Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1517178 > unrolled thread

[PATCH v6 0/7] Exynos IOMMU: proper runtime PM support (use device dependencies)

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  [PATCH v6 0/7] Exynos IOMMU: proper runtime PM support (use device  dependencies) Marek Szyprowski <m.szyprowski@samsung.com> - 2016-11-08 14:40 +0100
    [PATCH v6 5/7] iommu/exynos: Rework and fix internal locking Marek Szyprowski <m.szyprowski@samsung.com> - 2016-11-08 14:40 +0100
    [PATCH v6 7/7] iommu/exynos: Use device dependency links to control  runtime pm Marek Szyprowski <m.szyprowski@samsung.com> - 2016-11-08 14:40 +0100
      Re: [PATCH v6 7/7] iommu/exynos: Use device dependency links to  control runtime pm "Luis R. Rodriguez" <mcgrof@kernel.org> - 2016-11-10 00:40 +0100
        Re: [PATCH v6 7/7] iommu/exynos: Use device dependency links to  control runtime pm "Rafael J. Wysocki" <rafael@kernel.org> - 2016-11-10 00:50 +0100
    [PATCH v6 4/7] iommu/exynos: Set master device once on boot Marek Szyprowski <m.szyprowski@samsung.com> - 2016-11-08 14:40 +0100
    [PATCH v6 6/7] iommu/exynos: Add runtime pm support Marek Szyprowski <m.szyprowski@samsung.com> - 2016-11-08 14:40 +0100
    [PATCH v6 3/7] iommu/exynos: Simplify internal enable/disable functions Marek Szyprowski <m.szyprowski@samsung.com> - 2016-11-08 14:40 +0100
      Re: [PATCH v6 3/7] iommu/exynos: Simplify internal enable/disable  functions Joerg Roedel <joro@8bytes.org> - 2016-11-10 17:50 +0100

#1517178 — [PATCH v6 0/7] Exynos IOMMU: proper runtime PM support (use device dependencies)

Hello,

This is another update of the patchset for adding proper runtime PM
support to Exynos IOMMU driver. This has been achieved by using recently
introduced device links, which lets SYSMMU controller's runtime PM to
follow master's device runtime PM state (the device which actually
performs DMA transaction). The main idea behind this solution is an
observation that any DMA activity from master device can be done only
when master device is active, thus when master device is suspended
SYSMMU controller device can also be suspended.

This patchset solves the problem of all power domains being always
enabled. It happened, because all SYSMMU controllers (which belongs to
the same domains) are permanently kept active, because existing driver
was simplified and kept SYSMMU device active all the time after
initialization and attaching to the master device.

This patchset is based on sixth version of Rafael's "Functional dependencies
between devices" patchset [1], which has been merged to Greg's driver-core-next
branch [2] (last patch commit id is baa8809f60971d10220dfe79248f54b2b265f003).
As Greg pointed, the branch will not be rebased and can be used as a base
for applying my patchset [3].

Joerg: I hope you can merge this version on top of Greg's driver-core-next
branch to iommu tree.

[1] https://www.mail-archive.com/linux-kernel@vger.kernel.org/msg1261311.html
[2] git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core.git driver-core-next
[3] https://www.mail-archive.com/linux-kernel@vger.kernel.org/msg1264906.html

If one wants to test this patchset, I've provided a branch with all needed
patches (a small fix for Exynos4 FIMC-IS DTS is still needed, it is already
in samsung tree):
https://git.linaro.org/people/marek.szyprowski/linux-srpol.git v4.9-iommu-pm-v6

Best regards
Marek Szyprowski
Samsung R&D Institute Poland


Changelog:
v6:
- removed LATE_SYSTEM_SLEEP_PM_OPS-based workaround, because it is no longer
  needed after introducing device links (they also take care of proper system
  sleep suspend/resume sequence)
- updated some comments

v5: https://lkml.org/lkml/2016/10/20/70
- split main patch into several small changes for easier review (requested
  by Luis Rodriquez)
- fixed usage of runtime_pm_active, now it is guarded by pm_runtime_get_noresume()
  and pm_runtime_put() pair

v4: http://www.mail-archive.com/linux-kernel@vger.kernel.org/msg1241601.html
- rebased on top of v4 of device dependencies/links patchset, what resolved
  system hang on reboot

v3: http://www.spinics.net/lists/linux-samsung-soc/msg55256.html
- rebased on top of latest device dependencies/links patchset
- added proper locking between runtime pm, iommu_attach/detach and sysmmu
  enable/disable(added per iommu owner device's rpm lock)

v2: http://www.spinics.net/lists/arm-kernel/msg512082.html
- replaced PM notifiers with generic device dependencies/links developed
  by Rafael J. Wysocki

v1: http://www.spinics.net/lists/arm-kernel/msg509600.html
- initial version


Patch summary:

Marek Szyprowski (7):
  iommu/exynos: Remove excessive, useless debug
  iommu/exynos: Remove dead code
  iommu/exynos: Simplify internal enable/disable functions
  iommu/exynos: Set master device once on boot
  iommu/exynos: Rework and fix internal locking
  iommu/exynos: Add runtime pm support
  iommu/exynos: Use device dependency links to control runtime pm

 drivers/iommu/exynos-iommu.c | 230 ++++++++++++++++++-------------------------
 1 file changed, 95 insertions(+), 135 deletions(-)

-- 
1.9.1

[toc] | [next] | [standalone]

#1517183 — [PATCH v6 5/7] iommu/exynos: Rework and fix internal locking

This patch reworks locking in the exynos_iommu_attach/detach_device
functions to ensure that all entries of the sysmmu_drvdata and
exynos_iommu_owner structure are updated under the respective spinlocks,
while runtime pm functions are called without any spinlocks held.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
 drivers/iommu/exynos-iommu.c | 27 +++++++++++++++++++--------
 1 file changed, 19 insertions(+), 8 deletions(-)

diff --git a/drivers/iommu/exynos-iommu.c b/drivers/iommu/exynos-iommu.c
index 28e570b..a959443 100644
--- a/drivers/iommu/exynos-iommu.c
+++ b/drivers/iommu/exynos-iommu.c
@@ -731,10 +731,12 @@ static void exynos_iommu_domain_free(struct iommu_domain *iommu_domain)
 	spin_lock_irqsave(&domain->lock, flags);
 
 	list_for_each_entry_safe(data, next, &domain->clients, domain_node) {
+		spin_lock(&data->lock);
 		__sysmmu_disable(data);
 		data->pgtable = 0;
 		data->domain = NULL;
 		list_del_init(&data->domain_node);
+		spin_unlock(&data->lock);
 	}
 
 	spin_unlock_irqrestore(&domain->lock, flags);
@@ -772,17 +774,22 @@ static void exynos_iommu_detach_device(struct iommu_domain *iommu_domain,
 	if (!has_sysmmu(dev) || owner->domain != iommu_domain)
 		return;
 
+	list_for_each_entry(data, &owner->controllers, owner_node) {
+		__sysmmu_disable(data);
+		pm_runtime_put(data->sysmmu);
+	}
+
 	spin_lock_irqsave(&domain->lock, flags);
 	list_for_each_entry_safe(data, next, &domain->clients, domain_node) {
-		__sysmmu_disable(data);
+		spin_lock(&data->lock);
 		data->pgtable = 0;
 		data->domain = NULL;
 		list_del_init(&data->domain_node);
-		pm_runtime_put(data->sysmmu);
+		spin_unlock(&data->lock);
 	}
+	owner->domain = NULL;
 	spin_unlock_irqrestore(&domain->lock, flags);
 
-	owner->domain = NULL;
 
 	dev_dbg(dev, "%s: Detached IOMMU with pgtable %pa\n", __func__,
 		&pagetable);
@@ -803,18 +810,22 @@ static int exynos_iommu_attach_device(struct iommu_domain *iommu_domain,
 	if (owner->domain)
 		exynos_iommu_detach_device(owner->domain, dev);
 
+	spin_lock_irqsave(&domain->lock, flags);
 	list_for_each_entry(data, &owner->controllers, owner_node) {
+		spin_lock(&data->lock);
 		data->pgtable = pagetable;
 		data->domain = domain;
+		list_add_tail(&data->domain_node, &domain->clients);
+		spin_unlock(&data->lock);
+	}
+	owner->domain = iommu_domain;
+	spin_unlock_irqrestore(&domain->lock, flags);
+
+	list_for_each_entry(data, &owner->controllers, owner_node) {
 		pm_runtime_get_sync(data->sysmmu);
 		__sysmmu_enable(data);
-
-		spin_lock_irqsave(&domain->lock, flags);
-		list_add_tail(&data->domain_node, &domain->clients);
-		spin_unlock_irqrestore(&domain->lock, flags);
 	}
 
-	owner->domain = iommu_domain;
 	dev_dbg(dev, "%s: Attached IOMMU with pgtable %pa\n", __func__,
 		&pagetable);
 
-- 
1.9.1

[toc] | [prev] | [next] | [standalone]

#1517184 — [PATCH v6 7/7] iommu/exynos: Use device dependency links to control runtime pm

This patch uses recently introduced device dependency links to track the
runtime pm state of the master's device. The goal is to let SYSMMU
controller device's runtime PM to follow the runtime PM state of the
respective master's device. This way each SYSMMU controller is active
only when its master's device is active and can properly restore or save
its state instead on runtime PM transition of master's device.
This approach replaces old behavior, when SYSMMU controller was set to
runtime active once after attaching to the master device. In the new
approach SYSMMU controllers no longer prevents respective power domains
to be turned off when master's device is not being used.

