[PATCH 0/4] sched fixes

[PATCH 0/4] sched fixes

[Yury Norov]

Fixes for recently introduced sched_numa_find_nth_cpu(), and minor
improvements in sched/fair.

Yury Norov (4):
  numa: generalize numa_map_to_online_node()
  sched/fair: fix opencoded numa_nearest_node()
  sched: fix sched_numa_find_nth_cpu() in CPU-less case
  sched: fix sched_numa_find_nth_cpu() in non-NUMA case

 include/linux/numa.h     |  7 +++++--
 include/linux/topology.h |  2 +-
 kernel/sched/fair.c      | 14 +-------------
 kernel/sched/topology.c  |  6 +++++-
 mm/mempolicy.c           | 18 +++++++++++-------
 5 files changed, 23 insertions(+), 24 deletions(-)

-- 
2.39.2

[PATCH 1/4] numa: generalize numa_map_to_online_node()

[Yury Norov]

The function is in fact searches for the nearest node for a given one,
based on a N_ONLINE state. This is a common pattern to search for a
nearest node.

This patch converts numa_map_to_online_node(node) to
numa_nearest_node(node, state), so that others won't need opencode the
logic. The following patches apply it where applicable.

Signed-off-by: Yury Norov <yury.norov@gmail.com>
---
 include/linux/numa.h |  7 +++++--
 mm/mempolicy.c       | 18 +++++++++++-------
 2 files changed, 16 insertions(+), 9 deletions(-)

diff --git a/include/linux/numa.h b/include/linux/numa.h
index 59df211d051f..fb30a42f0700 100644
--- a/include/linux/numa.h
+++ b/include/linux/numa.h
@@ -25,7 +25,7 @@
 #include <asm/sparsemem.h>
 
 /* Generic implementation available */
-int numa_map_to_online_node(int node);
+int numa_nearest_node(int node, unsigned int state);
 
 #ifndef memory_add_physaddr_to_nid
 static inline int memory_add_physaddr_to_nid(u64 start)
@@ -44,10 +44,11 @@ static inline int phys_to_target_node(u64 start)
 }
 #endif
 #else /* !CONFIG_NUMA */
-static inline int numa_map_to_online_node(int node)
+static inline int numa_nearest_node(int node, unsigned int state)
 {
 	return NUMA_NO_NODE;
 }
+
 static inline int memory_add_physaddr_to_nid(u64 start)
 {
 	return 0;
@@ -58,6 +59,8 @@ static inline int phys_to_target_node(u64 start)
 }
 #endif
 
+#define numa_map_to_online_node(node) numa_nearest_node(node, N_ONLINE)
+
 #ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
 extern const struct attribute_group arch_node_dev_group;
 #endif
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index c53f8beeb507..db74dd957d2c 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -131,22 +131,26 @@ static struct mempolicy default_policy = {
 static struct mempolicy preferred_node_policy[MAX_NUMNODES];
 
 /**
- * numa_map_to_online_node - Find closest online node
+ * numa_nearest_node - Find nearest node by state
  * @node: Node id to start the search
+ * @state: State to filter the search
  *
- * Lookup the next closest node by distance if @nid is not online.
+ * Lookup the closest node by distance if @nid is not in state.
  *
- * Return: this @node if it is online, otherwise the closest node by distance
+ * Return: this @node if it is in state, otherwise the closest node by distance
  */
-int numa_map_to_online_node(int node)
+int numa_nearest_node(int node, unsigned int state)
 {
 	int min_dist = INT_MAX, dist, n, min_node;
 
-	if (node == NUMA_NO_NODE || node_online(node))
+	if (state >= NR_NODE_STATES)
+		return -EINVAL;
+
+	if (node == NUMA_NO_NODE || node_state(node, state))
 		return node;
 
 	min_node = node;
-	for_each_online_node(n) {
+	for_each_node_state(n, state) {
 		dist = node_distance(node, n);
 		if (dist < min_dist) {
 			min_dist = dist;
@@ -156,7 +160,7 @@ int numa_map_to_online_node(int node)
 
 	return min_node;
 }
-EXPORT_SYMBOL_GPL(numa_map_to_online_node);
+EXPORT_SYMBOL_GPL(numa_nearest_node);
 
 struct mempolicy *get_task_policy(struct task_struct *p)
 {
-- 
2.39.2

Re: [PATCH 1/4] numa: generalize numa_map_to_online_node()

[Andy Shevchenko]

On Thu, Aug 10, 2023 at 09:24:39AM -0700, Yury Norov wrote:
> The function is in fact searches for the nearest node for a given one,
> based on a N_ONLINE state. This is a common pattern to search for a
> nearest node.
> 
> This patch converts numa_map_to_online_node(node) to
> numa_nearest_node(node, state), so that others won't need opencode the
> logic.

