Merge branch 's390-bpf-describe-the-frame-using-a-struct-instead-of-constants'

Ilya Leoshkevich says:

====================
s390/bpf: Describe the frame using a struct instead of constants

Hi,

This series contains two small refactorings without functional changes.

The first one removes the code duplication around calculating the
distance from %r15 to the stack frame.

The second one simplifies how offsets to various values stored inside
the frame are calculated.

Best regards,
Ilya
====================

Link: https://patch.msgid.link/[e-mail]
Signed-off-by: Alexei Starovoitov <[e-mail]>

arch/s390/net/bpf_jit.h

···
-/* SPDX-License-Identifier: GPL-2.0 */
-/*
- * BPF Jit compiler defines
- *
- * Copyright IBM Corp. 2012,2015
- *
- * Author(s): Martin Schwidefsky <[e-mail]>
- *	      Michael Holzheu <[e-mail]>
- */
-
-#ifndef __ARCH_S390_NET_BPF_JIT_H
-#define __ARCH_S390_NET_BPF_JIT_H
-
-#ifndef __ASSEMBLY__
-
-#include <linux/filter.h>
-#include <linux/types.h>
-
-#endif /* __ASSEMBLY__ */
-
-/*
- * Stackframe layout (packed stack):
- *
- *				    ^ high
- *	      +---------------+     |
- *	      | old backchain |     |
- *	      +---------------+     |
- *	      |   r15 - r6    |     |
- *	      +---------------+     |
- *	      | 4 byte align  |     |
- *	      | tail_call_cnt |     |
- * BFP	   -> +===============+     |
- *	      |		      |     |
- *	      |   BPF stack   |     |
- *	      |		      |     |
- * R15+160 -> +---------------+     |
- *	      | new backchain |     |
- * R15+152 -> +---------------+     |
- *	      | + 152 byte SA |     |
- * R15	   -> +---------------+     + low
- *
- * We get 160 bytes stack space from calling function, but only use
- * 12 * 8 byte for old backchain, r15..r6, and tail_call_cnt.
- *
- * The stack size used by the BPF program ("BPF stack" above) is passed
- * via "aux->stack_depth".
- */
-#define STK_SPACE_ADD	(160)
-#define STK_160_UNUSED	(160 - 12 * 8)
-#define STK_OFF		(STK_SPACE_ADD - STK_160_UNUSED)
-
-#define STK_OFF_R6	(160 - 11 * 8)	/* Offset of r6 on stack */
-#define STK_OFF_TCCNT	(160 - 12 * 8)	/* Offset of tail_call_cnt on stack */
-
-#endif /* __ARCH_S390_NET_BPF_JIT_H */

arch/s390/net/bpf_jit_comp.c

···
 #include <asm/set_memory.h>
 #include <asm/text-patching.h>
 #include <asm/unwind.h>
-#include "bpf_jit.h"
 
 struct bpf_jit {
 	u32 seen;		/* Flags to remember seen eBPF instructions */
···
 	int prologue_plt;	/* Start of prologue hotpatch PLT */
 	int kern_arena;		/* Pool offset of kernel arena address */
 	u64 user_arena;		/* User arena address */
+	u32 frame_off;		/* Offset of struct bpf_prog from %r15 */
 };
 
 #define SEEN_MEM	BIT(0)		/* use mem[] for temporary storage */
···
 }
 
 /*
+ * Caller-allocated part of the frame.
+ * Thanks to packed stack, its otherwise unused initial part can be used for
+ * the BPF stack and for the next frame.
+ */
+struct prog_frame {
+	u64 unused[8];
+	/* BPF stack starts here and grows towards 0 */
+	u32 tail_call_cnt;
+	u32 pad;
+	u64 r6[10];  /* r6 - r15 */
+	u64 backchain;
+} __packed;
+
+/*
  * Save registers from "rs" (register start) to "re" (register end) on stack
  */
 static void save_regs(struct bpf_jit *jit, u32 rs, u32 re)
 {
-	u32 off = STK_OFF_R6 + (rs - 6) * 8;
+	u32 off = offsetof(struct prog_frame, r6) + (rs - 6) * 8;
 
