[PATCH v2 0/2] NVMEM cells in sysfs

[PATCH v2 0/2] NVMEM cells in sysfs

[Miquel Raynal]

Hello,

As part of a previous effort, support for dynamic NVMEM layouts was
brought into mainline, helping a lot in getting information from NVMEM
devices at non-static locations. One common example of NVMEM cell is the
MAC address that must be used. Sometimes the cell content is mainly (or
only) useful to the kernel, and sometimes it is not. Users might also
want to know the content of cells such as: the manufacturing place and
date, the hardware version, the unique ID, etc. Two possibilities in
this case: either the users re-implement their own parser to go through
the whole device and search for the information they want, or the kernel
can expose the content of the cells if deemed relevant. This second
approach sounds way more relevant than the first one to avoid useless
code duplication, so here is a series bringing NVMEM cells content to
the user through sysfs.

Here is a real life example with a Marvell Armada 7040 TN48m switch:

$ nvmem=/sys/bus/nvmem/devices/1-00563/
$ for i in `ls -1 $nvmem/cells/*`; do basename $i; hexdump -C $i | head -n1; done
country-code
00000000  54 57                                             |TW|
crc32
00000000  bb cd 51 98                                       |..Q.|
device-version
00000000  02                                                |.|
diag-version
00000000  56 31 2e 30 2e 30                                 |V1.0.0|
label-revision
00000000  44 31                                             |D1|
mac-address
00000000  18 be 92 13 9a 00                                 |......|
manufacture-date
00000000  30 32 2f 32 34 2f 32 30  32 31 20 31 38 3a 35 39  |02/24/2021 18:59|
manufacturer
00000000  44 4e 49                                          |DNI|
num-macs
00000000  00 40                                             |.@|
onie-version
00000000  32 30 32 30 2e 31 31 2d  56 30 31                 |2020.11-V01|
platform-name
00000000  38 38 46 37 30 34 30 2f  38 38 46 36 38 32 30     |88F7040/88F6820|
product-name
00000000  54 4e 34 38 4d 2d 50 2d  44 4e                    |TN48M-P-DN|
serial-number
00000000  54 4e 34 38 31 50 32 54  57 32 30 34 32 30 33 32  |TN481P2TW2042032|
vendor
00000000  44 4e 49                                          |DNI|

Here is a list of known limitations though:
* It is currently not possible to know whether the cell contains ASCII
  or binary data, so by default all cells are exposed in binary form.
* For now the implementation focuses on the read aspect. Technically
  speaking, in some cases, it could be acceptable to write the cells, I
  guess, but for now read-only files sound more than enough. A writable
  path can be added later anyway.
* The sysfs entries are created when the device probes, not when the
  NVMEM driver does. This means, if an NVMEM layout is used *and*
  compiled as a module *and* not installed properly in the system (a
  usermode helper tries to load the module otherwise), then the sysfs
  cells won't appear when the layout is actually insmod'ed because the
  sysfs folders/files have already been populated.

Changes in v2:
* Do not mention the cells might become writable in the future in the
  ABI documentation.
* Fix a wrong return value reported by Dan and kernel test robot.
* Implement .is_bin_visible().
* Avoid overwriting the list of attribute groups, but keep the cells
  attribute group writable as we need to populate it at run time.
* Improve the commit messages.
* Give a real life example in the cover letter.

Miquel Raynal (2):
  ABI: sysfs-nvmem-cells: Expose cells through sysfs
  nvmem: core: Expose cells through sysfs

 Documentation/ABI/testing/sysfs-nvmem-cells |  19 +++
 drivers/nvmem/core.c                        | 145 +++++++++++++++++++-
 2 files changed, 160 insertions(+), 4 deletions(-)
 create mode 100644 Documentation/ABI/testing/sysfs-nvmem-cells

-- 
2.34.1

[PATCH v2 1/2] ABI: sysfs-nvmem-cells: Expose cells through sysfs

[Miquel Raynal]

The binary content of nvmem devices is available to the user so in the
easiest cases, finding the content of a cell is rather easy as it is
just a matter of looking at a known and fixed offset. However, nvmem
layouts have been recently introduced to cope with more advanced
situations, where the offset and size of the cells is not known in
advance or is dynamic. When using layouts, more advanced parsers are
used by the kernel in order to give direct access to the content of each
cell regardless of their position/size in the underlying device, but
these information were not accessible to the user.

