[PATCH v3 0/2] KVM: arm64: ptdump: Shadow ptdump fixes

[PATCH v3 0/2] KVM: arm64: ptdump: Shadow ptdump fixes

[Wei-Lin Chang]

Hi,

This is v3 of fixing the sleep-in-atomic problem of the shadow ptdump.
The approach taken this time is simply creating a file for each nested
mmu.

The ptdump files are created alongside the nested mmus in
kvm_vcpu_init_nested(), in a sleepable context, and removed when the VM
is destroyed (kvm_destroy_vm_debugfs(), before the nested mmus are freed
in kvm_destroy_nested()). This allows us to store the kvm_s2_mmu pointers
in the ptdump code, after Marc's nested mmu lifecycle fixes [1].

For the shadow ptdump, for valid nested mmus the first line will show
the nested context's VTTBR and VTCR values, and whether s2 is enabled or
not. The ptdump then follows.

The first patch fixes mmu->pgt accesses after the VM MMU teardown in
ptdump.c, the second patch switches from per nested context ptdumps to
per nested mmu ptdumps.

Tested with CONFIG_PROVE_LOCKING, CONFIG_DEBUG_ATOMIC_SLEEP, and
CONFIG_KASAN.

Series based on kvmarm/fixes.

* Changes from v2 [2]:

  - Changed from creating a file that contains all nested mmus to
    creating one file for each nested mmu.

Thanks!

[1]: https://lore.kernel.org/kvmarm/20260911162203.1919330-1-maz@kernel.org/
[2]: https://lore.kernel.org/kvmarm/20260630121005.1130996-1-weilin.chang@arm.com/

Wei-Lin Chang (2):
  KVM: arm64: ptdump: Check the page tables aren't freed when accessing
  KVM: arm64: ptdump: Fix shadow ptdump sleep-in-atomic-context problem

 arch/arm64/include/asm/kvm_host.h |  4 --
 arch/arm64/include/asm/kvm_mmu.h  |  6 +--
 arch/arm64/kvm/nested.c           |  9 ++--
 arch/arm64/kvm/ptdump.c           | 75 ++++++++++++++++++++-----------
 4 files changed, 54 insertions(+), 40 deletions(-)

-- 
2.43.0

[PATCH v3 1/2] KVM: arm64: ptdump: Check the page tables aren't freed when accessing

[Wei-Lin Chang]

An open debugfs file keeps the KVM structure alive, but does not prevent
mmu notifier release from freeing the stage-2 page tables when the VMM’s
address space is torn down. Therefore page tables belonging to the mmus
could have been freed when a thread opens or reads the ptdump files.
Take the mmu_lock and check mmu->pgt is still alive before accessing the
page tables.

Also for the ipa_range file and stage2_levels file, switch from keeping
the pointer to the page tables to keeping the pointer to the mmu.

Fixes: 7c4f73548ed1 ("KVM: arm64: Register ptdump with debugfs on guest creation")
Reported-by: Sashiko AI <sashiko-bot@kernel.org>
Closes: https://lore.kernel.org/kvmarm/20260623144054.B91D21F000E9@smtp.kernel.org/
Signed-off-by: Wei-Lin Chang <weilin.chang@arm.com>
Cc: stable@vger.kernel.org
---
 arch/arm64/kvm/ptdump.c | 43 +++++++++++++++++++++++++++++++----------
 1 file changed, 33 insertions(+), 10 deletions(-)

diff --git a/arch/arm64/kvm/ptdump.c b/arch/arm64/kvm/ptdump.c
index 69899797dbad..4ead856c181a 100644
--- a/arch/arm64/kvm/ptdump.c
+++ b/arch/arm64/kvm/ptdump.c
@@ -115,15 +115,28 @@ static int kvm_ptdump_build_levels(struct ptdump_pg_level *level, u32 start_lvl)
 
 static struct kvm_ptdump_guest_state *kvm_ptdump_parser_create(struct kvm_s2_mmu *mmu)
 {
+	struct kvm *kvm = kvm_s2_mmu_to_kvm(mmu);
 	struct kvm_ptdump_guest_state *st;
-	struct kvm_pgtable *pgtable = mmu->pgt;
+	s8 start_level;
 	int ret;
 
+	/*
+	 * We only need the pgt start level to initialize the ptdump, get it
+	 * while holding the mmu_lock. It's fine if the pgt gets freed
+	 * afterwards, we'll check again when doing the actual dump.
+	 */
+	scoped_guard(write_lock, &kvm->mmu_lock) {
+		if (mmu->pgt)
+			start_level = mmu->pgt->start_level;
+		else
+			return ERR_PTR(-ENOENT);
+	}
+
 	st = kzalloc_obj(struct kvm_ptdump_guest_state, GFP_KERNEL_ACCOUNT);
 	if (!st)
 		return ERR_PTR(-ENOMEM);
 
-	ret = kvm_ptdump_build_levels(&st->level[0], pgtable->start_level);
+	ret = kvm_ptdump_build_levels(&st->level[0], start_level);
 	if (ret) {
 		kfree(st);
 		return ERR_PTR(ret);
@@ -149,6 +162,9 @@ static int kvm_ptdump_guest_show(struct seq_file *m, void *unused)
 	};
 
 	guard(write_lock)(&kvm->mmu_lock);
+	if (!mmu->pgt)
+		return 0;
+
 	st->parser_state = (struct ptdump_pg_state) {
 		.marker		= &st->ipa_marker[0],
 		.end_address	= BIT(mmu->pgt->ia_bits),
@@ -211,17 +227,27 @@ static const struct file_operations kvm_ptdump_guest_fops = {
 
 static int kvm_pgtable_range_show(struct seq_file *m, void *unused)
 {
-	struct kvm_pgtable *pgtable = m->private;
+	struct kvm_s2_mmu *mmu = m->private;
+	struct kvm *kvm = kvm_s2_mmu_to_kvm(mmu);
+
+	guard(write_lock)(&kvm->mmu_lock);
+
+	if (mmu->pgt)
+		seq_printf(m, "%2u\n", mmu->pgt->ia_bits);
 
-	seq_printf(m, "%2u\n", pgtable->ia_bits);
 	return 0;
 }
 
 static int kvm_pgtable_levels_show(struct seq_file *m, void *unused)
 {
-	struct kvm_pgtable *pgtable = m->private;
+	struct kvm_s2_mmu *mmu = m->private;
+	struct kvm *kvm = kvm_s2_mmu_to_kvm(mmu);
+
+	guard(write_lock)(&kvm->mmu_lock);
+
+	if (mmu->pgt)
+		seq_printf(m, "%1d\n", KVM_PGTABLE_MAX_LEVELS - mmu->pgt->start_level);
 
-	seq_printf(m, "%1d\n", KVM_PGTABLE_MAX_LEVELS - pgtable->start_level);
 	return 0;
 }
 
@@ -230,15 +256,12 @@ static int kvm_pgtable_debugfs_open(struct inode *m, struct file *file,
 {
 	struct kvm_s2_mmu *mmu = m->i_private;
 	struct kvm *kvm = kvm_s2_mmu_to_kvm(mmu);
-	struct kvm_pgtable *pgtable;
 	int ret;
 
 	if (!kvm_get_kvm_safe(kvm))
 		return -ENOENT;
 
-	pgtable = mmu->pgt;
-
-	ret = single_open(file, show, pgtable);
+	ret = single_open(file, show, mmu);
 	if (ret < 0)
 		kvm_put_kvm(kvm);
 	return ret;
-- 
2.43.0

[PATCH v3 2/2] KVM: arm64: ptdump: Fix shadow ptdump sleep-in-atomic-context problem

[Wei-Lin Chang]

Previously we exposed shadow page tables by creating a debugfs ptdump
file whenever a nested mmu instance gets bound to a new context, and
deleting the debugfs file whose context was getting unbound.

This turned out to be buggy, as the instance<->context binding process
is done with the mmu_lock held, and debugfs creation/deletion can sleep.

Change the approach and create a shadow ptdump file for each nested mmu
instead during kvm_vcpu_init_nested(), in a sleepable context. The files
will be named nested_mmu<index>, and reading it will return the nested
context's VTTBR, VTCR, and s2 enabled or not before dumping the shadow
page tables, given the nested mmu is valid.

