powerpc64/ftrace: workaround clang recording GEP in __patchable_function_entries

Support for -fpatchable-function-entry on ppc64le was added in Clang
with [1]. However, when no prefix NOPs are specified - as is the case
with CONFIG_PPC_FTRACE_OUT_OF_LINE - the first NOP is emitted at LEP,
but Clang records the Global Entry Point (GEP) unlike GCC which does
record the Local Entry Point (LEP). Issue [2] has been raised to align
Clang's behavior with GCC. As a temporary workaround to ensure ftrace
initialization works as expected with Clang, derive the LEP using
ppc_function_entry() for kernel symbols and by looking for the below
module GEP sequence for module addresses, until [2] is resolved:

	ld	r2, -8(r12)
	add	r2, r2, r12

[1] https://github.com/llvm/llvm-project/pull/151569
[2] https://github.com/llvm/llvm-project/issues/163706

Signed-off-by: Hari Bathini <hbathini@linux.ibm.com>
Tested-by: Venkat Rao Bagalkote <venkat88@linux.ibm.com>
Signed-off-by: Madhavan Srinivasan <maddy@linux.ibm.com>
Link: https://patch.msgid.link/20260127084926.34497-4-hbathini@linux.ibm.com

diff --git a/arch/powerpc/kernel/trace/ftrace.c b/arch/powerpc/kernel/trace/ftrace.c
index 841d077e28251abdff3b08d095d52cb4a31916d5..1b2f293e7dcb57b2ceafcf1699e8482802e1d704 100644 (file)
--- a/arch/powerpc/kernel/trace/ftrace.c
+++ b/arch/powerpc/kernel/trace/ftrace.c
@@ -37,11 +37,29 @@ unsigned long ftrace_call_adjust(unsigned long addr)
        if (addr >= (unsigned long)__exittext_begin && addr < (unsigned long)__exittext_end)
                return 0;
 
-       if (IS_ENABLED(CONFIG_ARCH_USING_PATCHABLE_FUNCTION_ENTRY) &&
-           !IS_ENABLED(CONFIG_PPC_FTRACE_OUT_OF_LINE)) {
-               addr += MCOUNT_INSN_SIZE;
-               if (IS_ENABLED(CONFIG_DYNAMIC_FTRACE_WITH_CALL_OPS))
+       if (IS_ENABLED(CONFIG_ARCH_USING_PATCHABLE_FUNCTION_ENTRY)) {
+               if (!IS_ENABLED(CONFIG_PPC_FTRACE_OUT_OF_LINE)) {
                        addr += MCOUNT_INSN_SIZE;
+                       if (IS_ENABLED(CONFIG_DYNAMIC_FTRACE_WITH_CALL_OPS))
+                               addr += MCOUNT_INSN_SIZE;
+               } else if (IS_ENABLED(CONFIG_CC_IS_CLANG) && IS_ENABLED(CONFIG_PPC64)) {
+                       /*
+                        * addr points to global entry point though the NOP was emitted at local
+                        * entry point due to https://github.com/llvm/llvm-project/issues/163706
+                        * Handle that here with ppc_function_entry() for kernel symbols while
+                        * adjusting module addresses in the else case, by looking for the below
+                        * module global entry point sequence:
+                        *      ld    r2, -8(r12)
+                        *      add   r2, r2, r12
+                        */
+                       if (is_kernel_text(addr) || is_kernel_inittext(addr))
+                               addr = ppc_function_entry((void *)addr);
+                       else if ((ppc_inst_val(ppc_inst_read((u32 *)addr)) ==
+                                 PPC_RAW_LD(_R2, _R12, -8)) &&
+                                (ppc_inst_val(ppc_inst_read((u32 *)(addr+4))) ==
+                                 PPC_RAW_ADD(_R2, _R2, _R12)))
+                               addr += 8;
+               }
        }
 
        return addr;