctx->flags & EBPF_SAVE_S4)
 		stack_adjust += sizeof(long);
+	if (ctx->flags & EBPF_SAVE_S3)
+		stack_adjust += sizeof(long);
+	if (ctx->flags & EBPF_SAVE_S2)
+		stack_adjust += sizeof(long);
+	if (ctx->flags & EBPF_SAVE_S1)
+		stack_adjust += sizeof(long);
+	if (ctx->flags & EBPF_SAVE_S0)
+		stack_adjust += sizeof(long);
+	if (tail_call_present(ctx) &&
+	    !(ctx->flags & EBPF_TCC_IN_RUN) && !tcc_sav)
+		/* Allocate scratch space for holding TCC if needed. */
+		stack_adjust += sizeof(long);
 
-	BUILD_BUG_ON(MAX_BPF_STACK & 7);
-	locals_size = (ctx->flags & EBPF_SEEN_FP) ? MAX_BPF_STACK : 0;
+	stack_adjust = ALIGN(stack_adjust, STACK_ALIGN);
+
+	locals_size = (ctx->flags & EBPF_SEEN_FP) ? prog->aux->stack_depth : 0;
+	locals_size = ALIGN(locals_size, STACK_ALIGN);
 
 	stack_adjust += locals_size;
 
 	ctx->stack_size = stack_adjust;
+	ctx->bpf_stack_off = locals_size;
+
+	/*
+	 * First instruction initializes the tail call count (TCC) if
+	 * called from kernel or via BPF tail call. A BPF tail-caller
+	 * will skip this instruction and pass the TCC via register.
+	 * As a BPF2BPF subprog, we are called directly and must avoid
+	 * resetting the TCC.
+	 */
+	if (!ctx->skf->is_func)
+		emit_instr(ctx, addiu, tcc_run, MIPS_R_ZERO, MAX_TAIL_CALL_CNT);
 
 	/*
-	 * First instruction initializes the tail call count (TCC).
-	 * On tail call we skip this instruction, and the TCC is
-	 * passed in $v1 from the caller.
+	 * If called from kernel under O32 ABI we must set up BPF R1 context,
+	 * since BPF R1 is an endian-order regster pair ($a0:$a1 or $a1:$a0)
+	 * but context is always passed in $a0 as 32-bit pointer. Entry from
+	 * a tail-call looks just like a kernel call, which means the caller
+	 * must set up R1 context according to the kernel call ABI. If we are
+	 * a BPF2BPF call then all registers are already correctly set up.
 	 */
-	emit_instr(ctx, addiu, MIPS_R_V1, MIPS_R_ZERO, MAX_TAIL_CALL_CNT);
+	if (!is64bit() && !ctx->skf->is_func) {
+		if (isbigend())
+			emit_instr(ctx, move, LO(r1), MIPS_R_A0);
+		/* Sanitize upper 32-bit reg */
+		gen_zext_insn(r1, true, ctx);
+	}
+
 	if (stack_adjust)
 		emit_instr_long(ctx, daddiu, addiu,
 					MIPS_R_SP, MIPS_R_SP, -stack_adjust);
@@ -316,24 +493,24 @@  static int gen_int_prologue(struct jit_ctx *ctx)
 					MIPS_R_RA, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
-	if (ctx->flags & EBPF_SAVE_S0) {
+	if (ctx->flags & EBPF_SAVE_S8) {
 		emit_instr_long(ctx, sd, sw,
-					MIPS_R_S0, store_offset, MIPS_R_SP);
+					MIPS_R_S8, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
-	if (ctx->flags & EBPF_SAVE_S1) {
+	if (ctx->flags & EBPF_SAVE_S7) {
 		emit_instr_long(ctx, sd, sw,
-					MIPS_R_S1, store_offset, MIPS_R_SP);
+					MIPS_R_S7, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
-	if (ctx->flags & EBPF_SAVE_S2) {
+	if (ctx->flags & EBPF_SAVE_S6) {
 		emit_instr_long(ctx, sd, sw,
-					MIPS_R_S2, store_offset, MIPS_R_SP);
+					MIPS_R_S6, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
-	if (ctx->flags & EBPF_SAVE_S3) {
+	if (ctx->flags & EBPF_SAVE_S5) {
 		emit_instr_long(ctx, sd, sw,
-					MIPS_R_S3, store_offset, MIPS_R_SP);
+					MIPS_R_S5, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
 	if (ctx->flags & EBPF_SAVE_S4) {
@@ -341,10 +518,40 @@  static int gen_int_prologue(struct jit_ctx *ctx)
 					MIPS_R_S4, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
+	if (ctx->flags & EBPF_SAVE_S3) {
+		emit_instr_long(ctx, sd, sw,
+					MIPS_R_S3, store_offset, MIPS_R_SP);
+		store_offset -= sizeof(long);
+	}
+	if (ctx->flags & EBPF_SAVE_S2) {
+		emit_instr_long(ctx, sd, sw,
+					MIPS_R_S2, store_offset, MIPS_R_SP);
+		store_offset -= sizeof(long);
+	}
+	if (ctx->flags & EBPF_SAVE_S1) {
+		emit_instr_long(ctx, sd, sw,
+					MIPS_R_S1, store_offset, MIPS_R_SP);
+		store_offset -= sizeof(long);
+	}
+	if (ctx->flags & EBPF_SAVE_S0) {
+		emit_instr_long(ctx, sd, sw,
+					MIPS_R_S0, store_offset, MIPS_R_SP);
+		store_offset -= sizeof(long);
+	}
+
+	/* Store TCC in backup register or stack scratch space if indicated. */
+	if (tail_call_present(ctx) && !(ctx->flags & EBPF_TCC_IN_RUN)) {
+		if (tcc_sav)
+			emit_instr(ctx, move, tcc_sav, tcc_run);
+		else
+			emit_instr_long(ctx, sd, sw,
+					tcc_run, ctx->bpf_stack_off, MIPS_R_SP);
+	}
 
