On Tuesday, March 08, 2016 04:25:30 PM Chen, Yu C wrote:

quoted

-----Original Message-----
From: linux-pm-owner@vger.kernel.org [mailto:linux-pm-
owner@vger.kernel.org] On Behalf Of Rafael J. Wysocki
Sent: Tuesday, March 08, 2016 9:54 AM
To: Chen, Yu C
Cc: Rafael J. Wysocki; ACPI Devel Maling List; x86@kernel.org; linux-
efi@vger.kernel.org; Linux Kernel Mailing List; linux-pm@vger.kernel.org;
Len Brown; Matt Fleming; Thomas Gleixner; Ingo Molnar; H. Peter Anvin;
Zhang, Rui
Subject: Re: [PATCH][RFC v3] ACPI / PM: Fix poweroff issue on HW-full
platforms without _S5

On Monday, March 07, 2016 03:53:13 PM Chen, Yu C wrote:

quoted

Hi Rafael,
(resend for broken content)

quoted

-----Original Message-----
From: rjwysocki@gmail.com [mailto:rjwysocki@gmail.com] On Behalf Of
Rafael J. Wysocki
Sent: Monday, March 07, 2016 9:19 PM
To: Chen, Yu C
Cc: ACPI Devel Maling List; x86@kernel.org;
linux-efi@vger.kernel.org; Linux Kernel Mailing List;
linux-pm@vger.kernel.org; Rafael J. Wysocki; Len Brown; Matt
Fleming; Thomas Gleixner; Ingo Molnar; H. Peter Anvin; Zhang, Rui
Subject: Re: [PATCH][RFC v3] ACPI / PM: Fix poweroff issue on
HW-full platforms without _S5

[cut]

quoted

quoted

 bool efi_poweroff_required(void)
 {
-       return !!acpi_gbl_reduced_hardware;
+       return acpi_gbl_reduced_hardware || (acpi_no_s5 &&
+ !pm_power_off);

What if CONFIG_ACPI is not set here?

If CONFIG_ACPI is not set, this file would not be compiled, because
CONFIG_EFI depends on CONFIG_ACPI.

OK

So the next question will be if efi_poweroff_required() is guaranteed to run
after all of the other code that may register alternative power off handling.

Hum. unfortunately it is not guaranteed to run after all of the other code,
because other components who register pm_power_off may be built as modules, and
we can not predict/control the sequence registration.   So this patch may
break the EFI platforms who use non-efi poweroff due to unstable EFI service
,  not sure if there are any released-products of this kind.

Currently I'm thinking of 3 possible solutions,  could you please give some advices on them:

1. Introduce bootopt of 'poweroff=efi'
     Set the pm_power_off to efi_power_off no matter whether there is _S5 or not

2. Introduce /sys/power/poweroff
    Allow the user to choose which  pm_power_off, for example:
 
# cat /sys/power/poweroff
*acpi		acpi_power_off
efi		efi_power_off	
gpio		gpio_poweroff_do_poweroff
user can echo string to enable which one.

And two APIs:
register_power_off(char *name, power_off func)
unregister_power_off(char *name)  


3. replace all the codes of  pm_power_off() with reliable_pm_power_off()

void reliable_pm_power_off(void)
{
	if (!pm_power_off) {
		if (acpi_no_s5)
			pm_power_off = efi_power_off;
	/* Other conditions added in the future. */
	}
	pm_power_off();
}

What about something like adding something like default_power_off that would
be used by pm_power_off if nothing else is available?

Thanks,
Rafael