By exposing the nvmem cells to the user through a dedicated cell/ folder
containing one file per cell, we provide a straightforward access to
useful user information without the need for re-writing a userland
parser. Content of nvmem cells is usually: product names, manufacturing
date, MAC addresses, etc,

Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
---
 Documentation/ABI/testing/sysfs-nvmem-cells | 19 +++++++++++++++++++
 1 file changed, 19 insertions(+)
 create mode 100644 Documentation/ABI/testing/sysfs-nvmem-cells

diff --git a/Documentation/ABI/testing/sysfs-nvmem-cells b/Documentation/ABI/testing/sysfs-nvmem-cells
new file mode 100644
index 000000000000..bee4acc5bfcb
--- /dev/null
+++ b/Documentation/ABI/testing/sysfs-nvmem-cells
@@ -0,0 +1,19 @@
+What:		/sys/bus/nvmem/devices/.../cells/<cell-name>
+Date:		May 2023
+KernelVersion:  6.5
+Contact:	Miquel Raynal <miquel.raynal@bootlin.com>
+Description:
+		The cells/ folder contains one file per cell exposed by
+		the nvmem device. The name of the file is the cell name.
+		The length of the file is the size of the cell (when
+		known). The content of the file is the binary content of
+		the cell (may sometimes be ASCII, likely without
+		trailing character).
+		Note: This file is only present if CONFIG_NVMEM_SYSFS
+		is enabled
+
+		ex::
+
+		  hexdump -C /sys/bus/nvmem/devices/1-00563/cells/product-name
+		  00000000  54 4e 34 38 4d 2d 50 2d  44 4e         |TN48M-P-DN|
+		  0000000a
-- 
2.34.1

[PATCH v2 2/2] nvmem: core: Expose cells through sysfs

[Miquel Raynal]

The binary content of nvmem devices is available to the user so in the
easiest cases, finding the content of a cell is rather easy as it is
just a matter of looking at a known and fixed offset. However, nvmem
layouts have been recently introduced to cope with more advanced
situations, where the offset and size of the cells is not known in
advance or is dynamic. When using layouts, more advanced parsers are
used by the kernel in order to give direct access to the content of each
cell, regardless of its position/size in the underlying
device. Unfortunately, these information are not accessible by users,
unless by fully re-implementing the parser logic in userland.

Let's expose the cells and their content through sysfs to avoid these
situations.

Exposed cells are read-only. There is, in practice, everything in the
core to support a write path, but as I don't see any need for that, I
prefer to keep the interface simple (and probably safer). The interface
is documented as being in the "testing" state which means we can later
add a write attribute if though relevant.

Of course the relevant NVMEM sysfs Kconfig option must be enabled for
this support to be compiled-in.

The current implementation leads to the 'cells' folder to be always
present even when no cell is actually exposed. This is due to a sysfs
limitation which might in the future be overcome. In order to be as
close as how sysfs was designed, despite the cells .bin_attrs attribute
group member being assigned at runtime (and thus, not movable to a RO
section), we provide a .is_bin_visible hook which might become really
useful if the series avoiding the creation of empty directories borns.

There is one limitation though: if a layout is built as a module but is
not properly installed in the system and loaded manually with insmod
while the nvmem device driver was built-in, the cells won't appear in
sysfs. But if done like that, the cells won't be usable by the built-in
kernel drivers anyway.

Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
---
 drivers/nvmem/core.c | 145 +++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 141 insertions(+), 4 deletions(-)

diff --git a/drivers/nvmem/core.c b/drivers/nvmem/core.c
index 342cd380b420..02b70f065ebc 100644
--- a/drivers/nvmem/core.c
+++ b/drivers/nvmem/core.c
@@ -325,6 +325,76 @@ static umode_t nvmem_bin_attr_is_visible(struct kobject *kobj,
 	return nvmem_bin_attr_get_umode(nvmem);
 }
 
+static struct nvmem_cell *nvmem_create_cell(struct nvmem_cell_entry *entry,
+					    const char *id, int index);
+
+static ssize_t nvmem_cell_attr_read(struct file *filp, struct kobject *kobj,
+				    struct bin_attribute *attr, char *buf,
+				    loff_t pos, size_t count)
+{
+	struct nvmem_cell_entry *entry;
+	struct nvmem_cell *cell = NULL;
+	struct nvmem_device *nvmem;
+	size_t cell_sz, read_len;
+	struct device *dev;
+	void *content;
+
+	if (attr->private)
+		dev = attr->private;
+	else
+		dev = kobj_to_dev(kobj);
+	nvmem = to_nvmem_device(dev);
+
+	mutex_lock(&nvmem_mutex);
+	list_for_each_entry(entry, &nvmem->cells, node) {
+		if (strncmp(entry->name, attr->attr.name, XATTR_NAME_MAX))
+			continue;
+
+		cell = nvmem_create_cell(entry, entry->name, 0);
+		if (IS_ERR(cell)) {
+			mutex_unlock(&nvmem_mutex);
+			return PTR_ERR(cell);
+		}
+
+		break;
+	}
+	mutex_unlock(&nvmem_mutex);
+
+	if (!cell)
+		return -EINVAL;
+
+	content = nvmem_cell_read(cell, &cell_sz);
+	if (IS_ERR(content)) {
+		read_len = PTR_ERR(content);
+		goto destroy_cell;
+	}
+
+	read_len = min_t(unsigned int, cell_sz - pos, count);
+	memcpy(buf, content + pos, read_len);
+	kfree(content);
+
+destroy_cell:
+	kfree_const(cell->id);
+	kfree(cell);
+
+	return read_len;
+}
+
+static umode_t nvmem_cells_is_visible(struct kobject *kobj,
+				      struct bin_attribute *attr, int i)
+{
+	struct device *dev = kobj_to_dev(kobj);
+	struct nvmem_device *nvmem = to_nvmem_device(dev);
+	umode_t mode = 0444;
+
+	mutex_lock(&nvmem_mutex);
+	if (list_empty(&nvmem->cells))
+		mode = 0;
+	mutex_unlock(&nvmem_mutex);
+
+	return mode;
+}
+
 /* default read/write permissions */
 static struct bin_attribute bin_attr_rw_nvmem = {
 	.attr	= {
@@ -346,8 +416,15 @@ static const struct attribute_group nvmem_bin_group = {
 	.is_bin_visible = nvmem_bin_attr_is_visible,
 };
 
+/* Cell attributes will be dynamically allocated */
+static struct attribute_group nvmem_cells_group = {
+	.name		= "cells",
+	.is_bin_visible = nvmem_cells_is_visible,
+};
+
 static const struct attribute_group *nvmem_dev_groups[] = {
 	&nvmem_bin_group,
+	&nvmem_cells_group,
 	NULL,
 };
 
@@ -406,6 +483,60 @@ static void nvmem_sysfs_remove_compat(struct nvmem_device *nvmem,
 		device_remove_bin_file(nvmem->base_dev, &nvmem->eeprom);
 }
 
+static int nvmem_populate_sysfs_cells(struct nvmem_device *nvmem)
+{
+	struct bin_attribute **cells_attrs, *attrs;
+	struct nvmem_cell_entry *entry;
+	unsigned int ncells = 0, i = 0;
+	int ret = 0;
+
+	mutex_lock(&nvmem_mutex);
+
+	list_for_each_entry(entry, &nvmem->cells, node)
+		ncells++;
+
+	/* Allocate an array of attributes with a sentinel */
+	cells_attrs = devm_kcalloc(&nvmem->dev, ncells + 1,
+				   sizeof(struct bin_attribute *), GFP_KERNEL);
+	if (!cells_attrs) {
+		ret = -ENOMEM;
+		goto unlock_mutex;
+	}
+
+	nvmem_cells_group.bin_attrs = cells_attrs;
+
+	/* Without exposed cells, successfully exit after assigning an empty attributes array */
+	if (!ncells)
+		goto unlock_mutex;
+
+	attrs = devm_kcalloc(&nvmem->dev, ncells, sizeof(struct bin_attribute), GFP_KERNEL);
+	if (!attrs) {
+		ret = -ENOMEM;
+		goto unlock_mutex;
+	}
+
+	/* Initialize each attribute to take the name and size of the cell */
+	list_for_each_entry(entry, &nvmem->cells, node) {
+		sysfs_bin_attr_init(&attrs[i]);
+		attrs[i].attr.name = devm_kstrdup(&nvmem->dev, entry->name, GFP_KERNEL);
+		attrs[i].attr.mode = 0444;
+		attrs[i].size = entry->bytes;
+		attrs[i].read = &nvmem_cell_attr_read;
+		if (!attrs[i].attr.name) {
+			ret = -ENOMEM;
+			goto unlock_mutex;
+		}
+
+		cells_attrs[i] = &attrs[i];
+		i++;
+	}
+
+unlock_mutex:
+	mutex_unlock(&nvmem_mutex);
+
+	return ret;
+}
+
 #else /* CONFIG_NVMEM_SYSFS */
 
 static int nvmem_sysfs_setup_compat(struct nvmem_device *nvmem,
@@ -976,16 +1107,22 @@ struct nvmem_device *nvmem_register(const struct nvmem_config *config)
 	if (rval)
 		goto err_remove_cells;
 
+	rval = nvmem_add_cells_from_layout(nvmem);
+	if (rval)
+		goto err_remove_cells;
+
+#ifdef CONFIG_NVMEM_SYSFS
+	rval = nvmem_populate_sysfs_cells(nvmem);
+	if (rval)
+		goto err_remove_cells;
+#endif
+
 	dev_dbg(&nvmem->dev, "Registering nvmem device %s\n", config->name);
 
 	rval = device_add(&nvmem->dev);
 	if (rval)
 		goto err_remove_cells;
 
-	rval = nvmem_add_cells_from_layout(nvmem);
-	if (rval)
-		goto err_remove_cells;
-
 	blocking_notifier_call_chain(&nvmem_notifier, NVMEM_ADD, nvmem);
 
 	return nvmem;
-- 
2.34.1

Re: [PATCH v2 2/2] nvmem: core: Expose cells through sysfs

[Miquel Raynal]

Hi Greg,

gregkh@linuxfoundation.org wrote on Wed, 31 May 2023 20:16:37 +0100:

