[PATCH mm-unstable RFC 17/26] powerpc/mm: support __HAVE_ARCH_PTE_SWP_EXCLUSIVE on 32bit book3s

We already implemented support for 64bit book3s in commit bff9beaa2e80
("powerpc/pgtable: support __HAVE_ARCH_PTE_SWP_EXCLUSIVE for book3s")

Let's support __HAVE_ARCH_PTE_SWP_EXCLUSIVE also in 32bit by reusing yet
unused LSB 2 / MSB 29. There seems to be no real reason why that bit cannot
be used, and reusing it avoids having to steal one bit from the swap
offset.

While at it, mask the type in __swp_entry().

Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Christophe Leroy <redacted>
Signed-off-by: David Hildenbrand <redacted>
---
 arch/powerpc/include/asm/book3s/32/pgtable.h | 38 +++++++++++++++++---
 1 file changed, 33 insertions(+), 5 deletions(-)

diff --git a/arch/powerpc/include/asm/book3s/32/pgtable.h b/arch/powerpc/include/asm/book3s/32/pgtable.h
index 75823f39e042..8107835b38c1 100644
--- a/arch/powerpc/include/asm/book3s/32/pgtable.h
+++ b/arch/powerpc/include/asm/book3s/32/pgtable.h

@@ -42,6 +42,9 @@
 #define _PMD_PRESENT_MASK (PAGE_MASK)
 #define _PMD_BAD	(~PAGE_MASK)
 
+/* We borrow the _PAGE_USER bit to store the exclusive marker in swap PTEs. */
+#define _PAGE_SWP_EXCLUSIVE	_PAGE_USER
+
 /* And here we include common definitions */
 
 #define _PAGE_KERNEL_RO		0

@@ -363,17 +366,42 @@ static inline void __ptep_set_access_flags(struct vm_area_struct *vma,
 #define pmd_page(pmd)		pfn_to_page(pmd_pfn(pmd))
 
 /*
- * Encode and decode a swap entry.
- * Note that the bits we use in a PTE for representing a swap entry
- * must not include the _PAGE_PRESENT bit or the _PAGE_HASHPTE bit (if used).
- *   -- paulus
+ * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
+ * are !pte_none() && !pte_present().
+ *
+ * Format of swap PTEs (32bit PTEs):
+ *
+ *                         1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
+ *   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+ *   E H P <- type --> <----------------- offset ------------------>
+ *
+ *   E is the exclusive marker that is not stored in swap entries.
+ *   _PAGE_PRESENT (P) and __PAGE_HASHPTE (H) must be 0.
+ *
+ * For 64bit PTEs, the offset is extended by 32bit.
  */
 #define __swp_type(entry)		((entry).val & 0x1f)
 #define __swp_offset(entry)		((entry).val >> 5)
-#define __swp_entry(type, offset)	((swp_entry_t) { (type) | ((offset) << 5) })
+#define __swp_entry(type, offset)	((swp_entry_t) { ((type) & 0x1f) | ((offset) << 5) })
 #define __pte_to_swp_entry(pte)		((swp_entry_t) { pte_val(pte) >> 3 })
 #define __swp_entry_to_pte(x)		((pte_t) { (x).val << 3 })
 
+#define __HAVE_ARCH_PTE_SWP_EXCLUSIVE
+static inline int pte_swp_exclusive(pte_t pte)
+{
+	return pte_val(pte) & _PAGE_SWP_EXCLUSIVE;
+}
+
+static inline pte_t pte_swp_mkexclusive(pte_t pte)
+{
+	return __pte(pte_val(pte) | _PAGE_SWP_EXCLUSIVE);
+}
+
+static inline pte_t pte_swp_clear_exclusive(pte_t pte)
+{
+	return __pte(pte_val(pte) & ~_PAGE_SWP_EXCLUSIVE);
+}
+
 /* Generic accessors to PTE bits */
 static inline int pte_write(pte_t pte)		{ return !!(pte_val(pte) & _PAGE_RW);}
 static inline int pte_read(pte_t pte)		{ return 1; }

-- 
2.38.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Le 06/12/2022 à 15:47, David Hildenbrand a écrit :

That's in reversed order. _PAGE_HASHPTE is bit 30 and should be on the 
right hand side. Etc ...

Some exemple in arch/powerpc/include/asm/nohash/32/pte-40x.h or 
arch/powerpc/include/asm/nohash/32/pte-85xx.h

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

On 07.12.22 14:55, Christophe Leroy wrote:

Le 06/12/2022 à 15:47, David Hildenbrand a écrit :

quoted

We already implemented support for 64bit book3s in commit bff9beaa2e80
("powerpc/pgtable: support __HAVE_ARCH_PTE_SWP_EXCLUSIVE for book3s")

Let's support __HAVE_ARCH_PTE_SWP_EXCLUSIVE also in 32bit by reusing yet
unused LSB 2 / MSB 29. There seems to be no real reason why that bit cannot
be used, and reusing it avoids having to steal one bit from the swap
offset.

While at it, mask the type in __swp_entry().

Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Christophe Leroy <redacted>
Signed-off-by: David Hildenbrand <redacted>
---
   arch/powerpc/include/asm/book3s/32/pgtable.h | 38 +++++++++++++++++---
   1 file changed, 33 insertions(+), 5 deletions(-)

diff --git a/arch/powerpc/include/asm/book3s/32/pgtable.h b/arch/powerpc/include/asm/book3s/32/pgtable.h
index 75823f39e042..8107835b38c1 100644
--- a/arch/powerpc/include/asm/book3s/32/pgtable.h
+++ b/arch/powerpc/include/asm/book3s/32/pgtable.h

@@ -42,6 +42,9 @@
   #define _PMD_PRESENT_MASK (PAGE_MASK)
   #define _PMD_BAD	(~PAGE_MASK)
   
+/* We borrow the _PAGE_USER bit to store the exclusive marker in swap PTEs. */
+#define _PAGE_SWP_EXCLUSIVE	_PAGE_USER
+
   /* And here we include common definitions */
   
   #define _PAGE_KERNEL_RO		0

@@ -363,17 +366,42 @@ static inline void __ptep_set_access_flags(struct vm_area_struct *vma,
   #define pmd_page(pmd)		pfn_to_page(pmd_pfn(pmd))
   
   /*
- * Encode and decode a swap entry.
- * Note that the bits we use in a PTE for representing a swap entry
- * must not include the _PAGE_PRESENT bit or the _PAGE_HASHPTE bit (if used).
- *   -- paulus
+ * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
+ * are !pte_none() && !pte_present().
+ *
+ * Format of swap PTEs (32bit PTEs):
+ *
+ *                         1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
+ *   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+ *   E H P <- type --> <----------------- offset ------------------>

That's in reversed order. _PAGE_HASHPTE is bit 30 and should be on the
right hand side. Etc ...

Ugh, messed it up while converting back and forth between LSB 0 and MSB 0.

/*
 * Format of swap PTEs (32bit PTEs):
 *
 *                         1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
 *   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
 *   <----------------- offset ------------------> <- type --> E H P


Now the patch description ("unused LSB 2 / MSB 29") makes sense.

Thanks!

Any feedback if the bit could be problematic?

-- 
Thanks,

David / dhildenb


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

On 08.12.22 09:52, David Hildenbrand wrote:

On 07.12.22 14:55, Christophe Leroy wrote:

quoted

Le 06/12/2022 à 15:47, David Hildenbrand a écrit :

quoted

We already implemented support for 64bit book3s in commit bff9beaa2e80
("powerpc/pgtable: support __HAVE_ARCH_PTE_SWP_EXCLUSIVE for book3s")

Let's support __HAVE_ARCH_PTE_SWP_EXCLUSIVE also in 32bit by reusing yet
unused LSB 2 / MSB 29. There seems to be no real reason why that bit cannot
be used, and reusing it avoids having to steal one bit from the swap
offset.

While at it, mask the type in __swp_entry().

Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Christophe Leroy <redacted>
Signed-off-by: David Hildenbrand <redacted>
---
    arch/powerpc/include/asm/book3s/32/pgtable.h | 38 +++++++++++++++++---
    1 file changed, 33 insertions(+), 5 deletions(-)

diff --git a/arch/powerpc/include/asm/book3s/32/pgtable.h b/arch/powerpc/include/asm/book3s/32/pgtable.h
index 75823f39e042..8107835b38c1 100644
--- a/arch/powerpc/include/asm/book3s/32/pgtable.h
+++ b/arch/powerpc/include/asm/book3s/32/pgtable.h

@@ -42,6 +42,9 @@
    #define _PMD_PRESENT_MASK (PAGE_MASK)
    #define _PMD_BAD	(~PAGE_MASK)
    
+/* We borrow the _PAGE_USER bit to store the exclusive marker in swap PTEs. */
+#define _PAGE_SWP_EXCLUSIVE	_PAGE_USER
+
    /* And here we include common definitions */
    
    #define _PAGE_KERNEL_RO		0

@@ -363,17 +366,42 @@ static inline void __ptep_set_access_flags(struct vm_area_struct *vma,
    #define pmd_page(pmd)		pfn_to_page(pmd_pfn(pmd))
    
    /*
- * Encode and decode a swap entry.
- * Note that the bits we use in a PTE for representing a swap entry
- * must not include the _PAGE_PRESENT bit or the _PAGE_HASHPTE bit (if used).
- *   -- paulus
+ * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
+ * are !pte_none() && !pte_present().
+ *
+ * Format of swap PTEs (32bit PTEs):
+ *
+ *                         1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
+ *   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+ *   E H P <- type --> <----------------- offset ------------------>

That's in reversed order. _PAGE_HASHPTE is bit 30 and should be on the
right hand side. Etc ...

Ugh, messed it up while converting back and forth between LSB 0 and MSB 0.

/*
   * Format of swap PTEs (32bit PTEs):
   *
   *                         1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
   *   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
   *   <----------------- offset ------------------> <- type --> E H P

Still wrong, the type is only 5 bits:

+ * Format of swap PTEs (32bit PTEs):
+ *
+ *                         1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
+ *   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+ *   <----------------- offset --------------------> < type -> E H P
+ *


-- 
Thanks,

David / dhildenb


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel