diff mbox series

[bpf-next,v2,1/2] bpf, x64: Fix a jit convergence issue

Message ID 20240903225949.2066234-1-yonghong.song@linux.dev (mailing list archive)
State Superseded
Delegated to: BPF
Headers show
Series [bpf-next,v2,1/2] bpf, x64: Fix a jit convergence issue | expand

Checks

Context Check Description
bpf/vmtest-bpf-next-PR success PR summary
bpf/vmtest-bpf-next-VM_Test-1 success Logs for ShellCheck
bpf/vmtest-bpf-next-VM_Test-0 success Logs for Lint
bpf/vmtest-bpf-next-VM_Test-2 success Logs for Unittests
bpf/vmtest-bpf-next-VM_Test-3 success Logs for Validate matrix.py
bpf/vmtest-bpf-next-VM_Test-5 success Logs for aarch64-gcc / build-release
bpf/vmtest-bpf-next-VM_Test-4 success Logs for aarch64-gcc / build / build for aarch64 with gcc
bpf/vmtest-bpf-next-VM_Test-10 success Logs for aarch64-gcc / veristat
bpf/vmtest-bpf-next-VM_Test-11 success Logs for s390x-gcc / build / build for s390x with gcc
bpf/vmtest-bpf-next-VM_Test-12 success Logs for s390x-gcc / build-release
bpf/vmtest-bpf-next-VM_Test-16 success Logs for s390x-gcc / veristat
bpf/vmtest-bpf-next-VM_Test-18 success Logs for x86_64-gcc / build / build for x86_64 with gcc
bpf/vmtest-bpf-next-VM_Test-17 success Logs for set-matrix
bpf/vmtest-bpf-next-VM_Test-19 success Logs for x86_64-gcc / build-release
bpf/vmtest-bpf-next-VM_Test-27 success Logs for x86_64-llvm-17 / build / build for x86_64 with llvm-17
bpf/vmtest-bpf-next-VM_Test-28 success Logs for x86_64-llvm-17 / build-release / build for x86_64 with llvm-17-O2
bpf/vmtest-bpf-next-VM_Test-34 success Logs for x86_64-llvm-18 / build / build for x86_64 with llvm-18
bpf/vmtest-bpf-next-VM_Test-41 success Logs for x86_64-llvm-18 / veristat
bpf/vmtest-bpf-next-VM_Test-35 success Logs for x86_64-llvm-18 / build-release / build for x86_64 with llvm-18-O2
bpf/vmtest-bpf-next-VM_Test-33 success Logs for x86_64-llvm-17 / veristat
bpf/vmtest-bpf-next-VM_Test-6 success Logs for aarch64-gcc / test (test_maps, false, 360) / test_maps on aarch64 with gcc
bpf/vmtest-bpf-next-VM_Test-9 success Logs for aarch64-gcc / test (test_verifier, false, 360) / test_verifier on aarch64 with gcc
bpf/vmtest-bpf-next-VM_Test-13 success Logs for s390x-gcc / test (test_progs, false, 360) / test_progs on s390x with gcc
bpf/vmtest-bpf-next-VM_Test-15 success Logs for s390x-gcc / test (test_verifier, false, 360) / test_verifier on s390x with gcc
bpf/vmtest-bpf-next-VM_Test-7 success Logs for aarch64-gcc / test (test_progs, false, 360) / test_progs on aarch64 with gcc
bpf/vmtest-bpf-next-VM_Test-8 success Logs for aarch64-gcc / test (test_progs_no_alu32, false, 360) / test_progs_no_alu32 on aarch64 with gcc
bpf/vmtest-bpf-next-VM_Test-20 success Logs for x86_64-gcc / test (test_maps, false, 360) / test_maps on x86_64 with gcc
bpf/vmtest-bpf-next-VM_Test-21 success Logs for x86_64-gcc / test (test_progs, false, 360) / test_progs on x86_64 with gcc
bpf/vmtest-bpf-next-VM_Test-22 success Logs for x86_64-gcc / test (test_progs_no_alu32, false, 360) / test_progs_no_alu32 on x86_64 with gcc
bpf/vmtest-bpf-next-VM_Test-26 success Logs for x86_64-gcc / veristat / veristat on x86_64 with gcc
bpf/vmtest-bpf-next-VM_Test-25 success Logs for x86_64-gcc / test (test_verifier, false, 360) / test_verifier on x86_64 with gcc
bpf/vmtest-bpf-next-VM_Test-30 success Logs for x86_64-llvm-17 / test (test_progs, false, 360) / test_progs on x86_64 with llvm-17
bpf/vmtest-bpf-next-VM_Test-32 success Logs for x86_64-llvm-17 / test (test_verifier, false, 360) / test_verifier on x86_64 with llvm-17
bpf/vmtest-bpf-next-VM_Test-36 success Logs for x86_64-llvm-18 / test (test_maps, false, 360) / test_maps on x86_64 with llvm-18
bpf/vmtest-bpf-next-VM_Test-40 success Logs for x86_64-llvm-18 / test (test_verifier, false, 360) / test_verifier on x86_64 with llvm-18
bpf/vmtest-bpf-next-VM_Test-23 success Logs for x86_64-gcc / test (test_progs_no_alu32_parallel, true, 30) / test_progs_no_alu32_parallel on x86_64 with gcc
bpf/vmtest-bpf-next-VM_Test-24 success Logs for x86_64-gcc / test (test_progs_parallel, true, 30) / test_progs_parallel on x86_64 with gcc
bpf/vmtest-bpf-next-VM_Test-29 success Logs for x86_64-llvm-17 / test (test_maps, false, 360) / test_maps on x86_64 with llvm-17
bpf/vmtest-bpf-next-VM_Test-31 success Logs for x86_64-llvm-17 / test (test_progs_no_alu32, false, 360) / test_progs_no_alu32 on x86_64 with llvm-17
bpf/vmtest-bpf-next-VM_Test-14 success Logs for s390x-gcc / test (test_progs_no_alu32, false, 360) / test_progs_no_alu32 on s390x with gcc
bpf/vmtest-bpf-next-VM_Test-37 success Logs for x86_64-llvm-18 / test (test_progs, false, 360) / test_progs on x86_64 with llvm-18
bpf/vmtest-bpf-next-VM_Test-39 success Logs for x86_64-llvm-18 / test (test_progs_no_alu32, false, 360) / test_progs_no_alu32 on x86_64 with llvm-18
bpf/vmtest-bpf-next-VM_Test-38 success Logs for x86_64-llvm-18 / test (test_progs_cpuv4, false, 360) / test_progs_cpuv4 on x86_64 with llvm-18
netdev/tree_selection success Clearly marked for bpf-next, async
netdev/apply success Patch already applied to bpf-next-0

