Skip to contentSkip to conversation

ARM: module: treat exit sections the same as init sections when !CONFIG_MODULE_UNLOAD

PATCH In review Jessica Yu 2 participants 3 messages May 7, 2021 Arm (32-bit)

lore
Share & tools
lore

Jessica Yu

author
#0 lore raw

PATCH 1/1 ARM: module: treat exit sections the same as init sections when !CONFIG_MODULE_UNLOAD

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")

---
 arch/arm/kernel/module.c | 7 +++++++
 1 file changed, 7 insertions(+)
arch/arm/kernel/module.c +7 -0
index beac45e89ba6..9d403fc1893b 100644
@@ -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)
Signature · 1 line
2.31.1
#1 lore raw
Quoted 3 lines · Jessica Yu

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.

...

Quoted 7 lines

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.

Quoted 6 lines

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?

Signature · 2 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!
author
#2 lore raw

+++ Russell King - ARM Linux admin [07/05/21 13:30 +0100]:

Quoted 19 lines · Jessica Yu
Quoted 3 lines

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.

...

Quoted 7 lines

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.

Quoted 6 lines

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.

Quoted 6 lines

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