Re: [PATCH] riscv: ftrace: Properly acquire text_mutex to fix a race condition

[Palmer Dabbelt]

On Tue, 08 Jul 2025 01:33:48 PDT (-0700), alexgh...@rivosinc.com wrote:
As reported by lockdep, some patching was done without acquiring
text_mutex, so there could be a race when mapping the page to patch
since we use the same fixmap entry.
Reported-by: Han Gao <rabenda...@gmail.com>
Reported-by: Vivian Wang <wangruik...@iscas.ac.cn>
Reported-by: Yao Zi <zi...@disroot.org>
Closes: https://lore.kernel.org/linux-riscv/agodmpq7tgind...@pie.lan/
Tested-by: Yao Zi <zi...@disroot.org>
Signed-off-by: Alexandre Ghiti <alexgh...@rivosinc.com>
---
 arch/riscv/kernel/ftrace.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

diff --git a/arch/riscv/kernel/ftrace.c b/arch/riscv/kernel/ftrace.c
index 
4c6c24380cfd9d6c51f0e4340cd674160b83a610..22e7bdf8de2b6ca950cf2c8b734bc82ea46ba8bf
 100644
--- a/arch/riscv/kernel/ftrace.c
+++ b/arch/riscv/kernel/ftrace.c
@@ -14,6 +14,16 @@
 #include <asm/text-patching.h>

 #ifdef CONFIG_DYNAMIC_FTRACE
+void ftrace_arch_code_modify_prepare(void)
+{
+       mutex_lock(&text_mutex);
+}
+
+void ftrace_arch_code_modify_post_process(void)
+{
+       mutex_unlock(&text_mutex);
+}

IIRC there's a reason we don't do it this way, we had (or had tried to) 
have it before.  It's been a while, though, and I'm just having some a 
coffee so may I'm just wrong...

 unsigned long ftrace_call_adjust(unsigned long addr)
 {
        if (IS_ENABLED(CONFIG_DYNAMIC_FTRACE_WITH_CALL_OPS))
@@ -29,10 +39,8 @@ unsigned long arch_ftrace_get_symaddr(unsigned long 
fentry_ip)

 void arch_ftrace_update_code(int command)
 {
-       mutex_lock(&text_mutex);
        command |= FTRACE_MAY_SLEEP;
        ftrace_modify_all_code(command);
-       mutex_unlock(&text_mutex);
        flush_icache_all();
 }

@@ -149,6 +157,8 @@ int ftrace_init_nop(struct module *mod, struct dyn_ftrace 
*rec)
        unsigned int nops[2], offset;
        int ret;

+       guard(mutex)(&text_mutex);
+
        ret = ftrace_rec_set_nop_ops(rec);
        if (ret)
                return ret;
@@ -157,9 +167,7 @@ int ftrace_init_nop(struct module *mod, struct dyn_ftrace 
*rec)
        nops[0] = to_auipc_t0(offset);
        nops[1] = RISCV_INSN_NOP4;

-       mutex_lock(&text_mutex);
        ret = patch_insn_write((void *)pc, nops, 2 * MCOUNT_INSN_SIZE);
-       mutex_unlock(&text_mutex);

        return ret;
 }

---
base-commit: d7b8f8e20813f0179d8ef519541a3527e7661d3a
change-id: 20250708-alex-fixes-1e719b9899f3

Best regards,