Fixes: 19e15dc73f0f ("KVM: arm64: nv: Expose shadow page tables in debugfs")
Reported-by: Itaru Kitayama <itaru.kitayama@fujitsu.com>
Closes: https://lore.kernel.org/kvmarm/aiuF0KSvvv-ZozI1@sm-arm-grace07/
Signed-off-by: Wei-Lin Chang <weilin.chang@arm.com>
Cc: stable@vger.kernel.org
---
 arch/arm64/include/asm/kvm_host.h |  4 ----
 arch/arm64/include/asm/kvm_mmu.h  |  6 ++----
 arch/arm64/kvm/nested.c           |  9 ++++-----
 arch/arm64/kvm/ptdump.c           | 32 +++++++++++++++----------------
 4 files changed, 21 insertions(+), 30 deletions(-)

diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
index cd9b9d2462f9..db177af9149f 100644
--- a/arch/arm64/include/asm/kvm_host.h
+++ b/arch/arm64/include/asm/kvm_host.h
@@ -217,10 +217,6 @@ struct kvm_s2_mmu {
 	 */
 	bool	nested_stage2_enabled;
 
-#ifdef CONFIG_PTDUMP_STAGE2_DEBUGFS
-	struct dentry *shadow_pt_debugfs_dentry;
-#endif
-
 	/*
 	 * true when this MMU needs to be unmapped before being used for a new
 	 * purpose.
diff --git a/arch/arm64/include/asm/kvm_mmu.h b/arch/arm64/include/asm/kvm_mmu.h
index 6eae7e7e2a68..ac2b0637692a 100644
--- a/arch/arm64/include/asm/kvm_mmu.h
+++ b/arch/arm64/include/asm/kvm_mmu.h
@@ -392,12 +392,10 @@ static inline bool kvm_supports_cacheable_pfnmap(void)
 
 #ifdef CONFIG_PTDUMP_STAGE2_DEBUGFS
 void kvm_s2_ptdump_create_debugfs(struct kvm *kvm);
-void kvm_nested_s2_ptdump_create_debugfs(struct kvm_s2_mmu *mmu);
-void kvm_nested_s2_ptdump_remove_debugfs(struct kvm_s2_mmu *mmu);
+void kvm_nested_s2_ptdump_create_debugfs(struct kvm_s2_mmu *mmu, int idx);
 #else
 static inline void kvm_s2_ptdump_create_debugfs(struct kvm *kvm) {}
-static inline void kvm_nested_s2_ptdump_create_debugfs(struct kvm_s2_mmu *mmu) {}
-static inline void kvm_nested_s2_ptdump_remove_debugfs(struct kvm_s2_mmu *mmu) {}
+static inline void kvm_nested_s2_ptdump_create_debugfs(struct kvm_s2_mmu *mmu, int idx) {}
 #endif /* CONFIG_PTDUMP_STAGE2_DEBUGFS */
 
 #endif /* __ASSEMBLER__ */
diff --git a/arch/arm64/kvm/nested.c b/arch/arm64/kvm/nested.c
index b191365d97cc..41253fe45941 100644
--- a/arch/arm64/kvm/nested.c
+++ b/arch/arm64/kvm/nested.c
@@ -124,6 +124,9 @@ int kvm_vcpu_init_nested(struct kvm_vcpu *vcpu)
 			return ret;
 		}
 
+		for (i = 0; i < S2_MMU_PER_VCPU; i++)
+			kvm_nested_s2_ptdump_create_debugfs(&tmp[i], i + kvm->arch.nested_mmus_size);
+
 		guard(write_lock)(&kvm->mmu_lock);
 
 		for (i = 0; i < S2_MMU_PER_VCPU; i++)
@@ -837,10 +840,8 @@ static struct kvm_s2_mmu *get_s2_mmu_nested(struct kvm_vcpu *vcpu)
 	kvm->arch.nested_mmus_next = (i + 1) % kvm->arch.nested_mmus_size;
 
 	/* Make sure we don't forget to do the laundry */
-	if (kvm_s2_mmu_valid(s2_mmu)) {
-		kvm_nested_s2_ptdump_remove_debugfs(s2_mmu);
+	if (kvm_s2_mmu_valid(s2_mmu))
 		s2_mmu->pending_unmap = true;
-	}
 
