[PATCH v2 0/2] arm64: ptdump flush fixes

[PATCH v2 0/2] arm64: ptdump flush fixes

[Wei-Lin Chang]

Hi,

This series fixes two problems around ptdumps:

1. note_page_flush(), which flushes out the last row of ptdumps, does
   not account for address spacees that have IA < 64. Other than making
   the last region extremely huge, the attributes of the last region
   within the address spaces appear to extend all the way to 1 << 64.

2. KVM/arm64's stage-2 ptdump missed calling note_page_flush().

Please see the patches for more detail.

* Changes from v1 [1]:

  - Instead of manually calling note_page() for flushing, fix
    note_page_flush() so that it ends the ptdump at the end of the
    address space.

  - Changed the start address of the second marker to ULONG_MAX for KVM
    ptdump, so we don't output extra marker names, and advance past the
    end of the marker array.

Thanks!

[1]: https://lore.kernel.org/kvmarm/20260717231233.2299068-1-weilin.chang@arm.com/

Wei-Lin Chang (2):
  arm64: ptdump: Make note_page_flush() range aware
  KVM: arm64: ptdump: Flush the last region

 arch/arm64/kvm/ptdump.c | 16 ++++++++++++----
 arch/arm64/mm/ptdump.c  | 17 ++++++++++++++++-
 2 files changed, 28 insertions(+), 5 deletions(-)


base-commit: 679d7201c1f09e37fa1c12ce28d84079c17fc87f
-- 
2.43.0

[PATCH v2 1/2] arm64: ptdump: Make note_page_flush() range aware

[Wei-Lin Chang]

note_page_flush() calls note_page() with addr == 0 and level == -1 to
dump the last row of a ptdump. addr == 0 (1 << 64 wrapped around)
renders a huge region with enormous size for address spaces with
IA bits < 64. For example the stage-2 page tables and the EFI runtime
page table.

More importantly, the last region of the address space and everything
after the address space up to 1 << 64 are merged into one row of
output. If the last region within the address space is valid, it will
appear to remain valid up to 1 << 64 with the same attributes.

Currently only the EFI runtime ptdump is affected by this, but KVM will
soon fix its stage-2 ptdump by using note_page_flush(). Here is an
example of an EFI runtime ptdump (last row):

0x0000008000000000-0x0000000000000000   17179868672G PGD

With this patch:

0x0000008000000000-0x0001000000000000      261632G PGD

To fix this, use ptdump_state.range[] to figure out where the address
space ends, and call note_page() with that.

Fixes: 9d80448ac92b ("efi/arm64: Add debugfs node to dump UEFI runtime page tables")
Signed-off-by: Wei-Lin Chang <weilin.chang@arm.com>
---
 arch/arm64/mm/ptdump.c | 17 ++++++++++++++++-
 1 file changed, 16 insertions(+), 1 deletion(-)

diff --git a/arch/arm64/mm/ptdump.c b/arch/arm64/mm/ptdump.c
index 1c20144700d7..5b34060ceb96 100644
--- a/arch/arm64/mm/ptdump.c
+++ b/arch/arm64/mm/ptdump.c
@@ -278,9 +278,24 @@ void note_page_pgd(struct ptdump_state *pt_st, unsigned long addr, pgd_t pgd)
 
 void note_page_flush(struct ptdump_state *pt_st)
 {
+	const struct ptdump_range *range = pt_st->range;
+	unsigned long end = 0;
 	pte_t pte_zero = {0};
 
-	note_page(pt_st, 0, -1, pte_val(pte_zero));
+	while (range->start != range->end) {
+		end = range->end;
+		range++;
+	}
+
+	/*
+	 * Address spaces that end at 1 << 64 have range->end == ULONG_MAX,
+	 * but note_page() expects the exclusive end. In this case adjust end
+	 * to the wraparound value 0.
+	 */
+	if (end == ULONG_MAX)
+		end = 0;
+
+	note_page(pt_st, end, -1, pte_val(pte_zero));
 }
 
 static void arm64_ptdump_walk_pgd(struct ptdump_state *st, struct mm_struct *mm)
-- 
2.43.0

[PATCH v2 2/2] KVM: arm64: ptdump: Flush the last region

[Wei-Lin Chang]

Currently the stage-2 ptdump calls note_page() at each leaf entry visit.
This simply misses the output of the last region, because note_page()
only dumps output when it detects a change in level/prot, or when the
walk enters a next marker section. The last region in the guest IPA
space with the same level/prot is not dumped since there is no change
after it.

