Skip to contentSkip to conversation

drm/fsl-dcu: fix register initialization

PATCH Applied Stefan Agner 1 participant 4 messages Feb 3, 2016 – Feb 26, 2016

Subsystems (3)
lore
Share & tools
lore

Stefan Agner

author maintainer
#0 lore raw

PATCH v2 1/2 drm/fsl-dcu: fix register initialization

The layer enumeration start with 0 (0-15 for LS1021a and 0-63 for Vybrid) whereas the register enumeration start from 1 (1-10 for LS1021a and 1-9 for Vybrid). The loop started off from 0 for both iterations and initialized the number of layers inclusive, which is one layer too many.

All extensively written registers seem to be unassigned, it seems that the write to those registers did not do any harm in practice.

---
This two patches apply cleanly on top of my earlier DCU fixes patchset:
https://lkml.org/lkml/2015/11/18/953

Changes since v1:
- (none)

 drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_crtc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_crtc.c +2 -2
index db69725..3c92889 100644
@@ -169,8 +169,8 @@ int fsl_dcu_drm_crtc_create(struct fsl_dcu_drm_device *fsl_dev)
 		reg_num = LS1021A_LAYER_REG_NUM;
 	else
 		reg_num = VF610_LAYER_REG_NUM;
-	for (i = 0; i <= fsl_dev->soc->total_layer; i++) {
-		for (j = 0; j < reg_num; j++)
+	for (i = 0; i < fsl_dev->soc->total_layer; i++) {
+		for (j = 1; j <= reg_num; j++)
 			regmap_write(fsl_dev->regmap, DCU_CTRLDESCLN(i, j), 0);
 	}
 	regmap_update_bits(fsl_dev->regmap, DCU_DCU_MODE,
Signature · 6 lines
2.7.0

_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
author maintainer
#1 lore raw

PATCH v2 2/2 drm/fsl-dcu: use flat regmap cache

Using flat regmap cache instead of RB-tree to avoid the following lockdep warning on driver load:

[    0.697285] WARNING: CPU: 0 PID: 1 at kernel/locking/lockdep.c:2755 lockdep_trace_alloc+0x15c/0x160()
[    0.697449] DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags))

The RB-tree regmap cache needs to allocate new space on first writes. However, allocations in an atomic context (e.g. when a spinlock is held) are not allowed. The function regmap_write calls map->lock, which acquires a spinlock in the fast_io case. Since the FSL DCU driver uses MMIO, the regmap bus of type regmap_mmio is being used which has fast_io set to true.

The MMIO space of the DCU driver is reasonable condense, hence using the much faster flat regmap cache is anyway the better choice.

---
Changes since v1:
- Do not move drm_dev_alloc

 drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c | 18 +++++++++++-------
 drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.h |  6 ++++--
 2 files changed, 15 insertions(+), 9 deletions(-)
drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c +11 -7
index e01c813..9c21aad 100644
@@ -36,11 +36,11 @@ static bool fsl_dcu_drm_is_volatile_reg(struct device *dev, unsigned int reg)
 	return false;
 }
 
-static const struct regmap_config fsl_dcu_regmap_config = {
+static struct regmap_config fsl_dcu_regmap_config = {
 	.reg_bits = 32,
 	.reg_stride = 4,
 	.val_bits = 32,
-	.cache_type = REGCACHE_RBTREE,
+	.cache_type = REGCACHE_FLAT,
 
 	.volatile_reg = fsl_dcu_drm_is_volatile_reg,
 };
@@ -260,12 +260,14 @@ static const struct fsl_dcu_soc_data fsl_dcu_ls1021a_data = {
 	.name = "ls1021a",
 	.total_layer = 16,
 	.max_layer = 4,
+	.max_register = LS1021A_DCU_MAX_REGISTER,
 };
 
 static const struct fsl_dcu_soc_data fsl_dcu_vf610_data = {
 	.name = "vf610",
 	.total_layer = 64,
 	.max_layer = 6,
+	.max_register = VF610_DCU_MAX_REGISTER,
 };
 
 static const struct of_device_id fsl_dcu_of_match[] = {
@@ -331,6 +333,13 @@ static int fsl_dcu_drm_probe(struct platform_device *pdev)
 		return ret;
 	}
 
+	id = of_match_node(fsl_dcu_of_match, pdev->dev.of_node);
+	if (!id)
+		return -ENODEV;
+
+	fsl_dev->soc = id->data;
+
+	fsl_dcu_regmap_config.max_register = fsl_dev->soc->max_register;
 	fsl_dev->regmap = devm_regmap_init_mmio(dev, base,
 			&fsl_dcu_regmap_config);
 	if (IS_ERR(fsl_dev->regmap)) {
@@ -338,11 +347,6 @@ static int fsl_dcu_drm_probe(struct platform_device *pdev)
 		return PTR_ERR(fsl_dev->regmap);
 	}
 
-	id = of_match_node(fsl_dcu_of_match, pdev->dev.of_node);
-	if (!id)
-		return -ENODEV;
-	fsl_dev->soc = id->data;
-
 	drm = drm_dev_alloc(driver, dev);
 	if (!drm)
 		return -ENOMEM;
drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.h +4 -2
index 2a724f3..7c296a0 100644
@@ -114,8 +114,6 @@
 #define DCU_UPDATE_MODE_MODE		BIT(31)
 #define DCU_UPDATE_MODE_READREG		BIT(30)
 
-#define DCU_DCFB_MAX			0x300
-
 #define DCU_CTRLDESCLN(layer, reg)	(0x200 + (reg - 1) * 4 + (layer) * 0x40)
 
 #define DCU_LAYER_HEIGHT(x)		((x) << 16)
@@ -155,6 +153,9 @@
 #define DCU_LAYER_POST_SKIP(x)		((x) << 16)
 #define DCU_LAYER_PRE_SKIP(x)		(x)
 
+#define VF610_DCU_MAX_REGISTER		0x11fc
+#define LS1021A_DCU_MAX_REGISTER	0x5fc
+
 #define FSL_DCU_RGB565			4
 #define FSL_DCU_RGB888			5
 #define FSL_DCU_ARGB8888		6
@@ -175,6 +176,7 @@ struct fsl_dcu_soc_data {
 	unsigned int total_layer;
 	/*max layer number DCU supported*/
 	unsigned int max_layer;
+	unsigned int max_register;
 };
 
 struct fsl_dcu_drm_device {
Signature · 6 lines
2.7.0

_______________________________________________
dri-devel mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/dri-devel
author maintainer
#2 lore raw

Re: [PATCH v2 2/2] drm/fsl-dcu: use flat regmap cache

Quoted 13 lines · Stefan Agner

Using flat regmap cache instead of RB-tree to avoid the following lockdep warning on driver load:

[    0.697285] WARNING: CPU: 0 PID: 1 at kernel/locking/lockdep.c:2755
lockdep_trace_alloc+0x15c/0x160()
[    0.697449] DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags))

The RB-tree regmap cache needs to allocate new space on first writes. However, allocations in an atomic context (e.g. when a spinlock is held) are not allowed. The function regmap_write calls map->lock, which acquires a spinlock in the fast_io case. Since the FSL DCU driver uses MMIO, the regmap bus of type regmap_mmio is being used which has fast_io set to true.

Quoted 3 lines

The MMIO space of the DCU driver is reasonable condense, hence using the much faster flat regmap cache is anyway the better choice.

As discussed with Thierry on IRC, using regmap cache often does not really make sense in the context of display controllers. While trying to use the suspend/resume, I realized that the current implementation in the DCU driver has bugs and does not work (e.g. an explicit READREG is missing, but even with that in place, restoring all registers and enable the controller then seem not to work reliable). Using the new suspend/resume helpers seem to work better...

Therefor I NACK my own patch here and vote for removing the regcache entirely (and use the new DRM supsend/resume helpers to implement suspend/resume)...

--
Stefan

Quoted 98 lines

---
Changes since v1:
- Do not move drm_dev_alloc

 drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c | 18 +++++++++++-------
 drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.h |  6 ++++--
 2 files changed, 15 insertions(+), 9 deletions(-)
drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c +2 -2
index e01c813..9c21aad 100644
@@ -36,11 +36,11 @@ static bool fsl_dcu_drm_is_volatile_reg(struct
 device *dev, unsigned int reg)
 	return false;
 }
 
-static const struct regmap_config fsl_dcu_regmap_config = {
+static struct regmap_config fsl_dcu_regmap_config = {
 	.reg_bits = 32,
 	.reg_stride = 4,
 	.val_bits = 32,
-	.cache_type = REGCACHE_RBTREE,
+	.cache_type = REGCACHE_FLAT,
 