The dependency links also enforces proper order of suspending/restoring
devices during system sleep transition, so there is no more need to use
LATE_SYSTEM_SLEEP_PM_OPS-based workaround for ensuring that SYSMMUs are
suspended after their master devices.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
 drivers/iommu/exynos-iommu.c | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/drivers/iommu/exynos-iommu.c b/drivers/iommu/exynos-iommu.c
index 5e6d7bb..4b05a15 100644
--- a/drivers/iommu/exynos-iommu.c
+++ b/drivers/iommu/exynos-iommu.c
@@ -633,8 +633,8 @@ static int __maybe_unused exynos_sysmmu_resume(struct device *dev)
 
 static const struct dev_pm_ops sysmmu_pm_ops = {
 	SET_RUNTIME_PM_OPS(exynos_sysmmu_suspend, exynos_sysmmu_resume, NULL)
-	SET_LATE_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
-				     pm_runtime_force_resume)
+	SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
+				pm_runtime_force_resume)
 };
 
 static const struct of_device_id sysmmu_of_match[] __initconst = {
@@ -781,10 +781,6 @@ static void exynos_iommu_detach_device(struct iommu_domain *iommu_domain,
 	if (!has_sysmmu(dev) || owner->domain != iommu_domain)
 		return;
 
-	list_for_each_entry(data, &owner->controllers, owner_node) {
-		pm_runtime_put_sync(data->sysmmu);
-	}
-
 	mutex_lock(&owner->rpm_lock);
 
 	list_for_each_entry(data, &owner->controllers, owner_node) {
@@ -848,10 +844,6 @@ static int exynos_iommu_attach_device(struct iommu_domain *iommu_domain,
 
 	mutex_unlock(&owner->rpm_lock);
 
-	list_for_each_entry(data, &owner->controllers, owner_node) {
-		pm_runtime_get_sync(data->sysmmu);
-	}
-
 	dev_dbg(dev, "%s: Attached IOMMU with pgtable %pa\n", __func__,
 		&pagetable);
 
@@ -1232,6 +1224,14 @@ static int exynos_iommu_of_xlate(struct device *dev,
 
 	list_add_tail(&data->owner_node, &owner->controllers);
 	data->master = dev;
+
+	/*
+	 * SYSMMU will be runtime activated via device link (dependency) to its
+	 * master device, so there are no direct calls to pm_runtime_get/put
+	 * in this driver.
+	 */
+	device_link_add(dev, data->sysmmu, DL_FLAG_PM_RUNTIME);
+
 	return 0;
 }
 
-- 
1.9.1

[toc] | [prev] | [next] | [standalone]

#1518566 — Re: [PATCH v6 7/7] iommu/exynos: Use device dependency links to control runtime pm

On Tue, Nov 08, 2016 at 02:29:24PM +0100, Marek Szyprowski wrote:
> This patch uses recently introduced device dependency links to track the
> runtime pm state of the master's device. The goal is to let SYSMMU
> controller device's runtime PM to follow the runtime PM state of the
> respective master's device. This way each SYSMMU controller is active
> only when its master's device is active and can properly restore or save
> its state instead on runtime PM transition of master's device.
> This approach replaces old behavior, when SYSMMU controller was set to
> runtime active once after attaching to the master device. In the new
> approach SYSMMU controllers no longer prevents respective power domains
> to be turned off when master's device is not being used.
> 
> The dependency links also enforces proper order of suspending/restoring
> devices during system sleep transition, so there is no more need to use
> LATE_SYSTEM_SLEEP_PM_OPS-based workaround for ensuring that SYSMMUs are
> suspended after their master devices.

Patches 1-6 seems reasonable to me, however in so far as this patch is
concerned I'd appreaciate if you and Rafael can reply to Lukas Wunner's
questions.

  Luis

> 
> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
> ---
>  drivers/iommu/exynos-iommu.c | 20 ++++++++++----------
>  1 file changed, 10 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/iommu/exynos-iommu.c b/drivers/iommu/exynos-iommu.c
> index 5e6d7bb..4b05a15 100644
> --- a/drivers/iommu/exynos-iommu.c
> +++ b/drivers/iommu/exynos-iommu.c
> @@ -633,8 +633,8 @@ static int __maybe_unused exynos_sysmmu_resume(struct device *dev)
>  
>  static const struct dev_pm_ops sysmmu_pm_ops = {
>  	SET_RUNTIME_PM_OPS(exynos_sysmmu_suspend, exynos_sysmmu_resume, NULL)
> -	SET_LATE_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
> -				     pm_runtime_force_resume)
> +	SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
> +				pm_runtime_force_resume)
>  };
>  
>  static const struct of_device_id sysmmu_of_match[] __initconst = {
> @@ -781,10 +781,6 @@ static void exynos_iommu_detach_device(struct iommu_domain *iommu_domain,
>  	if (!has_sysmmu(dev) || owner->domain != iommu_domain)
>  		return;
>  
> -	list_for_each_entry(data, &owner->controllers, owner_node) {
> -		pm_runtime_put_sync(data->sysmmu);
> -	}
> -
>  	mutex_lock(&owner->rpm_lock);
>  
>  	list_for_each_entry(data, &owner->controllers, owner_node) {
> @@ -848,10 +844,6 @@ static int exynos_iommu_attach_device(struct iommu_domain *iommu_domain,
>  
>  	mutex_unlock(&owner->rpm_lock);
>  
> -	list_for_each_entry(data, &owner->controllers, owner_node) {
> -		pm_runtime_get_sync(data->sysmmu);
> -	}
> -
>  	dev_dbg(dev, "%s: Attached IOMMU with pgtable %pa\n", __func__,
>  		&pagetable);
>  
> @@ -1232,6 +1224,14 @@ static int exynos_iommu_of_xlate(struct device *dev,
>  
>  	list_add_tail(&data->owner_node, &owner->controllers);
>  	data->master = dev;
> +
> +	/*
> +	 * SYSMMU will be runtime activated via device link (dependency) to its
> +	 * master device, so there are no direct calls to pm_runtime_get/put
> +	 * in this driver.
> +	 */
> +	device_link_add(dev, data->sysmmu, DL_FLAG_PM_RUNTIME);
> +
>  	return 0;
>  }
>  
> -- 
> 1.9.1
> 
> 

-- 
Luis Rodriguez, SUSE LINUX GmbH
Maxfeldstrasse 5; D-90409 Nuernberg

[toc] | [prev] | [next] | [standalone]

#1518570 — Re: [PATCH v6 7/7] iommu/exynos: Use device dependency links to control runtime pm

On Thu, Nov 10, 2016 at 12:31 AM, Luis R. Rodriguez <mcgrof@kernel.org> wrote:
> On Tue, Nov 08, 2016 at 02:29:24PM +0100, Marek Szyprowski wrote:
>> This patch uses recently introduced device dependency links to track the
>> runtime pm state of the master's device. The goal is to let SYSMMU
>> controller device's runtime PM to follow the runtime PM state of the
>> respective master's device. This way each SYSMMU controller is active
>> only when its master's device is active and can properly restore or save
>> its state instead on runtime PM transition of master's device.
>> This approach replaces old behavior, when SYSMMU controller was set to
>> runtime active once after attaching to the master device. In the new
>> approach SYSMMU controllers no longer prevents respective power domains
>> to be turned off when master's device is not being used.
>>
>> The dependency links also enforces proper order of suspending/restoring
>> devices during system sleep transition, so there is no more need to use
>> LATE_SYSTEM_SLEEP_PM_OPS-based workaround for ensuring that SYSMMUs are
>> suspended after their master devices.
>
> Patches 1-6 seems reasonable to me, however in so far as this patch is
> concerned I'd appreaciate if you and Rafael can reply to Lukas Wunner's
> questions.

Which questions exactly do you mean?

Thanks,
Rafael

[toc] | [prev] | [next] | [standalone]

#1517186 — [PATCH v6 4/7] iommu/exynos: Set master device once on boot

To avoid possible races, set master device pointer in each SYSMMU
controller once on boot. Suspend/resume callbacks now properly relies on
the configured iommu domain to enable or disable SYSMMU controller.
While changing the code, also update the sleep debug messages and make
them conditional.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
 drivers/iommu/exynos-iommu.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)

diff --git a/drivers/iommu/exynos-iommu.c b/drivers/iommu/exynos-iommu.c
index f45b274..28e570b 100644
--- a/drivers/iommu/exynos-iommu.c
+++ b/drivers/iommu/exynos-iommu.c
@@ -600,10 +600,12 @@ static int exynos_sysmmu_suspend(struct device *dev)
 	struct sysmmu_drvdata *data = dev_get_drvdata(dev);
 	struct device *master = data->master;
 
-	dev_dbg(dev, "suspend\n");
 	if (master) {
-		__sysmmu_disable(data);
 		pm_runtime_put(dev);
+		if (data->domain) {
+			dev_dbg(data->sysmmu, "saving state\n");
+			__sysmmu_disable(data);
+		}
 	}
 	return 0;
 }
@@ -613,10 +615,12 @@ static int exynos_sysmmu_resume(struct device *dev)
 	struct sysmmu_drvdata *data = dev_get_drvdata(dev);
 	struct device *master = data->master;
 
-	dev_dbg(dev, "resume\n");
 	if (master) {
 		pm_runtime_get_sync(dev);
-		__sysmmu_enable(data);
+		if (data->domain) {
+			dev_dbg(data->sysmmu, "restoring state\n");
+			__sysmmu_enable(data);
+		}
 	}
 	return 0;
 }
@@ -730,7 +734,6 @@ static void exynos_iommu_domain_free(struct iommu_domain *iommu_domain)
 		__sysmmu_disable(data);
 		data->pgtable = 0;
 		data->domain = NULL;
-		data->master = NULL;
 		list_del_init(&data->domain_node);
 	}
 
@@ -772,7 +775,6 @@ static void exynos_iommu_detach_device(struct iommu_domain *iommu_domain,
 	spin_lock_irqsave(&domain->lock, flags);
 	list_for_each_entry_safe(data, next, &domain->clients, domain_node) {
 		__sysmmu_disable(data);
-		data->master = NULL;
 		data->pgtable = 0;
 		data->domain = NULL;
 		list_del_init(&data->domain_node);
@@ -806,7 +808,6 @@ static int exynos_iommu_attach_device(struct iommu_domain *iommu_domain,
 		data->domain = domain;
 		pm_runtime_get_sync(data->sysmmu);
 		__sysmmu_enable(data);
-		data->master = dev;
 
 		spin_lock_irqsave(&domain->lock, flags);
 		list_add_tail(&data->domain_node, &domain->clients);
@@ -1192,6 +1193,7 @@ static int exynos_iommu_of_xlate(struct device *dev,
 	}
 
 	list_add_tail(&data->owner_node, &owner->controllers);
+	data->master = dev;
 	return 0;
 }
 
-- 
1.9.1

[toc] | [prev] | [next] | [standalone]

#1517191 — [PATCH v6 6/7] iommu/exynos: Add runtime pm support

This patch adds runtime pm implementation, which is based on previous
suspend/resume code. SYSMMU controller is now being enabled/disabled mainly
from the runtime pm callbacks. System sleep callbacks relies on generic
pm_runtime_force_suspend/pm_runtime_force_resume helpers. To ensure
internal state consistency, additional lock for runtime pm transitions
was introduced.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
 drivers/iommu/exynos-iommu.c | 45 +++++++++++++++++++++++++++++++++++---------
 1 file changed, 36 insertions(+), 9 deletions(-)

diff --git a/drivers/iommu/exynos-iommu.c b/drivers/iommu/exynos-iommu.c
index a959443..5e6d7bb 100644
--- a/drivers/iommu/exynos-iommu.c
+++ b/drivers/iommu/exynos-iommu.c
@@ -206,6 +206,7 @@ struct sysmmu_fault_info {
 struct exynos_iommu_owner {
 	struct list_head controllers;	/* list of sysmmu_drvdata.owner_node */
 	struct iommu_domain *domain;	/* domain this device is attached */
+	struct mutex rpm_lock;		/* for runtime pm of all sysmmus */
 };
 
 /*
@@ -594,40 +595,46 @@ static int __init exynos_sysmmu_probe(struct platform_device *pdev)
 	return 0;
 }
 
-#ifdef CONFIG_PM_SLEEP
-static int exynos_sysmmu_suspend(struct device *dev)
+static int __maybe_unused exynos_sysmmu_suspend(struct device *dev)
 {
 	struct sysmmu_drvdata *data = dev_get_drvdata(dev);
 	struct device *master = data->master;
 
 	if (master) {
-		pm_runtime_put(dev);
+		struct exynos_iommu_owner *owner = master->archdata.iommu;
+
+		mutex_lock(&owner->rpm_lock);
 		if (data->domain) {
 			dev_dbg(data->sysmmu, "saving state\n");
 			__sysmmu_disable(data);
 		}
+		mutex_unlock(&owner->rpm_lock);
 	}
 	return 0;
 }
 
-static int exynos_sysmmu_resume(struct device *dev)
+static int __maybe_unused exynos_sysmmu_resume(struct device *dev)
 {
 	struct sysmmu_drvdata *data = dev_get_drvdata(dev);
 	struct device *master = data->master;
 
 	if (master) {
-		pm_runtime_get_sync(dev);
+		struct exynos_iommu_owner *owner = master->archdata.iommu;
+
+		mutex_lock(&owner->rpm_lock);
 		if (data->domain) {
 			dev_dbg(data->sysmmu, "restoring state\n");
 			__sysmmu_enable(data);
 		}
+		mutex_unlock(&owner->rpm_lock);
 	}
 	return 0;
 }
-#endif
 
