Skip to contentSkip to conversation

aarch64: Enable ECV to allow access to CNTPOFF_EL2

PATCH In review Marc Zyngier 2 participants 4 messages Aug 11, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 aarch64: Enable ECV to allow access to CNTPOFF_EL2

If the implemnentation supports ID_AA64MMFR0_EL1.ECV==2, set SCR_EL3.ECVEn to allow EL2 access to CNTPOFF_EL2.

---
 arch/aarch64/boot.S | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)
arch/aarch64/boot.S +9 -2
index 7f208b5..f0aa3cb 100644
@@ -54,10 +54,17 @@ ASM_FUNC(_start)
 1:
 	/* Enable FGT if present */
 	mrs	x1, id_aa64mmfr0_el1
-	ubfx	x1, x1, #56, #4
-	cbz	x1, 1f
+	ubfx	x2, x1, #56, #4
+	cbz	x2, 1f
 
 	orr	x0, x0, #(1 << 27)		// FGT enable
+1:
+	/* Enable ECV2 if present (allows CNTPOFF_EL2) */
+	ubfx	x2, x1, #60, #4
+	cmp	x2, #2
+	bne	1f
+
+	orr	x0, x0, #(1 << 28)		// ECV enable
 1:
 	/* Enable MTE if present */
 	mrs	x10, id_aa64pfr1_el1
Signature · 7 lines
2.30.2


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#1 lore raw
Quoted 2 lines · Marc Zyngier

If the implemnentation supports ID_AA64MMFR0_EL1.ECV==2, set SCR_EL3.ECVEn to allow EL2 access to CNTPOFF_EL2.

I was about to ask if that was a typo (why is this in an MMFR?), but that is what the ARM ARM says!

Quoted 14 lines
---
 arch/aarch64/boot.S | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)
arch/aarch64/boot.S +0 -1
index 7f208b5..f0aa3cb 100644
@@ -54,10 +54,17 @@ ASM_FUNC(_start)
 1:
 	/* Enable FGT if present */
 	mrs	x1, id_aa64mmfr0_el1
-	ubfx	x1, x1, #56, #4

arch/aarch64/boot.S ASM_FUNC(_start) · line 65

Quoted 10 lines
arch/aarch64/boot.S +7 -1
-	cbz	x1, 1f
+	ubfx	x2, x1, #56, #4
+	cbz	x2, 1f
 
 	orr	x0, x0, #(1 << 27)		// FGT enable
+1:
+	/* Enable ECV2 if present (allows CNTPOFF_EL2) */
+	ubfx	x2, x1, #60, #4
+	cmp	x2, #2
+	bne	1f

We need to check ID_AA64MMFR0_EL1.ECV >= 2 (to handle any futrue variants) so the conditional branch needs to be something like b.lt.

arch/aarch64/boot.S ASM_FUNC(_start) · line 67

Quoted 2 lines
arch/aarch64/boot.S +2 -0
+
+	orr	x0, x0, #(1 << 28)		// ECV enable

Aside from the above, this looks good to me, and I plan to apply this with the below tweaks (updated patch below):

* Use b.lt, as above

* To keep each check self-contained, the ECV check will read id_aa64mmfr0_el1 itself. That removes the need to keep x1 around, and so we can leave the FGT check as-is.

As part of some cleanup, I'm planning to move this to C, where feature checks will read ID fields via a helepr that implicitly reads the register, like:

  if (read_reg_field(ID_AA64MMFR0_EL1, ECV) >= 2)
  	scr |= SCR_EL3_ECVEN;

Thanks,
Mark.
---->8----
From bc6a9380eb3c5afc96735a54d455f2487df48700 Mon Sep 17 00:00:00 2001
From: Marc Zyngier <[email protected]>
Date: Wed, 11 Aug 2021 10:22:26 +0100
Subject: [PATCH] aarch64: Enable ECV to allow access to CNTPOFF_EL2

If the implemnentation supports ID_AA64MMFR0_EL1.ECV==2, set SCR_EL3.ECVEn to allow EL2 access to CNTPOFF_EL2.

