2013-10-30 10:59:10

by Thierry Reding

[permalink] [raw]
Subject: [PATCH] drm/sysfs: Do not drop device reference twice

device_unregister() already drops its reference to the struct device, so
explicitly calling put_device() before device_unregister() can cause the
device to have been freed before it can be unregistered.

Signed-off-by: Thierry Reding <[email protected]>
---
drivers/gpu/drm/drm_sysfs.c | 2 --
1 file changed, 2 deletions(-)

diff --git a/drivers/gpu/drm/drm_sysfs.c b/drivers/gpu/drm/drm_sysfs.c
index dae42c7..db1c8f9 100644
--- a/drivers/gpu/drm/drm_sysfs.c
+++ b/drivers/gpu/drm/drm_sysfs.c
@@ -439,7 +439,6 @@ err_out_files:
device_remove_file(connector->kdev, &connector_attrs_opt1[i]);
for (i = 0; i < attr_cnt; i++)
device_remove_file(connector->kdev, &connector_attrs[i]);
- put_device(connector->kdev);
device_unregister(connector->kdev);

out:
@@ -472,7 +471,6 @@ void drm_sysfs_connector_remove(struct drm_connector *connector)
for (i = 0; i < ARRAY_SIZE(connector_attrs); i++)
device_remove_file(connector->kdev, &connector_attrs[i]);
sysfs_remove_bin_file(&connector->kdev->kobj, &edid_attr);
- put_device(connector->kdev);
device_unregister(connector->kdev);
connector->kdev = NULL;
}
--
1.8.4


2013-10-30 16:05:06

by Paulo Zanoni

[permalink] [raw]
Subject: Re: [PATCH] drm/sysfs: Do not drop device reference twice

2013/10/30 Thierry Reding <[email protected]>:
> device_unregister() already drops its reference to the struct device, so
> explicitly calling put_device() before device_unregister() can cause the
> device to have been freed before it can be unregistered.
>
> Signed-off-by: Thierry Reding <[email protected]>

I started investigating this problem yesterday and reached the same
conclusion. The connector path can be easily reproduced on i915.ko:
get a machine that has an eDP panel, physically disconnect the panel,
boot the machine, "modprobe i915" and watch the segfault.

Reviewed-by: Paulo Zanoni <[email protected]>
Tested-by: Paulo Zanoni <[email protected]>

I didn't really bisect, but I believe this is probably a regression
from "drm/sysfs: sort out minor and connector device object
lifetimes".

And kudos to whoever invented CONFIG_DEBUG_KOBJECT :)

> ---
> drivers/gpu/drm/drm_sysfs.c | 2 --
> 1 file changed, 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_sysfs.c b/drivers/gpu/drm/drm_sysfs.c
> index dae42c7..db1c8f9 100644
> --- a/drivers/gpu/drm/drm_sysfs.c
> +++ b/drivers/gpu/drm/drm_sysfs.c
> @@ -439,7 +439,6 @@ err_out_files:
> device_remove_file(connector->kdev, &connector_attrs_opt1[i]);
> for (i = 0; i < attr_cnt; i++)
> device_remove_file(connector->kdev, &connector_attrs[i]);
> - put_device(connector->kdev);
> device_unregister(connector->kdev);
>
> out:
> @@ -472,7 +471,6 @@ void drm_sysfs_connector_remove(struct drm_connector *connector)
> for (i = 0; i < ARRAY_SIZE(connector_attrs); i++)
> device_remove_file(connector->kdev, &connector_attrs[i]);
> sysfs_remove_bin_file(&connector->kdev->kobj, &edid_attr);
> - put_device(connector->kdev);
> device_unregister(connector->kdev);
> connector->kdev = NULL;
> }
> --
> 1.8.4
>
> _______________________________________________
> dri-devel mailing list
> [email protected]
> http://lists.freedesktop.org/mailman/listinfo/dri-devel



--
Paulo Zanoni

2013-10-30 18:06:46

by Ben Widawsky

[permalink] [raw]
Subject: Re: [PATCH] drm/sysfs: Do not drop device reference twice

On Wed, Oct 30, 2013 at 11:59:05AM +0100, Thierry Reding wrote:
> device_unregister() already drops its reference to the struct device, so
> explicitly calling put_device() before device_unregister() can cause the
> device to have been freed before it can be unregistered.
>
> Signed-off-by: Thierry Reding <[email protected]>

Thanks for fixing this. It was driving me nuts.
Tested-by: Ben Widawsky <[email protected]>

[snip]
--
Ben Widawsky, Intel Open Source Technology Center

2013-10-30 21:18:19

by Thierry Reding

[permalink] [raw]
Subject: Re: [PATCH] drm/sysfs: Do not drop device reference twice

On Wed, Oct 30, 2013 at 02:05:02PM -0200, Paulo Zanoni wrote:
> 2013/10/30 Thierry Reding <[email protected]>:
> > device_unregister() already drops its reference to the struct device, so
> > explicitly calling put_device() before device_unregister() can cause the
> > device to have been freed before it can be unregistered.
> >
> > Signed-off-by: Thierry Reding <[email protected]>
>
> I started investigating this problem yesterday and reached the same
> conclusion. The connector path can be easily reproduced on i915.ko:
> get a machine that has an eDP panel, physically disconnect the panel,
> boot the machine, "modprobe i915" and watch the segfault.
>
> Reviewed-by: Paulo Zanoni <[email protected]>
> Tested-by: Paulo Zanoni <[email protected]>
>
> I didn't really bisect, but I believe this is probably a regression
> from "drm/sysfs: sort out minor and connector device object
> lifetimes".

Yes, I think that's the one that broke it.

Thierry


Attachments:
(No filename) (980.00 B)
(No filename) (836.00 B)
Download all attachments