Package: linux-image-6.12.116+deb13-amd64 and 
Version: 6.12.116-1

Hardware
  Laptop:  ASUS Vivobook Pro N752VX
  CPU:     Intel Core i7-6700HQ @ 2.60GHz (family 6, model 94, stepping 3, 
signature 0x506e3)
  BIOS:    N752VX.303 (latest available from ASUS), UEFI boot
  Secure Boot: disabled

Software
  Distribution:    Debian 13 (trixie), amd64
  intel-microcode: 3.20251111.1~deb13u1 (revision 0xf0 for 06-5e-03)
  Kernels tested:  6.12.116+deb13-amd64 (Debian 13)
                   7.2.12+deb13-amd64 (from trixie-backports)
  Both kernels hang in the same way with the default command line.

Problem
  With early microcode loading enabled (the default), the system hangs
  very early during kernel boot. With the default command line the last
  lines on screen are from the EFI stub:

    EFI stub: Loaded initrd from LINUX_EFI_INITRD_MEDIA_GUID device path
    EFI stub: Measured initrd data into PCR9

  and nothing else is printed, even without "quiet". The machine is
  completely unresponsive (no Caps Lock LED, no VT switching); only a
  forced power-off is possible. The hang is 100% reproducible, on cold
  boot and on warm reboot.

  With "earlycon=efifb keep_bootcon" the last line printed is:

    smpboot: x86: Booting SMP configuration:

  so the hang occurs when the first secondary CPU is brought up.

  With "maxcpus=1" the system boots and the early update on the boot
  CPU succeeds:

    microcode: Current revision: 0xf0
    microcode: Updated early from: 0xba

  With "dis_ucode_ldr" the system boots normally with all CPUs (at the
  BIOS revision 0xba).

Timing dependence The hang is avoided only by serialized bring-up
  combined with an added delay. Results (all without dis_ucode_ldr,
  kernel 6.12.116 and 7.2.12 used for these tests]):

    (default)                                            -> hangs
    cpuhp.parallel=0                                     -> hangs
    cpu_init_udelay=10000                                -> hangs
    earlycon=efifb keep_bootcon                          -> hangs
    cpuhp.parallel=0 earlycon=efifb                      -> hangs
    cpuhp.parallel=0 earlycon=efifb keep_bootcon quiet   -> hangs
    cpuhp.parallel=0 earlycon=efifb keep_bootcon         -> boots
    cpuhp.parallel=0 cpu_init_udelay=10000               -> boots
    cpuhp.parallel=0 cpu_init_udelay=10000 quiet         -> boots

  In the booting cases all 8 CPUs come online with revision 0xf0. The
  slow early framebuffer console apparently acts as a delay; the
  legacy 10 ms INIT delay has the same effect, also with "quiet", so
  the delay in the secondary CPU start-up sequence is what matters,
  not the console output.This suggests a timing problem in the kernel
  during secondary CPU bring-up when a microcode update is applied on
  this platform; the microcode itself (0xf0) works once all CPUs are
  up.

BIOS history
  The same hang occurred with BIOS 204 (01/26/2016, BIOS microcode
  0x74). Updating to BIOS 303 (BIOS microcode 0xba, above the 0x80
  threshold documented for the known Skylake early-update hang) did
  not change the behaviour.

Similar reports on the same hardware family
  - ASUS N752VX, same last message, worked around with dis_ucode_ldr:
    https://forums.linuxmint.com/viewtopic.php?t=452805
  - ASUS N552V (i7-6700HQ), openSUSE Tumbleweed hangs without
    dis_ucode_ldr, while a 5.x kernel was reported to work:
   
https://forums.opensuse.org/t/opensuse-tw-wont-boot-without-the-dis-ucode-ldr-parameter/175229

Current workaround (confirmed on 6.12.116+deb13-amd64)
  quiet cpuhp.parallel=0 cpu_init_udelay=10000

-- 
Flavio Barbara
CNR-IFAC - Sesto Fiorentino (FI)

Reply via email to