By default the MBM and MBA tests use the "fill_buf" benchmark to
read from a buffer with the goal to measure the memory bandwidth
generated by this buffer access.

Care should be taken when sizing the buffer used by the "fill_buf"
benchmark. If the buffer is small enough to fit in the cache then
it cannot be expected that the benchmark will generate much memory
bandwidth. For example, on a system with 320MB L3 cache the existing
hardcoded default of 250MB is insufficient.

Use the measured cache size to determine a buffer size that can be
expected to trigger memory access while keeping the existing default
as minimum that has been appropriate for testing so far.

Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
---
 tools/testing/selftests/resctrl/mba_test.c | 8 +++++++-
 tools/testing/selftests/resctrl/mbm_test.c | 8 +++++++-
 2 files changed, 14 insertions(+), 2 deletions(-)

diff --git a/tools/testing/selftests/resctrl/mba_test.c b/tools/testing/selftests/resctrl/mba_test.c
index 8ad433495f61..cad473b81a64 100644
--- a/tools/testing/selftests/resctrl/mba_test.c
+++ b/tools/testing/selftests/resctrl/mba_test.c
@@ -170,11 +170,17 @@ static int mba_run_test(const struct resctrl_test *test, const struct user_param
 		.setup		= mba_setup,
 		.measure	= mba_measure,
 	};
+	unsigned long cache_total_size = 0;
 	int ret;
 
 	remove(RESULT_FILE_NAME);
 
-	param.fill_buf.buf_size = DEFAULT_SPAN;
+	ret = get_cache_size(uparams->cpu, "L3", &cache_total_size);
+	if (ret)
+		return ret;
+
+	param.fill_buf.buf_size = cache_total_size > DEFAULT_SPAN ?
+				  cache_total_size * 2 : DEFAULT_SPAN;
 	param.fill_buf.memflush = 1;
 	param.fill_buf.operation = 0;
 	param.fill_buf.once = false;
diff --git a/tools/testing/selftests/resctrl/mbm_test.c b/tools/testing/selftests/resctrl/mbm_test.c
index b6883f274c74..734bfa4f42b3 100644
--- a/tools/testing/selftests/resctrl/mbm_test.c
+++ b/tools/testing/selftests/resctrl/mbm_test.c
@@ -138,11 +138,17 @@ static int mbm_run_test(const struct resctrl_test *test, const struct user_param
 		.setup		= mbm_setup,
 		.measure	= mbm_measure,
 	};
+	unsigned long cache_total_size = 0;
 	int ret;
 
 	remove(RESULT_FILE_NAME);
 
-	param.fill_buf.buf_size = DEFAULT_SPAN;
+	ret = get_cache_size(uparams->cpu, "L3", &cache_total_size);
+	if (ret)
+		return ret;
+
+	param.fill_buf.buf_size = cache_total_size > DEFAULT_SPAN ?
+				  cache_total_size * 2 : DEFAULT_SPAN;
 	param.fill_buf.memflush = 1;
 	param.fill_buf.operation = 0;
 	param.fill_buf.once = false;
-- 
2.46.0

On Thu, 29 Aug 2024, Reinette Chatre wrote:

> By default the MBM and MBA tests use the "fill_buf" benchmark to
> read from a buffer with the goal to measure the memory bandwidth
> generated by this buffer access.
> 
> Care should be taken when sizing the buffer used by the "fill_buf"
> benchmark. If the buffer is small enough to fit in the cache then
> it cannot be expected that the benchmark will generate much memory
> bandwidth. For example, on a system with 320MB L3 cache the existing
> hardcoded default of 250MB is insufficient.
> 
> Use the measured cache size to determine a buffer size that can be
> expected to trigger memory access while keeping the existing default
> as minimum that has been appropriate for testing so far.
> 
> Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
> ---
>  tools/testing/selftests/resctrl/mba_test.c | 8 +++++++-
>  tools/testing/selftests/resctrl/mbm_test.c | 8 +++++++-
>  2 files changed, 14 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/testing/selftests/resctrl/mba_test.c b/tools/testing/selftests/resctrl/mba_test.c
> index 8ad433495f61..cad473b81a64 100644
> --- a/tools/testing/selftests/resctrl/mba_test.c
> +++ b/tools/testing/selftests/resctrl/mba_test.c
> @@ -170,11 +170,17 @@ static int mba_run_test(const struct resctrl_test *test, const struct user_param
>  		.setup		= mba_setup,
>  		.measure	= mba_measure,
>  	};
> +	unsigned long cache_total_size = 0;
>  	int ret;
>  
>  	remove(RESULT_FILE_NAME);
>  
> -	param.fill_buf.buf_size = DEFAULT_SPAN;
> +	ret = get_cache_size(uparams->cpu, "L3", &cache_total_size);
> +	if (ret)
> +		return ret;
> +
> +	param.fill_buf.buf_size = cache_total_size > DEFAULT_SPAN ?
> +				  cache_total_size * 2 : DEFAULT_SPAN;

Should the check leave a bit of safeguard so that the buf_size is at 
least 2x (or x1.25 or some other factor)?