 	/*
 	 * The virtual VMID (modulo CnP) will be used as a key when matching
@@ -854,8 +855,6 @@ static struct kvm_s2_mmu *get_s2_mmu_nested(struct kvm_vcpu *vcpu)
 	s2_mmu->tlb_vtcr = vcpu_read_sys_reg(vcpu, VTCR_EL2);
 	s2_mmu->nested_stage2_enabled = vcpu_read_sys_reg(vcpu, HCR_EL2) & HCR_VM;
 
-	kvm_nested_s2_ptdump_create_debugfs(s2_mmu);
-
 out:
 	atomic_inc(&s2_mmu->refcnt);
 
diff --git a/arch/arm64/kvm/ptdump.c b/arch/arm64/kvm/ptdump.c
index 4ead856c181a..9411d6c2dafb 100644
--- a/arch/arm64/kvm/ptdump.c
+++ b/arch/arm64/kvm/ptdump.c
@@ -17,7 +17,7 @@
 
 #define MARKERS_LEN		2
 #define KVM_PGTABLE_MAX_LEVELS	(KVM_PGTABLE_LAST_LEVEL + 1)
-#define S2FNAMESZ		sizeof("0x0123456789abcdef-0x0123456789abcdef-s2-disabled")
+#define S2FNAMESZ		sizeof("nested_mmu0000")
 
 struct kvm_ptdump_guest_state {
 	struct kvm_s2_mmu	*mmu;
@@ -173,6 +173,15 @@ static int kvm_ptdump_guest_show(struct seq_file *m, void *unused)
 		.seq		= m,
 	};
 
+	if (kvm_is_nested_s2_mmu(kvm, mmu)) {
+		if (kvm_s2_mmu_valid(mmu))
+			seq_printf(m, "VTCR: 0x%016llx VTTBR: 0x%016llx s2: %s\n",
+				   mmu->tlb_vtcr, mmu->tlb_vttbr,
+				   mmu->nested_stage2_enabled ? "enabled" : "disabled");
+		else
+			return 0;
+	}
+
 	ret = kvm_pgtable_walk(mmu->pgt, 0, BIT(mmu->pgt->ia_bits), &walker);
 	if (ret)
 		return ret;
@@ -299,26 +308,15 @@ static const struct file_operations kvm_pgtable_levels_fops = {
 	.release	= kvm_pgtable_debugfs_close,
 };
 
-void kvm_nested_s2_ptdump_create_debugfs(struct kvm_s2_mmu *mmu)
+void kvm_nested_s2_ptdump_create_debugfs(struct kvm_s2_mmu *mmu, int idx)
 {
-	struct dentry *dent;
 	char file_name[S2FNAMESZ];
 
-	snprintf(file_name, sizeof(file_name), "0x%016llx-0x%016llx-s2-%sabled",
-		 mmu->tlb_vttbr,
-		 mmu->tlb_vtcr,
-		 mmu->nested_stage2_enabled ? "en" : "dis");
-
-	dent = debugfs_create_file(file_name, 0400,
-				   mmu->arch->debugfs_nv_dentry, mmu,
-				   &kvm_ptdump_guest_fops);
+	snprintf(file_name, sizeof(file_name), "nested_mmu%d", idx);
 
-	mmu->shadow_pt_debugfs_dentry = dent;
-}
-
-void kvm_nested_s2_ptdump_remove_debugfs(struct kvm_s2_mmu *mmu)
-{
-	debugfs_remove(mmu->shadow_pt_debugfs_dentry);
+	debugfs_create_file(file_name, 0400,
+			    mmu->arch->debugfs_nv_dentry, mmu,
+			    &kvm_ptdump_guest_fops);
 }
 
 void kvm_s2_ptdump_create_debugfs(struct kvm *kvm)
-- 
2.43.0

Re: [PATCH v3 0/2] KVM: arm64: ptdump: Shadow ptdump fixes

[Itaru Kitayama]

Hi Wei-Lin,

On Thu, Sep 17, 2026 at 12:03:35AM +0100, Wei-Lin Chang wrote:
> Hi,
> 
> This is v3 of fixing the sleep-in-atomic problem of the shadow ptdump.
> The approach taken this time is simply creating a file for each nested
> mmu.
> 
> The ptdump files are created alongside the nested mmus in
> kvm_vcpu_init_nested(), in a sleepable context, and removed when the VM
> is destroyed (kvm_destroy_vm_debugfs(), before the nested mmus are freed
> in kvm_destroy_nested()). This allows us to store the kvm_s2_mmu pointers
> in the ptdump code, after Marc's nested mmu lifecycle fixes [1].
> 
> For the shadow ptdump, for valid nested mmus the first line will show
> the nested context's VTTBR and VTCR values, and whether s2 is enabled or
> not. The ptdump then follows.
> 
> The first patch fixes mmu->pgt accesses after the VM MMU teardown in
> ptdump.c, the second patch switches from per nested context ptdumps to
> per nested mmu ptdumps.
> 
> Tested with CONFIG_PROVE_LOCKING, CONFIG_DEBUG_ATOMIC_SLEEP, and
> CONFIG_KASAN.
> 
> Series based on kvmarm/fixes.
> 
> * Changes from v2 [2]:
> 
>   - Changed from creating a file that contains all nested mmus to
>     creating one file for each nested mmu.

I've booted your kernel built only CONFIG_DEBUG_ATOMIC_SLEEP enabled
mentioned above kernel configs; on Grace CPUs shadow_stage2 test did 
not trigger the sleep in atomic message.

Tested-by: Itaru Kitayama <itaru.kitayama@fujitsu.com>

Thanks,
Itaru.

> 
> Thanks!
> 
> [1]: https://lore.kernel.org/kvmarm/20260911162203.1919330-1-maz@kernel.org/
> [2]: https://lore.kernel.org/kvmarm/20260630121005.1130996-1-weilin.chang@arm.com/
> 
> Wei-Lin Chang (2):
>   KVM: arm64: ptdump: Check the page tables aren't freed when accessing
>   KVM: arm64: ptdump: Fix shadow ptdump sleep-in-atomic-context problem
> 
>  arch/arm64/include/asm/kvm_host.h |  4 --
>  arch/arm64/include/asm/kvm_mmu.h  |  6 +--
>  arch/arm64/kvm/nested.c           |  9 ++--
>  arch/arm64/kvm/ptdump.c           | 75 ++++++++++++++++++++-----------
>  4 files changed, 54 insertions(+), 40 deletions(-)
> 
> -- 
> 2.43.0
> 

Re: [PATCH v3 1/2] KVM: arm64: ptdump: Check the page tables aren't freed when accessing

[Vincent Donnefort]

On Thu, Sep 17, 2026 at 12:03:36AM +0100, Wei-Lin Chang wrote:
> An open debugfs file keeps the KVM structure alive, but does not prevent
> mmu notifier release from freeing the stage-2 page tables when the VMM’s
> address space is torn down. Therefore page tables belonging to the mmus
> could have been freed when a thread opens or reads the ptdump files.
> Take the mmu_lock and check mmu->pgt is still alive before accessing the
> page tables.

As Sashiko said, the read_lock is probably enough, including the existing one in
kvm_ptdump_guest_show()

With that change:

Reviewed-by: Vincent Donnefort <vdonnefort@google.com>
Tested-by: Vincent Donnefort <vdonnefort@google.com>