 	.volatile_reg = fsl_dcu_drm_is_volatile_reg,

};

drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c +2 -0
@@ -260,12 +260,14 @@ static const struct fsl_dcu_soc_data
 fsl_dcu_ls1021a_data = {
 	.name = "ls1021a",
 	.total_layer = 16,
 	.max_layer = 4,
+	.max_register = LS1021A_DCU_MAX_REGISTER,
 };
 
 static const struct fsl_dcu_soc_data fsl_dcu_vf610_data = {
 	.name = "vf610",
 	.total_layer = 64,
 	.max_layer = 6,
+	.max_register = VF610_DCU_MAX_REGISTER,
 };
 

static const struct of_device_id fsl_dcu_of_match[] = {

drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.c +7 -5
@@ -331,6 +333,13 @@ static int fsl_dcu_drm_probe(struct platform_device *pdev)
 		return ret;
 	}
 
+	id = of_match_node(fsl_dcu_of_match, pdev->dev.of_node);
+	if (!id)
+		return -ENODEV;
+
+	fsl_dev->soc = id->data;
+
+	fsl_dcu_regmap_config.max_register = fsl_dev->soc->max_register;
 	fsl_dev->regmap = devm_regmap_init_mmio(dev, base,
 			&fsl_dcu_regmap_config);
 	if (IS_ERR(fsl_dev->regmap)) {
@@ -338,11 +347,6 @@ static int fsl_dcu_drm_probe(struct platform_device *pdev)
 		return PTR_ERR(fsl_dev->regmap);
 	}
 
-	id = of_match_node(fsl_dcu_of_match, pdev->dev.of_node);
-	if (!id)
-		return -ENODEV;
-	fsl_dev->soc = id->data;
-
 	drm = drm_dev_alloc(driver, dev);
 	if (!drm)
 		return -ENOMEM;
drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.h +3 -2
index 2a724f3..7c296a0 100644
@@ -114,8 +114,6 @@
 #define DCU_UPDATE_MODE_MODE		BIT(31)
 #define DCU_UPDATE_MODE_READREG		BIT(30)
 
-#define DCU_DCFB_MAX			0x300
-
 #define DCU_CTRLDESCLN(layer, reg)	(0x200 + (reg - 1) * 4 + (layer) * 0x40)
 
 #define DCU_LAYER_HEIGHT(x)		((x) << 16)
@@ -155,6 +153,9 @@
 #define DCU_LAYER_POST_SKIP(x)		((x) << 16)
 #define DCU_LAYER_PRE_SKIP(x)		(x)
 
+#define VF610_DCU_MAX_REGISTER		0x11fc
+#define LS1021A_DCU_MAX_REGISTER	0x5fc
+
 #define FSL_DCU_RGB565			4
 #define FSL_DCU_RGB888			5
 #define FSL_DCU_ARGB8888		6

drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.h struct fsl_dcu_soc_data { · line 182

Quoted 7 lines
drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_drv.h +1 -0
@@ -175,6 +176,7 @@ struct fsl_dcu_soc_data {
 	unsigned int total_layer;
 	/*max layer number DCU supported*/
 	unsigned int max_layer;
+	unsigned int max_register;
 };
 
 struct fsl_dcu_drm_device {

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

author maintainer Applied
#3 lore raw
In reply to Stefan Agner
Quoted 10 lines · Stefan Agner

The layer enumeration start with 0 (0-15 for LS1021a and 0-63 for Vybrid) whereas the register enumeration start from 1 (1-10 for LS1021a and 1-9 for Vybrid). The loop started off from 0 for both iterations and initialized the number of layers inclusive, which is one layer too many.

All extensively written registers seem to be unassigned, it seems that the write to those registers did not do any harm in practice.

Applied.

Quoted 29 lines

---
This two patches apply cleanly on top of my earlier DCU fixes

patchset:

https://lkml.org/lkml/2015/11/18/953

Changes since v1:
- (none)

 drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_crtc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/fsl-dcu/fsl_dcu_drm_crtc.c +2 -2
index db69725..3c92889 100644
@@ -169,8 +169,8 @@ int fsl_dcu_drm_crtc_create(struct
 fsl_dcu_drm_device *fsl_dev)
 		reg_num = LS1021A_LAYER_REG_NUM;
 	else
 		reg_num = VF610_LAYER_REG_NUM;
-	for (i = 0; i <= fsl_dev->soc->total_layer; i++) {
-		for (j = 0; j < reg_num; j++)
+	for (i = 0; i < fsl_dev->soc->total_layer; i++) {
+		for (j = 1; j <= reg_num; j++)
 			regmap_write(fsl_dev->regmap, DCU_CTRLDESCLN(i, j), 0);
 	}
Quoted 1 line

regmap_update_bits(fsl_dev->regmap, DCU_DCU_MODE,

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel