Skip to contentSkip to conversation

mm/page_alloc: remove the static for local variable node_order

PATCH In review Hui Su 5 participants 8 messages Dec 30, 2020 – Jan 5, 2021

Subsystems (2)
lore
Share & tools
lore

Hui Su

author
#0 lore raw

PATCH 1/1 mm/page_alloc: remove the static for local variable node_order

local variable node_order do not need the static here.

---
 mm/page_alloc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
mm/page_alloc.c +1 -1
index bdbec4c98173..45e049ccf117 100644
@@ -5864,7 +5864,7 @@ static void build_thisnode_zonelists(pg_data_t *pgdat)
 
 static void build_zonelists(pg_data_t *pgdat)
 {
-	static int node_order[MAX_NUMNODES];
+	int node_order[MAX_NUMNODES];
 	int node, load, nr_nodes = 0;
 	nodemask_t used_mask = NODE_MASK_NONE;
 	int local_node, prev_node;
Signature · 1 line
2.25.1
#1 lore raw
Quoted 1 line · Hui Su

local variable node_order do not need the static here.

It bloody well does. It can be up to 2^10 entries on x86 (and larger on others) That's 4kB which you've now moved onto the stack.

Please, learn more about what you're doing. I suggest sending patches to drivers/staging; that will help you learn how to submit patches to linux.

author
#2 lore raw

Hi, Matthew:

Quoted 6 lines · Matthew Wilcox
Quoted 1 line · Hui Su

local variable node_order do not need the static here.

It bloody well does. It can be up to 2^10 entries on x86 (and larger on others) That's 4kB which you've now moved onto the stack.

Thanks for your explanation, this change will put too much onto the stack in some cases which i didn't consider.

Please ignore this change.

Quoted 3 lines

Please, learn more about what you're doing. I suggest sending patches to drivers/staging; that will help you learn how to submit patches to linux.

#3 lore raw
In reply to Matthew Wilcox
Quoted 6 lines · Matthew Wilcox
Quoted 1 line · Hui Su

local variable node_order do not need the static here.

It bloody well does. It can be up to 2^10 entries on x86 (and larger on others) That's 4kB which you've now moved onto the stack.

This is not the first time I have seen similar changes. So what do you think about adding a comment here to avoid another one do this in the feature?

Quoted 4 lines

Please, learn more about what you're doing. I suggest sending patches to drivers/staging; that will help you learn how to submit patches to linux.

reviewer
#4 lore raw
Quoted 11 lines · Muchun Song
Quoted 6 lines · Matthew Wilcox
Quoted 1 line · Hui Su

local variable node_order do not need the static here.

It bloody well does. It can be up to 2^10 entries on x86 (and larger on others) That's 4kB which you've now moved onto the stack.

This is not the first time I have seen similar changes. So what do you think about adding a comment here to avoid another one do this in the feature?

Well, this is not an unusual technieque to reduce the stack space. I am not really sure we really need to put an explicit comment about that. I would appreciate much more if patch submitters took an extra step when creating seemingly trivial patches and either consult the history of the respective code or look for a similar pattern elsewhere before sending them.

I do agree with Willy that mm code is usually not a great place to search for trivial patches. First of all most people tend to be pretty busy with other reviewes and the code has grown rather delicate and tricky so each review is non trivial.

Signature · 2 lines
Michal Hocko
SUSE Labs
maintainer
#5 lore raw
In reply to Matthew Wilcox
Quoted 5 lines · On Wed, 30 Dec 2020 12:42:33 +0000 Matthew Wilcox <[email protected]> wrote:
Quoted 1 line · Hui Su

local variable node_order do not need the static here.

It bloody well does. It can be up to 2^10 entries on x86 (and larger on others) That's 4kB which you've now moved onto the stack.

That being said, could we kmalloc the scratch area in __build_all_zonelists()? And maybe remove that static spinlock?

(what blocks node and cpu hotplug in there??)

#6 lore raw
Quoted 5 lines · Andrew Morton
Quoted 3 lines · On Wed, 30 Dec 2020 12:42:33 +0000 Matthew Wilcox <[email protected]> wrote:
Quoted 1 line · Hui Su

local variable node_order do not need the static here.

Quoted 7 lines
Quoted 2 lines

It bloody well does. It can be up to 2^10 entries on x86 (and larger on others) That's 4kB which you've now moved onto the stack.

That being said, could we kmalloc the scratch area in __build_all_zonelists()? And maybe remove that static spinlock?

(what blocks node and cpu hotplug in there??)

if we don't have the zonelists built yet, can slab possibly be operational?

reviewer
#7 lore raw
In reply to Andrew Morton
Quoted 10 lines · Andrew Morton
Quoted 5 lines · On Wed, 30 Dec 2020 12:42:33 +0000 Matthew Wilcox <[email protected]> wrote:
Quoted 1 line · Hui Su

local variable node_order do not need the static here.

It bloody well does. It can be up to 2^10 entries on x86 (and larger on others) That's 4kB which you've now moved onto the stack.

That being said, could we kmalloc the scratch area in __build_all_zonelists()? And maybe remove that static spinlock?

I am not sure we can (e.g. early init code) but even if we could, what would be an advantage. This code is called very seldom with a very shallow stacks so using the stack allocation sounds like the easiest thing to do.

Quoted 1 line

(what blocks node and cpu hotplug in there??)

Memory hotplug is excluded by the caller when it matters (e.g. no locking for the early init).

Signature · 2 lines
Michal Hocko
SUSE Labs