Skip to contentSkip to conversation

x86/reboot: Avoid EFI reboot when not running on EFI

PATCH In review Rob Bradford 3 participants 5 messages Aug 29, 2019 – Sep 13, 2019 x86

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86/reboot: Avoid EFI reboot when not running on EFI

Replace the check using efi_runtime_disabled() which only checks if EFI runtime was disabled on the kernel command line with a call to efi_enabled(EFI_RUNTIME_SERVICES) to check if EFI runtime services are available.

In the situation where the kernel was booted without an EFI environment then only efi_enabled(EFI_RUNTIME_SERVICES) correctly represents that no EFI is available. Setting "noefi" or "efi=noruntime" on the commandline continue to have the same effect as efi_enabled(EFI_RUNTIME_SERVICES) will return false.

---
 arch/x86/kernel/reboot.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kernel/reboot.c +1 -1
index 09d6bded3c1e..0b0a7fccdb00 100644
@@ -500,7 +500,7 @@ static int __init reboot_init(void)
 	 */
 	rv = dmi_check_system(reboot_dmi_table);
 
-	if (!rv && efi_reboot_required() && !efi_runtime_disabled())
+	if (!rv && efi_reboot_required() && efi_enabled(EFI_RUNTIME_SERVICES))
 		reboot_type = BOOT_EFI;
 
 	return 0;
Signature · 1 line
2.21.0
#1 lore raw

On Thu, 29 Aug 2019, Rob Bradford wrote:

CC+ Ard

Quoted 22 lines

Replace the check using efi_runtime_disabled() which only checks if EFI runtime was disabled on the kernel command line with a call to efi_enabled(EFI_RUNTIME_SERVICES) to check if EFI runtime services are available.

In the situation where the kernel was booted without an EFI environment then only efi_enabled(EFI_RUNTIME_SERVICES) correctly represents that no EFI is available. Setting "noefi" or "efi=noruntime" on the commandline continue to have the same effect as efi_enabled(EFI_RUNTIME_SERVICES) will return false.

---
 arch/x86/kernel/reboot.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kernel/reboot.c +0 -0
index 09d6bded3c1e..0b0a7fccdb00 100644

arch/x86/kernel/reboot.c static int __init reboot_init(void) · line 503

Quoted 5 lines
arch/x86/kernel/reboot.c +1 -1
@@ -500,7 +500,7 @@ static int __init reboot_init(void)
 	 */
 	rv = dmi_check_system(reboot_dmi_table);
 
-	if (!rv && efi_reboot_required() && !efi_runtime_disabled())
+	if (!rv && efi_reboot_required() && efi_enabled(EFI_RUNTIME_SERVICES))

Why is efi_reboot_required() set at all in that situation?

Thanks,

tglx

#2 lore raw
Quoted 35 lines · Thomas Gleixner

On Thu, 29 Aug 2019, Rob Bradford wrote:

CC+ Ard

Quoted 31 lines

Replace the check using efi_runtime_disabled() which only checks if EFI
runtime was disabled on the kernel command line with a call to efi_enabled(EFI_RUNTIME_SERVICES) to check if EFI runtime services are
available.

In the situation where the kernel was booted without an EFI environment
then only efi_enabled(EFI_RUNTIME_SERVICES) correctly represents that no
EFI is available. Setting "noefi" or "efi=noruntime" on the commandline
continue to have the same effect as efi_enabled(EFI_RUNTIME_SERVICES)
will return false.

---
 arch/x86/kernel/reboot.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kernel/reboot.c +0 -0
index 09d6bded3c1e..0b0a7fccdb00 100644
@@ -500,7 +500,7 @@ static int __init reboot_init(void)
 	 */
 	rv = dmi_check_system(reboot_dmi_table);
 
Quoted 6 lines

arch/x86/kernel/reboot.c static int __init reboot_init(void) · line 504

Quoted 3 lines
arch/x86/kernel/reboot.c +1 -1
-	if (!rv && efi_reboot_required() && !efi_runtime_disabled())
+	if (!rv && efi_reboot_required() &&
 efi_enabled(EFI_RUNTIME_SERVICES))

Why is efi_reboot_required() set at all in that situation?

Hi Thomas,

This platform uses HW Reduced ACPI (
https://github.com/intel/cloud-hypervisor) but also supports direct kernel boot without EFI.

/*
* For most modern platforms the preferred method of powering off is
via
* ACPI. However, there are some that are known to require the use of
* EFI runtime services and for which ACPI does not work at all. *
* Using EFI is a last resort, to be used only if no other option
* exists.
*/
bool efi_reboot_required(void)

{
	if (!acpi_gbl_reduced_hardware)
		return false;

	efi_reboot_quirk_mode = EFI_RESET_WARM;
	return true;
}

This is a hardware workaround that assumes that all ACPI HW reduced platforms do not support ACPI reset:

commit 44be28e9dd9880dca3e2cbf7a844f2114e67f2cb

Author: Matt Fleming <[email protected]>
Date:   Fri Jun 13 12:39:55 2014 +0100

x86/reboot: Add EFI reboot quirk for ACPI Hardware Reduced flag

It appears that the BayTrail-T class of hardware requires EFI in
order
to powerdown and reboot and no other reliable method exists.

This quirk is generally applicable to all hardware that has the
ACPI
Hardware Reduced bit set, since usually ACPI would be the preferred method.

Cc Len Brown <[email protected]>
Cc Mark Salter <[email protected]>
Cc "Rafael J. Wysocki" <[email protected]>

I'm reluctant to change the behaviour of efi_reboot_required() as it might break older hardware whereas checking if we have an EFI runtime is safer.

Rob

#3 lore raw
Quoted 87 lines · Robert
Quoted 37 lines · Thomas Gleixner

On Thu, 29 Aug 2019, Rob Bradford wrote:

CC+ Ard

Quoted 33 lines

Replace the check using efi_runtime_disabled() which only checks if
EFI
runtime was disabled on the kernel command line with a call to efi_enabled(EFI_RUNTIME_SERVICES) to check if EFI runtime services
are
available.

In the situation where the kernel was booted without an EFI environment
then only efi_enabled(EFI_RUNTIME_SERVICES) correctly represents that no
EFI is available. Setting "noefi" or "efi=noruntime" on the commandline
continue to have the same effect as efi_enabled(EFI_RUNTIME_SERVICES)
will return false.

---
 arch/x86/kernel/reboot.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kernel/reboot.c +0 -0
index 09d6bded3c1e..0b0a7fccdb00 100644
@@ -500,7 +500,7 @@ static int __init reboot_init(void)
 	 */
 	rv = dmi_check_system(reboot_dmi_table);
 
Quoted 6 lines

arch/x86/kernel/reboot.c static int __init reboot_init(void) · line 504

Quoted 3 lines
arch/x86/kernel/reboot.c +1 -1
-	if (!rv && efi_reboot_required() && !efi_runtime_disabled())
+	if (!rv && efi_reboot_required() &&
 efi_enabled(EFI_RUNTIME_SERVICES))

Why is efi_reboot_required() set at all in that situation?

Hi Thomas,

This platform uses HW Reduced ACPI (
https://github.com/intel/cloud-hypervisor) but also supports direct kernel boot without EFI.

/*
* For most modern platforms the preferred method of powering off is
via
* ACPI. However, there are some that are known to require the use of
* EFI runtime services and for which ACPI does not work at all. *
* Using EFI is a last resort, to be used only if no other option
* exists.
*/
bool efi_reboot_required(void)

{
	if (!acpi_gbl_reduced_hardware)
		return false;

	efi_reboot_quirk_mode = EFI_RESET_WARM;
	return true;
}

This is a hardware workaround that assumes that all ACPI HW reduced platforms do not support ACPI reset:

commit 44be28e9dd9880dca3e2cbf7a844f2114e67f2cb

Author: Matt Fleming <[email protected]>
Date:   Fri Jun 13 12:39:55 2014 +0100

x86/reboot: Add EFI reboot quirk for ACPI Hardware Reduced flag

It appears that the BayTrail-T class of hardware requires EFI in
order
to powerdown and reboot and no other reliable method exists.

This quirk is generally applicable to all hardware that has the
ACPI
Hardware Reduced bit set, since usually ACPI would be the
preferred
method.

Quoted 9 lines
Cc Len Brown <[email protected]>
Cc Mark Salter <[email protected]>
Cc "Rafael J. Wysocki" <[email protected]>

I'm reluctant to change the behaviour of efi_reboot_required() as it might break older hardware whereas checking if we have an EFI runtime is safer.

Hi Thomas,

Any update on your thoughts on this patch?

Rob

#4 lore raw
In reply to Bradford, Robert
Quoted 37 lines · Robert
Quoted 35 lines · Thomas Gleixner

On Thu, 29 Aug 2019, Rob Bradford wrote:

CC+ Ard

Quoted 31 lines

Replace the check using efi_runtime_disabled() which only checks if EFI
runtime was disabled on the kernel command line with a call to efi_enabled(EFI_RUNTIME_SERVICES) to check if EFI runtime services are
available.

In the situation where the kernel was booted without an EFI environment
then only efi_enabled(EFI_RUNTIME_SERVICES) correctly represents that no
EFI is available. Setting "noefi" or "efi=noruntime" on the commandline
continue to have the same effect as efi_enabled(EFI_RUNTIME_SERVICES)
will return false.

---
 arch/x86/kernel/reboot.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kernel/reboot.c +0 -0
index 09d6bded3c1e..0b0a7fccdb00 100644
@@ -500,7 +500,7 @@ static int __init reboot_init(void)
     */
    rv = dmi_check_system(reboot_dmi_table);
 
Quoted 3 lines
Quoted 3 lines

arch/x86/kernel/reboot.c static int __init reboot_init(void) · line 504

Quoted 3 lines
arch/x86/kernel/reboot.c +1 -1
-   if (!rv && efi_reboot_required() && !efi_runtime_disabled())
+   if (!rv && efi_reboot_required() &&
 efi_enabled(EFI_RUNTIME_SERVICES))

This change by itself seems correct to me - efi_runtime_disabled() only gives you whether EFI runtime services were disabled explicitly command line rather than whether they are available at all.

Quoted 27 lines
Quoted 3 lines

Why is efi_reboot_required() set at all in that situation?

Hi Thomas,

This platform uses HW Reduced ACPI (
https://github.com/intel/cloud-hypervisor) but also supports direct kernel boot without EFI.

/*
* For most modern platforms the preferred method of powering off is
via
* ACPI. However, there are some that are known to require the use of
* EFI runtime services and for which ACPI does not work at all. *
* Using EFI is a last resort, to be used only if no other option
* exists.
*/
bool efi_reboot_required(void)

{
        if (!acpi_gbl_reduced_hardware)
                return false;

        efi_reboot_quirk_mode = EFI_RESET_WARM;
        return true;
}
Quoted 3 lines

This is a hardware workaround that assumes that all ACPI HW reduced platforms do not support ACPI reset:

I'm not that familiar with the situation on x86, but I did discuss this with the MS Windows engineers back in April, and their policy is to strongly prefer ACPI reboot, and so that is most likely to work on arbitrary x86 hardware which was built to run Windows.

So I do think the patch below is overly broad, given that the ACPI spec is unambiguous about the fact the FADT reset is also covered by H/W reduced ACPI.

However, I do share Rob's reservation given how quirky x86 hardware can be in this respect.

TL;DR:

Quoted 25 lines

commit 44be28e9dd9880dca3e2cbf7a844f2114e67f2cb

Author: Matt Fleming <[email protected]>
Date:   Fri Jun 13 12:39:55 2014 +0100

x86/reboot: Add EFI reboot quirk for ACPI Hardware Reduced flag

It appears that the BayTrail-T class of hardware requires EFI in
order
to powerdown and reboot and no other reliable method exists.

This quirk is generally applicable to all hardware that has the
ACPI
Hardware Reduced bit set, since usually ACPI would be the preferred method.

Cc Len Brown <[email protected]>
Cc Mark Salter <[email protected]>
Cc "Rafael J. Wysocki" <[email protected]>

I'm reluctant to change the behaviour of efi_reboot_required() as it might break older hardware whereas checking if we have an EFI runtime is safer.

Rob