event: Add basic support for events

Add a way to create and dispatch events without needing to allocate
memory. Also add a way to 'spy' on events, thus allowing 'hooks' to be
created.

Use a linker list for static events, which we can use to replace functions
like arch_cpu_init_f(). Allow an EVENT_DEBUG option which makes it
easier to see what is going on at runtime, but uses more code space.

Dynamic events allow the creation of a spy at runtime. This is not always
necessary, but can be enabled with EVENT_DYNAMIC if needed.

A 'test' event is the only option for now.

Signed-off-by: Simon Glass <sjg@chromium.org>

MAINTAINERS

@@ -809,6 +809,12 @@ S:	Maintained
 F:	doc/usage/environment.rst
 F:	scripts/env2string.awk
 
+EVENTS
+M:	Simon Glass <sjg@chromium.org>
+S:	Maintained
+F:	common/event.c
+F:	include/event.h
+
 FASTBOOT
 S:	Orphaned
 F:	cmd/fastboot.c

common/Kconfig

@@ -492,6 +492,37 @@ config DISPLAY_BOARDINFO_LATE
 
 menu "Start-up hooks"
 
+config EVENT
+	bool "General-purpose event-handling mechanism"
+	default y if SANDBOX
+	help
+	  This enables sending and processing of events, to allow interested
+	  parties to be alerted when something happens. This is an attempt to
+	  step the flow of weak functions, hooks, functions in board_f.c
+	  and board_r.c and the Kconfig options below.
+
+	  See doc/develop/event.rst for more information.
+
+if EVENT
+
+config EVENT_DYNAMIC
+	bool "Support event registration at runtime"
+	default y if SANDBOX
+	help
+	  Enable this to support adding an event spy at runtime, without adding
+	  it to the EVENT_SPy() linker list. This increases code size slightly
+	  but provides more flexibility for boards and subsystems that need it.
+
+config EVENT_DEBUG
+	bool "Enable event debugging assistance"
+	default y if SANDBOX
+	help
+	  Enable this get usefui features for seeing what is happening with
+	  events, such as event-type names. This adds to the code size of
+	  U-Boot so can be turned off for production builds.
+
+endif # EVENT
+
 config ARCH_EARLY_INIT_R
 	bool "Call arch-specific init soon after relocation"
 	help

common/Makefile

@@ -89,6 +89,8 @@ obj-y += malloc_simple.o
 endif
 endif
 
+obj-$(CONFIG_$(SPL_TPL_)EVENT) += event.o
+
 obj-$(CONFIG_$(SPL_TPL_)HASH) += hash.o
 obj-$(CONFIG_IO_TRACE) += iotrace.o
 obj-y += memsize.o

common/board_r.c

@@ -594,6 +594,7 @@ static int run_main_loop(void)
 static init_fnc_t init_sequence_r[] = {
 	initr_trace,
 	initr_reloc,
+	event_init,
 	/* TODO: could x86/PPC have this also perhaps? */
 #if defined(CONFIG_ARM) || defined(CONFIG_RISCV)
 	initr_caches,

