[RFC] [v2 Patch 1/6] drivers: reset: TI: SoC reset controller support.

Felipe

Thanks for the comments

On 05/05/2014 04:33 PM, Felipe Balbi wrote:

Hi,

On Mon, May 05, 2014 at 03:09:22PM -0500, Dan Murphy wrote:

quoted

The TI SoC reset controller support utilizes the
reset controller framework to give device drivers or
function drivers a common set of APIs to call to reset
a module.

The reset-ti is a common interface to the reset framework.
 The register data is retrieved during initialization
 of the reset driver through the reset-ti-data
file.  The array of data is associated with the compatible from the
respective DT entry.

Once the data is available then this is derefenced within the common
interface.

The device driver has the ability to assert, deassert or perform a
complete reset.

This code was derived from previous work by Rajendra Nayak and Afzal Mohammed.
The code was changed to adopt to the reset core and abstract away the SoC information.

you should break your commit log at around 72 characters at most.

Do you mean 72 characters per line?

quoted

Signed-off-by: Dan Murphy <redacted>
---
 drivers/reset/Kconfig            |    1 +
 drivers/reset/Makefile           |    1 +
 drivers/reset/ti/Kconfig         |    8 ++
 drivers/reset/ti/Makefile        |    1 +
 drivers/reset/ti/reset-ti-data.h |   56 ++++++++
 drivers/reset/ti/reset-ti.c      |  267 ++++++++++++++++++++++++++++++++++++++
 6 files changed, 334 insertions(+)
 create mode 100644 drivers/reset/ti/Kconfig
 create mode 100644 drivers/reset/ti/Makefile
 create mode 100644 drivers/reset/ti/reset-ti-data.h
 create mode 100644 drivers/reset/ti/reset-ti.c

diff --git a/drivers/reset/Kconfig b/drivers/reset/Kconfig
index 0615f50..a58d789 100644
--- a/drivers/reset/Kconfig
+++ b/drivers/reset/Kconfig

@@ -13,3 +13,4 @@ menuconfig RESET_CONTROLLER
 	  If unsure, say no.
 
 source "drivers/reset/sti/Kconfig"
+source "drivers/reset/ti/Kconfig"

this driver is so small that I'm not sure you need a directory for it.

Yeah.  Carry over from v1 when we had the object data files which made
things bigger I will put them back.

quoted

diff --git a/drivers/reset/ti/Kconfig b/drivers/reset/ti/Kconfig
new file mode 100644
index 0000000..dcdce90
--- /dev/null
+++ b/drivers/reset/ti/Kconfig

@@ -0,0 +1,8 @@
+config RESET_TI
+	depends on RESET_CONTROLLER

don't you want to depend on ARCH_OMAP || COMPILE_TEST ?

Yes

quoted

diff --git a/drivers/reset/ti/reset-ti-data.h b/drivers/reset/ti/reset-ti-data.h
new file mode 100644
index 0000000..4d2a6d5
--- /dev/null
+++ b/drivers/reset/ti/reset-ti-data.h

@@ -0,0 +1,56 @@
+/*
+ * PRCM reset driver for TI SoC's
+ *
+ * Copyright 2014 Texas Instruments Inc.

this is not the TI aproved way for copyright notices. Yeah,
nit-picking...

Hmm.  Will need to look at other TI files to correct then.

quoted

+ * Author: Dan Murphy [off-list ref]
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef _RESET_TI_DATA_H_
+#define _RESET_TI_DATA_H_
+
+#include <linux/kernel.h>
+#include <linux/reset-controller.h>
+
+/**
+ * struct ti_reset_reg_data - Structure of the reset register information
+ *		for a particular SoC.
+ * @rstctrl_offs: This is the reset control offset value from
+ *		from the parent reset node.
+ * @rstst_offs: This is the reset status offset value from
+ *		from the parent reset node.
+ * @rstctrl_bit: This is the reset control bit for the module.
+ * @rstst_bit: This is the reset status bit for the module.
+ *
+ * This structure describes the reset register control and status offsets.
+ * The bits are also defined for the same.
+ */
+struct ti_reset_reg_data {
+	void __iomem *reg_base;
+	u32	rstctrl_offs;
+	u32	rstst_offs;
+	u32	rstctrl_bit;
+	u32	rstst_bit;

this structure can be folded into struct ti_reset_data and all of that
can be initialized during probe.

