On Mon, Jun 22, 2026 at 5:25 PM Laurent Pinchart
<[email protected]> wrote:
>
> On Tue, Jun 16, 2026 at 11:49:57PM +0530, Biren Pandya wrote:
> > Several GEM core functions manually managed mutex_lock() and
> > mutex_unlock() over single scopes or error paths. This adds boilerplate
> > and carries the risk of lock leaks if error paths are refactored.
> >
> > Modernize these locks by deploying the <linux/cleanup.h> scoped_guard()
> > macro. This ensures that the locks are reliably dropped when the block
> > exits, cleanly removing goto out_unlock paths and tightening the
> > lifecycle.
>
> What's the reason for doing so in in drm_gem and not other areas in DRM
> ?
Hi Laurent,
Thanks for taking a look.
No deeper reason than it being where I happened to start — I didn't
mean to single it out, and I'd rather the treatment be consistent than
piecemeal.
> > @@ -1021,37 +1018,34 @@ int drm_gem_change_handle_ioctl(struct drm_device
> > *dev, void *data,
> > goto out;
> > }
> >
> > - mutex_lock(&file_priv->prime.lock);
> > + scoped_guard(mutex, &file_priv->prime.lock) {
> > + spin_lock(&file_priv->table_lock);
> > + ret = idr_alloc(&file_priv->object_idr, obj, handle, handle +
> > 1,
> > + GFP_NOWAIT);
> > + spin_unlock(&file_priv->table_lock);
>
> And why don't you use guards for the spinlock as well ?
Fair point — the spinlocks here are equally good candidates; I only
kept v1 to mutexes to keep it small.
That said, this is a pure cleanup with no functional change, so it's
entirely your call whether it's worth carrying.
If you'd like it, I'll send a v2 that converts both the mutexes and
the spinlocks in drm_gem.c consistently. If you'd prefer not to take
cleanup-only churn, I'm happy to drop it — no problem either way.
Thanks,
Biren
