[IA64] - SGI SN hwperf enhancements -

Add a new exported function for determining the nearest node
with CPUs for I/O nodes and fix a bug where the hwperf dynamic
misc device was being registered before misc_init(). 

Signed-off-by: Mark Goodwin <markgw@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

2 files changed, 256 insertions, 20 deletions

diff --git a/arch/ia64/sn/kernel/sn2/sn_hwperf.c b/arch/ia64/sn/kernel/sn2/sn_hwperf.c
index 58e48fdf5b47..0261452d08df 100644
--- a/arch/ia64/sn/kernel/sn2/sn_hwperf.c
+++ b/arch/ia64/sn/kernel/sn2/sn_hwperf.c
@@ -59,7 +59,7 @@ static int sn_hwperf_enum_objects(int *nobj, struct sn_hwperf_object_info **ret)
         struct sn_hwperf_object_info *objbuf = NULL;
 
         if ((e = sn_hwperf_init()) < 0) {
-                printk("sn_hwperf_init failed: err %d\n", e);
+                printk(KERN_ERR "sn_hwperf_init failed: err %d\n", e);
                 goto out;
         }
 
@@ -111,7 +111,7 @@ static int sn_hwperf_geoid_to_cnode(char *location)
         if (sn_hwperf_location_to_bpos(location, &rack, &bay, &slot, &slab))
                 return -1;
 
-        for (cnode = 0; cnode < numionodes; cnode++) {
+        for_each_node(cnode) {
                 geoid = cnodeid_get_geoid(cnode);
                 module_id = geo_module(geoid);
                 this_rack = MODULE_GET_RACK(module_id);
@@ -124,11 +124,13 @@ static int sn_hwperf_geoid_to_cnode(char *location)
                 }
         }
 
-        return cnode < numionodes ? cnode : -1;
+        return node_possible(cnode) ? cnode : -1;
 }
 
 static int sn_hwperf_obj_to_cnode(struct sn_hwperf_object_info * obj)
 {
+        if (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj))
+                BUG();
         if (!obj->sn_hwp_this_part)
                 return -1;
         return sn_hwperf_geoid_to_cnode(obj->location);
@@ -192,6 +194,181 @@ static void print_pci_topology(struct seq_file *s)
         }
 }
 
+static inline int sn_hwperf_has_cpus(cnodeid_t node)
+{
+        return node_online(node) && nr_cpus_node(node);
+}
+
+static inline int sn_hwperf_has_mem(cnodeid_t node)
+{
+        return node_online(node) && NODE_DATA(node)->node_present_pages;
+}
+
+static struct sn_hwperf_object_info *
+sn_hwperf_findobj_id(struct sn_hwperf_object_info *objbuf,
+        int nobj, int id)
+{
+        int i;
+        struct sn_hwperf_object_info *p = objbuf;
+
+        for (i=0; i < nobj; i++, p++) {
+                if (p->id == id)
+                        return p;
+        }
+
+        return NULL;
+
+}
+
+static int sn_hwperf_get_nearest_node_objdata(struct sn_hwperf_object_info *objbuf,
+        int nobj, cnodeid_t node, cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
+{
+        int e;
+        struct sn_hwperf_object_info *nodeobj = NULL;
+        struct sn_hwperf_object_info *op;
+        struct sn_hwperf_object_info *dest;
+        struct sn_hwperf_object_info *router;
+        struct sn_hwperf_port_info ptdata[16];
+        int sz, i, j;
+        cnodeid_t c;
+        int found_mem = 0;
+        int found_cpu = 0;
+
+        if (!node_possible(node))
+                return -EINVAL;
+
+        if (sn_hwperf_has_cpus(node)) {
+                if (near_cpu_node)
+                        *near_cpu_node = node;
+                found_cpu++;
+        }
+
+        if (sn_hwperf_has_mem(node)) {
+                if (near_mem_node)
+                        *near_mem_node = node;
+                found_mem++;
+        }
+
+        if (found_cpu && found_mem)
+                return 0; /* trivially successful */
+
+        /* find the argument node object */
+        for (i=0, op=objbuf; i < nobj; i++, op++) {
+                if (!SN_HWPERF_IS_NODE(op) && !SN_HWPERF_IS_IONODE(op))
+                        continue;
+                if (node == sn_hwperf_obj_to_cnode(op)) {
+                        nodeobj = op;
+                        break;
+                }
+        }
+        if (!nodeobj) {
+                e = -ENOENT;
+                goto err;
+        }
+
+        /* get it's interconnect topology */
+        sz = op->ports * sizeof(struct sn_hwperf_port_info);
+        if (sz > sizeof(ptdata))
+                BUG();
+        e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
+                              SN_HWPERF_ENUM_PORTS, nodeobj->id, sz,
+                              (u64)&ptdata, 0, 0, NULL);
+        if (e != SN_HWPERF_OP_OK) {
+                e = -EINVAL;
+                goto err;
+        }
+
+        /* find nearest node with cpus and nearest memory */
+        for (router=NULL, j=0; j < op->ports; j++) {
+                dest = sn_hwperf_findobj_id(objbuf, nobj, ptdata[j].conn_id);
+                if (!dest || SN_HWPERF_FOREIGN(dest) ||
+                    !SN_HWPERF_IS_NODE(dest) || SN_HWPERF_IS_IONODE(dest)) {
+                        continue;
+                }
+                c = sn_hwperf_obj_to_cnode(dest);
+                if (!found_cpu && sn_hwperf_has_cpus(c)) {
+                        if (near_cpu_node)
+                                *near_cpu_node = c;
+                        found_cpu++;
+                }
+                if (!found_mem && sn_hwperf_has_mem(c)) {
+                        if (near_mem_node)
+                                *near_mem_node = c;
+                        found_mem++;
+                }
+                if (SN_HWPERF_IS_ROUTER(dest))
+                        router = dest;
+        }
+
+        if (router && (!found_cpu || !found_mem)) {
+                /* search for a node connected to the same router */
+                sz = router->ports * sizeof(struct sn_hwperf_port_info);
+                if (sz > sizeof(ptdata))
+                        BUG();
+                e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
+                                      SN_HWPERF_ENUM_PORTS, router->id, sz,
+                                      (u64)&ptdata, 0, 0, NULL);
+                if (e != SN_HWPERF_OP_OK) {
+                        e = -EINVAL;
+                        goto err;
+                }
+                for (j=0; j < router->ports; j++) {
+                        dest = sn_hwperf_findobj_id(objbuf, nobj,
+                                ptdata[j].conn_id);
+                        if (!dest || dest->id == node ||
+                            SN_HWPERF_FOREIGN(dest) ||
+                            !SN_HWPERF_IS_NODE(dest) ||
+                            SN_HWPERF_IS_IONODE(dest)) {
+                                continue;
+                        }
+                        c = sn_hwperf_obj_to_cnode(dest);
+                        if (!found_cpu && sn_hwperf_has_cpus(c)) {
+                                if (near_cpu_node)
+                                        *near_cpu_node = c;
+                                found_cpu++;
+                        }
+                        if (!found_mem && sn_hwperf_has_mem(c)) {
+                                if (near_mem_node)
+                                        *near_mem_node = c;
+                                found_mem++;
+                        }
+                        if (found_cpu && found_mem)
+                                break;
+                }
+        }
+
+        if (!found_cpu || !found_mem) {
+                /* resort to _any_ node with CPUs and memory */
+                for (i=0, op=objbuf; i < nobj; i++, op++) {
+                        if (SN_HWPERF_FOREIGN(op) ||
+                            SN_HWPERF_IS_IONODE(op) ||
+                            !SN_HWPERF_IS_NODE(op)) {
+                                continue;
+                        }
+                        c = sn_hwperf_obj_to_cnode(op);
+                        if (!found_cpu && sn_hwperf_has_cpus(c)) {
+                                if (near_cpu_node)
+                                        *near_cpu_node = c;
+                                found_cpu++;
+                        }
+                        if (!found_mem && sn_hwperf_has_mem(c)) {
+                                if (near_mem_node)
+                                        *near_mem_node = c;
+                                found_mem++;
+                        }
+                        if (found_cpu && found_mem)
+                                break;
+                }
+        }
+
+        if (!found_cpu || !found_mem)
+                e = -ENODATA;
+
+err:
+        return e;
+}
+
+
 static int sn_topology_show(struct seq_file *s, void *d)
 {
         int sz;
@@ -265,11 +442,24 @@ static int sn_topology_show(struct seq_file *s, void *d)
         if (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj))
                 seq_putc(s, '\n');
         else {
+                cnodeid_t near_mem = -1;
+                cnodeid_t near_cpu = -1;
+
                 seq_printf(s, ", nasid 0x%x", cnodeid_to_nasid(ordinal));
-                for (i=0; i < numionodes; i++) {
-                        seq_printf(s, i ? ":%d" : ", dist %d",
-                                node_distance(ordinal, i));
+
+                if (sn_hwperf_get_nearest_node_objdata(objs, sn_hwperf_obj_cnt,
+                        ordinal, &near_mem, &near_cpu) == 0) {
+                        seq_printf(s, ", near_mem_nodeid %d, near_cpu_nodeid %d",
+                                near_mem, near_cpu);
+                }
+
+                if (!SN_HWPERF_IS_IONODE(obj)) {
+                        for_each_online_node(i) {
+                                seq_printf(s, i ? ":%d" : ", dist %d",
+                                        node_distance(ordinal, i));
+                        }
                 }
+
                 seq_putc(s, '\n');
 
