On Fri, Feb 07, 2020 at 01:13:12PM +0100, Gerd Hoffmann wrote:

From: Gurchetan Singh gurchetansingh@chromium.org

Move bochs_unload call from bochs_remove() to the new bochs_release() callback. Also call drm_dev_unregister() first in bochs_remove().

Signed-off-by: Gerd Hoffmann kraxel@redhat.com

drivers/gpu/drm/bochs/bochs_drv.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/bochs/bochs_drv.c b/drivers/gpu/drm/bochs/bochs_drv.c index 10460878414e..87ee1eb21a4d 100644 --- a/drivers/gpu/drm/bochs/bochs_drv.c +++ b/drivers/gpu/drm/bochs/bochs_drv.c @@ -60,6 +60,11 @@ static int bochs_load(struct drm_device *dev)

DEFINE_DRM_GEM_FOPS(bochs_fops);

+static void bochs_release(struct drm_device *dev) +{

• bochs_unload(dev);

Hm I think this needs to be split, and bochs_hw_fini needs to stay in bochs_pci_remove. Otherwise we again release the iommu mappings too late (there was just a discussion about that on the fbdev side of things getting this wrong). Good to fix the lifetime fun here, but let's not move too far :-)

Ideally we'd move the iomapping to devm_, that would remove it at the right time I think.

Slight annoyance might be that we then need to shoot down the varios mmaps, which I frankly don't know how to do. But keeping them pointing at wrong stuff also isn't that great :-/

Cheers, Daniel

+}

static struct drm_driver bochs_driver = { .driver_features = DRIVER_GEM | DRIVER_MODESET | DRIVER_ATOMIC, .fops = &bochs_fops, @@ -69,6 +74,7 @@ static struct drm_driver bochs_driver = { .major = 1, .minor = 0, DRM_GEM_VRAM_DRIVER,

• .release = bochs_release,

};

/* ---------------------------------------------------------------------- */ @@ -148,9 +154,8 @@ static void bochs_pci_remove(struct pci_dev *pdev) { struct drm_device *dev = pci_get_drvdata(pdev);

• drm_atomic_helper_shutdown(dev); drm_dev_unregister(dev);
• bochs_unload(dev);

• drm_atomic_helper_shutdown(dev); drm_dev_put(dev);

}

-- 2.18.1