crypto decryption support in test-crypto-perf pmd-cycle-count test

crypto decryption support in test-crypto-perf pmd-cycle-count test

[Suanming Mou]

[-- Attachment #1: Type: text/plain, Size: 472 bytes --]

Hi Guys,

Sorry for the inconvenience.
I have one small question with pmd-cycle-count test in test-crypto-perf application.
As in pmd-cycle-count run code, I don't see the cperf_mbuf_set() function is called to fill the test data.
So for decryption, I assume it will always fails. Feel like pmd-cycle-count test does not expect to support decryption. Or is it a bug?

I have no idea with the application design background, is it expected?

Thanks,
Suanming Mou

[-- Attachment #2: Type: text/html, Size: 2479 bytes --]

RE: crypto decryption support in test-crypto-perf pmd-cycle-count test

[Akhil Goyal]

[-- Attachment #1: Type: text/plain, Size: 1148 bytes --]

Hi Suanming,

The pmd-cycle-count mode is for counting the cycles for enqueue and dequeue.
It does not matter what is the content of the packet, as we are not validating the content
after dequeue of the packet. PMD/Hw will assume it as plain/encrypted payload and
will process it as per the session attached.
Hence, doing memcpy to mbuf is not required.

Regards,
Akhil

From: Suanming Mou <suanmingm@nvidia.com>
Sent: Friday, July 21, 2023 4:06 PM
To: ciara.power@intel.com; Akhil Goyal <gakhil@marvell.com>
Cc: dev@dpdk.org
Subject: [EXT] crypto decryption support in test-crypto-perf pmd-cycle-count test

External Email
________________________________
Hi Guys,

Sorry for the inconvenience.
I have one small question with pmd-cycle-count test in test-crypto-perf application.
As in pmd-cycle-count run code, I don't see the cperf_mbuf_set() function is called to fill the test data.
So for decryption, I assume it will always fails. Feel like pmd-cycle-count test does not expect to support decryption. Or is it a bug?

I have no idea with the application design background, is it expected?

Thanks,
Suanming Mou

[-- Attachment #2: Type: text/html, Size: 4282 bytes --]

RE: crypto decryption support in test-crypto-perf pmd-cycle-count test

[Suanming Mou]

[-- Attachment #1: Type: text/plain, Size: 1714 bytes --]

Thanks Akhil.
Yes, I see.
But for decryption, we should prepare the correct cipher data, right? Or that test does not make any sense.
I assume decryption with incorrect data will cause PMD reject work properly.

From: Akhil Goyal <gakhil@marvell.com>
Sent: Friday, July 21, 2023 7:49 PM
To: Suanming Mou <suanmingm@nvidia.com>; ciara.power@intel.com
Cc: dev@dpdk.org
Subject: RE: crypto decryption support in test-crypto-perf pmd-cycle-count test

Hi Suanming,

The pmd-cycle-count mode is for counting the cycles for enqueue and dequeue.
It does not matter what is the content of the packet, as we are not validating the content
after dequeue of the packet. PMD/Hw will assume it as plain/encrypted payload and
will process it as per the session attached.
Hence, doing memcpy to mbuf is not required.

Regards,
Akhil

From: Suanming Mou <suanmingm@nvidia.com<mailto:suanmingm@nvidia.com>>
Sent: Friday, July 21, 2023 4:06 PM
To: ciara.power@intel.com<mailto:ciara.power@intel.com>; Akhil Goyal <gakhil@marvell.com<mailto:gakhil@marvell.com>>
Cc: dev@dpdk.org<mailto:dev@dpdk.org>
Subject: [EXT] crypto decryption support in test-crypto-perf pmd-cycle-count test

External Email
________________________________
Hi Guys,

Sorry for the inconvenience.
I have one small question with pmd-cycle-count test in test-crypto-perf application.
As in pmd-cycle-count run code, I don't see the cperf_mbuf_set() function is called to fill the test data.
So for decryption, I assume it will always fails. Feel like pmd-cycle-count test does not expect to support decryption. Or is it a bug?

I have no idea with the application design background, is it expected?

Thanks,
Suanming Mou

[-- Attachment #2: Type: text/html, Size: 5772 bytes --]

RE: crypto decryption support in test-crypto-perf pmd-cycle-count test

[Akhil Goyal]

[-- Attachment #1: Type: text/plain, Size: 2317 bytes --]

Yes, for auth cases, the PMD may give errors. However, cipher only cases would work without issues.
You can send a patch to fix this. But make sure not to add that inside the time keeping loop.

From: Suanming Mou <suanmingm@nvidia.com>
Sent: Friday, July 21, 2023 5:23 PM
To: Akhil Goyal <gakhil@marvell.com>; ciara.power@intel.com
Cc: dev@dpdk.org
Subject: [EXT] RE: crypto decryption support in test-crypto-perf pmd-cycle-count test

External Email
________________________________
Thanks Akhil.
Yes, I see.
But for decryption, we should prepare the correct cipher data, right? Or that test does not make any sense.
I assume decryption with incorrect data will cause PMD reject work properly.

