[RFCv1 11/12] drm: Atomic modeset ioctl
Ville Syrjälä
ville.syrjala at linux.intel.com
Mon Oct 7 17:05:22 CEST 2013
On Mon, Oct 07, 2013 at 10:39:01AM -0400, Rob Clark wrote:
> On Mon, Oct 7, 2013 at 10:18 AM, Ville Syrjälä
> <ville.syrjala at linux.intel.com> wrote:
> > On Mon, Oct 07, 2013 at 09:55:47AM -0400, Rob Clark wrote:
> >> On Mon, Oct 7, 2013 at 9:28 AM, Ville Syrjälä
> >> <ville.syrjala at linux.intel.com> wrote:
> >> > On Sat, Oct 05, 2013 at 08:45:49PM -0400, Rob Clark wrote:
> >> >> From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> >> >>
> >> >> The atomic modeset ioctl cna be used to push any number of new values
> >> >> for object properties. The driver can then check the full device
> >> >> configuration as single unit, and try to apply the changes atomically.
> >> >>
> >> >> The ioctl simply takes a list of object IDs and property IDs and their
> >> >> values. For setting values to blob properties, the property value
> >> >> indicates the length of the data, and the actual data is passed via
> >> >> another blob pointer.
> >> >>
> >> >> The caller can demand non-blocking operation from the ioctl, and if the
> >> >> driver can't satisfy that requirement an error will be returned.
> >> >>
> >> >> The caller can also request to receive asynchronous completion events
> >> >> after the operation has reached the hardware. An event is sent for each
> >> >> object specified by the caller, whether or not the actual state of
> >> >> that object changed. Each event also carries a framebuffer ID, which
> >> >> indicates to user space that the specified object is no longer
> >> >> accessing that framebuffer.
> >> >>
> >> >> TODO: detailed error reporting?
> >> >>
> >> >> v1: original
> >> >> v2: rebase on uapi changes, and drm state structs.. -- Rob Clark
> >> >> v3: rebase, missing padding in drm_event_atomic.. -- Rob Clark
> >> >>
> >> >> Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> >> >> ---
> >> >> drivers/gpu/drm/drm_crtc.c | 159 ++++++++++++++++++++++++++++++++++++++++++++
> >> >> drivers/gpu/drm/drm_drv.c | 1 +
> >> >> include/drm/drmP.h | 6 ++
> >> >> include/drm/drm_crtc.h | 2 +
> >> >> include/uapi/drm/drm.h | 12 ++++
> >> >> include/uapi/drm/drm_mode.h | 16 +++++
> >> >> 6 files changed, 196 insertions(+)
> >> >>
> >> >> diff --git a/drivers/gpu/drm/drm_crtc.c b/drivers/gpu/drm/drm_crtc.c
> >> >> index 09396a9..910e5c6 100644
> >> >> --- a/drivers/gpu/drm/drm_crtc.c
> >> >> +++ b/drivers/gpu/drm/drm_crtc.c
> >> >> @@ -4338,3 +4338,162 @@ void drm_mode_config_cleanup(struct drm_device *dev)
> >> >> idr_destroy(&dev->mode_config.crtc_idr);
> >> >> }
> >> >> EXPORT_SYMBOL(drm_mode_config_cleanup);
> >> >> +
> >> >> +int drm_mode_atomic_ioctl(struct drm_device *dev,
> >> >> + void *data, struct drm_file *file_priv)
> >> >> +{
> >> >> + struct drm_mode_atomic *arg = data;
> >> >> + uint32_t __user *objs_ptr = (uint32_t __user *)(unsigned long)(arg->objs_ptr);
> >> >> + uint32_t __user *count_props_ptr = (uint32_t __user *)(unsigned long)(arg->count_props_ptr);
> >> >> + uint32_t __user *props_ptr = (uint32_t __user *)(unsigned long)(arg->props_ptr);
> >> >> + uint64_t __user *prop_values_ptr = (uint64_t __user *)(unsigned long)(arg->prop_values_ptr);
> >> >> + uint64_t __user *blob_values_ptr = (uint64_t __user *)(unsigned long)(arg->blob_values_ptr);