 	if (rs == re)
 		/* stg %rs,off(%r15) */
···
 /*
  * Restore registers from "rs" (register start) to "re" (register end) on stack
  */
-static void restore_regs(struct bpf_jit *jit, u32 rs, u32 re, u32 stack_depth)
+static void restore_regs(struct bpf_jit *jit, u32 rs, u32 re)
 {
-	u32 off = STK_OFF_R6 + (rs - 6) * 8;
-
-	if (jit->seen & SEEN_STACK)
-		off += STK_OFF + stack_depth;
+	u32 off = jit->frame_off + offsetof(struct prog_frame, r6) + (rs - 6) * 8;
 
 	if (rs == re)
 		/* lg %rs,off(%r15) */
···
  * Save and restore clobbered registers (6-15) on stack.
  * We save/restore registers in chunks with gap >= 2 registers.
  */
-static void save_restore_regs(struct bpf_jit *jit, int op, u32 stack_depth,
-			      u16 extra_regs)
+static void save_restore_regs(struct bpf_jit *jit, int op, u16 extra_regs)
 {
 	u16 seen_regs = jit->seen_regs | extra_regs;
 	const int last = 15, save_restore_size = 6;
···
 		if (op == REGS_SAVE)
 			save_regs(jit, rs, re);
 		else
-			restore_regs(jit, rs, re, stack_depth);
+			restore_regs(jit, rs, re);
 		re++;
 	} while (re <= last);
 }
···
  * Emit function prologue
  *
  * Save registers and create stack frame if necessary.
- * See stack frame layout description in "bpf_jit.h"!
+ * Stack frame layout is described by struct prog_frame.
  */
-static void bpf_jit_prologue(struct bpf_jit *jit, struct bpf_prog *fp,
-			     u32 stack_depth)
+static void bpf_jit_prologue(struct bpf_jit *jit, struct bpf_prog *fp)
 {
+	BUILD_BUG_ON(sizeof(struct prog_frame) != STACK_FRAME_OVERHEAD);
+
 	/* No-op for hotpatching */
 	/* brcl 0,prologue_plt */
 	EMIT6_PCREL_RILC(0xc0040000, 0, jit->prologue_plt);
···
 
 	if (!bpf_is_subprog(fp)) {
 		/* Initialize the tail call counter in the main program. */
-		/* xc STK_OFF_TCCNT(4,%r15),STK_OFF_TCCNT(%r15) */
-		_EMIT6(0xd703f000 | STK_OFF_TCCNT, 0xf000 | STK_OFF_TCCNT);
+		/* xc tail_call_cnt(4,%r15),tail_call_cnt(%r15) */
+		_EMIT6(0xd703f000 | offsetof(struct prog_frame, tail_call_cnt),
+		       0xf000 | offsetof(struct prog_frame, tail_call_cnt));
 	} else {
 		/*
 		 * Skip the tail call counter initialization in subprograms.
···
 		jit->seen_regs |= NVREGS;
 	} else {
 		/* Save registers */
-		save_restore_regs(jit, REGS_SAVE, stack_depth,
+		save_restore_regs(jit, REGS_SAVE,
 				  fp->aux->exception_boundary ? NVREGS : 0);
 	}
 	/* Setup literal pool */
···
 	if (is_first_pass(jit) || (jit->seen & SEEN_STACK)) {
 		/* lgr %w1,%r15 (backchain) */
 		EMIT4(0xb9040000, REG_W1, REG_15);
-		/* la %bfp,STK_160_UNUSED(%r15) (BPF frame pointer) */
-		EMIT4_DISP(0x41000000, BPF_REG_FP, REG_15, STK_160_UNUSED);
-		/* aghi %r15,-STK_OFF */
-		EMIT4_IMM(0xa70b0000, REG_15, -(STK_OFF + stack_depth));
-		/* stg %w1,152(%r15) (backchain) */
+		/* la %bfp,unused_end(%r15) (BPF frame pointer) */
+		EMIT4_DISP(0x41000000, BPF_REG_FP, REG_15,
+			   offsetofend(struct prog_frame, unused));
+		/* aghi %r15,-frame_off */
+		EMIT4_IMM(0xa70b0000, REG_15, -jit->frame_off);
+		/* stg %w1,backchain(%r15) */
 		EMIT6_DISP_LH(0xe3000000, 0x0024, REG_W1, REG_0,
-			      REG_15, 152);
+			      REG_15,
+			      offsetof(struct prog_frame, backchain));
 	}
 }
 
