diff mbox

[v4,3/4] mm/sparse: Add a new parameter 'data_unit_size' for alloc_usemap_and_memmap

Message ID 20180521101555.25610-4-bhe@redhat.com (mailing list archive)
State New, archived
Headers show

Commit Message

Baoquan He May 21, 2018, 10:15 a.m. UTC
It's used to pass the size of map data unit into alloc_usemap_and_memmap,
and is preparation for next patch.

Signed-off-by: Baoquan He <bhe@redhat.com>
---
 mm/sparse.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

Comments

Dave Hansen June 7, 2018, 10:48 p.m. UTC | #1
On 05/21/2018 03:15 AM, Baoquan He wrote:
> It's used to pass the size of map data unit into alloc_usemap_and_memmap,
> and is preparation for next patch.

This is the "what", but not the "why".  Could you add another sentence
or two to explain why we need this?
Baoquan He June 8, 2018, 6:27 a.m. UTC | #2
On 06/07/18 at 03:48pm, Dave Hansen wrote:
> On 05/21/2018 03:15 AM, Baoquan He wrote:
> > It's used to pass the size of map data unit into alloc_usemap_and_memmap,
> > and is preparation for next patch.
> 
> This is the "what", but not the "why".  Could you add another sentence
> or two to explain why we need this?

Thanks for reviewing, Dave.

In alloc_usemap_and_memmap(), it will call
sparse_early_usemaps_alloc_node() or sparse_early_mem_maps_alloc_node()
to allocate usemap and memmap for each node and install them into
usemap_map[] and map_map[]. Here we need pass in the number of present
sections on this node so that we can move pointer of usemap_map[] and
map_map[] to right position.

How do think about above words?

Thanks
Baoquan
Dave Hansen June 8, 2018, 2:20 p.m. UTC | #3
On 06/07/2018 11:27 PM, Baoquan He wrote:
> In alloc_usemap_and_memmap(), it will call
> sparse_early_usemaps_alloc_node() or sparse_early_mem_maps_alloc_node()
> to allocate usemap and memmap for each node and install them into
> usemap_map[] and map_map[]. Here we need pass in the number of present
> sections on this node so that we can move pointer of usemap_map[] and
> map_map[] to right position.
> 
> How do think about above words?

But you're now passing in the size of the data structure.  Why is that
needed all of a sudden?
Baoquan He June 8, 2018, 3:17 p.m. UTC | #4
On 06/08/18 at 07:20am, Dave Hansen wrote:
> On 06/07/2018 11:27 PM, Baoquan He wrote:
> > In alloc_usemap_and_memmap(), it will call
> > sparse_early_usemaps_alloc_node() or sparse_early_mem_maps_alloc_node()
> > to allocate usemap and memmap for each node and install them into
> > usemap_map[] and map_map[]. Here we need pass in the number of present
> > sections on this node so that we can move pointer of usemap_map[] and
> > map_map[] to right position.
> > 
> > How do think about above words?
> 
> But you're now passing in the size of the data structure.  Why is that
> needed all of a sudden?

Oh, it's the size of the data structure. Because
alloc_usemap_and_memmap() is reused for both usemap and memmap
allocation, then inside alloc_usemap_and_memmap(), we need move forward
the passed in pointer which points at the starting address of usemap_map
or memmap_map, to a new position which points at the next starting
address on new node.

You can see we passed the usemap_map, map_map, and the array element
size.  

void __init sparse_init(void)
{
	...
        alloc_usemap_and_memmap(sparse_early_usemaps_alloc_node,                                                                                  
                                (void *)usemap_map,
                                sizeof(usemap_map[0]));
	...
        alloc_usemap_and_memmap(sparse_early_mem_maps_alloc_node,
                                (void *)map_map,
                                sizeof(map_map[0]));
	...
}

Then inside alloc_usemap_and_memmap(), For each node, we get how many
present sections on this node, call hook alloc_func(). Then we update
the pointer to point at a new position of usemap_map[] or map_map[].

