* [PATCH 1/3] iommu/exynos: Return proper errors from getting clocks
2016-05-23 9:30 [PATCH 0/3] Exynos IOMMU: improve clock management Marek Szyprowski
@ 2016-05-23 9:30 ` Marek Szyprowski
2016-05-23 9:30 ` [PATCH 2/3] iommu/exynos: Fix master clock management for inactive SYSMMU Marek Szyprowski
` (3 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Marek Szyprowski @ 2016-05-23 9:30 UTC (permalink / raw)
To: linux-arm-kernel
This patch reworks driver probe code to propagate error codes from
clk_get() operation. This will allow to properly handle deferred probe
in the future.
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
drivers/iommu/exynos-iommu.c | 56 +++++++++++++++++++-------------------------
1 file changed, 24 insertions(+), 32 deletions(-)
diff --git a/drivers/iommu/exynos-iommu.c b/drivers/iommu/exynos-iommu.c
index e27e3b7df4e7..989365682d14 100644
--- a/drivers/iommu/exynos-iommu.c
+++ b/drivers/iommu/exynos-iommu.c
@@ -602,37 +602,31 @@ static int __init exynos_sysmmu_probe(struct platform_device *pdev)
}
data->clk = devm_clk_get(dev, "sysmmu");
- if (!IS_ERR(data->clk)) {
- ret = clk_prepare(data->clk);
- if (ret) {
- dev_err(dev, "Failed to prepare clk\n");
- return ret;
- }
- } else {
+ if (PTR_ERR(data->clk) == -ENOENT)
data->clk = NULL;
- }
+ else if (IS_ERR(data->clk))
+ return PTR_ERR(data->clk);
+ ret = clk_prepare(data->clk);
+ if (ret)
+ return ret;
data->aclk = devm_clk_get(dev, "aclk");
- if (!IS_ERR(data->aclk)) {
- ret = clk_prepare(data->aclk);
- if (ret) {
- dev_err(dev, "Failed to prepare aclk\n");
- return ret;
- }
- } else {
+ if (PTR_ERR(data->aclk) == -ENOENT)
data->aclk = NULL;
- }
+ else if (IS_ERR(data->aclk))
+ return PTR_ERR(data->aclk);
+ ret = clk_prepare(data->aclk);
+ if (ret)
+ return ret;
data->pclk = devm_clk_get(dev, "pclk");
- if (!IS_ERR(data->pclk)) {
- ret = clk_prepare(data->pclk);
- if (ret) {
- dev_err(dev, "Failed to prepare pclk\n");
- return ret;
- }
- } else {
+ if (PTR_ERR(data->pclk) == -ENOENT)
data->pclk = NULL;
- }
+ else if (IS_ERR(data->pclk))
+ return PTR_ERR(data->pclk);
+ ret = clk_prepare(data->pclk);
+ if (ret)
+ return ret;
if (!data->clk && (!data->aclk || !data->pclk)) {
dev_err(dev, "Failed to get device clock(s)!\n");
@@ -640,15 +634,13 @@ static int __init exynos_sysmmu_probe(struct platform_device *pdev)
}
data->clk_master = devm_clk_get(dev, "master");
- if (!IS_ERR(data->clk_master)) {
- ret = clk_prepare(data->clk_master);
- if (ret) {
- dev_err(dev, "Failed to prepare master's clk\n");
- return ret;
- }
- } else {
+ if (PTR_ERR(data->clk_master) == -ENOENT)
data->clk_master = NULL;
- }
+ else if (IS_ERR(data->clk_master))
+ return PTR_ERR(data->clk_master);
+ ret = clk_prepare(data->clk_master);
+ if (ret)
+ return ret;
data->sysmmu = dev;
spin_lock_init(&data->lock);
--
1.9.2
^ permalink raw reply related [flat|nested] 6+ messages in thread* [PATCH 2/3] iommu/exynos: Fix master clock management for inactive SYSMMU
2016-05-23 9:30 [PATCH 0/3] Exynos IOMMU: improve clock management Marek Szyprowski
2016-05-23 9:30 ` [PATCH 1/3] iommu/exynos: Return proper errors from getting clocks Marek Szyprowski
@ 2016-05-23 9:30 ` Marek Szyprowski
2016-05-23 9:30 ` [PATCH 3/3] iommu/exynos: Prepare clocks when needed, not in driver probe Marek Szyprowski
` (2 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Marek Szyprowski @ 2016-05-23 9:30 UTC (permalink / raw)
To: linux-arm-kernel
If SYSMMU controller is not active, there is no point in enabling master's
clock just for doing the the of internal state. This patch moves enabling
that clock to the block which actually does the register access.
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
drivers/iommu/exynos-iommu.c | 9 ++++-----
1 file changed, 4 insertions(+), 5 deletions(-)
diff --git a/drivers/iommu/exynos-iommu.c b/drivers/iommu/exynos-iommu.c
index 989365682d14..018bcd5d5edc 100644
--- a/drivers/iommu/exynos-iommu.c
+++ b/drivers/iommu/exynos-iommu.c
@@ -524,16 +524,15 @@ static void sysmmu_tlb_invalidate_flpdcache(struct sysmmu_drvdata *data,
{
unsigned long flags;
- clk_enable(data->clk_master);
spin_lock_irqsave(&data->lock, flags);
- if (is_sysmmu_active(data)) {
- if (data->version >= MAKE_MMU_VER(3, 3))
- __sysmmu_tlb_invalidate_entry(data, iova, 1);
+ if (is_sysmmu_active(data) && 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);
- clk_disable(data->clk_master);
}
static void sysmmu_tlb_invalidate_entry(struct sysmmu_drvdata *data,
--
1.9.2
^ permalink raw reply related [flat|nested] 6+ messages in thread* [PATCH 3/3] iommu/exynos: Prepare clocks when needed, not in driver probe
2016-05-23 9:30 [PATCH 0/3] Exynos IOMMU: improve clock management Marek Szyprowski
2016-05-23 9:30 ` [PATCH 1/3] iommu/exynos: Return proper errors from getting clocks Marek Szyprowski
2016-05-23 9:30 ` [PATCH 2/3] iommu/exynos: Fix master clock management for inactive SYSMMU Marek Szyprowski
@ 2016-05-23 9:30 ` Marek Szyprowski
2016-05-23 11:01 ` [PATCH 4/3] iommu/exynos: Prepare for deferred probe support Marek Szyprowski
2016-06-15 12:00 ` [PATCH 0/3] Exynos IOMMU: improve clock management Joerg Roedel
4 siblings, 0 replies; 6+ messages in thread
From: Marek Szyprowski @ 2016-05-23 9:30 UTC (permalink / raw)
To: linux-arm-kernel
Make clock preparation together with clk_enable(). This way inactive
SYSMMU controllers will not keep clocks prepared all the time.
This change allows more fine graded power management in the future.
All the code assumes that clock management doesn't fail, so guard
clock_prepare_enable() it with BUG_ON().
