[PATCH bpf-next v2] selftests/bpf: Rewrite two infinite loops in bound check cases

[PATCH bpf-next v2] selftests/bpf: Rewrite two infinite loops in bound check cases

[Xu Kuohai]

From: Xu Kuohai <xukuohai@huawei.com>

The two infinite loops in bound check cases added by commit
1a3148fc171f ("selftests/bpf: Check when bounds are not in the 32-bit range")
increased the execution time of test_verifier from about 6 seconds to
about 9 seconds. Rewrite these two infinite loops to finite loops to get
rid of this extra time cost.

Signed-off-by: Xu Kuohai <xukuohai@huawei.com>
---
v2:
 - rewrite the infinite loops to finite loops instead of removing the
   test cases

v1: https://lore.kernel.org/bpf/20230327153538.850440-1-xukuohai@huaweicloud.com/
---
 tools/testing/selftests/bpf/verifier/bounds.c | 24 ++++++++++++-------
 1 file changed, 16 insertions(+), 8 deletions(-)

diff --git a/tools/testing/selftests/bpf/verifier/bounds.c b/tools/testing/selftests/bpf/verifier/bounds.c
index 74b1917d4208..43942ce8cf15 100644
--- a/tools/testing/selftests/bpf/verifier/bounds.c
+++ b/tools/testing/selftests/bpf/verifier/bounds.c
@@ -784,22 +784,26 @@
 	BPF_LDX_MEM(BPF_W, BPF_REG_3, BPF_REG_1, offsetof(struct xdp_md, data_end)),
 	BPF_MOV64_REG(BPF_REG_1, BPF_REG_2),
 	BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, 1),
-	BPF_JMP_REG(BPF_JGT, BPF_REG_1, BPF_REG_3, 8),
+	BPF_JMP_REG(BPF_JGT, BPF_REG_1, BPF_REG_3, 13),
 
 	BPF_LDX_MEM(BPF_B, BPF_REG_1, BPF_REG_2, 0),
 	BPF_LD_IMM64(BPF_REG_0, 0x7fffffffffffff10),
 	BPF_ALU64_REG(BPF_ADD, BPF_REG_1, BPF_REG_0),
 
+	BPF_LD_IMM64(BPF_REG_2, 0x8000000000000fff),
 	BPF_LD_IMM64(BPF_REG_0, 0x8000000000000000),
 	BPF_ALU64_IMM(BPF_ADD, BPF_REG_0, 1),
+	BPF_JMP_REG(BPF_JSGT, BPF_REG_0, BPF_REG_2, 3),
 	/* r1 signed range is [S64_MIN, S64_MAX] */
-	BPF_JMP_REG(BPF_JSLT, BPF_REG_0, BPF_REG_1, -2),
+	BPF_JMP_REG(BPF_JSLT, BPF_REG_0, BPF_REG_1, -3),
+
+	BPF_MOV64_IMM(BPF_REG_0, 1),
+	BPF_EXIT_INSN(),
 
 	BPF_MOV64_IMM(BPF_REG_0, 0),
 	BPF_EXIT_INSN(),
 	},
-	.errstr = "BPF program is too large",
-	.result = REJECT,
+	.result = ACCEPT,
 	.prog_type = BPF_PROG_TYPE_XDP,
 },
 {
@@ -856,21 +860,25 @@
 	BPF_LDX_MEM(BPF_W, BPF_REG_3, BPF_REG_1, offsetof(struct xdp_md, data_end)),
 	BPF_MOV64_REG(BPF_REG_1, BPF_REG_2),
 	BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, 1),
-	BPF_JMP_REG(BPF_JGT, BPF_REG_1, BPF_REG_3, 6),
+	BPF_JMP_REG(BPF_JGT, BPF_REG_1, BPF_REG_3, 10),
 
 	BPF_LDX_MEM(BPF_B, BPF_REG_1, BPF_REG_2, 0),
 	BPF_MOV32_IMM(BPF_REG_0, 0x7fffff10),
 	BPF_ALU32_REG(BPF_ADD, BPF_REG_1, BPF_REG_0),
 
+	BPF_MOV32_IMM(BPF_REG_2, 0x80000fff),
 	BPF_MOV32_IMM(BPF_REG_0, 0x80000000),
 	BPF_ALU32_IMM(BPF_ADD, BPF_REG_0, 1),
+	BPF_JMP32_REG(BPF_JSGT, BPF_REG_0, BPF_REG_2, 3),
 	/* r1 signed range is [S32_MIN, S32_MAX] */
-	BPF_JMP32_REG(BPF_JSLT, BPF_REG_0, BPF_REG_1, -2),
+	BPF_JMP32_REG(BPF_JSLT, BPF_REG_0, BPF_REG_1, -3),
+
+	BPF_MOV64_IMM(BPF_REG_0, 1),
+	BPF_EXIT_INSN(),
 
 	BPF_MOV64_IMM(BPF_REG_0, 0),
 	BPF_EXIT_INSN(),
 	},
-	.errstr = "BPF program is too large",
-	.result = REJECT,
+	.result = ACCEPT,
 	.prog_type = BPF_PROG_TYPE_XDP,
 },
-- 
2.30.2

Re: [PATCH bpf-next v2] selftests/bpf: Rewrite two infinite loops in bound check cases

[Alexei Starovoitov]

On Tue, Mar 28, 2023 at 5:11 AM Xu Kuohai <xukuohai@huaweicloud.com> wrote:
>
> From: Xu Kuohai <xukuohai@huawei.com>
>
> The two infinite loops in bound check cases added by commit
> 1a3148fc171f ("selftests/bpf: Check when bounds are not in the 32-bit range")
> increased the execution time of test_verifier from about 6 seconds to
> about 9 seconds. Rewrite these two infinite loops to finite loops to get
> rid of this extra time cost.
>
> Signed-off-by: Xu Kuohai <xukuohai@huawei.com>
> ---
> v2:
>  - rewrite the infinite loops to finite loops instead of removing the
>    test cases
>
> v1: https://lore.kernel.org/bpf/20230327153538.850440-1-xukuohai@huaweicloud.com/

Thanks.
On my VM with debug kernel it went
from:
real   0m37.508s
user   0m0.382s
sys    0m36.555s

to:
real   0m14.547s
user   0m0.378s
sys    0m13.661s

Applied.

Re: [PATCH bpf-next v2] selftests/bpf: Rewrite two infinite loops in bound check cases

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to bpf/bpf-next.git (master)
by Alexei Starovoitov <ast@kernel.org>:

On Tue, 28 Mar 2023 21:10:48 -0400 you wrote:
> From: Xu Kuohai <xukuohai@huawei.com>
> 
> The two infinite loops in bound check cases added by commit
> 1a3148fc171f ("selftests/bpf: Check when bounds are not in the 32-bit range")
> increased the execution time of test_verifier from about 6 seconds to
> about 9 seconds. Rewrite these two infinite loops to finite loops to get
> rid of this extra time cost.
> 
> [...]

Here is the summary with links:
  - [bpf-next,v2] selftests/bpf: Rewrite two infinite loops in bound check cases
    https://git.kernel.org/bpf/bpf-next/c/4ca13d1002f3

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html