This patch isn't the good one, I send the fix here: http://lists.freedesktop.org/archives/dri-devel/2015-July/086568.html
Regards, Benjamin
2015-07-16 3:13 GMT+02:00 Xinwei Kong kong.kongxinwei@hisilicon.com:
Thank you for Russel.
It is Right, when this sti_tvout (component) finish executing component ".bind" function while sti_hdmi or sti_hda is not registered. the bug will occur .
this patch will prepare this bug by calling master .bind of sti_tvout after sti_hdmi or sti_hda is register to finish binding sti_hdmi or sti_hda component, however, it also slove to bring the "drm_dev" struct into the sti_hdmi or sti_hda.
Thank you Xinwei
On 2015/7/15 21:17, Benjamin Gaignard wrote:
Thanks a lot Russell, I have now understand where was my mistake.
2015-07-15 12:42 GMT+02:00 Russell King - ARM Linux linux@arm.linux.org.uk:
On Wed, Jul 15, 2015 at 12:19:01PM +0200, Benjamin Gaignard wrote:
The build order in Makefile hasn't been change so the bug doesn't occur...
In addition of taking care of not changing build order in Makefile, I would like to understand what is the expected behavior of component framework when master call component_bind_all() before any component_add() calls.
Should component bind function be called in component_add() ? Is up to component to detect that master is already bounded ?
Russell can you tell us what to do in this case ?
I don't follow, and you certainly should never get into the situation you're alluding to (where the master is already bound but a component is not.)
The way this should work is:
- master and components register themselves into the component helper in a random order.
- when the master registers, it gives the component helper a set of matches which it uses to determine which components are required.
- when the component helper determines that all components and the master have been registered, it calls the master bind function.
- the master bind function is responsible for the classical subsystem initialisation, calling component_bind_all() to cause the individual components bind() method to be called.
So, you should never _ever_ be in the situation where initcall ordering matters, or where the master is already bound but a component hasn't registered.
-- FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up according to speedtest.net.