[PATCH v2] powerpc/mm: Check for _PAGE_PTE in *_devmap()

[PATCH v2] powerpc/mm: Check for _PAGE_PTE in *_devmap()

[Oliver O'Halloran]

The ISA radix translation tree contains two different types of entry,
directories and leaves. The formats of the two entries are different
with the directory entries containing no spare bits for use by software.
As a result we need to ensure that the *_devmap() family of functions
check fail for everything except leaf (PTE) entries.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
---
"i'll just tweak the mbox before i sent it, what's the worst that can happen"
*completely breaks KVM*
"..."
---
 arch/powerpc/include/asm/book3s/64/pgtable.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h b/arch/powerpc/include/asm/book3s/64/pgtable.h
index d1da415..6bc6248 100644
--- a/arch/powerpc/include/asm/book3s/64/pgtable.h
+++ b/arch/powerpc/include/asm/book3s/64/pgtable.h
@@ -610,7 +610,9 @@ static inline pte_t pte_mkdevmap(pte_t pte)
 
 static inline int pte_devmap(pte_t pte)
 {
-	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_DEVMAP));
+	uint64_t mask = cpu_to_be64(_PAGE_DEVMAP | _PAGE_PTE);
+
+	return (pte_raw(pte) & mask) == mask;
 }
 
 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
-- 
2.7.4

Re: [PATCH v2] powerpc/mm: Check for _PAGE_PTE in *_devmap()

[Laurent Vivier]

On 27/07/2017 17:35, oohall@gmail.com (Oliver O'Halloran) wrote:
> The ISA radix translation tree contains two different types of entry,
> directories and leaves. The formats of the two entries are different
> with the directory entries containing no spare bits for use by software.
> As a result we need to ensure that the *_devmap() family of functions
> check fail for everything except leaf (PTE) entries.
> 
> Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
> ---
> "i'll just tweak the mbox before i sent it, what's the worst that can happen"
> *completely breaks KVM*
> "..."
> ---
>  arch/powerpc/include/asm/book3s/64/pgtable.h | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h b/arch/powerpc/include/asm/book3s/64/pgtable.h
> index d1da415..6bc6248 100644
> --- a/arch/powerpc/include/asm/book3s/64/pgtable.h
> +++ b/arch/powerpc/include/asm/book3s/64/pgtable.h
> @@ -610,7 +610,9 @@ static inline pte_t pte_mkdevmap(pte_t pte)
>  
>  static inline int pte_devmap(pte_t pte)
>  {
> -	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_DEVMAP));
> +	uint64_t mask = cpu_to_be64(_PAGE_DEVMAP | _PAGE_PTE);
> +
> +	return (pte_raw(pte) & mask) == mask;
>  }
>  
>  static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
> 

Tested-by: Laurent Vivier <lvivier@redhat.com>

Re: [PATCH v2] powerpc/mm: Check for _PAGE_PTE in *_devmap()

[joserz]

On Fri, Jul 28, 2017 at 01:35:53AM +1000, Oliver O'Halloran wrote:
> The ISA radix translation tree contains two different types of entry,
> directories and leaves. The formats of the two entries are different
> with the directory entries containing no spare bits for use by software.
> As a result we need to ensure that the *_devmap() family of functions
> check fail for everything except leaf (PTE) entries.
> 
> Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
> ---
> "i'll just tweak the mbox before i sent it, what's the worst that can happen"
> *completely breaks KVM*
> "..."
> ---
>  arch/powerpc/include/asm/book3s/64/pgtable.h | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h b/arch/powerpc/include/asm/book3s/64/pgtable.h
> index d1da415..6bc6248 100644
> --- a/arch/powerpc/include/asm/book3s/64/pgtable.h
> +++ b/arch/powerpc/include/asm/book3s/64/pgtable.h
> @@ -610,7 +610,9 @@ static inline pte_t pte_mkdevmap(pte_t pte)
> 
>  static inline int pte_devmap(pte_t pte)
>  {
> -	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_DEVMAP));
> +	uint64_t mask = cpu_to_be64(_PAGE_DEVMAP | _PAGE_PTE);
> +
> +	return (pte_raw(pte) & mask) == mask;
>  }
> 
>  static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
> -- 
> 2.7.4
> 

Michael should thank you because I was about to send another insane
suggestion and ask him if that makes sense. :-D

Tested-by: Jose Ricardo Ziviani <joserz@linux.vnet.ibm.com>

Re: [PATCH v2] powerpc/mm: Check for _PAGE_PTE in *_devmap()

[Suraj Jitindar Singh]

On Fri, 2017-07-28 at 01:35 +1000, Oliver O'Halloran wrote:
> The ISA radix translation tree contains two different types of entry,
s/entry/entries
> > directories and leaves. The formats of the two entries are
> different
> with the directory entries containing no spare bits for use by
> software.

Rather than saying the directory entries contain no spare bits, would
it be better to say something like: the devmap property only relates to
pte (leaf) entries and so we shouldn't perform the check on/should
always return false for page directory entries?

> As a result we need to ensure that the *_devmap() family of functions
> check fail for everything except leaf (PTE) entries.
> 
> Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
> ---
> "i'll just tweak the mbox before i sent it, what's the worst that can
> happen"
> *completely breaks KVM*
> "..."
> ---
>  arch/powerpc/include/asm/book3s/64/pgtable.h | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h
> b/arch/powerpc/include/asm/book3s/64/pgtable.h
> index d1da415..6bc6248 100644
> --- a/arch/powerpc/include/asm/book3s/64/pgtable.h
> +++ b/arch/powerpc/include/asm/book3s/64/pgtable.h
> @@ -610,7 +610,9 @@ static inline pte_t pte_mkdevmap(pte_t pte)
>  
>  static inline int pte_devmap(pte_t pte)
>  {
> -	return !!(pte_raw(pte) & cpu_to_be64(_PAGE_DEVMAP));
> +	uint64_t mask = cpu_to_be64(_PAGE_DEVMAP | _PAGE_PTE);
> +
> +	return (pte_raw(pte) & mask) == mask;
>  }
>  
>  static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)

Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>

Re: [v2] powerpc/mm: Check for _PAGE_PTE in *_devmap()

[Michael Ellerman]

On Thu, 2017-07-27 at 15:35:53 UTC, Oliver O'Halloran wrote:
> The ISA radix translation tree contains two different types of entry,
> directories and leaves. The formats of the two entries are different
> with the directory entries containing no spare bits for use by software.
> As a result we need to ensure that the *_devmap() family of functions
> check fail for everything except leaf (PTE) entries.
> 
> Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
> Tested-by: Laurent Vivier <lvivier@redhat.com>
> Tested-by: Jose Ricardo Ziviani <joserz@linux.vnet.ibm.com>
> Reviewed-by: Suraj Jitindar Singh <sjitindarsingh@gmail.com>

Applied to powerpc fixes, thanks.

https://git.kernel.org/powerpc/c/c9c98bc5cc2ab4e0d5c9ad58286fa7

cheers