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
drivers/iommu/exynos-iommu.c | 54 +++++++++++++++++++++-----------------------
1 file changed, 26 insertions(+), 28 deletions(-)
diff --git a/drivers/iommu/exynos-iommu.c b/drivers/iommu/exynos-iommu.c
index 018bcd5d5edc..e0b834375f17 100644
--- a/drivers/iommu/exynos-iommu.c
+++ b/drivers/iommu/exynos-iommu.c
@@ -322,14 +322,27 @@ static void __sysmmu_set_ptbase(struct sysmmu_drvdata *data, phys_addr_t pgd)
__sysmmu_tlb_invalidate(data);
}
+static void __sysmmu_enable_clocks(struct sysmmu_drvdata *data)
+{
+ BUG_ON(clk_prepare_enable(data->clk_master));
+ BUG_ON(clk_prepare_enable(data->clk));
+ BUG_ON(clk_prepare_enable(data->pclk));
+ BUG_ON(clk_prepare_enable(data->aclk));
+}
+
+static void __sysmmu_disable_clocks(struct sysmmu_drvdata *data)
+{
+ clk_disable_unprepare(data->aclk);
+ clk_disable_unprepare(data->pclk);
+ clk_disable_unprepare(data->clk);
+ clk_disable_unprepare(data->clk_master);
+}
+
static void __sysmmu_get_version(struct sysmmu_drvdata *data)
{
u32 ver;
- clk_enable(data->clk_master);
- clk_enable(data->clk);
- clk_enable(data->pclk);
- clk_enable(data->aclk);
+ __sysmmu_enable_clocks(data);
ver = readl(data->sfrbase + REG_MMU_VERSION);
@@ -342,10 +355,7 @@ static void __sysmmu_get_version(struct sysmmu_drvdata *data)
dev_dbg(data->sysmmu, "hardware version: %d.%d\n",
MMU_MAJ_VER(data->version), MMU_MIN_VER(data->version));
- clk_disable(data->aclk);
- clk_disable(data->pclk);
- clk_disable(data->clk);
- clk_disable(data->clk_master);
+ __sysmmu_disable_clocks(data);
}
static void show_fault_information(struct sysmmu_drvdata *data,
@@ -427,10 +437,7 @@ static void __sysmmu_disable_nocount(struct sysmmu_drvdata *data)
writel(CTRL_DISABLE, data->sfrbase + REG_MMU_CTRL);
writel(0, data->sfrbase + REG_MMU_CFG);
- clk_disable(data->aclk);
- clk_disable(data->pclk);
- clk_disable(data->clk);
- clk_disable(data->clk_master);
+ __sysmmu_disable_clocks(data);
}
static bool __sysmmu_disable(struct sysmmu_drvdata *data)
@@ -475,10 +482,7 @@ static void __sysmmu_init_config(struct sysmmu_drvdata *data)
static void __sysmmu_enable_nocount(struct sysmmu_drvdata *data)
{
- clk_enable(data->clk_master);
- clk_enable(data->clk);
- clk_enable(data->pclk);
- clk_enable(data->aclk);
+ __sysmmu_enable_clocks(data);
writel(CTRL_BLOCK, data->sfrbase + REG_MMU_CTRL);
@@ -488,6 +492,12 @@ static void __sysmmu_enable_nocount(struct sysmmu_drvdata *data)
writel(CTRL_ENABLE, data->sfrbase + REG_MMU_CTRL);
+ /*
+ * SYSMMU driver keeps master's clock enabled only for the short
+ * time, while accessing the registers. For performing address
+ * translation during DMA transaction it relies on the client
+ * driver to enable it.
+ */
clk_disable(data->clk_master);
}
@@ -605,27 +615,18 @@ static int __init exynos_sysmmu_probe(struct platform_device *pdev)
data->clk = NULL;
else if (IS_ERR(data->clk))
return PTR_ERR(data->clk);
- ret = clk_prepare(data->clk);
- if (ret)
- return ret;
data->aclk = devm_clk_get(dev, "aclk");
if (PTR_ERR(data->aclk) == -ENOENT)
data->aclk = NULL;
else if (IS_ERR(data->aclk))
return PTR_ERR(data->aclk);
- ret = clk_prepare(data->aclk);
- if (ret)
- return ret;
data->pclk = devm_clk_get(dev, "pclk");
if (PTR_ERR(data->pclk) == -ENOENT)
data->pclk = NULL;
else if (IS_ERR(data->pclk))
return PTR_ERR(data->pclk);
- ret = clk_prepare(data->pclk);
- if (ret)
- return ret;
if (!data->clk && (!data->aclk || !data->pclk)) {
dev_err(dev, "Failed to get device clock(s)!\n");
@@ -637,9 +638,6 @@ static int __init exynos_sysmmu_probe(struct platform_device *pdev)
data->clk_master = NULL;
else if (IS_ERR(data->clk_master))
return PTR_ERR(data->clk_master);
- ret = clk_prepare(data->clk_master);
- if (ret)
- return ret;
data->sysmmu = dev;
spin_lock_init(&data->lock);
--
1.9.2
^ permalink raw reply related [flat|nested] 6+ messages in thread* [PATCH 4/3] iommu/exynos: Prepare for deferred probe support
2016-05-23 9:30 [PATCH 0/3] Exynos IOMMU: improve clock management Marek Szyprowski
` (2 preceding siblings ...)
