fix sysfs showing the correct C-states after AC->DC and DC->AC transitions

Hello,

I do have a Samsung NC20 netbook which provides the C-states C1 and C2 to the OS when connected to AC, and additionally provides the C3 C-state when disconnected from AC. With the current kernels I have these two problems (regressions):

1. The number of C-states shown in sysfs is fixed to the number of C-states present at boot.
If I boot with AC connected, I always only see the C-states up to C2 even if I disconnect AC.

The reason is commit 130a5f692425e6237229598a8624da0a247f33d5 "ACPI / cpuidle: remove dev->state_count setting". It removes the update of dev->state_count, but sysfs uses exactly this variable to show the C-states.

The fix is to use drv->state_count in sysfs. As this is currently the last user of dev->state_count, this variable can be completely removed. And this is exactly what Bartlomiej Zolnierkiewicz's not yet applied patch "[PATCH v2 9/9] cpuidle: remove state_count field from struct cpuidle_device"
http://marc.info/?m=138756533317836 does.

Long story short: That patch fixes problem 1 for me.

2. After a AC->DC or DC->AC transition, the name and description of the POLLING C-state become "<null>".

Here, the reason is commit d7c7f103262bc2248548ed0e113e916e843c4eeb "cpuidle: don't call poll_idle_init() for every cpu". It only calls poll_idle_init() during cpuidle_register_driver() instead of cpuidle_enable_device() and thus does not re-initialize the fields of drv->states[0] after acpi_processor_setup_cpuidle_states() cleared them.

Here, the fix is to _not_ clear drv->states[0] in acpi_processor_setup_cpuidle_states(). For this purpose I created a small patch.

For your convenience, both of the fixing patches are attached to this mail. The first one is necessary for all kernels since 3.14, the second one for all kernels since 3.13. So I'd propose to push both of them also to the necessary stable kernels.

Kind regards,
Thomas

From: Bartlomiej Zolnierkiewicz <[email protected]>
Date: Tue, 31 Mar 2015 20:15:09 +0200
Subject: [PATCH 1/2] cpuidle: remove state_count field from struct
cpuidle_device

dev->state_count is now always equal to drv->state_count so it can be removed.

---
 drivers/cpuidle/cpuidle.c | 3 ---
 drivers/cpuidle/sysfs.c   | 5 +++--
 include/linux/cpuidle.h   | 1 -
 3 files changed, 3 insertions(+), 6 deletions(-)

index 080bd2d..7a73a27 100644
@@ -330,9 +330,6 @@ int cpuidle_enable_device(struct cpuidle_device *dev)
 	if (!dev->registered)
 		return -EINVAL;
 
-	if (!dev->state_count)
-		dev->state_count = drv->state_count;
-
 	ret = cpuidle_add_device_sysfs(dev);
 	if (ret)
 		return ret;

index 97c5903..832a2c3 100644
@@ -401,7 +401,7 @@ static int cpuidle_add_state_sysfs(struct cpuidle_device *device)
 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(device);
 
 	/* state statistics */
-	for (i = 0; i < device->state_count; i++) {
+	for (i = 0; i < drv->state_count; i++) {
 		kobj = kzalloc(sizeof(struct cpuidle_state_kobj), GFP_KERNEL);
 		if (!kobj)
 			goto error_state;

@@ -433,9 +433,10 @@ error_state:
  */
 static void cpuidle_remove_state_sysfs(struct cpuidle_device *device)
 {
+	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(device);
 	int i;
 
-	for (i = 0; i < device->state_count; i++)
+	for (i = 0; i < drv->state_count; i++)
 		cpuidle_free_state_kobj(device, i);
 }
 

index 306178d..9c5e892 100644
@@ -77,7 +77,6 @@ struct cpuidle_device {
 	unsigned int		cpu;
 
 	int			last_residency;
-	int			state_count;
 	struct cpuidle_state_usage	states_usage[CPUIDLE_STATE_MAX];
 	struct cpuidle_state_kobj *kobjs[CPUIDLE_STATE_MAX];
 	struct cpuidle_driver_kobj *kobj_driver;

2.1.0

From: Thomas Schlichter <[email protected]>
Date: Tue, 31 Mar 2015 20:24:39 +0200
Subject: [PATCH 2/2] ACPI / cpuidle: do not overwrite name and description of
C0

This fixes a bug that leads to showing name and description of C-state C0 as "<null>" in sysfs after the ACPI C-states changed (e.g. after AC->DC or DC->AC transition).

The function poll_idle_init() in drivers/cpuidle/driver.c initializes the state 0 during cpuidle_register_driver(). So we better do not overwrite it again with '\0' during acpi_processor_cst_has_changed().

---
 drivers/acpi/processor_idle.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index c6bb9f1..f98db0b 100644
@@ -922,7 +922,7 @@ static int acpi_processor_setup_cpuidle_states(struct acpi_processor *pr)
 		return -EINVAL;
 
 	drv->safe_state_index = -1;
-	for (i = 0; i < CPUIDLE_STATE_MAX; i++) {
+	for (i = CPUIDLE_DRIVER_STATE_START; i < CPUIDLE_STATE_MAX; i++) {
 		drv->states[i].name[0] = '\0';
 		drv->states[i].desc[0] = '\0';
 	}

2.1.0

Hi,

Rafael, please apply that patch (as Thomas has noticed it is required to fix issue with dynamic dev/drv->state_count updates, sorry for not catching it earlier).

[ Originally the whole series was dropped when patch #8 ("[PATCH v2 8/9]
intel_idle: use the common cpuidle_[un]register() routines") turned out to introduce some problems. Patches #1-7 were later re-merged but not patch #9 (which is independent of patch #8). ]

This patch looks fine to me. FWIW:

Sounds good to me. Thank you for working on this.

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics