Re: [PATCH 3/3 -mm] kexec based hibernation -v6: kexec hibernate/resume

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Tuesday, 20 of November 2007, Huang, Ying wrote:
> On Mon, 2007-11-19 at 19:22 +0100, Rafael J. Wysocki wrote:
> > > +#ifdef CONFIG_KEXEC
> > > +static void kexec_hibernate_power_down(void)
> > > +{
> > > +	switch (hibernation_mode) {
> > > +	case HIBERNATION_TEST:
> > > +	case HIBERNATION_TESTPROC:
> > > +		break;
> > > +	case HIBERNATION_REBOOT:
> > > +		machine_restart(NULL);
> > > +		break;
> > > +	case HIBERNATION_PLATFORM:
> > > +		if (!hibernation_ops)
> > > +			break;
> > > +		hibernation_ops->enter();
> > 
> > hibernation_platform_enter() should be used here (as of the current mainline).
> 
> The power_down will be called with interrupt disabled, device suspended,
> non-boot CPU disabled. But the latest hibernate_platform_enter calls the
> device_suspend, disable_nonboot_cpus etc function. So, I use
> hibernation_ops->enter() directly instead of
> hibernation_platform_enter().

Hm, you need to call device_power_down(PMSG_SUSPEND) before
hibernation_ops->enter().

Also, all of the ACPI global calls need to be carried out before that and the
devices should be suspended rather than shut down in that case.

That's why hibernation_platform_enter() has been introduced, BTW.

> > > +		/* We should never get here */
> > > +		while (1);
> > > +		break;
> > > +	case HIBERNATION_SHUTDOWN:
> > > +		machine_power_off();
> > > +		break;
> > > +	}
> > > +	machine_halt();
> > > +	/*
> > > +	 * Valid image is on the disk, if we continue we risk serious data
> > > +	 * corruption after resume.
> > > +	 */
> > > +	printk(KERN_CRIT "Please power me down manually\n");
> > > +	while (1);
> > > +}
> > 
> > Hm, what's the difference between the above function and power_down(),
> > actually?
> 
> Same as above.

I see, but I'm not sure it's correct.

> > > +
> > > +int kexec_hibernate(struct kimage *image)
> > > +{
> > > +	int error;
> > > +	int platform_mode = (hibernation_mode == HIBERNATION_PLATFORM);
> > > +	unsigned long cmd_ret;
> > > +
> > > +	mutex_lock(&pm_mutex);
> > > +
> > > +	pm_prepare_console();
> > > +	suspend_console();
> > > +
> > > +	error = pm_notifier_call_chain(PM_HIBERNATION_PREPARE);
> > > +	if (error)
> > > +		goto Resume_console;
> > > +
> > > +	error = platform_start(platform_mode);
> > > +	if (error)
> > > +		goto Resume_console;
> > > +
> > > +	error = device_suspend(PMSG_FREEZE);
> > > +	if (error)
> > > +		goto Resume_console;
> > > +
> > > +	error = platform_pre_snapshot(platform_mode);
> > > +	if (error)
> > > +		goto Resume_devices;
> > > +
> > > +	error = disable_nonboot_cpus();
> > > +	if (error)
> > > +		goto Resume_devices;
> > 
> > I wonder if it's viable to merge the above with hibernate() and
> > hibernation_snapshot() somehow, to avoid code duplication?
> 
> Yes. Most code are duplicated. But there is one advantage not to merge
> them: power_down can called with IRQ disabled to make it possible to
> eliminate the freezer.

Please leave the freezer alone for now.  We'll clean up all that when we decide
to remove it.

> I think it is possible to merge the two implementation. I will try to do
> it.

Yes, please use as much of the existing code as reasonably possible, so that we
don't need to duplicate changes in the future.

> > Apart from the above, there's some new debug code to be added to disk.c
> > in 2.6.25.  It's in the ACPI test tree right now and you can get it as
> > individual patches from:
> > http://www.sisk.pl/kernel/hibernation_and_suspend/2.6.24-rc2/patches/
> > (patches 10-12).
> > 
> > Please base your changes on top of that.
> 
> OK, I will do it.

Thanks.

There will be some more cosmetic changes in this area, probably before 2.6.25,
so please stay tuned.

Greetings,
Rafael