sparc64: Setup sysfs to mark LDOM sockets, cores and threads correctly

commit 5f4826a362405748bbf73957027b77993e61e1af
Author: chris hyser <chris.hyser@oracle.com>
Date:   Tue Apr 21 10:31:38 2015 -0400

    sparc64: Setup sysfs to mark LDOM sockets, cores and threads correctly

    The current sparc kernel has no representation for sockets though tools
    like lscpu can pull this from sysfs. This patch walks the machine
    description cache and socket hierarchy and marks sockets as well as cores
    and threads such that a representative sysfs is created by
    drivers/base/topology.c.

    Before this patch:
        $ lscpu
        Architecture:          sparc64
        CPU op-mode(s):        32-bit, 64-bit
        Byte Order:            Big Endian
        CPU(s):                1024
        On-line CPU(s) list:   0-1023
        Thread(s) per core:    8
        Core(s) per socket:    1     <--- wrong
        Socket(s):             128   <--- wrong
        NUMA node(s):          4
        NUMA node0 CPU(s):     0-255
        NUMA node1 CPU(s):     256-511
        NUMA node2 CPU(s):     512-767
        NUMA node3 CPU(s):     768-1023

        After this patch:
        $ lscpu
        Architecture:          sparc64
        CPU op-mode(s):        32-bit, 64-bit
        Byte Order:            Big Endian
        CPU(s):                1024
        On-line CPU(s) list:   0-1023
        Thread(s) per core:    8
        Core(s) per socket:    32
        Socket(s):             4
        NUMA node(s):          4
        NUMA node0 CPU(s):     0-255
        NUMA node1 CPU(s):     256-511
        NUMA node2 CPU(s):     512-767
        NUMA node3 CPU(s):     768-1023

    Most of this patch was done by Chris with updates by David.
Signed-off-by: Chris Hyser <chris.hyser@oracle.com>
Signed-off-by: David Ahern <david.ahern@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

arch/sparc/include/asm/cpudata_64.h

@@ -24,7 +24,8 @@ typedef struct {
 	unsigned int	icache_line_size;
 	unsigned int	ecache_size;
 	unsigned int	ecache_line_size;
-	int		core_id;
+	unsigned short	sock_id;
+	unsigned short	core_id;
 	int		proc_id;
 } cpuinfo_sparc;
 

arch/sparc/include/asm/topology_64.h

@@ -40,11 +40,12 @@ static inline int pcibus_to_node(struct pci_bus *pbus)
 #ifdef CONFIG_SMP
 #define topology_physical_package_id(cpu)	(cpu_data(cpu).proc_id)
 #define topology_core_id(cpu)			(cpu_data(cpu).core_id)
-#define topology_core_cpumask(cpu)		(&cpu_core_map[cpu])
+#define topology_core_cpumask(cpu)		(&cpu_core_sib_map[cpu])
 #define topology_thread_cpumask(cpu)		(&per_cpu(cpu_sibling_map, cpu))
 #endif /* CONFIG_SMP */
 
 extern cpumask_t cpu_core_map[NR_CPUS];
+extern cpumask_t cpu_core_sib_map[NR_CPUS];
 static inline const struct cpumask *cpu_coregroup_mask(int cpu)
 {
         return &cpu_core_map[cpu];

arch/sparc/kernel/mdesc.c

@@ -614,45 +614,68 @@ static void fill_in_one_cache(cpuinfo_sparc *c, struct mdesc_handle *hp, u64 mp)
 	}
 }
 
-static void mark_core_ids(struct mdesc_handle *hp, u64 mp, int core_id)
+static void find_back_node_value(struct mdesc_handle *hp, u64 node,
+				 char *srch_val,
+				 void (*func)(struct mdesc_handle *, u64, int),
+				 u64 val, int depth)
 {
-	u64 a;
-
-	mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_BACK) {
-		u64 t = mdesc_arc_target(hp, a);
-		const char *name;
-		const u64 *id;
+	u64 arc;
 
-		name = mdesc_node_name(hp, t);
-		if (!strcmp(name, "cpu")) {
-			id = mdesc_get_property(hp, t, "id", NULL);
-			if (*id < NR_CPUS)
-				cpu_data(*id).core_id = core_id;
-		} else {
-			u64 j;
+	/* Since we have an estimate of recursion depth, do a sanity check. */
+	if (depth == 0)
+		return;
 
-			mdesc_for_each_arc(j, hp, t, MDESC_ARC_TYPE_BACK) {
-				u64 n = mdesc_arc_target(hp, j);
-				const char *n_name;
+	mdesc_for_each_arc(arc, hp, node, MDESC_ARC_TYPE_BACK) {
+		u64 n = mdesc_arc_target(hp, arc);
+		const char *name = mdesc_node_name(hp, n);
 
-				n_name = mdesc_node_name(hp, n);
-				if (strcmp(n_name, "cpu"))
-					continue;
+		if (!strcmp(srch_val, name))
+			(*func)(hp, n, val);
 
-				id = mdesc_get_property(hp, n, "id", NULL);
-				if (*id < NR_CPUS)
-					cpu_data(*id).core_id = core_id;
-			}
-		}
+		find_back_node_value(hp, n, srch_val, func, val, depth-1);
 	}
 }
 
+static void __mark_core_id(struct mdesc_handle *hp, u64 node,
+			   int core_id)
+{
+	const u64 *id = mdesc_get_property(hp, node, "id", NULL);
+
+	if (*id < num_possible_cpus())
+		cpu_data(*id).core_id = core_id;
+}
+
+static void __mark_sock_id(struct mdesc_handle *hp, u64 node,
+			   int sock_id)
+{
+	const u64 *id = mdesc_get_property(hp, node, "id", NULL);
+
+	if (*id < num_possible_cpus())
+		cpu_data(*id).sock_id = sock_id;
+}
+
+static void mark_core_ids(struct mdesc_handle *hp, u64 mp,
+			  int core_id)
+{
+	find_back_node_value(hp, mp, "cpu", __mark_core_id, core_id, 10);
+}
+
+static void mark_sock_ids(struct mdesc_handle *hp, u64 mp,
+			  int sock_id)
+{
+	find_back_node_value(hp, mp, "cpu", __mark_sock_id, sock_id, 10);
+}
+
 static void set_core_ids(struct mdesc_handle *hp)
 {
 	int idx;
 	u64 mp;
 
 	idx = 1;
+
+	/* Identify unique cores by looking for cpus backpointed to by
+	 * level 1 instruction caches.
+	 */
 	mdesc_for_each_node_by_name(hp, mp, "cache") {
 		const u64 *level;
 		const char *type;
@@ -667,11 +690,72 @@ static void set_core_ids(struct mdesc_handle *hp)
 			continue;
 
 		mark_core_ids(hp, mp, idx);
+		idx++;
+	}
+}
+
+static int set_sock_ids_by_cache(struct mdesc_handle *hp, int level)
+{
+	u64 mp;
+	int idx = 1;
+	int fnd = 0;
+
+	/* Identify unique sockets by looking for cpus backpointed to by
+	 * shared level n caches.
+	 */
+	mdesc_for_each_node_by_name(hp, mp, "cache") {
+		const u64 *cur_lvl;
+
+		cur_lvl = mdesc_get_property(hp, mp, "level", NULL);
+		if (*cur_lvl != level)
+			continue;
+
+		mark_sock_ids(hp, mp, idx);
+		idx++;
+		fnd = 1;
+	}
+	return fnd;
+}
+
+static void set_sock_ids_by_socket(struct mdesc_handle *hp, u64 mp)
+{
+	int idx = 1;
 
+	mdesc_for_each_node_by_name(hp, mp, "socket") {
+		u64 a;
+
+		mdesc_for_each_arc(a, hp, mp, MDESC_ARC_TYPE_FWD) {
+			u64 t = mdesc_arc_target(hp, a);
+			const char *name;
+			const u64 *id;
+
+			name = mdesc_node_name(hp, t);
+			if (strcmp(name, "cpu"))
+				continue;
+
+			id = mdesc_get_property(hp, t, "id", NULL);
+			if (*id < num_possible_cpus())
+				cpu_data(*id).sock_id = idx;
+		}
 		idx++;
 	}
 }
 
+static void set_sock_ids(struct mdesc_handle *hp)
+{
+	u64 mp;
+
+	/* If machine description exposes sockets data use it.
+	 * Otherwise fallback to use shared L3 or L2 caches.
+	 */
+	mp = mdesc_node_by_name(hp, MDESC_NODE_NULL, "sockets");
+	if (mp != MDESC_NODE_NULL)
+		return set_sock_ids_by_socket(hp, mp);
+
+	if (!set_sock_ids_by_cache(hp, 3))
+		set_sock_ids_by_cache(hp, 2);
+}
+
 static void mark_proc_ids(struct mdesc_handle *hp, u64 mp, int proc_id)
 {
 	u64 a;
@@ -707,7 +791,6 @@ static void __set_proc_ids(struct mdesc_handle *hp, const char *exec_unit_name)
 			continue;
 
 		mark_proc_ids(hp, mp, idx);
-
 		idx++;
 	}
 }
@@ -900,6 +983,7 @@ void mdesc_fill_in_cpu_data(cpumask_t *mask)
 
 	set_core_ids(hp);
 	set_proc_ids(hp);
+	set_sock_ids(hp);
 
 	mdesc_release(hp);
 

arch/sparc/kernel/smp_64.c

@@ -60,8 +60,12 @@ DEFINE_PER_CPU(cpumask_t, cpu_sibling_map) = CPU_MASK_NONE;
 cpumask_t cpu_core_map[NR_CPUS] __read_mostly =
 	{ [0 ... NR_CPUS-1] = CPU_MASK_NONE };
 
+cpumask_t cpu_core_sib_map[NR_CPUS] __read_mostly = {
+	[0 ... NR_CPUS-1] = CPU_MASK_NONE };
+
 EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
 EXPORT_SYMBOL(cpu_core_map);
+EXPORT_SYMBOL(cpu_core_sib_map);
 
 static cpumask_t smp_commenced_mask;
 
@@ -1243,6 +1247,15 @@ void smp_fill_in_sib_core_maps(void)
 		}
 	}
 
+	for_each_present_cpu(i)  {
+		unsigned int j;
+
+		for_each_present_cpu(j)  {
+			if (cpu_data(i).sock_id == cpu_data(j).sock_id)
+				cpumask_set_cpu(j, &cpu_core_sib_map[i]);
+		}
+	}
+
 	for_each_present_cpu(i) {
 		unsigned int j;