When an associativity level change is found for one thread, the
siblings threads need to be updated as well. This is done today
for PRRN in stage_topology_update() but is missing for VPHN in
update_cpu_associativity_changes_mask().
All threads should be updated to move to the new node. Without this
patch, a single thread may be flagged for a topology change, leaving it
in a different node from its siblings, which is incorrect. This causes
problems for the scheduler where overlapping scheduler groups are created
and a loop is formed in those groups.
Reported-by: Jan Stancek <jstan...@redhat.com>
Signed-off-by: Robert Jennings <r...@linux.vnet.ibm.com>
Cc: <sta...@vger.kernel.org>
---
Looking to get this fix in 3.11.
This is requested for the 3.10 stable tree as well, it fixes a
bug that presents as a scheduler issue when VPHN is active.
VPHN was enabled in the v3.9 window in commit b7abef0.
---
arch/powerpc/mm/numa.c | 59
+++++++++++++++++++++++++++++++++++++-------------
1 file changed, 44 insertions(+), 15 deletions(-)
diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
index 0839721..5850798 100644
--- a/arch/powerpc/mm/numa.c
+++ b/arch/powerpc/mm/numa.c
@@ -27,6 +27,7 @@
#include <linux/seq_file.h>
#include <linux/uaccess.h>
#include <linux/slab.h>
+#include <asm/cputhreads.h>
#include <asm/sparsemem.h>
#include <asm/prom.h>
#include <asm/smp.h>
@@ -1318,7 +1319,8 @@ static int update_cpu_associativity_changes_mask(void)
}
}
if (changed) {
- cpumask_set_cpu(cpu, changes);
+ cpumask_or(changes, changes, cpu_sibling_mask(cpu));
+ cpu = cpu_last_thread_sibling(cpu);
}
}
@@ -1426,7 +1428,7 @@ static int update_cpu_topology(void *data)
if (!data)
return -EINVAL;
- cpu = get_cpu();
+ cpu = smp_processor_id();
for (update = data; update; update = update->next) {
if (cpu != update->cpu)
@@ -1446,12 +1448,12 @@ static int update_cpu_topology(void *data)
*/
int arch_update_cpu_topology(void)
{
- unsigned int cpu, changed = 0;
+ unsigned int cpu, sibling, changed = 0;
struct topology_update_data *updates, *ud;
unsigned int associativity[VPHN_ASSOC_BUFSIZE] = {0};
cpumask_t updated_cpus;
struct device *dev;
- int weight, i = 0;
+ int weight, new_nid, i = 0;
weight = cpumask_weight(&cpu_associativity_changes_mask);
if (!weight)
@@ -1464,19 +1466,46 @@ int arch_update_cpu_topology(void)
cpumask_clear(&updated_cpus);
for_each_cpu(cpu, &cpu_associativity_changes_mask) {
- ud = &updates[i++];
- ud->cpu = cpu;
- vphn_get_associativity(cpu, associativity);
- ud->new_nid = associativity_to_nid(associativity);
-
- if (ud->new_nid < 0 || !node_online(ud->new_nid))
- ud->new_nid = first_online_node;
+ /*
+ * If siblings aren't flagged for changes, updates list
+ * will be too short. Skip on this update and set for next
+ * update.
+ */
+ if (!cpumask_subset(cpu_sibling_mask(cpu),
+ &cpu_associativity_changes_mask)) {
+ pr_info("Sibling bits not set for associativity "
+ "change, cpu%d\n", cpu);
+ cpumask_or(&cpu_associativity_changes_mask,
+ &cpu_associativity_changes_mask,
+ cpu_sibling_mask(cpu));
+ cpu = cpu_last_thread_sibling(cpu);
+ continue;
+ }
- ud->old_nid = numa_cpu_lookup_table[cpu];
- cpumask_set_cpu(cpu, &updated_cpus);
+ /* Use associativity from first thread for all siblings */
+ vphn_get_associativity(cpu, associativity);
+ new_nid = associativity_to_nid(associativity);
+ if (new_nid < 0 || !node_online(new_nid))
+ new_nid = first_online_node;
+
+ if (new_nid == numa_cpu_lookup_table[cpu]) {
+ cpumask_andnot(&cpu_associativity_changes_mask,
+ &cpu_associativity_changes_mask,
+ cpu_sibling_mask(cpu));
+ cpu = cpu_last_thread_sibling(cpu);
+ continue;
+ }
- if (i < weight)
- ud->next = &updates[i];
+ for_each_cpu(sibling, cpu_sibling_mask(cpu)) {
+ ud = &updates[i++];
+ ud->cpu = sibling;
+ ud->new_nid = new_nid;
+ ud->old_nid = numa_cpu_lookup_table[sibling];
+ cpumask_set_cpu(sibling, &updated_cpus);
+ if (i < weight)
+ ud->next = &updates[i];
+ }
+ cpu = cpu_last_thread_sibling(cpu);
}
stop_machine(update_cpu_topology, &updates[0], &updated_cpus);
--
1.8.1.2
_______________________________________________
Linuxppc-dev mailing list
Linuxppc-dev@lists.ozlabs.org
https://lists.ozlabs.org/listinfo/linuxppc-dev
