[PATCH] rt-tests: cyclictest: try not to share the same cpu with main thread

[PATCH] rt-tests: cyclictest: try not to share the same cpu with main thread

[Yunfeng Ye]

The main thread will interfere with the test thread and try not to share
the same CPU with the main thread when the number of thread is less than
max_cpus.

Signed-off-by: yeyunfeng <yeyunfeng@huawei.com>
---
 src/cyclictest/cyclictest.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/src/cyclictest/cyclictest.c b/src/cyclictest/cyclictest.c
index 989113fb3483..b3d72caa10ce 100644
--- a/src/cyclictest/cyclictest.c
+++ b/src/cyclictest/cyclictest.c
@@ -1080,7 +1080,8 @@ static int cpu_for_thread_sp(int thread_num, int max_cpus)
 		fatal("No allowable cpus to run on\n");
 	}

-	m = thread_num % num_cpus;
+	/* just don't try to share the same cpu with main thread */
+	m = (thread_num + 1) % num_cpus;

 	/* there are num_cpus bits set, we want position of m'th one */
 	for (i = 0, cpu = 0; i < max_cpus; i++) {
-- 
1.8.3.1

Re: [PATCH] rt-tests: cyclictest: try not to share the same cpu with main thread

[John Kacur]

<

On Sat, 20 Jun 2020, Yunfeng Ye wrote:

> The main thread will interfere with the test thread and try not to share
> the same CPU with the main thread when the number of thread is less than
> max_cpus.
> 
> Signed-off-by: yeyunfeng <yeyunfeng@huawei.com>
> ---
>  src/cyclictest/cyclictest.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/src/cyclictest/cyclictest.c b/src/cyclictest/cyclictest.c
> index 989113fb3483..b3d72caa10ce 100644
> --- a/src/cyclictest/cyclictest.c
> +++ b/src/cyclictest/cyclictest.c
> @@ -1080,7 +1080,8 @@ static int cpu_for_thread_sp(int thread_num, int max_cpus)
>  		fatal("No allowable cpus to run on\n");
>  	}
> 
> -	m = thread_num % num_cpus;
> +	/* just don't try to share the same cpu with main thread */
> +	m = (thread_num + 1) % num_cpus;
> 
>  	/* there are num_cpus bits set, we want position of m'th one */
>  	for (i = 0, cpu = 0; i < max_cpus; i++) {
> -- 
> 1.8.3.1
> 
> 
I'm not sure about this one. It would work, but the purpose of the test is 
to make sure we get good real-time results despite the other threads. Plus 
we have done it this way for so long, and tools might expect things to 
work this way. I'm inclined not to take this, even though your idea is 
correct, although note that you didn't apply it everywhere it would be 
needed.

Thanks

John

Re: [PATCH] rt-tests: cyclictest: try not to share the same cpu with main thread

[Yunfeng Ye]

On 2020/7/2 0:24, John Kacur wrote:
> <
> 
> On Sat, 20 Jun 2020, Yunfeng Ye wrote:
> 
>> The main thread will interfere with the test thread and try not to share
>> the same CPU with the main thread when the number of thread is less than
>> max_cpus.
>>
>> Signed-off-by: yeyunfeng <yeyunfeng@huawei.com>
>> ---
>>  src/cyclictest/cyclictest.c | 3 ++-
>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/src/cyclictest/cyclictest.c b/src/cyclictest/cyclictest.c
>> index 989113fb3483..b3d72caa10ce 100644
>> --- a/src/cyclictest/cyclictest.c
>> +++ b/src/cyclictest/cyclictest.c
>> @@ -1080,7 +1080,8 @@ static int cpu_for_thread_sp(int thread_num, int max_cpus)
>>  		fatal("No allowable cpus to run on\n");
>>  	}
>>
>> -	m = thread_num % num_cpus;
>> +	/* just don't try to share the same cpu with main thread */
>> +	m = (thread_num + 1) % num_cpus;
>>
>>  	/* there are num_cpus bits set, we want position of m'th one */
>>  	for (i = 0, cpu = 0; i < max_cpus; i++) {
>> -- 
>> 1.8.3.1
>>
>>
> I'm not sure about this one. It would work, but the purpose of the test is 
> to make sure we get good real-time results despite the other threads. Plus 
> we have done it this way for so long, and tools might expect things to 
> work this way. I'm inclined not to take this, even though your idea is 
> correct, although note that you didn't apply it everywhere it would be 
> needed.
> 
I think the cyclictest can use not only for rt-linux, but also for gp-linux
(General Purpose Linux) with isolation (for example, isolcpus, nohz_full,
and bind irq affinity to non isolation cpus, also deploy real-time application
with many constraints). In this situation, this path will work good. thanks.

> Thanks
> 
> John
> 
>