> >> >> + unsigned int copied_objs = 0;
> >> >> + unsigned int copied_props = 0;
> >> >> + unsigned int copied_blobs = 0;
> >> >> + void *state;
> >> >> + int ret = 0;
> >> >> + unsigned int i, j;
> >> >> +
> >> >> + if (arg->flags & ~(DRM_MODE_ATOMIC_TEST_ONLY |
> >> >> + DRM_MODE_ATOMIC_EVENT | DRM_MODE_ATOMIC_NONBLOCK))
> >> >> + return -EINVAL;
> >> >> +
> >> >> + /* can't test and expect an event at the same time. */
> >> >> + if ((arg->flags & DRM_MODE_ATOMIC_TEST_ONLY) &&
> >> >> + (arg->flags & DRM_MODE_ATOMIC_EVENT))
> >> >> + return -EINVAL;
> >> >> +
> >> >> + mutex_lock(&dev->mode_config.mutex);
> >> >
> >> > I'm pretty sure I had a version w/ fixed locking...
> >> >
> >> > git://gitorious.org/vsyrjala/linux.git rebased_drm_atomic_24
> >>
> >> oh, hmm.. actually the locking should no longer be in the ioctl fxn,
> >> but should be pushed down to the commit. looks like I missed this. I
> >> need to dig up some actual test code for atomic ioctl ;-)
> >
> > It can't be in commit. At the very least it has to be around
> > check+commit, and also you need to hold some locks when you're copying
> > the current config over to your temp storage. That's assuming you store
> > _everything_ in the temp storage and so it doesn't matter if the current
> > state can change between the copy and check+commit.
> >
>
> hmm.. I start to wish we did atomic first, before the fine grained
> locking re-work ;-)
I don't see it mattering for this particular problem if we just grab
the big lock anyway.
But as I already stated, I would prefer to solve the plane locking
before doing atomic. Whether we have per-plane locks or not can
actually matter for this code, especially when dealing with planes
that can move between crtcs.
And we still have locking issues to solve in the .check+commit steps
when we have to deal with shared resources across the entire device.
We can't simply use some small local lock in the .check step since
something relevant might have changed by the time we reach .commit.
I guess we could re-check such things at the very start of commit,
before we've clobbered any state, and then update the global state
to match if everything looks good, before we again drop the local
lock.
>
> >>
> >> >> +
> >> >> + state = dev->driver->atomic_begin(dev, arg->flags);
> >> >> + if (IS_ERR(state)) {
> >> >> + ret = PTR_ERR(state);
> >> >> + goto unlock;
> >> >> + }
> >> >> +
> >> >> + for (i = 0; i < arg->count_objs; i++) {
> >> >> + uint32_t obj_id, count_props;
> >> >> + struct drm_mode_object *obj;
> >> >> +
> >> >> + if (get_user(obj_id, objs_ptr + copied_objs)) {
> >> >> + ret = -EFAULT;
> >> >> + goto out;
> >> >> + }
> >> >> +
> >> >> + obj = drm_mode_object_find(dev, obj_id, DRM_MODE_OBJECT_ANY);
> >> >> + if (!obj || !obj->properties) {
> >> >> + ret = -ENOENT;
> >> >> + goto out;
> >> >> + }
> >> >> +
> >> >> + if (arg->flags & DRM_MODE_ATOMIC_EVENT) {
> >> >> + // XXX create atomic event instead..
> >> >
> >> > Some people are apparently more comfortable with a per-crtc event
> >> > rather than the per-obj events I added. So I think we might want
> >> > both in the end.
> >>
> >> yeah, sorting out events is a bit 'TODO' at the moment. I do kind of
> >> like per-crtc event.. it seems easier to implement and I'm not really
> >> sure what finer event granularity buys us.