> On Tue, May 30, 2023 at 12:09:29PM +0200, Miquel Raynal wrote:
> > The binary content of nvmem devices is available to the user so in the
> > easiest cases, finding the content of a cell is rather easy as it is
> > just a matter of looking at a known and fixed offset. However, nvmem
> > layouts have been recently introduced to cope with more advanced
> > situations, where the offset and size of the cells is not known in
> > advance or is dynamic. When using layouts, more advanced parsers are
> > used by the kernel in order to give direct access to the content of each
> > cell, regardless of its position/size in the underlying
> > device. Unfortunately, these information are not accessible by users,
> > unless by fully re-implementing the parser logic in userland.
> > 
> > Let's expose the cells and their content through sysfs to avoid these
> > situations.
> > 
> > Exposed cells are read-only. There is, in practice, everything in the
> > core to support a write path, but as I don't see any need for that, I
> > prefer to keep the interface simple (and probably safer). The interface
> > is documented as being in the "testing" state which means we can later
> > add a write attribute if though relevant.
> > 
> > Of course the relevant NVMEM sysfs Kconfig option must be enabled for
> > this support to be compiled-in.
> > 
> > The current implementation leads to the 'cells' folder to be always
> > present even when no cell is actually exposed. This is due to a sysfs
> > limitation which might in the future be overcome. In order to be as
> > close as how sysfs was designed, despite the cells .bin_attrs attribute
> > group member being assigned at runtime (and thus, not movable to a RO
> > section), we provide a .is_bin_visible hook which might become really
> > useful if the series avoiding the creation of empty directories borns.
> > 
> > There is one limitation though: if a layout is built as a module but is
> > not properly installed in the system and loaded manually with insmod
> > while the nvmem device driver was built-in, the cells won't appear in
> > sysfs. But if done like that, the cells won't be usable by the built-in
> > kernel drivers anyway.
> > 
> > Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
> > ---
> >  drivers/nvmem/core.c | 145 +++++++++++++++++++++++++++++++++++++++++--
> >  1 file changed, 141 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/nvmem/core.c b/drivers/nvmem/core.c
> > index 342cd380b420..02b70f065ebc 100644
> > --- a/drivers/nvmem/core.c
> > +++ b/drivers/nvmem/core.c
> > @@ -325,6 +325,76 @@ static umode_t nvmem_bin_attr_is_visible(struct kobject *kobj,
> >  	return nvmem_bin_attr_get_umode(nvmem);
> >  }
> >  
> > +static struct nvmem_cell *nvmem_create_cell(struct nvmem_cell_entry *entry,
> > +					    const char *id, int index);
> > +
> > +static ssize_t nvmem_cell_attr_read(struct file *filp, struct kobject *kobj,
> > +				    struct bin_attribute *attr, char *buf,
> > +				    loff_t pos, size_t count)
> > +{
> > +	struct nvmem_cell_entry *entry;
> > +	struct nvmem_cell *cell = NULL;
> > +	struct nvmem_device *nvmem;
> > +	size_t cell_sz, read_len;
> > +	struct device *dev;
> > +	void *content;
> > +
> > +	if (attr->private)
> > +		dev = attr->private;
> > +	else
> > +		dev = kobj_to_dev(kobj);
> > +	nvmem = to_nvmem_device(dev);
> > +
> > +	mutex_lock(&nvmem_mutex);
> > +	list_for_each_entry(entry, &nvmem->cells, node) {
> > +		if (strncmp(entry->name, attr->attr.name, XATTR_NAME_MAX))
> > +			continue;
> > +
> > +		cell = nvmem_create_cell(entry, entry->name, 0);
> > +		if (IS_ERR(cell)) {
> > +			mutex_unlock(&nvmem_mutex);
> > +			return PTR_ERR(cell);
> > +		}
> > +
> > +		break;
> > +	}
> > +	mutex_unlock(&nvmem_mutex);
> > +
> > +	if (!cell)
> > +		return -EINVAL;
> > +
> > +	content = nvmem_cell_read(cell, &cell_sz);
> > +	if (IS_ERR(content)) {
> > +		read_len = PTR_ERR(content);
> > +		goto destroy_cell;
> > +	}
> > +
> > +	read_len = min_t(unsigned int, cell_sz - pos, count);
> > +	memcpy(buf, content + pos, read_len);
> > +	kfree(content);
> > +
> > +destroy_cell:
> > +	kfree_const(cell->id);
> > +	kfree(cell);
> > +
> > +	return read_len;
> > +}
> > +
> > +static umode_t nvmem_cells_is_visible(struct kobject *kobj,
> > +				      struct bin_attribute *attr, int i)
> > +{
> > +	struct device *dev = kobj_to_dev(kobj);
> > +	struct nvmem_device *nvmem = to_nvmem_device(dev);
> > +	umode_t mode = 0444;
> > +
> > +	mutex_lock(&nvmem_mutex);
> > +	if (list_empty(&nvmem->cells))
> > +		mode = 0;
> > +	mutex_unlock(&nvmem_mutex);  
> 
> As the list can change right after you have unlocked this, why care
> about the locking at all?

Just to mimic the existing code which protects this list. I don't think
a race can happen here anyway, the locks are not needed indeed.

> 
> But in looking deeper here, is this really even needed?  As you manually
> create the attributes in here anyway, the group starts out empty and
> then you manually add them, so this should never fail, right?

Absolutely. I put these lines in the commit log:

"In order to be as close as how sysfs was designed, despite the cells
.bin_attrs attribute group member being assigned at runtime (and thus,
not movable to a RO section), we provide a .is_bin_visible hook which
might become really useful if the series avoiding the creation of empty
directories borns."

It was a try to prepare the future :) But I agree it is not needed,
statically defining the rights is more than enough, so I'll just get
rid of it.