> The following patches apply it where applicable.

Doesn't sound like should be part of the commit message.

-- 
With Best Regards,
Andy Shevchenko

[PATCH 2/4] sched/fair: fix opencoded numa_nearest_node()

[Yury Norov]

task_numa_placement() searches for a nearest node by calling
for_each_node_state(). Now that we have numa_nearest_node(), switch to
using it.

Signed-off-by: Yury Norov <yury.norov@gmail.com>
---
 kernel/sched/fair.c | 14 +-------------
 1 file changed, 1 insertion(+), 13 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index b3e25be58e2b..e7b7cf87937b 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -2645,19 +2645,7 @@ static void task_numa_placement(struct task_struct *p)
 	}
 
 	/* Cannot migrate task to CPU-less node */
-	if (max_nid != NUMA_NO_NODE && !node_state(max_nid, N_CPU)) {
-		int near_nid = max_nid;
-		int distance, near_distance = INT_MAX;
-
-		for_each_node_state(nid, N_CPU) {
-			distance = node_distance(max_nid, nid);
-			if (distance < near_distance) {
-				near_nid = nid;
-				near_distance = distance;
-			}
-		}
-		max_nid = near_nid;
-	}
+	max_nid = numa_nearest_node(max_nid, N_CPU);
 
 	if (ng) {
 		numa_group_count_active_nodes(ng);
-- 
2.39.2

Re: [PATCH 2/4] sched/fair: fix opencoded numa_nearest_node()

[Andy Shevchenko]

On Thu, Aug 10, 2023 at 09:24:40AM -0700, Yury Norov wrote:
> task_numa_placement() searches for a nearest node by calling
> for_each_node_state(). Now that we have numa_nearest_node(), switch to
> using it.

For better looking you may rewrap the above as

---8<---
task_numa_placement() searches for a nearest node by calling
for_each_node_state(). Now that we have numa_nearest_node(),
switch to using
---8<---

-- 
With Best Regards,
Andy Shevchenko

[PATCH 3/4] sched: fix sched_numa_find_nth_cpu() in CPU-less case

[Yury Norov]

When the node provided by user is CPU-less, corresponding record in
sched_domains_numa_masks is not set. Trying to dereference it in the
following code leads to kernel crash.

To avoid it, start searching from the nearest node with CPUs.

Fixes: cd7f55359c90 ("sched: add sched_numa_find_nth_cpu()")
Reported-by: Yicong Yang <yangyicong@hisilicon.com>
Closes: https://lore.kernel.org/lkml/CAAH8bW8C5humYnfpW3y5ypwx0E-09A3QxFE1JFzR66v+mO4XfA@mail.gmail.com/T/
Reported-by: Guenter Roeck <linux@roeck-us.net>
Closes: https://lore.kernel.org/lkml/ZMHSNQfv39HN068m@yury-ThinkPad/T/#mf6431cb0b7f6f05193c41adeee444bc95bf2b1c4
Signed-off-by: Yury Norov <yury.norov@gmail.com>
---

This has been discovered and fixed by Yicong Yang:

https://lore.kernel.org/lkml/CAAH8bW8C5humYnfpW3y5ypwx0E-09A3QxFE1JFzR66v+mO4XfA@mail.gmail.com/T/

When discovering Guenter's failure report for sparc64, I found it's due to
the same problem. And while fixing, I found an opportunity to generalize
nearest NUMA node search and avoid code duplication.

Yicong, if you like this approach, please feel free to add your co-developed-by
or any appropriate tags.

 kernel/sched/topology.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index d3a3b2646ec4..66b387172b6f 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -2113,10 +2113,14 @@ static int hop_cmp(const void *a, const void *b)
  */
 int sched_numa_find_nth_cpu(const struct cpumask *cpus, int cpu, int node)
 {
-	struct __cmp_key k = { .cpus = cpus, .node = node, .cpu = cpu };
+	struct __cmp_key k = { .cpus = cpus, .cpu = cpu };
 	struct cpumask ***hop_masks;
 	int hop, ret = nr_cpu_ids;
 
+	/* CPU-less node entries are uninitialized in sched_domains_numa_masks */
+	node = numa_nearest_node(node, N_CPU);
+	k.node = node;
+
 	rcu_read_lock();
 
 	k.masks = rcu_dereference(sched_domains_numa_masks);
-- 
2.39.2

Re: [PATCH 3/4] sched: fix sched_numa_find_nth_cpu() in CPU-less case

[Yicong Yang]

Hi Yury,

On 2023/8/11 0:24, Yury Norov wrote:
> When the node provided by user is CPU-less, corresponding record in
> sched_domains_numa_masks is not set. Trying to dereference it in the
> following code leads to kernel crash.
> 
> To avoid it, start searching from the nearest node with CPUs.
> 
> Fixes: cd7f55359c90 ("sched: add sched_numa_find_nth_cpu()")
> Reported-by: Yicong Yang <yangyicong@hisilicon.com>
> Closes: https://lore.kernel.org/lkml/CAAH8bW8C5humYnfpW3y5ypwx0E-09A3QxFE1JFzR66v+mO4XfA@mail.gmail.com/T/
> Reported-by: Guenter Roeck <linux@roeck-us.net>
> Closes: https://lore.kernel.org/lkml/ZMHSNQfv39HN068m@yury-ThinkPad/T/#mf6431cb0b7f6f05193c41adeee444bc95bf2b1c4
> Signed-off-by: Yury Norov <yury.norov@gmail.com>
> ---
> 
> This has been discovered and fixed by Yicong Yang:
> 
> https://lore.kernel.org/lkml/CAAH8bW8C5humYnfpW3y5ypwx0E-09A3QxFE1JFzR66v+mO4XfA@mail.gmail.com/T/
> 
> When discovering Guenter's failure report for sparc64, I found it's due to
> the same problem. And while fixing, I found an opportunity to generalize
> nearest NUMA node search and avoid code duplication.
> 
> Yicong, if you like this approach, please feel free to add your co-developed-by
> or any appropriate tags.
> 

Looks fine to me. One nit below.

Reviewed-by: Yicong Yang <yangyicong@hisilicon.com>

>  kernel/sched/topology.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
> index d3a3b2646ec4..66b387172b6f 100644
> --- a/kernel/sched/topology.c
> +++ b/kernel/sched/topology.c
> @@ -2113,10 +2113,14 @@ static int hop_cmp(const void *a, const void *b)
>   */
>  int sched_numa_find_nth_cpu(const struct cpumask *cpus, int cpu, int node)
>  {
> -	struct __cmp_key k = { .cpus = cpus, .node = node, .cpu = cpu };
> +	struct __cmp_key k = { .cpus = cpus, .cpu = cpu };
>  	struct cpumask ***hop_masks;
>  	int hop, ret = nr_cpu_ids;
>  
> +	/* CPU-less node entries are uninitialized in sched_domains_numa_masks */
> +	node = numa_nearest_node(node, N_CPU);
> +	k.node = node;
> +

We may also have problem if node == NUMA_NO_NODE, is it better to mention this
in the function comment or check it before we going on? Currently this function
is only used in cpumask_local_spread() and the caller has already checked it, but
considering this is an export function so somebody may use it directly.

I wondering whether we should put this block within the protection of rcu_read_lock()
for some issues like hotplug or not. Is it possible if @node become CPU-less subsequently?

>  	rcu_read_lock();
>  
>  	k.masks = rcu_dereference(sched_domains_numa_masks);
> 

[PATCH 4/4] sched: fix sched_numa_find_nth_cpu() in non-NUMA case

[Yury Norov]

When CONFIG_NUMA is enabled, sched_numa_find_nth_cpu() searches for a
CPU in sched_domains_numa_masks. The masks includes only online CPUs,
so effectively offline CPUs are skipped.

When CONFIG_NUMA is disabled, the fallback function should be consistent.

Fixes: cd7f55359c90 ("sched: add sched_numa_find_nth_cpu()")
Signed-off-by: Yury Norov <yury.norov@gmail.com>
---
 include/linux/topology.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/linux/topology.h b/include/linux/topology.h
index fea32377f7c7..52f5850730b3 100644
--- a/include/linux/topology.h
+++ b/include/linux/topology.h
@@ -251,7 +251,7 @@ extern const struct cpumask *sched_numa_hop_mask(unsigned int node, unsigned int
 #else
 static __always_inline int sched_numa_find_nth_cpu(const struct cpumask *cpus, int cpu, int node)
 {
-	return cpumask_nth(cpu, cpus);
+	return cpumask_nth_and(cpu, cpus, cpu_online_mask);
 }
 
 static inline const struct cpumask *
-- 
2.39.2