Re: [PATCH] staging: Platform device tester - Allow removal

Russell King - ARM Linux [off-list ref] writes:

On Tue, Aug 13, 2013 at 09:42:27PM +0300, Pantelis Antoniou wrote:

quoted

But creation just crashes.

quoted

root@beaglebone:/sys/bus/platform/drivers/omap_i2c# echo 4819c000.i2c >bind   
[  145.053929] Unable to handle kernel NULL pointer dereference at virtual address 00000001
[  145.062651] pgd = ca8c0000
[  145.065507] [00000001] *pgd=8f437831, *pte=00000000, *ppte=00000000
[  145.072163] Internal error: Oops: 17 [#1] SMP ARM
[  145.077105] Modules linked in: ipv6 autofs4
[  145.081537] CPU: 0 PID: 301 Comm: sh Not tainted 3.11.0-rc5-00125-g3b988fb #146
[  145.089222] task: cf5b5580 ti: cf464000 task.ti: cf464000
[  145.094918] PC is at omap_i2c_runtime_suspend+0x10/0xb4
[  145.100408] LR is at omap_i2c_runtime_suspend+0x8/0xb4

Well then, what you have here is a bug in the way that OMAP re-uses the
platform device stuff - and that _does_ need fixing somehow.

The problem is that i2c_dev->regs is NULL at the point where
pm_runtime_set_autosuspend_delay() is called, which as autosuspend
is still set from the time that the device was unbound, it means that
this triggers an immediate suspend.

I suspect that:

        pm_runtime_set_autosuspend_delay(dev->dev, OMAP_I2C_PM_TIMEOUT);
        pm_runtime_use_autosuspend(dev->dev);

should be moved later in the probe function, after the switch (scheme)
block.

That's correct.  

At a minimum, it should be after the _get_sync() call.  Without any
users (callers of pm_runtime_get*) the _set_autosuspend_delay() call
will try to rpm_idle().

Kevin