Skip to contentSkip to conversation

arm64: errata: refer to config ARM64_ERRATUM_2645198 to make workaround work

PATCH In review Lukas Bulwahn 3 participants 5 messages Dec 15, 2022

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 arm64: errata: refer to config ARM64_ERRATUM_2645198 to make workaround work

Commit 44ecda71fd8a ("arm64: errata: Workaround possible Cortex-A715 [ESR|FAR]_ELx corruption") implements a workaround for arm64 erratum 2645198. The arm64 cpucaps is called WORKAROUND_2645198; the kernel build configuration is called ARM64_ERRATUM_2645198.

In the functions huge_ptep_modify_prot_start() and ptep_modify_prot_start(), the code accidently refers to the non-existing config CONFIG_ARM64_WORKAROUND_2645198. Note that the config name uses ERRATUM, not WORKAROUND. By this accidental misreference, this condition is always false, the branch of the workaround is not reachable and the workaround is effectively not implemented at all.

Refer to the intended config ARM64_ERRATUM_2645198 and make the intended workaround effectively work.

Fixes 44ecda71fd8a ("arm64: errata: Workaround possible Cortex-A715 [ESR|FAR]_ELx corruption")
---
 arch/arm64/mm/hugetlbpage.c | 2 +-
 arch/arm64/mm/mmu.c         | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
arch/arm64/mm/hugetlbpage.c +1 -1
index cd8d96e1fa1a..95364e8bdc19 100644
@@ -562,7 +562,7 @@ bool __init arch_hugetlb_valid_size(unsigned long size)
 
 pte_t huge_ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
-	if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198) &&
+	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
 		/*
 		 * Break-before-make (BBM) is required for all user space mappings
arch/arm64/mm/mmu.c +1 -1
index 12915f379c22..d77c9f56b7b4 100644
@@ -1633,7 +1633,7 @@ early_initcall(prevent_bootmem_remove_init);
 
 pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
-	if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198) &&
+	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
 		/*
 		 * Break-before-make (BBM) is required for all user space mappings
Signature · 1 line
2.17.1
maintainer
#1 lore raw
Quoted 15 lines · Lukas Bulwahn

Commit 44ecda71fd8a ("arm64: errata: Workaround possible Cortex-A715 [ESR|FAR]_ELx corruption") implements a workaround for arm64 erratum 2645198. The arm64 cpucaps is called WORKAROUND_2645198; the kernel build configuration is called ARM64_ERRATUM_2645198.

In the functions huge_ptep_modify_prot_start() and ptep_modify_prot_start(), the code accidently refers to the non-existing config CONFIG_ARM64_WORKAROUND_2645198. Note that the config name uses ERRATUM, not WORKAROUND. By this accidental misreference, this condition is always false, the branch of the workaround is not reachable and the workaround is effectively not implemented at all.

Refer to the intended config ARM64_ERRATUM_2645198 and make the intended workaround effectively work.

Quoted 6 lines
Fixes 44ecda71fd8a ("arm64: errata: Workaround possible Cortex-A715 [ESR|FAR]_ELx corruption")
---
 arch/arm64/mm/hugetlbpage.c | 2 +-
 arch/arm64/mm/mmu.c         | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

Thanks for the report!

Quoted 18 lines
arch/arm64/mm/hugetlbpage.c +1 -1
index cd8d96e1fa1a..95364e8bdc19 100644
@@ -562,7 +562,7 @@ bool __init arch_hugetlb_valid_size(unsigned long size)
 
 pte_t huge_ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
-	if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198) &&
+	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
 		/*
 		 * Break-before-make (BBM) is required for all user space mappings
arch/arm64/mm/mmu.c +0 -0
index 12915f379c22..d77c9f56b7b4 100644

arch/arm64/mm/mmu.c early_initcall(prevent_bootmem_remove_init); · line 1639

Quoted 8 lines
arch/arm64/mm/mmu.c +1 -1
@@ -1633,7 +1633,7 @@ early_initcall(prevent_bootmem_remove_init);
 
 pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
-	if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198) &&
+	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
 		/*
 		 * Break-before-make (BBM) is required for all user space mappings

Grr, this bug seems to exist in all three versions of the patch reviewed on the list, so I can only draw the conclusion that this code has never been tested. Consequently, I'm more inclined to _revert_ the change for now and we can bring it back as a fix once somebody has checked that it actually works properly.

Will

#2 lore raw
Quoted 44 lines · Will Deacon
Quoted 15 lines · Lukas Bulwahn

Commit 44ecda71fd8a ("arm64: errata: Workaround possible Cortex-A715 [ESR|FAR]_ELx corruption") implements a workaround for arm64 erratum 2645198. The arm64 cpucaps is called WORKAROUND_2645198; the kernel build configuration is called ARM64_ERRATUM_2645198.

In the functions huge_ptep_modify_prot_start() and ptep_modify_prot_start(), the code accidently refers to the non-existing config CONFIG_ARM64_WORKAROUND_2645198. Note that the config name uses ERRATUM, not WORKAROUND. By this accidental misreference, this condition is always false, the branch of the workaround is not reachable and the workaround is effectively not implemented at all.

Refer to the intended config ARM64_ERRATUM_2645198 and make the intended workaround effectively work.

Quoted 6 lines
Fixes 44ecda71fd8a ("arm64: errata: Workaround possible Cortex-A715 [ESR|FAR]_ELx corruption")
---
 arch/arm64/mm/hugetlbpage.c | 2 +-
 arch/arm64/mm/mmu.c         | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

Thanks for the report!

