Instead of unconditionally performing push/pop on %rax/%rdx in case of
division/modulo, we can save a few bytes in case of destination register
being either BPF r0 (%rax) or r3 (%rdx) since the result is written in
there anyway.
Also, we do not need to copy the source to %r11 unless the source is either
%rax, %rdx or an immediate.
For example, before the patch:
22: push %rax
23: push %rdx
24: mov %rsi,%r11
27: xor %edx,%edx
29: div %r11
2c: mov %rax,%r11
2f: pop %rdx
30: pop %rax
31: mov %r11,%rax
After:
22: push %rdx
23: xor %edx,%edx
25: div %rsi
28: pop %rdx
Signed-off-by: Jie Meng <jmeng@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20211002035626.2041910-1-jmeng@fb.com
IMUL allows for multiple operands and saving and storing rax/rdx is no
longer needed. Signedness of the operands doesn't matter here because
the we only keep the lower 32/64 bit of the product for 32/64 bit
multiplications.
Signed-off-by: Jie Meng <jmeng@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210913211337.1564014-1-jmeng@fb.com
There are 3 tests added into verifier's jit tests to trigger x64
jit jump padding.
The first test can be represented as the following assembly code:
1: bpf_call bpf_get_prandom_u32
2: if r0 == 1 goto pc+128
3: if r0 == 2 goto pc+128
...
129: if r0 == 128 goto pc+128
130: goto pc+128
131: goto pc+127
...
256: goto pc+2
257: goto pc+1
258: r0 = 1
259: ret
We first store a random number to r0 and add the corresponding
conditional jumps (2~129) to make verifier believe that those jump
instructions from 130 to 257 are reachable. When the program is sent to
x64 jit, it starts to optimize out the NOP jumps backwards from 257.
Since there are 128 such jumps, the program easily reaches 15 passes and
triggers jump padding.
Here is the x64 jit code of the first test:
0: 0f 1f 44 00 00 nop DWORD PTR [rax+rax*1+0x0]
5: 66 90 xchg ax,ax
7: 55 push rbp
8: 48 89 e5 mov rbp,rsp
b: e8 4c 90 75 e3 call 0xffffffffe375905c
10: 48 83 f8 01 cmp rax,0x1
14: 0f 84 fe 04 00 00 je 0x518
1a: 48 83 f8 02 cmp rax,0x2
1e: 0f 84 f9 04 00 00 je 0x51d
...
f6: 48 83 f8 18 cmp rax,0x18
fa: 0f 84 8b 04 00 00 je 0x58b
100: 48 83 f8 19 cmp rax,0x19
104: 0f 84 86 04 00 00 je 0x590
10a: 48 83 f8 1a cmp rax,0x1a
10e: 0f 84 81 04 00 00 je 0x595
...
500: 0f 84 83 01 00 00 je 0x689
506: 48 81 f8 80 00 00 00 cmp rax,0x80
50d: 0f 84 76 01 00 00 je 0x689
513: e9 71 01 00 00 jmp 0x689
518: e9 6c 01 00 00 jmp 0x689
...
5fe: e9 86 00 00 00 jmp 0x689
603: e9 81 00 00 00 jmp 0x689
608: 0f 1f 00 nop DWORD PTR [rax]
60b: eb 7c jmp 0x689
60d: eb 7a jmp 0x689
...
683: eb 04 jmp 0x689
685: eb 02 jmp 0x689
687: 66 90 xchg ax,ax
689: b8 01 00 00 00 mov eax,0x1
68e: c9 leave
68f: c3 ret
As expected, a 3 bytes NOPs is inserted at 608 due to the transition
from imm32 jmp to imm8 jmp. A 2 bytes NOPs is also inserted at 687 to
replace a NOP jump.
The second test case is tricky. Here is the assembly code:
1: bpf_call bpf_get_prandom_u32
2: if r0 == 1 goto pc+2048
3: if r0 == 2 goto pc+2048
...
2049: if r0 == 2048 goto pc+2048
2050: goto pc+2048
2051: goto pc+16
2052: goto pc+15
...
