Skip to contentSkip to conversation

arm: gic-v3: deactivate SGI/PPI during initialization

PATCH Superseded Peng Fan 3 participants 6 messages Jan 30, 2019 – Feb 5, 2019

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Peng Fan

author
#0 lore raw

PATCH v2 1/1 arm: gic-v3: deactivate SGI/PPI during initialization

On i.MX8, we implemented partition reboot which means Cortex-A reboot will not impact M4 cores and System control Unit core. However GICv3 is not reset because we also need to support A72 Cluster reboot without affecting A53 Cluster.

The gic-v3 controller is configured with EOImode to 1, so during xen reboot, there is a function call "smp_call_function(halt_this_cpu, NULL, 0);" , but halt_this_cpu never return, that means other CPUs have no chance to deactivate the SGI interrupt, because the deactivate_irq operation is at the end of do_sgi. During the next boot of Xen, CPU0 will issue GIC_SGI_CALL_FUNCTION to other CPUs. As the Active state for SGI is left untouched during the reboot, the GIC_SGI_CALL_FUNCTION will still be active on the non-boot CPUs. This means the interrupt cannot be triggered again until it get deactivated.

And according to IHI0069D_gic_architecture_specification, chapter "8.11.3 GICR_ICACTIVER0, Interrupt Clear-Active Register 0", the RW field of GICR_ICACTIVER0 resets to a value that is architecturally UNKNOWN.

So set a fixed value during gic-v3 initialization to make sure interrupts are in deactivated state.

---

V2:
refine commit log
deactivate SGI/PPI.
No need for SPI, because Some or all RW fields of this GICD_ICACTIVER have defined reset values

 xen/arch/arm/gic-v3.c | 5 +++++
 1 file changed, 5 insertions(+)
xen/arch/arm/gic-v3.c +5 -0
index 6fbc106757..1c1d2604f3 100644
@@ -833,6 +833,11 @@ static int gicv3_cpu_init(void)
         writel_relaxed(priority,
                 GICD_RDIST_SGI_BASE + GICR_IPRIORITYR0 + (i / 4) * 4);
 
+    /*
+     * The activate state is unknown at boot, so make sure all
+     * SGIs and PPIs are de-activated.
+     */
+    writel_relaxed(0xffffffff, GICD_RDIST_SGI_BASE + GICR_ICACTIVER0);
     /*
      * Disable all PPI interrupts, ensure all SGI interrupts are
      * enabled.
Signature · 7 lines
2.14.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#1 lore raw

Hi,

We spoke about SPIs in the previous version. Why aren't they de-activated here?

Quoted 15 lines · Peng Fan

On i.MX8, we implemented partition reboot which means Cortex-A reboot will not impact M4 cores and System control Unit core. However GICv3 is not reset because we also need to support A72 Cluster reboot without affecting A53 Cluster.

The gic-v3 controller is configured with EOImode to 1, so during xen reboot, there is a function call "smp_call_function(halt_this_cpu, NULL, 0);" , but halt_this_cpu never return, that means other CPUs have no chance to deactivate the SGI interrupt, because the deactivate_irq operation is at the end of do_sgi. During the next boot of Xen, CPU0 will issue GIC_SGI_CALL_FUNCTION to other CPUs. As the Active state for SGI is left untouched during the reboot, the GIC_SGI_CALL_FUNCTION will still be active on the non-boot CPUs. This means the interrupt cannot be triggered again until it get deactivated.

Quoted 6 lines

And according to IHI0069D_gic_architecture_specification, chapter "8.11.3 GICR_ICACTIVER0, Interrupt Clear-Active Register 0", the RW field of GICR_ICACTIVER0 resets to a value that is architecturally UNKNOWN.

So set a fixed value during gic-v3 initialization to make sure interrupts are in deactivated state.

It is a bit unclear what you mean by "fixed value" here. The only thing you do is clearing active state. So a better wording is "So make sure all interrupts are deactivated at during initialization by clearing the state".

How about SPIs?

