Le 26/11/2021 à 08:39, Christophe Leroy a écrit :
Commit 8b8a8f0ab3f5 ("powerpc/code-patching: Improve verification of
patchability") introduced a stricter verification of the patched
area by checking it is proper kernel text.
But as least two usages of patch_instruction() fall outside:
- Code patching selftests, which use stack and vmalloc space.
- Ftrace
So for the time being, partially revert commit 8b8a8f0ab3f5 and add
a onetime warning:
Running code patching self-tests ...
patch_instruction() called on invalid text address 0xe1011e58 from
test_code_patching+0x34/0xd6c
Reported-by: Sachin Sant <sach...@linux.vnet.ibm.com>
Reported-by: Stephen Rothwell <s...@canb.auug.org.au>
Cc: Nicholas Piggin <npig...@gmail.com>
Fixes: 8b8a8f0ab3f5 ("powerpc/code-patching: Improve verification of
patchability")
Signed-off-by: Christophe Leroy <christophe.le...@csgroup.eu>
---
arch/powerpc/lib/code-patching.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/arch/powerpc/lib/code-patching.c b/arch/powerpc/lib/code-patching.c
index 1dd636a85cc1..c87eea773930 100644
--- a/arch/powerpc/lib/code-patching.c
+++ b/arch/powerpc/lib/code-patching.c
@@ -190,9 +190,13 @@ static int do_patch_instruction(u32 *addr, struct ppc_inst
instr)
int patch_instruction(u32 *addr, struct ppc_inst instr)
{
/* Make sure we aren't patching a freed init section */
- if (!kernel_text_address((unsigned long)addr))
+ if (system_state >= SYSTEM_FREEING_INITMEM &&
init_section_contains(addr, 4))
return 0;
+ if (!kernel_text_address((unsigned long)addr))
+ pr_warn_once("%s() called on invalid text address 0x%p from
%pS\n",
+ __func__, addr, __builtin_return_address(0));
+
May it be better to use pr_warn_ratelimited() instead in order to catch
more than the first occurence ?
return do_patch_instruction(addr, instr);
}
NOKPROBE_SYMBOL(patch_instruction);
