Skip to contentSkip to conversation

arm64 CTR_EL0 cpufeature fixes

PATCH In review Will Deacon 3 participants 7 messages Aug 5, 2019 – Aug 7, 2019

Subsystems (2)
lore
Share & tools
lore

Will Deacon

author maintainer
#0 lore raw

cover arm64 CTR_EL0 cpufeature fixes

Hi,

These two patches are backports for 4.4.y stable kernels after one of them failed to apply:

https://lkml.kernel.org/r/[email protected]

Cheers,

Will

--->8

Will Deacon (2):
arm64: cpufeature: Fix CTR_EL0 field definitions
arm64: cpufeature: Fix feature comparison for CTR_EL0.{CWG,ERG}

 arch/arm64/include/asm/cpufeature.h |  7 ++++---
 arch/arm64/kernel/cpufeature.c      | 14 ++++++++++----
 2 files changed, 14 insertions(+), 7 deletions(-)
Signature · 1 line
2.11.0
author maintainer
#1 lore raw

PATCH 1/2 arm64: cpufeature: Fix CTR_EL0 field definitions

From: Will Deacon <[email protected]>

commit be68a8aaf925aaf35574260bf820bb09d2f9e07f upstream.

Our field definitions for CTR_EL0 suffer from a number of problems:

- The IDC and DIC fields are missing, which causes us to enable CTR trapping on CPUs with either of these returning non-zero values.

- The ERG is FTR_LOWER_SAFE, whereas it should be treated like CWG as FTR_HIGHER_SAFE so that applications can use it to avoid false sharing.

- [nit] A RES1 field is described as "RAO"

This patch updates the CTR_EL0 field definitions to fix these issues.

Cc <[email protected]> # 4.4.y only
Cc Shanker Donthineni <[email protected]>
---
 arch/arm64/kernel/cpufeature.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
arch/arm64/kernel/cpufeature.c +5 -3
index c1eddc07d996..fff0bf2f889e 100644
@@ -126,10 +126,12 @@ static struct arm64_ftr_bits ftr_id_aa64mmfr1[] = {
 };
 
 static struct arm64_ftr_bits ftr_ctr[] = {
-	U_ARM64_FTR_BITS(FTR_STRICT, FTR_EXACT, 31, 1, 1),	/* RAO */
-	ARM64_FTR_BITS(FTR_STRICT, FTR_EXACT, 28, 3, 0),
+	U_ARM64_FTR_BITS(FTR_STRICT, FTR_EXACT, 31, 1, 1),	/* RES1 */
+	ARM64_FTR_BITS(FTR_STRICT, FTR_EXACT, 30, 1, 0),
+	U_ARM64_FTR_BITS(FTR_STRICT, FTR_LOWER_SAFE, 29, 1, 1),	/* DIC */
+	U_ARM64_FTR_BITS(FTR_STRICT, FTR_LOWER_SAFE, 28, 1, 1),	/* IDC */
 	U_ARM64_FTR_BITS(FTR_STRICT, FTR_HIGHER_SAFE, 24, 4, 0),	/* CWG */
-	U_ARM64_FTR_BITS(FTR_STRICT, FTR_LOWER_SAFE, 20, 4, 0),	/* ERG */
+	U_ARM64_FTR_BITS(FTR_STRICT, FTR_HIGHER_SAFE, 20, 4, 0),	/* ERG */
 	U_ARM64_FTR_BITS(FTR_STRICT, FTR_LOWER_SAFE, 16, 4, 1),	/* DminLine */
 	/*
 	 * Linux can handle differing I-cache policies. Userspace JITs will
Signature · 1 line
2.11.0
author maintainer
#2 lore raw

PATCH 2/2 arm64: cpufeature: Fix feature comparison for CTR_EL0.{CWG,ERG}

In reply to Will Deacon

commit 147b9635e6347104b91f48ca9dca61eb0fbf2a54 upstream.

If CTR_EL0.{CWG,ERG} are 0b0000 then they must be interpreted to have their architecturally maximum values, which defeats the use of FTR_HIGHER_SAFE when sanitising CPU ID registers on heterogeneous machines.

Introduce FTR_HIGHER_OR_ZERO_SAFE so that these fields effectively saturate at zero.

Fixes 3c739b571084 ("arm64: Keep track of CPU feature registers")
Cc <[email protected]> # 4.4.y only
---
 arch/arm64/include/asm/cpufeature.h | 7 ++++---
 arch/arm64/kernel/cpufeature.c      | 8 ++++++--
 2 files changed, 10 insertions(+), 5 deletions(-)
arch/arm64/include/asm/cpufeature.h +4 -3
index ad83c245781c..0a66f8241f18 100644
@@ -41,9 +41,10 @@
 
 /* CPU feature register tracking */
 enum ftr_type {
-	FTR_EXACT,	/* Use a predefined safe value */
-	FTR_LOWER_SAFE,	/* Smaller value is safe */
-	FTR_HIGHER_SAFE,/* Bigger value is safe */
+	FTR_EXACT,			/* Use a predefined safe value */
+	FTR_LOWER_SAFE,			/* Smaller value is safe */
+	FTR_HIGHER_SAFE,		/* Bigger value is safe */
+	FTR_HIGHER_OR_ZERO_SAFE,	/* Bigger value is safe, but 0 is biggest */
 };
 