From: Akhil Goyal <gakhil@marvell.com<mailto:gakhil@marvell.com>>
Sent: Friday, July 21, 2023 7:49 PM
To: Suanming Mou <suanmingm@nvidia.com<mailto:suanmingm@nvidia.com>>; ciara.power@intel.com<mailto:ciara.power@intel.com>
Cc: dev@dpdk.org<mailto:dev@dpdk.org>
Subject: RE: crypto decryption support in test-crypto-perf pmd-cycle-count test

Hi Suanming,

The pmd-cycle-count mode is for counting the cycles for enqueue and dequeue.
It does not matter what is the content of the packet, as we are not validating the content
after dequeue of the packet. PMD/Hw will assume it as plain/encrypted payload and
will process it as per the session attached.
Hence, doing memcpy to mbuf is not required.

Regards,
Akhil

From: Suanming Mou <suanmingm@nvidia.com<mailto:suanmingm@nvidia.com>>
Sent: Friday, July 21, 2023 4:06 PM
To: ciara.power@intel.com<mailto:ciara.power@intel.com>; Akhil Goyal <gakhil@marvell.com<mailto:gakhil@marvell.com>>
Cc: dev@dpdk.org<mailto:dev@dpdk.org>
Subject: [EXT] crypto decryption support in test-crypto-perf pmd-cycle-count test

External Email
________________________________
Hi Guys,

Sorry for the inconvenience.
I have one small question with pmd-cycle-count test in test-crypto-perf application.
As in pmd-cycle-count run code, I don't see the cperf_mbuf_set() function is called to fill the test data.
So for decryption, I assume it will always fails. Feel like pmd-cycle-count test does not expect to support decryption. Or is it a bug?

I have no idea with the application design background, is it expected?

Thanks,
Suanming Mou

[-- Attachment #2: Type: text/html, Size: 7064 bytes --]

RE: crypto decryption support in test-crypto-perf pmd-cycle-count test

[Suanming Mou]

[-- Attachment #1: Type: text/plain, Size: 2925 bytes --]

Sure, in fact I also find throughput testing with decryption and in-place-mode will override the cipher data in the single mbuf as well. So I assume out-of-place mode is valid in that case.
I will try to deliver some patches when have effort.  Thank you!

From: Akhil Goyal <gakhil@marvell.com>
Sent: Friday, July 21, 2023 8:04 PM
To: Suanming Mou <suanmingm@nvidia.com>; ciara.power@intel.com
Cc: dev@dpdk.org
Subject: RE: crypto decryption support in test-crypto-perf pmd-cycle-count test

Yes, for auth cases, the PMD may give errors. However, cipher only cases would work without issues.
You can send a patch to fix this. But make sure not to add that inside the time keeping loop.

From: Suanming Mou <suanmingm@nvidia.com<mailto:suanmingm@nvidia.com>>
Sent: Friday, July 21, 2023 5:23 PM
To: Akhil Goyal <gakhil@marvell.com<mailto:gakhil@marvell.com>>; ciara.power@intel.com<mailto:ciara.power@intel.com>
Cc: dev@dpdk.org<mailto:dev@dpdk.org>
Subject: [EXT] RE: crypto decryption support in test-crypto-perf pmd-cycle-count test

External Email
________________________________
Thanks Akhil.
Yes, I see.
But for decryption, we should prepare the correct cipher data, right? Or that test does not make any sense.
I assume decryption with incorrect data will cause PMD reject work properly.

From: Akhil Goyal <gakhil@marvell.com<mailto:gakhil@marvell.com>>
Sent: Friday, July 21, 2023 7:49 PM
To: Suanming Mou <suanmingm@nvidia.com<mailto:suanmingm@nvidia.com>>; ciara.power@intel.com<mailto:ciara.power@intel.com>
Cc: dev@dpdk.org<mailto:dev@dpdk.org>
Subject: RE: crypto decryption support in test-crypto-perf pmd-cycle-count test

Hi Suanming,

The pmd-cycle-count mode is for counting the cycles for enqueue and dequeue.
It does not matter what is the content of the packet, as we are not validating the content
after dequeue of the packet. PMD/Hw will assume it as plain/encrypted payload and
will process it as per the session attached.
Hence, doing memcpy to mbuf is not required.

Regards,
Akhil

From: Suanming Mou <suanmingm@nvidia.com<mailto:suanmingm@nvidia.com>>
Sent: Friday, July 21, 2023 4:06 PM
To: ciara.power@intel.com<mailto:ciara.power@intel.com>; Akhil Goyal <gakhil@marvell.com<mailto:gakhil@marvell.com>>
Cc: dev@dpdk.org<mailto:dev@dpdk.org>
Subject: [EXT] crypto decryption support in test-crypto-perf pmd-cycle-count test

External Email
________________________________
Hi Guys,

Sorry for the inconvenience.
I have one small question with pmd-cycle-count test in test-crypto-perf application.
As in pmd-cycle-count run code, I don't see the cperf_mbuf_set() function is called to fill the test data.
So for decryption, I assume it will always fails. Feel like pmd-cycle-count test does not expect to support decryption. Or is it a bug?

I have no idea with the application design background, is it expected?

Thanks,
Suanming Mou

[-- Attachment #2: Type: text/html, Size: 8430 bytes --]