Cheers,

Signature · 6 lines
Julien Grall

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
author
#2 lore raw

Hi Julien

Quoted 8 lines

-----Original Message-----
From: Julien Grall [mailto:[email protected]]
Sent: 2019年2月1日 18:41
To: Peng Fan <[email protected]>; [email protected]; [email protected]
Cc: [email protected]
Subject: Re: [PATCH for-4.12 v2] arm: gic-v3: deactivate SGI/PPI during initialization

Quoted 4 lines

Hi,

We spoke about SPIs in the previous version. Why aren't they de-activated here?

According to GIC-V3 " 8.9.5 GICD_ICACTIVER<n>, Interrupt Clear-Active Registers, n = 0 - 31" "
Some or all RW fields of this register have defined reset values.
When this register has an architecturally-defined reset value, this field resets to 0. "

So I think we no need to deactivated, because it has reset values 0.

Quoted 22 lines · Peng Fan
Quoted 15 lines

On i.MX8, we implemented partition reboot which means Cortex-A reboot will not impact M4 cores and System control Unit core. However GICv3 is not reset because we also need to support A72 Cluster reboot without affecting A53 Cluster.

The gic-v3 controller is configured with EOImode to 1, so during xen reboot, there is a function call "smp_call_function(halt_this_cpu, NULL, 0);" , but halt_this_cpu never return, that means other CPUs have no chance to deactivate the SGI interrupt, because the deactivate_irq operation is at the end of do_sgi. During the next boot of Xen, CPU0 will issue GIC_SGI_CALL_FUNCTION to other CPUs. As the Active state for SGI is left untouched during the reboot, the GIC_SGI_CALL_FUNCTION will still be active on the non-boot CPUs. This means the interrupt cannot be triggered again until it get deactivated.

Quoted 3 lines

And according to IHI0069D_gic_architecture_specification, chapter "8.11.3 GICR_ICACTIVER0, Interrupt Clear-Active Register 0", the RW field of GICR_ICACTIVER0 resets to a value that is architecturally

UNKNOWN.

Quoted 1 line
Quoted 8 lines
Quoted 2 lines

So set a fixed value during gic-v3 initialization to make sure interrupts are in deactivated state.

It is a bit unclear what you mean by "fixed value" here. The only thing you do is clearing active state. So a better wording is "So make sure all interrupts are deactivated at during initialization by clearing the state".

How about SPIs?

Replied above. I could add the following to the patch if you think it needed, I am not sure whether it is valid that SPI is in active state when Xen booting.

xen/arch/arm/gic-v3.c +4 -1
index 1c1d2604f3..5b9c5559a7 100644
@@ -622,9 +622,12 @@ static void __init gicv3_dist_init(void)
         writel_relaxed(priority, GICD + GICD_IPRIORITYR + (i / 4) * 4);
     }
 
-    /* Disable all global interrupts */
+    /* Disable/deactivate all global interrupts */
     for ( i = NR_GIC_LOCAL_IRQS; i < nr_lines; i += 32 )
+    {
         writel_relaxed(0xffffffff, GICD + GICD_ICENABLER + (i / 32) * 4);
+        writel_relaxed(0xffffffff, GICD + GICD_ICACTIVER + (i / 32) * 4);
+    }
 
     /*
      * Configure SPIs as non-secure Group-1. This will only matter

Thanks,
Peng.

Quoted 5 lines

Cheers,

--
Julien Grall

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#3 lore raw
Quoted 1 line · Peng Fan

Hi Julien

Hi Peng,

Quoted 10 lines
Quoted 10 lines

-----Original Message-----
From: Julien Grall [mailto:[email protected]]
Sent: 2019年2月1日 18:41
To: Peng Fan <[email protected]>; [email protected]; [email protected]
Cc: [email protected]
Subject: Re: [PATCH for-4.12 v2] arm: gic-v3: deactivate SGI/PPI during initialization

Hi,

Quoted 8 lines
Quoted 2 lines

We spoke about SPIs in the previous version. Why aren't they de-activated here?

According to GIC-V3 " 8.9.5 GICD_ICACTIVER<n>, Interrupt Clear-Active Registers, n = 0 - 31" "
Some or all RW fields of this register have defined reset values.
When this register has an architecturally-defined reset value, this field resets to 0. "

I can't find this wording in the last spec (IHI0069E). Can you please give the version of the specific version of the spec when quoting it?