[Mark: read id_aa64mmfr0_el1 separately, s/bne/b.lt/]
---
 arch/aarch64/boot.S | 8 ++++++++
 1 file changed, 8 insertions(+)
arch/aarch64/boot.S +8 -0
index 7f208b5..2215f7e 100644
@@ -59,6 +59,14 @@ ASM_FUNC(_start)
 
 	orr	x0, x0, #(1 << 27)		// FGT enable
 1:
+	/* Enable ECV2 if present (allows CNTPOFF_EL2) */
+	mrs	x1, id_aa64mmfr0_el1
+	ubfx	x1, x1, #60, #4
+	cmp	x1, #2
+	b.lt	1f
+
+	orr	x0, x0, #(1 << 28)		// ECV enable
+1:
 	/* Enable MTE if present */
 	mrs	x10, id_aa64pfr1_el1
 	ubfx	x10, x10, #8, #4
Signature · 7 lines
2.11.0


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#2 lore raw
Quoted 7 lines · Mark Rutland
Quoted 2 lines · Marc Zyngier

If the implemnentation supports ID_AA64MMFR0_EL1.ECV==2, set SCR_EL3.ECVEn to allow EL2 access to CNTPOFF_EL2.

I was about to ask if that was a typo (why is this in an MMFR?), but that is what the ARM ARM says!

I was just as surprised. PFR would have made a lot more sense, but I guess someone wanted to use these last four bits as quickly as possible... ;-)

Quoted 19 lines
Quoted 19 lines
---
 arch/aarch64/boot.S | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)
arch/aarch64/boot.S +2 -2
index 7f208b5..f0aa3cb 100644
@@ -54,10 +54,17 @@ ASM_FUNC(_start)
 1:
 	/* Enable FGT if present */
 	mrs	x1, id_aa64mmfr0_el1
-	ubfx	x1, x1, #56, #4
-	cbz	x1, 1f
+	ubfx	x2, x1, #56, #4
+	cbz	x2, 1f
 
Quoted 9 lines

arch/aarch64/boot.S ASM_FUNC(_start) · line 65

Quoted 6 lines
arch/aarch64/boot.S +5 -0
 	orr	x0, x0, #(1 << 27)		// FGT enable
+1:
+	/* Enable ECV2 if present (allows CNTPOFF_EL2) */
+	ubfx	x2, x1, #60, #4
+	cmp	x2, #2
+	bne	1f

We need to check ID_AA64MMFR0_EL1.ECV >= 2 (to handle any futrue variants) so the conditional branch needs to be something like b.lt.

Good point. I'm surprised ECV3 hasn't hit yet! :D

Quoted 12 lines

arch/aarch64/boot.S ASM_FUNC(_start) · line 68

Quoted 2 lines
arch/aarch64/boot.S +2 -0
+
+	orr	x0, x0, #(1 << 28)		// ECV enable

Aside from the above, this looks good to me, and I plan to apply this with the below tweaks (updated patch below):

* Use b.lt, as above

* To keep each check self-contained, the ECV check will read id_aa64mmfr0_el1 itself. That removes the need to keep x1 around, and so we can leave the FGT check as-is.

Quoted 6 lines

As part of some cleanup, I'm planning to move this to C, where feature checks will read ID fields via a helepr that implicitly reads the register, like:

  if (read_reg_field(ID_AA64MMFR0_EL1, ECV) >= 2)
  	scr |= SCR_EL3_ECVEN;

If you are actively rewriting this, maybe don't bother with the patch and just add this as part of your rewrite.

Quoted 28 lines

Thanks,
Mark.
---->8----
From bc6a9380eb3c5afc96735a54d455f2487df48700 Mon Sep 17 00:00:00 2001
From: Marc Zyngier <[email protected]>
Date: Wed, 11 Aug 2021 10:22:26 +0100
Subject: [PATCH] aarch64: Enable ECV to allow access to CNTPOFF_EL2

If the implemnentation supports ID_AA64MMFR0_EL1.ECV==2, set SCR_EL3.ECVEn to allow EL2 access to CNTPOFF_EL2.

