Hi all,
This is a bit nightmare fuel, but I think it's the best we can aim for:
- locking doesn't leak to callers/drivers, no headaches with locking inversions
- it allows us (once fbdev is fixed) to drop the modeset_lock_all from mst
unplug, cleaning up our locking hirarchy quite a bit
As a teaser of what this enables I've thrown three patches on top to drop a bit
of now superflous locking.
Todo:
- survive nightmares when reviewing this
- make sure all bits in i915 are covered (mostly depends upon Maarten's atomic
iterator cleanup, which is progressing well)
- apply same love to amdgpu/radeon to handle hot-unplugging there too
Some of the other approaches I've discussed with Dave on irc, but didn't
implemented here:
- Go with a reallocating (rcu-protected) array instead of connector_list. Has
the upside of catching every offender, but the downside of forcing churn onto
every driver which doesn't care.
- Rework locking to again protect the connector_list with mode_config.mutex. Imo
too hairy since that leaks the locking context out of every
drm_connector_unreference (since it's no longer a pure leaf lock protecting
things). Which tends to be real painful.
- Hope it works - we have the oopses already to prove otherwise :(
I think this implementation here strikes the best balance between invasiveness
and keeping locking concerns separated as much as possible. The one downside is
the implementation of the hot-unplug safe connector_list iterator.
Comments, screams of agony, testing and review highly welcome.
Cheers, Daniel
Daniel Vetter (10):
drm/amd-kfd: Clean up inline handling
drm: Don't compute obj counts expensively in get_resources
drm: Add explicit iter struct to drm_for_each_connector
drm/i915: Use use the drm_for_each_connector in i915_debugfs.c
drm/i915: Roll out drm_for_each_connector in intel_hotplug.c
drm: Drop cargo-culted modeset_lock_all from encoder/plane
init/cleanup
drm: Revamp connector_list protection
drm: Drop mode_config.mutex from connector_register_all
drm: Drop mode_config.mutex from get_resources ioctl
drm: Drop mode_config.mutex from _reset()
.../gpu/drm/amd/amdkfd/kfd_device_queue_manager.h | 4 +-
drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 -
drivers/gpu/drm/drm_atomic.c | 3 +-
drivers/gpu/drm/drm_atomic_helper.c | 12 ++-
drivers/gpu/drm/drm_crtc.c | 109 ++++++++++++---------
drivers/gpu/drm/drm_crtc_helper.c | 21 ++--
drivers/gpu/drm/drm_edid.c | 3 +-
drivers/gpu/drm/drm_fb_helper.c | 3 +-
drivers/gpu/drm/drm_plane_helper.c | 3 +-
drivers/gpu/drm/drm_probe_helper.c | 9 +-
drivers/gpu/drm/exynos/exynos_drm_drv.c | 6 +-
drivers/gpu/drm/i915/i915_debugfs.c | 22 ++---
drivers/gpu/drm/i915/intel_display.c | 6 +-
drivers/gpu/drm/i915/intel_dp_mst.c | 2 +-
drivers/gpu/drm/i915/intel_hotplug.c | 16 +--
drivers/gpu/drm/vc4/vc4_crtc.c | 3 +-
include/drm/drm_crtc.h | 79 +++++++++++----
17 files changed, 189 insertions(+), 115 deletions(-)
--
2.8.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