> 
> > +
> > +	return mode;
> > +}
> > +
> >  /* default read/write permissions */
> >  static struct bin_attribute bin_attr_rw_nvmem = {
> >  	.attr	= {
> > @@ -346,8 +416,15 @@ static const struct attribute_group
> > nvmem_bin_group = { .is_bin_visible = nvmem_bin_attr_is_visible,
> >  };
> >  
> > +/* Cell attributes will be dynamically allocated */
> > +static struct attribute_group nvmem_cells_group = {
> > +	.name		= "cells",
> > +	.is_bin_visible = nvmem_cells_is_visible,
> > +};
> > +
> >  static const struct attribute_group *nvmem_dev_groups[] = {
> >  	&nvmem_bin_group,
> > +	&nvmem_cells_group,
> >  	NULL,
> >  };
> >  
> > @@ -406,6 +483,60 @@ static void nvmem_sysfs_remove_compat(struct
> > nvmem_device *nvmem, device_remove_bin_file(nvmem->base_dev,
> > &nvmem->eeprom); }
> >  
> > +static int nvmem_populate_sysfs_cells(struct nvmem_device *nvmem)
> > +{
> > +	struct bin_attribute **cells_attrs, *attrs;
> > +	struct nvmem_cell_entry *entry;
> > +	unsigned int ncells = 0, i = 0;
> > +	int ret = 0;
> > +
> > +	mutex_lock(&nvmem_mutex);
> > +
> > +	list_for_each_entry(entry, &nvmem->cells, node)
> > +		ncells++;
> > +
> > +	/* Allocate an array of attributes with a sentinel */
> > +	cells_attrs = devm_kcalloc(&nvmem->dev, ncells + 1,
> > +				   sizeof(struct bin_attribute *),
> > GFP_KERNEL);
> > +	if (!cells_attrs) {
> > +		ret = -ENOMEM;
> > +		goto unlock_mutex;
> > +	}
> > +
> > +	nvmem_cells_group.bin_attrs = cells_attrs;
> > +
> > +	/* Without exposed cells, successfully exit after
> > assigning an empty attributes array */
> > +	if (!ncells)
> > +		goto unlock_mutex;  
> 
> Shouldn't this check be higher up _before_ you allocate any memory?
> If the attribute group list is empty, nothing should be created,
> right?  Or will the driver core crash?

As you rightfully guessed it, the core will crash if no list is
provided at all. I need to provide an empty list with just an empty
member and everything goes smoothly. 

Thanks,
Miquèl

Re: [PATCH v2 2/2] nvmem: core: Expose cells through sysfs

[Greg Kroah-Hartman]

On Thu, Jun 01, 2023 at 10:51:14AM +0200, Miquel Raynal wrote:
> > > +	/* Without exposed cells, successfully exit after
> > > assigning an empty attributes array */
> > > +	if (!ncells)
> > > +		goto unlock_mutex;  
> > 
> > Shouldn't this check be higher up _before_ you allocate any memory?
> > If the attribute group list is empty, nothing should be created,
> > right?  Or will the driver core crash?
> 
> As you rightfully guessed it, the core will crash if no list is
> provided at all. I need to provide an empty list with just an empty
> member and everything goes smoothly. 

Let's fix the core, it shouldn't crash like that :)

thanks,

greg k-h

Re: [PATCH v2 2/2] nvmem: core: Expose cells through sysfs

[Miquel Raynal]

Hi Greg,

gregkh@linuxfoundation.org wrote on Thu, 1 Jun 2023 10:03:01 +0100:

> On Thu, Jun 01, 2023 at 10:51:14AM +0200, Miquel Raynal wrote:
> > > > +	/* Without exposed cells, successfully exit after
> > > > assigning an empty attributes array */
> > > > +	if (!ncells)
> > > > +		goto unlock_mutex;  
> > > 
> > > Shouldn't this check be higher up _before_ you allocate any memory?
> > > If the attribute group list is empty, nothing should be created,
> > > right?  Or will the driver core crash?
> > 
> > As you rightfully guessed it, the core will crash if no list is
> > provided at all. I need to provide an empty list with just an empty
> > member and everything goes smoothly. 
> 
> Let's fix the core, it shouldn't crash like that :)

Perfectly fine by me, I'll give this idea a try.

Thanks,
Miquèl