 #define FTR_STRICT	true	/* SANITY check strict matching required */
arch/arm64/kernel/cpufeature.c +6 -2
index fff0bf2f889e..062484d34450 100644
@@ -130,8 +130,8 @@ static struct arm64_ftr_bits ftr_ctr[] = {
 	ARM64_FTR_BITS(FTR_STRICT, FTR_EXACT, 30, 1, 0),
 	U_ARM64_FTR_BITS(FTR_STRICT, FTR_LOWER_SAFE, 29, 1, 1),	/* DIC */
 	U_ARM64_FTR_BITS(FTR_STRICT, FTR_LOWER_SAFE, 28, 1, 1),	/* IDC */
-	U_ARM64_FTR_BITS(FTR_STRICT, FTR_HIGHER_SAFE, 24, 4, 0),	/* CWG */
-	U_ARM64_FTR_BITS(FTR_STRICT, FTR_HIGHER_SAFE, 20, 4, 0),	/* ERG */
+	U_ARM64_FTR_BITS(FTR_STRICT, FTR_HIGHER_OR_ZERO_SAFE, 24, 4, 0),	/* CWG */
+	U_ARM64_FTR_BITS(FTR_STRICT, FTR_HIGHER_OR_ZERO_SAFE, 20, 4, 0),	/* ERG */
 	U_ARM64_FTR_BITS(FTR_STRICT, FTR_LOWER_SAFE, 16, 4, 1),	/* DminLine */
 	/*
 	 * Linux can handle differing I-cache policies. Userspace JITs will
@@ -341,6 +341,10 @@ static s64 arm64_ftr_safe_value(struct arm64_ftr_bits *ftrp, s64 new, s64 cur)
 	case FTR_LOWER_SAFE:
 		ret = new < cur ? new : cur;
 		break;
+	case FTR_HIGHER_OR_ZERO_SAFE:
+		if (!cur || !new)
+			break;
+		/* Fallthrough */
 	case FTR_HIGHER_SAFE:
 		ret = new > cur ? new : cur;
 		break;
Signature · 1 line
2.11.0
maintainer
#3 lore raw
In reply to Will Deacon
Quoted 6 lines · Will Deacon

Hi,

These two patches are backports for 4.4.y stable kernels after one of them failed to apply:

https://lkml.kernel.org/r/[email protected]

I took these 2, but note that they have two fixes that are not in 4.4:

314d53d297980 arm64: Handle mismatched cache type 4c4a39dd5fe2d arm64: Fix mismatched cache line size detection

Will you send a backport for them?

--
Thanks,
Sasha

author maintainer
#4 lore raw

Hi Sasha, [+Suzuki]

Quoted 4 lines · Sasha Levin
Quoted 3 lines · Will Deacon

These two patches are backports for 4.4.y stable kernels after one of them failed to apply:

Quoted 8 lines
Quoted 1 line

https://lkml.kernel.org/r/[email protected]

I took these 2, but note that they have two fixes that are not in 4.4:

314d53d297980 arm64: Handle mismatched cache type 4c4a39dd5fe2d arm64: Fix mismatched cache line size detection

Will you send a backport for them?

4.4 doesn't handle mismatches for the cache type or the line sizes, and instead taints the kernel with a big splat at boot. If we wanted to backport this, we'd need to pick up more than just the above patches, since we don't have the means to emulate user cache maintenance operations either.

Given that the vast majority of systems don't suffer from this problem, I'd be inclined not to try shoe-horning all of this into 4.4, where I think it's more likely introduce other issues.

Suzuki, what do you think?

Will

#5 lore raw

Hi Will,

Quoted 21 lines · Will Deacon

Hi Sasha, [+Suzuki]

Quoted 4 lines · Sasha Levin
Quoted 3 lines · Will Deacon

These two patches are backports for 4.4.y stable kernels after one of them failed to apply:

Quoted 8 lines
Quoted 1 line

https://lkml.kernel.org/r/[email protected]

I took these 2, but note that they have two fixes that are not in 4.4:

314d53d297980 arm64: Handle mismatched cache type 4c4a39dd5fe2d arm64: Fix mismatched cache line size detection

Will you send a backport for them?

4.4 doesn't handle mismatches for the cache type or the line sizes, and instead taints the kernel with a big splat at boot. If we wanted to backport this, we'd need to pick up more than just the above patches, since we don't have the means to emulate user cache maintenance operations either.

Quoted 5 lines

Given that the vast majority of systems don't suffer from this problem, I'd be inclined not to try shoe-horning all of this into 4.4, where I think it's more likely introduce other issues.

Suzuki, what do you think?

I agree. It involves a lot of new code, with non-trivial backports.

Cheers
Suzuki

maintainer
#6 lore raw
Quoted 24 lines · Suzuki K Poulose

Hi Will,

Quoted 21 lines · Will Deacon

Hi Sasha, [+Suzuki]

Quoted 4 lines · Sasha Levin
Quoted 3 lines · Will Deacon

These two patches are backports for 4.4.y stable kernels after one of them failed to apply:

Quoted 8 lines
Quoted 1 line

https://lkml.kernel.org/r/[email protected]

I took these 2, but note that they have two fixes that are not in 4.4:

314d53d297980 arm64: Handle mismatched cache type 4c4a39dd5fe2d arm64: Fix mismatched cache line size detection

Will you send a backport for them?

4.4 doesn't handle mismatches for the cache type or the line sizes, and instead taints the kernel with a big splat at boot. If we wanted to backport this, we'd need to pick up more than just the above patches, since we don't have the means to emulate user cache maintenance operations either.

Quoted 7 lines
Quoted 5 lines

Given that the vast majority of systems don't suffer from this problem, I'd be inclined not to try shoe-horning all of this into 4.4, where I think it's more likely introduce other issues.

Suzuki, what do you think?

I agree. It involves a lot of new code, with non-trivial backports.

Okay, makes sense. Thanks for looking into it.

--
Thanks,
Sasha