Quoted 19 lines
arch/arm64/mm/hugetlbpage.c +1 -1
index cd8d96e1fa1a..95364e8bdc19 100644
@@ -562,7 +562,7 @@ bool __init arch_hugetlb_valid_size(unsigned long size)
 
 pte_t huge_ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
-	if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198) &&
+	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
 		/*
 		 * Break-before-make (BBM) is required for all user space mappings
arch/arm64/mm/mmu.c +0 -0
index 12915f379c22..d77c9f56b7b4 100644
@@ -1633,7 +1633,7 @@ early_initcall(prevent_bootmem_remove_init);
 
Quoted 10 lines

arch/arm64/mm/mmu.c early_initcall(prevent_bootmem_remove_init); · line 1639

Quoted 7 lines
arch/arm64/mm/mmu.c +1 -1
 pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
-	if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198) &&
+	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
 		/*
 		 * Break-before-make (BBM) is required for all user space mappings

Grr, this bug seems to exist in all three versions of the patch reviewed on the list, so I can only draw the conclusion that this code has never been

Ohh, my bad, apologies. I did not have a real system with this erratum, although had emulated and tested this workaround path via some other debug changes (which might have just forced the first condition to always evaluate true).

Quoted 3 lines

tested. Consequently, I'm more inclined to _revert_ the change for now and we can bring it back as a fix once somebody has checked that it actually works properly.

Please do not revert this change if possible.

maintainer
#3 lore raw
Quoted 32 lines · Anshuman Khandual
Quoted 20 lines · Will Deacon
Quoted 19 lines · Lukas Bulwahn
arch/arm64/mm/hugetlbpage.c +1 -1
index cd8d96e1fa1a..95364e8bdc19 100644
@@ -562,7 +562,7 @@ bool __init arch_hugetlb_valid_size(unsigned long size)
 
 pte_t huge_ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
-	if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198) &&
+	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
 		/*
 		 * Break-before-make (BBM) is required for all user space mappings
arch/arm64/mm/mmu.c +0 -0
index 12915f379c22..d77c9f56b7b4 100644
@@ -1633,7 +1633,7 @@ early_initcall(prevent_bootmem_remove_init);
 
Quoted 10 lines

arch/arm64/mm/mmu.c early_initcall(prevent_bootmem_remove_init); · line 1639

Quoted 7 lines
arch/arm64/mm/mmu.c +1 -1
 pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
-	if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198) &&
+	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
 		/*
 		 * Break-before-make (BBM) is required for all user space mappings

Grr, this bug seems to exist in all three versions of the patch reviewed on the list, so I can only draw the conclusion that this code has never been

Quoted 3 lines

Ohh, my bad, apologies. I did not have a real system with this erratum, although had emulated and tested this workaround path via some other debug changes (which might have just forced the first condition to always evaluate true).

"might have"?

Quoted 4 lines
Quoted 3 lines

tested. Consequently, I'm more inclined to _revert_ the change for now and we can bring it back as a fix once somebody has checked that it actually works properly.

Please do not revert this change if possible.

I've gone ahead with the revert anyway, just because it's the easy thing to do and we can bring back a fixed version of the patch as a fix in the new year. So please send a new version with this fix folded in after you've tested that it doesn't cause regressions for systems without the erratum.

Cheers,

Will

#4 lore raw
Quoted 42 lines · Will Deacon
Quoted 32 lines · Anshuman Khandual
Quoted 20 lines · Will Deacon
Quoted 19 lines · Lukas Bulwahn
arch/arm64/mm/hugetlbpage.c +1 -1
index cd8d96e1fa1a..95364e8bdc19 100644
@@ -562,7 +562,7 @@ bool __init arch_hugetlb_valid_size(unsigned long size)
 
 pte_t huge_ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
-	if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198) &&
+	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
 		/*
 		 * Break-before-make (BBM) is required for all user space mappings
arch/arm64/mm/mmu.c +0 -0
index 12915f379c22..d77c9f56b7b4 100644
@@ -1633,7 +1633,7 @@ early_initcall(prevent_bootmem_remove_init);
 
Quoted 10 lines

arch/arm64/mm/mmu.c early_initcall(prevent_bootmem_remove_init); · line 1639

Quoted 7 lines
arch/arm64/mm/mmu.c +1 -1
 pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
-	if (IS_ENABLED(CONFIG_ARM64_WORKAROUND_2645198) &&
+	if (IS_ENABLED(CONFIG_ARM64_ERRATUM_2645198) &&
 	    cpus_have_const_cap(ARM64_WORKAROUND_2645198)) {
 		/*
 		 * Break-before-make (BBM) is required for all user space mappings

Grr, this bug seems to exist in all three versions of the patch reviewed on the list, so I can only draw the conclusion that this code has never been

Quoted 3 lines

Ohh, my bad, apologies. I did not have a real system with this erratum, although had emulated and tested this workaround path via some other debug changes (which might have just forced the first condition to always evaluate true).

"might have"?

Quoted 3 lines
Quoted 3 lines

tested. Consequently, I'm more inclined to _revert_ the change for now and we can bring it back as a fix once somebody has checked that it actually works properly.

Quoted 6 lines
Quoted 1 line

Please do not revert this change if possible.

I've gone ahead with the revert anyway, just because it's the easy thing to do and we can bring back a fixed version of the patch as a fix in the new year. So please send a new version with this fix folded in after you've tested that it doesn't cause regressions for systems without the erratum.

Sure, will resend. Again, apologies for this last minute merge window trouble.