-	if ((ctx->flags & EBPF_SEEN_TC) && !(ctx->flags & EBPF_TCC_IN_V1))
-		emit_instr_long(ctx, daddu, addu,
-					MIPS_R_S4, MIPS_R_V1, MIPS_R_ZERO);
+	/* Prepare BPF FP as single-reg ptr, emulate upper 32-bits as needed.*/
+	if (ctx->flags & EBPF_SEEN_FP)
+		emit_instr_long(ctx, daddiu, addiu, r10,
+						MIPS_R_SP, ctx->bpf_stack_off);
 
 	return 0;
 }
@@ -354,14 +561,38 @@  static int build_int_epilogue(struct jit_ctx *ctx, int dest_reg)
 	const struct bpf_prog *prog = ctx->skf;
 	int stack_adjust = ctx->stack_size;
 	int store_offset = stack_adjust - sizeof(long);
+	int r1 = bpf2mips[BPF_REG_1].reg;
+	int r0 = bpf2mips[BPF_REG_0].reg;
 	enum reg_val_type td;
-	int r0 = MIPS_R_V0;
 
-	if (dest_reg == MIPS_R_RA) {
-		/* Don't let zero extended value escape. */
-		td = get_reg_val_type(ctx, prog->len, BPF_REG_0);
-		if (td == REG_64BIT)
-			emit_instr(ctx, sll, r0, r0, 0);
+	/*
+	 * Returns from BPF2BPF calls consistently use the BPF 64-bit ABI
+	 * i.e. register usage and mapping between JIT and OS is unchanged.
+	 * Returning to the kernel must follow the N64 or O32 ABI, and for
+	 * the latter requires fixup of BPF R0 to MIPS V0 register mapping.
+	 *
+	 * Tails calls must ensure the passed R1 context is consistent with
+	 * the kernel ABI, and requires fixup on MIPS32 bigendian systems.
+	 */
+	if (dest_reg == MIPS_R_RA && !ctx->skf->is_func) { /* kernel return */
+		if (is64bit()) {
+			/* Don't let zero extended value escape. */
+			td = get_reg_val_type(ctx, prog->len, BPF_REG_0);
+			if (td == REG_64BIT)
+				gen_sext_insn(r0, ctx);
+		} else if (isbigend()) { /* and 32-bit */
+			/*
+			 * O32 ABI specifies 32-bit return value always
+			 * placed in MIPS_R_V0 regardless of the native
+			 * endianness. This would be in the wrong position
+			 * in a BPF R0 reg pair on big-endian systems, so
+			 * we must relocate.
+			 */
+			emit_instr(ctx, move, MIPS_R_V0, LO(r0));
+		}
+	} else if (dest_reg == MIPS_R_T9) { /* tail call */
+		if (!is64bit() && isbigend())
+			emit_instr(ctx, move, MIPS_R_A0, LO(r1));
 	}
 
