[PATCH] arm64/mm: Create level specific section mappings in map_range()

[PATCH] arm64/mm: Create level specific section mappings in map_range()

[Anshuman Khandual]

Currently PMD section mapping mask i.e PMD_TYPE_SECT is used while creating
section mapping at all page table levels except the last level. This works
fine as the section mapping masks are exactly the same (0x1UL) for all page
table levels.

This will change in the future with D128 page tables that have unique skip
level values (SKL) required for creating section mapping at different page
table levels. Hence use page table level specific section mapping macros
instead of the common PMD_TYPE_SECT.

Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Ard Biesheuvel <ardb@kernel.org>
Cc: linux-kernel@vger.kernel.org
Cc: linux-arm-kernel@lists.infradead.org
Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
---
This patch applies on 6.14-rc5

PGD_TYPE_SECT for level -1 section map handling has been added for 4K
base pages with 52 bit VA configuration that has 5 page table levels.
In such cases (CONFIG_PGTABLE_LEVELS = 5) early_map_kernel() can call
map_range() eventually with -1 (i.e 4 - CONFIG_PGTABLE_LEVELS) as the
root_level.

 arch/arm64/include/asm/pgtable-hwdef.h |  1 +
 arch/arm64/kernel/pi/map_range.c       | 23 +++++++++++++++++++++--
 2 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/arch/arm64/include/asm/pgtable-hwdef.h b/arch/arm64/include/asm/pgtable-hwdef.h
index a9136cc551cc..fd0a82e8878c 100644
--- a/arch/arm64/include/asm/pgtable-hwdef.h
+++ b/arch/arm64/include/asm/pgtable-hwdef.h
@@ -99,6 +99,7 @@
 #define PGD_TYPE_TABLE		(_AT(pgdval_t, 3) << 0)
 #define PGD_TABLE_BIT		(_AT(pgdval_t, 1) << 1)
 #define PGD_TYPE_MASK		(_AT(pgdval_t, 3) << 0)
+#define PGD_TYPE_SECT		(_AT(pgdval_t, 1) << 0)
 #define PGD_TABLE_AF		(_AT(pgdval_t, 1) << 10)	/* Ignored if no FEAT_HAFT */
 #define PGD_TABLE_PXN		(_AT(pgdval_t, 1) << 59)
 #define PGD_TABLE_UXN		(_AT(pgdval_t, 1) << 60)
diff --git a/arch/arm64/kernel/pi/map_range.c b/arch/arm64/kernel/pi/map_range.c
index 2b69e3beeef8..9ea869f5745f 100644
--- a/arch/arm64/kernel/pi/map_range.c
+++ b/arch/arm64/kernel/pi/map_range.c
@@ -44,8 +44,27 @@ void __init map_range(u64 *pte, u64 start, u64 end, u64 pa, pgprot_t prot,
 	 * Set the right block/page bits for this level unless we are
 	 * clearing the mapping
 	 */
-	if (protval)
-		protval |= (level < 3) ? PMD_TYPE_SECT : PTE_TYPE_PAGE;
+	if (protval) {
+		switch (level) {
+		case 3:
+			protval |= PTE_TYPE_PAGE;
+			break;
+		case 2:
+			protval |= PMD_TYPE_SECT;
+			break;
+		case 1:
+			protval |= PUD_TYPE_SECT;
+			break;
+		case 0:
+			protval |= P4D_TYPE_SECT;
+			break;
+		case -1:
+			protval |= PGD_TYPE_SECT;
+			break;
+		default:
+			break;
+		}
+	}
 
 	while (start < end) {
 		u64 next = min((start | lmask) + 1, PAGE_ALIGN(end));
-- 
2.30.2

Re: [PATCH] arm64/mm: Create level specific section mappings in map_range()

[Ryan Roberts]

