[PATCH] mm: convert scan_control.priority int => byte

[PATCH] mm: convert scan_control.priority int => byte

[Greg Thelen]

Reclaim priorities range from 0..12(DEF_PRIORITY).
scan_control.priority is a 4 byte int, which is overkill.

Since commit 6538b8ea886e ("x86_64: expand kernel stack to 16K") x86_64
stack overflows are not an issue.  But it's inefficient to use 4 bytes
for priority.

Use s8 (signed byte) rather than u8 to allow for loops like:
	do {
		...
	} while (--sc.priority >= 0);

This reduces sizeof(struct scan_control) from 96 => 88 bytes (x86_64),
which saves some stack.

scan_control.priority field order is changed to occupy otherwise unused
padding.

Signed-off-by: Greg Thelen <gthelen@google.com>
---
 mm/vmscan.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 9b697323a88c..541c334bd176 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -83,9 +83,6 @@ struct scan_control {
 	 */
 	struct mem_cgroup *target_mem_cgroup;
 
-	/* Scan (total_size >> priority) pages at once */
-	int priority;
-
 	/* The highest zone to isolate pages for reclaim from */
 	enum zone_type reclaim_idx;
 
@@ -111,6 +108,9 @@ struct scan_control {
 	/* One of the zones is ready for compaction */
 	unsigned int compaction_ready:1;
 
+	/* Scan (total_size >> priority) pages at once */
+	s8 priority;
+
 	/* Incremented by the number of inactive pages that were scanned */
 	unsigned long nr_scanned;
 
-- 
2.17.0.921.gf22659ad46-goog

Re: [PATCH] mm: convert scan_control.priority int => byte

[Matthew Wilcox]

On Mon, May 28, 2018 at 07:40:25PM -0700, Greg Thelen wrote:
> Reclaim priorities range from 0..12(DEF_PRIORITY).
> scan_control.priority is a 4 byte int, which is overkill.
> 
> Since commit 6538b8ea886e ("x86_64: expand kernel stack to 16K") x86_64
> stack overflows are not an issue.  But it's inefficient to use 4 bytes
> for priority.

If you're looking to shave a few more bytes, allocation order can fit
in a u8 too (can't be more than 6 bits, and realistically won't be more
than 4 bits).  reclaim_idx likewise will fit in a u8, and actually won't
be more than 3 bits.

I am sceptical that nr_to_reclaim should really be an unsigned long; I
don't think we should be trying to free 4 billion pages in a single call.
nr_scanned might be over 4 billion (!) but nr_reclaimed can probably
shrink to unsigned int along with nr_to_reclaim.

Re: [PATCH] mm: convert scan_control.priority int => byte

[Greg Thelen]

Matthew Wilcox <willy@infradead.org> wrote:

> On Mon, May 28, 2018 at 07:40:25PM -0700, Greg Thelen wrote:
>> Reclaim priorities range from 0..12(DEF_PRIORITY).
>> scan_control.priority is a 4 byte int, which is overkill.
>> 
>> Since commit 6538b8ea886e ("x86_64: expand kernel stack to 16K") x86_64
>> stack overflows are not an issue.  But it's inefficient to use 4 bytes
>> for priority.
>
> If you're looking to shave a few more bytes, allocation order can fit
> in a u8 too (can't be more than 6 bits, and realistically won't be more
> than 4 bits).  reclaim_idx likewise will fit in a u8, and actually won't
> be more than 3 bits.

Nod.  Good tip.  Included in ("[PATCH v2] mm: condense scan_control").

> I am sceptical that nr_to_reclaim should really be an unsigned long; I
> don't think we should be trying to free 4 billion pages in a single call.
> nr_scanned might be over 4 billion (!) but nr_reclaimed can probably
> shrink to unsigned int along with nr_to_reclaim.

Agreed.  For patch simplicity, I'll pass on this for now.

[PATCH v2] mm: condense scan_control

[Greg Thelen]

Use smaller scan_control fields for order, priority, and reclaim_idx.
Convert fields from int => s8.  All easily fit within a byte:
* allocation order range: 0..MAX_ORDER(64?)
* priority range:         0..12(DEF_PRIORITY)
* reclaim_idx range:      0..6(__MAX_NR_ZONES)

Since commit 6538b8ea886e ("x86_64: expand kernel stack to 16K") x86_64
stack overflows are not an issue.  But it's inefficient to use ints.

Use s8 (signed byte) rather than u8 to allow for loops like:
	do {
		...
	} while (--sc.priority >= 0);

Add BUILD_BUG_ON to verify that s8 is capable of storing max values.

This reduces sizeof(struct scan_control):
* 96 => 80 bytes (x86_64)
* 68 => 56 bytes (i386)

scan_control structure field order is changed to utilize padding.
After this patch there is 1 bit of scan_control padding.

