Re: [PATCH net-next v7 0/8] fix two bugs related to page_pool

[Yunsheng Lin <linyunsheng@huawei.com>]

On 2025/1/16 1:40, Jesper Dangaard Brouer wrote:
> 
> 
> On 15/01/2025 12.33, Yunsheng Lin wrote:
>> On 2025/1/14 22:31, Jesper Dangaard Brouer wrote:
>>>
>>>
>>> On 10/01/2025 14.06, Yunsheng Lin wrote:
>>>> This patchset fix a possible time window problem for page_pool and
>>>> the dma API misuse problem as mentioned in [1], and try to avoid the
>>>> overhead of the fixing using some optimization.
>>>>
>>>>   From the below performance data, the overhead is not so obvious
>>>> due to performance variations for time_bench_page_pool01_fast_path()
>>>> and time_bench_page_pool02_ptr_ring, and there is about 20ns overhead
>>>> for time_bench_page_pool03_slow() for fixing the bug.
>>>>
>>>
>>> My benchmarking on x86_64 CPUs looks significantly different.
>>>   - CPU: Intel(R) Xeon(R) CPU E5-1650 v4 @ 3.60GHz
>>>
>>> Benchmark (bench_page_pool_simple) results from before and after patchset:
>>>
>>> | Test name  | Cycles |       |    |Nanosec |        |       |      % |
>>> | (tasklet_*)| Before | After |diff| Before |  After |  diff | change |
>>> |------------+--------+-------+----+--------+--------+-------+--------|
>>> | fast_path  |     19 |    24 |   5|  5.399 |  6.928 | 1.529 |   28.3 |
>>> | ptr_ring   |     54 |    79 |  25| 15.090 | 21.976 | 6.886 |   45.6 |
>>> | slow       |    238 |   299 |  61| 66.134 | 83.298 |17.164 |   26.0 |
>>> #+TBLFM: $4=$3-$2::$7=$6-$5::$8=(($7/$5)*100);%.1f
>>>
>>> My above testing show a clear performance regressions across three
>>> different page_pool operating modes.
>>
>> I retested it on arm64 server patch by patch as the raw performance
>> data in the attachment, it seems the result seemed similar as before.
>>
>> Before this patchset:
>>              fast_path              ptr_ring            slow
>> 1.         31.171 ns               60.980 ns          164.917 ns
>> 2.         28.824 ns               60.891 ns          170.241 ns
>> 3.         14.236 ns               60.583 ns          164.355 ns
>>
>> With patch 1-4:
>> 4.         31.443 ns               53.242 ns          210.148 ns
>> 5.         31.406 ns               53.270 ns          210.189 ns
>>
>> With patch 1-5:
>> 6.         26.163 ns               53.781 ns          189.450 ns
>> 7.         26.189 ns               53.798 ns          189.466 ns
>>
>> With patch 1-8:
>> 8.         28.108 ns               68.199 ns          202.516 ns
>> 9.         16.128 ns               55.904 ns          202.711 ns
>>
>> I am not able to get hold of a x86 server yet, I might be able
>> to get one during weekend.
>>
>> Theoretically, patch 1-4 or 1-5 should not have much performance
>> impact for fast_path and ptr_ring except for the rcu_lock mentioned
>> in page_pool_napi_local(), so it would be good if patch 1-5 is also
>> tested in your testlab with the rcu_lock removing in
>> page_pool_napi_local().
>>
> 
> What are you saying?
>  - (1) test patch 1-5
>  - or (2) test patch 1-5 but revert patch 2 with page_pool_napi_local()

patch 1-5 with below applied.

--- a/net/core/page_pool.c
+++ b/net/core/page_pool.c
@@ -1207,10 +1207,8 @@ static bool page_pool_napi_local(const struct page_pool *pool)
        /* Synchronizated with page_pool_destory() to avoid use-after-free
         * for 'napi'.
         */
-       rcu_read_lock();
        napi = READ_ONCE(pool->p.napi);
        napi_local = napi && READ_ONCE(napi->list_owner) == cpuid;
-       rcu_read_unlock();

        return napi_local;
 }