Re: [PATCH 2/6] module: add support for CONFIG_MODULE_DESTRUCTORS

[Johannes Berg <johannes@sipsolutions.net>]

On Fri, 2021-03-12 at 10:55 +0100, Johannes Berg wrote:
> From: Johannes Berg <johannes.berg@intel.com>
> 
> At least in ARCH=um with CONFIG_GCOV (which writes all the
> coverage data directly out from the userspace binary rather
> than presenting it in debugfs) it's necessary to run all
> the atexit handlers (dtors/fini_array) so that gcov actually
> does write out the data.
> 
> Add a new config option CONFIG_MODULE_DESTRUCTORS that can
> be selected via CONFIG_WANT_MODULE_DESTRUCTORS that the arch
> selects (this indirection exists so the architecture doesn't
> have to worry about whether or not CONFIG_MODULES is on).
> Additionally, the architecture must then (when it exits and
> no more module code can run) call run_all_module_destructors
> to run the code for all modules that are still loaded. When
> modules are unloaded, the handlers are called as well.

Oops, I forgot to add this bit to the patch:

--- a/scripts/module.lds.S
+++ b/scripts/module.lds.S
@@ -16,6 +16,8 @@ SECTIONS {
 
        .init_array             0 : ALIGN(8) { *(SORT(.init_array.*)) *(.init_array) }
 
+       .fini_array             0 : ALIGN(8) { *(SORT(.fini_array.*)) *(.fini_array) }
+
        __jump_table            0 : ALIGN(8) { KEEP(*(__jump_table)) }
 
        __patchable_function_entries : { *(__patchable_function_entries) }


Should that be under the ifdef? .init_array isn't, even though it's only
relevant for CONFIG_CONSTRUCTORS.

johannes