I could do that.  It will simplify the de-referencing as well.

quoted

+};
+
+/**
+ * struct ti_reset_data - Structure that contains the reset register data
+ *	as well as the total number of resets for a particular SoC.
+ * @reg_data:	Pointer to the register data structure.
+ * @nr_resets:	Total number of resets for the SoC in the reset array.
+ *
+ * This structure contains a pointer to the register data and the modules
+ * register base.  The number of resets and reset controller device data is
+ * stored within this structure.
+ *
+ */
+struct ti_reset_data {
+	struct ti_reset_reg_data *reg_data;
+	struct reset_controller_dev rcdev;
+};
+
+#endif

this entire file can be moved into the .c file. It's too small and the
only user for these new types is the .c driver anyway.

This was another carry over from v1 when I had other data within.
So I can collapse this now.

quoted

diff --git a/drivers/reset/ti/reset-ti.c b/drivers/reset/ti/reset-ti.c
new file mode 100644
index 0000000..349f4fb
--- /dev/null
+++ b/drivers/reset/ti/reset-ti.c

@@ -0,0 +1,267 @@
+/*
+ * PRCM reset driver for TI SoC's
+ *
+ * Copyright 2014 Texas Instruments Inc.
+ *
+ * Author: Dan Murphy <dmurphy@ti.com>

fix copyright notice here too

quoted

+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/device.h>
+#include <linux/err.h>
+#include <linux/io.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/of_address.h>
+#include <linux/of_device.h>
+#include <linux/platform_device.h>
+#include <linux/reset.h>
+#include <linux/slab.h>
+
+#include "reset-ti-data.h"
+
+#define DRIVER_NAME "prcm_reset_ti"
+
+static struct ti_reset_data *ti_data;

NAK, you *really* don't need and don't to have this global here. It only
creates problems. And avoid arguing that "there's only one reset
controller anyway" because that has bitten us in the past. Also, it's
not a lot of work to make proper use of dev_set/get_drvdata() and
container_of() to find your struct ti_reset_data pointer.

Agreed will fix

quoted

+static int ti_reset_get_of_data(struct ti_reset_reg_data *reset_data,
+				unsigned long id)
+{
+	struct device_node *dev_node;
+	struct device_node *parent;
+	struct device_node *prcm_parent;
+	struct device_node *reset_parent;
+	int ret = -EINVAL;
+
+	dev_node = of_find_node_by_phandle((phandle) id);
+	if (!dev_node) {
+		pr_err("%s: Cannot find phandle node\n", __func__);

pass a struct device pointer as argument so you can convert these to
dev_err().

Agreed

quoted

+		return ret;
+	}
+
+	/* node parent */
+	parent = of_get_parent(dev_node);
+	if (!parent) {
+		pr_err("%s: Cannot find parent reset node\n", __func__);
+		return ret;
+	}
+	/* prcm reset parent */
+	reset_parent = of_get_next_parent(parent);
+	if (!reset_parent) {
+		pr_err("%s: Cannot find parent reset node\n", __func__);
+		return ret;
+	}
+	/* PRCM Parent */
+	reset_parent = of_get_parent(reset_parent);
+	if (!prcm_parent) {
+		pr_err("%s: Cannot find parent reset node\n", __func__);
+		return ret;
+	}
+
+	reset_data->reg_base = of_iomap(reset_parent, 0);

please don't. of_iomap() is the stupidest helper in the kernel. Find
your resource using platform_get_resource() and map it with
devm_ioremap_resource() since that will properly request_mem_region()
and correctly map the resource with or without caching.

