diff mbox series

[RFC,v4,17/17] powerpc64/bpf: Add support for bpf trampolines

Message ID c522d78c5969ace231ccea3112d4ffe668ab82ff.1720942106.git.naveen@kernel.org (mailing list archive)
State Handled Elsewhere
Headers show
Series powerpc: Core ftrace rework, support for ftrace direct and bpf trampolines | expand

Commit Message

Naveen N Rao July 14, 2024, 8:27 a.m. UTC
Add support for bpf_arch_text_poke() and arch_prepare_bpf_trampoline()
for 64-bit powerpc. While the code is generic, BPF trampolines are only
enabled on 64-bit powerpc. 32-bit powerpc will need testing and some
updates.

BPF Trampolines adhere to the existing ftrace ABI utilizing a
two-instruction profiling sequence, as well as the newer ABI utilizing a
three-instruction profiling sequence enabling return with a 'blr'. The
trampoline code itself closely follows x86 implementation.

BPF prog JIT is extended to mimic 64-bit powerpc approach for ftrace
having a single nop at function entry, followed by the function
profiling sequence out-of-line and a separate long branch stub for calls
to trampolines that are out of range. A dummy_tramp is provided to
simplify synchronization similar to arm64.

When attaching a bpf trampoline to a bpf prog, we can patch up to three
things:
- the nop at bpf prog entry to go to the out-of-line stub
- the instruction in the out-of-line stub to either call the bpf trampoline
directly, or to branch to the long_branch stub.
- the trampoline address before the long_branch stub.

We do not need any synchronization here since we always have a valid
branch target regardless of the order in which the above stores are
seen. dummy_tramp ensures that the long_branch stub goes to a valid
destination on other cpus, even when the branch to the long_branch stub
is seen before the updated trampoline address.

However, when detaching a bpf trampoline from a bpf prog, or if changing
the bpf trampoline address, we need synchronization to ensure that other
cpus can no longer branch into the older trampoline so that it can be
safely freed. bpf_tramp_image_put() uses rcu_tasks to ensure all cpus
make forward progress, but we still need to ensure that other cpus
execute isync (or some CSI) so that they don't go back into the
trampoline again.

Signed-off-by: Naveen N Rao <naveen@kernel.org>
---
 arch/powerpc/include/asm/ppc-opcode.h |  14 +
 arch/powerpc/net/bpf_jit.h            |  12 +
 arch/powerpc/net/bpf_jit_comp.c       | 842 +++++++++++++++++++++++++-
 arch/powerpc/net/bpf_jit_comp32.c     |   7 +-
 arch/powerpc/net/bpf_jit_comp64.c     |   7 +-
 5 files changed, 879 insertions(+), 3 deletions(-)
diff mbox series

Patch

diff --git a/arch/powerpc/include/asm/ppc-opcode.h b/arch/powerpc/include/asm/ppc-opcode.h
index b98a9e982c03..4312bcb913a4 100644
--- a/arch/powerpc/include/asm/ppc-opcode.h
+++ b/arch/powerpc/include/asm/ppc-opcode.h
@@ -587,12 +587,26 @@ 
 #define PPC_RAW_MTSPR(spr, d)		(0x7c0003a6 | ___PPC_RS(d) | __PPC_SPR(spr))
 #define PPC_RAW_EIEIO()			(0x7c0006ac)
 
+/* bcl 20,31,$+4 */
+#define PPC_RAW_BCL4()			(0x429f0005)
 #define PPC_RAW_BRANCH(offset)		(0x48000000 | PPC_LI(offset))
 #define PPC_RAW_BL(offset)		(0x48000001 | PPC_LI(offset))
 #define PPC_RAW_TW(t0, a, b)		(0x7c000008 | ___PPC_RS(t0) | ___PPC_RA(a) | ___PPC_RB(b))
 #define PPC_RAW_TRAP()			PPC_RAW_TW(31, 0, 0)
 #define PPC_RAW_SETB(t, bfa)		(0x7c000100 | ___PPC_RT(t) | ___PPC_RA((bfa) << 2))
 
+#ifdef CONFIG_PPC32
+#define PPC_RAW_STL		PPC_RAW_STW
+#define PPC_RAW_STLU		PPC_RAW_STWU
+#define PPC_RAW_LL		PPC_RAW_LWZ
+#define PPC_RAW_CMPLI		PPC_RAW_CMPWI
+#else
+#define PPC_RAW_STL		PPC_RAW_STD
+#define PPC_RAW_STLU		PPC_RAW_STDU
+#define PPC_RAW_LL		PPC_RAW_LD
+#define PPC_RAW_CMPLI		PPC_RAW_CMPDI
+#endif
+
 /* Deal with instructions that older assemblers aren't aware of */
 #define	PPC_BCCTR_FLUSH		stringify_in_c(.long PPC_INST_BCCTR_FLUSH)
 #define	PPC_CP_ABORT		stringify_in_c(.long PPC_RAW_CP_ABORT)
diff --git a/arch/powerpc/net/bpf_jit.h b/arch/powerpc/net/bpf_jit.h
index cdea5dccaefe..2d04ce5a23da 100644
--- a/arch/powerpc/net/bpf_jit.h
+++ b/arch/powerpc/net/bpf_jit.h
@@ -12,6 +12,7 @@ 
 
 #include <asm/types.h>
 #include <asm/ppc-opcode.h>
+#include <linux/build_bug.h>
 
 #ifdef CONFIG_PPC64_ELF_ABI_V1
 #define FUNCTION_DESCR_SIZE	24
@@ -21,6 +22,9 @@ 
 
 #define CTX_NIA(ctx) ((unsigned long)ctx->idx * 4)
 
+#define SZL			sizeof(unsigned long)
+#define BPF_INSN_SAFETY		64
+
 #define PLANT_INSTR(d, idx, instr)					      \
 	do { if (d) { (d)[idx] = instr; } idx++; } while (0)
 #define EMIT(instr)		PLANT_INSTR(image, ctx->idx, instr)
@@ -81,6 +85,13 @@ 
 				EMIT(PPC_RAW_ORI(d, d, (uintptr_t)(i) &       \
 							0xffff));             \
 		} } while (0)
+#define PPC_LI_ADDR	PPC_LI64
+#define PPC64_LOAD_PACA()						      \
+	EMIT(PPC_RAW_LD(_R2, _R13, offsetof(struct paca_struct, kernel_toc)))
+#else
+#define PPC_LI64(d, i)	BUILD_BUG()
+#define PPC_LI_ADDR	PPC_LI32
+#define PPC64_LOAD_PACA() BUILD_BUG()
 #endif
 