Quoted 2 lines

So I think we no need to deactivated, because it has reset values 0

The specification is about the reset value from the hardware, it does not tell you how the firmware (or the previous kernel when using kexec) is going to leave the interrupts. For instance, as I pointed out in a different thread, Xen may reboot with SPIs activated. This can happen if you receive the reboot/power off request while handling an SPI.

Quoted 26 lines
Quoted 26 lines · Peng Fan
Quoted 15 lines

On i.MX8, we implemented partition reboot which means Cortex-A reboot will not impact M4 cores and System control Unit core. However GICv3 is not reset because we also need to support A72 Cluster reboot without affecting A53 Cluster.

The gic-v3 controller is configured with EOImode to 1, so during xen reboot, there is a function call "smp_call_function(halt_this_cpu, NULL, 0);" , but halt_this_cpu never return, that means other CPUs have no chance to deactivate the SGI interrupt, because the deactivate_irq operation is at the end of do_sgi. During the next boot of Xen, CPU0 will issue GIC_SGI_CALL_FUNCTION to other CPUs. As the Active state for SGI is left untouched during the reboot, the GIC_SGI_CALL_FUNCTION will still be active on the non-boot CPUs. This means the interrupt cannot be triggered again until it get deactivated.

Quoted 3 lines

And according to IHI0069D_gic_architecture_specification, chapter "8.11.3 GICR_ICACTIVER0, Interrupt Clear-Active Register 0", the RW field of GICR_ICACTIVER0 resets to a value that is architecturally

UNKNOWN.

Quoted 3 lines

So set a fixed value during gic-v3 initialization to make sure interrupts are in deactivated state.

Quoted 8 lines
Quoted 5 lines

It is a bit unclear what you mean by "fixed value" here. The only thing you do is clearing active state. So a better wording is "So make sure all interrupts are deactivated at during initialization by clearing the state".

How about SPIs?

Replied above. I could add the following to the patch if you think it needed, I am not sure whether it is valid that SPI is in active state when Xen booting.

See above, I think this is valid and can happen today.

Quoted 9 lines
xen/arch/arm/gic-v3.c +0 -0
index 1c1d2604f3..5b9c5559a7 100644
@@ -622,9 +622,12 @@ static void __init gicv3_dist_init(void)
          writel_relaxed(priority, GICD + GICD_IPRIORITYR + (i / 4) * 4);
      }
 

xen/arch/arm/gic-v3.c static void __init gicv3_dist_init(void) · line 633

Quoted 10 lines
xen/arch/arm/gic-v3.c +4 -1
-    /* Disable all global interrupts */
+    /* Disable/deactivate all global interrupts */
      for ( i = NR_GIC_LOCAL_IRQS; i < nr_lines; i += 32 )
+    {
          writel_relaxed(0xffffffff, GICD + GICD_ICENABLER + (i / 32) * 4);
+        writel_relaxed(0xffffffff, GICD + GICD_ICACTIVER + (i / 32) * 4);
+    }
 
      /*
       * Configure SPIs as non-secure Group-1. This will only matter

This chunk looks good. I will also write a patch for GICv2 doing the same.

Cheers,

Signature · 6 lines
Julien Grall

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
author
#4 lore raw
Quoted 33 lines

-----Original Message-----
From: Julien Grall [mailto:[email protected]]
Sent: 2019年2月3日 0:04
To: Peng Fan <[email protected]>; [email protected]; [email protected]
Cc: [email protected]
Subject: Re: [PATCH for-4.12 v2] arm: gic-v3: deactivate SGI/PPI during initialization

Quoted 1 line · Peng Fan

Hi Julien

Hi Peng,

Quoted 9 lines
Quoted 9 lines

-----Original Message-----
From: Julien Grall [mailto:[email protected]]
Sent: 2019年2月1日 18:41
To: Peng Fan <[email protected]>; [email protected]; [email protected]
Cc: [email protected]
Subject: Re: [PATCH for-4.12 v2] arm: gic-v3: deactivate SGI/PPI during initialization

Quoted 6 lines
Quoted 4 lines

Hi,

We spoke about SPIs in the previous version. Why aren't they de-activated here?

