Den 10.02.2020 16.06, skrev Daniel Vetter:
> On Mon, Feb 10, 2020 at 12:37:52PM +0100, Gerd Hoffmann wrote:
>> Move final cleanups to qxl_drm_release() callback.
>> Add drm_atomic_helper_shutdown() call to qxl_pci_remove().
>>
>> Signed-off-by: Gerd Hoffmann <kraxel@xxxxxxxxxx>
>> ---
>> drivers/gpu/drm/qxl/qxl_drv.c | 26 +++++++++++++++++++-------
>> 1 file changed, 19 insertions(+), 7 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/qxl/qxl_drv.c b/drivers/gpu/drm/qxl/qxl_drv.c
>> index 1d601f57a6ba..4fda3f9b29f4 100644
>> --- a/drivers/gpu/drm/qxl/qxl_drv.c
>> +++ b/drivers/gpu/drm/qxl/qxl_drv.c
>> @@ -34,6 +34,7 @@
>> #include <linux/pci.h>
>>
>> #include <drm/drm.h>
>> +#include <drm/drm_atomic_helper.h>
>> #include <drm/drm_drv.h>
>> #include <drm/drm_file.h>
>> #include <drm/drm_modeset_helper.h>
>> @@ -132,21 +133,30 @@ qxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
>> return ret;
>> }
>>
>> +static void qxl_drm_release(struct drm_device *dev)
>> +{
>> + struct qxl_device *qdev = dev->dev_private;
>> +
>> + /*
>> + * TODO: qxl_device_fini() call should be in qxl_pci_remove(),
>> + * reodering qxl_modeset_fini() + qxl_device_fini() calls is
>> + * non-trivial though.
>> + */
>> + qxl_modeset_fini(qdev);
>
> So the drm_mode_config_cleanup call in here belongs in ->release, but the
> qxl_destroy_monitors_object feels like should be perfectly fine in the
> remove hook. You might need to sprinkle a few drm_dev_enter/exit around to
> protect code paths thought.
>
> Aside from this lgtm, for the series
>
> Acked-by: Daniel Vetter <daniel.vetter@xxxxxxxx>
>
> And up to you whether you want to fix this or not really.
>
> Btw for testing all these patches that add a ->release hook I think it'd
> be good if the drm core checks that drm_device->dev is set to NULL, and
> that we do this in the remove hook. Since that's guaranteed to be gone at
> that point, so anything in ->release that still needs the device is
> broken. Ofc maybe do that check only for drivers which have a ->release
> hook, and we might need a few fixups.
>
We take a ref on the parent device in drm_dev_init() and release it in
drm_dev_fini(). I added this because of the DRM_DEV_* macros we have, to
protect access to the device struct after it was unregistered. Setting
drm_device->dev to NULL in drm_dev_unregister() instead will provide the
same protection I think.
commit 56be6503aab2
drm/drv: Hold ref on parent device during drm_device lifetime
Noralf.
> Cheers, Daniel
>
>> + qxl_device_fini(qdev);
>> + dev->dev_private = NULL;
>> + kfree(qdev);
>> +}
>> +
>> static void
>> qxl_pci_remove(struct pci_dev *pdev)
>> {
>> struct drm_device *dev = pci_get_drvdata(pdev);
>> - struct qxl_device *qdev = dev->dev_private;
>>
>> drm_dev_unregister(dev);
>> -
>> - qxl_modeset_fini(qdev);
>> - qxl_device_fini(qdev);
>> + drm_atomic_helper_shutdown(dev);
>> if (is_vga(pdev))
>> vga_put(pdev, VGA_RSRC_LEGACY_IO);
>> -
>> - dev->dev_private = NULL;
>> - kfree(qdev);
>> drm_dev_put(dev);
>> }
>>
>> @@ -279,6 +289,8 @@ static struct drm_driver qxl_driver = {
>> .major = 0,
>> .minor = 1,
>> .patchlevel = 0,
>> +
>> + .release = qxl_drm_release,
>> };
>>
>> static int __init qxl_init(void)
>> --
>> 2.18.1
>>
>
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
