Re: [PATCH v2] random: handle archrandom in plural words

From: Michael Ellerman
Date: Tue Jul 19 2022 - 23:03:34 EST


Gabriel Paubert <paubert@xxxxxxx> writes:
> On Mon, Jul 18, 2022 at 04:31:11PM +1000, Michael Ellerman wrote:
>> "Jason A. Donenfeld" <Jason@xxxxxxxxx> writes:
>> > The archrandom interface was originally designed for x86, which supplies
>> > RDRAND/RDSEED for receiving random words into registers, resulting in
>> > one function to generate an int and another to generate a long. However,
>> > other architectures don't follow this.
>> >
>> > On arm64, the SMCCC TRNG interface can return between 1 and 3 words. On
>> > s390, the CPACF TRNG interface can return between 1 and 32 words for the
>> > same cost as for one word. On UML, the os_getrandom() interface can return
>> > arbitrary amounts.
>> >
>> > So change the api signature to take a "words" parameter designating the
>> > maximum number of words requested, and then return the number of words
>> > generated.
>>
>> On powerpc a word is 32-bits and a doubleword is 64-bits (at least
>> according to the ISA). I think that's also true on other 64-bit
>> architectures.
>
> IIRC, this is (or was) not the case on Alpha, where word was defined as
> 16 bits. All assembly mnemonics had w for 16 bits, l for 32 bits, and q
> for 64 bits.

Yeah I should have said on *some* other 64-bit arches.

Seems to be a common feature/hack on arches that have evolved over time,
or been inspired by earlier arches.

The latest Power ISA has octwords :)

cheers