module: Optimize __module_address() using a latched RB-tree

Currently __module_address() is using a linear search through all
modules in order to find the module corresponding to the provided
address. With a lot of modules this can take a lot of time.

One of the users of this is kernel_text_address() which is employed
in many stack unwinders; which in turn are used by perf-callchain and
ftrace (possibly from NMI context).

So by optimizing __module_address() we optimize many stack unwinders
which are used by both perf and tracing in performance sensitive code.

Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

2 files changed, 136 insertions, 8 deletions

diff --git a/include/linux/module.h b/include/linux/module.h
index fb56dd85a862..ddf35a3368fb 100644
--- a/include/linux/module.h
+++ b/include/linux/module.h
@@ -17,6 +17,7 @@
 #include <linux/moduleparam.h>
 #include <linux/jump_label.h>
 #include <linux/export.h>
+#include <linux/rbtree_latch.h>
 
 #include <linux/percpu.h>
 #include <asm/module.h>
@@ -210,6 +211,13 @@ enum module_state {
         MODULE_STATE_UNFORMED,  /* Still setting it up. */
 };
 
+struct module;
+
+struct mod_tree_node {
+        struct module *mod;
+        struct latch_tree_node node;
+};
+
 struct module {
         enum module_state state;
 
@@ -269,8 +277,15 @@ struct module {
         /* Startup function. */
         int (*init)(void);
 
-        /* If this is non-NULL, vfree after init() returns */
-        void *module_init;
+        /*
+         * If this is non-NULL, vfree() after init() returns.
+         *
+         * Cacheline align here, such that:
+         *   module_init, module_core, init_size, core_size,
+         *   init_text_size, core_text_size and ltn_core.node[0]
+         * are on the same cacheline.
+         */
+        void *module_init       ____cacheline_aligned;
 
         /* Here is the actual code + data, vfree'd on unload. */
         void *module_core;
@@ -281,6 +296,14 @@ struct module {
         /* The size of the executable code in each section.  */
         unsigned int init_text_size, core_text_size;
 
+        /*
+         * We want mtn_core::{mod,node[0]} to be in the same cacheline as the
+         * above entries such that a regular lookup will only touch one
+         * cacheline.
+         */
+        struct mod_tree_node    mtn_core;
+        struct mod_tree_node    mtn_init;
+
         /* Size of RO sections of the module (text+rodata) */
         unsigned int init_ro_size, core_ro_size;
 