In here buf_size immediate jumps from 1x -> 2x when cache_total_size goes 
from DEFAULT_SPAN to DEFAULT_SPAN+1 (obviously L3 size won't be odd like 
that but I think you get my point).

Also, user might want to override this as mentioned in my reply to the 
previous patch.

-- 
 i.

>  	param.fill_buf.memflush = 1;
>  	param.fill_buf.operation = 0;
>  	param.fill_buf.once = false;
> diff --git a/tools/testing/selftests/resctrl/mbm_test.c b/tools/testing/selftests/resctrl/mbm_test.c
> index b6883f274c74..734bfa4f42b3 100644
> --- a/tools/testing/selftests/resctrl/mbm_test.c
> +++ b/tools/testing/selftests/resctrl/mbm_test.c
> @@ -138,11 +138,17 @@ static int mbm_run_test(const struct resctrl_test *test, const struct user_param
>  		.setup		= mbm_setup,
>  		.measure	= mbm_measure,
>  	};
> +	unsigned long cache_total_size = 0;
>  	int ret;
>  
>  	remove(RESULT_FILE_NAME);
>  
> -	param.fill_buf.buf_size = DEFAULT_SPAN;
> +	ret = get_cache_size(uparams->cpu, "L3", &cache_total_size);
> +	if (ret)
> +		return ret;
> +
> +	param.fill_buf.buf_size = cache_total_size > DEFAULT_SPAN ?
> +				  cache_total_size * 2 : DEFAULT_SPAN;
>  	param.fill_buf.memflush = 1;
>  	param.fill_buf.operation = 0;
>  	param.fill_buf.once = false;
>

Hi Ilpo,

On 8/30/24 4:25 AM, Ilpo Järvinen wrote:
> On Thu, 29 Aug 2024, Reinette Chatre wrote:
> 
>> By default the MBM and MBA tests use the "fill_buf" benchmark to
>> read from a buffer with the goal to measure the memory bandwidth
>> generated by this buffer access.
>>
>> Care should be taken when sizing the buffer used by the "fill_buf"
>> benchmark. If the buffer is small enough to fit in the cache then
>> it cannot be expected that the benchmark will generate much memory
>> bandwidth. For example, on a system with 320MB L3 cache the existing
>> hardcoded default of 250MB is insufficient.
>>
>> Use the measured cache size to determine a buffer size that can be
>> expected to trigger memory access while keeping the existing default
>> as minimum that has been appropriate for testing so far.
>>
>> Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
>> ---
>>   tools/testing/selftests/resctrl/mba_test.c | 8 +++++++-
>>   tools/testing/selftests/resctrl/mbm_test.c | 8 +++++++-
>>   2 files changed, 14 insertions(+), 2 deletions(-)
>>
>> diff --git a/tools/testing/selftests/resctrl/mba_test.c b/tools/testing/selftests/resctrl/mba_test.c
>> index 8ad433495f61..cad473b81a64 100644
>> --- a/tools/testing/selftests/resctrl/mba_test.c
>> +++ b/tools/testing/selftests/resctrl/mba_test.c
>> @@ -170,11 +170,17 @@ static int mba_run_test(const struct resctrl_test *test, const struct user_param
>>   		.setup		= mba_setup,
>>   		.measure	= mba_measure,
>>   	};
>> +	unsigned long cache_total_size = 0;
>>   	int ret;
>>   
>>   	remove(RESULT_FILE_NAME);
>>   
>> -	param.fill_buf.buf_size = DEFAULT_SPAN;
>> +	ret = get_cache_size(uparams->cpu, "L3", &cache_total_size);
>> +	if (ret)
>> +		return ret;
>> +
>> +	param.fill_buf.buf_size = cache_total_size > DEFAULT_SPAN ?
>> +				  cache_total_size * 2 : DEFAULT_SPAN;
> 
> Should the check leave a bit of safeguard so that the buf_size is at
> least 2x (or x1.25 or some other factor)?
> 
> In here buf_size immediate jumps from 1x -> 2x when cache_total_size goes
> from DEFAULT_SPAN to DEFAULT_SPAN+1 (obviously L3 size won't be odd like
> that but I think you get my point).

Good catch. Will fix.

> 
> Also, user might want to override this as mentioned in my reply to the
> previous patch.
> 

ack.

Reinette