[PATCH v2 04/16] test/bbdev: add timeout for latency tests

[Maxime Coquelin]

On 2/28/23 23:37, Chautru, Nicolas wrote:
> Hi Maxime,
> 
>> -----Original Message-----
>> From: Maxime Coquelin <maxime.coquelin at redhat.com>
>> Sent: Monday, February 27, 2023 1:45 AM
>> To: Vargas, Hernan <hernan.vargas at intel.com>; dev at dpdk.org;
>> gakhil at marvell.com; Rix, Tom <trix at redhat.com>
>> Cc: Chautru, Nicolas <nicolas.chautru at intel.com>; Zhang, Qi Z
>> <qi.z.zhang at intel.com>
>> Subject: Re: [PATCH v2 04/16] test/bbdev: add timeout for latency tests
>>
>>
>>
>> On 2/24/23 17:59, Vargas, Hernan wrote:
>>> Hi Maxime,
>>>
>>>> -----Original Message-----
>>>> From: Maxime Coquelin <maxime.coquelin at redhat.com>
>>>> Sent: Thursday, February 23, 2023 2:32 AM
>>>> To: Vargas, Hernan <hernan.vargas at intel.com>; dev at dpdk.org;
>>>> gakhil at marvell.com; Rix, Tom <trix at redhat.com>
>>>> Cc: Chautru, Nicolas <nicolas.chautru at intel.com>; Zhang, Qi Z
>>>> <qi.z.zhang at intel.com>
>>>> Subject: Re: [PATCH v2 04/16] test/bbdev: add timeout for latency
>>>> tests
>>>>
>>>>
>>>>
>>>> On 2/22/23 22:13, Vargas, Hernan wrote:
>>>>> Hi Maxime,
>>>>>
>>>>>> -----Original Message-----
>>>>>> From: Maxime Coquelin <maxime.coquelin at redhat.com>
>>>>>> Sent: Monday, February 20, 2023 10:33 AM
>>>>>> To: Vargas, Hernan <hernan.vargas at intel.com>; dev at dpdk.org;
>>>>>> gakhil at marvell.com; Rix, Tom <trix at redhat.com>
>>>>>> Cc: Chautru, Nicolas <nicolas.chautru at intel.com>; Zhang, Qi Z
>>>>>> <qi.z.zhang at intel.com>
>>>>>> Subject: Re: [PATCH v2 04/16] test/bbdev: add timeout for latency
>>>>>> tests
>>>>>>
>>>>>>
>>>>>>
>>>>>> On 2/15/23 18:09, Hernan Vargas wrote:
>>>>>>> Add a timeout to force exit the latency tests in case dequeue
>>>>>>> never happens.
>>>>>>>
>>>>>>> Signed-off-by: Hernan Vargas <hernan.vargas at intel.com>
>>>>>>> ---
>>>>>>>      app/test-bbdev/test_bbdev_perf.c | 24 +++++++++++++++++++-----
>>>>>>>      1 file changed, 19 insertions(+), 5 deletions(-)
>>>>>>>
>>>>>>> diff --git a/app/test-bbdev/test_bbdev_perf.c
>>>>>>> b/app/test-bbdev/test_bbdev_perf.c
>>>>>>> index 19b9a5b119..dede0f900e 100644
>>>>>>> --- a/app/test-bbdev/test_bbdev_perf.c
>>>>>>> +++ b/app/test-bbdev/test_bbdev_perf.c
>>>>>>> @@ -26,6 +26,7 @@
>>>>>>>
>>>>>>>      #define MAX_QUEUES RTE_MAX_LCORE
>>>>>>>      #define TEST_REPETITIONS 100
>>>>>>> +#define TIME_OUT_POLL 1e8
>>>>>>>      #define WAIT_OFFLOAD_US 1000
>>>>>>>
>>>>>>>      #ifdef RTE_BASEBAND_FPGA_LTE_FEC @@ -4546,6 +4547,7 @@
>>>>>>> latency_test_ldpc_dec(struct rte_mempool *mempool,
>>>>>>>
>>>>>>>      	for (i = 0, dequeued = 0; dequeued < num_to_process; ++i) {
>>>>>>>      		uint16_t enq = 0, deq = 0;
>>>>>>> +		uint32_t time_out = 0;
>>>>>>>      		bool first_time = true;
>>>>>>>      		last_time = 0;
>>>>>>>
>>>>>>> @@ -4597,7 +4599,8 @@ latency_test_ldpc_dec(struct rte_mempool
>>>>>> *mempool,
>>>>>>>      				last_time = rte_rdtsc_precise() -
>> start_time;
>>>>>>>      				first_time = false;
>>>>>>>      			}
>>>>>>> -		} while (unlikely(burst_sz != deq));
>>>>>>> +			time_out++;
>>>>>>> +		} while ((burst_sz != deq) && (time_out < TIME_OUT_POLL));
>>>>>>>
>>>>>>>      		*max_time = RTE_MAX(*max_time, last_time);
>>>>>>>      		*min_time = RTE_MIN(*min_time, last_time); @@ -
>> 4606,7
>>>>>> +4609,12 @@
>>>>>>> latency_test_ldpc_dec(struct rte_mempool *mempool,
>>>>>>>      		if (extDdr)
>>>>>>>      			retrieve_harq_ddr(dev_id, queue_id, ops_enq,
>>>>>> burst_sz);
>>>>>>>
>>>>>>> -		if (test_vector.op_type != RTE_BBDEV_OP_NONE) {
>>>>>>> +		if (burst_sz != deq) {
>>>>>>> +			struct rte_bbdev_info info;
>>>>>>> +			ret = TEST_FAILED;
>>>>>>> +			rte_bbdev_info_get(dev_id, &info);
>>>>>>
>>>>>> What is the point of calling rte_bbdev_info_get() here and below?
>>>>>> info is not used afterwards.
>>>>>
>>>>> The reason for calling this function is to check the device status
>>>>> and if there
>>>> is something wrong the PMD would display it to standard output.
>>>>
>>>> What kind of info exactly, I don't see much meaningful logs in
>>>> rte_bbdev_info_get() except printing error if dev_info == NULL.
>>>
>>> rte_bbdev_info_get() calls the device's info_get function that is specified in
>> the PMD.
>>> For example, for ACC100, acc100_dev_info_get() gets called to check the
>> device status.
>>
>> Ok, I looked at this function and it might be more relevant to mention
>> vrb1 than acc100, because acc100 does not support device status:
>>
>> 	/* Check the status of device */
>> 	dev_info->device_status = RTE_BBDEV_DEV_NOT_SUPPORTED;
>>
>> Also, this dev_info->device_status field is set by all the drivers but never read
>> (neither by the driver, nor library nor apps).
>>
>> So if this is the only reason rte_bbdev_info_get() is called, that is quite useless.
>> Am I missing someting?
> 
> Note the device status is printed in the bbdev-test app in that other patchset in the series https://patches.dpdk.org/project/dpdk/patch/20230215170949.60569-7-hernan.vargas@intel.com/

