[PATCH] riscv: patch: remove lockdep assertion on lock text_mutex

[Palmer Dabbelt]

On Fri, 16 Apr 2021 19:35:32 PDT (-0700), changbin.du at gmail.com wrote:
> The function patch_insn_write() expects that the text_mutex is already
> held. There's a case that text_mutex is acquired by ftrace_run_update_code()
> under syscall context but then patch_insn_write() will be executed under the
> migration kthread context as we involves stop machine. So we should remove
> the assertion, or it can cause warning storm in kernel message.
>
> [  104.641978] ------------[ cut here ]------------
> [  104.642327] WARNING: CPU: 0 PID: 13 at arch/riscv/kernel/patch.c:63 patch_insn_write+0x166/0x17c
> [  104.643587] Modules linked in:
> [  104.644691] CPU: 0 PID: 13 Comm: migration/0 Not tainted 5.12.0-rc7-00067-g9cdbf6467424 #102
> [  104.644907] Hardware name: riscv-virtio,qemu (DT)
> [  104.645068] Stopper: multi_cpu_stop+0x0/0x17e <- 0x0
> [  104.645349] epc : patch_insn_write+0x166/0x17c
> [  104.645467]  ra : patch_insn_write+0x162/0x17c
> [  104.645534] epc : ffffffe0000059c6 ra : ffffffe0000059c2 sp : ffffffe002a33c70
> [  104.645580]  gp : ffffffe0019e5518 tp : ffffffe002a232c0 t0 : ffffffe01295e8a8
> [  104.645622]  t1 : 0000000000000001 t2 : 0000000000000000 s0 : ffffffe002a33cc0
> [  104.645675]  s1 : ffffffe000007f72 a0 : 0000000000000000 a1 : ffffffffffffffff
> [  104.645716]  a2 : 0000000000000001 a3 : 0000000000000000 a4 : 0000000000000001
> [  104.645757]  a5 : ffffffe0799e45c8 a6 : 00000000000ca097 a7 : 0000000000000000
> [  104.645798]  s2 : 0000000000000008 s3 : 0000000000000f72 s4 : ffffffe002a33ce0
> [  104.645839]  s5 : 0000000000000f7a s6 : 0000000000000003 s7 : 0000000000000003
> [  104.645880]  s8 : 0000000000000004 s9 : 0000000000000002 s10: 0000000000000000
> [  104.645920]  s11: 0000000000000002 t3 : 0000000000000001 t4 : ffffffe000c615c8
> [  104.645958]  t5 : 0000000000007fff t6 : 0000000000000380
> [  104.645998] status: 0000000000000100 badaddr: 0000000000000000 cause: 0000000000000003
> [  104.646081] Call Trace:
> [  104.646147] [<ffffffe0000059c6>] patch_insn_write+0x166/0x17c
> [  104.646280] [<ffffffe0000059ec>] patch_text_nosync+0x10/0x32
> [  104.646317] [<ffffffe000007d0c>] ftrace_update_ftrace_func+0x74/0xac
> [  104.646352] [<ffffffe0000d338a>] ftrace_modify_all_code+0x9c/0x144
> [  104.646387] [<ffffffe0000d3444>] __ftrace_modify_code+0x12/0x1c
> [  104.646420] [<ffffffe0000c4c2c>] multi_cpu_stop+0xa8/0x17e
> [  104.646451] [<ffffffe0000c4474>] cpu_stopper_thread+0xb2/0x156
> [  104.646489] [<ffffffe00003645e>] smpboot_thread_fn+0x102/0x1ea
> [  104.646524] [<ffffffe0000311b0>] kthread+0x132/0x148
> [  104.646556] [<ffffffe000002e02>] ret_from_exception+0x0/0x14
> [  104.646657] ---[ end trace ccf71babb9de4d5b ]---
> [  104.647444] ------------[ cut here ]------------
>
> Signed-off-by: Changbin Du <changbin.du at gmail.com>
> ---
>  arch/riscv/kernel/patch.c | 13 ++++++-------
>  1 file changed, 6 insertions(+), 7 deletions(-)
>
> diff --git a/arch/riscv/kernel/patch.c b/arch/riscv/kernel/patch.c
> index 0b552873a577..6d2ed9c15065 100644
> --- a/arch/riscv/kernel/patch.c
> +++ b/arch/riscv/kernel/patch.c
> @@ -49,19 +49,18 @@ static void patch_unmap(int fixmap)
>  }
>  NOKPROBE_SYMBOL(patch_unmap);
>
> +
> +/*
> + * Before reaching here, it was expected to lock the text_mutex
> + * already, so we don't need to give another lock here and could
> + * ensure that it was safe between each cores.
> + */
>  static int patch_insn_write(void *addr, const void *insn, size_t len)
>  {
>  	void *waddr = addr;
>  	bool across_pages = (((uintptr_t) addr & ~PAGE_MASK) + len) > PAGE_SIZE;
>  	int ret;
>
> -	/*
> -	 * Before reaching here, it was expected to lock the text_mutex
> -	 * already, so we don't need to give another lock here and could
> -	 * ensure that it was safe between each cores.
> -	 */
> -	lockdep_assert_held(&text_mutex);
> -
>  	if (across_pages)
>  		patch_map(addr + len, FIX_TEXT_POKE1);

Well, we definately need the lock -- otherwise concurrent callers will 
install conflicting maps for FIX_TEXT_POKE*, step on each other, and 
install the wrong patch.  I just sent out a patch to use an explicit 
lock here, as sharing text_mutex doesn't seem safe.