> 
> Also for the ipa_range file and stage2_levels file, switch from keeping
> the pointer to the page tables to keeping the pointer to the mmu.
> 
> Fixes: 7c4f73548ed1 ("KVM: arm64: Register ptdump with debugfs on guest creation")
> Reported-by: Sashiko AI <sashiko-bot@kernel.org>
> Closes: https://lore.kernel.org/kvmarm/20260623144054.B91D21F000E9@smtp.kernel.org/
> Signed-off-by: Wei-Lin Chang <weilin.chang@arm.com>
> Cc: stable@vger.kernel.org
> ---
>  arch/arm64/kvm/ptdump.c | 43 +++++++++++++++++++++++++++++++----------
>  1 file changed, 33 insertions(+), 10 deletions(-)
> 
> diff --git a/arch/arm64/kvm/ptdump.c b/arch/arm64/kvm/ptdump.c
> index 69899797dbad..4ead856c181a 100644
> --- a/arch/arm64/kvm/ptdump.c
> +++ b/arch/arm64/kvm/ptdump.c
> @@ -115,15 +115,28 @@ static int kvm_ptdump_build_levels(struct ptdump_pg_level *level, u32 start_lvl)
>  
>  static struct kvm_ptdump_guest_state *kvm_ptdump_parser_create(struct kvm_s2_mmu *mmu)
>  {
> +	struct kvm *kvm = kvm_s2_mmu_to_kvm(mmu);
>  	struct kvm_ptdump_guest_state *st;
> -	struct kvm_pgtable *pgtable = mmu->pgt;
> +	s8 start_level;
>  	int ret;
>  
> +	/*
> +	 * We only need the pgt start level to initialize the ptdump, get it
> +	 * while holding the mmu_lock. It's fine if the pgt gets freed
> +	 * afterwards, we'll check again when doing the actual dump.
> +	 */
> +	scoped_guard(write_lock, &kvm->mmu_lock) {
> +		if (mmu->pgt)
> +			start_level = mmu->pgt->start_level;
> +		else
> +			return ERR_PTR(-ENOENT);
> +	}
> +
>  	st = kzalloc_obj(struct kvm_ptdump_guest_state, GFP_KERNEL_ACCOUNT);
>  	if (!st)
>  		return ERR_PTR(-ENOMEM);
>  
> -	ret = kvm_ptdump_build_levels(&st->level[0], pgtable->start_level);
> +	ret = kvm_ptdump_build_levels(&st->level[0], start_level);
>  	if (ret) {
>  		kfree(st);
>  		return ERR_PTR(ret);
> @@ -149,6 +162,9 @@ static int kvm_ptdump_guest_show(struct seq_file *m, void *unused)
>  	};
>  
>  	guard(write_lock)(&kvm->mmu_lock);
> +	if (!mmu->pgt)
> +		return 0;
> +
>  	st->parser_state = (struct ptdump_pg_state) {
>  		.marker		= &st->ipa_marker[0],
>  		.end_address	= BIT(mmu->pgt->ia_bits),
> @@ -211,17 +227,27 @@ static const struct file_operations kvm_ptdump_guest_fops = {
>  
>  static int kvm_pgtable_range_show(struct seq_file *m, void *unused)
>  {
> -	struct kvm_pgtable *pgtable = m->private;
> +	struct kvm_s2_mmu *mmu = m->private;
> +	struct kvm *kvm = kvm_s2_mmu_to_kvm(mmu);
> +
> +	guard(write_lock)(&kvm->mmu_lock);
> +
> +	if (mmu->pgt)
> +		seq_printf(m, "%2u\n", mmu->pgt->ia_bits);
>  
> -	seq_printf(m, "%2u\n", pgtable->ia_bits);
>  	return 0;
>  }
>  
>  static int kvm_pgtable_levels_show(struct seq_file *m, void *unused)
>  {
> -	struct kvm_pgtable *pgtable = m->private;
> +	struct kvm_s2_mmu *mmu = m->private;
> +	struct kvm *kvm = kvm_s2_mmu_to_kvm(mmu);
> +
> +	guard(write_lock)(&kvm->mmu_lock);
> +
> +	if (mmu->pgt)
> +		seq_printf(m, "%1d\n", KVM_PGTABLE_MAX_LEVELS - mmu->pgt->start_level);
>  
> -	seq_printf(m, "%1d\n", KVM_PGTABLE_MAX_LEVELS - pgtable->start_level);
>  	return 0;
>  }
>  
> @@ -230,15 +256,12 @@ static int kvm_pgtable_debugfs_open(struct inode *m, struct file *file,
>  {
>  	struct kvm_s2_mmu *mmu = m->i_private;
>  	struct kvm *kvm = kvm_s2_mmu_to_kvm(mmu);
> -	struct kvm_pgtable *pgtable;
>  	int ret;
>  
>  	if (!kvm_get_kvm_safe(kvm))
>  		return -ENOENT;
>  
> -	pgtable = mmu->pgt;
> -
> -	ret = single_open(file, show, pgtable);
> +	ret = single_open(file, show, mmu);
>  	if (ret < 0)
>  		kvm_put_kvm(kvm);
>  	return ret;
> -- 
> 2.43.0
> 

-- 
Vincent