> >
> > Mainly it gets you the old fb. If you limit youself to < vrefresh it's
> > not a big deal, but going faster than that and you start to want that
> > information.
> >
>
> oh, right.. well I guess it wouldn't be too hard to add the current
> primary plane fb_id to the existing event, which might be sufficient.
>
> At any rate, I don't mind adding new event flags to atomic ioctl once
> we get atomic in place.
Yeah adding new events isn't that hard. We just need to make sure
things make sense. The code you posted doesn't :P
You add the event to every object, even though you only wanted one
event per crtc. The same object could also be listed multiple times in
the object list, so the code would even add multiple events for the
same object.
>
> BR,
> -R
>
> >>
> >> BR,
> >> -R
> >>
> > >> + struct drm_pending_vblank_event *e =
> >> >> + create_vblank_event(dev, file_priv, arg->user_data);
> >> >> + if (!e) {
> >> >> + ret = -ENOMEM;
> >> >> + goto out;
> >> >> + }
> >> >> + ret = dev->driver->atomic_set_event(dev, state, obj, e);
> >> >> + if (ret) {
> >> >> + destroy_vblank_event(dev, file_priv, e);
> >> >> + goto out;
> >> >> + }
> >> >> + }
> >> >> +
> >> >> + if (get_user(count_props, count_props_ptr + copied_objs)) {
> >> >> + ret = -EFAULT;
> >> >> + goto out;
> >> >> + }
> >> >> +
> >> >> + copied_objs++;
> >> >> +
> >> >> + for (j = 0; j < count_props; j++) {
> >> >> + uint32_t prop_id;
> >> >> + uint64_t prop_value;
> >> >> + struct drm_mode_object *prop_obj;
> >> >> + struct drm_property *prop;
> >> >> + void *blob_data = NULL;
> >> >> +
> >> >> + if (get_user(prop_id, props_ptr + copied_props)) {
> >> >> + ret = -EFAULT;
> >> >> + goto out;
> >> >> + }
> >> >> +
> >> >> + if (!object_has_prop(obj, prop_id)) {
> >> >> + ret = -EINVAL;
> >> >> + goto out;
> >> >> + }
> >> >> +
> >> >> + prop_obj = drm_mode_object_find(dev, prop_id,
> >> >> + DRM_MODE_OBJECT_PROPERTY);
> >> >> + if (!prop_obj) {
> >> >> + ret = -ENOENT;
> >> >> + goto out;
> >> >> + }
> >> >> + prop = obj_to_property(prop_obj);
> >> >> +
> >> >> + if (get_user(prop_value, prop_values_ptr + copied_props)) {
> >> >> + ret = -EFAULT;
> >> >> + goto out;
> >> >> + }
> >> >> +
> >> >> + if (!drm_property_change_is_valid(prop, prop_value)) {
> >> >> + ret = -EINVAL;
> >> >> + goto out;
> >> >> + }
> >> >> +
> >> >> + if ((prop->flags & DRM_MODE_PROP_BLOB) && prop_value) {
> >> >> + uint64_t blob_ptr;
> >> >> +
> >> >> + if (get_user(blob_ptr, blob_values_ptr + copied_blobs)) {
> >> >> + ret = -EFAULT;
> >> >> + goto out;
> >> >> + }
> >> >> +
> >> >> + blob_data = kmalloc(prop_value, GFP_KERNEL);
> >> >> + if (!blob_data) {
> >> >> + ret = -ENOMEM;
> >> >> + goto out;
> >> >> + }
> >> >> +
> >> >> + if (copy_from_user(blob_data, (void __user *)(unsigned long)blob_ptr, prop_value)) {
> >> >> + kfree(blob_data);
> >> >> + ret = -EFAULT;
> >> >> + goto out;
> >> >> + }
> >> >> + }
> >> >> +
> >> >> + /* User space sends the blob pointer even if we
> >> >> + * don't use it (length==0).