 	if (ctx->flags & EBPF_SAVE_RA) {
@@ -369,24 +600,24 @@  static int build_int_epilogue(struct jit_ctx *ctx, int dest_reg)
 					MIPS_R_RA, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
-	if (ctx->flags & EBPF_SAVE_S0) {
+	if (ctx->flags & EBPF_SAVE_S8) {
 		emit_instr_long(ctx, ld, lw,
-					MIPS_R_S0, store_offset, MIPS_R_SP);
+					MIPS_R_S8, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
-	if (ctx->flags & EBPF_SAVE_S1) {
+	if (ctx->flags & EBPF_SAVE_S7) {
 		emit_instr_long(ctx, ld, lw,
-					MIPS_R_S1, store_offset, MIPS_R_SP);
+					MIPS_R_S7, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
-	if (ctx->flags & EBPF_SAVE_S2) {
+	if (ctx->flags & EBPF_SAVE_S6) {
 		emit_instr_long(ctx, ld, lw,
-				MIPS_R_S2, store_offset, MIPS_R_SP);
+					MIPS_R_S6, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
-	if (ctx->flags & EBPF_SAVE_S3) {
+	if (ctx->flags & EBPF_SAVE_S5) {
 		emit_instr_long(ctx, ld, lw,
-					MIPS_R_S3, store_offset, MIPS_R_SP);
+					MIPS_R_S5, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
 	if (ctx->flags & EBPF_SAVE_S4) {
@@ -394,8 +625,29 @@  static int build_int_epilogue(struct jit_ctx *ctx, int dest_reg)
 					MIPS_R_S4, store_offset, MIPS_R_SP);
 		store_offset -= sizeof(long);
 	}
+	if (ctx->flags & EBPF_SAVE_S3) {
+		emit_instr_long(ctx, ld, lw,
+					MIPS_R_S3, store_offset, MIPS_R_SP);
+		store_offset -= sizeof(long);
+	}
+	if (ctx->flags & EBPF_SAVE_S2) {
+		emit_instr_long(ctx, ld, lw,
+					MIPS_R_S2, store_offset, MIPS_R_SP);
+		store_offset -= sizeof(long);
+	}
+	if (ctx->flags & EBPF_SAVE_S1) {
+		emit_instr_long(ctx, ld, lw,
+					MIPS_R_S1, store_offset, MIPS_R_SP);
+		store_offset -= sizeof(long);
+	}
+	if (ctx->flags & EBPF_SAVE_S0) {
+		emit_instr_long(ctx, ld, lw,
+					MIPS_R_S0, store_offset, MIPS_R_SP);
+		store_offset -= sizeof(long);
+	}
 	emit_instr(ctx, jr, dest_reg);
 
+	/* Delay slot */
 	if (stack_adjust)
 		emit_instr_long(ctx, daddiu, addiu,
 					MIPS_R_SP, MIPS_R_SP, stack_adjust);
@@ -415,7 +667,9 @@  static void gen_imm_to_reg(const struct bpf_insn *insn, int reg,
 		int upper = insn->imm - lower;
 
 		emit_instr(ctx, lui, reg, upper >> 16);
-		emit_instr(ctx, addiu, reg, reg, lower);
+		/* lui already clears lower halfword */
+		if (lower)
+			emit_instr(ctx, addiu, reg, reg, lower);
 	}
 }
 
@@ -564,12 +818,12 @@  static int gen_imm_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
 	return 0;
 }
 
-static void emit_const_to_reg(struct jit_ctx *ctx, int dst, u64 value)
+static void emit_const_to_reg(struct jit_ctx *ctx, int dst, unsigned long value)
 {
-	if (value >= 0xffffffffffff8000ull || value < 0x8000ull) {
-		emit_instr(ctx, daddiu, dst, MIPS_R_ZERO, (int)value);
-	} else if (value >= 0xffffffff80000000ull ||
-		   (value < 0x80000000 && value > 0xffff)) {
+	if (value >= S16_MIN || value <= S16_MAX) {
+		emit_instr_long(ctx, daddiu, addiu, dst, MIPS_R_ZERO, (int)value);
+	} else if (value >= S32_MIN ||
+		   (value <= S32_MAX && value > U16_MAX)) {
 		emit_instr(ctx, lui, dst, (s32)(s16)(value >> 16));
 		emit_instr(ctx, ori, dst, dst, (unsigned int)(value & 0xffff));
 	} else {
@@ -601,54 +855,145 @@  static void emit_const_to_reg(struct jit_ctx *ctx, int dst, u64 value)
 	}
 }
 
+/*
+ * Push BPF regs R3-R5 to the stack, skipping BPF regs R1-R2 which are
+ * passed via MIPS register pairs in $a0-$a3. Register order within pairs
+ * and the memory storage order are identical i.e. endian native.
+ */
+static void emit_push_args(struct jit_ctx *ctx)
+{
+	int store_offset = 2 * sizeof(u64); /* Skip R1-R2 in $a0-$a3 */
+	int bpf, reg;
+
+	for (bpf = BPF_REG_3; bpf <= BPF_REG_5; bpf++) {
+		reg = bpf2mips[bpf].reg;
+
+		emit_instr(ctx, sw, LO(reg), OFFLO(store_offset), MIPS_R_SP);
+		emit_instr(ctx, sw, HI(reg), OFFHI(store_offset), MIPS_R_SP);
+		store_offset += sizeof(u64);
+	}
+}
+
+/*
+ * Common helper for BPF_CALL insn, handling TCC and ABI variations.
+ * Kernel calls under O32 ABI require arguments passed on the stack,
+ * while BPF2BPF calls need the TCC passed via register as expected
+ * by the subprog's prologue.
+ *
+ * Under MIPS32 O32 ABI calling convention, u64 BPF regs R1-R2 are passed
+ * via reg pairs in $a0-$a3, while BPF regs R3-R5 are passed via the stack.
+ * Stack space is still reserved for $a0-$a3, and the whole area aligned.
+ */
+#define ARGS_SIZE (5 * sizeof(u64))
+
+void emit_bpf_call(struct jit_ctx *ctx, const struct bpf_insn *insn)
+{
+	int stack_adjust = ALIGN(ARGS_SIZE, STACK_ALIGN);
+	int tcc_run = bpf2mips[JIT_RUN_TCC].reg ?
+		      bpf2mips[JIT_RUN_TCC].reg :
+		      TEMP_PASS_TCC;
+	int tcc_sav = bpf2mips[JIT_SAV_TCC].reg;
+	long func_addr;
+
+	ctx->flags |= EBPF_SAVE_RA;
+
+	/* Ensure TCC passed into BPF subprog */
+	if ((insn->src_reg == BPF_PSEUDO_CALL) &&
+	    tail_call_present(ctx) && !(ctx->flags & EBPF_TCC_IN_RUN)) {
+		/* Set TCC from reg or stack */
+		if (tcc_sav)
+			emit_instr(ctx, move, tcc_run, tcc_sav);
+		else
+			emit_instr_long(ctx, ld, lw, tcc_run,
+						ctx->bpf_stack_off, MIPS_R_SP);
+	}
+
+	/* Push O32 stack args for kernel call */
+	if (!is64bit() && (insn->src_reg != BPF_PSEUDO_CALL)) {
+		emit_instr(ctx, addiu, MIPS_R_SP, MIPS_R_SP, -stack_adjust);
+		emit_push_args(ctx);
+	}
+
+	func_addr = (long)__bpf_call_base + insn->imm;
+	emit_const_to_reg(ctx, MIPS_R_T9, func_addr);
+	emit_instr(ctx, jalr, MIPS_R_RA, MIPS_R_T9);
+	/* Delay slot */
+	emit_instr(ctx, nop);
+
+	/* Restore stack */
+	if (!is64bit() && (insn->src_reg != BPF_PSEUDO_CALL))
+		emit_instr(ctx, addiu, MIPS_R_SP, MIPS_R_SP, stack_adjust);
+}
+
+/*
+ * Tail call helper arguments passed via BPF ABI as u64 parameters. On
+ * MIPS64 N64 ABI systems these are native regs, while on MIPS32 O32 ABI
+ * systems these are reg pairs:
+ *
+ * R1 -> &ctx
+ * R2 -> &array
+ * R3 -> index
+ */
 static int emit_bpf_tail_call(struct jit_ctx *ctx, int this_idx)
 {
+	int tcc_run = bpf2mips[JIT_RUN_TCC].reg ?
+		      bpf2mips[JIT_RUN_TCC].reg :
+		      TEMP_PASS_TCC;
+	int tcc_sav = bpf2mips[JIT_SAV_TCC].reg;
+	int r2 = bpf2mips[BPF_REG_2].reg;
+	int r3 = bpf2mips[BPF_REG_3].reg;
 	int off, b_off;
-	int tcc_reg;
+	int tcc;
 
 	ctx->flags |= EBPF_SEEN_TC;
 	/*
 	 * if (index >= array->map.max_entries)
 	 *     goto out;
 	 */
-	/* Mask index as 32-bit */
-	emit_instr(ctx, dinsu, MIPS_R_A2, MIPS_R_ZERO, 32, 32);
+	if (is64bit())
+		/* Mask index as 32-bit */
+		gen_zext_insn(r3, true, ctx);
 	off = offsetof(struct bpf_array, map.max_entries);
-	emit_instr(ctx, lwu, MIPS_R_T5, off, MIPS_R_A1);
-	emit_instr(ctx, sltu, MIPS_R_AT, MIPS_R_T5, MIPS_R_A2);
+	emit_instr_long(ctx, lwu, lw, MIPS_R_AT, off, LO(r2));
+	emit_instr(ctx, sltu, MIPS_R_AT, MIPS_R_AT, LO(r3));
 	b_off = b_imm(this_idx + 1, ctx);
-	emit_instr(ctx, bne, MIPS_R_AT, MIPS_R_ZERO, b_off);
+	emit_instr(ctx, bnez, MIPS_R_AT, b_off);
 	/*
 	 * if (TCC-- < 0)
 	 *     goto out;
 	 */
 	/* Delay slot */
-	tcc_reg = (ctx->flags & EBPF_TCC_IN_V1) ? MIPS_R_V1 : MIPS_R_S4;
-	emit_instr(ctx, daddiu, MIPS_R_T5, tcc_reg, -1);
+	tcc = (ctx->flags & EBPF_TCC_IN_RUN) ? tcc_run : tcc_sav;
+	/* Get TCC from reg or stack */
+	if (tcc)
+		emit_instr(ctx, move, MIPS_R_T8, tcc);
+	else
+		emit_instr_long(ctx, ld, lw, MIPS_R_T8,
+						ctx->bpf_stack_off, MIPS_R_SP);
 	b_off = b_imm(this_idx + 1, ctx);
-	emit_instr(ctx, bltz, tcc_reg, b_off);
+	emit_instr(ctx, bltz, MIPS_R_T8, b_off);
 	/*
 	 * prog = array->ptrs[index];
 	 * if (prog == NULL)
 	 *     goto out;
 	 */
 	/* Delay slot */
-	emit_instr(ctx, dsll, MIPS_R_T8, MIPS_R_A2, 3);
-	emit_instr(ctx, daddu, MIPS_R_T8, MIPS_R_T8, MIPS_R_A1);
+	emit_instr_long(ctx, dsll, sll, MIPS_R_AT, LO(r3), ilog2(sizeof(long)));
+	emit_instr_long(ctx, daddu, addu, MIPS_R_AT, MIPS_R_AT, LO(r2));
 	off = offsetof(struct bpf_array, ptrs);
-	emit_instr(ctx, ld, MIPS_R_AT, off, MIPS_R_T8);
+	emit_instr_long(ctx, ld, lw, MIPS_R_AT, off, MIPS_R_AT);
 	b_off = b_imm(this_idx + 1, ctx);
-	emit_instr(ctx, beq, MIPS_R_AT, MIPS_R_ZERO, b_off);
+	emit_instr(ctx, beqz, MIPS_R_AT, b_off);
 	/* Delay slot */
 	emit_instr(ctx, nop);
 
