[PATCH 01/13] proc tty: introduce ->proc_fops

[PATCH 01/13] proc tty: introduce ->proc_fops

[Alexey Dobriyan]

Add struct tty_operations:proc_fops .

The intent is gradual switch of TTY drivers from ->read_proc usage.
proc entries are created with proc_create_data() which even fixes
early-read races.

Eventually ->read_proc code will be removed from TTY code, thus helping
remove ->read_proc from whole proc code.

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
---

 fs/proc/proc_tty.c         |   21 +++++++++++++--------
 include/linux/tty_driver.h |    3 +++
 2 files changed, 16 insertions(+), 8 deletions(-)

--- a/fs/proc/proc_tty.c
+++ b/fs/proc/proc_tty.c
@@ -144,17 +144,22 @@ void proc_tty_register_driver(struct tty_driver *driver)
 {
 	struct proc_dir_entry *ent;
 		
-	if (!driver->ops->read_proc || !driver->driver_name ||
-	    driver->proc_entry)
+	if (!driver->driver_name || driver->proc_entry)
 		return;
-
-	ent = create_proc_entry(driver->driver_name, 0, proc_tty_driver);
-	if (!ent)
+	if (driver->ops->proc_fops) {
+		ent = proc_create_data(driver->driver_name, 0, proc_tty_driver,
+				       driver->ops->proc_fops, driver);
+		if (!ent)
+			return;
+	} else if (driver->ops->read_proc) {
+		ent = create_proc_entry(driver->driver_name, 0, proc_tty_driver);
+		if (!ent)
+			return;
+		ent->read_proc = driver->ops->read_proc;
+		ent->data = driver;
+	} else
 		return;
-	ent->read_proc = driver->ops->read_proc;
 	ent->owner = driver->owner;
-	ent->data = driver;
-
 	driver->proc_entry = ent;
 }
 
--- a/include/linux/tty_driver.h
+++ b/include/linux/tty_driver.h
@@ -203,6 +203,9 @@ struct tty_operations {
 	void (*send_xchar)(struct tty_struct *tty, char ch);
 	int (*read_proc)(char *page, char **start, off_t off,
 			  int count, int *eof, void *data);
+#ifdef CONFIG_PROC_FS
+	const struct file_operations *proc_fops;
+#endif
 	int (*tiocmget)(struct tty_struct *tty, struct file *file);
 	int (*tiocmset)(struct tty_struct *tty, struct file *file,
 			unsigned int set, unsigned int clear);
-- 
1.5.4.5

Re: [PATCH 01/13] proc tty: introduce ->proc_fops

[Alan Cox]

On Tue, 29 Jul 2008 05:29:08 +0400
Alexey Dobriyan <adobriyan@gmail.com> wrote:

> Add struct tty_operations:proc_fops .
> 
> The intent is gradual switch of TTY drivers from ->read_proc usage.
> proc entries are created with proc_create_data() which even fixes
> early-read races.
> 
> Eventually ->read_proc code will be removed from TTY code, thus helping
> remove ->read_proc from whole proc code.

I'm going to NAK this but not because I think the concept is wrong. I
think the way its been done is perhaps wrong.

You've added ifdefs to a lot of drivers and more basically duplicate
code. Is there a reason you can't keep the ->read_proc method in the tty
code but as a function called by a single instance of proc_fops and
seq_file methods for the whole tty driver layer.

Ie have a single tty seq_file method that calls driver->ops->read_proc as
the seqfile iterator ?

>  int count, int *eof, void *data);
> +#ifdef CONFIG_PROC_FS
> +	const struct file_operations *proc_fops;
> +#endif

I'd prefer you didn't put variables in the middle of the methods. Also
the ifdef isn't needed. It's not worth one pointer to create a load of
ifdefs

Alan

Re: [PATCH 01/13] proc tty: introduce ->proc_fops

[Alexey Dobriyan]

On Tue, Jul 29, 2008 at 10:53:50AM +0100, Alan Cox wrote:
> On Tue, 29 Jul 2008 05:29:08 +0400
> Alexey Dobriyan <adobriyan@gmail.com> wrote:
> 
> > Add struct tty_operations:proc_fops .
> > 
> > The intent is gradual switch of TTY drivers from ->read_proc usage.
> > proc entries are created with proc_create_data() which even fixes
> > early-read races.
> > 
> > Eventually ->read_proc code will be removed from TTY code, thus helping
> > remove ->read_proc from whole proc code.
> 
> I'm going to NAK this but not because I think the concept is wrong. I
> think the way its been done is perhaps wrong.
> 
> You've added ifdefs to a lot of drivers and more basically duplicate
> code.

It's boileplate code which is hard to screwup, unlike boilerplate at the
end of every ->read_proc hook.

> Is there a reason you can't keep the ->read_proc method in the tty
> code but as a function called by a single instance of proc_fops and
> seq_file methods for the whole tty driver layer.
> 
> Ie have a single tty seq_file method that calls driver->ops->read_proc as
> the seqfile iterator ?

I think it should be struct tty_operations::proc_fops .

In fact, posted conversions are bit incorrect.

Let's look at cyclades. It prints banner, then information about each
card. The very right way to print it via seq_files is implement
struct seq_operations::start, next, stop, show.

If TTY driver prints something simple, just one show hook is enough.
If TTY driver wants something more complex, right way start to be
seq_operations and one hook is not enough anymore.

But I don't really care about this exact issue, as long as ->read_proc
TTY hook is gone as ->read_proc proc hook.

> >  int count, int *eof, void *data);
> > +#ifdef CONFIG_PROC_FS
> > +	const struct file_operations *proc_fops;
> > +#endif
> 
> I'd prefer you didn't put variables in the middle of the methods. Also
> the ifdef isn't needed. It's not worth one pointer to create a load of
> ifdefs

OK.