> >> >> + */
> >> >> + if (prop->flags & DRM_MODE_PROP_BLOB)
> >> >> + copied_blobs++;
> >> >> +
> >> >> + /* The driver will be in charge of blob_data from now on. */
> >> >> + ret = drm_mode_set_obj_prop(obj, state, prop,
> >> >> + prop_value, blob_data);
> >> >> + if (ret)
> >> >> + goto out;
> >> >> +
> >> >> + copied_props++;
> >> >> + }
> >> >> + }
> >> >> +
> >> >> + ret = dev->driver->atomic_check(dev, state);
> >> >> + if (ret)
> >> >> + goto out;
> >> >> +
> >> >> + if (arg->flags & DRM_MODE_ATOMIC_TEST_ONLY)
> >> >> + goto out;
> >> >> +
> >> >> + ret = dev->driver->atomic_commit(dev, state);
> >> >> +
> >> >> + out:
> >> >> + dev->driver->atomic_end(dev, state);
> >> >> + unlock:
> >> >> + mutex_unlock(&dev->mode_config.mutex);
> >> >> +
> >> >> + return ret;
> >> >> +}
> >> >> diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c
> >> >> index e572dd2..43e04ae 100644
> >> >> --- a/drivers/gpu/drm/drm_drv.c
> >> >> +++ b/drivers/gpu/drm/drm_drv.c
> >> >> @@ -166,6 +166,7 @@ static const struct drm_ioctl_desc drm_ioctls[] = {
> >> >> DRM_IOCTL_DEF(DRM_IOCTL_MODE_OBJ_GETPROPERTIES, drm_mode_obj_get_properties_ioctl, DRM_CONTROL_ALLOW|DRM_UNLOCKED),
> >> >> DRM_IOCTL_DEF(DRM_IOCTL_MODE_OBJ_SETPROPERTY, drm_mode_obj_set_property_ioctl, DRM_MASTER|DRM_CONTROL_ALLOW|DRM_UNLOCKED),
> >> >> DRM_IOCTL_DEF(DRM_IOCTL_MODE_CURSOR2, drm_mode_cursor2_ioctl, DRM_MASTER|DRM_CONTROL_ALLOW|DRM_UNLOCKED),
> >> >> + DRM_IOCTL_DEF(DRM_IOCTL_MODE_ATOMIC, drm_mode_atomic_ioctl, DRM_MASTER|DRM_CONTROL_ALLOW|DRM_UNLOCKED),
> >> >> };
> >> >>
> >> >> #define DRM_CORE_IOCTL_COUNT ARRAY_SIZE( drm_ioctls )
> >> >> diff --git a/include/drm/drmP.h b/include/drm/drmP.h
> >> >> index 4c3de22..f282733 100644
> >> >> --- a/include/drm/drmP.h
> >> >> +++ b/include/drm/drmP.h
> >> >> @@ -1158,6 +1158,12 @@ struct drm_pending_vblank_event {
> >> >> struct drm_event_vblank event;
> >> >> };
> >> >>
> >> >> +struct drm_pending_atomic_event {
> >> >> + struct drm_pending_event base;
> >> >> + int pipe;
> >> >> + struct drm_event_atomic event;
> >> >> +};
> >> >> +
> >> >> /**
> >> >> * DRM device structure. This structure represent a complete card that
> >> >> * may contain multiple heads.