common/event.c

@@ -0,0 +1,186 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * Events provide a general-purpose way to react to / subscribe to changes
+ * within U-Boot
+ *
+ * Copyright 2021 Google LLC
+ * Written by Simon Glass <sjg@chromium.org>
+ */
+
+#define LOG_CATEGORY	LOGC_EVENT
+
+#include <common.h>
+#include <event.h>
+#include <event_internal.h>
+#include <log.h>
+#include <linker_lists.h>
+#include <malloc.h>
+#include <asm/global_data.h>
+#include <linux/list.h>
+
+DECLARE_GLOBAL_DATA_PTR;
+
+#if CONFIG_IS_ENABLED(EVENT_DEBUG)
+const char *const type_name[] = {
+	"none",
+	"test",
+};
+
+_Static_assert(ARRAY_SIZE(type_name) == EVT_COUNT, "event type_name size");
+#endif
+
+static const char *event_type_name(enum event_t type)
+{
+#if CONFIG_IS_ENABLED(EVENT_DEBUG)
+	return type_name[type];
+#else
+	return "(unknown)";
+#endif
+}
+
+static int notify_static(struct event *ev)
+{
+	struct evspy_info *start =
+		ll_entry_start(struct evspy_info, evspy_info);
+	const int n_ents = ll_entry_count(struct evspy_info, evspy_info);
+	struct evspy_info *spy;
+
+	for (spy = start; spy != start + n_ents; spy++) {
+		if (spy->type == ev->type) {
+			int ret;
+
+			log_debug("Sending event %x/%s to spy '%s'\n", ev->type,
+				  event_type_name(ev->type), event_spy_id(spy));
+			ret = spy->func(NULL, ev);
+
+			/*
+			 * TODO: Handle various return codes to
+			 *
+			 * - claim an event (no others will see it)
+			 * - return an error from the event
+			 */
+			if (ret)
+				return log_msg_ret("spy", ret);
+		}
+	}
+
+	return 0;
+}
+
+static int notify_dynamic(struct event *ev)
+{
+	struct event_state *state = gd_event_state();
+	struct event_spy *spy, *next;
+
+	list_for_each_entry_safe(spy, next, &state->spy_head, sibling_node) {
+		if (spy->type == ev->type) {
+			int ret;
+
+			log_debug("Sending event %x/%s to spy '%s'\n", ev->type,
+				  event_type_name(ev->type), spy->id);
+			ret = spy->func(spy->ctx, ev);
+
+			/*
+			 * TODO: Handle various return codes to
+			 *
+			 * - claim an event (no others will see it)
+			 * - return an error from the event
+			 */
+			if (ret)
+				return log_msg_ret("spy", ret);
+		}
+	}
+
+	return 0;
+}
+
+int event_notify(enum event_t type, void *data, int size)
+{
+	struct event event;
+	int ret;
+
+	event.type = type;
+	if (size > sizeof(event.data))
+		return log_msg_ret("size", -E2BIG);
+	memcpy(&event.data, data, size);
+
+	ret = notify_static(&event);
+	if (ret)
+		return log_msg_ret("dyn", ret);
+
+	if (CONFIG_IS_ENABLED(EVENT_DYNAMIC)) {
+		ret = notify_dynamic(&event);
+		if (ret)
+			return log_msg_ret("dyn", ret);
+	}
+
+	return 0;
+}
+
+int event_notify_null(enum event_t type)
+{
+	return event_notify(type, NULL, 0);
+}
+
+void event_show_spy_list(void)
+{
+	struct evspy_info *start =
+		ll_entry_start(struct evspy_info, evspy_info);
+	const int n_ents = ll_entry_count(struct evspy_info, evspy_info);
+	struct evspy_info *spy;
+	const int size = sizeof(ulong) * 2;
+
+	printf("Seq  %-24s  %*s  %s\n", "Type", size, "Function", "ID");
+	for (spy = start; spy != start + n_ents; spy++) {
+		printf("%3x  %-3x %-20s  %*p  %s\n", (uint)(spy - start),
+		       spy->type, event_type_name(spy->type), size, spy->func,
+		       event_spy_id(spy));
+	}
+}
+
+#if CONFIG_IS_ENABLED(EVENT_DYNAMIC)
+static void spy_free(struct event_spy *spy)
+{
+	list_del(&spy->sibling_node);
+}
+
+int event_register(const char *id, enum event_t type, event_handler_t func, void *ctx)
+{
+	struct event_state *state = gd_event_state();
+	struct event_spy *spy;
+
+	if (!CONFIG_IS_ENABLED(EVENT_DYNAMIC))
+		return -ENOSYS;
+	spy = malloc(sizeof(*spy));
+	if (!spy)
+		return log_msg_ret("alloc", -ENOMEM);
+
+	spy->id = id;
+	spy->type = type;
+	spy->func = func;
+	spy->ctx = ctx;
+	list_add_tail(&spy->sibling_node, &state->spy_head);
+
+	return 0;
+}
+
+int event_uninit(void)
+{
+	struct event_state *state = gd_event_state();
+	struct event_spy *spy, *next;
+
+	list_for_each_entry_safe(spy, next, &state->spy_head, sibling_node)
+		spy_free(spy);
+
+	return 0;
+}
+
+int event_init(void)
+{
+	struct event_state *state = gd_event_state();
+
+	INIT_LIST_HEAD(&state->spy_head);
+
+	return 0;
+}
+#endif /* EVENT_DYNAMIC */

common/log.c

@@ -28,6 +28,7 @@ static const char *const log_cat_name[] = {
 	"devres",
 	"acpi",
 	"boot",
+	"event",
 };
 
 _Static_assert(ARRAY_SIZE(log_cat_name) == LOGC_COUNT - LOGC_NONE,

include/asm-generic/global_data.h

@@ -20,6 +20,7 @@
  */
 
 #ifndef __ASSEMBLY__
+#include <event_internal.h>
 #include <fdtdec.h>
 #include <membuff.h>
 #include <linux/list.h>
@@ -467,6 +468,12 @@ struct global_data {
 	 */
 	char *smbios_version;
 #endif
+#if CONFIG_IS_ENABLED(EVENT)
+	/**
+	 * @event_state: Points to the current state of events
+	 */
+	struct event_state event_state;
+#endif
 };
 #ifndef DO_DEPS_ONLY
 static_assert(sizeof(struct global_data) == GD_SIZE);
@@ -532,6 +539,12 @@ static_assert(sizeof(struct global_data) == GD_SIZE);
 #define gd_set_multi_dtb_fit(_dtb)
 #endif
 
+#if CONFIG_IS_ENABLED(EVENT_DYNAMIC)
+#define gd_event_state()	((struct event_state *)&gd->event_state)
+#else
+#define gd_event_state()	NULL
+#endif
+
 /**
  * enum gd_flags - global data flags
  *

include/event.h

@@ -0,0 +1,188 @@
+/* SPDX-License-Identifier: GPL-2.0+ */
+/*
+ * Events provide a general-purpose way to react to / subscribe to changes
+ * within U-Boot
+ *
+ * Copyright 2021 Google LLC
+ * Written by Simon Glass <sjg@chromium.org>
+ */
+
+#ifndef __event_h
+#define __event_h
+
+/**
+ * enum event_t - Types of events supported by U-Boot
+ *
+ * @EVT_DM_PRE_PROBE: Device is about to be probed
+ */
+enum event_t {
+	EVT_NONE,
+	EVT_TEST,
+
+	EVT_COUNT
+};
+
+union event_data {
+	/**
+	 * struct event_data_test  - test data
+	 *
+	 * @signal: A value to update the state with
+	 */
+	struct event_data_test {
+		int signal;
+	} test;
+};
+
+/**
+ * struct event - an event that can be sent and received
+ *
+ * @type: Event type
+ * @data: Data for this particular event
+ */
+struct event {
+	enum event_t type;
+	union event_data data;
+};
+
+/** Function type for event handlers */
+typedef int (*event_handler_t)(void *ctx, struct event *event);
+
+/**
+ * struct evspy_info - information about an event spy
+ *
+ * @func: Function to call when the event is activated (must be first)
+ * @type: Event type
+ * @id: Event id string
+ */
+struct evspy_info {
+	event_handler_t func;
+	enum event_t type;
+#if CONFIG_IS_ENABLED(EVENT_DEBUG)
+	const char *id;
+#endif
+};
+
+/* Declare a new event spy */
+#if CONFIG_IS_ENABLED(EVENT_DEBUG)
+#define _ESPY_REC(_type, _func)   { _func, _type, #_func, }
+#else
+#define _ESPY_REC(_type, _func)   { _func, _type, }
+#endif
+
+static inline const char *event_spy_id(struct evspy_info *spy)
+{
+#if CONFIG_IS_ENABLED(EVENT_DEBUG)
+	return spy->id;
+#else
+	return "?";
+#endif
+}
+
+/*
+ * It seems that LTO will drop list entries if it decides they are not used,
+ * although the conditions that cause this are unclear.
+ *
+ * The example found is the following:
+ *
+ * static int sandbox_misc_init_f(void *ctx, struct event *event)
+ * {
+ *    return sandbox_early_getopt_check();
+ * }
+ * EVENT_SPY(EVT_MISC_INIT_F, sandbox_misc_init_f);
+ *
+ * where EVENT_SPY uses ll_entry_declare()
+ *
+ * In this case, LTO decides to drop the sandbox_misc_init_f() function
+ * (which is fine) but then drops the linker-list entry too. This means
+ * that the code no longer works, in this case sandbox no-longer checks its
+ * command-line arguments properly.
+ *
+ * Without LTO, the KEEP() command in the .lds file is enough to keep the