On 03/03/2025 04:18, Anshuman Khandual wrote:
> Currently PMD section mapping mask i.e PMD_TYPE_SECT is used while creating
> section mapping at all page table levels except the last level. This works
> fine as the section mapping masks are exactly the same (0x1UL) for all page
> table levels.
> 
> This will change in the future with D128 page tables that have unique skip
> level values (SKL) required for creating section mapping at different page
> table levels. Hence use page table level specific section mapping macros
> instead of the common PMD_TYPE_SECT.
> 
> Cc: Catalin Marinas <catalin.marinas@arm.com>
> Cc: Will Deacon <will@kernel.org>
> Cc: Ryan Roberts <ryan.roberts@arm.com>
> Cc: Ard Biesheuvel <ardb@kernel.org>
> Cc: linux-kernel@vger.kernel.org
> Cc: linux-arm-kernel@lists.infradead.org
> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
> ---
> This patch applies on 6.14-rc5
> 
> PGD_TYPE_SECT for level -1 section map handling has been added for 4K
> base pages with 52 bit VA configuration that has 5 page table levels.
> In such cases (CONFIG_PGTABLE_LEVELS = 5) early_map_kernel() can call
> map_range() eventually with -1 (i.e 4 - CONFIG_PGTABLE_LEVELS) as the
> root_level.

Table Table D8-16 on page D8-6459 of ARM DDI 0487 L.a says that block mappings
at level -1 are not permitted for 4K pages; only levels 0-3 support leaf
mappings. Similarly for 16K, table D8-26 says only levels 1-3 permit leaf
mappings. And for 64K, table D8-35 says only levels 1-3 permit leaf mappings.

So I don't think PGD_TYPE_SECT is the right solution. Perhaps we need to
explicitly force the unsupported levels to be table entries even if the
alignment is correct?

Thanks,
Ryan

> 
>  arch/arm64/include/asm/pgtable-hwdef.h |  1 +
>  arch/arm64/kernel/pi/map_range.c       | 23 +++++++++++++++++++++--
>  2 files changed, 22 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/include/asm/pgtable-hwdef.h b/arch/arm64/include/asm/pgtable-hwdef.h
> index a9136cc551cc..fd0a82e8878c 100644
> --- a/arch/arm64/include/asm/pgtable-hwdef.h
> +++ b/arch/arm64/include/asm/pgtable-hwdef.h
> @@ -99,6 +99,7 @@
>  #define PGD_TYPE_TABLE		(_AT(pgdval_t, 3) << 0)
>  #define PGD_TABLE_BIT		(_AT(pgdval_t, 1) << 1)
>  #define PGD_TYPE_MASK		(_AT(pgdval_t, 3) << 0)
> +#define PGD_TYPE_SECT		(_AT(pgdval_t, 1) << 0)
>  #define PGD_TABLE_AF		(_AT(pgdval_t, 1) << 10)	/* Ignored if no FEAT_HAFT */
>  #define PGD_TABLE_PXN		(_AT(pgdval_t, 1) << 59)
>  #define PGD_TABLE_UXN		(_AT(pgdval_t, 1) << 60)
> diff --git a/arch/arm64/kernel/pi/map_range.c b/arch/arm64/kernel/pi/map_range.c
> index 2b69e3beeef8..9ea869f5745f 100644
> --- a/arch/arm64/kernel/pi/map_range.c
> +++ b/arch/arm64/kernel/pi/map_range.c
> @@ -44,8 +44,27 @@ void __init map_range(u64 *pte, u64 start, u64 end, u64 pa, pgprot_t prot,
>  	 * Set the right block/page bits for this level unless we are
>  	 * clearing the mapping
>  	 */
> -	if (protval)
> -		protval |= (level < 3) ? PMD_TYPE_SECT : PTE_TYPE_PAGE;
> +	if (protval) {
> +		switch (level) {
> +		case 3:
> +			protval |= PTE_TYPE_PAGE;
> +			break;
> +		case 2:
> +			protval |= PMD_TYPE_SECT;
> +			break;
> +		case 1:
> +			protval |= PUD_TYPE_SECT;
> +			break;
> +		case 0:
> +			protval |= P4D_TYPE_SECT;
> +			break;
> +		case -1:
> +			protval |= PGD_TYPE_SECT;
> +			break;
> +		default:
> +			break;
> +		}
> +	}
>  
>  	while (start < end) {
>  		u64 next = min((start | lmask) + 1, PAGE_ALIGN(end));

