CPU load is increased by 6% when switched CMAC operations to SECO HSM

[Moheb_Basta]

Hello Community,

We are using SECO MB_4.2.2_IMX8_RTM_4.0.0_CRYPTO_HSM_4.3.0_RTM_1.0.0_D2103

SCFW 1.7.4 and SECOFW 3.8.2

We were executing CMAC operations on a SW crypto driver, but we have a requirement to switch the CMAC operation to use the SECO HSM.

To do so we have moved the crypto buffers to a non-cacheable shared MPU region and assigned the crypto Jobs to the SECO HSM.

Moving the buffers to a non-cacheable shared MPU region increased the CPU load by 5%.

And assigning the crypto Jobs to the SECO HSM increased the CPU load by 1% further.

so in total, we have a 6% CPU load increase compared to the SW solution, while we were expecting the CPU load to actually decrease as we have offloaded the CMAC operations from the M4 to the HW accelerator SECO HSM.

Do you have any proposal on how to decrease the CPU load in this case?

Thanks for your support.

[igorpadykov]

Hi Moheb

below answer from internal team:

------------------

The possible reason is that CMAC crypto cost very little resource, but the overhead of loading SECO HSM is much higher.

If you execute CMAC operation based on larger size of data, it will occupy more resource so that the impact of loading SECO HSM could be not that obvious.

------------------

Best regards
igor