Here usemap_map[] is (unsigned long *), map_map[] is (struct page*).
Even though both of them are pointer, I think it might be not good to
assume that in alloc_usemap_and_memmap() because
alloc_usemap_and_memmap() doesn't know what is stored in its 2nd
parameter, namely (void * data). So add this new parameter to tell it.

Combine with patch 4/4, it might be easier to understand.

static void __init alloc_usemap_and_memmap(void (*alloc_func)
..
{
	...
	for_each_present_section_nr(pnum_begin + 1, pnum) {
                alloc_func(data, pnum_begin, pnum,                                                                                                
                                                map_count, nodeid_begin);
		...
		data += map_count * data_unit_size; 
		...
	}
	...
}
Dave Hansen June 8, 2018, 4:13 p.m. UTC | #5
On 06/08/2018 08:17 AM, Baoquan He wrote:
> 
> Then inside alloc_usemap_and_memmap(), For each node, we get how many
> present sections on this node, call hook alloc_func(). Then we update
> the pointer to point at a new position of usemap_map[] or map_map[].

I think this is the key.

alloc_usemap_and_memmap() is passed in a "void *" that it needs to
update as things get consumed.  But, it knows only the quantity of
objects consumed and not the type.  This effectively tells it enough
about the type to let it update the pointer as objects are consumed.

Right?

Can we get that in the changelog?
Baoquan He June 10, 2018, 11:32 p.m. UTC | #6
On 06/08/18 at 09:13am, Dave Hansen wrote:
> On 06/08/2018 08:17 AM, Baoquan He wrote:
> > 
> > Then inside alloc_usemap_and_memmap(), For each node, we get how many
> > present sections on this node, call hook alloc_func(). Then we update
> > the pointer to point at a new position of usemap_map[] or map_map[].
> 
> I think this is the key.
> 
> alloc_usemap_and_memmap() is passed in a "void *" that it needs to
> update as things get consumed.  But, it knows only the quantity of
> objects consumed and not the type.  This effectively tells it enough
> about the type to let it update the pointer as objects are consumed.
> 
> Right?
> 
> Can we get that in the changelog?

Hmm, I like above sentences very much, thanks.

Do you means putting it in changelog, but not commit log of patch 3/4,
right? I can do this when repost.

Thanks
Baoquan
diff mbox

Patch

diff --git a/mm/sparse.c b/mm/sparse.c
index 3d697292be08..4a58f8809542 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -491,10 +491,12 @@  void __weak __meminit vmemmap_populate_print_last(void)
 /**
  *  alloc_usemap_and_memmap - memory alloction for pageblock flags and vmemmap
  *  @map: usemap_map for pageblock flags or mmap_map for vmemmap
+ *  @unit_size: size of map unit
  */
 static void __init alloc_usemap_and_memmap(void (*alloc_func)
 					(void *, unsigned long, unsigned long,
-					unsigned long, int), void *data)
+					unsigned long, int), void *data,
+					int data_unit_size)
 {
 	unsigned long pnum;
 	unsigned long map_count;
@@ -571,7 +573,8 @@  void __init sparse_init(void)
 	if (!usemap_map)
 		panic("can not allocate usemap_map\n");
 	alloc_usemap_and_memmap(sparse_early_usemaps_alloc_node,
-							(void *)usemap_map);
+				(void *)usemap_map,
+				sizeof(usemap_map[0]));
 
 #ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
 	size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
@@ -579,7 +582,8 @@  void __init sparse_init(void)
 	if (!map_map)
 		panic("can not allocate map_map\n");
 	alloc_usemap_and_memmap(sparse_early_mem_maps_alloc_node,
-							(void *)map_map);
+				(void *)map_map,
+				sizeof(map_map[0]));
 #endif
 
 	for_each_present_section_nr(0, pnum) {