                 /*
@@ -554,6 +744,8 @@ sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
                 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
                         memset(p, 0, a.sz);
                         for (i = 0; i < nobj; i++) {
+                                if (!SN_HWPERF_IS_NODE(objs + i))
+                                        continue;
                                 node = sn_hwperf_obj_to_cnode(objs + i);
                                 for_each_online_cpu(j) {
                                         if (node != cpu_to_node(j))
@@ -580,7 +772,7 @@ sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
 
         case SN_HWPERF_GET_NODE_NASID:
                 if (a.sz != sizeof(u64) ||
-                   (node = a.arg) < 0 || node >= numionodes) {
+                   (node = a.arg) < 0 || !node_possible(node)) {
                         r = -EINVAL;
                         goto error;
                 }
@@ -609,6 +801,14 @@ sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
                                 vfree(objs);
                                 goto error;
                         }
+
+                        if (!SN_HWPERF_IS_NODE(objs + i) &&
+                            !SN_HWPERF_IS_IONODE(objs + i)) {
+                                r = -ENOENT;
+                                vfree(objs);
+                                goto error;
+                        }
+
                         *(u64 *)p = (u64)sn_hwperf_obj_to_cnode(objs + i);
                         vfree(objs);
                 }
@@ -674,6 +874,7 @@ static int sn_hwperf_init(void)
 
         /* single threaded, once-only initialization */
         down(&sn_hwperf_init_mutex);
+
         if (sn_hwperf_salheap) {
                 up(&sn_hwperf_init_mutex);
                 return e;
@@ -724,19 +925,6 @@ out:
                 sn_hwperf_salheap = NULL;
                 sn_hwperf_obj_cnt = 0;
         }
-
-        if (!e) {
-                /*
-                 * Register a dynamic misc device for ioctl. Platforms
-                 * supporting hotplug will create /dev/sn_hwperf, else
-                 * user can to look up the minor number in /proc/misc.
-                 */
-                if ((e = misc_register(&sn_hwperf_dev)) != 0) {
-                        printk(KERN_ERR "sn_hwperf_init: misc register "
-                               "for \"sn_hwperf\" failed, err %d\n", e);
-                }
-        }
-
         up(&sn_hwperf_init_mutex);
         return e;
 }
@@ -764,3 +952,41 @@ int sn_topology_release(struct inode *inode, struct file *file)
         vfree(seq->private);
         return seq_release(inode, file);
 }
+
+int sn_hwperf_get_nearest_node(cnodeid_t node,
+        cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
+{
+        int e;
+        int nobj;
+        struct sn_hwperf_object_info *objbuf;
+
+        if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
+                e = sn_hwperf_get_nearest_node_objdata(objbuf, nobj,
+                        node, near_mem_node, near_cpu_node);
+                vfree(objbuf);
+        }
+
+        return e;
+}
+
+static int __devinit sn_hwperf_misc_register_init(void)
+{
+        int e;
+
+        sn_hwperf_init();
+
+        /*
+         * Register a dynamic misc device for hwperf ioctls. Platforms
+         * supporting hotplug will create /dev/sn_hwperf, else user
+         * can to look up the minor number in /proc/misc.
+         */
+        if ((e = misc_register(&sn_hwperf_dev)) != 0) {
+                printk(KERN_ERR "sn_hwperf_misc_register_init: failed to "
+                "register misc device for \"%s\"\n", sn_hwperf_dev.name);
+        }
+
+        return e;
+}
+
+device_initcall(sn_hwperf_misc_register_init); /* after misc_init() */
+EXPORT_SYMBOL(sn_hwperf_get_nearest_node);
diff --git a/include/asm-ia64/sn/sn2/sn_hwperf.h b/include/asm-ia64/sn/sn2/sn_hwperf.h
index df75f4c4aec3..291ef3d69da2 100644
--- a/include/asm-ia64/sn/sn2/sn_hwperf.h
+++ b/include/asm-ia64/sn/sn2/sn_hwperf.h
@@ -43,6 +43,7 @@ struct sn_hwperf_object_info {
 
 /* macros for object classification */
 #define SN_HWPERF_IS_NODE(x)            ((x) && strstr((x)->name, "SHub"))
+#define SN_HWPERF_IS_NODE_SHUB2(x)      ((x) && strstr((x)->name, "SHub 2."))
 #define SN_HWPERF_IS_IONODE(x)          ((x) && strstr((x)->name, "TIO"))
 #define SN_HWPERF_IS_ROUTER(x)          ((x) && strstr((x)->name, "Router"))
 #define SN_HWPERF_IS_NL3ROUTER(x)       ((x) && strstr((x)->name, "NL3Router"))
@@ -214,6 +215,15 @@ struct sn_hwperf_ioctl_args {
  */
 #define SN_HWPERF_GET_NODE_NASID        (102|SN_HWPERF_OP_MEM_COPYOUT)
 
+/*
+ * Given a node id, determine the id of the nearest node with CPUs
+ * and the id of the nearest node that has memory. The argument
+ * node would normally be a "headless" node, e.g. an "IO node".
+ * Return 0 on success.
+ */
+extern int sn_hwperf_get_nearest_node(cnodeid_t node,
+        cnodeid_t *near_mem, cnodeid_t *near_cpu);
+
 /* return codes */
 #define SN_HWPERF_OP_OK                 0
 #define SN_HWPERF_OP_NOMEM              1