Re: [PATCH 00/11] Support for grouping cores

[Srikar Dronamraju <srikar@linux.vnet.ibm.com>]

* Srikar Dronamraju <srikar@linux.vnet.ibm.com> [2020-07-14 10:06:13]:

> 
> On Power 9 (with device-tree enablement to show coregroups).
> -----------------------------------------------------------

With help of Gautham, I tried to kexec into a newer kernel with a modified
dtb. However when passing with the dtb option, kexec would get stuck.

Hence alternatively, I hardcoded some assumptions to test the same.
These hunks apply on top of all the patches.

diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
index 8ec7ff05ae47..762aa573c313 100644
--- a/arch/powerpc/mm/numa.c
+++ b/arch/powerpc/mm/numa.c
@@ -56,6 +56,12 @@ static int n_mem_addr_cells, n_mem_size_cells;
 static int form1_affinity;
 
 #define MAX_DISTANCE_REF_POINTS 4
+/*
+ *  HAck2: On Power9, there are 12 SMT8 cores per chip.
+ *  Hard code this for now.
+ */
+#define CORES_PER_CHIP 12
+#define CORES_PER_COREGROUP (CORES_PER_CHIP / 2)
 static int distance_ref_points_depth;
 static const __be32 *distance_ref_points;
 static int distance_lookup_table[MAX_NUMNODES][MAX_DISTANCE_REF_POINTS];
@@ -326,6 +332,13 @@ static int __init find_min_common_depth(void)
 
 	if (form1_affinity) {
 		depth = of_read_number(distance_ref_points, 1);
+		/*
+		 * Hack: Hack: Hack:
+		 * For now only on Phyp machines.
+		 */
+		if (!firmware_has_feature(FW_FEATURE_OPAL))
+			depth--;
 	} else {
 		if (distance_ref_points_depth < 2) {
 			printk(KERN_WARNING "NUMA: "
@@ -1247,8 +1260,8 @@ int cpu_to_coregroup_id(int cpu)
 		goto out;
 
 	index = of_read_number(associativity, 1);
-	if ((index > min_common_depth + 1) && coregroup_enabled)
-		return of_read_number(&associativity[index - 1], 1);
+	if ((index > min_common_depth + 1) && coregroup_enabled && has_big_cores)
+		return of_read_number(&associativity[index], 1) / CORES_PER_COREGROUP;
 
 out:
 	return cpu_to_core_id(cpu);
-- 
Thanks and Regards
Srikar Dronamraju