Commit Message

Yonghong Song Sept. 3, 2024, 10:59 p.m. UTC
Daniel Hodges reported a jit error when playing with a sched-ext program.
The error message is:
  unexpected jmp_cond padding: -4 bytes

But further investigation shows the error is actual due to failed
convergence. The following are some analysis:

  ...
  pass4, final_proglen=4391:
    ...
    20e:    48 85 ff                test   rdi,rdi
    211:    74 7d                   je     0x290
    213:    48 8b 77 00             mov    rsi,QWORD PTR [rdi+0x0]
    ...
    289:    48 85 ff                test   rdi,rdi
    28c:    74 17                   je     0x2a5
    28e:    e9 7f ff ff ff          jmp    0x212
    293:    bf 03 00 00 00          mov    edi,0x3

Note that insn at 0x211 is 2-byte cond jump insn for offset 0x7d (-125)
and insn at 0x28e is 5-byte jmp insn with offset -129.

  pass5, final_proglen=4392:
    ...
    20e:    48 85 ff                test   rdi,rdi
    211:    0f 84 80 00 00 00       je     0x297
    217:    48 8b 77 00             mov    rsi,QWORD PTR [rdi+0x0]
    ...
    28d:    48 85 ff                test   rdi,rdi
    290:    74 1a                   je     0x2ac
    292:    eb 84                   jmp    0x218
    294:    bf 03 00 00 00          mov    edi,0x3

Note that insn at 0x211 is 6-byte cond jump insn now since its offset
becomes 0x80 based on previous round (0x293 - 0x213 = 0x80). At the same
time, insn at 0x292 is a 2-byte insn since its offset is -124.

pass6 will repeat the same code as in pass4. pass7 will repeat the same
code as in pass5, and so on. This will prevent eventual convergence.

Passes 1-14 are with padding = 0. At pass15, padding is 1 and related
insn looks like:

    211:    0f 84 80 00 00 00       je     0x297
    217:    48 8b 77 00             mov    rsi,QWORD PTR [rdi+0x0]
    ...
    24d:    48 85 d2                test   rdx,rdx

The similar code in pass14:
    211:    74 7d                   je     0x290
    213:    48 8b 77 00             mov    rsi,QWORD PTR [rdi+0x0]
    ...
    249:    48 85 d2                test   rdx,rdx
    24c:    74 21                   je     0x26f
    24e:    48 01 f7                add    rdi,rsi
    ...

Before generating the following insn,
  250:    74 21                   je     0x273
"padding = 1" enables some checking to ensure nops is either 0 or 4
where
  #define INSN_SZ_DIFF (((addrs[i] - addrs[i - 1]) - (prog - temp)))
  nops = INSN_SZ_DIFF - 2

In this specific case,
  addrs[i] = 0x24e // from pass14
  addrs[i-1] = 0x24d // from pass15
  prog - temp = 3 // from 'test rdx,rdx' in pass15
so
  nops = -4
and this triggers the failure.

To fix the issue, we need to break cycles of je <-> jmp. For example,
in the above case, we have
  211:    74 7d                   je     0x290
the offset is 0x7d. If 2-byte je insn is generated only if
the offset is less than 0x7d (<= 0x7c), the cycle can be
break and we can achieve the convergence.

I did some study on other cases like je <-> je, jmp <-> je and
jmp <-> jmp which may cause cycles. Those cases are not from actual
reproducible cases since it is pretty hard to construct a test case
for them. the results show that the offset <= 0x7b (0x7b = 123) should
be enough to cover all cases. This patch added a new helper to generate 8-bit
cond/uncond jmp insns only if the offset range is [-128, 123].

Reported-by: Daniel Hodges <hodgesd@meta.com>
Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
---
 arch/x86/net/bpf_jit_comp.c | 55 +++++++++++++++++++++++++++++++++++--
 1 file changed, 53 insertions(+), 2 deletions(-)

Comments

Alexei Starovoitov Sept. 4, 2024, 4:56 p.m. UTC | #1
On Tue, Sep 3, 2024 at 4:00 PM Yonghong Song <yonghong.song@linux.dev> wrote:
>
>
> diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
> index 074b41fafbe3..63c4816ed4e7 100644
> --- a/arch/x86/net/bpf_jit_comp.c
> +++ b/arch/x86/net/bpf_jit_comp.c
> @@ -64,6 +64,57 @@ static bool is_imm8(int value)
>         return value <= 127 && value >= -128;
>  }
>
> +/*
> + * Let us limit the positive offset to be <= 124.

I think the comment will read better if the above says "<= 123",
since that's the final outcome.
124 above is a bit confusing.
I can tweak while applying if you agree.

> + * to mamximum 123 (0x7b). This way, the jit pass can eventually converge.

Can fix this typo too.
Yonghong Song Sept. 4, 2024, 6:32 p.m. UTC | #2
On 9/4/24 9:56 AM, Alexei Starovoitov wrote:
> On Tue, Sep 3, 2024 at 4:00 PM Yonghong Song <yonghong.song@linux.dev> wrote:
>>
>> diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
>> index 074b41fafbe3..63c4816ed4e7 100644
>> --- a/arch/x86/net/bpf_jit_comp.c
>> +++ b/arch/x86/net/bpf_jit_comp.c
>> @@ -64,6 +64,57 @@ static bool is_imm8(int value)
>>          return value <= 127 && value >= -128;
>>   }
>>
>> +/*
>> + * Let us limit the positive offset to be <= 124.
> I think the comment will read better if the above says "<= 123",
> since that's the final outcome.
> 124 above is a bit confusing.
> I can tweak while applying if you agree.

Correct. Let us do "<= 123".

>
>> + * to mamximum 123 (0x7b). This way, the jit pass can eventually converge.
> Can fix this typo too.

Sounds good.

Since I need to change test in the second patch (enabling tests for arm64/s390),
let me send v3 to address all comments.
diff mbox series

Patch

diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
index 074b41fafbe3..63c4816ed4e7 100644
--- a/arch/x86/net/bpf_jit_comp.c
+++ b/arch/x86/net/bpf_jit_comp.c
@@ -64,6 +64,57 @@  static bool is_imm8(int value)
 	return value <= 127 && value >= -128;
 }
 
+/*
+ * Let us limit the positive offset to be <= 124.
+ * This is to ensure eventual jit convergence For the following patterns:
+ * ...
+ * pass4, final_proglen=4391:
+ *   ...
+ *   20e:    48 85 ff                test   rdi,rdi
+ *   211:    74 7d                   je     0x290
+ *   213:    48 8b 77 00             mov    rsi,QWORD PTR [rdi+0x0]
+ *   ...
+ *   289:    48 85 ff                test   rdi,rdi
+ *   28c:    74 17                   je     0x2a5
+ *   28e:    e9 7f ff ff ff          jmp    0x212
+ *   293:    bf 03 00 00 00          mov    edi,0x3
+ * Note that insn at 0x211 is 2-byte cond jump insn for offset 0x7d (-125)
+ * and insn at 0x28e is 5-byte jmp insn with offset -129.
+ *
+ * pass5, final_proglen=4392:
+ *   ...
+ *   20e:    48 85 ff                test   rdi,rdi
+ *   211:    0f 84 80 00 00 00       je     0x297
+ *   217:    48 8b 77 00             mov    rsi,QWORD PTR [rdi+0x0]
+ *   ...
+ *   28d:    48 85 ff                test   rdi,rdi
+ *   290:    74 1a                   je     0x2ac
+ *   292:    eb 84                   jmp    0x218
+ *   294:    bf 03 00 00 00          mov    edi,0x3
+ * Note that insn at 0x211 is 6-byte cond jump insn now since its offset
+ * becomes 0x80 based on previous round (0x293 - 0x213 = 0x80).
+ * At the same time, insn at 0x292 is a 2-byte insn since its offset is
+ * -124.
+ *
+ * pass6 will repeat the same code as in pass4 and this will prevent
+ * eventual convergence.
+ *
+ * To fix this issue, we need to break je (2->6 bytes) <-> jmp (5->2 bytes)
+ * cycle in the above. In the above example that we found je offset <= 0x7c
+ * should work.
+ *
+ * For other cases, je <-> je needs offset <= 0x7b to avoid no convergence
+ * issue. For jmp <-> je and jmp <-> jmp cases, jmp offset <= 0x7c should
+ * avoid no convergence issue.
+ *
+ * Overall, let us limit the positive offset for 8bit cond/uncond jmp insn
+ * to mamximum 123 (0x7b). This way, the jit pass can eventually converge.
+ */
+static bool is_imm8_jmp_offset(int value)
+{
+	return value <= 123 && value >= -128;
+}
+
 static bool is_simm32(s64 value)
 {
 	return value == (s64)(s32)value;
@@ -2231,7 +2282,7 @@  st:			if (is_imm8(insn->off))
 				return -EFAULT;
 			}
 			jmp_offset = addrs[i + insn->off] - addrs[i];
-			if (is_imm8(jmp_offset)) {
+			if (is_imm8_jmp_offset(jmp_offset)) {
 				if (jmp_padding) {
 					/* To keep the jmp_offset valid, the extra bytes are
 					 * padded before the jump insn, so we subtract the
@@ -2313,7 +2364,7 @@  st:			if (is_imm8(insn->off))
 				break;
 			}
 emit_jmp:
-			if (is_imm8(jmp_offset)) {
+			if (is_imm8_jmp_offset(jmp_offset)) {
 				if (jmp_padding) {
 					/* To avoid breaking jmp_offset, the extra bytes
 					 * are padded before the actual jmp insn, so