Re: [RFC PATCH bpf-next 3/6] selftests/test: test gen_prologue and gen_epilogue

[Martin KaFai Lau <martin.lau@linux.dev>]

On 8/14/24 1:48 PM, Eduard Zingerman wrote:
> Hi Martin,
> 
> Please note that after changes for struct_ops map autoload by libbpf,
> test_loader could be use to test struct_ops related changes.
> Also, test_loader now supports __xlated macro which allows to verify
> rewrites applied by verifier.
> For example, the sample below works:
> 
>      struct st_ops_args;
>      
>      struct bpf_testmod_st_ops {
>      	int (*test_prologue)(struct st_ops_args *args);
>      	int (*test_epilogue)(struct st_ops_args *args);
>      	int (*test_pro_epilogue)(struct st_ops_args *args);
>      	struct module *owner;
>      };
>      
>      __success
>      __xlated("0: *(u64 *)(r10 -8) = r1")
>      __xlated("1: r0 = 0")
>      __xlated("2: r1 = *(u64 *)(r10 -8)")
>      __xlated("3: r1 = *(u64 *)(r1 +0)")
>      __xlated("4: r6 = *(u32 *)(r1 +0)")
>      __xlated("5: w6 += 10000")
>      __xlated("6: *(u32 *)(r1 +0) = r6")
>      __xlated("7: r6 = r1")
>      __xlated("8: call kernel-function")
>      __xlated("9: r1 = r6")
>      __xlated("10: call kernel-function")
>      __xlated("11: w0 *= 2")
>      __xlated("12: exit")

It is appealing to be able to check at the xlated instruction level for 
.gen_pro/epilogue.

>      SEC("struct_ops/test_epilogue")
>      __naked int test_epilogue(void)
>      {
>      	asm volatile (
>      	"r0 = 0;"

I also want to test a struct_ops prog making kfunc call, e.g. the 
BPF_PROG(test_epilogue_kfunc) in this patch. I have never tried this in asm, so 
a n00b question. Do you know if there is an example how to call kfunc?

>      	"exit;"
>      	::: __clobber_all);
>      }
>      
>      SEC(".struct_ops.link")
>      struct bpf_testmod_st_ops st_ops = {
>      	.test_epilogue = (void *)test_epilogue,
>      };
> 
> (Complete example is in the attachment).
> test_loader based tests can also trigger program execution via __retval() macro.
> The only (minor) shortcoming that I see, is that test_loader would
> load/unload st_ops map multiple times because of the following
> interaction:
> - test_loader assumes that each bpf program defines a test;
> - test_loader re-creates all maps before each test;
> - libbpf struct_ops autocreate logic marks all programs referenced
>    from struct_ops map as autoloaded.

If I understand correctly, there are redundant works but still work?

Potentially the test_loader can check all the loaded struct_ops progs of a 
st_ops map at once which is an optimization.

Re: __retval(), the struct_ops progs is triggered by a SEC("syscall") prog. 
Before calling this syscall prog, the st_ops map needs to be attached first. I 
think the attach part is missing also? or there is a way?