Thanks for the information.  The reason this is here so that if there are multiple PRCM
modules I can just get the base address for the phandle of interest.

quoted

+	if (!reset_data->reg_base) {
+		pr_err("%s: Cannot map reset parent.\n", __func__);
+		return ret;
+	}
+
+	ret = of_property_read_u32_index(parent, "reg", 0,
+					 &reset_data->rstctrl_offs);
+	if (ret)
+		return ret;
+
+	ret = of_property_read_u32_index(parent, "reg", 1,
+					 &reset_data->rstst_offs);
+	if (ret)
+		return ret;
+
+	ret = of_property_read_u32(dev_node, "control-bit",
+				   &reset_data->rstctrl_bit);
+	if (ret < 0)
+		pr_err("%s: No entry in %s for rstst_offs\n", __func__,
+			   dev_node->name);
+
+	ret = of_property_read_u32(dev_node, "status-bit",
+				   &reset_data->rstst_bit);
+	if (ret < 0)
+		pr_err("%s: No entry in %s for rstst_offs\n", __func__,
+			   dev_node->name);
+
+	return 0;
+}
+
+static void ti_reset_wait_on_reset(struct reset_controller_dev *rcdev,
+				unsigned long id)
+{

right here you should have:

	struct ti_reset_data *ti_data = container_of(rcdev,
		struct ti_reset_data, rcdev);

I had that in v1.  Should have kept that.
I will change

quoted

+	struct ti_reset_reg_data *temp_reg_data;
+	void __iomem *status_reg;
+	u32 bit_mask = 0;
+	u32 val = 0;
+
+	temp_reg_data = kzalloc(sizeof(struct ti_reset_reg_data), GFP_KERNEL);

let's think about this for a second:

user asks for a reset, then ->assert() method gets called, which will
allocate memory, do some crap, free the allocated memory, then you call
your ->deassert() method which will, allocate memory, do something, free
memory, then this method is called which will allocate memory, do
something, free memory.

Note that *all* of your allocations a) are unnecessary and b) will break
resets from inside IRQ context...

This made also think that this is not thread safe as this reset calls can be called
from multiple callers so I think a semaphore is order for this function plus the other calls.

quoted

+	ti_reset_get_of_data(temp_reg_data, id);

this means that *every time* a reset is asserted/deasserted/waited on,
you will parse *ONE MORE TIME* the DTB. Why ? Why don't you parse it
once during probe() and cache the results ?

Cache it into what, a list?
I had implemented a list before and received comments do not use a list.  Because you
have to iterate through the list every time.  And a list would need to contain some sort
of indexing agent which defeats the purpose of a phandle.  Then the DTB and kernel images
would have to be in sync for the indexes.

If I put it into an array I run into issues with a bounded array and need to introduce
the index anyway.  So passing the phandle is futile which means I have to introduce the
indexing from the V1 series again.

For my information why is parsing the DTB worse then iterating through a list?
Is there a possibility that the DTB will be over written?

quoted

+	/* Clear the reset status bit to reflect the current status */
+	status_reg = temp_reg_data->reg_base + temp_reg_data->rstst_offs;
+	bit_mask = temp_reg_data->rstst_bit;
+	do {
+		val = readl(status_reg);
+		if (!(val & (1 << bit_mask)))
+			break;
+	} while (1);

also, none of these loops have a timeout. You are creating the
possibility of hanging the platform completely if the HW misbehaves.

Agreed.  Philip commented on that on v1 I overlooked that comment.

quoted

+static int ti_reset_assert(struct reset_controller_dev *rcdev,
+			       unsigned long id)
+{
+	struct ti_reset_reg_data *temp_reg_data;
+	void __iomem *reg;
+	void __iomem *status_reg;
+	u32 status_bit = 0;
+	u32 bit_mask = 0;
+	u32 val = 0;
+
+	temp_reg_data = kzalloc(sizeof(struct ti_reset_reg_data), GFP_KERNEL);
+	ti_reset_get_of_data(temp_reg_data, id);
+
+	/* Clear the reset status bit to reflect the current status */
+	status_reg = temp_reg_data->reg_base + temp_reg_data->rstst_offs;
+	status_bit = temp_reg_data->rstst_bit;
+	writel(1 << status_bit, status_reg);
+
+	reg = temp_reg_data->reg_base + temp_reg_data->rstctrl_offs;
+	bit_mask = temp_reg_data->rstctrl_bit;
+	val = readl(reg);
+	if (!(val & bit_mask)) {
+		val |= bit_mask;
+		writel(val, reg);
+	}
+
+	kfree(temp_reg_data);

