[PATCH 1/2] mm/sparse: pass the __highest_present_section_nr + 1 to alloc_func()

[PATCH 1/2] mm/sparse: pass the __highest_present_section_nr + 1 to alloc_func()

[Wei Yang]

In 'commit c4e1be9ec113 ("mm, sparsemem: break out of loops early")',
__highest_present_section_nr is introduced to reduce the loop counts for
present section. This is also helpful for usemap and memmap allocation.

This patch uses __highest_present_section_nr + 1 to optimize the loop.

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
---
 mm/sparse.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/sparse.c b/mm/sparse.c
index 7af5e7a92528..505050346249 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -561,7 +561,7 @@ static void __init alloc_usemap_and_memmap(void (*alloc_func)
 		map_count = 1;
 	}
 	/* ok, last chunk */
-	alloc_func(data, pnum_begin, NR_MEM_SECTIONS,
+	alloc_func(data, pnum_begin, __highest_present_section_nr+1,
 						map_count, nodeid_begin);
 }
 
-- 
2.15.1

[PATCH 2/2] mm/sparse: check __highest_present_section_nr only for a present section

[Wei Yang]

When searching a present section, there are two boundaries:

    * __highest_present_section_nr
    * NR_MEM_SECTIONS

And it is konwn, __highest_present_section_nr is a more strict boundary
than NR_MEM_SECTIONS. This means it would be necessary to check
__highest_present_section_nr only.

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
---
 mm/sparse.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/mm/sparse.c b/mm/sparse.c
index 505050346249..b6560029a16c 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -190,15 +190,13 @@ static inline int next_present_section_nr(int section_nr)
 		section_nr++;
 		if (present_section_nr(section_nr))
 			return section_nr;
-	} while ((section_nr < NR_MEM_SECTIONS) &&
-		 (section_nr <= __highest_present_section_nr));
+	} while ((section_nr <= __highest_present_section_nr));
 
 	return -1;
 }
 #define for_each_present_section_nr(start, section_nr)		\
 	for (section_nr = next_present_section_nr(start-1);	\
 	     ((section_nr >= 0) &&				\
-	      (section_nr < NR_MEM_SECTIONS) &&			\
 	      (section_nr <= __highest_present_section_nr));	\
 	     section_nr = next_present_section_nr(section_nr))
 
-- 
2.15.1

Re: [PATCH 1/2] mm/sparse: pass the __highest_present_section_nr + 1 to alloc_func()

[David Rientjes]

On Mon, 26 Mar 2018, Wei Yang wrote:

> In 'commit c4e1be9ec113 ("mm, sparsemem: break out of loops early")',
> __highest_present_section_nr is introduced to reduce the loop counts for
> present section. This is also helpful for usemap and memmap allocation.
> 
> This patch uses __highest_present_section_nr + 1 to optimize the loop.
> 
> Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
> ---
>  mm/sparse.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/sparse.c b/mm/sparse.c
> index 7af5e7a92528..505050346249 100644
> --- a/mm/sparse.c
> +++ b/mm/sparse.c
> @@ -561,7 +561,7 @@ static void __init alloc_usemap_and_memmap(void (*alloc_func)
>  		map_count = 1;
>  	}
>  	/* ok, last chunk */
> -	alloc_func(data, pnum_begin, NR_MEM_SECTIONS,
> +	alloc_func(data, pnum_begin, __highest_present_section_nr+1,
>  						map_count, nodeid_begin);
>  }
>  

What happens if s/NR_MEM_SECTIONS/pnum/?