 /*
@@ -165,6 +176,7 @@  int bpf_jit_build_body(struct bpf_prog *fp, u32 *image, u32 *fimage, struct code
 		       u32 *addrs, int pass, bool extra_pass);
 void bpf_jit_build_prologue(u32 *image, struct codegen_context *ctx);
 void bpf_jit_build_epilogue(u32 *image, struct codegen_context *ctx);
+void bpf_jit_build_fentry_stubs(u32 *image, struct codegen_context *ctx);
 void bpf_jit_realloc_regs(struct codegen_context *ctx);
 int bpf_jit_emit_exit_insn(u32 *image, struct codegen_context *ctx, int tmp_reg, long exit_addr);
 
diff --git a/arch/powerpc/net/bpf_jit_comp.c b/arch/powerpc/net/bpf_jit_comp.c
index 984655419da5..bed1d26e7069 100644
--- a/arch/powerpc/net/bpf_jit_comp.c
+++ b/arch/powerpc/net/bpf_jit_comp.c
@@ -22,11 +22,81 @@ 
 
 #include "bpf_jit.h"
 
+/* These offsets are from bpf prog end and stay the same across progs */
+static int bpf_jit_ool_stub, bpf_jit_long_branch_stub;
+
 static void bpf_jit_fill_ill_insns(void *area, unsigned int size)
 {
 	memset32(area, BREAKPOINT_INSTRUCTION, size / 4);
 }
 
+void dummy_tramp(void);
+
+asm (
+"	.pushsection .text, \"ax\", @progbits	;"
+"	.global dummy_tramp			;"
+"	.type dummy_tramp, @function		;"
+"dummy_tramp:					;"
+#ifdef CONFIG_PPC_FTRACE_OUT_OF_LINE
+"	blr					;"
+#else
+/* LR is always in r11, so we don't need a 'mflr r11' here */
+"	mtctr	11				;"
+"	mtlr	0				;"
+"	bctr					;"
+#endif
+"	.size dummy_tramp, .-dummy_tramp	;"
+"	.popsection				;"
+);
+
+void bpf_jit_build_fentry_stubs(u32 *image, struct codegen_context *ctx)
+{
+	int ool_stub_idx, long_branch_stub_idx;
+
+	/*
+	 * Out-of-line stub:
+	 *	mflr	r0
+	 *	[b|bl]	tramp
+	 *	mtlr	r0 // only with CONFIG_PPC_FTRACE_OUT_OF_LINE
+	 *	b	bpf_func + 4
+	 */
+	ool_stub_idx = ctx->idx;
+	EMIT(PPC_RAW_MFLR(_R0));
+	EMIT(PPC_RAW_NOP());
+	if (IS_ENABLED(CONFIG_PPC_FTRACE_OUT_OF_LINE))
+		EMIT(PPC_RAW_MTLR(_R0));
+	WARN_ON_ONCE(!is_offset_in_branch_range(4 - (long)ctx->idx * 4));
+	EMIT(PPC_RAW_BRANCH(4 - (long)ctx->idx * 4));
+
+	/*
+	 * Long branch stub:
+	 *	.long	<dummy_tramp_addr>
+	 *	mflr	r11
+	 *	bcl	20,31,$+4
+	 *	mflr	r12
+	 *	ld	r12, -8-SZL(r12)
+	 *	mtctr	r12
+	 *	mtlr	r11 // needed to retain ftrace ABI
+	 *	bctr
+	 */
+	if (image)
+		*((unsigned long *)&image[ctx->idx]) = (unsigned long)dummy_tramp;
+	ctx->idx += SZL / 4;
+	long_branch_stub_idx = ctx->idx;
+	EMIT(PPC_RAW_MFLR(_R11));
+	EMIT(PPC_RAW_BCL4());
+	EMIT(PPC_RAW_MFLR(_R12));
+	EMIT(PPC_RAW_LL(_R12, _R12, -8-SZL));
+	EMIT(PPC_RAW_MTCTR(_R12));
+	EMIT(PPC_RAW_MTLR(_R11));
+	EMIT(PPC_RAW_BCTR());
+
+	if (!bpf_jit_ool_stub) {
+		bpf_jit_ool_stub = (ctx->idx - ool_stub_idx) * 4;
+		bpf_jit_long_branch_stub = (ctx->idx - long_branch_stub_idx) * 4;
+	}
+}
+
 int bpf_jit_emit_exit_insn(u32 *image, struct codegen_context *ctx, int tmp_reg, long exit_addr)
 {
 	if (!exit_addr || is_offset_in_branch_range(exit_addr - (ctx->idx * 4))) {
@@ -222,7 +292,7 @@  struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
 
 	fp->bpf_func = (void *)fimage;
 	fp->jited = 1;
-	fp->jited_len = proglen + FUNCTION_DESCR_SIZE;
+	fp->jited_len = cgctx.idx * 4 + FUNCTION_DESCR_SIZE;
 
 	if (!fp->is_func || extra_pass) {
 		if (bpf_jit_binary_pack_finalize(fp, fhdr, hdr)) {
@@ -369,3 +439,773 @@  bool bpf_jit_supports_far_kfunc_call(void)
 {
 	return IS_ENABLED(CONFIG_PPC64);
 }
+
+void *arch_alloc_bpf_trampoline(unsigned int size)
+{
+	return bpf_prog_pack_alloc(size, bpf_jit_fill_ill_insns);
+}
+
+void arch_free_bpf_trampoline(void *image, unsigned int size)
+{
+	bpf_prog_pack_free(image, size);
+}
+
+int arch_protect_bpf_trampoline(void *image, unsigned int size)
+{
+	return 0;
+}
+
+static int invoke_bpf_prog(u32 *image, u32 *ro_image, struct codegen_context *ctx,
+			   struct bpf_tramp_link *l, int regs_off, int retval_off,
+			   int run_ctx_off, bool save_ret)
+{
+	struct bpf_prog *p = l->link.prog;
+	ppc_inst_t branch_insn;
+	u32 jmp_idx;
+	int ret = 0;
+
+	/* Save cookie */
+	if (IS_ENABLED(CONFIG_PPC64)) {
+		PPC_LI64(_R3, l->cookie);
+		EMIT(PPC_RAW_STD(_R3, _R1, run_ctx_off + offsetof(struct bpf_tramp_run_ctx,
+				 bpf_cookie)));
+	} else {
+		PPC_LI32(_R3, l->cookie >> 32);
+		PPC_LI32(_R4, l->cookie);
+		EMIT(PPC_RAW_STW(_R3, _R1,
+				 run_ctx_off + offsetof(struct bpf_tramp_run_ctx, bpf_cookie)));
+		EMIT(PPC_RAW_STW(_R4, _R1,
+				 run_ctx_off + offsetof(struct bpf_tramp_run_ctx, bpf_cookie) + 4));
+	}
+
+	/* __bpf_prog_enter(p, &bpf_tramp_run_ctx) */
+	PPC_LI_ADDR(_R3, p);
+	EMIT(PPC_RAW_MR(_R25, _R3));
+	EMIT(PPC_RAW_ADDI(_R4, _R1, run_ctx_off));
+	ret = bpf_jit_emit_func_call_rel(image, ro_image, ctx, (unsigned long)bpf_trampoline_enter(p));
+	if (ret)
+		return ret;
+
+	/* Remember prog start time returned by __bpf_prog_enter */
+	EMIT(PPC_RAW_MR(_R26, _R3));
+
+	/*
+	 * if (__bpf_prog_enter(p) == 0)
+	 *	goto skip_exec_of_prog;
+	 *
+	 * Emit a nop to be later patched with conditional branch, once offset is known
+	 */
+	EMIT(PPC_RAW_CMPLI(_R3, 0));
+	jmp_idx = ctx->idx;
+	EMIT(PPC_RAW_NOP());
+
+	/* p->bpf_func(ctx) */
+	EMIT(PPC_RAW_ADDI(_R3, _R1, regs_off));
+	if (!p->jited)
+		PPC_LI_ADDR(_R4, (unsigned long)p->insnsi);
+	if (!create_branch(&branch_insn, (u32 *)&ro_image[ctx->idx], (unsigned long)p->bpf_func,
+			   BRANCH_SET_LINK)) {
+		if (image)
+			image[ctx->idx] = ppc_inst_val(branch_insn);
+		ctx->idx++;
+	} else {
+		EMIT(PPC_RAW_LL(_R12, _R25, offsetof(struct bpf_prog, bpf_func)));
+		EMIT(PPC_RAW_MTCTR(_R12));
+		EMIT(PPC_RAW_BCTRL());
+	}
+
+	if (save_ret)
+		EMIT(PPC_RAW_STL(_R3, _R1, retval_off));
+
+	/* Fix up branch */
+	if (image) {
+		if (create_cond_branch(&branch_insn, &image[jmp_idx],
+				       (unsigned long)&image[ctx->idx], COND_EQ << 16))
+			return -EINVAL;
+		image[jmp_idx] = ppc_inst_val(branch_insn);
+	}
+
+	/* __bpf_prog_exit(p, start_time, &bpf_tramp_run_ctx) */
+	EMIT(PPC_RAW_MR(_R3, _R25));
+	EMIT(PPC_RAW_MR(_R4, _R26));
+	EMIT(PPC_RAW_ADDI(_R5, _R1, run_ctx_off));
+	ret = bpf_jit_emit_func_call_rel(image, ro_image, ctx, (unsigned long)bpf_trampoline_exit(p));
+
+	return ret;
+}
+
+static int invoke_bpf_mod_ret(u32 *image, u32 *ro_image, struct codegen_context *ctx,
+			      struct bpf_tramp_links *tl, int regs_off, int retval_off,
+			      int run_ctx_off, u32 *branches)
+{
+	int i;
+
+	/*
+	 * The first fmod_ret program will receive a garbage return value.
+	 * Set this to 0 to avoid confusing the program.
+	 */
+	EMIT(PPC_RAW_LI(_R3, 0));
+	EMIT(PPC_RAW_STL(_R3, _R1, retval_off));
+	for (i = 0; i < tl->nr_links; i++) {
+		if (invoke_bpf_prog(image, ro_image, ctx, tl->links[i], regs_off, retval_off,
+				    run_ctx_off, true))
+			return -EINVAL;
+
+		/*
+		 * mod_ret prog stored return value after prog ctx. Emit:
+		 * if (*(u64 *)(ret_val) !=  0)
+		 *	goto do_fexit;
+		 */
+		EMIT(PPC_RAW_LL(_R3, _R1, retval_off));
+		EMIT(PPC_RAW_CMPLI(_R3, 0));
+
+		/*
+		 * Save the location of the branch and generate a nop, which is
+		 * replaced with a conditional jump once do_fexit (i.e. the
+		 * start of the fexit invocation) is finalized.
+		 */
+		branches[i] = ctx->idx;
+		EMIT(PPC_RAW_NOP());
+	}
+
+	return 0;
+}
+
+static void bpf_trampoline_setup_tail_call_cnt(u32 *image, struct codegen_context *ctx,
+					       int func_frame_offset, int r4_off)
+{
+	if (IS_ENABLED(CONFIG_PPC64)) {
+		/* See bpf_jit_stack_tailcallcnt() */
+		int tailcallcnt_offset = 6 * 8;
+
+		EMIT(PPC_RAW_LL(_R3, _R1, func_frame_offset - tailcallcnt_offset));
+		EMIT(PPC_RAW_STL(_R3, _R1, -tailcallcnt_offset));
+	} else {
+		/* See bpf_jit_stack_offsetof() and BPF_PPC_TC */
+		EMIT(PPC_RAW_LL(_R4, _R1, r4_off));
+	}
+}
+
+static void bpf_trampoline_restore_tail_call_cnt(u32 *image, struct codegen_context *ctx,
+						 int func_frame_offset, int r4_off)
+{
+	if (IS_ENABLED(CONFIG_PPC64)) {
+		/* See bpf_jit_stack_tailcallcnt() */
+		int tailcallcnt_offset = 6 * 8;
+
+		EMIT(PPC_RAW_LL(_R3, _R1, -tailcallcnt_offset));
+		EMIT(PPC_RAW_STL(_R3, _R1, func_frame_offset - tailcallcnt_offset));
+	} else {
+		/* See bpf_jit_stack_offsetof() and BPF_PPC_TC */
+		EMIT(PPC_RAW_STL(_R4, _R1, r4_off));
+	}
+}
+
+static void bpf_trampoline_save_args(u32 *image, struct codegen_context *ctx, int func_frame_offset,
+				     int nr_regs, int regs_off)
+{
+	int param_save_area_offset;
