Commit 6c21a2a6 authored by David S. Miller's avatar David S. Miller
Browse files

Merge branch 'bpf-stack-tracker' 



Alexei Starovoitov says:

====================
bpf: stack depth tracking

Introduce tracking of bpf program stack depth in the verifier and use that
info to reduce bpf program stack consumption in the interpreter and x64 JIT.
Other JITs can take advantage of it as well in the future.
Most of the programs consume very little stack, so it's good optimization
in general and it's the first step toward bpf to bpf function calls.

Also use internal opcode for bpf_tail_call() marking to make clear
that jmp|call|x opcode is not uapi and may be used for actual
indirect call opcode in the future.
====================

Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parents d2e0ef49 2960ae48

arch/arm64/net/bpf_jit_comp.c

+1 −1
Original line number Diff line number Diff line
@@ -586,7 +586,7 @@ emit_cond_jmp: 
		break;

	}

	/* tail call */

	case BPF_JMP | BPF_CALL | BPF_X:

	case BPF_JMP | BPF_TAIL_CALL:

		if (emit_bpf_tail_call(ctx))

			return -EFAULT;

		break;

arch/powerpc/net/bpf_jit_comp64.c

+1 −1
Original line number Diff line number Diff line
@@ -938,7 +938,7 @@ common_load: 
		/*

		 * Tail call

		 */

		case BPF_JMP | BPF_CALL | BPF_X:

		case BPF_JMP | BPF_TAIL_CALL:

			ctx->seen |= SEEN_TAILCALL;

			bpf_jit_emit_tail_call(image, ctx, addrs[i + 1]);

			break;

arch/s390/net/bpf_jit_comp.c

+1 −1
Original line number Diff line number Diff line
@@ -991,7 +991,7 @@ static noinline int bpf_jit_insn(struct bpf_jit *jit, struct bpf_prog *fp, int i 
		}

		break;

	}

	case BPF_JMP | BPF_CALL | BPF_X:

	case BPF_JMP | BPF_TAIL_CALL:

		/*

		 * Implicit input:

		 *  B1: pointer to ctx

arch/sparc/net/bpf_jit_comp_64.c

+1 −1
Original line number Diff line number Diff line
@@ -1217,7 +1217,7 @@ static int build_insn(const struct bpf_insn *insn, struct jit_ctx *ctx) 
	}



	/* tail call */

	case BPF_JMP | BPF_CALL |BPF_X:

	case BPF_JMP | BPF_TAIL_CALL:

		emit_tail_call(ctx);

		break;

arch/x86/net/bpf_jit.S

+9 −11
Original line number Diff line number Diff line
@@ -19,9 +19,6 @@ 
 */

#define SKBDATA	%r10

#define SKF_MAX_NEG_OFF    $(-0x200000) /* SKF_LL_OFF from filter.h */

#define MAX_BPF_STACK (512 /* from filter.h */ + \

	32 /* space for rbx,r13,r14,r15 */ + \

	8 /* space for skb_copy_bits */)



#define FUNC(name) \

	.globl name; \
@@ -66,7 +63,7 @@ FUNC(sk_load_byte_positive_offset) 


/* rsi contains offset and can be scratched */

#define bpf_slow_path_common(LEN)		\

	lea	-MAX_BPF_STACK + 32(%rbp), %rdx;\

	lea	32(%rbp), %rdx;\

	FRAME_BEGIN;				\

	mov	%rbx, %rdi; /* arg1 == skb */	\

	push	%r9;				\
@@ -83,14 +80,14 @@ FUNC(sk_load_byte_positive_offset) 
bpf_slow_path_word:

	bpf_slow_path_common(4)

	js	bpf_error

	mov	- MAX_BPF_STACK + 32(%rbp),%eax

	mov	32(%rbp),%eax

	bswap	%eax

	ret



bpf_slow_path_half:

	bpf_slow_path_common(2)

	js	bpf_error

	mov	- MAX_BPF_STACK + 32(%rbp),%ax

	mov	32(%rbp),%ax

	rol	$8,%ax

	movzwl	%ax,%eax

	ret
@@ -98,7 +95,7 @@ bpf_slow_path_half: 
bpf_slow_path_byte:

	bpf_slow_path_common(1)

	js	bpf_error

	movzbl	- MAX_BPF_STACK + 32(%rbp),%eax

	movzbl	32(%rbp),%eax

	ret



#define sk_negative_common(SIZE)				\
@@ -148,9 +145,10 @@ FUNC(sk_load_byte_negative_offset) 
bpf_error:

# force a return 0 from jit handler

	xor	%eax,%eax

	mov	- MAX_BPF_STACK(%rbp),%rbx

	mov	- MAX_BPF_STACK + 8(%rbp),%r13

	mov	- MAX_BPF_STACK + 16(%rbp),%r14

	mov	- MAX_BPF_STACK + 24(%rbp),%r15

	mov	(%rbp),%rbx

	mov	8(%rbp),%r13

	mov	16(%rbp),%r14

	mov	24(%rbp),%r15

	add	$40, %rbp

	leaveq

	ret

CRA Git | Maintained and supported by SUSTech CRA and CCSE