2020-03-26 22:25:21

by Wei Yang

[permalink] [raw]
Subject: [PATCH 1/2] mm/page_alloc.c: leverage compiler to zero out used_mask

Since we always clear used_mask before getting node order, we can
leverage compiler to do this instead of at run time.

Signed-off-by: Wei Yang <[email protected]>
---
mm/page_alloc.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 0e823bca3f2f..2144b6ceb119 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -5587,14 +5587,13 @@ static void build_zonelists(pg_data_t *pgdat)
{
static int node_order[MAX_NUMNODES];
int node, load, nr_nodes = 0;
- nodemask_t used_mask;
+ nodemask_t used_mask = {.bits = {0}};
int local_node, prev_node;

/* NUMA-aware ordering of nodes */
local_node = pgdat->node_id;
load = nr_online_nodes;
prev_node = local_node;
- nodes_clear(used_mask);

memset(node_order, 0, sizeof(node_order));
while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
--
2.23.0


2020-03-26 22:26:33

by Wei Yang

[permalink] [raw]
Subject: [PATCH 2/2] mm/page_alloc.c: define node_order with all zero

Since we always clear node_order before getting it, we can leverage
compiler to do this instead of at run time.

Signed-off-by: Wei Yang <[email protected]>
---
mm/page_alloc.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 2144b6ceb119..7a6435cffdb7 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -5585,7 +5585,7 @@ static void build_thisnode_zonelists(pg_data_t *pgdat)

static void build_zonelists(pg_data_t *pgdat)
{
- static int node_order[MAX_NUMNODES];
+ static int node_order[MAX_NUMNODES] = {0};
int node, load, nr_nodes = 0;
nodemask_t used_mask = {.bits = {0}};
int local_node, prev_node;
@@ -5595,7 +5595,6 @@ static void build_zonelists(pg_data_t *pgdat)
load = nr_online_nodes;
prev_node = local_node;

- memset(node_order, 0, sizeof(node_order));
while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
/*
* We don't want to pressure a particular node.
--
2.23.0

2020-03-26 22:36:52

by Jason Gunthorpe

[permalink] [raw]
Subject: Re: [PATCH 1/2] mm/page_alloc.c: leverage compiler to zero out used_mask

On Thu, Mar 26, 2020 at 10:24:44PM +0000, Wei Yang wrote:
> Since we always clear used_mask before getting node order, we can
> leverage compiler to do this instead of at run time.
>
> Signed-off-by: Wei Yang <[email protected]>
> mm/page_alloc.c | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 0e823bca3f2f..2144b6ceb119 100644
> +++ b/mm/page_alloc.c
> @@ -5587,14 +5587,13 @@ static void build_zonelists(pg_data_t *pgdat)
> {
> static int node_order[MAX_NUMNODES];
> int node, load, nr_nodes = 0;
> - nodemask_t used_mask;
> + nodemask_t used_mask = {.bits = {0}};

If this style is to be done it should just be '= {}';

This case demonstrates why the popular '= {0}' idiom is not such a
good idea, as it only works if the first member is an integral type.

Jason

2020-03-27 09:33:28

by David Hildenbrand

[permalink] [raw]
Subject: Re: [PATCH 1/2] mm/page_alloc.c: leverage compiler to zero out used_mask

On 26.03.20 23:24, Wei Yang wrote:
> Since we always clear used_mask before getting node order, we can
> leverage compiler to do this instead of at run time.
>
> Signed-off-by: Wei Yang <[email protected]>
> ---
> mm/page_alloc.c | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 0e823bca3f2f..2144b6ceb119 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -5587,14 +5587,13 @@ static void build_zonelists(pg_data_t *pgdat)
> {
> static int node_order[MAX_NUMNODES];
> int node, load, nr_nodes = 0;
> - nodemask_t used_mask;
> + nodemask_t used_mask = {.bits = {0}};
> int local_node, prev_node;
>
> /* NUMA-aware ordering of nodes */
> local_node = pgdat->node_id;
> load = nr_online_nodes;
> prev_node = local_node;
> - nodes_clear(used_mask);
>
> memset(node_order, 0, sizeof(node_order));
> while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
>

t480s: ~/git/linux default_online_type $ git grep "nodemask_t " | grep "="
arch/x86/mm/numa.c: nodemask_t reserved_nodemask = NODE_MASK_NONE;
arch/x86/mm/numa_emulation.c: nodemask_t physnode_mask = numa_nodes_parsed;
arch/x86/mm/numa_emulation.c: nodemask_t physnode_mask = numa_nodes_parsed;
arch/x86/mm/numa_emulation.c: nodemask_t physnode_mask = numa_nodes_parsed;
drivers/acpi/numa/srat.c:static nodemask_t nodes_found_map = NODE_MASK_NONE;
kernel/irq/affinity.c: nodemask_t nodemsk = NODE_MASK_NONE;
kernel/sched/fair.c: nodemask_t max_group = NODE_MASK_NONE;
mm/memory_hotplug.c: nodemask_t nmask = node_states[N_MEMORY];
mm/mempolicy.c: nodemask_t mems = cpuset_mems_allowed(current);
mm/mempolicy.c: nodemask_t nodes = NODE_MASK_NONE;
mm/oom_kill.c: const nodemask_t *mask = oc->nodemask;
mm/page_alloc.c:nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
mm/page_alloc.c: nodemask_t saved_node_state = node_states[N_MEMORY];

Should this be NODE_MASK_NONE?

