PROBLEM: Kernel BUG in mfgpt_tick (cs5535-clockevt.c) on ALIX 2c3 - null call

[David Kozub]

Hi all,

booting up kernel 4.14-rc3 with CS5535_CLOCK_EVENT_SRC on an ALIX 2c3 
(http://pcengines.ch/alix2c3.htm) dies with:

[    2.313086] cs5535-smb cs5535-smb: SCx200 device 'CS5535 ACB0' registered
[    2.338711] cs5535-mfgpt cs5535-mfgpt: registered timer 0
[    2.355676] ledtrig-cpu: registered to indicate activity on CPUs
[    2.373745] cs5535-mfgpt cs5535-mfgpt: registered timer 1
[    2.389976] cs5535-clockevt: Registering MFGPT timer as a clock event, using IRQ 7
[    2.412687] BUG: unable to handle kernel NULL pointer dereference at   (null)
[    2.412698] IP:   (null)
[    2.412702] *pde = 00000000
[    2.412713] Oops: 0000 [#1]
[    2.412716] Modules linked in:
[    2.412732] CPU: 0 PID: 1 Comm: swapper Not tainted 4.14.0-rc3-humel-test17 #36
[    2.412739] task: c0010000 task.stack: c008e000
[    2.412744] EIP:   (null)
[    2.412749] EFLAGS: 00210093 CPU: 0
[    2.412758] EAX: c05fb8c0 EBX: c05fb880 ECX: 00000000 EDX: 0000620c
[    2.412769] ESI: c0009fd4 EDI: c014e14e EBP: c0009fac ESP: c0009fa8
[    2.412780]  DS: 007b ES: 007b FS: 0000 GS: 0000 SS: 0068
[    2.412790] CR0: 80050033 CR2: 00000000 CR3: 0064f000 CR4: 00000090
[    2.412793] Call Trace:
[    2.412800]  <IRQ>
[    2.412825]  ? mfgpt_tick+0x5d/0x81
[    2.412845]  __handle_irq_event_percpu+0x56/0xb6
[    2.412864]  ? handle_fasteoi_irq+0xf3/0xf3
[    2.412878]  handle_irq_event_percpu+0x17/0x3f
[    2.412892]  handle_irq_event+0x1d/0x29
[    2.412905]  handle_level_irq+0x57/0xc6
[    2.412924]  handle_irq+0x47/0x52
[    2.412929]  </IRQ>
[    2.412934]  <SOFTIRQ>
[    2.412949]  do_IRQ+0x32/0x9b
[    2.412963]  ? __irqentry_text_end+0x7/0x7
[    2.412976]  common_interrupt+0x2e/0x34
...

The problem seems to be in drivers/clocksource/cs5535-clockevt.c in 
mfgpt_tick() on line 132:
...
	cs5535_clockevent.event_handler(&cs5535_clockevent);
...
cs5535_clockevent.event_handler is null.

Adding some more traces I see mfgpt_tick() gets called before 
clockevents_config_and_register() finishes (invoked from 
cs5535_mfgpt_init() on line 178). So when mfgpt_tick() accessess the 
event_handler, it's NULL. Wrapping the event_handler call on line 132 in a 
null pointer check results in a working system.

The issue is present at least also in kernel 4.13.5. In kernel versions <= 
4.1-rc6 the cs5535_clockevent worked OK. Kernels >= 4.1-rc7 never booted 
at all on my ALIX 2c3 (last I tried 4.5, then gave up and tried 4.13.5 
recently), so I don't know when exactly this issue appeared.

My guess is that the timer interrupt is enabled too early. If I change the 
order of clockevents_config_and_register() and 
cs5535_mfgpt_write(cs5535_event_clock, MFGPT_REG_SETUP, val) in 
cs5535_mfgpt_init() like this:
...
	/* Set up the clock event */
	printk(KERN_INFO DRV_NAME
		": Registering MFGPT timer as a clock event, using IRQ %d\n",
		timer_irq);
	clockevents_config_and_register(&cs5535_clockevent, MFGPT_HZ,
					0xF, 0xFFFE);

	/* Set the clock scale and enable the event mode for CMP2 */
	val = MFGPT_SCALE | (3 << 8);
	cs5535_mfgpt_write(cs5535_event_clock, MFGPT_REG_SETUP, val);
...
the system boots and seems to be OK.


Best regards,
David