+
+	param_save_area_offset = func_frame_offset; /* the two frames we alloted */
+	param_save_area_offset += STACK_FRAME_MIN_SIZE; /* param save area is past the frame header */
+
+	for (int i = 0; i < nr_regs; i++) {
+		if (i < 8) {
+			EMIT(PPC_RAW_STL(_R3 + i, _R1, regs_off + i * SZL));
+		} else {
+			EMIT(PPC_RAW_LL(_R3, _R1, param_save_area_offset + i * SZL));
+			EMIT(PPC_RAW_STL(_R3, _R1, regs_off + i * SZL));
+		}
+	}
+}
+
+/* Used when restoring just the register parameters when returning back */
+static void bpf_trampoline_restore_args_regs(u32 *image, struct codegen_context *ctx,
+					     int nr_regs, int regs_off)
+{
+	for (int i = 0; i < nr_regs && i < 8; i++)
+		EMIT(PPC_RAW_LL(_R3 + i, _R1, regs_off + i * SZL));
+}
+
+/* Used when we call into the traced function. Replicate parameter save area */
+static void bpf_trampoline_restore_args_stack(u32 *image, struct codegen_context *ctx,
+					      int func_frame_offset, int nr_regs, int regs_off)
+{
+	int param_save_area_offset;
+
+	param_save_area_offset = func_frame_offset; /* the two frames we alloted */
+	param_save_area_offset += STACK_FRAME_MIN_SIZE; /* param save area is past the frame header */
+
+	for (int i = 8; i < nr_regs; i++) {
+		EMIT(PPC_RAW_LL(_R3, _R1, param_save_area_offset + i * SZL));
+		EMIT(PPC_RAW_STL(_R3, _R1, STACK_FRAME_MIN_SIZE + i * SZL));
+	}
+	bpf_trampoline_restore_args_regs(image, ctx, nr_regs, regs_off);
+}
+
+static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *rw_image,
+					 void *rw_image_end, void *ro_image,
+					 const struct btf_func_model *m, u32 flags,
+					 struct bpf_tramp_links *tlinks,
+					 void *func_addr)
+{
+	int regs_off, nregs_off, ip_off, run_ctx_off, retval_off, nvr_off, alt_lr_off, r4_off;
+	int i, ret, nr_regs, bpf_frame_size = 0, bpf_dummy_frame_size = 0, func_frame_offset;
+	struct bpf_tramp_links *fmod_ret = &tlinks[BPF_TRAMP_MODIFY_RETURN];
+	struct bpf_tramp_links *fentry = &tlinks[BPF_TRAMP_FENTRY];
+	struct bpf_tramp_links *fexit = &tlinks[BPF_TRAMP_FEXIT];
+	struct codegen_context codegen_ctx, *ctx;
+	u32 *image = (u32 *)rw_image;
+	ppc_inst_t branch_insn;
+	u32 *branches = NULL;
+	bool save_ret;
+
+	if (IS_ENABLED(CONFIG_PPC32))
+		return -ENOTSUPP;
+
+	nr_regs = m->nr_args;
+	/* Extra registers for struct arguments */
+	for (i = 0; i < m->nr_args; i++)
+		if (m->arg_size[i] > SZL)
+			nr_regs += round_up(m->arg_size[i], SZL) / SZL - 1;
+
+	if (nr_regs > MAX_BPF_FUNC_ARGS)
+		return -ENOTSUPP;
+
+	ctx = &codegen_ctx;
+	memset(ctx, 0, sizeof(*ctx));
+
+	/*
+	 * Generated stack layout:
+	 *
+	 * func prev back chain         [ back chain        ]
+	 *                              [                   ]
+	 * bpf prog redzone/tailcallcnt [ ...               ] 64 bytes (64-bit powerpc)
+	 *                              [                   ] --
+	 * LR save area                 [ r0 save (64-bit)  ]   | header
+	 *                              [ r0 save (32-bit)  ]   |
+	 * dummy frame for unwind       [ back chain 1      ] --
+	 *                              [ padding           ] align stack frame
+	 *       r4_off                 [ r4 (tailcallcnt)  ] optional - 32-bit powerpc
+	 *       alt_lr_off             [ real lr (ool stub)] optional - actual lr
+	 *                              [ r26               ]
+	 *       nvr_off                [ r25               ] nvr save area
+	 *       retval_off             [ return value      ]
+	 *                              [ reg argN          ]
+	 *                              [ ...               ]
+	 *       regs_off               [ reg_arg1          ] prog ctx context
+	 *       nregs_off              [ args count        ]
+	 *       ip_off                 [ traced function   ]
+	 *                              [ ...               ]
+	 *       run_ctx_off            [ bpf_tramp_run_ctx ]
+	 *                              [ reg argN          ]
+	 *                              [ ...               ]
+	 *       param_save_area        [ reg_arg1          ] min 8 doublewords, per ABI
+	 *                              [ TOC save (64-bit) ] --
+	 *                              [ LR save (64-bit)  ]   | header
+	 *                              [ LR save (32-bit)  ]   |
+	 * bpf trampoline frame	        [ back chain 2      ] --
+	 *
+	 */
+
+	/* Minimum stack frame header */
+	bpf_frame_size = STACK_FRAME_MIN_SIZE;
+
+	/*
+	 * Room for parameter save area.
+	 *
+	 * Per the ABI, this is required if we call into the traced function (BPF_TRAMP_F_CALL_ORIG):
+	 * - if the function takes more than 8 arguments for the rest to spill onto the stack
+	 * - or, if the function has variadic arguments
+	 * - or, if this functions's prototype was not available to the caller
+	 *
+	 * Reserve space for at least 8 registers for now. This can be optimized later.
+	 */
+	bpf_frame_size += (nr_regs > 8 ? nr_regs : 8) * SZL;
+
+	/* Room for struct bpf_tramp_run_ctx */
+	run_ctx_off = bpf_frame_size;
+	bpf_frame_size += round_up(sizeof(struct bpf_tramp_run_ctx), SZL);
+
+	/* Room for IP address argument */
+	ip_off = bpf_frame_size;
+	if (flags & BPF_TRAMP_F_IP_ARG)