@@ -367,7 +390,7 @@ struct module {
         ctor_fn_t *ctors;
         unsigned int num_ctors;
 #endif
-};
+} ____cacheline_aligned;
 #ifndef MODULE_ARCH_INIT
 #define MODULE_ARCH_INIT {}
 #endif
diff --git a/kernel/module.c b/kernel/module.c
index a15899e00ca9..e0db5c31cb53 100644
--- a/kernel/module.c
+++ b/kernel/module.c
@@ -101,6 +101,108 @@
 DEFINE_MUTEX(module_mutex);
 EXPORT_SYMBOL_GPL(module_mutex);
 static LIST_HEAD(modules);
+
+/*
+ * Use a latched RB-tree for __module_address(); this allows us to use
+ * RCU-sched lookups of the address from any context.
+ *
+ * Because modules have two address ranges: init and core, we need two
+ * latch_tree_nodes entries. Therefore we need the back-pointer from
+ * mod_tree_node.
+ *
+ * Because init ranges are short lived we mark them unlikely and have placed
+ * them outside the critical cacheline in struct module.
+ */
+
+static __always_inline unsigned long __mod_tree_val(struct latch_tree_node *n)
+{
+        struct mod_tree_node *mtn = container_of(n, struct mod_tree_node, node);
+        struct module *mod = mtn->mod;
+
+        if (unlikely(mtn == &mod->mtn_init))
+                return (unsigned long)mod->module_init;
+
+        return (unsigned long)mod->module_core;
+}
+
+static __always_inline unsigned long __mod_tree_size(struct latch_tree_node *n)
+{
+        struct mod_tree_node *mtn = container_of(n, struct mod_tree_node, node);
+        struct module *mod = mtn->mod;
+
+        if (unlikely(mtn == &mod->mtn_init))
+                return (unsigned long)mod->init_size;
+
+        return (unsigned long)mod->core_size;
+}
+
+static __always_inline bool
+mod_tree_less(struct latch_tree_node *a, struct latch_tree_node *b)
+{
+        return __mod_tree_val(a) < __mod_tree_val(b);
+}
+
+static __always_inline int
+mod_tree_comp(void *key, struct latch_tree_node *n)
+{
+        unsigned long val = (unsigned long)key;
+        unsigned long start, end;
+
+        start = __mod_tree_val(n);
+        if (val < start)
+                return -1;
+
+        end = start + __mod_tree_size(n);
+        if (val >= end)
+                return 1;
+
+        return 0;
+}
+
+static const struct latch_tree_ops mod_tree_ops = {
+        .less = mod_tree_less,
+        .comp = mod_tree_comp,
+};
+
+static struct latch_tree_root mod_tree __cacheline_aligned;
+
+/*
+ * These modifications: insert, remove_init and remove; are serialized by the
+ * module_mutex.
+ */
+static void mod_tree_insert(struct module *mod)
+{
+        mod->mtn_core.mod = mod;
+        mod->mtn_init.mod = mod;
+
+        latch_tree_insert(&mod->mtn_core.node, &mod_tree, &mod_tree_ops);
+        if (mod->init_size)
+                latch_tree_insert(&mod->mtn_init.node, &mod_tree, &mod_tree_ops);
+}
+
+static void mod_tree_remove_init(struct module *mod)
+{
+        if (mod->init_size)
+                latch_tree_erase(&mod->mtn_init.node, &mod_tree, &mod_tree_ops);
+}
+
+static void mod_tree_remove(struct module *mod)
+{
+        latch_tree_erase(&mod->mtn_core.node, &mod_tree, &mod_tree_ops);
+        mod_tree_remove_init(mod);
+}
+
+static struct module *mod_tree_find(unsigned long addr)
+{
+        struct latch_tree_node *ltn;
+
+        ltn = latch_tree_find((void *)addr, &mod_tree, &mod_tree_ops);
+        if (!ltn)
+                return NULL;
+
+        return container_of(ltn, struct mod_tree_node, node)->mod;
+}
+
 #ifdef CONFIG_KGDB_KDB
 struct list_head *kdb_modules = &modules; /* kdb needs the list of modules */
 #endif /* CONFIG_KGDB_KDB */
@@ -1878,6 +1980,7 @@ static void free_module(struct module *mod)
         mutex_lock(&module_mutex);
         /* Unlink carefully: kallsyms could be walking list. */
         list_del_rcu(&mod->list);
+        mod_tree_remove(mod);
         /* Remove this module from bug list, this uses list_del_rcu */
         module_bug_cleanup(mod);
         /* Wait for RCU-sched synchronizing before releasing mod->list and buglist. */
@@ -3145,6 +3248,7 @@ static noinline int do_init_module(struct module *mod)
         mod->symtab = mod->core_symtab;
         mod->strtab = mod->core_strtab;
 #endif
+        mod_tree_remove_init(mod);
         unset_module_init_ro_nx(mod);
         module_arch_freeing_init(mod);
         mod->module_init = NULL;
@@ -3215,6 +3319,7 @@ again:
                 goto out;
         }
         list_add_rcu(&mod->list, &modules);
+        mod_tree_insert(mod);
         err = 0;
 
 out:
@@ -3861,13 +3966,13 @@ struct module *__module_address(unsigned long addr)
 
         module_assert_mutex_or_preempt();
 
-        list_for_each_entry_rcu(mod, &modules, list) {
+        mod = mod_tree_find(addr);
+        if (mod) {
+                BUG_ON(!within_module(addr, mod));
                 if (mod->state == MODULE_STATE_UNFORMED)
-                        continue;
-                if (within_module(addr, mod))
-                        return mod;
+                        mod = NULL;
         }
-        return NULL;
+        return mod;
 }
 EXPORT_SYMBOL_GPL(__module_address);