• Eric W. Biederman's avatar
    [PATCH] Don't attempt to power off if power off is not implemented · 5e38291d
    Eric W. Biederman authored
    The problem.  It is expected that /sbin/halt -p works exactly like
    /sbin/halt, when the kernel does not implement power off functionality.
    The kernel can do a lot of work in the reboot notifiers and in
    device_shutdown before we even get to machine_power_off.  Some of that
    shutdown is not safe if you are leaving the power on, and it definitely
    gets in the way of using sysrq or pressing ctrl-alt-del.  Since the
    shutdown happens in generic code there is no way to fix this in
    architecture specific code :(
    Some machines are kernel oopsing today because of this.
    The simple solution is to turn LINUX_REBOOT_CMD_POWER_OFF into
    LINUX_REBOOT_CMD_HALT if power_off functionality is not implemented.
    This has the unfortunate side effect of disabling the power off
    functionality on architectures that leave pm_power_off to null and still
    implement something in machine_power_off.  And it will break the build on
    some architectures that don't have a pm_power_off variable at all.
    On both counts I say tough.
    For architectures like alpha that don't implement the pm_power_off variable
    pm_power_off is declared in linux/pm.h and it is a generic part of our
    power management code, and all architectures should implement it.
    For architectures like parisc that have a default power off method in
    machine_power_off if pm_power_off is not implemented or fails.  It is easy
    enough to set the pm_power_off variable.  And nothing bad happens there,
    the machines just stop powering off.
    The current semantics are impossible without a flag at the top level so we
    can avoid the problem code if a power off is not implemented.  pm_power_off
    is as good a flag as any with the bonus that it works without modification
    on at least x86, x86_64, powerpc, and ppc today.
    Andrew can you pick this up and put this in the mm tree.  Kernels that
    don't compile or don't power off seem saner than kernels that oops or
    panic.  Until we get the arch specific patches for the problem
    architectures this probably isn't smart to push into the stable kernel.
    Unfortunately I don't have the time at the moment to walk through every
    architecture and make them work.  And even if I did I couldn't test it :(
    From: Hirokazu Takata <takata@linux-m32r.org>
        Add pm_power_off() for build fix of arch/m32r/kernel/process.c.
    From: Miklos Szeredi <miklos@szeredi.hu>
        UML build fix
    Signed-off-by: default avatarEric W. Biederman <ebiederm@xmission.com>
    Signed-off-by: default avatarHayato Fujiwara <fujiwara@linux-m32r.org>
    Signed-off-by: default avatarHirokazu Takata <takata@linux-m32r.org>
    Signed-off-by: default avatarMiklos Szeredi <miklos@szeredi.hu>
    Signed-off-by: default avatarAndrew Morton <akpm@osdl.org>
    Signed-off-by: default avatarLinus Torvalds <torvalds@osdl.org>