same comments as previous callback

Same answer here

quoted

+	return 0;
+}
+
+static int ti_reset_deassert(struct reset_controller_dev *rcdev,
+			       unsigned long id)
+{
+
+	struct ti_reset_reg_data *temp_reg_data;
+	void __iomem *reg;
+	void __iomem *status_reg;
+	u32 status_bit = 0;
+	u32 bit_mask = 0;
+	u32 val = 0;
+
+	temp_reg_data = kzalloc(sizeof(struct ti_reset_reg_data), GFP_KERNEL);
+	ti_reset_get_of_data(temp_reg_data, id);
+
+	/* Clear the reset status bit to reflect the current status */
+	status_reg = temp_reg_data->reg_base + temp_reg_data->rstst_offs;
+	status_bit = temp_reg_data->rstst_bit;
+	writel(1 << status_bit, status_reg);
+
+	reg = temp_reg_data->reg_base + temp_reg_data->rstctrl_offs;
+	bit_mask = temp_reg_data->rstctrl_bit;
+	val = readl(reg);
+	if (val & bit_mask) {
+		val &= ~bit_mask;
+		writel(val, reg);
+	}

and here. Also, this is leaking temp_reg_data.

and here

quoted

+
+	return 0;
+}
+
+static int ti_reset_reset(struct reset_controller_dev *rcdev,
+				  unsigned long id)
+{
+	ti_reset_assert(rcdev, id);
+	ti_reset_deassert(rcdev, id);
+	ti_reset_wait_on_reset(rcdev, id);
+
+	return 0;
+}
+
+static int ti_reset_xlate(struct reset_controller_dev *rcdev,
+			const struct of_phandle_args *reset_spec)
+{
+	struct device_node *dev_node;
+
+	if (WARN_ON(reset_spec->args_count != rcdev->of_reset_n_cells))
+		return -EINVAL;
+
+	/* Verify that the phandle exists */
+	dev_node = of_find_node_by_phandle((phandle) reset_spec->args[0]);
+	if (!dev_node) {
+		pr_err("%s: Cannot find phandle node\n", __func__);
+		return -EINVAL;
+	}
+
+	return reset_spec->args[0];
+}
+
+static struct reset_control_ops ti_reset_ops = {
+	.reset = ti_reset_reset,
+	.assert = ti_reset_assert,
+	.deassert = ti_reset_deassert,
+};
+
+static int ti_reset_probe(struct platform_device *pdev)
+{
+	struct device_node *resets;

here you should have something like:

	struct ti_reset_data *ti_data;

	[...]

	ti_data = devm_kzalloc(sizeof(*ti_data), GFP_KERNEL);
	if (!ti_data)
		return -ENOMEM;

	platform_get_resource(...);

	ti_data->base = devm_ioremap_resource(res);
	if (IS_ERR(ti_data->base))
		return PTR_ERR(ti_data->base);

	platform_set_drvdata(pdev, ti_data);

agreed.  Did not think of this when I made this a module.

quoted

+	resets = of_find_node_by_name(NULL, "resets");
+	if (!resets) {
+		pr_err("%s: missing 'resets' child node.\n", __func__);
+		return -EINVAL;
+	}
+
+	ti_data = kzalloc(sizeof(*ti_data), GFP_KERNEL);
+	if (!ti_data)
+		return -ENOMEM;
+
+	ti_data->rcdev.owner = THIS_MODULE;
+	ti_data->rcdev.of_node = resets;
+	ti_data->rcdev.ops = &ti_reset_ops;
+
+	ti_data->rcdev.of_reset_n_cells = 1;
+	ti_data->rcdev.of_xlate = &ti_reset_xlate;
+
+	reset_controller_register(&ti_data->rcdev);
+
+	return 0;
+}
+
+static int ti_reset_remove(struct platform_device *pdev)
+{

and here:

	struct ti_reset_data *ti_data = platform_get_drvdata(pdev);

	reset_controller_unregister(&ti_data->rcdev);

quoted

+	reset_controller_unregister(&ti_data->rcdev);
+
+	return 0;
+}

-- 
------------------
Dan Murphy