[Mark: read id_aa64mmfr0_el1 separately, s/bne/b.lt/]
---
 arch/aarch64/boot.S | 8 ++++++++
 1 file changed, 8 insertions(+)
arch/aarch64/boot.S +1 -0
index 7f208b5..2215f7e 100644
@@ -59,6 +59,14 @@ ASM_FUNC(_start)
 
 	orr	x0, x0, #(1 << 27)		// FGT enable
 1:
+	/* Enable ECV2 if present (allows CNTPOFF_EL2) */

arch/aarch64/boot.S ASM_FUNC(_start) · line 72

Quoted 10 lines
arch/aarch64/boot.S +7 -0
+	mrs	x1, id_aa64mmfr0_el1
+	ubfx	x1, x1, #60, #4
+	cmp	x1, #2
+	b.lt	1f
+
+	orr	x0, x0, #(1 << 28)		// ECV enable
+1:
 	/* Enable MTE if present */
 	mrs	x10, id_aa64pfr1_el1
 	ubfx	x10, x10, #8, #4

Otherwise looks fine to me.

Thanks,

	M.
Signature · 6 lines
Without deviation from the norm, progress is not possible.

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
#3 lore raw
Quoted 61 lines · Marc Zyngier
Quoted 7 lines · Mark Rutland
Quoted 2 lines · Marc Zyngier

If the implemnentation supports ID_AA64MMFR0_EL1.ECV==2, set SCR_EL3.ECVEn to allow EL2 access to CNTPOFF_EL2.

I was about to ask if that was a typo (why is this in an MMFR?), but that is what the ARM ARM says!

I was just as surprised. PFR would have made a lot more sense, but I guess someone wanted to use these last four bits as quickly as possible... ;-)

Quoted 19 lines
Quoted 19 lines
---
 arch/aarch64/boot.S | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)
arch/aarch64/boot.S +2 -2
index 7f208b5..f0aa3cb 100644
@@ -54,10 +54,17 @@ ASM_FUNC(_start)
 1:
 	/* Enable FGT if present */
 	mrs	x1, id_aa64mmfr0_el1
-	ubfx	x1, x1, #56, #4
-	cbz	x1, 1f
+	ubfx	x2, x1, #56, #4
+	cbz	x2, 1f
 
Quoted 9 lines

arch/aarch64/boot.S ASM_FUNC(_start) · line 65

Quoted 6 lines
arch/aarch64/boot.S +5 -0
 	orr	x0, x0, #(1 << 27)		// FGT enable
+1:
+	/* Enable ECV2 if present (allows CNTPOFF_EL2) */
+	ubfx	x2, x1, #60, #4
+	cmp	x2, #2
+	bne	1f

We need to check ID_AA64MMFR0_EL1.ECV >= 2 (to handle any futrue variants) so the conditional branch needs to be something like b.lt.

Good point. I'm surprised ECV3 hasn't hit yet! :D

Quoted 16 lines

arch/aarch64/boot.S ASM_FUNC(_start) · line 68

Quoted 2 lines
arch/aarch64/boot.S +2 -0
+
+	orr	x0, x0, #(1 << 28)		// ECV enable

Aside from the above, this looks good to me, and I plan to apply this with the below tweaks (updated patch below):

* Use b.lt, as above

* To keep each check self-contained, the ECV check will read id_aa64mmfr0_el1 itself. That removes the need to keep x1 around, and so we can leave the FGT check as-is.

As part of some cleanup, I'm planning to move this to C, where feature checks will read ID fields via a helepr that implicitly reads the register, like:

Quoted 5 lines
Quoted 2 lines
  if (read_reg_field(ID_AA64MMFR0_EL1, ECV) >= 2)
  	scr |= SCR_EL3_ECVEN;

If you are actively rewriting this, maybe don't bother with the patch and just add this as part of your rewrite.

There's a bunch of preparatory stuff to fix first, so I've applied this patch (with those alterations) for now.

Thanks,
Mark.

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel