arm64/efi: use UEFI ResetSystem() Runtime Service for system reset

PATCH New Ard Biesheuvel 2 participants 3 messages Mar 5, 2015 – Mar 6, 2015

Subsystems (2)

arm64 EFI

lore

Share & tools

lore

Ard Biesheuvel

author

2015-03-05 12:51 UTC #0 lore raw

PATCH 1/1 arm64/efi: use UEFI ResetSystem() Runtime Service for system reset

If UEFI Runtime Services are available, the ResetSystem() service should be preferred over direct PSCI calls or other methods to reset the system. The reason is that the UpdateCapsule() UEFI Runtime Service, which is used to perform firmware updates, relies on this.

Signed-off-by Ard Biesheuvel <[email protected]>

---
I sent roughly the same patch ~6 months ago, but at the time, we were still waiting for the restart notifier call chain patches to land. Since that code got rejected, I am proposing this again. Note that efi_enabled(x) always evaluates to 'false' on !CONFIG_EFI.

This fixes reboot on my Seattle [although it doesn't make it any faster :-)]

 arch/arm64/kernel/process.c | 9 +++++++++
 1 file changed, 9 insertions(+)

arch/arm64/kernel/process.c +9 -0

index fde9923af859..a52bc0c316a8 100644
@@ -21,6 +21,7 @@
 #include <stdarg.h>
 
 #include <linux/compat.h>
+#include <linux/efi.h>
 #include <linux/export.h>
 #include <linux/sched.h>
 #include <linux/kernel.h>

@@ -150,6 +151,14 @@ void machine_restart(char *cmd)
 	local_irq_disable();
 	smp_send_stop();
 
+	/*
+	 * Prefer reboot via EFI if available, so that capsule updates [which
+	 * rely on UEFI's ResetSystem() being called with the return value of
+	 * UpdateCapsule()] have a chance of working as expected.
+	 */
+	if (efi_enabled(EFI_RUNTIME_SERVICES))
+		efi_reboot(reboot_mode, NULL);
+
 	/* Now call the architecture specific reboot code. */
 	if (arm_pm_restart)
 		arm_pm_restart(reboot_mode, cmd);

Signature · 1 line

1.8.3.2

Mark Rutland

reviewer

2015-03-05 14:22 UTC #1 lore raw

PATCH 1/1 arm64/efi: use UEFI ResetSystem() Runtime Service for system reset

Hi Ard,

Quoted 11 lines · Ard Biesheuvel

If UEFI Runtime Services are available, the ResetSystem() service should be preferred over direct PSCI calls or other methods to reset the system. The reason is that the UpdateCapsule() UEFI Runtime Service, which is used to perform firmware updates, relies on this.
Signed-off-by Ard Biesheuvel <[email protected]>
---
I sent roughly the same patch ~6 months ago, but at the time, we were still waiting for the restart notifier call chain patches to land. Since that code got rejected, I am proposing this again. Note that efi_enabled(x) always evaluates to 'false' on !CONFIG_EFI.

Also, efi_reboot is a static inline for !CONFIG_EFI, so I can't see any possibility of a build failure.

Quoted 18 lines

This fixes reboot on my Seattle [although it doesn't make it any faster :-)]

 arch/arm64/kernel/process.c | 9 +++++++++
 1 file changed, 9 insertions(+)

arch/arm64/kernel/process.c +1 -0

index fde9923af859..a52bc0c316a8 100644
@@ -21,6 +21,7 @@
 #include <stdarg.h>
 
 #include <linux/compat.h>
+#include <linux/efi.h>
 #include <linux/export.h>
 #include <linux/sched.h>
 #include <linux/kernel.h>

arch/arm64/kernel/process.c void machine_restart(char *cmd) · line 160

Quoted 10 lines

arch/arm64/kernel/process.c +7 -0

@@ -150,6 +151,14 @@ void machine_restart(char *cmd)
 	local_irq_disable();
 	smp_send_stop();
 
+	/*
+	 * Prefer reboot via EFI if available, so that capsule updates [which
+	 * rely on UEFI's ResetSystem() being called with the return value of
+	 * UpdateCapsule()] have a chance of working as expected.
+	 */
+	if (efi_enabled(EFI_RUNTIME_SERVICES))
+		efi_reboot(reboot_mode, NULL);

I expect that the particulars of the UpdateCapsule() will be handled within efi_reboot and won't require any additions here. So the comment could just be trimmed to say that UpdateCapsule() depends on the system being reset with ResetSystem().

Also, we need to make sure we call efi_poweroff to make UpdateCapsule() work when shutting the machine down (behind the scenes efi_poweroff calls ResetSystem(EfiResetShutdown, ...)).

For that I think adding the following to arch/arm64/kernel/efi.c is sufficient:

/*
* UpdateCapsule() depends on the system being shutdown via
* ResetSystem().
*/
bool efi_poweroff_required(void)

{
	return efi_enabled(EFI_RUNTIME_SERVICES);
}

I've given the patch a spin (with and without that addition) on Juno and Seattle. So with that folded in:

Tested-by Mark Rutland <[email protected]>

Reviewed-by Mark Rutland <[email protected]>

Thanks,
Mark.

Ard Biesheuvel

author

2015-03-06 12:14 UTC #2 lore raw

PATCH 1/1 arm64/efi: use UEFI ResetSystem() Runtime Service for system reset

Quoted 39 lines · Mark Rutland

Hi Ard,
Quoted 11 lines · Ard Biesheuvel
If UEFI Runtime Services are available, the ResetSystem() service should be preferred over direct PSCI calls or other methods to reset the system. The reason is that the UpdateCapsule() UEFI Runtime Service, which is used to perform firmware updates, relies on this.
Signed-off-by Ard Biesheuvel <[email protected]>
---
I sent roughly the same patch ~6 months ago, but at the time, we were still waiting for the restart notifier call chain patches to land. Since that code got rejected, I am proposing this again. Note that efi_enabled(x) always evaluates to 'false' on !CONFIG_EFI.
Also, efi_reboot is a static inline for !CONFIG_EFI, so I can't see any possibility of a build failure.
Quoted 21 lines
This fixes reboot on my Seattle [although it doesn't make it any faster :-)]
 arch/arm64/kernel/process.c | 9 +++++++++
 1 file changed, 9 insertions(+)
arch/arm64/kernel/process.c +1 -0
index fde9923af859..a52bc0c316a8 100644
@@ -21,6 +21,7 @@
 #include <stdarg.h>
 
 #include <linux/compat.h>
+#include <linux/efi.h>
 #include <linux/export.h>
 #include <linux/sched.h>
 #include <linux/kernel.h>
@@ -150,6 +151,14 @@ void machine_restart(char *cmd)
      local_irq_disable();
      smp_send_stop();
 

Quoted 13 lines

arch/arm64/kernel/process.c void machine_restart(char *cmd) · line 160
Quoted 7 lines
arch/arm64/kernel/process.c +7 -0
+     /*
+      * Prefer reboot via EFI if available, so that capsule updates [which
+      * rely on UEFI's ResetSystem() being called with the return value of
+      * UpdateCapsule()] have a chance of working as expected.
+      */
+     if (efi_enabled(EFI_RUNTIME_SERVICES))
+             efi_reboot(reboot_mode, NULL);
I expect that the particulars of the UpdateCapsule() will be handled within efi_reboot and won't require any additions here. So the comment could just be trimmed to say that UpdateCapsule() depends on the system being reset with ResetSystem().

OK.

Quoted 7 lines

Also, we need to make sure we call efi_poweroff to make UpdateCapsule() work when shutting the machine down (behind the scenes efi_poweroff calls ResetSystem(EfiResetShutdown, ...)).
For that I think adding the following to arch/arm64/kernel/efi.c is sufficient:

Quoted 9 lines

/*
* UpdateCapsule() depends on the system being shutdown via
* ResetSystem().
*/
bool efi_poweroff_required(void)
{
        return efi_enabled(EFI_RUNTIME_SERVICES);
}

Yes. I used to have this as a separate patch as well ~6 months ago, but I did not realise at the time that UpdateCapsule() depends on ResetSystem() for poweroff as well.

Quoted 6 lines

I've given the patch a spin (with and without that addition) on Juno and Seattle. So with that folded in:
Tested-by Mark Rutland <[email protected]>
Reviewed-by Mark Rutland <[email protected]>

Thanks. I will fold it in and resend.

Signature · 1 line

Ard.