Skip to contentSkip to conversation

pinctrl: pinmux: Don't free pins requested by other devices

PATCH Superseded Sonic Zhang 3 participants 5 messages Jul 17, 2013 – Aug 8, 2013 Pin control

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pinctrl: pinmux: Don't free pins requested by other devices

From: Sonic Zhang <[email protected]>

in pinmux_disable_setting after current device fails to request the same pins.

---
 drivers/pinctrl/pinmux.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
drivers/pinctrl/pinmux.c +5 -4
index 88cc509..9ebcf3b 100644
@@ -482,13 +482,14 @@ void pinmux_disable_setting(struct pinctrl_setting const *setting)
 				 pins[i]);
 			continue;
 		}
+		/* And release the pins */
+		if (desc->mux_usecount &&
+			!strcmp(desc->mux_owner, setting->dev_name))
+			pin_free(pctldev, pins[i], NULL);
+
 		desc->mux_setting = NULL;
 	}
 
-	/* And release the pins */
-	for (i = 0; i < num_pins; i++)
-		pin_free(pctldev, pins[i], NULL);
-
 	if (ops->disable)
 		ops->disable(pctldev, setting->data.mux.func, setting->data.mux.group);
 }
Signature · 1 line
1.8.2.3
#1 lore raw

On Wed, Jul 17, 2013 at 7:31 AM, Sonic Zhang <[email protected]> wrote:

I'd like Stephen and Axel to have a look at this as well...

Quoted 6 lines

From: Sonic Zhang <[email protected]>

in pinmux_disable_setting after current device fails to request the same pins.

I don't quite understand the patch. Can you provide more context?

Quoted 10 lines
drivers/pinctrl/pinmux.c +2 -0
index 88cc509..9ebcf3b 100644
@@ -482,13 +482,14 @@ void pinmux_disable_setting(struct pinctrl_setting const *setting)
                                 pins[i]);
                        continue;
                }
+               /* And release the pins */
+               if (desc->mux_usecount &&

drivers/pinctrl/pinmux.c void pinmux_disable_setting(struct pinctrl_setting const *setting) · line 491

Quoted 10 lines
drivers/pinctrl/pinmux.c +3 -4
+                       !strcmp(desc->mux_owner, setting->dev_name))
+                       pin_free(pctldev, pins[i], NULL);
+
                desc->mux_setting = NULL;
        }
 
-       /* And release the pins */
-       for (i = 0; i < num_pins; i++)
-               pin_free(pctldev, pins[i], NULL);
-

For pinmux_disable_setting() to inspect desc->mux_usecount seems assymetric. This is something pin_free() should do, shouldn't it?

Should not this codepath be kept and a change made inside pin_free() for the check above instead?

Yours,
Linus Walleij

#2 lore raw
Quoted 4 lines · Linus Walleij

On Wed, Jul 17, 2013 at 7:31 AM, Sonic Zhang <[email protected]> wrote:

I'd like Stephen and Axel to have a look at this as well...

Quoted 8 lines
Quoted 6 lines

From: Sonic Zhang <[email protected]>

in pinmux_disable_setting after current device fails to request the same pins.

I don't quite understand the patch. Can you provide more context?

Yes, the commit description needs to describe the problem this solves.

I'm *guessing* the issue is:

Something tries to enable a new mux setting on some pins. One of those pins is already owned by something else. So, applying the current setting fails. So, pinctrl core attempts to unapply the partially applied setting. This ends up incorrectly over-writing the conflicting ownership of the pins with NULL, and hence forgetting about it.

I think a better change would be something more along the lines of:

for (i = 0; i < num_pins; i++)
+ if (this_device_owns_pin(pins[i])
pin_free(pctldev, pins[i], NULL);

?

Where this_device_owns_pin() might be someting like:

	desc->owning_setting == setting

(which would be a new field that needed to be assigned during pinmux_enable_setting).

Or perhaps the strcmp() is fine.

Quoted 26 lines
Quoted 10 lines
drivers/pinctrl/pinmux.c +2 -0
index 88cc509..9ebcf3b 100644
@@ -482,13 +482,14 @@ void pinmux_disable_setting(struct pinctrl_setting const *setting)
                                 pins[i]);
                        continue;
                }
+               /* And release the pins */
+               if (desc->mux_usecount &&

drivers/pinctrl/pinmux.c void pinmux_disable_setting(struct pinctrl_setting const *setting) · line 491

Quoted 10 lines
drivers/pinctrl/pinmux.c +3 -4
+                       !strcmp(desc->mux_owner, setting->dev_name))
+                       pin_free(pctldev, pins[i], NULL);
+
                desc->mux_setting = NULL;
        }
 
-       /* And release the pins */
-       for (i = 0; i < num_pins; i++)
-               pin_free(pctldev, pins[i], NULL);
-

For pinmux_disable_setting() to inspect desc->mux_usecount seems assymetric. This is something pin_free() should do, shouldn't it?

Should not this codepath be kept and a change made inside pin_free() for the check above instead?

#3 lore raw

Hi Stephen,

Quoted 32 lines · Stephen Warren
Quoted 4 lines · Linus Walleij

On Wed, Jul 17, 2013 at 7:31 AM, Sonic Zhang <[email protected]> wrote:

I'd like Stephen and Axel to have a look at this as well...

Quoted 8 lines
Quoted 6 lines

From: Sonic Zhang <[email protected]>

in pinmux_disable_setting after current device fails to request the same pins.

I don't quite understand the patch. Can you provide more context?

Yes, the commit description needs to describe the problem this solves.

I'm *guessing* the issue is:

Something tries to enable a new mux setting on some pins. One of those pins is already owned by something else. So, applying the current setting fails. So, pinctrl core attempts to unapply the partially applied setting. This ends up incorrectly over-writing the conflicting ownership of the pins with NULL, and hence forgetting about it.

I think a better change would be something more along the lines of:

for (i = 0; i < num_pins; i++)
+ if (this_device_owns_pin(pins[i])

                pin_free(pctldev, pins[i], NULL);

?

Quoted 9 lines

Where this_device_owns_pin() might be someting like:

        desc->owning_setting == setting

(which would be a new field that needed to be assigned during pinmux_enable_setting).

Or perhaps the strcmp() is fine.

You are right. One peripheral may share part of its pins with the 2nd peripheral and the other pins with the 3rd. If it requests all pins when part of them has already be requested and owned by the 2nd peripheral, this request fails and pinmux_disable_setting() is called. The pinmux_disable_setting() frees all pins of the first peripheral without checking if the pin is owned by itself or the 2nd, which results in the malfunction of the 2nd peripheral driver.

I am fine to compare owner's pinctrl_setting structure other than name string.

Regards,

Sonic

#4 lore raw
In reply to Linus Walleij

Hi Linus,

Quoted 37 lines · Linus Walleij

On Wed, Jul 17, 2013 at 7:31 AM, Sonic Zhang <[email protected]> wrote:

I'd like Stephen and Axel to have a look at this as well...

Quoted 6 lines

From: Sonic Zhang <[email protected]>

in pinmux_disable_setting after current device fails to request the same pins.

I don't quite understand the patch. Can you provide more context?

Quoted 10 lines
drivers/pinctrl/pinmux.c +2 -0
index 88cc509..9ebcf3b 100644
@@ -482,13 +482,14 @@ void pinmux_disable_setting(struct pinctrl_setting const *setting)
                                 pins[i]);
                        continue;
                }
+               /* And release the pins */
+               if (desc->mux_usecount &&

drivers/pinctrl/pinmux.c void pinmux_disable_setting(struct pinctrl_setting const *setting) · line 491

Quoted 10 lines
drivers/pinctrl/pinmux.c +3 -4
+                       !strcmp(desc->mux_owner, setting->dev_name))
+                       pin_free(pctldev, pins[i], NULL);
+
                desc->mux_setting = NULL;
        }
 
-       /* And release the pins */
-       for (i = 0; i < num_pins; i++)
-               pin_free(pctldev, pins[i], NULL);
-

For pinmux_disable_setting() to inspect desc->mux_usecount seems assymetric. This is something pin_free() should do, shouldn't it?

Quoted 3 lines

Should not this codepath be kept and a change made inside pin_free() for the check above instead?

You can't move this codepath into pin_free(), because the pointer to structure pinctrl_setting is not passed through pin_free(). But yes, checking desc->mux_usecount is not necessary here, because pin_free() has already handled that.

I will remove desc->mux_usecount checking in next patch.

Regards,

Sonic