Hi, On Wed, Jun 15, 2016 at 01:25:53PM +0000, Usyskin, Alexander wrote: > It may be that fdd9b8655933 uncovered the bug in iTCO_wdt because > it actually removes iAMT watchdog and new implementation does not > creates one for not provisioned ME. > > So before that patch there was two watchdog devices and systemd > maybe worked with iAMT one. After that patch only one watchdog > exists in the system - iTCO and system works with it. Yes, looks like /dev/watchdog was iAMT watchdog on older kernels: 4.5: Jun 15 01:26:36 earth systemd[1]: Hardware watchdog 'INTCAMT', version 0 Jun 15 01:26:36 earth systemd[1]: Failed to set timeout to 30s: Invalid argument 4.6: Jun 15 01:24:49 earth systemd[1]: Hardware watchdog 'iTCO_wdt', version 0 Jun 15 01:24:49 earth systemd[1]: Set hardware watchdog to 30s. Thanks for you help. So I guess there are two bugs, which are both unrelated to iAMT watchdog, but uncovered by this change: 1. systemd's watchdog logic seems to be broken after suspend/resume, so that the watchdog times out. It must be systemd's fault, since wd_keepalive does not trigger the bug. Previous kernels worked, since operations on INTCAMT were basically no-ops without provisioned ME. 2. iTCO_wdt hangs the system instead of restarting it. -- Sebastian
Attachment:
signature.asc
Description: PGP signature