On Fri, 2024-07-05 at 10:12 +0200, Peter Hilber wrote:

On 03.07.24 12:40, David Woodhouse wrote:

[...]

quoted

This is what I currently have for 'struct vmclock_abi' that I'd like to
persuade you to adopt. I need to tweak it some more, for at least the
following reasons, as well as any more you can see:

 • size isn't big enough for 64KiB pages
 • Should be explicitly little-endian
 • Does it need esterror as well as maxerror?

I have no opinion about this. I can drop esterror if unwanted.

I also don't care. I'm just observing the inconsistency.

quoted

 • Why is maxerror in picoseconds? It's the only use of that unit

Between us we now have picoseconds, nanoseconds, (seconds >> 64) and
(seconds >> 64+n).

The power-of-two fractions seem to make a lot of sense for the counter
period, because they mean we don't have to perform divisions.

Does it makes sense to harmonise on (seconds >> 64) for all of the
fractional seconds? Again I don't have a strong opinion; I only want us
to have a *reason* for any differences that exist.

quoted

 • Where do the clock_status values come from? Do they make sense?
 • Are signed integers OK? (I think so!).

Signed integers would need to be introduced to Virtio, which so far only
uses explicitly unsigned types: u8, le16 etc.

Perhaps. Although it would also be possible (if not ideal) to define
that e.g. the tai_offset field is a 16-bit "unsigned" integer according
to virtio, but to be interpreted as follows:

If the number is <= 32767 then the TAI offset is that value, but if the
number is >= 32768 then the TAI offset is that value minus 65536.

Perhaps not pretty, but there isn't a *fundamental* dependency on
virtio supporting signed integers as a primary type.