+		bpf_frame_size += SZL;
+
+	/* Room for args count */
+	nregs_off = bpf_frame_size;
+	bpf_frame_size += SZL;
+
+	/* Room for args */
+	regs_off = bpf_frame_size;
+	bpf_frame_size += nr_regs * SZL;
+
+	/* Room for return value of func_addr or fentry prog */
+	retval_off = bpf_frame_size;
+	save_ret = flags & (BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_RET_FENTRY_RET);
+	if (save_ret)
+		bpf_frame_size += SZL;
+
+	/* Room for nvr save area */
+	nvr_off = bpf_frame_size;
+	bpf_frame_size += 2 * SZL;
+
+	/* Optional save area for actual LR in case of ool ftrace */
+	if (IS_ENABLED(CONFIG_PPC_FTRACE_OUT_OF_LINE)) {
+		alt_lr_off = bpf_frame_size;
+		bpf_frame_size += SZL;
+	}
+
+	if (IS_ENABLED(CONFIG_PPC32)) {
+		if (nr_regs < 2) {
+			r4_off = bpf_frame_size;
+			bpf_frame_size += SZL;
+		} else {
+			r4_off = regs_off + SZL;
+		}
+	}
+
+	/* Padding to align stack frame, if any */
+	bpf_frame_size = round_up(bpf_frame_size, SZL * 2);
+
+	/* Dummy frame size for proper unwind - includes 64-bytes red zone for 64-bit powerpc */
+	bpf_dummy_frame_size = STACK_FRAME_MIN_SIZE + 64;
+
+	/* Offset to the traced function's stack frame */
+	func_frame_offset = bpf_dummy_frame_size + bpf_frame_size;
+
+	/* Create dummy frame for unwind, store original return value */
+	EMIT(PPC_RAW_STL(_R0, _R1, PPC_LR_STKOFF));
+	/* Protect red zone where tail call count goes */
+	EMIT(PPC_RAW_STLU(_R1, _R1, -bpf_dummy_frame_size));
+
+	/* Create our stack frame */
+	EMIT(PPC_RAW_STLU(_R1, _R1, -bpf_frame_size));
+
+	/* 64-bit: Save TOC and load kernel TOC */
+	if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2) && !IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) {
+		EMIT(PPC_RAW_STD(_R2, _R1, 24));
+		PPC64_LOAD_PACA();
+	}
+
+	/* 32-bit: save tail call count in r4 */
+	if (IS_ENABLED(CONFIG_PPC32) && nr_regs < 2)
+		EMIT(PPC_RAW_STL(_R4, _R1, r4_off));
+
+	bpf_trampoline_save_args(image, ctx, func_frame_offset, nr_regs, regs_off);
+
+	/* Save our return address */
+	EMIT(PPC_RAW_MFLR(_R3));
+	if (IS_ENABLED(CONFIG_PPC_FTRACE_OUT_OF_LINE))
+		EMIT(PPC_RAW_STL(_R3, _R1, alt_lr_off));
+	else
+		EMIT(PPC_RAW_STL(_R3, _R1, bpf_frame_size + PPC_LR_STKOFF));
+
+	/*
+	 * Save ip address of the traced function.
+	 * We could recover this from LR, but we will need to address for OOL trampoline,
+	 * and optional GEP area.
+	 */
+	if (IS_ENABLED(CONFIG_PPC_FTRACE_OUT_OF_LINE) || flags & BPF_TRAMP_F_IP_ARG) {
+		EMIT(PPC_RAW_LWZ(_R4, _R3, 4));
+		EMIT(PPC_RAW_SLWI(_R4, _R4, 6));
+		EMIT(PPC_RAW_SRAWI(_R4, _R4, 6));
+		EMIT(PPC_RAW_ADD(_R3, _R3, _R4));
+		EMIT(PPC_RAW_ADDI(_R3, _R3, 4));
+	}
+
+	if (flags & BPF_TRAMP_F_IP_ARG)
+		EMIT(PPC_RAW_STL(_R3, _R1, ip_off));
+
+	if (IS_ENABLED(CONFIG_PPC_FTRACE_OUT_OF_LINE))
+		/* Fake our LR for unwind */
+		EMIT(PPC_RAW_STL(_R3, _R1, bpf_frame_size + PPC_LR_STKOFF));
+
+	/* Save function arg count -- see bpf_get_func_arg_cnt() */
+	EMIT(PPC_RAW_LI(_R3, nr_regs));
+	EMIT(PPC_RAW_STL(_R3, _R1, nregs_off));
+
+	/* Save nv regs */
+	EMIT(PPC_RAW_STL(_R25, _R1, nvr_off));
+	EMIT(PPC_RAW_STL(_R26, _R1, nvr_off + SZL));
+
+	if (flags & BPF_TRAMP_F_CALL_ORIG) {
+		PPC_LI_ADDR(_R3, (unsigned long)im);
+		ret = bpf_jit_emit_func_call_rel(image, ro_image, ctx, (unsigned long)__bpf_tramp_enter);
+		if (ret)
+			return ret;
+	}
+
+	for (i = 0; i < fentry->nr_links; i++)
+		if (invoke_bpf_prog(image, ro_image, ctx, fentry->links[i], regs_off, retval_off,
+				    run_ctx_off, flags & BPF_TRAMP_F_RET_FENTRY_RET))
+			return -EINVAL;
+
+	if (fmod_ret->nr_links) {
+		branches = kcalloc(fmod_ret->nr_links, sizeof(u32), GFP_KERNEL);
+		if (!branches)
+			return -ENOMEM;
+
+		if (invoke_bpf_mod_ret(image, ro_image, ctx, fmod_ret, regs_off, retval_off,
+				       run_ctx_off, branches)) {
+			ret = -EINVAL;
+			goto cleanup;
+		}
+	}
+
+	/* Call the traced function */
+	if (flags & BPF_TRAMP_F_CALL_ORIG) {
+		/*
+		 * The address in LR save area points to the correct point in the original function
+		 * with both PPC_FTRACE_OUT_OF_LINE as well as with traditional ftrace instruction
+		 * sequence
+		 */
+		EMIT(PPC_RAW_LL(_R3, _R1, bpf_frame_size + PPC_LR_STKOFF));
+		EMIT(PPC_RAW_MTCTR(_R3));
+
+		/* Replicate tail_call_cnt before calling the original BPF prog */
+		if (flags & BPF_TRAMP_F_TAIL_CALL_CTX)
+			bpf_trampoline_setup_tail_call_cnt(image, ctx, func_frame_offset, r4_off);
+
+		/* Restore args */
+		bpf_trampoline_restore_args_stack(image, ctx, func_frame_offset, nr_regs, regs_off);
+
+		/* Restore TOC for 64-bit */
+		if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2) && !IS_ENABLED(CONFIG_PPC_KERNEL_PCREL))
+			EMIT(PPC_RAW_LD(_R2, _R1, 24));
+		EMIT(PPC_RAW_BCTRL());
+		if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2) && !IS_ENABLED(CONFIG_PPC_KERNEL_PCREL))
+			PPC64_LOAD_PACA();
+
+		/* Store return value for bpf prog to access */
+		EMIT(PPC_RAW_STL(_R3, _R1, retval_off));
+
+		/* Restore updated tail_call_cnt */
+		if (flags & BPF_TRAMP_F_TAIL_CALL_CTX)
+			bpf_trampoline_restore_tail_call_cnt(image, ctx, func_frame_offset, r4_off);
+
+		/* Reserve space to patch branch instruction to skip fexit progs */
+		im->ip_after_call = &((u32 *)ro_image)[ctx->idx];
+		EMIT(PPC_RAW_NOP());
+	}
+
+	/* Update branches saved in invoke_bpf_mod_ret with address of do_fexit */
+	for (i = 0; i < fmod_ret->nr_links && image; i++) {
+		if (create_cond_branch(&branch_insn, &image[branches[i]],
+				       (unsigned long)&image[ctx->idx], COND_NE << 16)) {
+			ret = -EINVAL;
+			goto cleanup;
+		}
+
+		image[branches[i]] = ppc_inst_val(branch_insn);
+	}
+
+	for (i = 0; i < fexit->nr_links; i++)
+		if (invoke_bpf_prog(image, ro_image, ctx, fexit->links[i], regs_off, retval_off,
+				    run_ctx_off, false)) {
+			ret = -EINVAL;
+			goto cleanup;
+		}
+
+	if (flags & BPF_TRAMP_F_CALL_ORIG) {
+		im->ip_epilogue = &((u32 *)ro_image)[ctx->idx];
+		PPC_LI_ADDR(_R3, im);
+		ret = bpf_jit_emit_func_call_rel(image, ro_image, ctx, (unsigned long)__bpf_tramp_exit);
+		if (ret)
+			goto cleanup;
+	}
+
+	if (flags & BPF_TRAMP_F_RESTORE_REGS)
+		bpf_trampoline_restore_args_regs(image, ctx, nr_regs, regs_off);
+
+	/* Restore return value of func_addr or fentry prog */
+	if (save_ret)
+		EMIT(PPC_RAW_LL(_R3, _R1, retval_off));
+
+	/* Restore nv regs */
+	EMIT(PPC_RAW_LL(_R26, _R1, nvr_off + SZL));
+	EMIT(PPC_RAW_LL(_R25, _R1, nvr_off));
+
+	/* Epilogue */
+	if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2) && !IS_ENABLED(CONFIG_PPC_KERNEL_PCREL))
+		EMIT(PPC_RAW_LD(_R2, _R1, 24));
+	if (flags & BPF_TRAMP_F_SKIP_FRAME) {
+		/* Skip the traced function and return to parent */
+		EMIT(PPC_RAW_ADDI(_R1, _R1, func_frame_offset));
+		EMIT(PPC_RAW_LL(_R0, _R1, PPC_LR_STKOFF));
+		EMIT(PPC_RAW_MTLR(_R0));
+		EMIT(PPC_RAW_BLR());
+	} else {
+		if (IS_ENABLED(CONFIG_PPC_FTRACE_OUT_OF_LINE)) {
+			EMIT(PPC_RAW_LL(_R0, _R1, alt_lr_off));
+			EMIT(PPC_RAW_MTLR(_R0));
+			EMIT(PPC_RAW_ADDI(_R1, _R1, func_frame_offset));
+			EMIT(PPC_RAW_LL(_R0, _R1, PPC_LR_STKOFF));
+			EMIT(PPC_RAW_BLR());
+		} else {
+			EMIT(PPC_RAW_LL(_R0, _R1, bpf_frame_size + PPC_LR_STKOFF));
+			EMIT(PPC_RAW_MTCTR(_R0));
+			EMIT(PPC_RAW_ADDI(_R1, _R1, func_frame_offset));
+			EMIT(PPC_RAW_LL(_R0, _R1, PPC_LR_STKOFF));
+			EMIT(PPC_RAW_MTLR(_R0));
+			EMIT(PPC_RAW_BCTR());
+		}
+	}
+
+	/* Make sure the trampoline generation logic doesn't overflow */
+	if (image && WARN_ON_ONCE(&image[ctx->idx] > (u32 *)rw_image_end - BPF_INSN_SAFETY)) {
+		ret = -EFAULT;
+		goto cleanup;
+	}
+	ret = ctx->idx * 4 + BPF_INSN_SAFETY * 4;
+
+cleanup:
+	kfree(branches);
+	return ret;
+}
+
+int arch_bpf_trampoline_size(const struct btf_func_model *m, u32 flags,
+			     struct bpf_tramp_links *tlinks, void *func_addr)
+{
+	struct bpf_tramp_image im;
+	void *image;
+	int ret;
+
+	/*
+	 * Allocate a temporary buffer for __arch_prepare_bpf_trampoline().
+	 * This will NOT cause fragmentation in direct map, as we do not
+	 * call set_memory_*() on this buffer.
+	 *
+	 * We cannot use kvmalloc here, because we need image to be in
+	 * module memory range.
+	 */
+	image = bpf_jit_alloc_exec(PAGE_SIZE);
+	if (!image)
+		return -ENOMEM;
+
+	ret = __arch_prepare_bpf_trampoline(&im, image, image + PAGE_SIZE, image,
+					    m, flags, tlinks, func_addr);
+	bpf_jit_free_exec(image);
+
+	return ret;
+}
+
+int arch_prepare_bpf_trampoline(struct bpf_tramp_image *im, void *image, void *image_end,
+				const struct btf_func_model *m, u32 flags,
+				struct bpf_tramp_links *tlinks,
+				void *func_addr)
+{
+	u32 size = image_end - image;
+	void *rw_image, *tmp;
+	int ret;
+
+	/*
+	 * rw_image doesn't need to be in module memory range, so we can
+	 * use kvmalloc.
+	 */
+	rw_image = kvmalloc(size, GFP_KERNEL);
+	if (!rw_image)
+		return -ENOMEM;
+
+	ret = __arch_prepare_bpf_trampoline(im, rw_image, rw_image + size, image, m,
+					    flags, tlinks, func_addr);
+	if (ret < 0)
+		goto out;
+
+	if (bpf_jit_enable > 1)
+		bpf_jit_dump(1, ret - BPF_INSN_SAFETY * 4, 1, rw_image);
+
+	tmp = bpf_arch_text_copy(image, rw_image, size);
+	if (IS_ERR(tmp))
+		ret = PTR_ERR(tmp);
+
+out:
+	kvfree(rw_image);
+	return ret;
+}
+
+static int bpf_modify_inst(void *ip, ppc_inst_t old_inst, ppc_inst_t new_inst)
+{
+	ppc_inst_t org_inst;
+
+	if (copy_inst_from_kernel_nofault(&org_inst, ip)) {
+		pr_err("0x%lx: fetching instruction failed\n", (unsigned long)ip);
+		return -EFAULT;
+	}
+
+	if (!ppc_inst_equal(org_inst, old_inst)) {
+		pr_err("0x%lx: expected (%08lx) != found (%08lx)\n",
+		       (unsigned long)ip, ppc_inst_as_ulong(old_inst), ppc_inst_as_ulong(org_inst));
+		return -EINVAL;
+	}
+
+	if (ppc_inst_equal(old_inst, new_inst))
+		return 0;
+
+	return patch_instruction(ip, new_inst);
+}
+
+static void do_isync(void *info __maybe_unused)
+{
+	isync();
+}
+
+/*
+ * A 3-step process for bpf prog entry:
+ * 1. At bpf prog entry, a single nop/b:
+ * bpf_func:
+ *	[nop|b]	ool_stub
+ * 2. Out-of-line stub:
+ * ool_stub:
+ *	mflr	r0
+ *	[b|bl]	<bpf_prog>/<long_branch_stub>
+ *	mtlr	r0 // CONFIG_PPC_FTRACE_OUT_OF_LINE only
+ *	b	bpf_func + 4
+ * 3. Long branch stub:
+ * long_branch_stub:
+ *	.long	<branch_addr>/<dummy_tramp>
+ *	mflr	r11
+ *	bcl	20,31,$+4
+ *	mflr	r12
+ *	ld	r12, -16(r12)
+ *	mtctr	r12
+ *	mtlr	r11 // needed to retain ftrace ABI
+ *	bctr
+ *
+ * dummy_tramp is used to reduce synchronization requirements.
+ *
+ * When attaching a bpf trampoline to a bpf prog, we do not need any
+ * synchronization here since we always have a valid branch target regardless
+ * of the order in which the above stores are seen. dummy_tramp ensures that
+ * the long_branch stub goes to a valid destination on other cpus, even when
+ * the branch to the long_branch stub is seen before the updated trampoline
+ * address.
+ *
+ * However, when detaching a bpf trampoline from a bpf prog, or if changing
+ * the bpf trampoline address, we need synchronization to ensure that other
+ * cpus can no longer branch into the older trampoline so that it can be
+ * safely freed. bpf_tramp_image_put() uses rcu_tasks to ensure all cpus
+ * make forward progress, but we still need to ensure that other cpus
+ * execute isync (or some CSI) so that they don't go back into the
+ * trampoline again.
+ */
+int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type poke_type,
+		       void *old_addr, void *new_addr)
+{
+	unsigned long bpf_func, bpf_func_end, size, offset;
+	ppc_inst_t old_inst, new_inst;
+	int ret = 0, branch_flags;
+	char name[KSYM_NAME_LEN];
+
+	if (IS_ENABLED(CONFIG_PPC32))
+		return -ENOTSUPP;
+
+	bpf_func = (unsigned long)ip;
+	branch_flags = poke_type == BPF_MOD_CALL ? BRANCH_SET_LINK : 0;
+
+	/* We currently only support poking bpf programs */
+	if (!__bpf_address_lookup(bpf_func, &size, &offset, name)) {
+		pr_err("%s (0x%lx): kernel/modules are not supported\n", __func__, bpf_func);
+		return -ENOTSUPP;
+	}
+
+	/*
+	 * If we are not poking at bpf prog entry, then we are simply patching in/out
+	 * an unconditional branch instruction at im->ip_after_call
+	 */
+	if (offset) {
+		if (poke_type != BPF_MOD_JUMP) {
+			pr_err("%s (0x%lx): calls are not supported in bpf prog body\n", __func__,
+			       bpf_func);
+			return -ENOTSUPP;
+		}
+		old_inst = ppc_inst(PPC_RAW_NOP());
+		if (old_addr)
+			if (create_branch(&old_inst, ip, (unsigned long)old_addr, 0))
+				return -ERANGE;
+		new_inst = ppc_inst(PPC_RAW_NOP());
+		if (new_addr)
+			if (create_branch(&new_inst, ip, (unsigned long)new_addr, 0))
+				return -ERANGE;
+		mutex_lock(&text_mutex);
+		ret = bpf_modify_inst(ip, old_inst, new_inst);
+		mutex_unlock(&text_mutex);
+
+		/* Make sure all cpus see the new instruction */
+		smp_call_function(do_isync, NULL, 1);
+		return ret;
+	}
+
+	bpf_func_end = bpf_func + size;
+
+	/* Address of the jmp/call instruction in the out-of-line stub */
+	ip = (void *)(bpf_func_end - bpf_jit_ool_stub + 4);
+
+	if (!is_offset_in_branch_range((long)ip - 4 - bpf_func)) {
+		pr_err("%s (0x%lx): bpf prog too large, ool stub out of branch range\n", __func__,
+		       bpf_func);
+		return -ERANGE;
+	}
+
+	old_inst = ppc_inst(PPC_RAW_NOP());
+	if (old_addr) {
+		if (is_offset_in_branch_range(ip - old_addr))
+			create_branch(&old_inst, ip, (unsigned long)old_addr, branch_flags);
+		else
+			create_branch(&old_inst, ip, bpf_func_end - bpf_jit_long_branch_stub,
+				      branch_flags);
+	}
+	new_inst = ppc_inst(PPC_RAW_NOP());
+	if (new_addr) {
+		if (is_offset_in_branch_range(ip - new_addr))
+			create_branch(&new_inst, ip, (unsigned long)new_addr, branch_flags);
+		else
+			create_branch(&new_inst, ip, bpf_func_end - bpf_jit_long_branch_stub,
+				      branch_flags);
+	}
+
+	mutex_lock(&text_mutex);
+
+	/*
+	 * 1. Update the address in the long branch stub:
+	 * If new_addr is out of range, we will have to use the long branch stub, so patch new_addr
+	 * here. Otherwise, revert to dummy_tramp, but only if we had patched old_addr here.
+	 */
+	if ((new_addr && !is_offset_in_branch_range(new_addr - ip)) ||
+	    (old_addr && !is_offset_in_branch_range(old_addr - ip)))
+		ret = patch_ulong((void *)(bpf_func_end - bpf_jit_long_branch_stub - SZL),
+				  (new_addr && !is_offset_in_branch_range(new_addr - ip)) ?
+				  (unsigned long)new_addr : (unsigned long)dummy_tramp);
+	if (ret)
+		goto out;
+
+	/* 2. Update the branch/call in the out-of-line stub */
+	ret = bpf_modify_inst(ip, old_inst, new_inst);
+	if (ret)
+		goto out;
+
+	/* 3. Update instruction at bpf prog entry */
+	ip = (void *)bpf_func;
+	if (!old_addr || !new_addr) {
+		if (!old_addr) {
+			old_inst = ppc_inst(PPC_RAW_NOP());
+			create_branch(&new_inst, ip, bpf_func_end - bpf_jit_ool_stub, 0);
+		} else {
+			new_inst = ppc_inst(PPC_RAW_NOP());
+			create_branch(&old_inst, ip, bpf_func_end - bpf_jit_ool_stub, 0);
+		}
+		ret = bpf_modify_inst(ip, old_inst, new_inst);
+	}
+
+out:
+	mutex_unlock(&text_mutex);
+
+	/*
+	 * Sync only if we are not attaching a trampoline to a bpf prog so the older
+	 * trampoline can be freed safely.
+	 */
+	if (old_addr)
+		smp_call_function(do_isync, NULL, 1);
+
+	return ret;
+}
diff --git a/arch/powerpc/net/bpf_jit_comp32.c b/arch/powerpc/net/bpf_jit_comp32.c
index a0c4f1bde83e..c4db278dae36 100644
--- a/arch/powerpc/net/bpf_jit_comp32.c
+++ b/arch/powerpc/net/bpf_jit_comp32.c
@@ -127,13 +127,16 @@  void bpf_jit_build_prologue(u32 *image, struct codegen_context *ctx)
 {
 	int i;
 
+	/* Instruction for trampoline attach */
+	EMIT(PPC_RAW_NOP());
+
 	/* Initialize tail_call_cnt, to be skipped if we do tail calls. */
 	if (ctx->seen & SEEN_TAILCALL)
 		EMIT(PPC_RAW_LI(_R4, 0));
 	else
 		EMIT(PPC_RAW_NOP());
 
-#define BPF_TAILCALL_PROLOGUE_SIZE	4
+#define BPF_TAILCALL_PROLOGUE_SIZE	8
 
 	if (bpf_has_stack_frame(ctx))
 		EMIT(PPC_RAW_STWU(_R1, _R1, -BPF_PPC_STACKFRAME(ctx)));
@@ -198,6 +201,8 @@  void bpf_jit_build_epilogue(u32 *image, struct codegen_context *ctx)
 	bpf_jit_emit_common_epilogue(image, ctx);
 
 	EMIT(PPC_RAW_BLR());
+
+	bpf_jit_build_fentry_stubs(image, ctx);
 }
 
 /* Relative offset needs to be calculated based on final image location */
diff --git a/arch/powerpc/net/bpf_jit_comp64.c b/arch/powerpc/net/bpf_jit_comp64.c
index f3be024fc685..dcf339788e58 100644
--- a/arch/powerpc/net/bpf_jit_comp64.c
+++ b/arch/powerpc/net/bpf_jit_comp64.c
@@ -126,6 +126,9 @@  void bpf_jit_build_prologue(u32 *image, struct codegen_context *ctx)
 {
 	int i;
 
+	/* Instruction for trampoline attach */
+	EMIT(PPC_RAW_NOP());
+
 #ifndef CONFIG_PPC_KERNEL_PCREL
 	if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2))
 		EMIT(PPC_RAW_LD(_R2, _R13, offsetof(struct paca_struct, kernel_toc)));
@@ -200,6 +203,8 @@  void bpf_jit_build_epilogue(u32 *image, struct codegen_context *ctx)
 	EMIT(PPC_RAW_MR(_R3, bpf_to_ppc(BPF_REG_0)));
 
 	EMIT(PPC_RAW_BLR());
+
+	bpf_jit_build_fentry_stubs(image, ctx);
 }
 
 int bpf_jit_emit_func_call_rel(u32 *image, u32 *fimage, struct codegen_context *ctx, u64 func)
@@ -303,7 +308,7 @@  static int bpf_jit_emit_tail_call(u32 *image, struct codegen_context *ctx, u32 o
 	 */
 	int b2p_bpf_array = bpf_to_ppc(BPF_REG_2);
 	int b2p_index = bpf_to_ppc(BPF_REG_3);
-	int bpf_tailcall_prologue_size = 8;
+	int bpf_tailcall_prologue_size = 12;
 
 	if (!IS_ENABLED(CONFIG_PPC_KERNEL_PCREL) && IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2))
 		bpf_tailcall_prologue_size += 4; /* skip past the toc load */