According to GIC-V3 " 8.9.5 GICD_ICACTIVER<n>, Interrupt Clear-Active

Registers, n = 0 - 31"

Quoted 8 lines
Quoted 3 lines

"
Some or all RW fields of this register have defined reset values. When this register has an architecturally-defined reset value, this field

resets to 0.

Quoted 1 line

"

I can't find this wording in the last spec (IHI0069E). Can you please give the version of the specific version of the spec when quoting it?

The spec I use is IHI0069D_gic_architecture_specification. Will add the version in future patches when needed.

Quoted 9 lines
Quoted 3 lines

So I think we no need to deactivated, because it has reset values 0

The specification is about the reset value from the hardware, it does not tell you how the firmware (or the previous kernel when using kexec) is going to leave the interrupts. For instance, as I pointed out in a different thread, Xen may reboot with SPIs activated. This can happen if you receive the reboot/power off request while handling an SPI.

Thanks, just sent out v3 including the deactivation of SPI.

Quoted 60 lines
Quoted 11 lines
Quoted 11 lines · Peng Fan
Quoted 7 lines

On i.MX8, we implemented partition reboot which means Cortex-A reboot will not impact M4 cores and System control Unit core. However GICv3 is not reset because we also need to support A72 Cluster reboot without affecting A53 Cluster.

The gic-v3 controller is configured with EOImode to 1, so during xen reboot, there is a function call "smp_call_function(halt_this_cpu, NULL,

0);"

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

, but halt_this_cpu never return, that means other CPUs have no chance to deactivate the SGI interrupt, because the deactivate_irq operation is at the end of do_sgi. During the next boot of Xen, CPU0 will issue GIC_SGI_CALL_FUNCTION to other CPUs. As the Active state for SGI is left untouched during the reboot, the GIC_SGI_CALL_FUNCTION will still be active on the non-boot CPUs. This means the interrupt cannot be triggered again until it get

deactivated.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

And according to IHI0069D_gic_architecture_specification, chapter "8.11.3 GICR_ICACTIVER0, Interrupt Clear-Active Register 0", the RW field of GICR_ICACTIVER0 resets to a value that is architecturally

Quoted 8 lines
Quoted 8 lines

UNKNOWN.

Quoted 3 lines

So set a fixed value during gic-v3 initialization to make sure interrupts are in deactivated state.

It is a bit unclear what you mean by "fixed value" here. The only thing you do is clearing active state. So a better wording is "So make sure all interrupts are deactivated at during initialization by clearing

the state".

Quoted 5 lines
Quoted 2 lines

How about SPIs?

Replied above. I could add the following to the patch if you think it needed, I am not sure whether it is valid that SPI is in active state when Xen

booting.

See above, I think this is valid and can happen today.

Quoted 18 lines
xen/arch/arm/gic-v3.c index renamed from xen/arch/arm/gic-v3.c +0 -0

1c1d2604f3..5b9c5559a7 100644

xen/arch/arm/gic-v3.c +4 -1
@@ -622,9 +622,12 @@ static void __init gicv3_dist_init(void)
          writel_relaxed(priority, GICD + GICD_IPRIORITYR + (i / 4) * 4);
      }
 
-    /* Disable all global interrupts */
+    /* Disable/deactivate all global interrupts */
      for ( i = NR_GIC_LOCAL_IRQS; i < nr_lines; i += 32 )
+    {
          writel_relaxed(0xffffffff, GICD + GICD_ICENABLER + (i / 32)
 * 4);
+        writel_relaxed(0xffffffff, GICD + GICD_ICACTIVER + (i / 32) * 4);
+    }
 
Quoted 4 lines
Quoted 2 lines
xen/arch/arm/gic-v3.c +0 -0
      /*

* Configure SPIs as non-secure Group-1. This will only matter

This chunk looks good. I will also write a patch for GICv2 doing the same.

Thanks,
Peng.

Quoted 5 lines

Cheers,

--
Julien Grall

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel