[PATCH 1/1] ARM: Fix mapping in alloc_init_section for unaligned addresses.

[PATCH 1/1] ARM: Fix mapping in alloc_init_section for unaligned addresses.

[R Sricharan]

When either the start address or end address or physical address
to be mapped is unaligned, alloc_init_section creates
page granularity mappings. alloc_init_section calls
alloc_init_pte which populates one pmd entry and sets up
the ptes. But if the size is greater than what can be mapped
by one pmd entry, then the rest remains unmapped.

The issue becomes visible when LPAE is enabled, where we have
the 3 levels with seperate pgd and pmd's.
When a static mapping for 3MB is requested, only 2MB is mapped
and the remaining 1MB is unmapped. Fixing this here, by looping
in to map the entire unaligned address range.

Signed-off-by: R Sricharan <r.sricharan@ti.com>
Reviewed-by: Santosh Shilimkar <santosh.shilimkar@ti.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
---
 arch/arm/mm/mmu.c |   54 ++++++++++++++++++++++++++++++-----------------------
 1 file changed, 31 insertions(+), 23 deletions(-)

diff --git a/arch/arm/mm/mmu.c b/arch/arm/mm/mmu.c
index cf4528d..c8c405f 100644
--- a/arch/arm/mm/mmu.c
+++ b/arch/arm/mm/mmu.c
@@ -597,34 +597,42 @@ static void __init alloc_init_section(pud_t *pud, unsigned long addr,
 				      const struct mem_type *type)
 {
 	pmd_t *pmd = pmd_offset(pud, addr);
-
-	/*
-	 * Try a section mapping - end, addr and phys must all be aligned
-	 * to a section boundary.  Note that PMDs refer to the individual
-	 * L1 entries, whereas PGDs refer to a group of L1 entries making
-	 * up one logical pointer to an L2 table.
-	 */
-	if (type->prot_sect && ((addr | end | phys) & ~SECTION_MASK) == 0) {
-		pmd_t *p = pmd;
+	unsigned long next;
 
 #ifndef CONFIG_ARM_LPAE
-		if (addr & SECTION_SIZE)
-			pmd++;
+	if ((addr & SECTION_SIZE) &&
+	      (type->prot_sect && ((addr | next | phys) & ~SECTION_MASK) == 0))
+		pmd++;
 #endif
-
-		do {
-			*pmd = __pmd(phys | type->prot_sect);
-			phys += SECTION_SIZE;
-		} while (pmd++, addr += SECTION_SIZE, addr != end);
-
-		flush_pmd_entry(p);
-	} else {
+	do {
+		if ((end - addr) & SECTION_MASK)
+			next = (addr + SECTION_SIZE) & SECTION_MASK;
+		else
+			next = end;
 		/*
-		 * No need to loop; pte's aren't interested in the
-		 * individual L1 entries.
+		 * Try a section mapping - end, addr and phys must all be
+		 * aligned to a section boundary.  Note that PMDs refer to
+		 * the individual L1 entries, whereas PGDs refer to a group
+		 * of L1 entries making up one logical pointer to an L2 table.
 		 */
-		alloc_init_pte(pmd, addr, end, __phys_to_pfn(phys), type);
-	}
+		if (type->prot_sect &&
+				((addr | next | phys) & ~SECTION_MASK) == 0) {
+			*pmd = __pmd(phys | type->prot_sect);
+			flush_pmd_entry(pmd);
+		} else {
+			/*
+			 * when addresses are not aligned,
+			 * we may be required to map address range greater
+			 * than a section size. So loop in here to map the
+			 * complete range.
+			 */
+			alloc_init_pte(pmd, addr, next,
+						__phys_to_pfn(phys), type);
+		}
+
+		phys += next - addr;
+
+	} while (pmd++, addr = next, addr != end);
 }
 
 static void __init alloc_init_pud(pgd_t *pgd, unsigned long addr,
-- 
1.7.9.5

[PATCH 1/1] ARM: Fix mapping in alloc_init_section for unaligned addresses.

[Russell King - ARM Linux]

On Sat, Aug 04, 2012 at 12:30:22PM +0530, R Sricharan wrote:
> When either the start address or end address or physical address
> to be mapped is unaligned, alloc_init_section creates
> page granularity mappings. alloc_init_section calls
> alloc_init_pte which populates one pmd entry and sets up
> the ptes. But if the size is greater than what can be mapped
> by one pmd entry, then the rest remains unmapped.
> 
> The issue becomes visible when LPAE is enabled, where we have
> the 3 levels with seperate pgd and pmd's.
> When a static mapping for 3MB is requested, only 2MB is mapped
> and the remaining 1MB is unmapped. Fixing this here, by looping
> in to map the entire unaligned address range.

This doesn't look like a nice fix.  The implication above is that it's
only alloc_init_pte() which is affected - so why not add a loop there?

Remember that pte's cover two sections, not one, so you need to use a
different increment to sections.

[PATCH 1/1] ARM: Fix mapping in alloc_init_section for unaligned addresses.

[R, Sricharan]

Hi Russell,

On Sat, Aug 4, 2012 at 2:14 PM, Russell King - ARM Linux
<linux@arm.linux.org.uk> wrote:
> On Sat, Aug 04, 2012 at 12:30:22PM +0530, R Sricharan wrote:
>> When either the start address or end address or physical address
>> to be mapped is unaligned, alloc_init_section creates
>> page granularity mappings. alloc_init_section calls
>> alloc_init_pte which populates one pmd entry and sets up
>> the ptes. But if the size is greater than what can be mapped
>> by one pmd entry, then the rest remains unmapped.
>>
>> The issue becomes visible when LPAE is enabled, where we have
>> the 3 levels with seperate pgd and pmd's.
>> When a static mapping for 3MB is requested, only 2MB is mapped
>> and the remaining 1MB is unmapped. Fixing this here, by looping
>> in to map the entire unaligned address range.
>
> This doesn't look like a nice fix.  The implication above is that it's
> only alloc_init_pte() which is affected - so why not add a loop there?
>
    Ok, will move the loop to alloc_init_pte instead.

> Remember that pte's cover two sections, not one, so you need to use a
> different increment to sections.

     Yes, and that varies when LPAE is enabled. Will take care of this
     as well in V2 post.

Thanks,
 Sricharan