2021-09-21 10:35:39

by Tony Lindgren

Subject: [PATCH 1/6] n_tty: Start making use of -EAGAIN returned from process_output_block()

We check for -EAGAIN in n_tty_write() but never currently get it from
process_output_block(). Let's add -EAGAIN handling and break out with 0
bytes processed. Note that if we return -EAGAIN from n_tty_write(), it
will be treated as error by the caller rather than a retry.

Looking at the patch description for commit 9ef8927f45f2 ("n_tty: check
for negative and zero space return from tty_write_room") it looks like we
have not made use of -EGAIN from process_output_block() so far, so this
does not seem like it's currently needed as a fix.

We can use -EAGAIN for serial layer power management changes as we now
will make use of write_room() returning 0 for an idled serial port.

Signed-off-by: Tony Lindgren <[email protected]>
---
drivers/tty/n_tty.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/drivers/tty/n_tty.c b/drivers/tty/n_tty.c
--- a/drivers/tty/n_tty.c
+++ b/drivers/tty/n_tty.c
@@ -549,6 +549,8 @@ static ssize_t process_output_block(struct tty_struct *tty,
space = tty_write_room(tty);
if (space <= 0) {
mutex_unlock(&ldata->output_lock);
+ if (!space)
+ space = -EAGAIN;
return space;
}
if (nr > space)
@@ -2287,8 +2289,10 @@ static ssize_t n_tty_write(struct tty_struct *tty, struct file *file,
while (nr > 0) {
ssize_t num = process_output_block(tty, b, nr);
if (num < 0) {
- if (num == -EAGAIN)
- break;
+ if (num == -EAGAIN) {
+ retval = 0;
+ goto break_out;
+ }
retval = num;
goto break_out;
}
--
2.33.0

2021-09-21 12:00:33

by Andy Shevchenko

Subject: Re: [PATCH 1/6] n_tty: Start making use of -EAGAIN returned from process_output_block()

On Tue, Sep 21, 2021 at 1:34 PM Tony Lindgren <[email protected]> wrote:
>
> We check for -EAGAIN in n_tty_write() but never currently get it from
> process_output_block(). Let's add -EAGAIN handling and break out with 0
> bytes processed. Note that if we return -EAGAIN from n_tty_write(), it
> will be treated as error by the caller rather than a retry.
>
> Looking at the patch description for commit 9ef8927f45f2 ("n_tty: check
> for negative and zero space return from tty_write_room") it looks like we
> have not made use of -EGAIN from process_output_block() so far, so this

-EGAIN?

> does not seem like it's currently needed as a fix.
>
> We can use -EAGAIN for serial layer power management changes as we now
> will make use of write_room() returning 0 for an idled serial port.
>
> Signed-off-by: Tony Lindgren <[email protected]>
> ---
> drivers/tty/n_tty.c | 8 ++++++--
> 1 file changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/tty/n_tty.c b/drivers/tty/n_tty.c
> --- a/drivers/tty/n_tty.c
> +++ b/drivers/tty/n_tty.c
> @@ -549,6 +549,8 @@ static ssize_t process_output_block(struct tty_struct *tty,
> space = tty_write_room(tty);
> if (space <= 0) {
> mutex_unlock(&ldata->output_lock);
> + if (!space)
> + space = -EAGAIN;
> return space;
> }
> if (nr > space)
> @@ -2287,8 +2289,10 @@ static ssize_t n_tty_write(struct tty_struct *tty, struct file *file,
> while (nr > 0) {
> ssize_t num = process_output_block(tty, b, nr);
> if (num < 0) {
> - if (num == -EAGAIN)
> - break;
> + if (num == -EAGAIN) {
> + retval = 0;
> + goto break_out;
> + }
> retval = num;
> goto break_out;
> }
> --
> 2.33.0



--
With Best Regards,
Andy Shevchenko