Skip to contentSkip to conversation

ep93xx: Use ioremap for backlight driver

PATCH Superseded Ryan Mallon 4 participants 5 messages Mar 12, 2012 – Mar 13, 2012

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

PATCH v4 4/11 ep93xx: Use ioremap for backlight driver

The ep93xx backlight driver uses a single register within the framebuffer's register space. Currently the backlight driver uses a static IO mapping for the register since the memory cannot be requested by both drivers.

Convert the static mapping to use ioremap so that we can remove the dependency on mach/hardware.h. To do so, we need remove the request_mem_region from both the backlight and framebuffer drivers, since whichever driver is loaded second will fail with -EBUSY otherwise.

A proper fix is still required, and a FIXME comment has been added to both drivers.

Cc H Hartley Sweeten <[email protected]>
Cc Mika Westerberg <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Florian Tobias Schandinat <[email protected]>
---
 arch/arm/mach-ep93xx/core.c         |   10 ++++++++++
 drivers/video/backlight/ep93xx_bl.c |   25 ++++++++++++++-----------
 drivers/video/ep93xx-fb.c           |   15 +++++++++------
 3 files changed, 33 insertions(+), 17 deletions(-)
arch/arm/mach-ep93xx/core.c +10 -0
index b5c1dae..c73e299 100644
@@ -682,9 +682,19 @@ static struct platform_device ep93xx_fb_device = {
 	.resource		= ep93xx_fb_resource,
 };
 
+/* The backlight use a single register in the framebuffer's register space */
+#define EP93XX_RASTER_REG_BRIGHTNESS 0x20
+
+static struct resource ep93xx_bl_resources[] = {
+	DEFINE_RES_MEM(EP93XX_RASTER_PHYS_BASE +
+		       EP93XX_RASTER_REG_BRIGHTNESS, 0x04),
+};
+
 static struct platform_device ep93xx_bl_device = {
 	.name		= "ep93xx-bl",
 	.id		= -1,
+	.num_resources	= ARRAY_SIZE(ep93xx_bl_resources),
+	.resource	= ep93xx_bl_resources,
 };
 
 /**
drivers/video/backlight/ep93xx_bl.c +14 -11
index b62b8b9..08214e1 100644
@@ -17,11 +17,6 @@
 #include <linux/fb.h>
 #include <linux/backlight.h>
 
-#include <mach/hardware.h>
-
-#define EP93XX_RASTER_REG(x)		(EP93XX_RASTER_BASE + (x))
-#define EP93XX_RASTER_BRIGHTNESS	EP93XX_RASTER_REG(0x20)
-
 #define EP93XX_MAX_COUNT		255
 #define EP93XX_MAX_BRIGHT		255
 #define EP93XX_DEF_BRIGHT		128
@@ -35,7 +30,7 @@ static int ep93xxbl_set(struct backlight_device *bl, int brightness)
 {
 	struct ep93xxbl *ep93xxbl = bl_get_data(bl);
 
-	__raw_writel((brightness << 8) | EP93XX_MAX_COUNT, ep93xxbl->mmio);
+	writel((brightness << 8) | EP93XX_MAX_COUNT, ep93xxbl->mmio);
 
 	ep93xxbl->brightness = brightness;
 
@@ -70,21 +65,29 @@ static int __init ep93xxbl_probe(struct platform_device *dev)
 	struct ep93xxbl *ep93xxbl;
 	struct backlight_device *bl;
 	struct backlight_properties props;
+	struct resource *res;
 
 	ep93xxbl = devm_kzalloc(&dev->dev, sizeof(*ep93xxbl), GFP_KERNEL);
 	if (!ep93xxbl)
 		return -ENOMEM;
 
+	res = platform_get_resource(dev, IORESOURCE_MEM, 0);
+	if (!res)
+		return -ENXIO;
+
 	/*
-	 * This register is located in the range already ioremap'ed by
-	 * the framebuffer driver.  A MFD driver seems a bit of overkill
-	 * to handle this so use the static I/O mapping; this address
-	 * is already virtual.
+	 * FIXME - We don't do a request_mem_region here because we are
+	 * sharing the register space with the framebuffer driver (see
+	 * drivers/video/ep93xx-fb.c) and doing so will cause the second
+	 * loaded driver to return -EBUSY.
 	 *
 	 * NOTE: No locking is required; the framebuffer does not touch
 	 * this register.
 	 */
-	ep93xxbl->mmio = EP93XX_RASTER_BRIGHTNESS;
+	ep93xxbl->mmio = devm_ioremap(&dev->dev, res->start,
+				      resource_size(res));
+	if (!ep93xxbl->mmio)
+		return -ENXIO;
 
 	memset(&props, 0, sizeof(struct backlight_properties));
 	props.type = BACKLIGHT_RAW;
drivers/video/ep93xx-fb.c +9 -6
index 2e830ec..5301275 100644
@@ -519,12 +519,15 @@ static int __devinit ep93xxfb_probe(struct platform_device *pdev)
 		goto failed;
 	}
 
-	res = request_mem_region(res->start, resource_size(res), pdev->name);
-	if (!res) {
-		err = -EBUSY;
-		goto failed;
-	}
-
+	/*
+	 * FIXME - We don't do a request_mem_region here because we are
+	 * sharing the register space with the backlight driver (see
+	 * drivers/video/backlight/ep93xx_bl.c) and doing so will cause
+	 * the second loaded driver to return -EBUSY.
+	 *
+	 * NOTE: No locking is required; the backlight does not touch
+	 * any of the framebuffer registers.
+	 */
 	fbi->res = res;
 	fbi->mmio_base = ioremap(res->start, resource_size(res));
 	if (!fbi->mmio_base) {
Signature · 1 line
1.7.0.4
#1 lore raw

PATCH v4 4/11 ep93xx: Use ioremap for backlight driver

Quoted 11 lines · Ryan Mallon

The ep93xx backlight driver uses a single register within the framebuffer's register space. Currently the backlight driver uses a static IO mapping for the register since the memory cannot be requested by both drivers.

Convert the static mapping to use ioremap so that we can remove the dependency on mach/hardware.h. To do so, we need remove the request_mem_region from both the backlight and framebuffer drivers, since whichever driver is loaded second will fail with -EBUSY otherwise.

Quoted 7 lines

A proper fix is still required, and a FIXME comment has been added to both drivers.

Cc H Hartley Sweeten <[email protected]>
Cc Mika Westerberg <[email protected]>
#2 lore raw

PATCH v4 4/11 ep93xx: Use ioremap for backlight driver

In reply to Ryan Mallon
Quoted 5 lines · Ryan Mallon

The ep93xx backlight driver uses a single register within the framebuffer's register space. Currently the backlight driver uses a static IO mapping for the register since the memory cannot be requested by both drivers.

Quoted 8 lines

Convert the static mapping to use ioremap so that we can remove the dependency on mach/hardware.h. To do so, we need remove the request_mem_region from both the backlight and framebuffer drivers, since whichever driver is loaded second will fail with -EBUSY otherwise.

A proper fix is still required, and a FIXME comment has been added to both drivers.

#3 lore raw

PATCH v4 4/11 ep93xx: Use ioremap for backlight driver

In reply to Ryan Mallon
Quoted 21 lines · Ryan Mallon

The ep93xx backlight driver uses a single register within the framebuffer's register space. Currently the backlight driver uses a static IO mapping for the register since the memory cannot be requested by both drivers.

Convert the static mapping to use ioremap so that we can remove the dependency on mach/hardware.h. To do so, we need remove the request_mem_region from both the backlight and framebuffer drivers, since whichever driver is loaded second will fail with -EBUSY otherwise.

A proper fix is still required, and a FIXME comment has been added to both drivers.

Cc H Hartley Sweeten <[email protected]>
Cc Mika Westerberg <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Florian Tobias Schandinat <[email protected]>
Quoted 5 lines
---
 arch/arm/mach-ep93xx/core.c         |   10 ++++++++++
 drivers/video/backlight/ep93xx_bl.c |   25 ++++++++++++++-----------
 drivers/video/ep93xx-fb.c           |   15 +++++++++------
 3 files changed, 33 insertions(+), 17 deletions(-)
#4 lore raw

PATCH v4 4/11 ep93xx: Use ioremap for backlight driver

In reply to Ryan Mallon
Quoted 115 lines · Ryan Mallon

The ep93xx backlight driver uses a single register within the framebuffer's register space. Currently the backlight driver uses a static IO mapping for the register since the memory cannot be requested by both drivers.

Convert the static mapping to use ioremap so that we can remove the dependency on mach/hardware.h. To do so, we need remove the request_mem_region from both the backlight and framebuffer drivers, since whichever driver is loaded second will fail with -EBUSY otherwise.

A proper fix is still required, and a FIXME comment has been added to both drivers.

Cc H Hartley Sweeten <[email protected]>
Cc Mika Westerberg <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Florian Tobias Schandinat <[email protected]>
---
 arch/arm/mach-ep93xx/core.c         |   10 ++++++++++
 drivers/video/backlight/ep93xx_bl.c |   25 ++++++++++++++-----------
 drivers/video/ep93xx-fb.c           |   15 +++++++++------
 3 files changed, 33 insertions(+), 17 deletions(-)
arch/arm/mach-ep93xx/core.c +10 -0
index b5c1dae..c73e299 100644
@@ -682,9 +682,19 @@ static struct platform_device ep93xx_fb_device = {
 	.resource		= ep93xx_fb_resource,
 };
 
+/* The backlight use a single register in the framebuffer's register space */
+#define EP93XX_RASTER_REG_BRIGHTNESS 0x20
+
+static struct resource ep93xx_bl_resources[] = {
+	DEFINE_RES_MEM(EP93XX_RASTER_PHYS_BASE +
+		       EP93XX_RASTER_REG_BRIGHTNESS, 0x04),
+};
+
 static struct platform_device ep93xx_bl_device = {
 	.name		= "ep93xx-bl",
 	.id		= -1,
+	.num_resources	= ARRAY_SIZE(ep93xx_bl_resources),
+	.resource	= ep93xx_bl_resources,
 };
 
 /**
drivers/video/backlight/ep93xx_bl.c +14 -11
index b62b8b9..08214e1 100644
@@ -17,11 +17,6 @@
 #include <linux/fb.h>
 #include <linux/backlight.h>
 
-#include <mach/hardware.h>
-
-#define EP93XX_RASTER_REG(x)		(EP93XX_RASTER_BASE + (x))
-#define EP93XX_RASTER_BRIGHTNESS	EP93XX_RASTER_REG(0x20)
-
 #define EP93XX_MAX_COUNT		255
 #define EP93XX_MAX_BRIGHT		255
 #define EP93XX_DEF_BRIGHT		128
@@ -35,7 +30,7 @@ static int ep93xxbl_set(struct backlight_device *bl, int brightness)
 {
 	struct ep93xxbl *ep93xxbl = bl_get_data(bl);
 
-	__raw_writel((brightness << 8) | EP93XX_MAX_COUNT, ep93xxbl->mmio);
+	writel((brightness << 8) | EP93XX_MAX_COUNT, ep93xxbl->mmio);
 
 	ep93xxbl->brightness = brightness;
 
@@ -70,21 +65,29 @@ static int __init ep93xxbl_probe(struct platform_device *dev)
 	struct ep93xxbl *ep93xxbl;
 	struct backlight_device *bl;
 	struct backlight_properties props;
+	struct resource *res;
 
 	ep93xxbl = devm_kzalloc(&dev->dev, sizeof(*ep93xxbl), GFP_KERNEL);
 	if (!ep93xxbl)
 		return -ENOMEM;
 
+	res = platform_get_resource(dev, IORESOURCE_MEM, 0);
+	if (!res)
+		return -ENXIO;
+
 	/*
-	 * This register is located in the range already ioremap'ed by
-	 * the framebuffer driver.  A MFD driver seems a bit of overkill
-	 * to handle this so use the static I/O mapping; this address
-	 * is already virtual.
+	 * FIXME - We don't do a request_mem_region here because we are
+	 * sharing the register space with the framebuffer driver (see
+	 * drivers/video/ep93xx-fb.c) and doing so will cause the second
+	 * loaded driver to return -EBUSY.
 	 *
 	 * NOTE: No locking is required; the framebuffer does not touch
 	 * this register.
 	 */
-	ep93xxbl->mmio = EP93XX_RASTER_BRIGHTNESS;
+	ep93xxbl->mmio = devm_ioremap(&dev->dev, res->start,
+				      resource_size(res));
+	if (!ep93xxbl->mmio)
+		return -ENXIO;
 
 	memset(&props, 0, sizeof(struct backlight_properties));
 	props.type = BACKLIGHT_RAW;
drivers/video/ep93xx-fb.c +0 -0
index 2e830ec..5301275 100644

drivers/video/ep93xx-fb.c static int __devinit ep93xxfb_probe(struct platform_device *pdev) · line 527

Quoted 9 lines
drivers/video/ep93xx-fb.c +0 -6
@@ -519,12 +519,15 @@ static int __devinit ep93xxfb_probe(struct platform_device *pdev)
 		goto failed;
 	}
 
-	res = request_mem_region(res->start, resource_size(res), pdev->name);
-	if (!res) {
-		err = -EBUSY;
-		goto failed;
-	}
-

Oops... Just took another look at this...

You removed the request_mem_region but left the release_mem_region in the error path and in ep93xxfb_remove.

Regards,
Hartley

Quoted 12 lines
drivers/video/ep93xx-fb.c +9 -0
+	/*
+	 * FIXME - We don't do a request_mem_region here because we are
+	 * sharing the register space with the backlight driver (see
+	 * drivers/video/backlight/ep93xx_bl.c) and doing so will cause
+	 * the second loaded driver to return -EBUSY.
+	 *
+	 * NOTE: No locking is required; the backlight does not touch
+	 * any of the framebuffer registers.
+	 */
 	fbi->res = res;
 	fbi->mmio_base = ioremap(res->start, resource_size(res));
 	if (!fbi->mmio_base) {