Re: [PATCH bpf-next 3/3] selftests/bpf: Fix test dynptr/test_dynptr_memset_xdp_chunks failure

[Yonghong Song <yonghong.song@linux.dev>]

On 7/25/25 4:29 PM, Martin KaFai Lau wrote:
> On 7/24/25 9:34 PM, Yonghong Song wrote:
>> For arm64 64K page size, the xdp data size was set to be more than 64K
>> in one of previous patches. This will cause failure for 
>> bpf_dynptr_memset().
>> Since the failure of bpf_dynptr_memset() is expected with 64K page size,
>> return success.
>>
>> Signed-off-by: Yonghong Song <yonghong.song@linux.dev>
>> ---
>>   tools/testing/selftests/bpf/progs/dynptr_success.c | 13 ++++++++++++-
>>   1 file changed, 12 insertions(+), 1 deletion(-)
>>
>> diff --git a/tools/testing/selftests/bpf/progs/dynptr_success.c 
>> b/tools/testing/selftests/bpf/progs/dynptr_success.c
>> index 3094a1e4ee91..8315273cb900 100644
>> --- a/tools/testing/selftests/bpf/progs/dynptr_success.c
>> +++ b/tools/testing/selftests/bpf/progs/dynptr_success.c
>> @@ -9,6 +9,8 @@
>>   #include "bpf_misc.h"
>>   #include "errno.h"
>>   +#define PAGE_SIZE_64K 65536
>> +
>>   char _license[] SEC("license") = "GPL";
>>     int pid, err, val;
>> @@ -821,8 +823,17 @@ int test_dynptr_memset_xdp_chunks(struct xdp_md 
>> *xdp)
>>       data_sz = bpf_dynptr_size(&ptr_xdp);
>>         err = bpf_dynptr_memset(&ptr_xdp, 0, data_sz, 
>> DYNPTR_MEMSET_VAL);
>> -    if (err)
>> +    if (err) {
>> +        /* bpf_dynptr_memset() eventually called bpf_xdp_pointer()
>
> I don't think I understand why the test fixed in patch 1 (e.g. 
> test_probe_read_user_dynptr) can pass the bpf_xdp_pointer test on 
> 0xffff. I thought the bpf_probe_read_user_str_dynptr will eventually 
> call the __bpf_xdp_store_bytes which also does a bpf_xdp_pointer?

For example, for test_probe_read_user_dynptr, for function test_dynptr_probe_xdp(),
for this one:
    off = xdp_near_frag_end_offset();

the off = 64928. Note that xdp_near_frag_end_offset() return value depends page size.

__u32 xdp_near_frag_end_offset(void)
{
         const __u32 headroom = 256;
         const __u32 max_frag_size =  __PAGE_SIZE - headroom - sizeof(struct skb_shared_info);
         
         /* 32 bytes before the approximate end of the fragment */
         return max_frag_size - 32;
}

The 'len' depends on 'test_len[i]' and test_len is
    __u32 test_len[7] = {0/* placeholder */, 0, 1, 2, 255, 256, 257};

In bpf_xdp_pointer, we have the following test

         if (unlikely(offset > 0xffff || len > 0xffff))
                 return ERR_PTR(-EFAULT);

In this particular case, offset = 64928, len = {0, 1, 2, 255, 256, 257}, so
it won't return -EFAULT.

For this patch 3, test_dynptr_memset_xdp_chunks, eventually we reached here:

         for (write_off = 0; write_off < size; write_off += chunk_sz) {
                 chunk_sz = min_t(u32, sizeof(buf), size - write_off);
                 err = __bpf_dynptr_write(ptr, offset + write_off, buf, chunk_sz, 0);
                 if (err)
                         return err;
         }

So the 'size' is 90000, chunk_sz is 256.
So 'offset + write_off' will be 0, 256, 512, ..., 65536
Once it reached 65536, 'offset > 0xffff' will become true since 0xffff = 65535.
and the -EFAULT will be returned.

>
>> +         * where if data_sz is greater than 0xffff, -EFAULT will be
>> +         * returned. For 64K page size, data_sz is greater than
>> +         * 64K, so error is expected and let us zero out error and
>> +         * return success.
>> +         */
>> +        if (data_sz >= PAGE_SIZE_64K)
>> +            err = 0;
>>           goto out;
>> +    }
>>         bpf_for(i, 0, max_chunks) {
>>           offset = i * sizeof(buf);
>