[PATCH] video: fbdev: Request IO ports

[PATCH] video: fbdev: Request IO ports

[Daniel Matyas]

Some fbdev drivers do not request their memory correctly. Request the IO
ports' memory so no other driver may access them. Only request 1 byte,
because the inb() and out() are used for reading and writing.

Did not test on hardware.

Signed-off-by: Daniel Matyas <matyasdaniel006@gmail.com>
---
 drivers/video/fbdev/arcfb.c | 8 ++++++++
 drivers/video/fbdev/n411.c  | 5 +++++
 2 files changed, 13 insertions(+)

diff --git a/drivers/video/fbdev/arcfb.c b/drivers/video/fbdev/arcfb.c
index b807cf07522d..6771a7409cde 100644
--- a/drivers/video/fbdev/arcfb.c
+++ b/drivers/video/fbdev/arcfb.c
@@ -481,6 +481,14 @@ static int arcfb_probe(struct platform_device *dev)
 		printk(KERN_WARNING "no IO addresses supplied\n");
 		goto err_addr;
 	}
+
+	if (!devm_request_region(&dev->dev, dio_addr, 1, "arcfb") ||
+		!devm_request_region(&dev->dev, cio_addr, 1, "arcfb") ||
+		!devm_request_region(&dev->dev, c2io_addr, 1, "arcfb")) {
+		retval = -EBUSY;
+		goto err_addr;
+	}
+
 	par->dio_addr = dio_addr;
 	par->cio_addr = cio_addr;
 	par->c2io_addr = c2io_addr;
diff --git a/drivers/video/fbdev/n411.c b/drivers/video/fbdev/n411.c
index a3677313396e..ed0281939d37 100644
--- a/drivers/video/fbdev/n411.c
+++ b/drivers/video/fbdev/n411.c
@@ -165,6 +165,11 @@ static int __init n411_init(void)
 	if (!n411_device)
 		return -ENOMEM;
 
+	if (!devm_request_region(&n411_device->dev, dio_addr, 1, "n411") ||
+		!devm_request_region(&n411_device->dev, cio_addr, 1, "n411") ||
+		!devm_request_region(&n411_device->dev, c2io_addr, 1, "n411"))
+		return -EBUSY;
+
 	ret = platform_device_add_data(n411_device, &n411_board,
 				       sizeof(n411_board));
 	if (ret)
-- 
2.55.0

Re: [PATCH] video: fbdev: Request IO ports

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 2 potential
issues to consider:
- [High] fbdev: n411: devm_request_region() in n411_init() breaks
  hecubafb probing
- [Medium] fbdev: n411: leak of n411_device and I/O regions in
  n411_init() on error
--

commit 233a6d07f214976aca6e2d815559082f7aaf9dc7
Author: Daniel Matyas <matyasdaniel006@gmail.com>

video: fbdev: Request IO ports

This commit requests the I/O port regions used by the arcfb and n411
framebuffer drivers via devm_request_region().

> diff --git a/drivers/video/fbdev/n411.c b/drivers/video/fbdev/n411.c
> index a3677313396e2..ed0281939d37d 100644
> --- a/drivers/video/fbdev/n411.c
> +++ b/drivers/video/fbdev/n411.c
> @@ -165,6 +165,11 @@ static int __init n411_init(void)
>  	if (!n411_device)
>  		return -ENOMEM;
>  
> +	if (!devm_request_region(&n411_device->dev, dio_addr, 1, "n411") ||
> +		!devm_request_region(&n411_device->dev, cio_addr, 1, "n411") ||
> +		!devm_request_region(&n411_device->dev, c2io_addr, 1, "n411"))

[Severity: High]
Does calling devm_request_region() on &n411_device->dev before
platform_device_add() prevent the hecubafb driver from binding to
n411_device?

Each devm_request_region() call adds an entry to
n411_device->dev.devres_head.  When platform_device_add(n411_device) later
attempts to bind the driver, really_probe() rejects any device that already
has devres entries:

drivers/base/dd.c:really_probe() {
	...
	if (!list_empty(&dev->devres_head)) {
		dev_crit(dev, "Resources present before probing\n");
		ret = -EBUSY;
		goto done;
	}
	...
}

As a result, probing always fails with -EBUSY and the display device remains
unbound.

> +		return -EBUSY;

[Severity: Medium]
Does returning -EBUSY directly here leak n411_device and any I/O port
regions already claimed by the preceding devm_request_region() calls?
Jumping to put_plat_device instead would call platform_device_put() and
release the allocated device and its managed resources.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261111150219.239706-1-matyasdaniel006@gmail.com?part=1