Re: [PATCH] arm64/mm: Create level specific section mappings in map_range()

[Anshuman Khandual]

On 3/4/25 21:51, Ryan Roberts wrote:
> On 03/03/2025 04:18, Anshuman Khandual wrote:
>> Currently PMD section mapping mask i.e PMD_TYPE_SECT is used while creating
>> section mapping at all page table levels except the last level. This works
>> fine as the section mapping masks are exactly the same (0x1UL) for all page
>> table levels.
>>
>> This will change in the future with D128 page tables that have unique skip
>> level values (SKL) required for creating section mapping at different page
>> table levels. Hence use page table level specific section mapping macros
>> instead of the common PMD_TYPE_SECT.
>>
>> Cc: Catalin Marinas <catalin.marinas@arm.com>
>> Cc: Will Deacon <will@kernel.org>
>> Cc: Ryan Roberts <ryan.roberts@arm.com>
>> Cc: Ard Biesheuvel <ardb@kernel.org>
>> Cc: linux-kernel@vger.kernel.org
>> Cc: linux-arm-kernel@lists.infradead.org
>> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
>> ---
>> This patch applies on 6.14-rc5
>>
>> PGD_TYPE_SECT for level -1 section map handling has been added for 4K
>> base pages with 52 bit VA configuration that has 5 page table levels.
>> In such cases (CONFIG_PGTABLE_LEVELS = 5) early_map_kernel() can call
>> map_range() eventually with -1 (i.e 4 - CONFIG_PGTABLE_LEVELS) as the
>> root_level.
> 
> Table Table D8-16 on page D8-6459 of ARM DDI 0487 L.a says that block mappings
> at level -1 are not permitted for 4K pages; only levels 0-3 support leaf
> mappings. Similarly for 16K, table D8-26 says only levels 1-3 permit leaf
> mappings. And for 64K, table D8-35 says only levels 1-3 permit leaf mappings.

Then seems like the current code is actually wrong because PMD_TYPE_SECT
is being set at all levels (except level 3) regardless of the configured
page size ?

> 
> So I don't think PGD_TYPE_SECT is the right solution. Perhaps we need to
> explicitly force the unsupported levels to be table entries even if the
> alignment is correct?

Just wondering - something like the following will work instead ? Tested
on both 4K and 64K page sizes.

--- a/arch/arm64/kernel/pi/map_range.c
+++ b/arch/arm64/kernel/pi/map_range.c
@@ -26,6 +26,21 @@
  * @va_offset:         Offset between a physical page and its current mapping
  *                     in the VA space
  */
+static bool sect_supported(int level)
+{
+       switch(level) {
+       case -1:
+               return false;
+       case 0:
+               if(IS_ENABLED(CONFIG_ARM64_16K_PAGES) ||
+                  IS_ENABLED(CONFIG_ARM64_64K_PAGES))
+                       return false;
+               fallthrough;
+       default:
+               return true;
+       }
+}
+
 void __init map_range(u64 *pte, u64 start, u64 end, u64 pa, pgprot_t prot,
                      int level, pte_t *tbl, bool may_use_cont, u64 va_offset)
 {
@@ -44,13 +59,30 @@ void __init map_range(u64 *pte, u64 start, u64 end, u64 pa, pgprot_t prot,
         * Set the right block/page bits for this level unless we are
         * clearing the mapping
         */
-       if (protval)
-               protval |= (level < 3) ? PMD_TYPE_SECT : PTE_TYPE_PAGE;
+       if (protval && sect_supported(level)) {
+               switch (level) {
+               case 3:
+                       protval |= PTE_TYPE_PAGE;
+                       break;
+               case 2:
+                       protval |= PMD_TYPE_SECT;
+                       break;
+               case 1:
+                       protval |= PUD_TYPE_SECT;
+                       break;
+               case 0:
+                       protval |= P4D_TYPE_SECT;
+                       break;
+               default:
+                       break;
+               }
+       }
 
