Dynamic code patching (alternatives, jump_label and static_call) can
have sites in __exit code, even if __exit is never executed. Therefore
__exit must be present at runtime, at least for as long as __init code is.
Additionally, for jump_label and static_call, the __exit sites must also
identify as within_module_init(), such that the infrastructure is aware
to never touch them after module init -- alternatives are only ran once
at init and hence don't have this particular constraint.
Previously, the module loader never loaded the exit sections in the first
place when CONFIG_MODULE_UNLOAD=n. Commit 33121347fb1c ("module: treat exit
sections the same as init sections when !CONFIG_MODULE_UNLOAD") addressed
the issue by having the module loader load the exit sections and then making
__exit identify as __init for !MODULE_UNLOAD. Then, since they are treated
like init sections, they will be also discarded after init.
That commit satisfied the above requirements for jump_labels and
static_calls by modifying the checks in the core module_init_section()
function in kernel/module.c to include exit sections. However, ARM
overrides these and implements their own module_{init,exit}_section()
functions. Add a similar check for exit sections to ARM's
module_init_section() function so that all arches are on the same page.
Fixes: 33121347fb1c ("module: treat exit sections the same as init
sections when !CONFIG_MODULE_UNLOAD")
Link: https://lore.kernel.org/lkml/YFiuphGw0RKehWsQ@gunter/
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jessica Yu <[email protected]>
---
arch/arm/kernel/module.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/arch/arm/kernel/module.c b/arch/arm/kernel/module.c
index beac45e89ba6..9d403fc1893b 100644
--- a/arch/arm/kernel/module.c
+++ b/arch/arm/kernel/module.c
@@ -56,9 +56,16 @@ void *module_alloc(unsigned long size)
bool module_init_section(const char *name)
{
+#ifndef CONFIG_MODULE_UNLOAD
+ return strstarts(name, ".init") ||
+ strstarts(name, ".ARM.extab.init") ||
+ strstarts(name, ".ARM.exidx.init") ||
+ module_exit_section(name);
+#else
return strstarts(name, ".init") ||
strstarts(name, ".ARM.extab.init") ||
strstarts(name, ".ARM.exidx.init");
+#endif
}
bool module_exit_section(const char *name)
--
2.31.1
On Fri, May 07, 2021 at 02:13:22PM +0200, Jessica Yu wrote:
> Dynamic code patching (alternatives, jump_label and static_call) can
> have sites in __exit code, even if __exit is never executed. Therefore
> __exit must be present at runtime, at least for as long as __init code is.
...
> Previously, the module loader never loaded the exit sections in the first
> place when CONFIG_MODULE_UNLOAD=n. Commit 33121347fb1c ("module: treat exit
> sections the same as init sections when !CONFIG_MODULE_UNLOAD") addressed
> the issue by having the module loader load the exit sections and then making
> __exit identify as __init for !MODULE_UNLOAD. Then, since they are treated
> like init sections, they will be also discarded after init.
>
> That commit satisfied the above requirements for jump_labels and
> static_calls by modifying the checks in the core module_init_section()
> function in kernel/module.c to include exit sections. However, ARM
> overrides these and implements their own module_{init,exit}_section()
> functions. Add a similar check for exit sections to ARM's
> module_init_section() function so that all arches are on the same page.
Shouldn't the module core code itself be doing:
module_init_section(name) || module_exit_section(name)
itself when CONFIG_MODULE_UNLOAD is not set, rather than pushing this
logic down into every module_init_section() implementation?
--
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!
+++ Russell King - ARM Linux admin [07/05/21 13:30 +0100]:
>On Fri, May 07, 2021 at 02:13:22PM +0200, Jessica Yu wrote:
>> Dynamic code patching (alternatives, jump_label and static_call) can
>> have sites in __exit code, even if __exit is never executed. Therefore
>> __exit must be present at runtime, at least for as long as __init code is.
>...
>> Previously, the module loader never loaded the exit sections in the first
>> place when CONFIG_MODULE_UNLOAD=n. Commit 33121347fb1c ("module: treat exit
>> sections the same as init sections when !CONFIG_MODULE_UNLOAD") addressed
>> the issue by having the module loader load the exit sections and then making
>> __exit identify as __init for !MODULE_UNLOAD. Then, since they are treated
>> like init sections, they will be also discarded after init.
>>
>> That commit satisfied the above requirements for jump_labels and
>> static_calls by modifying the checks in the core module_init_section()
>> function in kernel/module.c to include exit sections. However, ARM
>> overrides these and implements their own module_{init,exit}_section()
>> functions. Add a similar check for exit sections to ARM's
>> module_init_section() function so that all arches are on the same page.
>
>Shouldn't the module core code itself be doing:
>
> module_init_section(name) || module_exit_section(name)
>
>itself when CONFIG_MODULE_UNLOAD is not set, rather than pushing this
>logic down into every module_init_section() implementation?
Yeah, that sounds better. Originally, I had wanted to keep the #ifndef
in one place to keep the churn to a minimum.
But seeing that we have to patch up ARM too, it's probably the less
ugly option now. Let me cook up an alternative patch and resend.
Thanks,
Jessica