2064: goto pc+3
2065: goto pc+2
2066: goto pc+1
...
[repeat "goto pc+16".."goto pc+1" 127 times]
...
4099: r0 = 2
4100: ret
There are 4 major parts of the program.
1) 1~2049: Those are instructions to make 2050~4098 reachable. Some of
them also could generate the padding for jmp_cond.
2) 2050: This is the target instruction for the imm32 nop jmp padding.
3) 2051~4098: The repeated "goto 1~16" instructions are designed to be
consumed by the nop jmp optimization. In the end, those
instrucitons become 128 continuous 0 offset jmp and are
optimized out in 1 pass, and this make insn 2050 an imm32
nop jmp in the next pass, so that we can trigger the
5 bytes padding.
4) 4099~4100: Those are the instructions to end the program.
The x64 jit code is like this:
0: 0f 1f 44 00 00 nop DWORD PTR [rax+rax*1+0x0]
5: 66 90 xchg ax,ax
7: 55 push rbp
8: 48 89 e5 mov rbp,rsp
b: e8 bc 7b d5 d3 call 0xffffffffd3d57bcc
10: 48 83 f8 01 cmp rax,0x1
14: 0f 84 7e 66 00 00 je 0x6698
1a: 48 83 f8 02 cmp rax,0x2
1e: 0f 84 74 66 00 00 je 0x6698
24: 48 83 f8 03 cmp rax,0x3
28: 0f 84 6a 66 00 00 je 0x6698
2e: 48 83 f8 04 cmp rax,0x4
32: 0f 84 60 66 00 00 je 0x6698
38: 48 83 f8 05 cmp rax,0x5
3c: 0f 84 56 66 00 00 je 0x6698
42: 48 83 f8 06 cmp rax,0x6
46: 0f 84 4c 66 00 00 je 0x6698
...
666c: 48 81 f8 fe 07 00 00 cmp rax,0x7fe
6673: 0f 1f 40 00 nop DWORD PTR [rax+0x0]
6677: 74 1f je 0x6698
6679: 48 81 f8 ff 07 00 00 cmp rax,0x7ff
6680: 0f 1f 40 00 nop DWORD PTR [rax+0x0]
6684: 74 12 je 0x6698
6686: 48 81 f8 00 08 00 00 cmp rax,0x800
668d: 0f 1f 40 00 nop DWORD PTR [rax+0x0]
6691: 74 05 je 0x6698
6693: 0f 1f 44 00 00 nop DWORD PTR [rax+rax*1+0x0]
6698: b8 02 00 00 00 mov eax,0x2
669d: c9 leave
669e: c3 ret
Since insn 2051~4098 are optimized out right before the padding pass,
there are several conditional jumps from the first part are replaced with
imm8 jmp_cond, and this triggers the 4 bytes padding, for example at
6673, 6680, and 668d. On the other hand, Insn 2050 is replaced with the
5 bytes nops at 6693.
The third test is to invoke the first and second tests as subprogs to test
bpf2bpf. Per the system log, there was one more jit happened with only
one pass and the same jit code was produced.
v4:
- Add the second test case which triggers jmp_cond padding and imm32 nop
jmp padding.
- Add the new test case as another subprog
Signed-off-by: Gary Lin <glin@suse.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20210119102501.511-4-glin@suse.com
The current method to compare 64-bit numbers for conditional jump is:
1) Compare the high 32-bit first.
2) If the high 32-bit isn't the same, then goto step 4.
3) Compare the low 32-bit.
4) Check the desired condition.
This method is right for unsigned comparison, but it is buggy for signed
comparison, because it does signed comparison for low 32-bit too.
There is only one sign bit in 64-bit number, that is the MSB in the 64-bit
number, it is wrong to treat low 32-bit as signed number and do the signed
comparison for it.
This patch fixes the bug and adds a testcase in selftests/bpf for such bug.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Break up the rest of test_verifier tests into separate
files.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Jiong Wang <jiong.wang@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
