a bit of a random question... is there a way to tie a device instance and an smf service in a contract so a fma failure of the device would fault the corresponding service?

i think i can use ddi_fm_service_impact() in the driver, but then associating the device with the smf service.. the contract documentation is rather... abstract

jbk: I believe the process in the SMF service instance could itself obtain a contract on the device to monitor, and exit with the error code that puts the service into maintenance (or possibly use libscf to put itself into maintenance) if it hits the right conditions

I've not ever used the device contract stuff myself

But I gather it's supposed to give a user process the kind of interaction you'd be able to get when in the kernel using LDI callbacks etc

danmcd: maybe the hack that was in omnios bloody to enable/disable the new NVME naming could be brought back? Having issues? set the entry in /etc/system, reboot, pick old pi/be, reboot ?

sure its an extra step to rollback but you shouldnt need the correct intermedia be/pi?

If anyone has some time to think about locales, I'd appreciate another pair of eyes in the conclusion I wrote up in illumos.org/issues/15234.

→ BUG 15234: strxfrm() segfaults for Vietnamese locale (vi_VN.UTF-8) and character 168 (\xA8 / \250) (New)

jclulow: doing more digging, it looks like the SES topo module is doing that sort of monitoring (obviously not in an SMF service, but that's not the unclear part)... so I might use that as a guide

sjorge - OmniOS has that in place in r151044, but the new devid types will come in with r151046 (and are already in bloody). The reason it was delayed was to give people a large easy rollback window if they need to go back to an old BE for any reason.

[illumos-gate] 15173 mdb iob_addr2str clobbers memory -- Robert Mustacchi <rm⊙fo>