Re: [PATCH 2/2] selftests/bpf: add verifier test for PTR_TO_MEM spill

[Yonghong Song <yhs@fb.com>]

On 1/12/21 7:43 AM, Daniel Borkmann wrote:
> On 1/12/21 4:35 PM, Gilad Reti wrote:
>> On Tue, Jan 12, 2021 at 4:56 PM KP Singh <kpsingh@kernel.org> wrote:
>>> On Tue, Jan 12, 2021 at 10:16 AM Gilad Reti <gilad.reti@gmail.com> 
>>> wrote:
>>>>
>>>> Add test to check that the verifier is able to recognize spilling of
>>>> PTR_TO_MEM registers.
>>>
>>> It would be nice to have some explanation of what the test does to
>>> recognize the spilling of the PTR_TO_MEM registers in the commit
>>> log as well.
>>>
>>> Would it be possible to augment an existing test_progs
>>> program like tools/testing/selftests/bpf/progs/test_ringbuf.c to test
>>> this functionality?
> 
> How would you guarantee that LLVM generates the spill/fill, via inline asm?

You can make the following change to force the return value ("sample" 
here) of bpf_ringbuf_reserve() to spill on the stack.

diff --git a/tools/testing/selftests/bpf/progs/test_ringbuf.c 
b/tools/testing/selftests/bpf/progs/test_ringbuf.c
index 8ba9959b036b..011521170856 100644
--- a/tools/testing/selftests/bpf/progs/test_ringbuf.c
+++ b/tools/testing/selftests/bpf/progs/test_ringbuf.c
@@ -40,7 +40,7 @@ SEC("tp/syscalls/sys_enter_getpgid")
  int test_ringbuf(void *ctx)
  {
         int cur_pid = bpf_get_current_pid_tgid() >> 32;
-       struct sample *sample;
+       struct sample * volatile sample;
         int zero = 0;

         if (cur_pid != pid)

This change will cause verifier failure without Patch #1.

> 
>> It may be possible, but from what I understood from Daniel's comment here
>>
>> https://lore.kernel.org/bpf/17629073-4fab-a922-ecc3-25b019960f44@iogearbox.net/ 
>>
>>
>> the test should be a part of the verifier tests (which is reasonable
>> to me since it is
>> a verifier bugfix)
> 
> Yeah, the test_verifier case as you have is definitely the most straight
> forward way to add coverage in this case.