Re: [PATCH v2 1/2] eeprom: Add IDT 89HPESx EEPROM/CSR driver

[Greg KH]

On Wed, Nov 30, 2016 at 12:16:25AM +0300, Serge Semin wrote:
> On Tue, Nov 29, 2016 at 08:37:50PM +0100, Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > On Tue, Nov 29, 2016 at 01:38:20AM +0300, Serge Semin wrote:
> > > +struct idt_89hpesx_dev {
> > > +	u32 eesize;
> > > +	bool eero;
> > > +	u8 eeaddr;
> > > +
> > > +	u8 inieecmd;
> > > +	u8 inicsrcmd;
> > > +	u8 iniccode;
> > > +
> > > +	atomic_t csr;
> > > +
> > > +	int (*smb_write)(struct idt_89hpesx_dev *, const struct idt_smb_seq *);
> > > +	int (*smb_read)(struct idt_89hpesx_dev *, struct idt_smb_seq *);
> > > +	struct mutex smb_mtx;
> > > +
> > > +	struct i2c_client *client;
> > > +
> > > +	struct bin_attribute *ee_file;
> > > +	struct dentry *csr_dir;
> > > +	struct dentry *csr_file;
> > > +};
> > > +#define to_pdev_kobj(__kobj) \
> > > +	dev_get_drvdata(container_of(__kobj, struct device, kobj))
> > 
> > Is it a struct device, or a kobject?  This is totally confusing to me.
> > 
> > And can't you just use kobj_to_dev()?
> > 
> 
> I just didn't know about kobj_to_dev() inline function. Totally agree that
> container_of() should be replaced with it.
> What does look confusing to you? Do you mean the name "to_pdev_kobj" of the
> macro?

Yes, the macro is odd.  As you are doing two different things here, just
spell it out in the code and use kobj_to_dev() to make it easier to
read please.

> > > +/*
> > > + * eeprom_attribute - EEPROM sysfs-node attributes
> > > + *
> > > + * NOTE Size will be changed in compliance with OF node. EEPROM attribute will
> > > + * be read-only as well if the corresponding flag is specified in OF node.
> > > + */
> > > +BIN_ATTR(eeprom, 0644, idt_sysfs_eeprom_read, idt_sysfs_eeprom_write,
> > > +	 EEPROM_DEF_SIZE);
> > 
> > static?
> > 
> > And BIN_ATTR_RW()?
> > 
> > thanks,
> > 
> > greg k-h
> 
> Of course it should be static. Thanks for noticing that.
> But I intentionally utilized BIN_ATTR() instead of BIN_ATTR_RW(), because
> the last one implies to define the read/write methods with names
> "_name##_read"/"_name##_write", which totally get out of naming within the
> driver source code.

That's ok, use the names the macro wants you to, that's the best way,
and it ensures that I don't have to audit your permissions are correct
for the file.

> To tell the truth macro BIN_ATTR_RW() isn't that popular in the
> kernel.

Yes, but it should be, I have patches floating around somewhere to fix
almost all of these up.

> Neither is BIN_ATTR() macro, but it suites my driver better than the
> another one.

a "raw" BIN_ATTR() shouldn't be used either, please use the _RW()
variant.

thanks,

greg k-h