static_call: Allow early init

In order to use static_call() to wire up x86_pmu, we need to
initialize earlier, specifically before memory allocation works; copy
some of the tricks from jump_label to enable this.

Primarily we overload key->next to store a sites pointer when there
are no modules, this avoids having to use kmalloc() to initialize the
sites and allows us to run much earlier.

Signed-off-by: Peter Zijlstra (Intel) <[e-mail]>
Signed-off-by: Ingo Molnar <[e-mail]>
Reviewed-by: Steven Rostedt (VMware) <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]

arch/x86/kernel/setup.c

···
 #include <linux/hugetlb.h>
 #include <linux/tboot.h>
 #include <linux/usb/xhci-dbgp.h>
+#include <linux/static_call.h>
 
 #include <uapi/linux/mount.h>
 
···
 	early_cpu_init();
 	arch_init_ideal_nops();
 	jump_label_init();
+	static_call_init();
 	early_ioremap_init();
 
 	setup_olpc_ofw_pgd();

arch/x86/kernel/static_call.c

···
 	RET = 3,  /* tramp / site cond-tail-call */
 };
 
-static void __static_call_transform(void *insn, enum insn_type type, void *func)
+static void __ref __static_call_transform(void *insn, enum insn_type type, void *func)
 {
 	int size = CALL_INSN_SIZE;
 	const void *code;
···
 
 	if (memcmp(insn, code, size) == 0)
 		return;
+
+	if (unlikely(system_state == SYSTEM_BOOTING))
+		return text_poke_early(insn, code, size);
 
 	text_poke_bp(insn, code, size, NULL);
 }

include/linux/static_call.h

···
 
 #ifdef CONFIG_HAVE_STATIC_CALL_INLINE
 
+extern void __init static_call_init(void);
+
 struct static_call_mod {
 	struct static_call_mod *next;
 	struct module *mod; /* for vmlinux, mod == NULL */
···
 
 struct static_call_key {
 	void *func;
-	struct static_call_mod *mods;
+	union {
+		/* bit 0: 0 = mods, 1 = sites */
+		unsigned long type;
+		struct static_call_mod *mods;
+		struct static_call_site *sites;
+	};
 };
 
 extern void __static_call_update(struct static_call_key *key, void *tramp, void *func);
···
 	DECLARE_STATIC_CALL(name, _func);				\
 	struct static_call_key STATIC_CALL_KEY(name) = {		\
 		.func = _func,						\
-		.mods = NULL,						\
+		.type = 1,						\
 	};								\
 	ARCH_DEFINE_STATIC_CALL_TRAMP(name, _func)
 
···
 	EXPORT_SYMBOL_GPL(STATIC_CALL_TRAMP(name))
 
 #elif defined(CONFIG_HAVE_STATIC_CALL)
+
+static inline void static_call_init(void) { }
 
 struct static_call_key {
 	void *func;
···
 	EXPORT_SYMBOL_GPL(STATIC_CALL_TRAMP(name))
 
 #else /* Generic implementation */
+
+static inline void static_call_init(void) { }
 
 struct static_call_key {
 	void *func;

kernel/static_call.c

···
 	     static_call_site_cmp, static_call_site_swap);
 }
 
+static inline bool static_call_key_has_mods(struct static_call_key *key)
+{
+	return !(key->type & 1);
+}
+
+static inline struct static_call_mod *static_call_key_next(struct static_call_key *key)
+{
+	if (!static_call_key_has_mods(key))
+		return NULL;
+
+	return key->mods;
+}
+
+static inline struct static_call_site *static_call_key_sites(struct static_call_key *key)
+{
+	if (static_call_key_has_mods(key))
+		return NULL;
+
+	return (struct static_call_site *)(key->type & ~1);
+}
+
 void __static_call_update(struct static_call_key *key, void *tramp, void *func)
 {
 	struct static_call_site *site, *stop;
-	struct static_call_mod *site_mod;
+	struct static_call_mod *site_mod, first;
 
 	cpus_read_lock();
 	static_call_lock();
···
 	if (WARN_ON_ONCE(!static_call_initialized))
 		goto done;
 
-	for (site_mod = key->mods; site_mod; site_mod = site_mod->next) {
+	first = (struct static_call_mod){
+		.next = static_call_key_next(key),
+		.mod = NULL,
+		.sites = static_call_key_sites(key),
+	};
+
+	for (site_mod = &first; site_mod; site_mod = site_mod->next) {
 		struct module *mod = site_mod->mod;
 
 		if (!site_mod->sites) {
 			/*
 			 * This can happen if the static call key is defined in
 			 * a module which doesn't use it.
+			 *
+			 * It also happens in the has_mods case, where the
+			 * 'first' entry has no sites associated with it.
 			 */
 			continue;
 		}
···
 		if (key != prev_key) {
 			prev_key = key;
 
+			/*
+			 * For vmlinux (!mod) avoid the allocation by storing
+			 * the sites pointer in the key itself. Also see
+			 * __static_call_update()'s @first.
+			 *
+			 * This allows architectures (eg. x86) to call
+			 * static_call_init() before memory allocation works.
+			 */
+			if (!mod) {
+				key->sites = site;
+				key->type |= 1;
+				goto do_transform;
+			}
+
 			site_mod = kzalloc(sizeof(*site_mod), GFP_KERNEL);
 			if (!site_mod)
 				return -ENOMEM;
 
+			/*
+			 * When the key has a direct sites pointer, extract
+			 * that into an explicit struct static_call_mod, so we
+			 * can have a list of modules.
+			 */
+			if (static_call_key_sites(key)) {
+				site_mod->mod = NULL;
+				site_mod->next = NULL;
+				site_mod->sites = static_call_key_sites(key);
+
+				key->mods = site_mod;
+
+				site_mod = kzalloc(sizeof(*site_mod), GFP_KERNEL);
+				if (!site_mod)
+					return -ENOMEM;
+			}
+
 			site_mod->mod = mod;
 			site_mod->sites = site;
-			site_mod->next = key->mods;
+			site_mod->next = static_call_key_next(key);
 			key->mods = site_mod;
 		}
 
+do_transform:
 		arch_static_call_transform(site_addr, NULL, key->func,
 				static_call_is_tail(site));
 	}
···
 	return __static_call_mod_text_reserved(start, end);
 }
 
-static void __init static_call_init(void)
+void __init static_call_init(void)
 {
 	int ret;