iv/linux
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

bpf: change x86 JITed program stack layout

Browse Source 
in order to JIT programs with different stack sizes we need to
make epilogue and exception path to be stack size independent,
hence move auxiliary stack space from the bottom of the stack
to the top of the stack.
Nice side effect is that JITed function prologue becomes shorter
due to imm8 offset encoding vs imm32.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit is contained in:
Alexei Starovoitov 2017-05-30 13:31:34 -07:00 committed by David S. Miller
parent b870aa901f
commit 177366bf7c
2 changed files with 39 additions and 37 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
arch/x86/net/bpf_jit.S
View File

@ -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

56
arch/x86/net/bpf_jit_comp.c
View File

@ -197,12 +197,11 @@ struct jit_context {
#define BPF_MAX_INSN_SIZE 128
#define BPF_INSN_SAFETY 64
#define STACKSIZE \
(MAX_BPF_STACK + \
32 /* space for rbx, r13, r14, r15 */ + \
#define AUX_STACK_SPACE \
(32 /* space for rbx, r13, r14, r15 */ + \
8 /* space for skb_copy_bits() buffer */)
#define PROLOGUE_SIZE 48
#define PROLOGUE_SIZE 37
/* emit x64 prologue code for BPF program and check it's size.
* bpf_tail_call helper will skip it while jumping into another program
@ -215,13 +214,16 @@ static void emit_prologue(u8 **pprog)
EMIT1(0x55); /* push rbp */
EMIT3(0x48, 0x89, 0xE5); /* mov rbp,rsp */
/* sub rsp, STACKSIZE */
EMIT3_off32(0x48, 0x81, 0xEC, STACKSIZE);
/* sub rsp, MAX_BPF_STACK + AUX_STACK_SPACE */
EMIT3_off32(0x48, 0x81, 0xEC, MAX_BPF_STACK + AUX_STACK_SPACE);
/* sub rbp, AUX_STACK_SPACE */
EMIT4(0x48, 0x83, 0xED, AUX_STACK_SPACE);
/* all classic BPF filters use R6(rbx) save it */
/* mov qword ptr [rbp-X],rbx */
EMIT3_off32(0x48, 0x89, 0x9D, -STACKSIZE);
/* mov qword ptr [rbp+0],rbx */
EMIT4(0x48, 0x89, 0x5D, 0);
/* bpf_convert_filter() maps classic BPF register X to R7 and uses R8
* as temporary, so all tcpdump filters need to spill/fill R7(r13) and
@ -231,12 +233,12 @@ static void emit_prologue(u8 **pprog)
* than synthetic ones. Therefore not worth adding complexity.
*/
/* mov qword ptr [rbp-X],r13 */
EMIT3_off32(0x4C, 0x89, 0xAD, -STACKSIZE + 8);
/* mov qword ptr [rbp-X],r14 */
EMIT3_off32(0x4C, 0x89, 0xB5, -STACKSIZE + 16);
/* mov qword ptr [rbp-X],r15 */
EMIT3_off32(0x4C, 0x89, 0xBD, -STACKSIZE + 24);
/* mov qword ptr [rbp+8],r13 */
EMIT4(0x4C, 0x89, 0x6D, 8);
/* mov qword ptr [rbp+16],r14 */
EMIT4(0x4C, 0x89, 0x75, 16);
/* mov qword ptr [rbp+24],r15 */
EMIT4(0x4C, 0x89, 0x7D, 24);
/* Clear the tail call counter (tail_call_cnt): for eBPF tail calls
* we need to reset the counter to 0. It's done in two instructions,
@ -246,8 +248,8 @@ static void emit_prologue(u8 **pprog)
/* xor eax, eax */
EMIT2(0x31, 0xc0);
/* mov qword ptr [rbp-X], rax */
EMIT3_off32(0x48, 0x89, 0x85, -STACKSIZE + 32);
/* mov qword ptr [rbp+32], rax */
EMIT4(0x48, 0x89, 0x45, 32);
BUILD_BUG_ON(cnt != PROLOGUE_SIZE);
*pprog = prog;
@ -289,13 +291,13 @@ static void emit_bpf_tail_call(u8 **pprog)
/* if (tail_call_cnt > MAX_TAIL_CALL_CNT)
* goto out;
*/
EMIT2_off32(0x8B, 0x85, -STACKSIZE + 36); /* mov eax, dword ptr [rbp - 516] */
EMIT2_off32(0x8B, 0x85, 36); /* mov eax, dword ptr [rbp + 36] */
EMIT3(0x83, 0xF8, MAX_TAIL_CALL_CNT); /* cmp eax, MAX_TAIL_CALL_CNT */
#define OFFSET2 36
EMIT2(X86_JA, OFFSET2); /* ja out */
label2 = cnt;
EMIT3(0x83, 0xC0, 0x01); /* add eax, 1 */
EMIT2_off32(0x89, 0x85, -STACKSIZE + 36); /* mov dword ptr [rbp - 516], eax */
EMIT2_off32(0x89, 0x85, 36); /* mov dword ptr [rbp + 36], eax */
/* prog = array->ptrs[index]; */
EMIT4_off32(0x48, 0x8D, 0x84, 0xD6, /* lea rax, [rsi + rdx * 8 + offsetof(...)] */
@ -1036,15 +1038,17 @@ common_load:
seen_exit = true;
/* update cleanup_addr */
ctx->cleanup_addr = proglen;
/* mov rbx, qword ptr [rbp-X] */
EMIT3_off32(0x48, 0x8B, 0x9D, -STACKSIZE);
/* mov r13, qword ptr [rbp-X] */
EMIT3_off32(0x4C, 0x8B, 0xAD, -STACKSIZE + 8);
/* mov r14, qword ptr [rbp-X] */
EMIT3_off32(0x4C, 0x8B, 0xB5, -STACKSIZE + 16);
/* mov r15, qword ptr [rbp-X] */
EMIT3_off32(0x4C, 0x8B, 0xBD, -STACKSIZE + 24);
/* mov rbx, qword ptr [rbp+0] */
EMIT4(0x48, 0x8B, 0x5D, 0);
/* mov r13, qword ptr [rbp+8] */
EMIT4(0x4C, 0x8B, 0x6D, 8);
/* mov r14, qword ptr [rbp+16] */
EMIT4(0x4C, 0x8B, 0x75, 16);
/* mov r15, qword ptr [rbp+24] */
EMIT4(0x4C, 0x8B, 0x7D, 24);
/* add rbp, AUX_STACK_SPACE */
EMIT4(0x48, 0x83, 0xC5, AUX_STACK_SPACE);
EMIT1(0xC9); /* leave */
EMIT1(0xC3); /* ret */
break;