Call note_page_flush() to dump the final region. note_page_flush()
expects ptdump_state.range[] to be initialized to the range of the
address space, so also do that.

Also change the second marker's start address to ULONG_MAX so we never
cross it. This avoids dumping redundant marker names (which are NULL),
and advancing beyond the end of the marker array.

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/20260630122758.891011F00A3A@smtp.kernel.org/
Signed-off-by: Wei-Lin Chang <weilin.chang@arm.com>
---
 arch/arm64/kvm/ptdump.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

diff --git a/arch/arm64/kvm/ptdump.c b/arch/arm64/kvm/ptdump.c
index c9140e22abcf..222577163d2b 100644
--- a/arch/arm64/kvm/ptdump.c
+++ b/arch/arm64/kvm/ptdump.c
@@ -130,7 +130,7 @@ static struct kvm_ptdump_guest_state *kvm_ptdump_parser_create(struct kvm_s2_mmu
 	}
 
 	st->ipa_marker[0].name		= "Guest IPA";
-	st->ipa_marker[1].start_address = BIT(pgtable->ia_bits);
+	st->ipa_marker[1].start_address = ULONG_MAX;
 
 	st->mmu				= mmu;
 	return st;
@@ -148,18 +148,26 @@ static int kvm_ptdump_guest_show(struct seq_file *m, void *unused)
 		.flags	= KVM_PGTABLE_WALK_LEAF,
 	};
 
+	guard(write_lock)(&kvm->mmu_lock);
 	st->parser_state = (struct ptdump_pg_state) {
 		.marker		= &st->ipa_marker[0],
 		.level		= -1,
 		.pg_level	= &st->level[0],
 		.seq		= m,
+		.ptdump		= {
+			.range = (struct ptdump_range[]){
+				{0, BIT(mmu->pgt->ia_bits)},
+				{0, 0}
+			}
+		}
 	};
 
-	write_lock(&kvm->mmu_lock);
 	ret = kvm_pgtable_walk(mmu->pgt, 0, BIT(mmu->pgt->ia_bits), &walker);
-	write_unlock(&kvm->mmu_lock);
+	if (ret)
+		return ret;
+	note_page_flush(&st->parser_state.ptdump);
 
-	return ret;
+	return 0;
 }
 
 static int kvm_ptdump_guest_open(struct inode *m, struct file *file)
-- 
2.43.0

Re: [PATCH v2 2/2] KVM: arm64: ptdump: Flush the last region

[Dev Jain]

On 25/07/26 12:24 am, Wei-Lin Chang wrote:
> Currently the stage-2 ptdump calls note_page() at each leaf entry visit.
> This simply misses the output of the last region, because note_page()
> only dumps output when it detects a change in level/prot, or when the
> walk enters a next marker section. The last region in the guest IPA
> space with the same level/prot is not dumped since there is no change
> after it.
> 
> Call note_page_flush() to dump the final region. note_page_flush()
> expects ptdump_state.range[] to be initialized to the range of the
> address space, so also do that.
> 
> Also change the second marker's start address to ULONG_MAX so we never
> cross it. This avoids dumping redundant marker names (which are NULL),
> and advancing beyond the end of the marker array.
> 
> 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/20260630122758.891011F00A3A@smtp.kernel.org/
> Signed-off-by: Wei-Lin Chang <weilin.chang@arm.com>
> ---
>  arch/arm64/kvm/ptdump.c | 16 ++++++++++++----
>  1 file changed, 12 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/arm64/kvm/ptdump.c b/arch/arm64/kvm/ptdump.c
> index c9140e22abcf..222577163d2b 100644
> --- a/arch/arm64/kvm/ptdump.c
> +++ b/arch/arm64/kvm/ptdump.c
> @@ -130,7 +130,7 @@ static struct kvm_ptdump_guest_state *kvm_ptdump_parser_create(struct kvm_s2_mmu
>  	}
>  
>  	st->ipa_marker[0].name		= "Guest IPA";
> -	st->ipa_marker[1].start_address = BIT(pgtable->ia_bits);
> +	st->ipa_marker[1].start_address = ULONG_MAX;
>  
>  	st->mmu				= mmu;
>  	return st;
> @@ -148,18 +148,26 @@ static int kvm_ptdump_guest_show(struct seq_file *m, void *unused)
>  		.flags	= KVM_PGTABLE_WALK_LEAF,
>  	};
>  
> +	guard(write_lock)(&kvm->mmu_lock);
>  	st->parser_state = (struct ptdump_pg_state) {
>  		.marker		= &st->ipa_marker[0],
>  		.level		= -1,
>  		.pg_level	= &st->level[0],
>  		.seq		= m,
> +		.ptdump		= {
> +			.range = (struct ptdump_range[]){
> +				{0, BIT(mmu->pgt->ia_bits)},
> +				{0, 0}
> +			}

I am having a real hard time reading that indentation :) you
could make it look like what we have currently in ptdump_walk().

The code looks sensible to me. I have also tested by booting a
Linux VM inside a Linux VM on a Mac, and I see the stage2 ptdump
prints the last range upto the ia_bits correctly. So with the
nit addressed above:

Reviewed-by: Dev Jain <dev.jain@arm.com>
Tested-by: Dev Jain <dev.jain@arm.com>




> +		}
>  	};
>  
> -	write_lock(&kvm->mmu_lock);
>  	ret = kvm_pgtable_walk(mmu->pgt, 0, BIT(mmu->pgt->ia_bits), &walker);
> -	write_unlock(&kvm->mmu_lock);
> +	if (ret)
> +		return ret;
> +	note_page_flush(&st->parser_state.ptdump);
>  
> -	return ret;
> +	return 0;
>  }
>  
>  static int kvm_ptdump_guest_open(struct inode *m, struct file *file)

Re: [PATCH v2 2/2] KVM: arm64: ptdump: Flush the last region

[Marc Zyngier]

On Sat, 25 Jul 2026 18:16:37 +0100,
Dev Jain <dev.jain@arm.com> wrote:
> 
> 
> 
> On 25/07/26 12:24 am, Wei-Lin Chang wrote:
> > Currently the stage-2 ptdump calls note_page() at each leaf entry visit.
> > This simply misses the output of the last region, because note_page()
> > only dumps output when it detects a change in level/prot, or when the
> > walk enters a next marker section. The last region in the guest IPA
> > space with the same level/prot is not dumped since there is no change
> > after it.
> > 
> > Call note_page_flush() to dump the final region. note_page_flush()
> > expects ptdump_state.range[] to be initialized to the range of the
> > address space, so also do that.
> > 
> > Also change the second marker's start address to ULONG_MAX so we never
> > cross it. This avoids dumping redundant marker names (which are NULL),
> > and advancing beyond the end of the marker array.
> > 
> > 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/20260630122758.891011F00A3A@smtp.kernel.org/
> > Signed-off-by: Wei-Lin Chang <weilin.chang@arm.com>
> > ---
> >  arch/arm64/kvm/ptdump.c | 16 ++++++++++++----
> >  1 file changed, 12 insertions(+), 4 deletions(-)
> > 
> > diff --git a/arch/arm64/kvm/ptdump.c b/arch/arm64/kvm/ptdump.c
> > index c9140e22abcf..222577163d2b 100644
> > --- a/arch/arm64/kvm/ptdump.c
> > +++ b/arch/arm64/kvm/ptdump.c
> > @@ -130,7 +130,7 @@ static struct kvm_ptdump_guest_state *kvm_ptdump_parser_create(struct kvm_s2_mmu
> >  	}
> >  
> >  	st->ipa_marker[0].name		= "Guest IPA";
> > -	st->ipa_marker[1].start_address = BIT(pgtable->ia_bits);
> > +	st->ipa_marker[1].start_address = ULONG_MAX;
> >  
> >  	st->mmu				= mmu;
> >  	return st;
> > @@ -148,18 +148,26 @@ static int kvm_ptdump_guest_show(struct seq_file *m, void *unused)
> >  		.flags	= KVM_PGTABLE_WALK_LEAF,
> >  	};
> >  
> > +	guard(write_lock)(&kvm->mmu_lock);
> >  	st->parser_state = (struct ptdump_pg_state) {
> >  		.marker		= &st->ipa_marker[0],
> >  		.level		= -1,
> >  		.pg_level	= &st->level[0],
> >  		.seq		= m,
> > +		.ptdump		= {
> > +			.range = (struct ptdump_range[]){
> > +				{0, BIT(mmu->pgt->ia_bits)},
> > +				{0, 0}
> > +			}
> 
> I am having a real hard time reading that indentation :) you
> could make it look like what we have currently in ptdump_walk().

This patch:

		.ptdump		= {
			.range = (struct ptdump_range[]){
				{0, BIT(mmu->pgt->ia_bits)},
				{0, 0}
			}
		}

ptdump_walk():

		.ptdump = {
			[...]
			.range = (struct ptdump_range[]){
				{info->base_addr, end},
				{0, 0}
			}
		}

Zero difference in indentation. Consider using a better email client,
or look at the applied patches.

	M.

-- 
Jazz isn't dead. It just smells funny.