> >> >> diff --git a/include/drm/drm_crtc.h b/include/drm/drm_crtc.h
> >> >> index 86a5a00..fa3956e 100644
> >> >> --- a/include/drm/drm_crtc.h
> >> >> +++ b/include/drm/drm_crtc.h
> >> >> @@ -1313,6 +1313,8 @@ extern int drm_mode_obj_get_properties_ioctl(struct drm_device *dev, void *data,
> >> >> struct drm_file *file_priv);
> >> >> extern int drm_mode_obj_set_property_ioctl(struct drm_device *dev, void *data,
> >> >> struct drm_file *file_priv);
> >> >> +extern int drm_mode_atomic_ioctl(struct drm_device *dev,
> >> >> + void *data, struct drm_file *file_priv);
> >> >>
> >> >> extern void drm_fb_get_bpp_depth(uint32_t format, unsigned int *depth,
> >> >> int *bpp);
> >> >> diff --git a/include/uapi/drm/drm.h b/include/uapi/drm/drm.h
> >> >> index ece8678..efb1461 100644
> >> >> --- a/include/uapi/drm/drm.h
> >> >> +++ b/include/uapi/drm/drm.h
> >> >> @@ -733,6 +733,7 @@ struct drm_prime_handle {
> >> >> #define DRM_IOCTL_MODE_OBJ_GETPROPERTIES DRM_IOWR(0xB9, struct drm_mode_obj_get_properties)
> >> >> #define DRM_IOCTL_MODE_OBJ_SETPROPERTY DRM_IOWR(0xBA, struct drm_mode_obj_set_property)
> >> >> #define DRM_IOCTL_MODE_CURSOR2 DRM_IOWR(0xBB, struct drm_mode_cursor2)
> >> >> +#define DRM_IOCTL_MODE_ATOMIC DRM_IOWR(0xBC, struct drm_mode_atomic)
> >> >>
> >> >> /**
> >> >> * Device specific ioctls should only be in their respective headers
> >> >> @@ -774,6 +775,17 @@ struct drm_event_vblank {
> >> >> __u32 reserved;
> >> >> };
> >> >>
> >> >> +struct drm_event_atomic {
> >> >> + struct drm_event base;
> >> >> + __u64 user_data;
> >> >> + __u32 tv_sec;
> >> >> + __u32 tv_usec;
> >> >> + __u32 sequence;
> >> >> + __u32 obj_id;
> >> >> + __u32 old_fb_id;
> >> >> + __u32 reserved;
> >> >> +};
> >> >> +
> >> >> #define DRM_CAP_DUMB_BUFFER 0x1
> >> >> #define DRM_CAP_VBLANK_HIGH_CRTC 0x2
> >> >> #define DRM_CAP_DUMB_PREFERRED_DEPTH 0x3
> >> >> diff --git a/include/uapi/drm/drm_mode.h b/include/uapi/drm/drm_mode.h
> >> >> index 6d4f089..03a473c 100644
> >> >> --- a/include/uapi/drm/drm_mode.h
> >> >> +++ b/include/uapi/drm/drm_mode.h
> >> >> @@ -489,4 +489,20 @@ struct drm_mode_destroy_dumb {
> >> >> uint32_t handle;
> >> >> };
> >> >>
> >> >> +#define DRM_MODE_ATOMIC_TEST_ONLY (1<<0)
> >> >> +#define DRM_MODE_ATOMIC_EVENT (1<<1)
> >> >> +#define DRM_MODE_ATOMIC_NONBLOCK (1<<2)
> >> >> +
> >> >> +/* FIXME come up with some sane error reporting mechanism? */
> >> >> +struct drm_mode_atomic {
> >> >> + __u32 flags;
> >> >> + __u32 count_objs;
> >> >> + __u64 objs_ptr;
> >> >> + __u64 count_props_ptr;
> >> >> + __u64 props_ptr;
> >> >> + __u64 prop_values_ptr;
> >> >> + __u64 blob_values_ptr;
> >> >> + __u64 user_data;
> >> >> +};
> >> >> +
> >> >> #endif
> >> >> --
> >> >> 1.8.3.1
> >> >
> >> > --
> >> > Ville Syrjälä
> >> > Intel OTC
> >
> > --
> > Ville Syrjälä
> > Intel OTC
--
Ville Syrjälä
Intel OTC
More information about the dri-devel
mailing list