AM625: TRNG SAMPLE_CYCLES value

Let me look into your question, and get back to you next week as we'll have holiday in US.
Best,
-Hong

Hi Hong,

Any updates to this?

Thanks and regards,
Koichiro Tashiro

Sorry for my delayed response.
"2.10.4 TRNG Clock Configuration for First Random Value Generation"
"The [11-8] SAMPLE_DIV, [31-16] SAMPLE_CYCLES, [7-6] SCALE and [5-0] NOISE_BLOCKS bit fields in the
register determine the number of clocks needed to generate the first random value.
...
The run time for a single output block is always the same: SAMPLE_CYCLES * (4SCALE) * NOISE_BLOCKS *
512 * trng_clock_period * (SAMPLE_DIV + 1)."

It is not necessary to configure SAMPLE_CYCLES = "0".
For example, on AM62x HS-FS and HS-SE, TIFS initialize TRNG with SAMPLE_CYCLES = "5".
Best,
-Hong

Hi Hong,

Thanks for your reply.

Hong Guan64 said:

It is not necessary to configure SAMPLE_CYCLES = "0".

So the old answer "Typically these should be set to the normal operational settings (0)" was misleading.

Hong Guan64 said:

For example, on AM62x HS-FS and HS-SE, TIFS initialize TRNG with SAMPLE_CYCLES = "5".

What is the reason why "5" was chosen?
The customer is asking how to select SAMPLE_CYCLES value.

And in case SAMPLE_CYCLES is changed from the value used in SDK, what should be done?  

Koichiro Tashiro said:

Is it OK just overwrite that value to zero and use TRNG?
Or 
TRNG initialization sequence needs to be done again?

Thanks and regards,
Koichiro Tashiro

I'd recommend referring to
2.10.4 TRNG Clock Configuration for First Random Value Generation
2.10.3 TRNG Operation Sequences
Best,
-Hong

Hi Hong,

Hong Guan64 said:

I'd recommend referring to
2.10.4 TRNG Clock Configuration for First Random Value Generation
2.10.3 TRNG Operation Sequences

I checked these TRM sections, but it does not seem answer to my questions.
I understood changing SAMPLE_CYCLES affects the number of FRO samples that are XOR-ed.
Then the run time for a single output block varies depending on SAMPLE_SYCLES and other configuraiton.
But it is not clearly explained *how* SAMPLE_CYCLES value is chosen.

Thanks and regards,
Koichiro Tashiro

Refer to <Table 5-199. TRNG_CONFIG Register Field Descriptions> on SAMPLE_CYCLES
"31-16 SAMPLE_CYCLES R/W 0h
This field, together with the SCALE field, sets the number of FRO
samples that are XOR-ed together into one bit to be shifted into the
main shift register. This value must be such that there is at least one
bit of entropy (in total) in each 8 bits that are shifted. This field can
only be modified while ENABLE_TRNG in TRNG_CONTROL is ‘0’
or when either of the TEST_KNOW_NOISE or TEST_COND_FUNC
bits in TRNG_TEST is ‘1’. Value 0 in this field selects 65536 * 4SCALE
FRO samples to be XOR-ed together."

Best,
-Hong

Hi Hong,

What does below description exactly mean?

Hong Guan64 said:

This value must be such that there is at least one
bit of entropy (in total) in each 8 bits that are shifted.

Thanks and regards,
Koichiro Tashiro

Here're some reference on RNG/Entropy
https://www.cryptoquantique.com/blog/what-is-a-random-number/
https://lwn.net/Articles/884875/
Best,
-Hong

Hi Hong,

Thanks for the reference. I checked them.
But honestly speaking, I still cannot understand how SAMPLE_CYCLES and Entropy are related.
Could you explain below?
- Why '5' is used in our SDK?
- What is difference using '5' and '4' (or '3')?

Thanks and regards,
Koichiro Tashiro

Hi Hong,

Could you reply to this one?

Thanks and regards,
Koichiro Tashiro

SAMPLE_CYCLES value is re-configurable in SW as necessary, for example,
- OPTEE
https://git.ti.com/cgit/optee/ti-optee-os/tree/core/arch/arm/plat-k3/drivers/sa2ul_rng.c?h=4.0.0#n134
- Linux kernel driver
https://git.ti.com/cgit/ti-linux-kernel/ti-linux-kernel/tree/drivers/char/hw_random/omap-rng.c?h=10.00.08#n267
Best,
-Hong

Hong,

I understood SAMPLE_CYCLES value is re-configurable.
The customer is asking how the value for SAMPLE_CYCLES is selected.
TRM does not provide that information or it is not easy to understand.

Thanks and regards,
Koichiro Tashiro

Hello Tashiro-san

These selected values have been there on several of devices - customer is free to change these values as they are configurable. Older OMAP4 uses a value here of 0x21 and AM6x uses 0x5, both should work fine.

It is a tradeoff between speed and "entropy". The value of 5 we derived from example calculations given in the internal IP specifications. In general anything more than 4 will work (at least 4 FROs running before automatic shutdown, at 1 bit per clock, worst case half-rate), so we select the lowest(fastest) safe value. If your customer is concerned about the amount of "entropy", and speed is not a concern, they can use the default of 0, which actually sets a value of 65536.