 	/* goto *(prog->bpf_func + 4); */
 	off = offsetof(struct bpf_prog, bpf_func);
-	emit_instr(ctx, ld, MIPS_R_T9, off, MIPS_R_AT);
-	/* All systems are go... propagate TCC */
-	emit_instr(ctx, daddu, MIPS_R_V1, MIPS_R_T5, MIPS_R_ZERO);
+	emit_instr_long(ctx, ld, lw, MIPS_R_T9, off, MIPS_R_AT);
+	/* All systems are go... decrement and propagate TCC */
+	emit_instr_long(ctx, daddiu, addiu, tcc_run, MIPS_R_T8, -1);
 	/* Skip first instruction (TCC initialization) */
-	emit_instr(ctx, daddiu, MIPS_R_T9, MIPS_R_T9, 4);
+	emit_instr_long(ctx, daddiu, addiu, MIPS_R_T9, MIPS_R_T9, 4);
 	return build_int_epilogue(ctx, MIPS_R_T9);
 }
 
@@ -828,15 +1173,7 @@  static int build_one_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
 		if (get_reg_val_type(ctx, this_idx, insn->dst_reg) == REG_32BIT)
 			emit_instr(ctx, dinsu, dst, MIPS_R_ZERO, 32, 32);
 		did_move = false;
-		if (insn->src_reg == BPF_REG_10) {
-			if (bpf_op == BPF_MOV) {
-				emit_instr(ctx, daddiu, dst, MIPS_R_SP, MAX_BPF_STACK);
-				did_move = true;
-			} else {
-				emit_instr(ctx, daddiu, MIPS_R_AT, MIPS_R_SP, MAX_BPF_STACK);
-				src = MIPS_R_AT;
-			}
-		} else if (get_reg_val_type(ctx, this_idx, insn->src_reg) == REG_32BIT) {
+		if (get_reg_val_type(ctx, this_idx, insn->src_reg) == REG_32BIT) {
 			int tmp_reg = MIPS_R_AT;
 
 			if (bpf_op == BPF_MOV) {
@@ -917,7 +1254,7 @@  static int build_one_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
 	case BPF_ALU | BPF_LSH | BPF_X: /* ALU_REG */
 	case BPF_ALU | BPF_RSH | BPF_X: /* ALU_REG */
 	case BPF_ALU | BPF_ARSH | BPF_X: /* ALU_REG */
-		src = ebpf_to_mips_reg(ctx, insn, REG_SRC_NO_FP);
+		src = ebpf_to_mips_reg(ctx, insn, REG_SRC_FP_OK);
 		dst = ebpf_to_mips_reg(ctx, insn, REG_DST_NO_FP);
 		if (src < 0 || dst < 0)
 			return -EINVAL;
@@ -1029,8 +1366,8 @@  static int build_one_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
 	case BPF_JMP | BPF_JGT | BPF_X:
 	case BPF_JMP | BPF_JGE | BPF_X:
 	case BPF_JMP | BPF_JSET | BPF_X:
-		src = ebpf_to_mips_reg(ctx, insn, REG_SRC_NO_FP);
-		dst = ebpf_to_mips_reg(ctx, insn, REG_DST_NO_FP);
+		src = ebpf_to_mips_reg(ctx, insn, REG_SRC_FP_OK);
+		dst = ebpf_to_mips_reg(ctx, insn, REG_DST_FP_OK);
 		if (src < 0 || dst < 0)
 			return -EINVAL;
 		td = get_reg_val_type(ctx, this_idx, insn->dst_reg);
@@ -1311,12 +1648,7 @@  static int build_one_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
 		return 2; /* Double slot insn */
 
 	case BPF_JMP | BPF_CALL:
-		ctx->flags |= EBPF_SAVE_RA;
-		t64s = (s64)insn->imm + (long)__bpf_call_base;
-		emit_const_to_reg(ctx, MIPS_R_T9, (u64)t64s);
-		emit_instr(ctx, jalr, MIPS_R_RA, MIPS_R_T9);
-		/* delay slot */
-		emit_instr(ctx, nop);
+		emit_bpf_call(ctx, insn);
 		break;
 