--
Thanks,

David / dhildenb

2020-03-27 13:22:47

by Wei Yang

[permalink] [raw]
Subject: Re: [PATCH 1/2] mm/page_alloc.c: leverage compiler to zero out used_mask

On Fri, Mar 27, 2020 at 10:32:45AM +0100, David Hildenbrand wrote:
>On 26.03.20 23:24, Wei Yang wrote:
>> Since we always clear used_mask before getting node order, we can
>> leverage compiler to do this instead of at run time.
>>
>> Signed-off-by: Wei Yang <[email protected]>
>> ---
>> mm/page_alloc.c | 3 +--
>> 1 file changed, 1 insertion(+), 2 deletions(-)
>>
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 0e823bca3f2f..2144b6ceb119 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -5587,14 +5587,13 @@ static void build_zonelists(pg_data_t *pgdat)
>> {
>> static int node_order[MAX_NUMNODES];
>> int node, load, nr_nodes = 0;
>> - nodemask_t used_mask;
>> + nodemask_t used_mask = {.bits = {0}};
>> int local_node, prev_node;
>>
>> /* NUMA-aware ordering of nodes */
>> local_node = pgdat->node_id;
>> load = nr_online_nodes;
>> prev_node = local_node;
>> - nodes_clear(used_mask);
>>
>> memset(node_order, 0, sizeof(node_order));
>> while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
>>
>
>t480s: ~/git/linux default_online_type $ git grep "nodemask_t " | grep "="
>arch/x86/mm/numa.c: nodemask_t reserved_nodemask = NODE_MASK_NONE;
>arch/x86/mm/numa_emulation.c: nodemask_t physnode_mask = numa_nodes_parsed;
>arch/x86/mm/numa_emulation.c: nodemask_t physnode_mask = numa_nodes_parsed;
>arch/x86/mm/numa_emulation.c: nodemask_t physnode_mask = numa_nodes_parsed;
>drivers/acpi/numa/srat.c:static nodemask_t nodes_found_map = NODE_MASK_NONE;
>kernel/irq/affinity.c: nodemask_t nodemsk = NODE_MASK_NONE;
>kernel/sched/fair.c: nodemask_t max_group = NODE_MASK_NONE;
>mm/memory_hotplug.c: nodemask_t nmask = node_states[N_MEMORY];
>mm/mempolicy.c: nodemask_t mems = cpuset_mems_allowed(current);
>mm/mempolicy.c: nodemask_t nodes = NODE_MASK_NONE;
>mm/oom_kill.c: const nodemask_t *mask = oc->nodemask;
>mm/page_alloc.c:nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
>mm/page_alloc.c: nodemask_t saved_node_state = node_states[N_MEMORY];
>
>Should this be NODE_MASK_NONE?

Thanks, this is gcc extension. Learn something new.

I would update this in v2.

>
>--
>Thanks,
>
>David / dhildenb

--
Wei Yang
Help you, Help me

2020-03-27 13:28:25

by Wei Yang

[permalink] [raw]
Subject: Re: [PATCH 1/2] mm/page_alloc.c: leverage compiler to zero out used_mask

On Thu, Mar 26, 2020 at 07:36:04PM -0300, Jason Gunthorpe wrote:
>On Thu, Mar 26, 2020 at 10:24:44PM +0000, Wei Yang wrote:
>> Since we always clear used_mask before getting node order, we can
>> leverage compiler to do this instead of at run time.
>>
>> Signed-off-by: Wei Yang <[email protected]>
>> mm/page_alloc.c | 3 +--
>> 1 file changed, 1 insertion(+), 2 deletions(-)
>>
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 0e823bca3f2f..2144b6ceb119 100644
>> +++ b/mm/page_alloc.c
>> @@ -5587,14 +5587,13 @@ static void build_zonelists(pg_data_t *pgdat)
>> {
>> static int node_order[MAX_NUMNODES];
>> int node, load, nr_nodes = 0;
>> - nodemask_t used_mask;
>> + nodemask_t used_mask = {.bits = {0}};
>
>If this style is to be done it should just be '= {}';
>
>This case demonstrates why the popular '= {0}' idiom is not such a
>good idea, as it only works if the first member is an integral type.
>

Thanks for your comment. I think David found a better solution.

>Jason

--
Wei Yang
Help you, Help me