From: Arnd Bergmann <[email protected]>
__symbol_put() is really meant as an internal helper and is not available
when module unloading is disabled, unlike the previously used symbol_put():
samples/hw_breakpoint/data_breakpoint.c: In function 'hw_break_module_exit':
samples/hw_breakpoint/data_breakpoint.c:73:9: error: implicit declaration of function '__symbol_put'; did you mean '__symbol_get'? [-Werror=implicit-function-declaration]
The hw_break_module_exit() function is not actually used when module
unloading is disabled, but it still causes the build failure for an
undefined identifier. Enclose this one call in an appropriate #ifdef to
clarify what the requirement is. Leaving out the entire exit function
would also work but feels less clar in this case.
Fixes: 910e230d5f1bb ("samples/hw_breakpoint: Fix kernel BUG 'invalid opcode: 0000'")
Fixes: d8a84d33a4954 ("samples/hw_breakpoint: drop use of kallsyms_lookup_name()")
Signed-off-by: Arnd Bergmann <[email protected]>
---
samples/hw_breakpoint/data_breakpoint.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/samples/hw_breakpoint/data_breakpoint.c b/samples/hw_breakpoint/data_breakpoint.c
index 9debd128b2ab8..b99322f188e59 100644
--- a/samples/hw_breakpoint/data_breakpoint.c
+++ b/samples/hw_breakpoint/data_breakpoint.c
@@ -70,7 +70,9 @@ static int __init hw_break_module_init(void)
static void __exit hw_break_module_exit(void)
{
unregister_wide_hw_breakpoint(sample_hbp);
+#ifdef CONFIG_MODULE_UNLOAD
__symbol_put(ksym_name);
+#endif
printk(KERN_INFO "HW Breakpoint for %s write uninstalled\n", ksym_name);
}
--
2.39.2
On Tue 2023-07-25 10:25:36, Arnd Bergmann wrote:
> From: Arnd Bergmann <[email protected]>
>
> __symbol_put() is really meant as an internal helper and is not available
> when module unloading is disabled, unlike the previously used symbol_put():
Ah, I have missed this.
> samples/hw_breakpoint/data_breakpoint.c: In function 'hw_break_module_exit':
> samples/hw_breakpoint/data_breakpoint.c:73:9: error: implicit declaration of function '__symbol_put'; did you mean '__symbol_get'? [-Werror=implicit-function-declaration]
>
> The hw_break_module_exit() function is not actually used when module
> unloading is disabled, but it still causes the build failure for an
> undefined identifier. Enclose this one call in an appropriate #ifdef to
> clarify what the requirement is. Leaving out the entire exit function
> would also work but feels less clar in this case.
> --- a/samples/hw_breakpoint/data_breakpoint.c
> +++ b/samples/hw_breakpoint/data_breakpoint.c
> @@ -70,7 +70,9 @@ static int __init hw_break_module_init(void)
> static void __exit hw_break_module_exit(void)
> {
> unregister_wide_hw_breakpoint(sample_hbp);
> +#ifdef CONFIG_MODULE_UNLOAD
> __symbol_put(ksym_name);
> +#endif
> printk(KERN_INFO "HW Breakpoint for %s write uninstalled\n", ksym_name);
> }
I see that both symbol_put() and symbol_put_addr() are defined
as noops twice:
+ when !defined(CONFIG_MODULE_UNLOAD)
+ when !defined(CONFIG_MODULES)
This patch is enough because samples/hw_breakpoint/data_breakpoint.c
could be built only as a module, so feel free to use:
Reviewed-by: Petr Mladek <[email protected]>
But even better solution would be to define symbol_put_name() as
a noop in both situations and use it.
Best Regards,
Petr
On Tue, Jul 25, 2023 at 11:38:48AM +0200, Petr Mladek wrote:
> Reviewed-by: Petr Mladek <[email protected]>
Thanks, patch applied and pushed!
> But even better solution would be to define symbol_put_name() as
> a noop in both situations and use it.
Patch welcomed!
Luis