Skip to contentSkip to conversation

riscv: Assorted bug fixes

PATCH In review Michael Neuling 4 participants 19 messages Apr 9 – May 2

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

cover riscv: Assorted bug fixes

This series contains five independent bug fixes across the RISC-V architecture code, found with the help of Claude AI (claude-4.6-opus):

1. riscv: errata: Fix bitwise vs logical AND in MIPS errata patching
- Logical AND (&&) was used instead of bitwise AND (&) when checking errata flags, causing all errata patches to be applied when any single one was detected.

2. riscv: ptrace: Fix register corruption in compat_riscv_gpr_set on error
- On copyin failure, uninitialized stack data was written into the target task's pt_regs, corrupting registers and potentially leaking kernel stack contents.

3. riscv: mm: Fix NULL pointer dereference in __set_memory
- find_vm_area() can return NULL but was dereferenced without a check.

4. riscv: mm: Fix NULL dereferences in napot hugetlb functions
- huge_pte_offset() can return NULL in the napot path but was used without NULL checks in huge_ptep_set_access_flags() and huge_ptep_set_wrprotect().

5. riscv: mm: Fix TOCTOU race in remove_pte_mapping
- The PTE was read twice (once via ptep_get, once via *ptep), creating a race window where another CPU could modify the PTE between reads.

All patches are small and self-contained. These bugs were identified through AI-assisted code review using Claude, which also assisted in writing the fixes. Each patch has been manually reviewed for correctness.

Michael Neuling (5):
riscv: errata: Fix bitwise vs logical AND in MIPS errata patching
riscv: ptrace: Fix register corruption in compat_riscv_gpr_set on
error
riscv: mm: Fix NULL pointer dereference in __set_memory
riscv: mm: Fix NULL dereferences in napot hugetlb functions
riscv: mm: Fix TOCTOU race in remove_pte_mapping

 arch/riscv/errata/mips/errata.c | 2 +-
 arch/riscv/kernel/ptrace.c      | 4 ++--
 arch/riscv/mm/hugetlbpage.c     | 4 ++++
 arch/riscv/mm/init.c            | 2 +-
 arch/riscv/mm/pageattr.c        | 4 ++++
 5 files changed, 12 insertions(+), 4 deletions(-)
Signature · 1 line
2.43.0
#1 lore raw

PATCH 1/5 riscv: errata: Fix bitwise vs logical AND in MIPS errata patching

The condition checking whether a specific errata needs patching uses logical AND (&&) instead of bitwise AND (&). Since logical AND only checks that both operands are non-zero, this causes all errata patches to be applied whenever any single errata is detected, rather than only applying the matching one.

The SiFive errata implementation correctly uses bitwise AND for the same check.

Fixes 0b0ca959d2 ("riscv: errata: Fix the PAUSE Opcode for MIPS P8700")
Assisted-by Cursor:claude-4.6-opus-high-thinking
---
 arch/riscv/errata/mips/errata.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/riscv/errata/mips/errata.c +1 -1
index e984a81522..2c3dc2259e 100644
@@ -57,7 +57,7 @@ void mips_errata_patch_func(struct alt_entry *begin, struct alt_entry *end,
 		}
 
 		tmp = (1U << alt->patch_id);
-		if (cpu_req_errata && tmp) {
+		if (cpu_req_errata & tmp) {
 			mutex_lock(&text_mutex);
 			patch_text_nosync(ALT_OLD_PTR(alt), ALT_ALT_PTR(alt),
 					  alt->alt_len);
Signature · 1 line
2.43.0
maintainer
#2 lore raw

Re: [PATCH 1/5] riscv: errata: Fix bitwise vs logical AND in MIPS errata patching

Quoted 6 lines · Michael Neuling

The condition checking whether a specific errata needs patching uses logical AND (&&) instead of bitwise AND (&). Since logical AND only checks that both operands are non-zero, this causes all errata patches to be applied whenever any single errata is detected, rather than only applying the matching one.

Quoted 6 lines

The SiFive errata implementation correctly uses bitwise AND for the same check.

Fixes 0b0ca959d2 ("riscv: errata: Fix the PAUSE Opcode for MIPS P8700")
Assisted-by Cursor:claude-4.6-opus-high-thinking

Thanks, after fixing the checkpatch warning regarding the Fixes: line in the patch description, queued for v7.1-rc.

- Paul

#3 lore raw

PATCH 2/5 riscv: ptrace: Fix register corruption in compat_riscv_gpr_set on error

In reply to Michael Neuling

compat_riscv_gpr_set() calls cregs_to_regs() unconditionally, even when user_regset_copyin() fails. Since cregs is an uninitialized stack variable, a copyin failure causes uninitialized stack data to be written into the target task's pt_regs, corrupting its register state and potentially leaking kernel stack contents.

Only call cregs_to_regs() when user_regset_copyin() succeeds.

Fixes 4608c15959 ("riscv: compat: ptrace: Add compat_arch_ptrace implement")
Assisted-by Cursor:claude-4.6-opus-high-thinking
---
 arch/riscv/kernel/ptrace.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/riscv/kernel/ptrace.c +2 -2
index 93de2e7a30..793bcee461 100644
@@ -577,8 +577,8 @@ static int compat_riscv_gpr_set(struct task_struct *target,
 	struct compat_user_regs_struct cregs;
 
 	ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &cregs, 0, -1);
-
-	cregs_to_regs(&cregs, task_pt_regs(target));
+	if (!ret)
+		cregs_to_regs(&cregs, task_pt_regs(target));
 
 	return ret;
 }
Signature · 1 line
2.43.0
maintainer
#4 lore raw

Re: [PATCH 2/5] riscv: ptrace: Fix register corruption in compat_riscv_gpr_set on error

Hi Mikey,

Quoted 11 lines · Michael Neuling

compat_riscv_gpr_set() calls cregs_to_regs() unconditionally, even when user_regset_copyin() fails. Since cregs is an uninitialized stack variable, a copyin failure causes uninitialized stack data to be written into the target task's pt_regs, corrupting its register state and potentially leaking kernel stack contents.

Only call cregs_to_regs() when user_regset_copyin() succeeds.

Fixes 4608c15959 ("riscv: compat: ptrace: Add compat_arch_ptrace implement")
Assisted-by Cursor:claude-4.6-opus-high-thinking

Thanks. I'm wondering if we have the same issue in compat_restore_sigcontext(). Care to add that to your patch if you agree?

- Paul

#5 lore raw

Re: [PATCH 2/5] riscv: ptrace: Fix register corruption in compat_riscv_gpr_set on error

Quoted 2 lines

Thanks. I'm wondering if we have the same issue in compat_restore_sigcontext(). Care to add that to your patch if you agree?

Agreed, resending.

Mikey

#6 lore raw

PATCH v2 1/1 riscv: Fix register corruption from uninitialized cregs on error

In reply to Paul Walmsley

compat_riscv_gpr_set() calls cregs_to_regs() unconditionally, even when user_regset_copyin() fails. Since cregs is an uninitialized stack variable, a copyin failure causes uninitialized stack data to be written into the target task's pt_regs, corrupting its register state and potentially leaking kernel stack contents.

compat_restore_sigcontext() has the same issue: it calls cregs_to_regs() even when __copy_from_user() fails, leading to the same corruption of the signal-returning task's register state on error.

Only call cregs_to_regs() when the user copy succeeds.

Fixes 4608c159594f ("riscv: compat: ptrace: Add compat_arch_ptrace implement")
Fixes 7383ee05314b ("riscv: compat: signal: Add rt_frame implementation")
Assisted-by Cursor:claude-4.6-opus-high-thinking
---
 arch/riscv/kernel/compat_signal.c | 2 ++
 arch/riscv/kernel/ptrace.c        | 4 ++--
 2 files changed, 4 insertions(+), 2 deletions(-)
arch/riscv/kernel/compat_signal.c +2 -0
index 6ec4e34255..cf3eb33a11 100644
@@ -107,6 +107,8 @@ static long compat_restore_sigcontext(struct pt_regs *regs,
 
 	/* sc_regs is structured the same as the start of pt_regs */
 	err = __copy_from_user(&cregs, &sc->sc_regs, sizeof(sc->sc_regs));
+	if (unlikely(err))
+		return err;
 
 	cregs_to_regs(&cregs, regs);
 
arch/riscv/kernel/ptrace.c +2 -2
index 93de2e7a30..793bcee461 100644
@@ -577,8 +577,8 @@ static int compat_riscv_gpr_set(struct task_struct *target,
 	struct compat_user_regs_struct cregs;
 
 	ret = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &cregs, 0, -1);
-
-	cregs_to_regs(&cregs, task_pt_regs(target));
+	if (!ret)
+		cregs_to_regs(&cregs, task_pt_regs(target));
 
 	return ret;
 }
Signature · 1 line
2.43.0
maintainer
#7 lore raw

Re: [PATCH v2] riscv: Fix register corruption from uninitialized cregs on error

Quoted 10 lines · Michael Neuling

compat_riscv_gpr_set() calls cregs_to_regs() unconditionally, even when user_regset_copyin() fails. Since cregs is an uninitialized stack variable, a copyin failure causes uninitialized stack data to be written into the target task's pt_regs, corrupting its register state and potentially leaking kernel stack contents.

compat_restore_sigcontext() has the same issue: it calls cregs_to_regs() even when __copy_from_user() fails, leading to the same corruption of the signal-returning task's register state on error.

Quoted 6 lines

Only call cregs_to_regs() when the user copy succeeds.

Fixes 4608c159594f ("riscv: compat: ptrace: Add compat_arch_ptrace implement")
Fixes 7383ee05314b ("riscv: compat: signal: Add rt_frame implementation")
Assisted-by Cursor:claude-4.6-opus-high-thinking

Thanks very much; queued for v7.1-rc.

- Paul

#8 lore raw

PATCH 3/5 riscv: mm: Fix NULL pointer dereference in __set_memory

In reply to Michael Neuling

find_vm_area() can return NULL if no vm_struct covers the given address. The code immediately dereferences area->addr without a NULL check. While is_vmalloc_or_module_addr() confirms the address falls within the vmalloc/module address range, it does not guarantee the address belongs to an active allocation, so find_vm_area() may still return NULL.

Add the missing NULL check.

Fixes 311cd2f6e2 ("riscv: Fix set_memory_XX() and set_direct_map_XX() by splitting huge linear mappings")
Assisted-by Cursor:claude-4.6-opus-high-thinking
---
 arch/riscv/mm/pageattr.c | 4 ++++
 1 file changed, 4 insertions(+)
arch/riscv/mm/pageattr.c +4 -0
index 3f76db3d27..46a999c86b 100644
@@ -289,6 +289,10 @@ static int __set_memory(unsigned long addr, int numpages, pgprot_t set_mask,
 		int i, page_start;
 
 		area = find_vm_area((void *)start);
+		if (!area) {
+			ret = -EINVAL;
+			goto unlock;
+		}
 		page_start = (start - (unsigned long)area->addr) >> PAGE_SHIFT;
 
 		for (i = page_start; i < page_start + numpages; ++i) {
Signature · 1 line
2.43.0
#9 lore raw

Re: [PATCH 3/5] riscv: mm: Fix NULL pointer dereference in __set_memory

Quoted 20 lines · Michael Neuling

find_vm_area() can return NULL if no vm_struct covers the given address. The code immediately dereferences area->addr without a NULL check. While is_vmalloc_or_module_addr() confirms the address falls within the vmalloc/module address range, it does not guarantee the address belongs to an active allocation, so find_vm_area() may still return NULL.

Add the missing NULL check.

