delphij: as the approver for secteam of D6888 which introduced the code and logic implemented in this driver, are you comfortable to approve this one patch here, which does not introduce functional changes to the callout mode?

Schedule the initial harvesting as soon as one second after the RNG is started, which should give it plenty of time to generate the first random bytes.

Remove IRQ lines from our own rpi[2].dts as suggested.

Cosmetic change: rename callback argument from "param" to "arg" as per callout_reset documentation.

Fix indentation (use tabs instead of spaces) of compat_data per style(9)

Fix indentation of compat_data in RNG driver.

Fixed previous patch which misused the FDT_PLATFORM_DEF2 macro.

Use two FDT_PLATFORM_DEF2 directives instead of changing the FDT_PLATFORM_DEF one to allow compatibility with previous compatibility strings.

This breaks the kernel build if WITHOUT_ZFS is defined in src.conf.

Same here, good port!

Fixed version has been discussed in https://reviews.freebsd.org/D11475 and committed to head. I think we can close this one.

I'm so glad I cc'ed you on this PR, you rock. I shall be able to confirm my armv6 kernel no longer panic with this fix within the next 24h. THANKS!

What I am not sure is why it broke only 32-bit architectures?

Please see PR 220454. This code generates kernel panics.