[PATCH 0/1] rpmsg: char - treat `ENOMEM` as `EAGAIN`

[PATCH 0/1] rpmsg: char - treat `ENOMEM` as `EAGAIN`

[Tim Blechmann]

This patch should allow the use of `rpmsg_char` character devices with
libraries that do a `poll`/`write` loop and expect `EAGAIN` when sending
fails and the user space application needs to `poll` to wait for more
space to be available.
`boost::asio::write` is a notable example of a library, which implements
such a loop.


Tim Blechmann (1):
  rpmsg: char - treat `ENOMEM` as `EAGAIN`

 drivers/rpmsg/rpmsg_char.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

-- 
2.35.1

[PATCH 1/1] rpmsg: char - treat `ENOMEM` as `EAGAIN`

[Tim Blechmann]

`rpmsg_trysend` returns `-ENOMEM` when no rpmsg buffer can be allocated.
this causes `::write` to fail with this error as opposed to `-EAGAIN`.
this is what user space applications (and libraries like boost.asio)
would expect when using normal character devices.

Signed-off-by: Tim Blechmann <tim@klingt.org>
CC: Arnaud Pouliquen <arnaud.pouliquen@foss.st.com>
---
 drivers/rpmsg/rpmsg_char.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/rpmsg/rpmsg_char.c b/drivers/rpmsg/rpmsg_char.c
index 5663cf799c95..5b9e708d595a 100644
--- a/drivers/rpmsg/rpmsg_char.c
+++ b/drivers/rpmsg/rpmsg_char.c
@@ -239,14 +239,17 @@ static ssize_t rpmsg_eptdev_write_iter(struct kiocb *iocb,
 
 	if (!eptdev->ept) {
 		ret = -EPIPE;
 		goto unlock_eptdev;
 	}
 
-	if (filp->f_flags & O_NONBLOCK)
+	if (filp->f_flags & O_NONBLOCK) {
 		ret = rpmsg_trysendto(eptdev->ept, kbuf, len, eptdev->chinfo.dst);
+		if (ret == -ENOMEM)
+			ret = -EAGAIN;
+	}
 	else
 		ret = rpmsg_sendto(eptdev->ept, kbuf, len, eptdev->chinfo.dst);
 
 unlock_eptdev:
 	mutex_unlock(&eptdev->ept_lock);
 
-- 
2.35.1

Re: [PATCH 1/1] rpmsg: char - treat `ENOMEM` as `EAGAIN`

[Bjorn Andersson]

On Mon 14 Feb 03:30 CST 2022, Tim Blechmann wrote:

> `rpmsg_trysend` returns `-ENOMEM` when no rpmsg buffer can be allocated.

Please use the form rpmsg_trysend() - without ``. You can omit that
around -ENOMEM as well.

> this causes `::write` to fail with this error as opposed to `-EAGAIN`.

Please drop the :: here as well.

> this is what user space applications (and libraries like boost.asio)
> would expect when using normal character devices.
> 
> Signed-off-by: Tim Blechmann <tim@klingt.org>

I attempted to fix up the above details, but unfortunately your
Signed-off-by doesn't match your From:, so I can't apply the patch
anyways.

Can you please resubmit this with appropriate author/s-o-b?

> CC: Arnaud Pouliquen <arnaud.pouliquen@foss.st.com>
> ---
>  drivers/rpmsg/rpmsg_char.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/rpmsg/rpmsg_char.c b/drivers/rpmsg/rpmsg_char.c
> index 5663cf799c95..5b9e708d595a 100644
> --- a/drivers/rpmsg/rpmsg_char.c
> +++ b/drivers/rpmsg/rpmsg_char.c
> @@ -239,14 +239,17 @@ static ssize_t rpmsg_eptdev_write_iter(struct kiocb *iocb,
>  
>  	if (!eptdev->ept) {
>  		ret = -EPIPE;
>  		goto unlock_eptdev;
>  	}
>  
> -	if (filp->f_flags & O_NONBLOCK)
> +	if (filp->f_flags & O_NONBLOCK) {
>  		ret = rpmsg_trysendto(eptdev->ept, kbuf, len, eptdev->chinfo.dst);
> +		if (ret == -ENOMEM)
> +			ret = -EAGAIN;
> +	}
>  	else
>  		ret = rpmsg_sendto(eptdev->ept, kbuf, len, eptdev->chinfo.dst);

./script/checkpatch.pl --strict tells me that you should have {} around
the else block as well..

Thanks,
Bjorn

>  
>  unlock_eptdev:
>  	mutex_unlock(&eptdev->ept_lock);
>  
> -- 
> 2.35.1
> 

[PATCH 0/1 v2] rpmsg: char - treat `ENOMEM` as `EAGAIN`

[tim.blechmann]

From: Tim Blechmann <tim@klingt.org>

This patch should allow the use of `rpmsg_char` character devices with
libraries that do a `poll`/`write` loop and expect `EAGAIN` when sending
fails and the user space application needs to `poll` to wait for more
space to be available.
`boost::asio::write` is a notable example of a library, which implements
such a loop.


Tim Blechmann (1):
  rpmsg: char - treat `ENOMEM` as `EAGAIN`

 drivers/rpmsg/rpmsg_char.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

-- 
2.35.1

[PATCH 1/1] rpmsg: char - treat ENOMEM as EAGAIN

[tim.blechmann]

From: Tim Blechmann <tim@klingt.org>

rpmsg_trysend() returns -ENOMEM when no rpmsg buffer can be allocated.
this causes write to fail with this error as opposed to -EAGAIN.
this is what user space applications (and libraries like boost.asio)
would expect when using normal character devices.

Signed-off-by: Tim Blechmann <tim@klingt.org>
CC: Arnaud Pouliquen <arnaud.pouliquen@foss.st.com>
---
 drivers/rpmsg/rpmsg_char.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/rpmsg/rpmsg_char.c b/drivers/rpmsg/rpmsg_char.c
index 5663cf799c95..5b9e708d595a 100644
--- a/drivers/rpmsg/rpmsg_char.c
+++ b/drivers/rpmsg/rpmsg_char.c
@@ -239,14 +239,17 @@ static ssize_t rpmsg_eptdev_write_iter(struct kiocb *iocb,
 
 	if (!eptdev->ept) {
 		ret = -EPIPE;
 		goto unlock_eptdev;
 	}
 
-	if (filp->f_flags & O_NONBLOCK)
+	if (filp->f_flags & O_NONBLOCK) {
 		ret = rpmsg_trysendto(eptdev->ept, kbuf, len, eptdev->chinfo.dst);
+		if (ret == -ENOMEM)
+			ret = -EAGAIN;
+	}
 	else
 		ret = rpmsg_sendto(eptdev->ept, kbuf, len, eptdev->chinfo.dst);
 
 unlock_eptdev:
 	mutex_unlock(&eptdev->ept_lock);
 
-- 
2.35.1