[PATCH] x86/mce: Initialize workqueues only once

[Borislav Petkov <bp@suse.de>]

On Wed, Jun 17, 2015 at 11:53:53PM +0000, Luck, Tony wrote:
> > if you want to give those changes a run, I've uploaded them here:
> >
> > git://git.kernel.org/pub/scm/linux/kernel/git/ras/ras.git#tip-ras
> 
> Latest experiments show that sometimes checking kventd_up() before calling schedule_work()
> helps ... but mostly only when I fake some early logs from low numbered  cpus.  I added some
> traces to the real case of a left-over fatal error and got this splat:

Here's the fix:

--
From: Borislav Petkov <bp@suse.de>
Subject: [PATCH] x86/mce: Initialize workqueues only once

96d98bfd0366 ("x86/mce: Don't use percpu workqueues") dropped the
per-CPU workqueues in the MCE code but left the initialization per-CPU.
This lead to early boot time splats (below) in the workqueues code
because we were overwriting the workqueue during INIT_WORK() on each new
CPU which would appear.

And since mcheck_cpu_init() happens very early, using an initcall to do
this one-time initialization doesn't fly. Using work->func as a check
whether the workqueue has been initialized already might break in the
future if someone changes ->func, so doesn't work either.

So let's have a simple static boolean flag to do that one-time work.

  mce: [Hardware Error]: Machine check events logged
  BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
  IP: [<ffffffff810980a1>] process_one_work+0x31/0x420
   PGD 0
  Oops: 0000 [#1] SMP
  Modules linked in:
  CPU: 36 PID: 263 Comm: kworker/36:0 Not tainted 4.1.0-rc8 #1
  Hardware name: Intel Corporation BRICKLAND/BRICKLAND, BIOS BRHSXSD1.86B.0065.R01.1505011640 05/01/2015
  task: ffff88181c284470 ti: ffff88181bd94000 task.ti: ffff88181bd94000
  RIP: 0010:[<ffffffff810980a1>] process_one_work+0x31/0x420
  RSP: 0000:ffff88181bd97e08  EFLAGS: 00010046
  RAX: 0000000fffffffe0 RBX: ffffffff81d0fa20 RCX: 0000000000000000
  RDX: 0000000fffffff00 RSI: ffffffff81d0fa20 RDI: ffff88181c2660c0
  RBP: ffff88181bd97e48 R08: ffff88181f416ec0 R09: ffff88181c284470
  R10: 0000000000000002 R11: ffffffff8109e5ac R12: ffff88181c2660c0
  R13: ffff88181f416ec0 R14: 0000000000000000 R15: ffff88181c2660f0
  			     ^^^^^^^^^^^^^^^^^

  27:   4c 0f 45 f2             cmovne %rdx,%r14
  2b:*  49 8b 46 08             mov    0x8(%r14),%rax           <-- trapping instruction
  2f:   44 8b b8 00 01 00 00    mov    0x100(%rax),%r15d

  ...

  Call Trace:
   worker_thread
   ? rescuer_thread
   kthread
   ? kthread_create_on_node
   ret_from_fork
   ? kthread_create_on_node
  Code: 48 89 e5 41 57 41 56 45 31 f6 41 55 41 54 49 89 fc 53 48 89 f3 48 83 ec 18 48 8b 06 4c 8b 6f 48 48 89 c2 30 d2 a8 04 4c 0f 45 f2 <49> 8b 46 08 44 8b b8 00 01 00 00 41 c1 ef 05 44 89 f8 83 e0 01
  RIP  [<ffffffff810980a1>] process_one_work
   RSP <ffff88181bd97e08>
  CR2: 0000000000000008
  ---[ end trace 8229a011b97532a0 ]---
  Kernel panic - not syncing: Fatal exception
  ---[ end Kernel panic - not syncing: Fatal exception

Signed-off-by: Borislav Petkov <bp@suse.de>
---
 arch/x86/kernel/cpu/mcheck/mce.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c
index 478f81a6d824..216d44d074df 100644
--- a/arch/x86/kernel/cpu/mcheck/mce.c
+++ b/arch/x86/kernel/cpu/mcheck/mce.c
@@ -1645,6 +1645,8 @@ void (*machine_check_vector)(struct pt_regs *, long error_code) =
  */
 void mcheck_cpu_init(struct cpuinfo_x86 *c)
 {
+	static bool __mce_init_once;
+
 	if (mca_cfg.disabled)
 		return;
 
@@ -1665,8 +1667,11 @@ void mcheck_cpu_init(struct cpuinfo_x86 *c)
 		return;
 	}
 
-	INIT_WORK(&mce_work, mce_process_work);
-	init_irq_work(&mce_irq_work, mce_irq_work_cb);
+	if (!__mce_init_once) {
+		INIT_WORK(&mce_work, mce_process_work);
+		init_irq_work(&mce_irq_work, mce_irq_work_cb);
+		__mce_init_once = 1;
+	}
 
 	machine_check_vector = do_machine_check;
 
-- 
2.3.5

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--