 	case BPF_JMP | BPF_TAIL_CALL:
@@ -1364,16 +1696,10 @@  static int build_one_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
 	case BPF_ST | BPF_H | BPF_MEM:
 	case BPF_ST | BPF_W | BPF_MEM:
 	case BPF_ST | BPF_DW | BPF_MEM:
-		if (insn->dst_reg == BPF_REG_10) {
-			ctx->flags |= EBPF_SEEN_FP;
-			dst = MIPS_R_SP;
-			mem_off = insn->off + MAX_BPF_STACK;
-		} else {
-			dst = ebpf_to_mips_reg(ctx, insn, REG_DST_NO_FP);
-			if (dst < 0)
-				return dst;
-			mem_off = insn->off;
-		}
+		dst = ebpf_to_mips_reg(ctx, insn, REG_DST_FP_OK);
+		if (dst < 0)
+			return dst;
+		mem_off = insn->off;
 		gen_imm_to_reg(insn, MIPS_R_AT, ctx);
 		switch (BPF_SIZE(insn->code)) {
 		case BPF_B:
@@ -1395,19 +1721,11 @@  static int build_one_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
 	case BPF_LDX | BPF_H | BPF_MEM:
 	case BPF_LDX | BPF_W | BPF_MEM:
 	case BPF_LDX | BPF_DW | BPF_MEM:
-		if (insn->src_reg == BPF_REG_10) {
-			ctx->flags |= EBPF_SEEN_FP;
-			src = MIPS_R_SP;
-			mem_off = insn->off + MAX_BPF_STACK;
-		} else {
-			src = ebpf_to_mips_reg(ctx, insn, REG_SRC_NO_FP);
-			if (src < 0)
-				return src;
-			mem_off = insn->off;
-		}
 		dst = ebpf_to_mips_reg(ctx, insn, REG_DST_NO_FP);
-		if (dst < 0)
-			return dst;
+		src = ebpf_to_mips_reg(ctx, insn, REG_SRC_FP_OK);
+		if (dst < 0 || src < 0)
+			return -EINVAL;
+		mem_off = insn->off;
 		switch (BPF_SIZE(insn->code)) {
 		case BPF_B:
 			emit_instr(ctx, lbu, dst, mem_off, src);
@@ -1430,25 +1748,16 @@  static int build_one_insn(const struct bpf_insn *insn, struct jit_ctx *ctx,
 	case BPF_STX | BPF_DW | BPF_MEM:
 	case BPF_STX | BPF_W | BPF_ATOMIC:
 	case BPF_STX | BPF_DW | BPF_ATOMIC:
-		if (insn->dst_reg == BPF_REG_10) {
-			ctx->flags |= EBPF_SEEN_FP;
-			dst = MIPS_R_SP;
-			mem_off = insn->off + MAX_BPF_STACK;
-		} else {
-			dst = ebpf_to_mips_reg(ctx, insn, REG_DST_NO_FP);
-			if (dst < 0)
-				return dst;
-			mem_off = insn->off;
-		}
-		src = ebpf_to_mips_reg(ctx, insn, REG_SRC_NO_FP);
-		if (src < 0)
-			return src;
+		dst = ebpf_to_mips_reg(ctx, insn, REG_DST_FP_OK);
+		src = ebpf_to_mips_reg(ctx, insn, REG_SRC_FP_OK);
+		if (src < 0 || dst < 0)
+			return -EINVAL;
+		mem_off = insn->off;
 		if (BPF_MODE(insn->code) == BPF_ATOMIC) {
 			if (insn->imm != BPF_ADD) {
 				pr_err("ATOMIC OP %02x NOT HANDLED\n", insn->imm);
 				return -EINVAL;
 			}
-
 			/*
 			 * If mem_off does not fit within the 9 bit ll/sc
 			 * instruction immediate field, use a temp reg.
@@ -1829,6 +2138,73 @@  static void jit_fill_hole(void *area, unsigned int size)
 		uasm_i_break(&p, BRK_BUG); /* Increments p */
 }
 