Fixes 311cd2f6e2 ("riscv: Fix set_memory_XX() and set_direct_map_XX() by splitting huge linear mappings")
Assisted-by Cursor:claude-4.6-opus-high-thinking
---
 arch/riscv/mm/pageattr.c | 4 ++++
 1 file changed, 4 insertions(+)
arch/riscv/mm/pageattr.c +0 -0
index 3f76db3d27..46a999c86b 100644

arch/riscv/mm/pageattr.c static int __set_memory(unsigned long addr, int numpages, pgprot_t set_mask, · line 298

Quoted 10 lines
arch/riscv/mm/pageattr.c +4 -0
@@ -289,6 +289,10 @@ static int __set_memory(unsigned long addr, int numpages, pgprot_t set_mask,
 		int i, page_start;
 
 		area = find_vm_area((void *)start);
+		if (!area) {
+			ret = -EINVAL;
+			goto unlock;
+		}
 		page_start = (start - (unsigned long)area->addr) >> PAGE_SHIFT;
 
 		for (i = page_start; i < page_start + numpages; ++i) {

Which caller would end up calling __set_memory() in such a way?

Signature · 3 lines
Cheers,

David
#10 lore raw

Re: [PATCH 3/5] riscv: mm: Fix NULL pointer dereference in __set_memory

Quoted 10 lines

arch/riscv/mm/pageattr.c static int __set_memory(unsigned long addr, int numpages, pgprot_t set_mask, · line 298

Quoted 8 lines
              area = find_vm_area((void *)start);
arch/riscv/mm/pageattr.c +4 -0
+             if (!area) {
+                     ret = -EINVAL;
+                     goto unlock;
+             }
              page_start = (start - (unsigned long)area->addr) >> PAGE_SHIFT;
 
              for (i = page_start; i < page_start + numpages; ++i) {

Which caller would end up calling __set_memory() in such a way?

It wouldn't. You're right on this and the other two. Sorry for the noise.

Mikey

#11 lore raw

Re: [PATCH 3/5] riscv: mm: Fix NULL pointer dereference in __set_memory

Quoted 7 lines · Michael Neuling
Quoted 7 lines
Quoted 7 lines
              area = find_vm_area((void *)start);
arch/riscv/mm/pageattr.c +4 -0
+             if (!area) {
+                     ret = -EINVAL;
+                     goto unlock;
+             }
              page_start = (start - (unsigned long)area->addr) >> PAGE_SHIFT;
Quoted 5 lines
Quoted 3 lines
Quoted 1 line
              for (i = page_start; i < page_start + numpages; ++i) {

Which caller would end up calling __set_memory() in such a way?

It wouldn't. You're right on this and the other two. Sorry for the noise.

Slopped-by: Cursor:claude-4.6-opus-high-thinking

:)

Patch #1 seems reasonable (although non-critical); not sure about patch #2, leaving both to Risc-V people!

Signature · 3 lines
Cheers,

David
#12 lore raw

Re: [PATCH 3/5] riscv: mm: Fix NULL pointer dereference in __set_memory

Quoted 10 lines · David Hildenbrand (Arm)
Quoted 9 lines · Michael Neuling
Quoted 9 lines
Quoted 8 lines
              area = find_vm_area((void *)start);
arch/riscv/mm/pageattr.c +4 -0
+             if (!area) {
+                     ret = -EINVAL;
+                     goto unlock;
+             }
              page_start = (start - (unsigned long)area->addr) >> PAGE_SHIFT;
 
              for (i = page_start; i < page_start + numpages; ++i) {
Quoted 7 lines
Quoted 3 lines
Quoted 1 line

Which caller would end up calling __set_memory() in such a way?

It wouldn't. You're right on this and the other two. Sorry for the noise.

Slopped-by: Cursor:claude-4.6-opus-high-thinking

:)

It's a general tendency of LLMs to generate overly defensive code and raise concerns about theoretical issues.
Maybe it's worth taking into account when writing prompts specific to kernel development.

Quoted 7 lines

Patch #1 seems reasonable (although non-critical); not sure about patch #2, leaving both to Risc-V people!