···
 /*
  * Function epilogue
  */
-static void bpf_jit_epilogue(struct bpf_jit *jit, u32 stack_depth)
+static void bpf_jit_epilogue(struct bpf_jit *jit)
 {
 	jit->exit_ip = jit->prg;
 	/* Load exit code: lgr %r2,%b0 */
 	EMIT4(0xb9040000, REG_2, BPF_REG_0);
 	/* Restore registers */
-	save_restore_regs(jit, REGS_RESTORE, stack_depth, 0);
+	save_restore_regs(jit, REGS_RESTORE, 0);
 	EMIT_JUMP_REG(14);
 
 	jit->prg = ALIGN(jit->prg, 8);
···
  * stack space for the large switch statement.
  */
 static noinline int bpf_jit_insn(struct bpf_jit *jit, struct bpf_prog *fp,
-				 int i, bool extra_pass, u32 stack_depth)
+				 int i, bool extra_pass)
 {
 	struct bpf_insn *insn = &fp->insnsi[i];
 	s32 branch_oc_off = insn->off;
···
 		 * Note 2: We assume that the verifier does not let us call the
 		 * main program, which clears the tail call counter on entry.
 		 */
-		/* mvc STK_OFF_TCCNT(4,%r15),N(%r15) */
-		_EMIT6(0xd203f000 | STK_OFF_TCCNT,
-		       0xf000 | (STK_OFF_TCCNT + STK_OFF + stack_depth));
+		/* mvc tail_call_cnt(4,%r15),frame_off+tail_call_cnt(%r15) */
+		_EMIT6(0xd203f000 | offsetof(struct prog_frame, tail_call_cnt),
+		       0xf000 | (jit->frame_off +
+				 offsetof(struct prog_frame, tail_call_cnt)));
 
 		/* Sign-extend the kfunc arguments. */
 		if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) {
···
 		 *         goto out;
 		 */
 
-		if (jit->seen & SEEN_STACK)
-			off = STK_OFF_TCCNT + STK_OFF + stack_depth;
-		else
-			off = STK_OFF_TCCNT;
+		off = jit->frame_off +
+		      offsetof(struct prog_frame, tail_call_cnt);
 		/* lhi %w0,1 */
 		EMIT4_IMM(0xa7080000, REG_W0, 1);
 		/* laal %w1,%w0,off(%r15) */
···
 		/*
 		 * Restore registers before calling function
 		 */
-		save_restore_regs(jit, REGS_RESTORE, stack_depth, 0);
+		save_restore_regs(jit, REGS_RESTORE, 0);
 
 		/*
 		 * goto *(prog->bpf_func + tail_call_start);
···
  * Compile eBPF program into s390x code
  */
 static int bpf_jit_prog(struct bpf_jit *jit, struct bpf_prog *fp,
-			bool extra_pass, u32 stack_depth)
+			bool extra_pass)
 {
 	int i, insn_count, lit32_size, lit64_size;
 	u64 kern_arena;
···
 	jit->lit64 = jit->lit64_start;
 	jit->prg = 0;
 	jit->excnt = 0;
+	if (is_first_pass(jit) || (jit->seen & SEEN_STACK))
+		jit->frame_off = sizeof(struct prog_frame) -
+				 offsetofend(struct prog_frame, unused) +
+				 round_up(fp->aux->stack_depth, 8);
+	else
+		jit->frame_off = 0;
 
 	kern_arena = bpf_arena_get_kern_vm_start(fp->aux->arena);
 	if (kern_arena)
 		jit->kern_arena = _EMIT_CONST_U64(kern_arena);
 	jit->user_arena = bpf_arena_get_user_vm_start(fp->aux->arena);
 
-	bpf_jit_prologue(jit, fp, stack_depth);
+	bpf_jit_prologue(jit, fp);
 	if (bpf_set_addr(jit, 0) < 0)
 		return -1;
 	for (i = 0; i < fp->len; i += insn_count) {
-		insn_count = bpf_jit_insn(jit, fp, i, extra_pass, stack_depth);
+		insn_count = bpf_jit_insn(jit, fp, i, extra_pass);
 		if (insn_count < 0)
 			return -1;
 		/* Next instruction address */
 		if (bpf_set_addr(jit, i + insn_count) < 0)
 			return -1;
 	}
-	bpf_jit_epilogue(jit, stack_depth);
+	bpf_jit_epilogue(jit);
 
 	lit32_size = jit->lit32 - jit->lit32_start;
 	lit64_size = jit->lit64 - jit->lit64_start;
···
  */
 struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
 {
-	u32 stack_depth = round_up(fp->aux->stack_depth, 8);
 	struct bpf_prog *tmp, *orig_fp = fp;
 	struct bpf_binary_header *header;
 	struct s390_jit_data *jit_data;
···
 	 *   - 3:   Calculate program size and addrs array
 	 */
 	for (pass = 1; pass <= 3; pass++) {
-		if (bpf_jit_prog(&jit, fp, extra_pass, stack_depth)) {
+		if (bpf_jit_prog(&jit, fp, extra_pass)) {
 			fp = orig_fp;
 			goto free_addrs;
 		}
···
 		goto free_addrs;
 	}
 skip_init_ctx:
-	if (bpf_jit_prog(&jit, fp, extra_pass, stack_depth)) {
+	if (bpf_jit_prog(&jit, fp, extra_pass)) {
 		bpf_jit_binary_free(header);
 		fp = orig_fp;
 		goto free_addrs;
···
 	/* stg %r1,backchain_off(%r15) */
 	EMIT6_DISP_LH(0xe3000000, 0x0024, REG_1, REG_0, REG_15,
 		      tjit->backchain_off);
-	/* mvc tccnt_off(4,%r15),stack_size+STK_OFF_TCCNT(%r15) */
+	/* mvc tccnt_off(4,%r15),stack_size+tail_call_cnt(%r15) */
 	_EMIT6(0xd203f000 | tjit->tccnt_off,
-	       0xf000 | (tjit->stack_size + STK_OFF_TCCNT));
+	       0xf000 | (tjit->stack_size +
+			 offsetof(struct prog_frame, tail_call_cnt)));
 	/* stmg %r2,%rN,fwd_reg_args_off(%r15) */
 	if (nr_reg_args)
 		EMIT6_DISP_LH(0xeb000000, 0x0024, REG_2,
···
 				       (nr_stack_args * sizeof(u64) - 1) << 16 |
 				       tjit->stack_args_off,
 			       0xf000 | tjit->orig_stack_args_off);
-		/* mvc STK_OFF_TCCNT(4,%r15),tccnt_off(%r15) */
-		_EMIT6(0xd203f000 | STK_OFF_TCCNT, 0xf000 | tjit->tccnt_off);
+		/* mvc tail_call_cnt(4,%r15),tccnt_off(%r15) */
+		_EMIT6(0xd203f000 | offsetof(struct prog_frame, tail_call_cnt),
+		       0xf000 | tjit->tccnt_off);
 		/* lgr %r1,%r8 */
 		EMIT4(0xb9040000, REG_1, REG_8);
 		/* %r1() */
···
 	if (flags & (BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_RET_FENTRY_RET))
 		EMIT6_DISP_LH(0xe3000000, 0x0004, REG_2, REG_0, REG_15,
 			      tjit->retval_off);
-	/* mvc stack_size+STK_OFF_TCCNT(4,%r15),tccnt_off(%r15) */
-	_EMIT6(0xd203f000 | (tjit->stack_size + STK_OFF_TCCNT),
+	/* mvc stack_size+tail_call_cnt(4,%r15),tccnt_off(%r15) */
+	_EMIT6(0xd203f000 | (tjit->stack_size +
+			     offsetof(struct prog_frame, tail_call_cnt)),
 	       0xf000 | tjit->tccnt_off);
 	/* aghi %r15,stack_size */
 	EMIT4_IMM(0xa70b0000, REG_15, tjit->stack_size);