 static const struct dev_pm_ops sysmmu_pm_ops = {
-	SET_LATE_SYSTEM_SLEEP_PM_OPS(exynos_sysmmu_suspend, exynos_sysmmu_resume)
+	SET_RUNTIME_PM_OPS(exynos_sysmmu_suspend, exynos_sysmmu_resume, NULL)
+	SET_LATE_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
+				     pm_runtime_force_resume)
 };
 
 static const struct of_device_id sysmmu_of_match[] __initconst = {
@@ -775,7 +782,15 @@ static void exynos_iommu_detach_device(struct iommu_domain *iommu_domain,
 		return;
 
 	list_for_each_entry(data, &owner->controllers, owner_node) {
-		__sysmmu_disable(data);
+		pm_runtime_put_sync(data->sysmmu);
+	}
+
+	mutex_lock(&owner->rpm_lock);
+
+	list_for_each_entry(data, &owner->controllers, owner_node) {
+		pm_runtime_get_noresume(data->sysmmu);
+		if (pm_runtime_active(data->sysmmu))
+			__sysmmu_disable(data);
 		pm_runtime_put(data->sysmmu);
 	}
 
@@ -790,6 +805,7 @@ static void exynos_iommu_detach_device(struct iommu_domain *iommu_domain,
 	owner->domain = NULL;
 	spin_unlock_irqrestore(&domain->lock, flags);
 
+	mutex_unlock(&owner->rpm_lock);
 
 	dev_dbg(dev, "%s: Detached IOMMU with pgtable %pa\n", __func__,
 		&pagetable);
@@ -810,6 +826,8 @@ static int exynos_iommu_attach_device(struct iommu_domain *iommu_domain,
 	if (owner->domain)
 		exynos_iommu_detach_device(owner->domain, dev);
 
+	mutex_lock(&owner->rpm_lock);
+
 	spin_lock_irqsave(&domain->lock, flags);
 	list_for_each_entry(data, &owner->controllers, owner_node) {
 		spin_lock(&data->lock);
@@ -822,8 +840,16 @@ static int exynos_iommu_attach_device(struct iommu_domain *iommu_domain,
 	spin_unlock_irqrestore(&domain->lock, flags);
 
 	list_for_each_entry(data, &owner->controllers, owner_node) {
+		pm_runtime_get_noresume(data->sysmmu);
+		if (pm_runtime_active(data->sysmmu))
+			__sysmmu_enable(data);
+		pm_runtime_put(data->sysmmu);
+	}
+
+	mutex_unlock(&owner->rpm_lock);
+
+	list_for_each_entry(data, &owner->controllers, owner_node) {
 		pm_runtime_get_sync(data->sysmmu);
-		__sysmmu_enable(data);
 	}
 
 	dev_dbg(dev, "%s: Attached IOMMU with pgtable %pa\n", __func__,
@@ -1200,6 +1226,7 @@ static int exynos_iommu_of_xlate(struct device *dev,
 			return -ENOMEM;
 
 		INIT_LIST_HEAD(&owner->controllers);
+		mutex_init(&owner->rpm_lock);
 		dev->archdata.iommu = owner;
 	}
 
-- 
1.9.1

[toc] | [prev] | [next] | [standalone]

#1517193 — [PATCH v6 3/7] iommu/exynos: Simplify internal enable/disable functions

Remove remaining leftovers of the ref-count related code in the
__sysmmu_enable/disable functions inline __sysmmu_enable/disable_nocount
to them. Suspend/resume callbacks now checks if master device is set for
given SYSMMU controller instead of relying on the activation count.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
 drivers/iommu/exynos-iommu.c | 104 ++++++++++++-------------------------------
 1 file changed, 29 insertions(+), 75 deletions(-)

diff --git a/drivers/iommu/exynos-iommu.c b/drivers/iommu/exynos-iommu.c
index 4056228..f45b274 100644
--- a/drivers/iommu/exynos-iommu.c
+++ b/drivers/iommu/exynos-iommu.c
@@ -237,8 +237,8 @@ struct sysmmu_drvdata {
 	struct clk *aclk;		/* SYSMMU's aclk clock */
 	struct clk *pclk;		/* SYSMMU's pclk clock */
 	struct clk *clk_master;		/* master's device clock */
-	int activations;		/* number of calls to sysmmu_enable */
 	spinlock_t lock;		/* lock for modyfying state */
+	int active;			/* current status */
 	struct exynos_iommu_domain *domain; /* domain we belong to */
 	struct list_head domain_node;	/* node for domain clients list */
 	struct list_head owner_node;	/* node for owner controllers list */
@@ -251,25 +251,6 @@ static struct exynos_iommu_domain *to_exynos_domain(struct iommu_domain *dom)
 	return container_of(dom, struct exynos_iommu_domain, domain);
 }
 
-static bool set_sysmmu_active(struct sysmmu_drvdata *data)
-{
-	/* return true if the System MMU was not active previously
-	   and it needs to be initialized */
-	return ++data->activations == 1;
-}
-
-static bool set_sysmmu_inactive(struct sysmmu_drvdata *data)
-{
-	/* return true if the System MMU is needed to be disabled */
-	BUG_ON(data->activations < 1);
-	return --data->activations == 0;
-}
-
-static bool is_sysmmu_active(struct sysmmu_drvdata *data)
-{
-	return data->activations > 0;
-}
-
 static void sysmmu_unblock(struct sysmmu_drvdata *data)
 {
 	writel(CTRL_ENABLE, data->sfrbase + REG_MMU_CTRL);
@@ -388,7 +369,7 @@ static irqreturn_t exynos_sysmmu_irq(int irq, void *dev_id)
 	unsigned short reg_status, reg_clear;
 	int ret = -ENOSYS;
 
-	WARN_ON(!is_sysmmu_active(data));
+	WARN_ON(!data->active);
 
 	if (MMU_MAJ_VER(data->version) < 5) {
 		reg_status = REG_INT_STATUS;
@@ -434,37 +415,19 @@ static irqreturn_t exynos_sysmmu_irq(int irq, void *dev_id)
 	return IRQ_HANDLED;
 }
 
-static void __sysmmu_disable_nocount(struct sysmmu_drvdata *data)
+static void __sysmmu_disable(struct sysmmu_drvdata *data)
 {
+	unsigned long flags;
+
 	clk_enable(data->clk_master);
 
+	spin_lock_irqsave(&data->lock, flags);
 	writel(CTRL_DISABLE, data->sfrbase + REG_MMU_CTRL);
 	writel(0, data->sfrbase + REG_MMU_CFG);
-
-	__sysmmu_disable_clocks(data);
-}
-
-static bool __sysmmu_disable(struct sysmmu_drvdata *data)
-{
-	bool disabled;
-	unsigned long flags;
-
-	spin_lock_irqsave(&data->lock, flags);
-
-	disabled = set_sysmmu_inactive(data);
-
-	if (disabled) {
-		data->pgtable = 0;
-		data->domain = NULL;
-
-		__sysmmu_disable_nocount(data);
-
-		dev_dbg(data->sysmmu, "Disabled\n");
-	}
-
+	data->active = false;
 	spin_unlock_irqrestore(&data->lock, flags);
 
-	return disabled;
+	__sysmmu_disable_clocks(data);
 }
 
 static void __sysmmu_init_config(struct sysmmu_drvdata *data)
@@ -481,17 +444,19 @@ static void __sysmmu_init_config(struct sysmmu_drvdata *data)
 	writel(cfg, data->sfrbase + REG_MMU_CFG);
 }
 
-static void __sysmmu_enable_nocount(struct sysmmu_drvdata *data)
+static void __sysmmu_enable(struct sysmmu_drvdata *data)
 {
+	unsigned long flags;
+
 	__sysmmu_enable_clocks(data);
 
+	spin_lock_irqsave(&data->lock, flags);
 	writel(CTRL_BLOCK, data->sfrbase + REG_MMU_CTRL);
-
 	__sysmmu_init_config(data);
-
 	__sysmmu_set_ptbase(data, data->pgtable);
-
 	writel(CTRL_ENABLE, data->sfrbase + REG_MMU_CTRL);
+	data->active = true;
+	spin_unlock_irqrestore(&data->lock, flags);
 
 	/*
 	 * SYSMMU driver keeps master's clock enabled only for the short
@@ -502,37 +467,18 @@ static void __sysmmu_enable_nocount(struct sysmmu_drvdata *data)
 	clk_disable(data->clk_master);
 }
 
-static int __sysmmu_enable(struct sysmmu_drvdata *data, phys_addr_t pgtable,
-			   struct exynos_iommu_domain *domain)
-{
-	unsigned long flags;
-
-	spin_lock_irqsave(&data->lock, flags);
-	if (set_sysmmu_active(data)) {
-		data->pgtable = pgtable;
-		data->domain = domain;
-		__sysmmu_enable_nocount(data);
-		dev_dbg(data->sysmmu, "Enabled\n");
-	}
-	spin_unlock_irqrestore(&data->lock, flags);
-
-	return 0;
-}
-
 static void sysmmu_tlb_invalidate_flpdcache(struct sysmmu_drvdata *data,
 					    sysmmu_iova_t iova)
 {
 	unsigned long flags;
 
