Intel Revises Work On Timed I/O PPS Functionality For Linux

[phoronix]

Phoronix: Intel Revises Work On Timed I/O PPS Functionality For Linux

Back in January I wrote about Intel working on a Timed I/O driver for Linux with PPS (Pulse Per Second) to help with time synchronization between multiple devices. There hasn't been much external talk about this Timed I/O / PPS functionality while today as we approach the end of the year a new set of Linux kernel patches were posted for wiring up this Intel PPS Generator functionality...

https://www.phoronix.com/news/Intel-Timed-IO-PPS-Linux-v2

[milkylainen]

Isn't PPS typically generated by GNSS devices and similar? Driven by an atomic watch?
Also, I don't get it. Isn't the "timed I/O" description similar to what the kernel uses it's already existing hw timers for?
I can't tell the difference.

[elbar]

Writing to floppy kills Desktop still in year 2024?

Or renewed: copy from HDD to SSD still kills Desktop in year 2024?

Or renewed: copy from SATA SSD to NVME still kills Desktop in year 2024?
___________________________
some accounting
___________________________
is still IOPS a Desktop disk benchmark or EXT2 is still with its features supported(because EXT3 and newer filesystems has it not...)
___________________________
some accounting
___________________________
Windows is looking here and wishing: wining required

[halloc]

I wonder if this can replace the common need for a Parallel Port when using CNC (Computer Numerical Control) equipment (mills, lathes, etc) with modern hardware. Would still need a hardware interface of some sort.

[pkese]

Isn't PPS typically generated by GNSS devices and similar? Driven by an atomic watch?
Also, I don't get it. Isn't the "timed I/O" description similar to what the kernel uses it's already existing hw timers for?
I can't tell the difference.

I think the problem is, once you have a PPS signal on the wire, how do you get it into your PC.
There aren't any proper GPIO pins with interrupts in modern hardware and all periphery (USB, thunderbolt, etc) is just a form of packetized network communication optimized for throughput rather than small/stable/predictable latency.
It doesn't help if you have a nanosecond accurate clock pulse on the wire outside, when anything that gets that pulse into the kernel has varying latency of up to 1 millisecond.

Edit:
in this case they are solving the opposite problem of how to get pulse from kernel to the wire ... again solving the issue with nonpredictable latency.

[mb_q]

USB by design is sending 48MHz clock and expects device to sync to that, so it is possible to control drift... Given that, a device should be able to figure out its offset with an NTP-like protocol over an isochronous transfer. And a dedicated MB pin? 1Hz signal? This is super odd.

[partcyborg]

Writing to floppy kills Desktop still in year 2024?

Or renewed: copy from HDD to SSD still kills Desktop in year 2024?

Or renewed: copy from SATA SSD to NVME still kills Desktop in year 2024?
___________________________
some accounting
___________________________
is still IOPS a Desktop disk benchmark or EXT2 is still with its features supported(because EXT3 and newer filesystems has it not...)
___________________________
some accounting
___________________________
Windows is looking here and wishing: wining required

Congratulations! This just might be the least coherent post I have read on Phoronix in 2023.

[discordian]

most ieee1588 network hardware already has support for using a couple if pins for both the in- and out- part of a PPS signal.
and since its network hardware, well it works over ethernet aswell