        while (start < end) {
                u64 next = min((start | lmask) + 1, PAGE_ALIGN(end));
 
-               if (level < 3 && (start | next | pa) & lmask) {
+               if ((level < 3 && (start | next | pa) & lmask) ||
+                    !sect_supported(level)){
                        /*
                         * This chunk needs a finer grained mapping. Create a
                         * table mapping if necessary and recurse.

> 
> Thanks,
> Ryan
> 
>>
>>  arch/arm64/include/asm/pgtable-hwdef.h |  1 +
>>  arch/arm64/kernel/pi/map_range.c       | 23 +++++++++++++++++++++--
>>  2 files changed, 22 insertions(+), 2 deletions(-)
>>
>> diff --git a/arch/arm64/include/asm/pgtable-hwdef.h b/arch/arm64/include/asm/pgtable-hwdef.h
>> index a9136cc551cc..fd0a82e8878c 100644
>> --- a/arch/arm64/include/asm/pgtable-hwdef.h
>> +++ b/arch/arm64/include/asm/pgtable-hwdef.h
>> @@ -99,6 +99,7 @@
>>  #define PGD_TYPE_TABLE		(_AT(pgdval_t, 3) << 0)
>>  #define PGD_TABLE_BIT		(_AT(pgdval_t, 1) << 1)
>>  #define PGD_TYPE_MASK		(_AT(pgdval_t, 3) << 0)
>> +#define PGD_TYPE_SECT		(_AT(pgdval_t, 1) << 0)
>>  #define PGD_TABLE_AF		(_AT(pgdval_t, 1) << 10)	/* Ignored if no FEAT_HAFT */
>>  #define PGD_TABLE_PXN		(_AT(pgdval_t, 1) << 59)
>>  #define PGD_TABLE_UXN		(_AT(pgdval_t, 1) << 60)
>> diff --git a/arch/arm64/kernel/pi/map_range.c b/arch/arm64/kernel/pi/map_range.c
>> index 2b69e3beeef8..9ea869f5745f 100644
>> --- a/arch/arm64/kernel/pi/map_range.c
>> +++ b/arch/arm64/kernel/pi/map_range.c
>> @@ -44,8 +44,27 @@ void __init map_range(u64 *pte, u64 start, u64 end, u64 pa, pgprot_t prot,
>>  	 * Set the right block/page bits for this level unless we are
>>  	 * clearing the mapping
>>  	 */
>> -	if (protval)
>> -		protval |= (level < 3) ? PMD_TYPE_SECT : PTE_TYPE_PAGE;
>> +	if (protval) {
>> +		switch (level) {
>> +		case 3:
>> +			protval |= PTE_TYPE_PAGE;
>> +			break;
>> +		case 2:
>> +			protval |= PMD_TYPE_SECT;
>> +			break;
>> +		case 1:
>> +			protval |= PUD_TYPE_SECT;
>> +			break;
>> +		case 0:
>> +			protval |= P4D_TYPE_SECT;
>> +			break;
>> +		case -1:
>> +			protval |= PGD_TYPE_SECT;
>> +			break;
>> +		default:
>> +			break;
>> +		}
>> +	}
>>  
>>  	while (start < end) {
>>  		u64 next = min((start | lmask) + 1, PAGE_ALIGN(end));
> 

Re: [PATCH] arm64/mm: Create level specific section mappings in map_range()

[Ryan Roberts]

On 06/03/2025 03:37, Anshuman Khandual wrote:
> 
> 
> On 3/4/25 21:51, Ryan Roberts wrote:
>> On 03/03/2025 04:18, Anshuman Khandual wrote:
>>> Currently PMD section mapping mask i.e PMD_TYPE_SECT is used while creating
>>> section mapping at all page table levels except the last level. This works
>>> fine as the section mapping masks are exactly the same (0x1UL) for all page
>>> table levels.
>>>
>>> This will change in the future with D128 page tables that have unique skip
>>> level values (SKL) required for creating section mapping at different page
>>> table levels. Hence use page table level specific section mapping macros
>>> instead of the common PMD_TYPE_SECT.
>>>
>>> Cc: Catalin Marinas <catalin.marinas@arm.com>
>>> Cc: Will Deacon <will@kernel.org>
>>> Cc: Ryan Roberts <ryan.roberts@arm.com>
>>> Cc: Ard Biesheuvel <ardb@kernel.org>
>>> Cc: linux-kernel@vger.kernel.org
>>> Cc: linux-arm-kernel@lists.infradead.org
>>> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
>>> ---
>>> This patch applies on 6.14-rc5
>>>
>>> PGD_TYPE_SECT for level -1 section map handling has been added for 4K
>>> base pages with 52 bit VA configuration that has 5 page table levels.
>>> In such cases (CONFIG_PGTABLE_LEVELS = 5) early_map_kernel() can call
>>> map_range() eventually with -1 (i.e 4 - CONFIG_PGTABLE_LEVELS) as the
>>> root_level.
>>
>> Table Table D8-16 on page D8-6459 of ARM DDI 0487 L.a says that block mappings
>> at level -1 are not permitted for 4K pages; only levels 0-3 support leaf
>> mappings. Similarly for 16K, table D8-26 says only levels 1-3 permit leaf
>> mappings. And for 64K, table D8-35 says only levels 1-3 permit leaf mappings.
> 
> Then seems like the current code is actually wrong because PMD_TYPE_SECT
> is being set at all levels (except level 3) regardless of the configured
> page size ?

Yes, I think so. In that case, you should mark it with Fixes: and Cc: stable.

But I feel like there might be some subtlely here that means that the problem
can't happen in practice. Ard might know?

> 
>>
>> So I don't think PGD_TYPE_SECT is the right solution. Perhaps we need to
>> explicitly force the unsupported levels to be table entries even if the
>> alignment is correct?
> 
> Just wondering - something like the following will work instead ? Tested
> on both 4K and 64K page sizes.

LGTM!

> 
> --- a/arch/arm64/kernel/pi/map_range.c
> +++ b/arch/arm64/kernel/pi/map_range.c
> @@ -26,6 +26,21 @@
>   * @va_offset:         Offset between a physical page and its current mapping
>   *                     in the VA space
>   */
> +static bool sect_supported(int level)
> +{
> +       switch(level) {
> +       case -1:
> +               return false;
> +       case 0:
> +               if(IS_ENABLED(CONFIG_ARM64_16K_PAGES) ||
> +                  IS_ENABLED(CONFIG_ARM64_64K_PAGES))
> +                       return false;
> +               fallthrough;
> +       default:
> +               return true;
> +       }
> +}
> +
>  void __init map_range(u64 *pte, u64 start, u64 end, u64 pa, pgprot_t prot,
>                       int level, pte_t *tbl, bool may_use_cont, u64 va_offset)
>  {
> @@ -44,13 +59,30 @@ void __init map_range(u64 *pte, u64 start, u64 end, u64 pa, pgprot_t prot,
>          * Set the right block/page bits for this level unless we are
>          * clearing the mapping
>          */
> -       if (protval)
> -               protval |= (level < 3) ? PMD_TYPE_SECT : PTE_TYPE_PAGE;
> +       if (protval && sect_supported(level)) {
> +               switch (level) {
> +               case 3:
> +                       protval |= PTE_TYPE_PAGE;
> +                       break;
> +               case 2:
> +                       protval |= PMD_TYPE_SECT;
> +                       break;
> +               case 1:
> +                       protval |= PUD_TYPE_SECT;
> +                       break;
> +               case 0:
> +                       protval |= P4D_TYPE_SECT;
> +                       break;
> +               default:
> +                       break;
> +               }
> +       }
>  
>         while (start < end) {
>                 u64 next = min((start | lmask) + 1, PAGE_ALIGN(end));
>  
> -               if (level < 3 && (start | next | pa) & lmask) {
> +               if ((level < 3 && (start | next | pa) & lmask) ||
> +                    !sect_supported(level)){
>                         /*
>                          * This chunk needs a finer grained mapping. Create a
>                          * table mapping if necessary and recurse.
> 
>>
>> Thanks,
>> Ryan
>>
>>>
>>>  arch/arm64/include/asm/pgtable-hwdef.h |  1 +
>>>  arch/arm64/kernel/pi/map_range.c       | 23 +++++++++++++++++++++--
>>>  2 files changed, 22 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/arch/arm64/include/asm/pgtable-hwdef.h b/arch/arm64/include/asm/pgtable-hwdef.h
>>> index a9136cc551cc..fd0a82e8878c 100644
>>> --- a/arch/arm64/include/asm/pgtable-hwdef.h
>>> +++ b/arch/arm64/include/asm/pgtable-hwdef.h
>>> @@ -99,6 +99,7 @@
>>>  #define PGD_TYPE_TABLE		(_AT(pgdval_t, 3) << 0)
>>>  #define PGD_TABLE_BIT		(_AT(pgdval_t, 1) << 1)
>>>  #define PGD_TYPE_MASK		(_AT(pgdval_t, 3) << 0)
>>> +#define PGD_TYPE_SECT		(_AT(pgdval_t, 1) << 0)
>>>  #define PGD_TABLE_AF		(_AT(pgdval_t, 1) << 10)	/* Ignored if no FEAT_HAFT */
>>>  #define PGD_TABLE_PXN		(_AT(pgdval_t, 1) << 59)
>>>  #define PGD_TABLE_UXN		(_AT(pgdval_t, 1) << 60)
>>> diff --git a/arch/arm64/kernel/pi/map_range.c b/arch/arm64/kernel/pi/map_range.c
>>> index 2b69e3beeef8..9ea869f5745f 100644
>>> --- a/arch/arm64/kernel/pi/map_range.c
>>> +++ b/arch/arm64/kernel/pi/map_range.c
>>> @@ -44,8 +44,27 @@ void __init map_range(u64 *pte, u64 start, u64 end, u64 pa, pgprot_t prot,
>>>  	 * Set the right block/page bits for this level unless we are
>>>  	 * clearing the mapping
>>>  	 */
>>> -	if (protval)
>>> -		protval |= (level < 3) ? PMD_TYPE_SECT : PTE_TYPE_PAGE;
>>> +	if (protval) {
>>> +		switch (level) {
>>> +		case 3:
>>> +			protval |= PTE_TYPE_PAGE;
>>> +			break;
>>> +		case 2:
>>> +			protval |= PMD_TYPE_SECT;
>>> +			break;
>>> +		case 1:
>>> +			protval |= PUD_TYPE_SECT;
>>> +			break;
>>> +		case 0:
>>> +			protval |= P4D_TYPE_SECT;
>>> +			break;
>>> +		case -1:
>>> +			protval |= PGD_TYPE_SECT;
>>> +			break;
>>> +		default:
>>> +			break;
>>> +		}
>>> +	}
>>>  
>>>  	while (start < end) {
>>>  		u64 next = min((start | lmask) + 1, PAGE_ALIGN(end));
>>