Signature · 3 lines
Cheers,

David
Signature · 2 lines
Sincerely yours,
Mike.
#13 lore raw

Re: [PATCH 3/5] riscv: mm: Fix NULL pointer dereference in __set_memory

Quoted 5 lines · Mike Rapoport
Quoted 4 lines · David Hildenbrand (Arm)
Quoted 2 lines · Michael Neuling

It wouldn't. You're right on this and the other two. Sorry for the noise.

Quoted 8 lines
Quoted 3 lines

Slopped-by: Cursor:claude-4.6-opus-high-thinking

:)

It's a general tendency of LLMs to generate overly defensive code and raise concerns about theoretical issues.
Maybe it's worth taking into account when writing prompts specific to kernel development.

Yes, and I can only urge people to actually check what the LLM spits out, so they won't lose their credibility with maintainers ;)

Signature · 3 lines
Cheers,

David
#14 lore raw

Re: [PATCH 3/5] riscv: mm: Fix NULL pointer dereference in __set_memory

Quoted 7 lines · David Hildenbrand (Arm)
Quoted 5 lines · Mike Rapoport
Quoted 4 lines · David Hildenbrand (Arm)
Quoted 2 lines · Michael Neuling

It wouldn't. You're right on this and the other two. Sorry for the noise.

Quoted 8 lines
Quoted 8 lines
Quoted 3 lines

Slopped-by: Cursor:claude-4.6-opus-high-thinking

:)

It's a general tendency of LLMs to generate overly defensive code and raise concerns about theoretical issues.
Maybe it's worth taking into account when writing prompts specific to kernel development.

We've had some success using it to review some patches for maintainers.

I was attempting to use it on existing code but I agree this was overly defensive
and didn't consider the broader context.

Quoted 2 lines

Yes, and I can only urge people to actually check what the LLM spits out, so they won't lose their credibility with maintainers ;)

I did actually dismiss more than half of what it produced. Maybe that makes it worse :-)

I'm not as familiar with mm so it was a mistake on my part to send these.

Mikey

#15 lore raw

PATCH 4/5 riscv: mm: Fix NULL dereferences in napot hugetlb functions

In reply to Michael Neuling

huge_pte_offset() can return NULL when any level of the page table walk encounters a non-present entry. Both huge_ptep_set_access_flags() and huge_ptep_set_wrprotect() re-derive ptep via huge_pte_offset() in the napot path but use the result without a NULL check, leading to NULL pointer dereferences in get_clear_contig_flush() and set_pte_at().

Add NULL checks after huge_pte_offset() in both functions.

Fixes 82a1a1f3bf ("riscv: mm: support Svnapot in hugetlb page")
Assisted-by Cursor:claude-4.6-opus-high-thinking
---
 arch/riscv/mm/hugetlbpage.c | 4 ++++
 1 file changed, 4 insertions(+)
arch/riscv/mm/hugetlbpage.c +4 -0
index a6d217112c..7d155341cf 100644
@@ -288,6 +288,8 @@ int huge_ptep_set_access_flags(struct vm_area_struct *vma,
 	order = napot_cont_order(pte);
 	pte_num = napot_pte_num(order);
 	ptep = huge_pte_offset(mm, addr, napot_cont_size(order));
+	if (!ptep)
+		return 0;
 	orig_pte = get_clear_contig_flush(mm, addr, ptep, pte_num);
 
 	if (pte_dirty(orig_pte))
@@ -335,6 +337,8 @@ void huge_ptep_set_wrprotect(struct mm_struct *mm,
 	order = napot_cont_order(pte);
 	pte_num = napot_pte_num(order);
 	ptep = huge_pte_offset(mm, addr, napot_cont_size(order));
+	if (!ptep)
+		return;
 	orig_pte = get_clear_contig_flush(mm, addr, ptep, pte_num);
 
 	orig_pte = pte_wrprotect(orig_pte);
Signature · 1 line
2.43.0
#16 lore raw

Re: [PATCH 4/5] riscv: mm: Fix NULL dereferences in napot hugetlb functions

Quoted 29 lines · Michael Neuling

huge_pte_offset() can return NULL when any level of the page table walk encounters a non-present entry. Both huge_ptep_set_access_flags() and huge_ptep_set_wrprotect() re-derive ptep via huge_pte_offset() in the napot path but use the result without a NULL check, leading to NULL pointer dereferences in get_clear_contig_flush() and set_pte_at().

Add NULL checks after huge_pte_offset() in both functions.

Fixes 82a1a1f3bf ("riscv: mm: support Svnapot in hugetlb page")
Assisted-by Cursor:claude-4.6-opus-high-thinking
---
 arch/riscv/mm/hugetlbpage.c | 4 ++++
 1 file changed, 4 insertions(+)
arch/riscv/mm/hugetlbpage.c +2 -0
index a6d217112c..7d155341cf 100644
@@ -288,6 +288,8 @@ int huge_ptep_set_access_flags(struct vm_area_struct *vma,
 	order = napot_cont_order(pte);
 	pte_num = napot_pte_num(order);
 	ptep = huge_pte_offset(mm, addr, napot_cont_size(order));
+	if (!ptep)
+		return 0;
 	orig_pte = get_clear_contig_flush(mm, addr, ptep, pte_num);
 
 	if (pte_dirty(orig_pte))

arch/riscv/mm/hugetlbpage.c void huge_ptep_set_wrprotect(struct mm_struct *mm, · line 344

Quoted 8 lines
arch/riscv/mm/hugetlbpage.c +2 -0
@@ -335,6 +337,8 @@ void huge_ptep_set_wrprotect(struct mm_struct *mm,
 	order = napot_cont_order(pte);
 	pte_num = napot_pte_num(order);
 	ptep = huge_pte_offset(mm, addr, napot_cont_size(order));
+	if (!ptep)
+		return;
 	orig_pte = get_clear_contig_flush(mm, addr, ptep, pte_num);
 
 	orig_pte = pte_wrprotect(orig_pte);

These functions are called when we previously verified that there is indeed a pte mapped. And while holding the PTL to protect concurrent unmapping.

So how should this possibly trigger?

Signature · 3 lines
Cheers,

David
#17 lore raw

PATCH 5/5 riscv: mm: Fix TOCTOU race in remove_pte_mapping

In reply to Michael Neuling

remove_pte_mapping() reads the PTE via ptep_get() (a READ_ONCE) into a local variable, but then checks pte_present(*ptep) by dereferencing the pointer directly, reading the PTE a second time. If another CPU modifies the PTE between the two reads, pte_present may check a different value than what was captured, and the subsequent pte_page() could derive the wrong page to free.

Use the already-captured local pte variable for the pte_present check.

Fixes c75a74f4ba ("riscv: mm: Add memory hotplugging support")
Assisted-by Cursor:claude-4.6-opus-high-thinking
---
 arch/riscv/mm/init.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/riscv/mm/init.c +1 -1
index 23cc1b81fa..873cc860a1 100644
@@ -1562,7 +1562,7 @@ static void __meminit remove_pte_mapping(pte_t *pte_base, unsigned long addr, un
 
 		ptep = pte_base + pte_index(addr);
 		pte = ptep_get(ptep);
-		if (!pte_present(*ptep))
+		if (!pte_present(pte))
 			continue;
 
 		pte_clear(&init_mm, addr, ptep);
Signature · 1 line
2.43.0
#18 lore raw

Re: [PATCH 5/5] riscv: mm: Fix TOCTOU race in remove_pte_mapping

Quoted 4 lines · Michael Neuling

remove_pte_mapping() reads the PTE via ptep_get() (a READ_ONCE) into a local variable, but then checks pte_present(*ptep) by dereferencing the pointer directly, reading the PTE a second time. If another CPU modifies the PTE between the two reads

Is that even possible?

The code does not use any locking, so nothing would be safe here if races could happen, no?

Signature · 3 lines
Cheers,

David