module load notification try 2
PATCH 1/1 module load notification try 2
Rusty, any happier with this one ?
The below implements a module load notification for oprofile's benefit. The oprofile part is not attached here, but seems to work OK.
regards
john
include/linux/module.h +15 -0
@@ -142,6 +142,7 @@const struct exception_table_entry *entry; }; +struct notifier_block ;#ifdef CONFIG_MODULES
@@ -352,6 +353,9 @@/* For extable.c to search modules' exception tables. */ const struct exception_table_entry *search_module_extables (unsigned long addr); +int register_module_notifier (struct notifier_block * nb); +int unregister_module_notifier (struct notifier_block * nb); +#else /* !CONFIG_MODULES... */ #define EXPORT_SYMBOL (sym)#define EXPORT_SYMBOL_GPL (sym)
@@ -396,6 +400,17 @@ {return NULL ; } + +static inline int register_module_notifier (struct notifier_block * nb) +{ +return -ENOSYS ; +} + +static inline int unregister_module_notifier (struct notifier_block * nb) +{ +return -ENOSYS ; +} +#endif /* CONFIG_MODULES */ #ifdef MODULE
kernel/module.c +28 -0
@@ -31,6 +31,7 @@#include <linux/errno.h> #include <linux/err.h> #include <linux/vermagic.h> +#include <linux/notifier.h> #include <asm/uaccess.h> #include <asm/semaphore.h> #include <asm/pgalloc.h>
@@ -61,6 +62,29 @@static LIST_HEAD (symbols);static LIST_HEAD (extables); +static DECLARE_MUTEX (notify_mutex); +static struct notifier_block * module_notify_list; + +int register_module_notifier (struct notifier_block * nb) +{ +int err; +down (¬ify_mutex); + err =notifier_chain_register (&module_notify_list, nb); +up (¬ify_mutex); +return err; +} +EXPORT_SYMBOL (register_module_notifier); + +int unregister_module_notifier (struct notifier_block * nb) +{ +int err; +down (¬ify_mutex); + err =notifier_chain_unregister (&module_notify_list, nb); +up (¬ify_mutex); +return err; +} +EXPORT_SYMBOL (unregister_module_notifier); +/* We require a truly strong try_module_get() */ static inline int strong_try_module_get (struct module *mod) {
@@ -1368,6 +1392,10 @@/* Drop lock so they can recurse */ up (&module_mutex); +down (¬ify_mutex); +notifier_call_chain (&module_notify_list,MODULE_STATE_COMING , mod); +up (¬ify_mutex); +/* Start the module */ ret = mod->init ();if (ret <0 ) {
In message <[email protected]> you write:
Quoted 2 lines
Rusty, any happier with this one ?
Yep, much simpler. Couple of comments:
include/linux/module.h line 414
Quoted 11 lines
include/linux/module.h +10
-0+static inline int register_module_notifier (struct notifier_block * nb) +{ +return -ENOSYS ; +} + +static inline int unregister_module_notifier (struct notifier_block * nb) +{ +return -ENOSYS ; +} +#endif /* CONFIG_MODULES */
Shouldn't fail just because !CONFIG_MODULES: should just return 0. Otherwise there's no way to sanely use them without wrapping in #ifdef CONFIG_MODULES (in which case why have the non-module definition at all?).
Quoted 1 line
+static
DECLARE_MUTEX(notify_mutex);
Hmm, yes, you need to use your own protection around notifier_chain_register and notifier_call_chain. Wierd, because notifier.c does its own locking for register and unregister, but not for calling, which AFAICT makes it useless...
Rusty.
--
Anyone who quotes me in their sig is an idiot. -- Rusty Russell.
Quoted 6 lines · Rusty Russell
include/linux/module.h line 407
Quoted 4 lines
include/linux/module.h +4
-0+static inline int register_module_notifier (struct notifier_block * nb) +{ +return -ENOSYS ; +}Shouldn't fail just because !CONFIG_MODULES: should just return 0.
Ho-de-hum, flipped a coin, lost the toss.
Quoted 1 line
Otherwise there's no way to sanely use them without wrapping in #ifdef
Well, "if (err && err != -ENOSYS)". The relative sanity of that is debatable. I don't care either way, hence the coin toss.
I Cc:ed Zwane who was advocating -ENOSYS on IRC ...
Quoted 6 lines
Quoted 1 line
+static
DECLARE_MUTEX(notify_mutex);Hmm, yes, you need to use your own protection around
notifier_chain_registerandnotifier_call_chain. Wierd, becausenotifier.cdoes its own locking for register and unregister, but not for calling, which AFAICT makes it useless...
I mentioned about this some time ago on lkml to massive indifference. Later on someone from OSDL reworked all the notifier stuff, dunno what happened to the patch.
regards,
john
In message <[email protected]> you write:
Quoted 4 lines
Quoted 1 line
Otherwise there's no way to sanely use them without wrapping in #ifdef
Well, "if (err && err !=
-ENOSYS)". The relative sanity of that is debatable. I don't care either way, hence the coin toss.
OK, well, I have an opinion and it's the other one. 8)
Quoted 6 lines
Quoted 4 lines
Hmm, yes, you need to use your own protection around
notifier_chain_registerandnotifier_call_chain. Wierd, becausenotifier.cdoes its own locking for register and unregister, but not for calling, which AFAICT makes it useless...I mentioned about this some time ago on lkml to massive indifference.
Yeah, it's a PITA.
Cheers,
Rusty.
--
Anyone who quotes me in their sig is an idiot. -- Rusty Russell.