livepatch: Remove .klp.arch

After the previous patch, vmlinux-specific KLP relocations are now
applied early during KLP module load.  This means that .klp.arch
sections are no longer needed for *vmlinux-specific* KLP relocations.

One might think they're still needed for *module-specific* KLP
relocations.  If a to-be-patched module is loaded *after* its
corresponding KLP module is loaded, any corresponding KLP relocations
will be delayed until the to-be-patched module is loaded.  If any
special sections (.parainstructions, for example) rely on those
relocations, their initializations (apply_paravirt) need to be done
afterwards.  Thus the apparent need for arch_klp_init_object_loaded()
and its corresponding .klp.arch sections -- it allows some of the
special section initializations to be done at a later time.

But... if you look closer, that dependency between the special sections
and the module-specific KLP relocations doesn't actually exist in
reality.  Looking at the contents of the .altinstructions and
.parainstructions sections, there's not a realistic scenario in which a
KLP module's .altinstructions or .parainstructions section needs to
access a symbol in a to-be-patched module.  It might need to access a
local symbol or even a vmlinux symbol; but not another module's symbol.
When a special section needs to reference a local or vmlinux symbol, a
normal rela can be used instead of a KLP rela.

Since the special section initializations don't actually have any real
dependency on module-specific KLP relocations, .klp.arch and
arch_klp_init_object_loaded() no longer have a reason to exist.  So
remove them.

As Peter said much more succinctly:

  So the reason for .klp.arch was that .klp.rela.* stuff would overwrite
  paravirt instructions. If that happens you're doing it wrong. Those
  RELAs are core kernel, not module, and thus should've happened in
  .rela.* sections at patch-module loading time.

  Reverting this removes the two apply_{paravirt,alternatives}() calls
  from the late patching path, and means we don't have to worry about
  them when removing module_disable_ro().

[ jpoimboe: Rewrote patch description.  Tweaked klp_init_object_loaded()
	    error path. ]

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Joe Lawrence <joe.lawrence@redhat.com>
Acked-by: Miroslav Benes <mbenes@suse.cz>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
This commit is contained in:
Peter Zijlstra
2020-04-29 10:24:45 -05:00
committed by Jiri Kosina
parent 7c8e2bdd5f
commit 1d05334d28
5 changed files with 11 additions and 88 deletions

View File

@@ -14,8 +14,7 @@ This document outlines the Elf format requirements that livepatch modules must f
4. Livepatch symbols
4.1 A livepatch module's symbol table
4.2 Livepatch symbol format
5. Architecture-specific sections
6. Symbol table and Elf section access
5. Symbol table and Elf section access
1. Background and motivation
============================
@@ -298,17 +297,7 @@ Examples:
Note that the 'Ndx' (Section index) for these symbols is SHN_LIVEPATCH (0xff20).
"OS" means OS-specific.
5. Architecture-specific sections
=================================
Architectures may override arch_klp_init_object_loaded() to perform
additional arch-specific tasks when a target module loads, such as applying
arch-specific sections. On x86 for example, we must apply per-object
.altinstructions and .parainstructions sections when a target module loads.
These sections must be prefixed with ".klp.arch.$objname." so that they can
be easily identified when iterating through a patch module's Elf sections
(See arch/x86/kernel/livepatch.c for a complete example).
6. Symbol table and Elf section access
5. Symbol table and Elf section access
======================================
A livepatch module's symbol table is accessible through module->symtab.