Re: [PATCH 1/2] mm/sparse: pass the __highest_present_section_nr + 1 to alloc_func()

[Wei Yang]

On Mon, Mar 26, 2018 at 01:56:50PM -0700, David Rientjes wrote:
>On Mon, 26 Mar 2018, Wei Yang wrote:
>
>> In 'commit c4e1be9ec113 ("mm, sparsemem: break out of loops early")',
>> __highest_present_section_nr is introduced to reduce the loop counts for
>> present section. This is also helpful for usemap and memmap allocation.
>> 
>> This patch uses __highest_present_section_nr + 1 to optimize the loop.
>> 
>> Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
>> ---
>>  mm/sparse.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>> 
>> diff --git a/mm/sparse.c b/mm/sparse.c
>> index 7af5e7a92528..505050346249 100644
>> --- a/mm/sparse.c
>> +++ b/mm/sparse.c
>> @@ -561,7 +561,7 @@ static void __init alloc_usemap_and_memmap(void (*alloc_func)
>>  		map_count = 1;
>>  	}
>>  	/* ok, last chunk */
>> -	alloc_func(data, pnum_begin, NR_MEM_SECTIONS,
>> +	alloc_func(data, pnum_begin, __highest_present_section_nr+1,
>>  						map_count, nodeid_begin);
>>  }
>>  
>
>What happens if s/NR_MEM_SECTIONS/pnum/?

I have tried this :-)

The last pnum is -1 from next_present_section_nr().

-- 
Wei Yang
Help you, Help me

Re: [PATCH 1/2] mm/sparse: pass the __highest_present_section_nr + 1 to alloc_func()

[David Rientjes]

On Tue, 27 Mar 2018, Wei Yang wrote:

> >> In 'commit c4e1be9ec113 ("mm, sparsemem: break out of loops early")',
> >> __highest_present_section_nr is introduced to reduce the loop counts for
> >> present section. This is also helpful for usemap and memmap allocation.
> >> 
> >> This patch uses __highest_present_section_nr + 1 to optimize the loop.
> >> 
> >> Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
> >> ---
> >>  mm/sparse.c | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >> 
> >> diff --git a/mm/sparse.c b/mm/sparse.c
> >> index 7af5e7a92528..505050346249 100644
> >> --- a/mm/sparse.c
> >> +++ b/mm/sparse.c
> >> @@ -561,7 +561,7 @@ static void __init alloc_usemap_and_memmap(void (*alloc_func)
> >>  		map_count = 1;
> >>  	}
> >>  	/* ok, last chunk */
> >> -	alloc_func(data, pnum_begin, NR_MEM_SECTIONS,
> >> +	alloc_func(data, pnum_begin, __highest_present_section_nr+1,
> >>  						map_count, nodeid_begin);
> >>  }
> >>  
> >
> >What happens if s/NR_MEM_SECTIONS/pnum/?
> 
> I have tried this :-)
> 
> The last pnum is -1 from next_present_section_nr().
> 

Lol.  I think it would make more sense for the second patch to come before 
the first, but feel free to add

Acked-by: David Rientjes <rientjes@google.com>

Re: [PATCH 1/2] mm/sparse: pass the __highest_present_section_nr + 1 to alloc_func()

[Wei Yang]

On Mon, Mar 26, 2018 at 03:47:03PM -0700, David Rientjes wrote:
>On Tue, 27 Mar 2018, Wei Yang wrote:
>
>> >> In 'commit c4e1be9ec113 ("mm, sparsemem: break out of loops early")',
>> >> __highest_present_section_nr is introduced to reduce the loop counts for
>> >> present section. This is also helpful for usemap and memmap allocation.
>> >> 
>> >> This patch uses __highest_present_section_nr + 1 to optimize the loop.
>> >> 
>> >> Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
>> >> ---
>> >>  mm/sparse.c | 2 +-
>> >>  1 file changed, 1 insertion(+), 1 deletion(-)
>> >> 
>> >> diff --git a/mm/sparse.c b/mm/sparse.c
>> >> index 7af5e7a92528..505050346249 100644
>> >> --- a/mm/sparse.c
>> >> +++ b/mm/sparse.c
>> >> @@ -561,7 +561,7 @@ static void __init alloc_usemap_and_memmap(void (*alloc_func)
>> >>  		map_count = 1;
>> >>  	}
>> >>  	/* ok, last chunk */
>> >> -	alloc_func(data, pnum_begin, NR_MEM_SECTIONS,
>> >> +	alloc_func(data, pnum_begin, __highest_present_section_nr+1,
>> >>  						map_count, nodeid_begin);
>> >>  }
>> >>  
>> >
>> >What happens if s/NR_MEM_SECTIONS/pnum/?
>> 
>> I have tried this :-)
>> 
>> The last pnum is -1 from next_present_section_nr().
>> 
>
>Lol.  I think it would make more sense for the second patch to come before 
>the first, but feel free to add
>

Thanks for your comment.

