Revert "x86/smp: Put CPUs into INIT on shutdown if possible"

This reverts commit 45e34c8af58f23db4474e2bfe79183efec09a18b, and the
two subsequent fixes to it:

  3f874c9b2aae ("x86/smp: Don't send INIT to non-present and non-booted CPUs")
  b1472a60a584 ("x86/smp: Don't send INIT to boot CPU")

because it seems to result in hung machines at shutdown.  Particularly
some Dell machines, but Thomas says

 "The rest seems to be Lenovo and Sony with Alderlake/Raptorlake CPUs -
  at least that's what I could figure out from the various bug reports.

  I don't know which CPUs the DELL machines have, so I can't say it's a

  I agree with the revert for now"

Ashok Raj chimes in:

 "There was a report (probably this same one), and it turns out it was a
  bug in the BIOS SMI handler.

  The client BIOS's were waiting for the lowest APICID to be the SMI
  rendevous master. If this is MeteorLake, the BSP wasn't the one with
  the lowest APIC and it triped here.

  The BIOS change is also being pushed to others for assimilation :)

  Server BIOS's had this correctly for a while now"

and it does look likely to be some bad interaction between SMI and the
non-BSP cores having put into INIT (and thus unresponsive until reset).

Acked-by: Thomas Gleixner <>
Cc: Ashok Raj <>
Signed-off-by: Linus Torvalds <>
3 files changed