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

From: Miquel Raynal
Date: Thu Jun 01 2023 - 04:51:23 EST


Hi Greg,

gregkh@xxxxxxxxxxxxxxxxxxx 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@xxxxxxxxxxx>
> > ---
> > 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