Do I need to reorder the patch and send v2?

>Acked-by: David Rientjes <rientjes@google.com>

-- 
Wei Yang
Help you, Help me

Re: [PATCH 1/2] mm/sparse: pass the __highest_present_section_nr + 1 to alloc_func()

[David Rientjes]

On Tue, 27 Mar 2018, Wei Yang wrote:

> >Lol.  I think it would make more sense for the second patch to come before 
> >the first
> 
> Thanks for your comment.
> 
> Do I need to reorder the patch and send v2?
> 

I think we can just ask Andrew to apply backwards, but it's not crucial.  
The ordering of patch 2 before patch 1 simply helped me to understand the 
boundaries better.

Re: [PATCH 1/2] mm/sparse: pass the __highest_present_section_nr + 1 to alloc_func()

[Wei Yang]

On Mon, Mar 26, 2018 at 03:58:28PM -0700, David Rientjes wrote:
>On Tue, 27 Mar 2018, Wei Yang wrote:
>
>> >Lol.  I think it would make more sense for the second patch to come before 
>> >the first
>> 
>> Thanks for your comment.
>> 
>> Do I need to reorder the patch and send v2?
>> 
>
>I think we can just ask Andrew to apply backwards, but it's not crucial.  
>The ordering of patch 2 before patch 1 simply helped me to understand the 
>boundaries better.

Ah, got it.

Actually, the original order is what you expected. While for some mysterious
reasons, I reordered them :-(

Maybe you are right, it would be more easy to understand with patch 2 before
patch 1. :-)

Have a good day~

-- 
Wei Yang
Help you, Help me

[PATCH] mm: check __highest_present_sectioin_nr directly in memory_dev_init()

[Wei Yang]

__highest_present_section_nr is a more strict boundary than
NR_MEM_SECTIONS. So check __highest_present_sectioin_nr directly is enough.

Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
---
 drivers/base/memory.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/drivers/base/memory.c b/drivers/base/memory.c
index fe4b24f05f6a..e79e3361f632 100644
--- a/drivers/base/memory.c
+++ b/drivers/base/memory.c
@@ -833,11 +833,8 @@ int __init memory_dev_init(void)
 	 * during boot and have been initialized
 	 */
 	mutex_lock(&mem_sysfs_mutex);
-	for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) {
-		/* Don't iterate over sections we know are !present: */
-		if (i > __highest_present_section_nr)
-			break;
-
+	for (i = 0; i <= __highest_present_section_nr;
+		i += sections_per_block) {
 		err = add_memory_block(i);
 		if (!ret)
 			ret = err;
-- 
2.15.1

Re: [PATCH 1/2] mm/sparse: pass the __highest_present_section_nr + 1 to alloc_func()

[Wei Yang]

Hi, Andrew

I saw you merged one related patch recently, not sure you would take these two?

On Mon, Mar 26, 2018 at 04:19:55PM +0800, Wei Yang wrote:
>In 'commit c4e1be9ec113 ("mm, sparsemem: break out of loops early")',
>__highest_present_section_nr is introduced to reduce the loop counts for
>present section. This is also helpful for usemap and memmap allocation.
>
>This patch uses __highest_present_section_nr + 1 to optimize the loop.
>
>Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
>---
> mm/sparse.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
>diff --git a/mm/sparse.c b/mm/sparse.c
>index 7af5e7a92528..505050346249 100644
>--- a/mm/sparse.c
>+++ b/mm/sparse.c
>@@ -561,7 +561,7 @@ static void __init alloc_usemap_and_memmap(void (*alloc_func)
> 		map_count = 1;
> 	}
> 	/* ok, last chunk */
>-	alloc_func(data, pnum_begin, NR_MEM_SECTIONS,
>+	alloc_func(data, pnum_begin, __highest_present_section_nr+1,
> 						map_count, nodeid_begin);
> }
> 
>-- 
>2.15.1

-- 
Wei Yang
Help you, Help me