Page MenuHomeFreeBSD
Differential D58563

apex: add a driver for the Google Apex Edge TPU
Needs ReviewPublic
Actions

Authored by seuros on Jul 30 2026, 8:12 PM.
Tags
Referenced Files
F173332368: D58563.id183133.diff
Fri, Sep 25, 7:09 AM
F173240757: D58563.id186612.diff
Thu, Sep 24, 2:54 PM
Unknown Object (File)
Thu, Sep 24, 6:37 AM
Unknown Object (File)
Sun, Sep 20, 9:33 AM
Unknown Object (File)
Sat, Sep 19, 6:46 PM
Unknown Object (File)
Sun, Sep 13, 1:53 AM
Unknown Object (File)
Sat, Sep 12, 9:00 PM
Unknown Object (File)
Fri, Sep 11, 2:55 AM
View All 63 Files
Subscribers
imp
ivy
ziaee

Details

Reviewers
adrian
ziaee
Group Reviewers
drivers
manpages
Summary

The Apex is the accelerator on Coral PCI Express and M.2 modules,
appearing as 1ac1:089a
Google's Linux driver left staging in 5.16 and the out of tree
repository was archived; no BSD has had one.

Tested on: amd64, Coral M.2 Accelerator, running MobileNet v1
with the TensorFlow Lite 2.16 and LiteRT runtimes

Diff Detail

Repository
rG FreeBSD src repository
Lint
Lint Passed
Unit
No Test Coverage
Build Status
Buildable 75304
Build 72187: arc lint + arc unit

Event Timeline

seuros created this revision.Jul 30 2026, 8:12 PM
Herald added subscribers: ivy, ziaee, imp. · View Herald TranscriptJul 30 2026, 8:12 PM
seuros requested review of this revision.Jul 30 2026, 8:12 PM
Harbormaster completed remote builds in B75304: Diff 183133.Jul 30 2026, 8:12 PM
seuros added a comment.Jul 30 2026, 8:18 PM

The USB variant is more common than the PCIE/M2 in the wild, but i did not want to port it without testing it on hardware.

I will check with some frigate or home-assistant friends if they have they can undust the coral .

PS: the linux driver has a bug that will never be fixed (the repo is archived) which pushed people in stopped the coral usage. this port address it.

seuros added a parent revision: D58561: gasket: add a framework for Google accelerator ASICs.Jul 30 2026, 8:40 PM
seuros updated this revision to Diff 183142.Jul 30 2026, 10:36 PM

fix docs

Harbormaster completed remote builds in B75310: Diff 183142.Jul 30 2026, 10:36 PM
adrian added a comment.Sat, Sep 12, 8:41 PM

in general good; just update your email address!

sys/dev/apex/apex.c
60

Maybe create a #define for how big this can get instead of '3' and use that for bounds checking the array?

ziaee requested changes to this revision.Sun, Sep 13, 12:19 AM
ziaee added a reviewer: manpages.

looks cool; please also relnotes: yes for any new driver

share/man/man4/apex.4
14–27

can also add tunables and sysctls if you want, see share/examples/mdoc/example.4

44

see share/man/man5/style.mdoc.5

This revision now requires changes to proceed.Sun, Sep 13, 12:19 AM
seuros updated this revision to Diff 186603.Sun, Sep 13, 2:49 PM

refactor/doc/split driver

Harbormaster completed remote builds in B76843: Diff 186603.Sun, Sep 13, 2:49 PM
seuros updated this revision to Diff 186612.Sun, Sep 13, 5:18 PM

rebase on main

Harbormaster completed remote builds in B76851: Diff 186612.Sun, Sep 13, 5:18 PM

Revision Contents
Changeset List

PathSize
etc/
mtree/
2 lines
include/
5 lines
share/
man/
man4/
1 line
104 lines
sys/
amd64/
conf/
6 lines
conf/
1 line
dev/
apex/
1177 lines
39 lines
145 lines
modules/
2 lines
apex/
8 lines

Diff 183133

View Options

etc/mtree/BSD.include.dist

Loading...
View Options

include/Makefile

Loading...
View Options

share/man/man4/Makefile

Loading...
View Options

share/man/man4/apex.4

Loading...
View Options

sys/amd64/conf/NOTES

Loading...
View Options

sys/conf/files.amd64

Loading...
View Options

sys/dev/apex/apex.c

Loading...
View Options

sys/dev/apex/apex_ioctl.h

Loading...
View Options

sys/dev/apex/apexreg.h

Loading...
View Options

sys/modules/Makefile

Loading...
View Options

sys/modules/apex/Makefile

Loading...