Not related to this patch, but maybe you should use an assert in the
patch you link above. Indeed, what is the point of continuing the init
if the device status is "FATAL_ERR"?

> Also calling that function then exercise interaction with underlying HW and PF driver which is valuable and the expected usecase.
> This is indeed relevant from vrb1 in term of current  intel PMD implementation, but the bbdev-test is HW agnostic.
> Let us know if unclear,

OK, that's clearer.

For this patch:
Reviewed-by: Maxime Coquelin <maxime.coquelin at redhat.com>

Thanks,
Maxime

> Thanks,
> Nic
> 
> 
>>
>> Thanks,
>> Maxime
>>
>>> Thanks,
>>> Hernan
>>>>
>>>> Regards,
>>>> Maxime
>>>>
>>>>>>
>>>>>>> +			TEST_ASSERT_SUCCESS(ret, "Dequeue timeout!");
>>>>>>> +		} else if (test_vector.op_type != RTE_BBDEV_OP_NONE) {
>>>>>>>      			ret = validate_ldpc_dec_op(ops_deq, burst_sz,
>>>>>> ref_op,
>>>>>>>      					vector_mask);
>>>>>>>      			TEST_ASSERT_SUCCESS(ret, "Validation
>> failed!"); @@
>>>>>> -4632,6
>>>>>>> +4640,7 @@ latency_test_enc(struct rte_mempool *mempool,
>>>>>>>
>>>>>>>      	for (i = 0, dequeued = 0; dequeued < num_to_process; ++i) {
>>>>>>>      		uint16_t enq = 0, deq = 0;
>>>>>>> +		uint32_t time_out = 0;
>>>>>>>      		bool first_time = true;
>>>>>>>      		last_time = 0;
>>>>>>>
>>>>>>> @@ -4667,13 +4676,18 @@ latency_test_enc(struct rte_mempool
>>>>>> *mempool,
>>>>>>>      				last_time += rte_rdtsc_precise() -
>> start_time;
>>>>>>>      				first_time = false;
>>>>>>>      			}
>>>>>>> -		} while (unlikely(burst_sz != deq));
>>>>>>> +			time_out++;
>>>>>>> +		} while ((burst_sz != deq) && (time_out < TIME_OUT_POLL));
>>>>>>>
>>>>>>>      		*max_time = RTE_MAX(*max_time, last_time);
>>>>>>>      		*min_time = RTE_MIN(*min_time, last_time);
>>>>>>>      		*total_time += last_time;
>>>>>>> -
>>>>>>> -		if (test_vector.op_type != RTE_BBDEV_OP_NONE) {
>>>>>>> +		if (burst_sz != deq) {
>>>>>>> +			struct rte_bbdev_info info;
>>>>>>> +			ret = TEST_FAILED;
>>>>>>> +			rte_bbdev_info_get(dev_id, &info);
>>>>>>
>>>>>> Same here.
>>>>>>
>>>>>>> +			TEST_ASSERT_SUCCESS(ret, "Dequeue timeout!");
>>>>>>> +		} else if (test_vector.op_type != RTE_BBDEV_OP_NONE) {
>>>>>>>      			ret = validate_enc_op(ops_deq, burst_sz,
>> ref_op);
>>>>>>>      			TEST_ASSERT_SUCCESS(ret, "Validation
>> failed!");
>>>>>>>      		}
>>>>>
>>>
>