| Message ID | 1631158350-3661-1-git-send-email-yangtiezhu@loongson.cn (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [bpf-next] bpf: Change value of MAX_TAIL_CALL_CNT from 32 to 33 | expand |
On Wed, Sep 8, 2021 at 8:33 PM Tiezhu Yang <yangtiezhu@loongson.cn> wrote: > > In the current code, the actual max tail call count is 33 which is greater > than MAX_TAIL_CALL_CNT (defined as 32), the actual limit is not consistent > with the meaning of MAX_TAIL_CALL_CNT, there is some confusion and need to > spend some time to think the reason at the first glance. think *about* the reason > > We can see the historical evolution from commit 04fd61ab36ec ("bpf: allow > bpf programs to tail-call other bpf programs") and commit f9dabe016b63 > ("bpf: Undo off-by-one in interpreter tail call count limit"). > > In order to avoid changing existing behavior, the actual limit is 33 now, > this is resonable. typo: reasonable > > After commit 874be05f525e ("bpf, tests: Add tail call test suite"), we can > see there exists failed testcase. > > On all archs when CONFIG_BPF_JIT_ALWAYS_ON is not set: > # echo 0 > /proc/sys/net/core/bpf_jit_enable > # modprobe test_bpf > # dmesg | grep -w FAIL > Tail call error path, max count reached jited:0 ret 34 != 33 FAIL > > On some archs: > # echo 1 > /proc/sys/net/core/bpf_jit_enable > # modprobe test_bpf > # dmesg | grep -w FAIL > Tail call error path, max count reached jited:1 ret 34 != 33 FAIL > > So it is necessary to change the value of MAX_TAIL_CALL_CNT from 32 to 33, > then do some small changes of the related code. > > With this patch, it does not change the current limit, MAX_TAIL_CALL_CNT > can reflect the actual max tail call count, and the above failed testcase > can be fixed. > > Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn> > --- This change breaks selftests ([0]), please fix them at the same time as you are changing the kernel behavior: test_tailcall_2:PASS:tailcall 128 nsec test_tailcall_2:PASS:tailcall 128 nsec test_tailcall_2:FAIL:tailcall err 0 errno 2 retval 4 #135/2 tailcalls/tailcall_2:FAIL test_tailcall_3:PASS:tailcall 128 nsec test_tailcall_3:FAIL:tailcall count err 0 errno 2 count 34 test_tailcall_3:PASS:tailcall 128 nsec #135/3 tailcalls/tailcall_3:FAIL #135/4 tailcalls/tailcall_4:OK #135/5 tailcalls/tailcall_5:OK #135/6 tailcalls/tailcall_bpf2bpf_1:OK test_tailcall_bpf2bpf_2:PASS:tailcall 128 nsec test_tailcall_bpf2bpf_2:FAIL:tailcall count err 0 errno 2 count 34 test_tailcall_bpf2bpf_2:PASS:tailcall 128 nsec #135/7 tailcalls/tailcall_bpf2bpf_2:FAIL #135/8 tailcalls/tailcall_bpf2bpf_3:OK test_tailcall_bpf2bpf_4:PASS:tailcall 54 nsec test_tailcall_bpf2bpf_4:FAIL:tailcall count err 0 errno 2 count 32 #135/9 tailcalls/tailcall_bpf2bpf_4:FAIL test_tailcall_bpf2bpf_4:PASS:tailcall 54 nsec test_tailcall_bpf2bpf_4:FAIL:tailcall count err 0 errno 2 count 32 #135/10 tailcalls/tailcall_bpf2bpf_5:FAIL #135 tailcalls:FAIL [0] https://github.com/kernel-patches/bpf/pull/1747/checks?check_run_id=3552002906 > arch/arm/net/bpf_jit_32.c | 11 ++++++----- > arch/arm64/net/bpf_jit_comp.c | 7 ++++--- > arch/mips/net/ebpf_jit.c | 4 ++-- > arch/powerpc/net/bpf_jit_comp32.c | 4 ++-- > arch/powerpc/net/bpf_jit_comp64.c | 12 ++++++------ > arch/riscv/net/bpf_jit_comp32.c | 4 ++-- > arch/riscv/net/bpf_jit_comp64.c | 4 ++-- > arch/sparc/net/bpf_jit_comp_64.c | 8 ++++---- > include/linux/bpf.h | 2 +- > kernel/bpf/core.c | 4 ++-- > 10 files changed, 31 insertions(+), 29 deletions(-) > [...]
On 9/9/21 7:50 AM, Andrii Nakryiko wrote: > On Wed, Sep 8, 2021 at 8:33 PM Tiezhu Yang <yangtiezhu@loongson.cn> wrote: >> >> In the current code, the actual max tail call count is 33 which is greater >> than MAX_TAIL_CALL_CNT (defined as 32), the actual limit is not consistent >> with the meaning of MAX_TAIL_CALL_CNT, there is some confusion and need to >> spend some time to think the reason at the first glance. > > think *about* the reason > >> We can see the historical evolution from commit 04fd61ab36ec ("bpf: allow >> bpf programs to tail-call other bpf programs") and commit f9dabe016b63 >> ("bpf: Undo off-by-one in interpreter tail call count limit"). >> >> In order to avoid changing existing behavior, the actual limit is 33 now, >> this is resonable. > > typo: reasonable > >> After commit 874be05f525e ("bpf, tests: Add tail call test suite"), we can >> see there exists failed testcase. >> >> On all archs when CONFIG_BPF_JIT_ALWAYS_ON is not set: >> # echo 0 > /proc/sys/net/core/bpf_jit_enable >> # modprobe test_bpf >> # dmesg | grep -w FAIL >> Tail call error path, max count reached jited:0 ret 34 != 33 FAIL >> >> On some archs: >> # echo 1 > /proc/sys/net/core/bpf_jit_enable >> # modprobe test_bpf >> # dmesg | grep -w FAIL >> Tail call error path, max count reached jited:1 ret 34 != 33 FAIL >> >> So it is necessary to change the value of MAX_TAIL_CALL_CNT from 32 to 33, >> then do some small changes of the related code. >> >> With this patch, it does not change the current limit, MAX_TAIL_CALL_CNT >> can reflect the actual max tail call count, and the above failed testcase >> can be fixed. >> >> Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn> >> --- > > This change breaks selftests ([0]), please fix them at the same time > as you are changing the kernel behavior: The below selftests shouldn't have to change given there is no change in behavior intended (MAX_TAIL_CALL_CNT is bumped to 33 but counter inc'ed prior to the comparison). It just means that /all/ JITs must be changed and in particular properly _tested_. > test_tailcall_2:PASS:tailcall 128 nsec > test_tailcall_2:PASS:tailcall 128 nsec > test_tailcall_2:FAIL:tailcall err 0 errno 2 retval 4 > #135/2 tailcalls/tailcall_2:FAIL > test_tailcall_3:PASS:tailcall 128 nsec > test_tailcall_3:FAIL:tailcall count err 0 errno 2 count 34 > test_tailcall_3:PASS:tailcall 128 nsec > #135/3 tailcalls/tailcall_3:FAIL > #135/4 tailcalls/tailcall_4:OK > #135/5 tailcalls/tailcall_5:OK > #135/6 tailcalls/tailcall_bpf2bpf_1:OK > test_tailcall_bpf2bpf_2:PASS:tailcall 128 nsec > test_tailcall_bpf2bpf_2:FAIL:tailcall count err 0 errno 2 count 34 > test_tailcall_bpf2bpf_2:PASS:tailcall 128 nsec > #135/7 tailcalls/tailcall_bpf2bpf_2:FAIL > #135/8 tailcalls/tailcall_bpf2bpf_3:OK > test_tailcall_bpf2bpf_4:PASS:tailcall 54 nsec > test_tailcall_bpf2bpf_4:FAIL:tailcall count err 0 errno 2 count 32 > #135/9 tailcalls/tailcall_bpf2bpf_4:FAIL > test_tailcall_bpf2bpf_4:PASS:tailcall 54 nsec > test_tailcall_bpf2bpf_4:FAIL:tailcall count err 0 errno 2 count 32 > #135/10 tailcalls/tailcall_bpf2bpf_5:FAIL > #135 tailcalls:FAIL > > [0] https://github.com/kernel-patches/bpf/pull/1747/checks?check_run_id=3552002906 > >> arch/arm/net/bpf_jit_32.c | 11 ++++++----- >> arch/arm64/net/bpf_jit_comp.c | 7 ++++--- >> arch/mips/net/ebpf_jit.c | 4 ++-- >> arch/powerpc/net/bpf_jit_comp32.c | 4 ++-- >> arch/powerpc/net/bpf_jit_comp64.c | 12 ++++++------ >> arch/riscv/net/bpf_jit_comp32.c | 4 ++-- >> arch/riscv/net/bpf_jit_comp64.c | 4 ++-- >> arch/sparc/net/bpf_jit_comp_64.c | 8 ++++---- >> include/linux/bpf.h | 2 +- >> kernel/bpf/core.c | 4 ++-- >> 10 files changed, 31 insertions(+), 29 deletions(-) >> > > [...] >
diff --git a/arch/arm/net/bpf_jit_32.c b/arch/arm/net/bpf_jit_32.c index a951276..39d9ae9 100644 --- a/arch/arm/net/bpf_jit_32.c +++ b/arch/arm/net/bpf_jit_32.c @@ -1180,18 +1180,19 @@ static int emit_bpf_tail_call(struct jit_ctx *ctx) /* tmp2[0] = array, tmp2[1] = index */ - /* if (tail_call_cnt > MAX_TAIL_CALL_CNT) - * goto out; + /* * tail_call_cnt++; + * if (tail_call_cnt > MAX_TAIL_CALL_CNT) + * goto out; */ + tc = arm_bpf_get_reg64(tcc, tmp, ctx); + emit(ARM_ADDS_I(tc[1], tc[1], 1), ctx); + emit(ARM_ADC_I(tc[0], tc[0], 0), ctx); lo = (u32)MAX_TAIL_CALL_CNT; hi = (u32)((u64)MAX_TAIL_CALL_CNT >> 32); - tc = arm_bpf_get_reg64(tcc, tmp, ctx); emit(ARM_CMP_I(tc[0], hi), ctx); _emit(ARM_COND_EQ, ARM_CMP_I(tc[1], lo), ctx); _emit(ARM_COND_HI, ARM_B(jmp_offset), ctx); - emit(ARM_ADDS_I(tc[1], tc[1], 1), ctx); - emit(ARM_ADC_I(tc[0], tc[0], 0), ctx); arm_bpf_put_reg64(tcc, tmp, ctx); /* prog = array->ptrs[index] diff --git a/arch/arm64/net/bpf_jit_comp.c b/arch/arm64/net/bpf_jit_comp.c index 41c23f4..5d6c843 100644 --- a/arch/arm64/net/bpf_jit_comp.c +++ b/arch/arm64/net/bpf_jit_comp.c @@ -286,14 +286,15 @@ static int emit_bpf_tail_call(struct jit_ctx *ctx) emit(A64_CMP(0, r3, tmp), ctx); emit(A64_B_(A64_COND_CS, jmp_offset), ctx); - /* if (tail_call_cnt > MAX_TAIL_CALL_CNT) - * goto out; + /* * tail_call_cnt++; + * if (tail_call_cnt > MAX_TAIL_CALL_CNT) + * goto out; */ + emit(A64_ADD_I(1, tcc, tcc, 1), ctx); emit_a64_mov_i64(tmp, MAX_TAIL_CALL_CNT, ctx); emit(A64_CMP(1, tcc, tmp), ctx); emit(A64_B_(A64_COND_HI, jmp_offset), ctx); - emit(A64_ADD_I(1, tcc, tcc, 1), ctx); /* prog = array->ptrs[index]; * if (prog == NULL) diff --git a/arch/mips/net/ebpf_jit.c b/arch/mips/net/ebpf_jit.c index 3a73e93..029fc34 100644 --- a/arch/mips/net/ebpf_jit.c +++ b/arch/mips/net/ebpf_jit.c @@ -617,14 +617,14 @@ static int emit_bpf_tail_call(struct jit_ctx *ctx, int this_idx) b_off = b_imm(this_idx + 1, ctx); emit_instr(ctx, bne, MIPS_R_AT, MIPS_R_ZERO, b_off); /* - * if (TCC-- < 0) + * 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); b_off = b_imm(this_idx + 1, ctx); - emit_instr(ctx, bltz, tcc_reg, b_off); + emit_instr(ctx, bltz, MIPS_R_T5, b_off); /* * prog = array->ptrs[index]; * if (prog == NULL) diff --git a/arch/powerpc/net/bpf_jit_comp32.c b/arch/powerpc/net/bpf_jit_comp32.c index beb12cb..b5585ad 100644 --- a/arch/powerpc/net/bpf_jit_comp32.c +++ b/arch/powerpc/net/bpf_jit_comp32.c @@ -221,12 +221,12 @@ static void bpf_jit_emit_tail_call(u32 *image, struct codegen_context *ctx, u32 PPC_BCC(COND_GE, out); /* + * tail_call_cnt++; * if (tail_call_cnt > MAX_TAIL_CALL_CNT) * goto out; */ - EMIT(PPC_RAW_CMPLWI(_R0, MAX_TAIL_CALL_CNT)); - /* tail_call_cnt++; */ EMIT(PPC_RAW_ADDIC(_R0, _R0, 1)); + EMIT(PPC_RAW_CMPLWI(_R0, MAX_TAIL_CALL_CNT)); PPC_BCC(COND_GT, out); /* prog = array->ptrs[index]; */ diff --git a/arch/powerpc/net/bpf_jit_comp64.c b/arch/powerpc/net/bpf_jit_comp64.c index b87a63d..bb15cc4 100644 --- a/arch/powerpc/net/bpf_jit_comp64.c +++ b/arch/powerpc/net/bpf_jit_comp64.c @@ -227,6 +227,12 @@ static void bpf_jit_emit_tail_call(u32 *image, struct codegen_context *ctx, u32 PPC_BCC(COND_GE, out); /* + * tail_call_cnt++; + */ + EMIT(PPC_RAW_ADDI(b2p[TMP_REG_1], b2p[TMP_REG_1], 1)); + PPC_BPF_STL(b2p[TMP_REG_1], 1, bpf_jit_stack_tailcallcnt(ctx)); + + /* * if (tail_call_cnt > MAX_TAIL_CALL_CNT) * goto out; */ @@ -234,12 +240,6 @@ static void bpf_jit_emit_tail_call(u32 *image, struct codegen_context *ctx, u32 EMIT(PPC_RAW_CMPLWI(b2p[TMP_REG_1], MAX_TAIL_CALL_CNT)); PPC_BCC(COND_GT, out); - /* - * tail_call_cnt++; - */ - EMIT(PPC_RAW_ADDI(b2p[TMP_REG_1], b2p[TMP_REG_1], 1)); - PPC_BPF_STL(b2p[TMP_REG_1], 1, bpf_jit_stack_tailcallcnt(ctx)); - /* prog = array->ptrs[index]; */ EMIT(PPC_RAW_MULI(b2p[TMP_REG_1], b2p_index, 8)); EMIT(PPC_RAW_ADD(b2p[TMP_REG_1], b2p[TMP_REG_1], b2p_bpf_array)); diff --git a/arch/riscv/net/bpf_jit_comp32.c b/arch/riscv/net/bpf_jit_comp32.c index e649742..1608d94 100644 --- a/arch/riscv/net/bpf_jit_comp32.c +++ b/arch/riscv/net/bpf_jit_comp32.c @@ -800,12 +800,12 @@ static int emit_bpf_tail_call(int insn, struct rv_jit_context *ctx) /* * temp_tcc = tcc - 1; - * if (tcc < 0) + * if (temp_tcc < 0) * goto out; */ emit(rv_addi(RV_REG_T1, RV_REG_TCC, -1), ctx); off = ninsns_rvoff(tc_ninsn - (ctx->ninsns - start_insn)); - emit_bcc(BPF_JSLT, RV_REG_TCC, RV_REG_ZERO, off, ctx); + emit_bcc(BPF_JSLT, RV_REG_T1, RV_REG_ZERO, off, ctx); /* * prog = array->ptrs[index]; diff --git a/arch/riscv/net/bpf_jit_comp64.c b/arch/riscv/net/bpf_jit_comp64.c index 3af4131..6e9ba83 100644 --- a/arch/riscv/net/bpf_jit_comp64.c +++ b/arch/riscv/net/bpf_jit_comp64.c @@ -311,12 +311,12 @@ static int emit_bpf_tail_call(int insn, struct rv_jit_context *ctx) off = ninsns_rvoff(tc_ninsn - (ctx->ninsns - start_insn)); emit_branch(BPF_JGE, RV_REG_A2, RV_REG_T1, off, ctx); - /* if (TCC-- < 0) + /* if (--TCC < 0) * goto out; */ emit_addi(RV_REG_T1, tcc, -1, ctx); off = ninsns_rvoff(tc_ninsn - (ctx->ninsns - start_insn)); - emit_branch(BPF_JSLT, tcc, RV_REG_ZERO, off, ctx); + emit_branch(BPF_JSLT, RV_REG_T1, RV_REG_ZERO, off, ctx); /* prog = array->ptrs[index]; * if (!prog) diff --git a/arch/sparc/net/bpf_jit_comp_64.c b/arch/sparc/net/bpf_jit_comp_64.c index 9a2f20c..50d914c 100644 --- a/arch/sparc/net/bpf_jit_comp_64.c +++ b/arch/sparc/net/bpf_jit_comp_64.c @@ -863,6 +863,10 @@ static void emit_tail_call(struct jit_ctx *ctx) emit_branch(BGEU, ctx->idx, ctx->idx + OFFSET1, ctx); emit_nop(ctx); + emit_alu_K(ADD, tmp, 1, ctx); + off = BPF_TAILCALL_CNT_SP_OFF; + emit(ST32 | IMMED | RS1(SP) | S13(off) | RD(tmp), ctx); + off = BPF_TAILCALL_CNT_SP_OFF; emit(LD32 | IMMED | RS1(SP) | S13(off) | RD(tmp), ctx); emit_cmpi(tmp, MAX_TAIL_CALL_CNT, ctx); @@ -870,10 +874,6 @@ static void emit_tail_call(struct jit_ctx *ctx) emit_branch(BGU, ctx->idx, ctx->idx + OFFSET2, ctx); emit_nop(ctx); - emit_alu_K(ADD, tmp, 1, ctx); - off = BPF_TAILCALL_CNT_SP_OFF; - emit(ST32 | IMMED | RS1(SP) | S13(off) | RD(tmp), ctx); - emit_alu3_K(SLL, bpf_index, 3, tmp, ctx); emit_alu(ADD, bpf_array, tmp, ctx); off = offsetof(struct bpf_array, ptrs); diff --git a/include/linux/bpf.h b/include/linux/bpf.h index f4c16f1..224cc7e 100644 --- a/include/linux/bpf.h +++ b/include/linux/bpf.h @@ -1046,7 +1046,7 @@ struct bpf_array { }; #define BPF_COMPLEXITY_LIMIT_INSNS 1000000 /* yes. 1M insns */ -#define MAX_TAIL_CALL_CNT 32 +#define MAX_TAIL_CALL_CNT 33 #define BPF_F_ACCESS_MASK (BPF_F_RDONLY | \ BPF_F_RDONLY_PROG | \ diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c index 9f4636d..8edb1c3 100644 --- a/kernel/bpf/core.c +++ b/kernel/bpf/core.c @@ -1564,10 +1564,10 @@ static u64 ___bpf_prog_run(u64 *regs, const struct bpf_insn *insn) if (unlikely(index >= array->map.max_entries)) goto out; - if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT)) - goto out; tail_call_cnt++; + if (unlikely(tail_call_cnt > MAX_TAIL_CALL_CNT)) + goto out; prog = READ_ONCE(array->ptrs[index]); if (!prog)
In the current code, the actual max tail call count is 33 which is greater than MAX_TAIL_CALL_CNT (defined as 32), the actual limit is not consistent with the meaning of MAX_TAIL_CALL_CNT, there is some confusion and need to spend some time to think the reason at the first glance. We can see the historical evolution from commit 04fd61ab36ec ("bpf: allow bpf programs to tail-call other bpf programs") and commit f9dabe016b63 ("bpf: Undo off-by-one in interpreter tail call count limit"). In order to avoid changing existing behavior, the actual limit is 33 now, this is resonable. After commit 874be05f525e ("bpf, tests: Add tail call test suite"), we can see there exists failed testcase. On all archs when CONFIG_BPF_JIT_ALWAYS_ON is not set: # echo 0 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf # dmesg | grep -w FAIL Tail call error path, max count reached jited:0 ret 34 != 33 FAIL On some archs: # echo 1 > /proc/sys/net/core/bpf_jit_enable # modprobe test_bpf # dmesg | grep -w FAIL Tail call error path, max count reached jited:1 ret 34 != 33 FAIL So it is necessary to change the value of MAX_TAIL_CALL_CNT from 32 to 33, then do some small changes of the related code. With this patch, it does not change the current limit, MAX_TAIL_CALL_CNT can reflect the actual max tail call count, and the above failed testcase can be fixed. Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn> --- arch/arm/net/bpf_jit_32.c | 11 ++++++----- arch/arm64/net/bpf_jit_comp.c | 7 ++++--- arch/mips/net/ebpf_jit.c | 4 ++-- arch/powerpc/net/bpf_jit_comp32.c | 4 ++-- arch/powerpc/net/bpf_jit_comp64.c | 12 ++++++------ arch/riscv/net/bpf_jit_comp32.c | 4 ++-- arch/riscv/net/bpf_jit_comp64.c | 4 ++-- arch/sparc/net/bpf_jit_comp_64.c | 8 ++++---- include/linux/bpf.h | 2 +- kernel/bpf/core.c | 4 ++-- 10 files changed, 31 insertions(+), 29 deletions(-)