Signed-off-by: Greg Thelen <gthelen@google.com>
Suggested-by: Matthew Wilcox <willy@infradead.org>
---
 mm/vmscan.c | 32 ++++++++++++++++++++------------
 1 file changed, 20 insertions(+), 12 deletions(-)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 9b697323a88c..42731faea306 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -65,12 +65,6 @@ struct scan_control {
 	/* How many pages shrink_list() should reclaim */
 	unsigned long nr_to_reclaim;
 
-	/* This context's GFP mask */
-	gfp_t gfp_mask;
-
-	/* Allocation order */
-	int order;
-
 	/*
 	 * Nodemask of nodes allowed by the caller. If NULL, all nodes
 	 * are scanned.
@@ -83,12 +77,6 @@ struct scan_control {
 	 */
 	struct mem_cgroup *target_mem_cgroup;
 
-	/* Scan (total_size >> priority) pages at once */
-	int priority;
-
-	/* The highest zone to isolate pages for reclaim from */
-	enum zone_type reclaim_idx;
-
 	/* Writepage batching in laptop mode; RECLAIM_WRITE */
 	unsigned int may_writepage:1;
 
@@ -111,6 +99,18 @@ struct scan_control {
 	/* One of the zones is ready for compaction */
 	unsigned int compaction_ready:1;
 
+	/* Allocation order */
+	s8 order;
+
+	/* Scan (total_size >> priority) pages at once */
+	s8 priority;
+
+	/* The highest zone to isolate pages for reclaim from */
+	s8 reclaim_idx;
+
+	/* This context's GFP mask */
+	gfp_t gfp_mask;
+
 	/* Incremented by the number of inactive pages that were scanned */
 	unsigned long nr_scanned;
 
@@ -3047,6 +3047,14 @@ unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
 		.may_swap = 1,
 	};
 
+	/*
+	 * scan_control uses s8 fields for order, priority, and reclaim_idx.
+	 * Confirm they are large enough for max values.
+	 */
+	BUILD_BUG_ON(MAX_ORDER > S8_MAX);
+	BUILD_BUG_ON(DEF_PRIORITY > S8_MAX);
+	BUILD_BUG_ON(MAX_NR_ZONES > S8_MAX);
+
 	/*
 	 * Do not enter reclaim if fatal signal was delivered while throttled.
 	 * 1 is returned so that the page allocator does not OOM kill at this
-- 
2.17.0.921.gf22659ad46-goog

Re: [PATCH v2] mm: condense scan_control

[Greg Thelen]

On Tue, May 29, 2018 at 11:12 PM Greg Thelen <gthelen@google.com> wrote:
>
> Use smaller scan_control fields for order, priority, and reclaim_idx.
> Convert fields from int => s8.  All easily fit within a byte:
> * allocation order range: 0..MAX_ORDER(64?)
> * priority range:         0..12(DEF_PRIORITY)
> * reclaim_idx range:      0..6(__MAX_NR_ZONES)
>
> Since commit 6538b8ea886e ("x86_64: expand kernel stack to 16K") x86_64
> stack overflows are not an issue.  But it's inefficient to use ints.
>
> Use s8 (signed byte) rather than u8 to allow for loops like:
>         do {
>                 ...
>         } while (--sc.priority >= 0);
>
> Add BUILD_BUG_ON to verify that s8 is capable of storing max values.
>
> This reduces sizeof(struct scan_control):
> * 96 => 80 bytes (x86_64)
> * 68 => 56 bytes (i386)
>
> scan_control structure field order is changed to utilize padding.
> After this patch there is 1 bit of scan_control padding.
>
> Signed-off-by: Greg Thelen <gthelen@google.com>
> Suggested-by: Matthew Wilcox <willy@infradead.org>

Is there any interest in this?  Less stack usage could mean less
dcache and dtlb pressure.  But I understand if the complexity is
distasteful.

> ---
>  mm/vmscan.c | 32 ++++++++++++++++++++------------
>  1 file changed, 20 insertions(+), 12 deletions(-)
>
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index 9b697323a88c..42731faea306 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -65,12 +65,6 @@ struct scan_control {
>         /* How many pages shrink_list() should reclaim */
>         unsigned long nr_to_reclaim;
>
> -       /* This context's GFP mask */
> -       gfp_t gfp_mask;
> -
> -       /* Allocation order */
> -       int order;
> -
>         /*
>          * Nodemask of nodes allowed by the caller. If NULL, all nodes
>          * are scanned.
> @@ -83,12 +77,6 @@ struct scan_control {
>          */
>         struct mem_cgroup *target_mem_cgroup;
>
> -       /* Scan (total_size >> priority) pages at once */
> -       int priority;
> -
> -       /* The highest zone to isolate pages for reclaim from */
> -       enum zone_type reclaim_idx;
> -
>         /* Writepage batching in laptop mode; RECLAIM_WRITE */
>         unsigned int may_writepage:1;
>
> @@ -111,6 +99,18 @@ struct scan_control {
>         /* One of the zones is ready for compaction */
>         unsigned int compaction_ready:1;
>
> +       /* Allocation order */
> +       s8 order;
> +
> +       /* Scan (total_size >> priority) pages at once */
> +       s8 priority;
> +
> +       /* The highest zone to isolate pages for reclaim from */
> +       s8 reclaim_idx;
> +
> +       /* This context's GFP mask */
> +       gfp_t gfp_mask;
> +
>         /* Incremented by the number of inactive pages that were scanned */
>         unsigned long nr_scanned;
>
> @@ -3047,6 +3047,14 @@ unsigned long try_to_free_pages(struct zonelist *zonelist, int order,
>                 .may_swap = 1,
>         };
>
> +       /*
> +        * scan_control uses s8 fields for order, priority, and reclaim_idx.
> +        * Confirm they are large enough for max values.
> +        */
> +       BUILD_BUG_ON(MAX_ORDER > S8_MAX);
> +       BUILD_BUG_ON(DEF_PRIORITY > S8_MAX);
> +       BUILD_BUG_ON(MAX_NR_ZONES > S8_MAX);
> +
>         /*
>          * Do not enter reclaim if fatal signal was delivered while throttled.
>          * 1 is returned so that the page allocator does not OOM kill at this
> --
> 2.17.0.921.gf22659ad46-goog
>