On 4/4/24 1:09 PM, Andrii Nakryiko wrote:
On Thu, Apr 4, 2024 at 11:48 AM Yonghong Song <yonghong.song@xxxxxxxxx> wrote:
On 4/2/24 4:26 PM, Andrii Nakryiko wrote:
On Tue, Apr 2, 2024 at 3:50 PM Andrii Nakryiko <andrii@xxxxxxxxxx> wrote:
Add selftests validating that BPF verifier handles precision marking
for SCALAR registers derived from r10 (fp) register correctly.
Signed-off-by: Andrii Nakryiko <andrii@xxxxxxxxxx>
---
.../bpf/progs/verifier_subprog_precision.c | 86 +++++++++++++++++++
1 file changed, 86 insertions(+)
diff --git a/tools/testing/selftests/bpf/progs/verifier_subprog_precision.c b/tools/testing/selftests/bpf/progs/verifier_subprog_precision.c
index 6f5d19665cf6..e1a8f107f0a7 100644
--- a/tools/testing/selftests/bpf/progs/verifier_subprog_precision.c
+++ b/tools/testing/selftests/bpf/progs/verifier_subprog_precision.c
@@ -76,6 +76,92 @@ __naked int subprog_result_precise(void)
);
}
+__naked __noinline __used
+static unsigned long fp_leaking_subprog()
+{
+ asm volatile (
+ "r0 = (s8)r10;"
Our CI's clang doesn't like this instruction. I guess I'll have to
encode it in binary form :(
This patch disappeared from CI so I am not able to check the result.
But I tried with the following small example.
$ cat t.c
__attribute__((naked)) unsigned long t(void)
{
asm volatile("r0 = (s8)r10;"
"exit;"
);
}
$ clang --target=bpf -O2 -mcpu=v2 -g -c t.c && llvm-objdump -d t.o
You are using local clang built from source code, right? I think our
BPF CI still is on Clang 17 or something, so it doesn't yet understand
"(s8)r10" syntax, unfortunately.
Yes, it makes sense. Indeed in that case, either using bytes or guarding
with >= llvm18 is needed.
t.o: file format elf64-bpf
Disassembly of section .text:
0000000000000000 <t>:
0: bf a0 08 00 00 00 00 00 r0 = (s8)r10
1: 95 00 00 00 00 00 00 00 exit
-mcpu=v3/v4 has the same result.
Not sure what clang complains.
+ "exit;"
+ );
+}
+
[...]