+/*
+ * Save and restore the BPF VM state across a direct kernel call. This
+ * includes the caller-saved registers used for BPF_REG_0 .. BPF_REG_5
+ * and BPF_REG_AX used by the verifier for blinding and other dark arts.
+ * Restore avoids clobbering bpf_ret, which holds the call return value.
+ * BPF_REG_6 .. BPF_REG_10 and TCC are already callee-saved or on stack.
+ */
+static const int bpf_caller_save[] = {
+	BPF_REG_0,
+	BPF_REG_1,
+	BPF_REG_2,
+	BPF_REG_3,
+	BPF_REG_4,
+	BPF_REG_5,
+	BPF_REG_AX,
+};
+
+#define CALLER_ENV_SIZE (ARRAY_SIZE(bpf_caller_save) * sizeof(u64))
+
+void emit_caller_save(struct jit_ctx *ctx)
+{
+	int stack_adj = ALIGN(CALLER_ENV_SIZE, STACK_ALIGN);
+	int i, bpf, reg, store_offset;
+
+	emit_instr_long(ctx, daddiu, addiu, MIPS_R_SP, MIPS_R_SP, -stack_adj);
+
+	for (i = 0; i < ARRAY_SIZE(bpf_caller_save); i++) {
+		bpf = bpf_caller_save[i];
+		reg = bpf2mips[bpf].reg;
+		store_offset = i * sizeof(u64);
+
+		if (is64bit()) {
+			emit_instr(ctx, sd, reg, store_offset, MIPS_R_SP);
+		} else {
+			emit_instr(ctx, sw, LO(reg),
+						OFFLO(store_offset), MIPS_R_SP);
+			emit_instr(ctx, sw, HI(reg),
+						OFFHI(store_offset), MIPS_R_SP);
+		}
+	}
+}
+
+void emit_caller_restore(struct jit_ctx *ctx, int bpf_ret)
+{
+	int stack_adj = ALIGN(CALLER_ENV_SIZE, STACK_ALIGN);
+	int i, bpf, reg, store_offset;
+
+	for (i = 0; i < ARRAY_SIZE(bpf_caller_save); i++) {
+		bpf = bpf_caller_save[i];
+		reg = bpf2mips[bpf].reg;
+		store_offset = i * sizeof(u64);
+		if (bpf == bpf_ret)
+			continue;
+
+		if (is64bit()) {
+			emit_instr(ctx, ld, reg, store_offset, MIPS_R_SP);
+		} else {
+			emit_instr(ctx, lw, LO(reg),
+						OFFLO(store_offset), MIPS_R_SP);
+			emit_instr(ctx, lw, HI(reg),
+						OFFHI(store_offset), MIPS_R_SP);
+		}
+	}
+
+	emit_instr_long(ctx, daddiu, addiu, MIPS_R_SP, MIPS_R_SP, stack_adj);
+}
+
 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
 {
 	struct bpf_prog *orig_prog = prog;
@@ -1889,14 +2265,14 @@  struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
 		goto out_err;
 
 	/*
-	 * If no calls are made (EBPF_SAVE_RA), then tail call count
-	 * in $v1, else we must save in n$s4.
+	 * If no calls are made (EBPF_SAVE_RA), then tailcall count located
+	 * in runtime reg if defined, else we backup to save reg or stack.
 	 */
-	if (ctx.flags & EBPF_SEEN_TC) {
+	if (tail_call_present(&ctx)) {
 		if (ctx.flags & EBPF_SAVE_RA)
-			ctx.flags |= EBPF_SAVE_S4;
-		else
-			ctx.flags |= EBPF_TCC_IN_V1;
+			ctx.flags |= bpf2mips[JIT_SAV_TCC].flags;
+		else if (bpf2mips[JIT_RUN_TCC].reg)
+			ctx.flags |= EBPF_TCC_IN_RUN;
 	}
 
 	/*
@@ -1910,7 +2286,7 @@  struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
 		ctx.idx = 0;
 		ctx.gen_b_offsets = 1;
 		ctx.long_b_conversion = 0;
-		if (gen_int_prologue(&ctx))
+		if (build_int_prologue(&ctx))
 			goto out_err;
 		if (build_int_body(&ctx))
 			goto out_err;
@@ -1929,7 +2305,7 @@  struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
 
 	/* Third pass generates the code */
 	ctx.idx = 0;
-	if (gen_int_prologue(&ctx))
+	if (build_int_prologue(&ctx))
 		goto out_err;
 	if (build_int_body(&ctx))
 		goto out_err;