powerpc32/modules: Reserve a PLT entry for ftrace_regs_caller

Commit 7dfbfb87c243 ("powerpc/ftrace: Activate
HAVE_DYNAMIC_FTRACE_WITH_REGS on PPC32") made count_relocs()
undercount the PLT entries the loader
consumes: module_finalize_ftrace() now creates two trampolines
(mod->arch.tramp for ftrace_caller() and mod->arch.tramp_regs for
ftrace_regs_caller()), while count_relocs() reserves only one entry
for ftrace.

So the core .plt ends up one ppc_plt_entry (16 bytes) short. The
walk in do_plt_call() scans for a free slot with
"while (entry->jump[0])" and has no bound check, so once the last
reserved slot is used up, the walk steps past the end of the section
and patch_instruction() writes the 4-instruction trampoline into
whatever module memory follows the .plt.

Count the second entry, mirroring what module_64.c does for the same
pair of stubs.

Signed-off-by: Bradley Morgan <redacted>
---
 arch/powerpc/kernel/module_32.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/arch/powerpc/kernel/module_32.c b/arch/powerpc/kernel/module_32.c
index f930e3395a7f..61152deffc55 100644
--- a/arch/powerpc/kernel/module_32.c
+++ b/arch/powerpc/kernel/module_32.c

@@ -41,6 +41,9 @@ static unsigned int count_relocs(const Elf32_Rela *rela, unsigned int num)
 
 #ifdef CONFIG_DYNAMIC_FTRACE
 	_count_relocs++;	/* add one for ftrace_caller */
+#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
+	_count_relocs++;	/* add one for ftrace_regs_caller */
+#endif
 #endif
 	return _count_relocs;
 }

-- 
2.53.0

Le 23/09/2026 à 19:56, Bradley Morgan a écrit :

Commit 7dfbfb87c243 ("powerpc/ftrace: Activate
HAVE_DYNAMIC_FTRACE_WITH_REGS on PPC32") made count_relocs()
undercount the PLT entries the loader
consumes: module_finalize_ftrace() now creates two trampolines
(mod->arch.tramp for ftrace_caller() and mod->arch.tramp_regs for
ftrace_regs_caller()), while count_relocs() reserves only one entry
for ftrace.

So the core .plt ends up one ppc_plt_entry (16 bytes) short. The
walk in do_plt_call() scans for a free slot with
"while (entry->jump[0])" and has no bound check, so once the last
reserved slot is used up, the walk steps past the end of the section
and patch_instruction() writes the 4-instruction trampoline into
whatever module memory follows the .plt.

Count the second entry, mirroring what module_64.c does for the same
pair of stubs.

Signed-off-by: Bradley Morgan <redacted>

I would have had a preference for something similar to commit c12cfe9dee07 ("powerpc/module_64: Convert #ifdef to IS_ENABLED()")

Nevertheless,

Reviewed-by: Christophe Leroy (CS GROUP) <chleroy@kernel.org>