linux-stable.git/drivers/pinctrl/core.c, branch linux-3.19.y

pinctrl: Fix two deadlocks

2015-01-14T13:20:33+00:00

This patch is to fix two deadlock cases.
Deadlock 1:
CPU #1
 pinctrl_register-> pinctrl_get ->
 create_pinctrl
 (Holding lock pinctrl_maps_mutex)
 -> get_pinctrl_dev_from_devname
 (Trying to acquire lock pinctrldev_list_mutex)
CPU #0
 pinctrl_unregister
 (Holding lock pinctrldev_list_mutex)
 -> pinctrl_put ->> pinctrl_free ->
 pinctrl_dt_free_maps -> pinctrl_unregister_map
 (Trying to acquire lock pinctrl_maps_mutex)

Simply to say
CPU#1 is holding lock A and trying to acquire lock B,
CPU#0 is holding lock B and trying to acquire lock A.

Deadlock 2:
CPU #3
 pinctrl_register-> pinctrl_get ->
 create_pinctrl
 (Holding lock pinctrl_maps_mutex)
 -> get_pinctrl_dev_from_devname
 (Trying to acquire lock pinctrldev_list_mutex)
CPU #2
 pinctrl_unregister
 (Holding lock pctldev->mutex)
 -> pinctrl_put ->> pinctrl_free ->
 pinctrl_dt_free_maps -> pinctrl_unregister_map
 (Trying to acquire lock pinctrl_maps_mutex)
CPU #0
 tegra_gpio_request
 (Holding lock pinctrldev_list_mutex)
 -> pinctrl_get_device_gpio_range
 (Trying to acquire lock pctldev->mutex)

Simply to say
CPU#3 is holding lock A and trying to acquire lock D,
CPU#2 is holding lock B and trying to acquire lock A,
CPU#0 is holding lock D and trying to acquire lock B.

Cc: Stable 
Signed-off-by: Jim Lin 
Signed-off-by: Linus Walleij

pinctrl: avoid duplicated calling enable_pinmux_setting for a pin

2014-07-11T12:08:26+00:00

What the patch does:
1. Call pinmux_disable_setting ahead of pinmux_enable_setting
  each time pinctrl_select_state is called
2. Remove the HW disable operation in pinmux_disable_setting function.
3. Remove the disable ops in struct pinmux_ops
4. Remove all the disable ops users in current code base.

Notes:
1. Great thanks for the suggestion from Linus, Tony Lindgren and
   Stephen Warren and Everyone that shared comments on this patch.
2. The patch also includes comment fixes from Stephen Warren.

The reason why we do this:
1. To avoid duplicated calling of the enable_setting operation
   without disabling operation inbetween which will let the pin
   descriptor desc->mux_usecount increase monotonously.
2. The HW pin disable operation is not useful for any of the
   existing platforms.
   And this can be used to avoid the HW glitch after using the
   item #1 modification.

In the following case, the issue can be reproduced:
1. There is a driver that need to switch pin state dynamically,
   e.g. between "sleep" and "default" state
2. The pin setting configuration in a DTS node may be like this:

  component a {
	pinctrl-names = "default", "sleep";
	pinctrl-0 = <&a_grp_setting &c_grp_setting>;
	pinctrl-1 = <&b_grp_setting &c_grp_setting>;
  }

  The "c_grp_setting" config node is totally identical, maybe like
  following one:

  c_grp_setting: c_grp_setting {
	pinctrl-single,pins = ;
  }

3. When switching the pin state in the following official pinctrl
   sequence:
	pin = pinctrl_get();
	state = pinctrl_lookup_state(wanted_state);
	pinctrl_select_state(state);
	pinctrl_put();

Test Result:
1. The switch is completed as expected, that is: the device's
   pin configuration is changed according to the description in the
   "wanted_state" group setting
2. The "desc->mux_usecount" of the corresponding pins in "c_group"
   is increased without being decreased, because the "desc" is for
   each physical pin while the setting is for each setting node
   in the DTS.
   Thus, if the "c_grp_setting" in pinctrl-0 is not disabled ahead
   of enabling "c_grp_setting" in pinctrl-1, the desc->mux_usecount
   will keep increasing without any chance to be decreased.

According to the comments in the original code, only the setting,
in old state but not in new state, will be "disabled" (calling
pinmux_disable_setting), which is correct logic but not intact. We
still need consider case that the setting is in both old state
and new state. We can do this in the following two ways:

1. Avoid to "enable"(calling pinmux_enable_setting) the "same pin
   setting" repeatedly
2. "Disable"(calling pinmux_disable_setting) the "same pin setting",
   actually two setting instances, ahead of enabling them.

Analysis:
1. The solution #2 is better because it can avoid too much
   iteration.
2. If we disable all of the settings in the old state and one of
   the setting(s) exist in the new state, the pins mux function
   change may happen when some SoC vendors defined the
   "pinctrl-single,function-off"
   in their DTS file.
   old_setting => disabled_setting => new_setting.
3. In the pinmux framework, when a pin state is switched, the
   setting in the old state should be marked as "disabled".

Conclusion:
1. To Remove the HW disabling operation to above the glitch mentioned
   above.
2. Handle the issue mentioned above by disabling all of the settings
   in old state and then enable the all of the settings in new state.

Signed-off-by: Fan Wu 
Acked-by: Stephen Warren 
Acked-by: Patrice Chotard 
Acked-by: Heiko Stuebner 
Acked-by: Maxime Coquelin 
Signed-off-by: Linus Walleij

pinctrl: allows not to define the get_group_pins operation

2014-04-22T12:47:02+00:00

When using a group only pinctrl driver, which does not have any
information on the pins it is useless to define a get_group_pins
always returning an empty list of pins.

When not using get_group_pin[1], a driver must implement it so
pins = NULL and num_pins = 0. This patch makes it the default
behaviour if not defined in the pinctrl driver when used in
pinmux enable and disable funtions and in pinctrl_groups_show.

It also adds a check in pinctrl_get_group_pins and return -EINVAL if
not defined. This function is called in the gpiolib when adding when
pingroup range. It cannot be used if no group is defined, so this seams
reasonable.

[1] get_group_pin(struct pinctrl_dev *pctldev,
		  unsigned selector,
		  const unsigned **pins,
		  unsigned *num_pins);

Signed-off-by: Antoine TÃ©nart 
Signed-off-by: Linus Walleij

pinctrl: do not init debugfs entries for unimplemented functionalities

2014-02-06T12:48:17+00:00

Commit c420619 "pinctrl: pinconf: remove checks on ops->pin_config_get"
removed the check on (ops != NULL) when performing pinconf_pins_show() or
pinconf_groups_show(). As these entries are always enabled, even if
pinconf is not supported, reading will result in an oops due to NULL
ops.

Instead of checking for ops, remove the corresponding debugfs entries if
pinconf and/or pinmux are not implemented.

Tested on OMAP3 (pinctrl-single).

Cc: stable@vger.kernel.org
Signed-off-by: Florian Vaussard 
Signed-off-by: Linus Walleij

pinctrl: protect pinctrl_list add

2014-02-04T20:59:45+00:00

We have few fedora bug reports about list corruption on pinctrl,
for example:
https://bugzilla.redhat.com/show_bug.cgi?id=1051918

Most likely corruption happen due lack of protection of pinctrl_list
when adding new nodes to it. Patch corrects that.

Fixes: 42fed7ba44e ("pinctrl: move subsystem mutex to pinctrl_dev struct")
Cc: stable@vger.kernel.org
Signed-off-by: Stanislaw Gruszka 
Acked-by: Stephen Warren 
Signed-off-by: Linus Walleij

pinctrl/gpio: non-linear GPIO ranges accesible from gpiolib

2013-10-16T13:33:50+00:00

This patch adds the infrastructure required to register non-linear gpio
ranges through gpiolib and the standard GPIO device tree bindings.

Signed-off-by: Christian Ruppert 
Signed-off-by: Linus Walleij

Merge tag 'v3.11-rc7' into devel

2013-08-29T07:46:30+00:00

Merged in this to avoid conflicts with the big locking fixes
from upstream.

Signed-off-by: Linus Walleij 

Conflicts:
	drivers/pinctrl/pinctrl-sunxi.c

pinctrl: core: Add proper mutex lock in pinctrl_request_gpio

2013-08-28T11:13:05+00:00

This one is missed in commit 42fed7ba "pinctrl: move subsystem mutex to
pinctrl_dev struct".

I think this fixes the race between pin_free() and pin_request() calls.
It protects accessing the members of pctldev->desc.
(e.g. update desc->mux_usecount, desc->gpio_owner, desc->mux_owner, etc)
Current code grabs pctldev->mutex before calling pinmux_free_gpio(),
but did not grab the mutex while calling pinmux_request_gpio().

Signed-off-by: Axel Lin 
Signed-off-by: Linus Walleij

pinctrl: core: Remove unnecessary test for desc->name

2013-08-23T06:56:29+00:00

The implementation in pinctrl_register_one_pin() ensures pindesc->name is always
not NULL before insert the pindesc to radix tree.
If the desc return from pin_desc_get is not NULL, desc->name is always not NULL.

Signed-off-by: Axel Lin 
Signed-off-by: Linus Walleij

pinctrl: core: Hold pctldev->mutex mutex lock while traversing gpio_ranges list

2013-08-23T06:56:28+00:00

Hold pctldev->mutex mutex_lock when traverse the list.

Signed-off-by: Axel Lin 
Signed-off-by: Linus Walleij