| Message ID | 20171130151948.GF12606@linutronix.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, 2017-11-30 at 16:19 +0100, Sebastian Andrzej Siewior wrote: > The code has an ifdef and uses two functions to either init the bare > spinlock or init it and set a lock-class. It is possible to do the same > thing without an ifdef. > With this patch (in debug case) we first use the "default" lock class > which is later overwritten to the supplied one. Without lockdep the set > name/class function vanishes. > > Reported-by: kbuild test robot <fengguang.wu@intel.com> How exactly did kbuild test robot figure this out? At least according to the source there doesn't seem to be clarity about what is the right thing to do, this being just one option. Regards, Joonas > +++ b/drivers/gpu/drm/i915/i915_gem_timeline.c > @@ -33,11 +33,8 @@ static void __intel_timeline_init(struct > { > tl->fence_context = context; > tl->common = parent; > -#ifdef CONFIG_DEBUG_SPINLOCK > - __raw_spin_lock_init(&tl->lock.rlock, lockname, lockclass); > -#else > spin_lock_init(&tl->lock); > -#endif > + lockdep_set_class_and_name(&tl->lock, lockclass, lockname); > init_request_active(&tl->last_request, NULL); > INIT_LIST_HEAD(&tl->requests); > i915_syncmap_init(&tl->sync);
On 2017-12-13 16:00:49 [+0200], Joonas Lahtinen wrote: > On Thu, 2017-11-30 at 16:19 +0100, Sebastian Andrzej Siewior wrote: > > The code has an ifdef and uses two functions to either init the bare > > spinlock or init it and set a lock-class. It is possible to do the same > > thing without an ifdef. > > With this patch (in debug case) we first use the "default" lock class > > which is later overwritten to the supplied one. Without lockdep the set > > name/class function vanishes. > > > > Reported-by: kbuild test robot <fengguang.wu@intel.com> > > How exactly did kbuild test robot figure this out? I fixed it up for RT, then robot found a way to complain about it. Then I slapped myself for not submitting this patch in the first place. > At least according to the source there doesn't seem to be clarity about > what is the right thing to do, this being just one option. I don' think `ifdef CONFIG_DEBUG_SPINLOCK' is an option. Especially since the i915 driver here is the only user in tree doing this kind of thing. Then we have lockdep_set_class_and_name() (which I promote here). This looks like the official way of doing lockdep related things and it has even more than ten users in tree. > Regards, Joonas Sebastian
On Wed, 2017-12-13 at 16:06 +0100, Sebastian Andrzej Siewior wrote: > On 2017-12-13 16:00:49 [+0200], Joonas Lahtinen wrote: > > On Thu, 2017-11-30 at 16:19 +0100, Sebastian Andrzej Siewior wrote: > > > The code has an ifdef and uses two functions to either init the bare > > > spinlock or init it and set a lock-class. It is possible to do the same > > > thing without an ifdef. > > > With this patch (in debug case) we first use the "default" lock class > > > which is later overwritten to the supplied one. Without lockdep the set > > > name/class function vanishes. > > > > > > Reported-by: kbuild test robot <fengguang.wu@intel.com> > > > > How exactly did kbuild test robot figure this out? > > I fixed it up for RT, then robot found a way to complain about it. Then > I slapped myself for not submitting this patch in the first place. Right, I was kinda wondering if the robot has gained consciuousness and is developing new checks... > > At least according to the source there doesn't seem to be clarity about > > what is the right thing to do, this being just one option. > > I don' think `ifdef CONFIG_DEBUG_SPINLOCK' is an option. Especially > since the i915 driver here is the only user in tree doing this kind of > thing. > Then we have lockdep_set_class_and_name() (which I promote here). This > looks like the official way of doing lockdep related things and it has > even more than ten users in tree. I think it be worthwhile to suggest would be the addition of __spin_lock_init where you can pass in the the lockclass and name. Regards, Joonas
+peterz context: http://www.spinics.net/lists/intel-gfx/msg149011.html On 2017-12-13 17:37:21 [+0200], Joonas Lahtinen wrote: > On Wed, 2017-12-13 at 16:06 +0100, Sebastian Andrzej Siewior wrote: > > On 2017-12-13 16:00:49 [+0200], Joonas Lahtinen wrote: > > > On Thu, 2017-11-30 at 16:19 +0100, Sebastian Andrzej Siewior wrote: > > > > The code has an ifdef and uses two functions to either init the bare > > > > spinlock or init it and set a lock-class. It is possible to do the same > > > > thing without an ifdef. > > > > With this patch (in debug case) we first use the "default" lock class > > > > which is later overwritten to the supplied one. Without lockdep the set > > > > name/class function vanishes. … > > > At least according to the source there doesn't seem to be clarity about > > > what is the right thing to do, this being just one option. > > > > I don' think `ifdef CONFIG_DEBUG_SPINLOCK' is an option. Especially > > since the i915 driver here is the only user in tree doing this kind of > > thing. > > Then we have lockdep_set_class_and_name() (which I promote here). This > > looks like the official way of doing lockdep related things and it has > > even more than ten users in tree. > > I think it be worthwhile to suggest would be the addition of > __spin_lock_init where you can pass in the the lockclass and name. I don't due to the existing interface. However I don't call the shots here and added peterz instead :) > Regards, Joonas Sebastian
On Wed, Dec 13, 2017 at 06:36:33PM +0100, Sebastian Andrzej Siewior wrote: > +peterz > context: http://www.spinics.net/lists/intel-gfx/msg149011.html > > On 2017-12-13 17:37:21 [+0200], Joonas Lahtinen wrote: > > On Wed, 2017-12-13 at 16:06 +0100, Sebastian Andrzej Siewior wrote: > > > On 2017-12-13 16:00:49 [+0200], Joonas Lahtinen wrote: > > > > On Thu, 2017-11-30 at 16:19 +0100, Sebastian Andrzej Siewior wrote: > > > > > The code has an ifdef and uses two functions to either init the bare > > > > > spinlock or init it and set a lock-class. It is possible to do the same > > > > > thing without an ifdef. > > > > > With this patch (in debug case) we first use the "default" lock class > > > > > which is later overwritten to the supplied one. Without lockdep the set > > > > > name/class function vanishes. > … > > > > At least according to the source there doesn't seem to be clarity about > > > > what is the right thing to do, this being just one option. The proposed patch is definitely the right thing to do. The fact that it doesn't require #ifdef is a very big clue.
--- a/drivers/gpu/drm/i915/i915_gem_timeline.c +++ b/drivers/gpu/drm/i915/i915_gem_timeline.c @@ -33,11 +33,8 @@ static void __intel_timeline_init(struct { tl->fence_context = context; tl->common = parent; -#ifdef CONFIG_DEBUG_SPINLOCK - __raw_spin_lock_init(&tl->lock.rlock, lockname, lockclass); -#else spin_lock_init(&tl->lock); -#endif + lockdep_set_class_and_name(&tl->lock, lockclass, lockname); init_request_active(&tl->last_request, NULL); INIT_LIST_HEAD(&tl->requests); i915_syncmap_init(&tl->sync);
The code has an ifdef and uses two functions to either init the bare spinlock or init it and set a lock-class. It is possible to do the same thing without an ifdef. With this patch (in debug case) we first use the "default" lock class which is later overwritten to the supplied one. Without lockdep the set name/class function vanishes. Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> --- drivers/gpu/drm/i915/i915_gem_timeline.c | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-)