[PATCH] arm64: Fix 5-level paging support in kexec/hibernate trampoline

[PATCH] arm64: Fix 5-level paging support in kexec/hibernate trampoline

[Ard Biesheuvel]

From: Ard Biesheuvel <ardb@kernel.org>

Add the missing code to allocate P4D level page tables when cloning the
the kernel page tables. This fixes a crash that may be observed when
attempting to resume from hibernation on an LPA2 capable system with 4k
pages, which therefore uses 5 levels of paging.

Presumably, kexec is equally affected.

Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
---
 arch/arm64/mm/trans_pgd.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/arch/arm64/mm/trans_pgd.c b/arch/arm64/mm/trans_pgd.c
index 0f7b484cb2ff..84594f02f780 100644
--- a/arch/arm64/mm/trans_pgd.c
+++ b/arch/arm64/mm/trans_pgd.c
@@ -162,6 +162,13 @@ static int copy_p4d(struct trans_pgd_info *info, pgd_t *dst_pgdp,
 	unsigned long next;
 	unsigned long addr = start;
 
+	if (pgd_none(READ_ONCE(*dst_pgdp))) {
+		dst_p4dp = trans_alloc(info);
+		if (!dst_p4dp)
+			return -ENOMEM;
+		pgd_populate(NULL, dst_pgdp, dst_p4dp);
+	}
+
 	dst_p4dp = p4d_offset(dst_pgdp, start);
 	src_p4dp = p4d_offset(src_pgdp, start);
 	do {
-- 
2.47.1.688.g23fc6f90ad-goog

Re: [PATCH] arm64: Fix 5-level paging support in kexec/hibernate trampoline

[Anshuman Khandual]

On 1/10/25 23:21, Ard Biesheuvel wrote:
> From: Ard Biesheuvel <ardb@kernel.org>
> 
> Add the missing code to allocate P4D level page tables when cloning the
> the kernel page tables. This fixes a crash that may be observed when
> attempting to resume from hibernation on an LPA2 capable system with 4k
> pages, which therefore uses 5 levels of paging.
> 
> Presumably, kexec is equally affected.
> 
> Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
> ---
>  arch/arm64/mm/trans_pgd.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/arch/arm64/mm/trans_pgd.c b/arch/arm64/mm/trans_pgd.c
> index 0f7b484cb2ff..84594f02f780 100644
> --- a/arch/arm64/mm/trans_pgd.c
> +++ b/arch/arm64/mm/trans_pgd.c
> @@ -162,6 +162,13 @@ static int copy_p4d(struct trans_pgd_info *info, pgd_t *dst_pgdp,
>  	unsigned long next;
>  	unsigned long addr = start;
>  
> +	if (pgd_none(READ_ONCE(*dst_pgdp))) {
> +		dst_p4dp = trans_alloc(info);
> +		if (!dst_p4dp)
> +			return -ENOMEM;
> +		pgd_populate(NULL, dst_pgdp, dst_p4dp);

Just wondering would not dst_p4dp goes unused (and probably leaked) when
pgtable_l5_enabled() is not enabled ? Similar scenario might also exist
in copy_pud() as well with respect to pgtable_l4_enabled().

pgd_populate()
	__pgd_populate()
		if (pgtable_l5_enabled())
			set_pgd()

> +	}
> +
>  	dst_p4dp = p4d_offset(dst_pgdp, start);
>  	src_p4dp = p4d_offset(src_pgdp, start);
>  	do {

Re: [PATCH] arm64: Fix 5-level paging support in kexec/hibernate trampoline

[Ard Biesheuvel]

On Mon, 13 Jan 2025 at 06:01, Anshuman Khandual
<anshuman.khandual@arm.com> wrote:
>
>
> On 1/10/25 23:21, Ard Biesheuvel wrote:
> > From: Ard Biesheuvel <ardb@kernel.org>
> >
> > Add the missing code to allocate P4D level page tables when cloning the
> > the kernel page tables. This fixes a crash that may be observed when
> > attempting to resume from hibernation on an LPA2 capable system with 4k
> > pages, which therefore uses 5 levels of paging.
> >
> > Presumably, kexec is equally affected.
> >
> > Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
> > ---
> >  arch/arm64/mm/trans_pgd.c | 7 +++++++
> >  1 file changed, 7 insertions(+)
> >
> > diff --git a/arch/arm64/mm/trans_pgd.c b/arch/arm64/mm/trans_pgd.c
> > index 0f7b484cb2ff..84594f02f780 100644
> > --- a/arch/arm64/mm/trans_pgd.c
> > +++ b/arch/arm64/mm/trans_pgd.c
> > @@ -162,6 +162,13 @@ static int copy_p4d(struct trans_pgd_info *info, pgd_t *dst_pgdp,
> >       unsigned long next;
> >       unsigned long addr = start;
> >
> > +     if (pgd_none(READ_ONCE(*dst_pgdp))) {
> > +             dst_p4dp = trans_alloc(info);
> > +             if (!dst_p4dp)
> > +                     return -ENOMEM;
> > +             pgd_populate(NULL, dst_pgdp, dst_p4dp);
>
> Just wondering would not dst_p4dp goes unused (and probably leaked) when
> pgtable_l5_enabled() is not enabled ?

Would pgd_none() ever return TRUE in that case?

> Similar scenario might also exist
> in copy_pud() as well with respect to pgtable_l4_enabled().
>
> pgd_populate()
>         __pgd_populate()
>                 if (pgtable_l5_enabled())
>                         set_pgd()
>
> > +     }
> > +
> >       dst_p4dp = p4d_offset(dst_pgdp, start);
> >       src_p4dp = p4d_offset(src_pgdp, start);
> >       do {

Re: [PATCH] arm64: Fix 5-level paging support in kexec/hibernate trampoline

[Will Deacon]

On Fri, 10 Jan 2025 18:51:46 +0100, Ard Biesheuvel wrote:
> Add the missing code to allocate P4D level page tables when cloning the
> the kernel page tables. This fixes a crash that may be observed when
> attempting to resume from hibernation on an LPA2 capable system with 4k
> pages, which therefore uses 5 levels of paging.
> 
> Presumably, kexec is equally affected.
> 
> [...]

Applied to arm64 (for-next/fixes), thanks!

[1/1] arm64: Fix 5-level paging support in kexec/hibernate trampoline
      https://git.kernel.org/arm64/c/3648027de1fa

Cheers,
-- 
Will

https://fixes.arm64.dev
https://next.arm64.dev
https://will.arm64.dev