+ * entry around. But with LTO it seems that the entry has already been
+ * dropped before the link script is considered.
+ *
+ * The only solution I can think of is to mark linker-list entries as 'used'
+ * using an attribute. This should be safe, since we don't actually want to drop
+ * any of these. However this does slightly limit LTO's optimisation choices.
+ */
+#define EVENT_SPY(_type, _func) \
+	static __attribute__((used)) ll_entry_declare(struct evspy_info, \
+						      _type, evspy_info) = \
+	_ESPY_REC(_type, _func)
+
+/**
+ * event_register - register a new spy
+ *
+ * @id: Spy ID
+ * @type: Event type to subscribe to
+ * @func: Function to call when the event is sent
+ * @ctx: Context to pass to the function
+ * @return 0 if OK, -ve on error
+ */
+int event_register(const char *id, enum event_t type, event_handler_t func,
+		   void *ctx);
+
+#if CONFIG_IS_ENABLED(EVENT)
+/**
+ * event_notify() - notify spies about an event
+ *
+ * It is possible to pass in union event_data here but that may not be
+ * convenient if the data is elsewhere, or is one of the members of the union.
+ * So this uses a void * for @data, with a separate @size.
+ *
+ * @type: Event type
+ * @data: Event data to be sent (e.g. union_event_data)
+ * @size: Size of data in bytes
+ * @return 0 if OK, -ve on error
+ */
+int event_notify(enum event_t type, void *data, int size);
+
+/**
+ * event_notify_null() - notify spies about an event
+ *
+ * Data is NULL and the size is 0
+ *
+ * @type: Event type
+ * @return 0 if OK, -ve on error
+ */
+int event_notify_null(enum event_t type);
+#else
+static inline int event_notify(enum event_t type, void *data, int size)
+{
+	return 0;
+}
+
+static inline int event_notify_null(enum event_t type)
+{
+	return 0;
+}
+#endif
+
+#if CONFIG_IS_ENABLED(EVENT_DYNAMIC)
+/**
+ * event_uninit() - Clean up dynamic events
+ *
+ * This removes all dynamic event handlers
+ */
+int event_uninit(void);
+
+/**
+ * event_uninit() - Set up dynamic events
+ *
+ * Init a list of dynamic event handlers, so that these can be added as
+ * needed
+ */
+int event_init(void);
+#else
+static inline int event_uninit(void)
+{
+	return 0;
+}
+
+static inline int event_init(void)
+{
+	return 0;
+}
+#endif
+
+#endif

include/event_internal.h

@@ -0,0 +1,35 @@
+/* SPDX-License-Identifier: GPL-2.0+ */
+/*
+ * Internal definitions for events
+ *
+ * Copyright 2021 Google LLC
+ * Written by Simon Glass <sjg@chromium.org>
+ */
+
+#ifndef __event_internal_h
+#define __event_internal_h
+
+#include <event.h>
+#include <linux/list.h>
+
+/**
+ * struct event_spy - a spy that watches for an event of a particular type
+ *
+ * @id: Spy ID
+ * @type: Event type to subscribe to
+ * @func: Function to call when the event is sent
+ * @ctx: Context to pass to the function
+ */
+struct event_spy {
+	struct list_head sibling_node;
+	const char *id;
+	enum event_t type;
+	event_handler_t func;
+	void *ctx;
+};
+
+struct event_state {
+	struct list_head spy_head;
+};
+
+#endif

include/log.h

@@ -98,6 +98,8 @@ enum log_category_t {
 	LOGC_ACPI,
 	/** @LOGC_BOOT: Related to boot process / boot image processing */
 	LOGC_BOOT,
+	/** @LOGC_EVENT: Related to event and event handling */
+	LOGC_EVENT,
 	/** @LOGC_COUNT: Number of log categories */
 	LOGC_COUNT,
 	/** @LOGC_END: Sentinel value for lists of log categories */