2016-05-23 9:30 ` [PATCH 3/3] iommu/exynos: Prepare clocks when needed, not in driver probe Marek Szyprowski
@ 2016-05-23 11:01 ` Marek Szyprowski
2016-06-15 12:00 ` [PATCH 0/3] Exynos IOMMU: improve clock management Joerg Roedel
4 siblings, 0 replies; 6+ messages in thread
From: Marek Szyprowski @ 2016-05-23 11:01 UTC (permalink / raw)
To: linux-arm-kernel
Register iommu_ops at the end of successful probe instead of doing that
unconditionally. This makes Exynos IOMMU driver ready for deferred probe
caused by not-yet-available clocks.
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
---
I'm sorry for the confussing patch number. This patch got lost while
rebasing, but it should belong to the "Exynos IOMMU: improve clock
management" patchset.
Best regards
Marek Szyprowski
Samsung R&D Institute Poland
---
drivers/iommu/exynos-iommu.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/drivers/iommu/exynos-iommu.c b/drivers/iommu/exynos-iommu.c
index e0b834375f17..633e6d023c0d 100644
--- a/drivers/iommu/exynos-iommu.c
+++ b/drivers/iommu/exynos-iommu.c
@@ -581,6 +581,8 @@ static void sysmmu_tlb_invalidate_entry(struct sysmmu_drvdata *data,
spin_unlock_irqrestore(&data->lock, flags);
}
+static struct iommu_ops exynos_iommu_ops;
+
static int __init exynos_sysmmu_probe(struct platform_device *pdev)
{
int irq, ret;
@@ -654,6 +656,8 @@ static int __init exynos_sysmmu_probe(struct platform_device *pdev)
pm_runtime_enable(dev);
+ of_iommu_set_ops(dev->of_node, &exynos_iommu_ops);
+
return 0;
}
@@ -1347,7 +1351,6 @@ static int __init exynos_iommu_of_setup(struct device_node *np)
if (!dma_dev)
dma_dev = &pdev->dev;
- of_iommu_set_ops(np, &exynos_iommu_ops);
return 0;
}
--
1.9.2
^ permalink raw reply related [flat|nested] 6+ messages in thread* [PATCH 0/3] Exynos IOMMU: improve clock management
2016-05-23 9:30 [PATCH 0/3] Exynos IOMMU: improve clock management Marek Szyprowski
` (3 preceding siblings ...)
2016-05-23 11:01 ` [PATCH 4/3] iommu/exynos: Prepare for deferred probe support Marek Szyprowski
@ 2016-06-15 12:00 ` Joerg Roedel
4 siblings, 0 replies; 6+ messages in thread
From: Joerg Roedel @ 2016-06-15 12:00 UTC (permalink / raw)
To: linux-arm-kernel
On Mon, May 23, 2016 at 11:30:06AM +0200, Marek Szyprowski wrote:
> Hello,
>
> This patch series improves clock management in Exynos IOMMU driver, so
> the driver will be ready for potential deferred probe caused by
> not-yet-available clocks. Such case doesn't happen with current kernel,
> but it may happen with the planned rework on clocks and power domain
> code.
>
> Best regards
> Marek Szyprowski
> Samsung R&D Institute Poland
>
>
> Marek Szyprowski (3):
> iommu/exynos: Return proper errors from getting clocks
> iommu/exynos: Fix master clock management for inactive SYSMMU
> iommu/exynos: Prepare clocks when needed, not in driver probe
>
> drivers/iommu/exynos-iommu.c | 95 ++++++++++++++++++++------------------------
> 1 file changed, 42 insertions(+), 53 deletions(-)
Applied all, thanks.
^ permalink raw reply [flat|nested] 6+ messages in thread