Skip to contentSkip to conversation

w1 fixes for broken subsystem

PATCH In review Hans-Frieder Vogt 3 participants 3 messages Oct 6, 2013

lore
Share & tools
lore
#0 lore raw

cover w1 fixes for broken subsystem

In recent kernels the 1-wire kernel modules do not work any more. The module wire got broken in 3.11-rc1 and 3.12-rc1 introduced a further problem:

1: 3.11-rc1 introduced the automatic loading of slave modules. This blocks modprobe, because request_module is called with a locked mutex of the w1 master.
Loading of the slave module calls w1_register_family, which calls in turn w1_reconnect_slaves and there mutex_lock(&dev->mutex) will wait forever....

2: 3.12-rc1 introduced w1_bus_notify which assumes that all slaves have fops defined. This is not the case for e.g. w1_smem and maybe also breaks custom written modules.

This very small patch set is split in 2 parts:

PATCH 1/2: w1 - fix fops in w1_bus_notify
fixes problem 2 above

PATCH 2/2: w1 - call request_module with w1 master mutex unlocked fixes problem 1 above

Please include these patches in 3.12-rc4 and in 3.11 stable.

Best regards,

Hans-Frieder Vogt e-mail: hfvogt <at> gmx .dot. net

Acked-by on the v1 series
#1 lore raw

Hi

06.10.2013, 23:13, "Hans-Frieder Vogt" <[email protected]>:

Quoted 2 lines

In recent kernels the 1-wire kernel modules do not work any more. The module wire got broken in 3.11-rc1 and 3.12-rc1 introduced a further problem:

Looks like I somehow missed it and bugs got in

Quoted 5 lines

1: 3.11-rc1 introduced the automatic loading of slave modules. This blocks modprobe, because request_module is called with a locked mutex of the w1 master.
Loading of the slave module calls  w1_register_family, which calls in turn w1_reconnect_slaves and there mutex_lock(&dev->mutex) will wait forever....

That's correct

Quoted 3 lines

2: 3.12-rc1 introduced w1_bus_notify which assumes that all slaves have fops defined. This is not the case for e.g. w1_smem and maybe also breaks custom written modules.

Isn't fops are supposed to be in the slave device?
Although they look meaningless for most of devices, I would introduce them everywhere...

But anyway, your patches fix those problems, thanks a lot. Greg, please queue them into appropriate trees.

#2 lore raw
Quoted 20 lines · Evgeniy Polyakov

Hi

06.10.2013, 23:13, "Hans-Frieder Vogt" <[email protected]>:

Quoted 2 lines

In recent kernels the 1-wire kernel modules do not work any more. The module wire got broken in 3.11-rc1 and 3.12-rc1 introduced a further problem:

Looks like I somehow missed it and bugs got in

Quoted 5 lines

1: 3.11-rc1 introduced the automatic loading of slave modules. This blocks modprobe, because request_module is called with a locked mutex of the w1 master.
Loading of the slave module calls  w1_register_family, which calls in turn w1_reconnect_slaves and there mutex_lock(&dev->mutex) will wait forever....

That's correct

Quoted 3 lines

2: 3.12-rc1 introduced w1_bus_notify which assumes that all slaves have fops defined. This is not the case for e.g. w1_smem and maybe also breaks custom written modules.

Quoted 7 lines

Isn't fops are supposed to be in the slave device?
Although they look meaningless for most of devices, I would introduce them everywhere...

But anyway, your patches fix those problems, thanks a lot. Greg, please queue them into appropriate trees.

Thanks for reviewing this, I'll queue them up in a bit.

greg k-h