mps(4): add support for hostboot
Needs ReviewPublic
Actions

Authored by allanjude on Sun, Sep 20, 1:52 AM.

Details

Reviewers

imp
adrian
kevans
slm
ziaee

Summary

If the mps/mpr(4) device has corrupted or missing firmware then
it will fail to attach at boot.

Add a new /dev/mp{s,r}_hostboot device, and a corresponding
mpsutil hostboot command, to reset the device and boot
from firmware loaded from host memory.

Also add an interface to use firmware(9) to load the firmware at boot.

Loosely based on https://github.com/marcan/lsirec/

Assisted by: Claude Code (Sonnet 5)
Sponsored by: Klara, Inc.

Diff Detail

Repository

rG FreeBSD src repository

Lint

Lint Skipped

Unit

Tests Skipped

Build Status

Buildable 77102
Build 73985: arc lint + arc unit

Event Timeline

allanjude created this revision.Sun, Sep 20, 1:52 AM

Herald added a reviewer: slm. · View Herald TranscriptSun, Sep 20, 1:52 AM

Herald added a subscriber: ziaee. · View Herald Transcript

allanjude requested review of this revision.Sun, Sep 20, 1:52 AM

Harbormaster completed remote builds in B77102: Diff 187256.Sun, Sep 20, 1:52 AM

allanjude added inline comments.Sun, Sep 20, 1:54 AM

sys/dev/mpr/mpr.c
979	I've never done this kind of driver stuff (DMA) before, so this needs extra careful review.

allanjude added a reviewer: ziaee.Sun, Sep 20, 1:55 AM

imp added inline comments.Mon, Sep 21, 2:32 AM

sys/dev/mpr/mpr_user.c
272	silly question. If there's no driver attached, then how do we open a device to all this ioctl? How could you possibly recover a device that mpr doesn't attach to? Why do you need that code to allocate resources? You might want to explain the exact mechanism how this comes to pass. the missing bit from the comments is that that mpr associate with the device because probe succeeds. But the bigger issue, I'd think, is that if probe succeeds and attach doesn't, then I think there'd be no devsw registered...

allanjude added inline comments.Mon, Sep 21, 11:33 PM

sys/dev/mpr/mpr_user.c
272	Indeed, probe succeeds, and mpr_attach() fails, because the device stays in the MPI2_IOC_STATE_RESET state, because it has no firmware. So we end up with no /dev/mprN device. So, this patch adds a new /dev/mpr_hostboot device, that isn't associated with any hardware, and will just "always exist". Then `mprutil` can IOCTL to it, and ask it "could you please boot pci0:7:0:0 with this firmware (pointer + length)" And then you can `devctl attach pci0:7:0:0` and suddenly you have /dev/mpr0 that you didn't before. The primary use case is actually to preload the firmware in loader.conf and stop the initial attach from failing, be having the card booted to said firmware so it doesn't fail to attach. This is some oddball hardware I bought, a single card that is a PCI-e switch, a Mellanox ConnectX-3, and 2x LSI SAS2308 chips (basically 9305-8e), all on one card. When making this custom card, they did not include any flash storage for the firmware. see: https://forums.servethehome.com/index.php?threads/unique-oem-dual-lsi-2308-mellanox-connectx-3-vpi-combo-expansion-cards.30532/ If you'd like one to play with, they are only $29.99: https://www.ebay.com/itm/188115617273?itmmeta=01M2V5GZ247M7TP012W8M9FMF2&hash=item2bcc9085f9:g:YPcAAOSwpI5eCaDK