-
 	spin_lock_irqsave(&data->lock, flags);
-	if (is_sysmmu_active(data) && data->version >= MAKE_MMU_VER(3, 3)) {
+	if (data->active && data->version >= MAKE_MMU_VER(3, 3)) {
 		clk_enable(data->clk_master);
 		__sysmmu_tlb_invalidate_entry(data, iova, 1);
 		clk_disable(data->clk_master);
 	}
 	spin_unlock_irqrestore(&data->lock, flags);
-
 }
 
 static void sysmmu_tlb_invalidate_entry(struct sysmmu_drvdata *data,
@@ -541,7 +487,7 @@ static void sysmmu_tlb_invalidate_entry(struct sysmmu_drvdata *data,
 	unsigned long flags;
 
 	spin_lock_irqsave(&data->lock, flags);
-	if (is_sysmmu_active(data)) {
+	if (data->active) {
 		unsigned int num_inv = 1;
 
 		clk_enable(data->clk_master);
@@ -652,10 +598,11 @@ static int __init exynos_sysmmu_probe(struct platform_device *pdev)
 static int exynos_sysmmu_suspend(struct device *dev)
 {
 	struct sysmmu_drvdata *data = dev_get_drvdata(dev);
+	struct device *master = data->master;
 
 	dev_dbg(dev, "suspend\n");
-	if (is_sysmmu_active(data)) {
-		__sysmmu_disable_nocount(data);
+	if (master) {
+		__sysmmu_disable(data);
 		pm_runtime_put(dev);
 	}
 	return 0;
@@ -664,11 +611,12 @@ static int exynos_sysmmu_suspend(struct device *dev)
 static int exynos_sysmmu_resume(struct device *dev)
 {
 	struct sysmmu_drvdata *data = dev_get_drvdata(dev);
+	struct device *master = data->master;
 
 	dev_dbg(dev, "resume\n");
-	if (is_sysmmu_active(data)) {
+	if (master) {
 		pm_runtime_get_sync(dev);
-		__sysmmu_enable_nocount(data);
+		__sysmmu_enable(data);
 	}
 	return 0;
 }
@@ -780,6 +728,8 @@ static void exynos_iommu_domain_free(struct iommu_domain *iommu_domain)
 
 	list_for_each_entry_safe(data, next, &domain->clients, domain_node) {
 		__sysmmu_disable(data);
+		data->pgtable = 0;
+		data->domain = NULL;
 		data->master = NULL;
 		list_del_init(&data->domain_node);
 	}
@@ -823,6 +773,8 @@ static void exynos_iommu_detach_device(struct iommu_domain *iommu_domain,
 	list_for_each_entry_safe(data, next, &domain->clients, domain_node) {
 		__sysmmu_disable(data);
 		data->master = NULL;
+		data->pgtable = 0;
+		data->domain = NULL;
 		list_del_init(&data->domain_node);
 		pm_runtime_put(data->sysmmu);
 	}
@@ -850,8 +802,10 @@ static int exynos_iommu_attach_device(struct iommu_domain *iommu_domain,
 		exynos_iommu_detach_device(owner->domain, dev);
 
 	list_for_each_entry(data, &owner->controllers, owner_node) {
+		data->pgtable = pagetable;
+		data->domain = domain;
 		pm_runtime_get_sync(data->sysmmu);
-		__sysmmu_enable(data, pagetable, domain);
+		__sysmmu_enable(data);
 		data->master = dev;
 
 		spin_lock_irqsave(&domain->lock, flags);
-- 
1.9.1

[toc] | [prev] | [next] | [standalone]

#1519139 — Re: [PATCH v6 3/7] iommu/exynos: Simplify internal enable/disable functions

On Tue, Nov 08, 2016 at 02:29:20PM +0100, Marek Szyprowski wrote:
> Remove remaining leftovers of the ref-count related code in the
> __sysmmu_enable/disable functions inline __sysmmu_enable/disable_nocount
> to them. Suspend/resume callbacks now checks if master device is set for
> given SYSMMU controller instead of relying on the activation count.
> 
> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
> ---
>  drivers/iommu/exynos-iommu.c | 104 ++++++++++++-------------------------------
>  1 file changed, 29 insertions(+), 75 deletions(-)
> 
> diff --git a/drivers/iommu/exynos-iommu.c b/drivers/iommu/exynos-iommu.c
> index 4056228..f45b274 100644
> --- a/drivers/iommu/exynos-iommu.c
> +++ b/drivers/iommu/exynos-iommu.c
> @@ -237,8 +237,8 @@ struct sysmmu_drvdata {
>  	struct clk *aclk;		/* SYSMMU's aclk clock */
>  	struct clk *pclk;		/* SYSMMU's pclk clock */
>  	struct clk *clk_master;		/* master's device clock */
> -	int activations;		/* number of calls to sysmmu_enable */
>  	spinlock_t lock;		/* lock for modyfying state */
> +	int active;			/* current status */

Any reason for this being int and not bool? You only assign true/false
later.


	Joerg

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web