Add support for tracking the mode in which DMA is being used in overlay.
Remove unnecessary tracking variables, and their set,reset inconsistency.
With such varibles in hwc, it still leads to overlay giving out DMA pipes.
Make overlay as the single tracking place.
Change-Id: Ib949f3d0ef5918ef323c81cb1768b68a68c83da4
Log a frame-by-frame succession of HWComposer layers' data and write
their buffers, if any, into raw or png files based on system property
values.
Change-Id: Icceccf5fff5ab4d78dbc6c28c11210b8837ddfd9
With this change we are moving MDP composition from all-or-nothing
implementation to mixed mode implementation where layers can
partially be composited through MDP and while rest of the layers are
composed / cached in framebuffer.
- Mixed mode design is based on layer caching
- Mixed mode path is configured only when non-bypassing
layers are cached.
- Never allow mixed mode when FB needs to be udpated.
- If we cannot bypass all MDP comp marked layers, bail
out completely.
Change-Id: Ie08f39db07e032b537f042d0d2bfe772ebfed049
This is logged in the kernel and if there are excessive delays,
logged in the PowerManager.
Bug:8485194
Change-Id: Ia83b186ccae1fb6de59f200faf28346f84d832b6
- On suspend, SurfaceFlinger calls hwc_blank on primary and
external separately, inbetween these calls there might be a
update, which leads to acquiring overlay pipes for external
which will not be cleared and results in showing up stale
frame on external display
- This change clears all the overlay pipes used on every blank
Change-Id: I44d62b7d0fcd992cf8a26bb0c21df77952bc1381
CRs-fixed: 464070
This is indicative of how many frames are posted to the display
by SurfaceFlinger. Set debug.gr.calcfps to 1 and restart the
framework to enable.
Change-Id: Ic503c14380fd7d8a5b062aa5da0d61bfc3991cb2
CRs-Fixed: 462010
Even if we exceed max supported layers, configure pipe for FB
Because FB contents are always displayed using hwc
Also some minor code fixes
Change-Id: I740ab7e908c916373b1d87facd831e09098ce753
This change adds support to MDP composition to use DMA pipes when
layer scaling is not needed. Also makes sure not to invoke
MDP rotator when DMA pipes are configured since MDP rotator will
expect both the DMA pipes to available for performing rotation.
In such cases, we let the first frame of rotated video frame to
fall back to FB.
Change-Id: I3c3697dd2713ce9b86ba1e23192ec5841abd25ec
This patch removes the usage of the framebuffer HAL which is
deprecated in JB MR1 onwards. The code is left for compatibility
such as conformance tests but it is unused for normal display
usage.
Change-Id: If98133bdaa759cdc41d4503ff695b225ee43cb6f
Conflicts:
libhwcomposer/hwc_utils.cpp
-Add support for Video via overlay on 4k2k external panel.
-Add support for rotating videos on 4k2k panels. We use
pre-rotation in hwc to rotate a video into a single buffer,
irrespective of panel size. Then this buffer is fed to MDP.
Rotator objects are managed by the new RotMgr.
-Cleaup mdpcomp and overlay.
Change-Id: Ifb08534747e8e18b6c58dd8a3e1a9947409100f1
Sometimes vsync would not be disabled before the display is
blanked.
This was due to the vsync disable ioctl being kicked off in a
thread different from the event control which surfaceflinger
called. So, while the calls from surfaceflinger are in order, the
order of execution wasn't always the same. Hence, making sure
ioctls are called in the same context to ensure order.
Also
- Make blanking/unblanking logs show the operation as a string.
- Add a debug property to dump vsync timestamps (needs framework
reboot)
- Remove a log which showed delay in reading vsync. This log was
unreliable when vsync is disabled and we're still reading from
the sysfs node.
Change-Id: Ibec04e9ffebd0ac6e1d32b7031e3668abd9390ff
CRs-fixed: 443113
If the incoming layer count exceeds MAX_NUM_LAYERS,
let surfaceflinger handle the list.
CRs-fixed: 456906
Change-Id: I53159104e5e714ba313d0196ca959f55c82e18ae
Surfaceflinger does not send Blank/unblank event to hwc for virtual
display, handle it explicitly when blank for primary is invoked, so
that any pipes unset gets committed.
Change-Id: Iec7a1611a8963408f65c8a73a1c15e47cd7faab6
CRs-Fixed: 456810
The earlier copybit solution involved blitting directly into
the SurfaceFlinger's framebuffer target. That solution involved
unnecessary framework changes and caused issues when the
framebuffer was being written to both by GL and copybit.
Update hwc_copybit to use our own buffers for this purpose.
We also make sure we display only the region we're interested in
so that unnecessary artifacts from previous blits do not show up
on the display. This way, we can avoid clearing the intermediate
buffers every frame.
Change-Id: I713b3fc606e0768444c621af76853ece41964da1
Close acquireFenceFds always even in case of failures if a layer is either
OVERLAY or FRAMEBUFFER_TARGET to prevent leaks in case of failures.
The framework is *not* responsible for closing acquire fds, for layers
marked as above two.
Change-Id: Ia6c751d3ec25f196f5503120894fc6cc692b9d25
This change allows virtual display to be composed
by HWC for supporting Google WFD App + QCOM WFD stack.
Change-Id: If8892230256e72fa34e3fb5ae715c3ad8cbd5b64
Call MSMFB_DISPLAY_COMMIT for primary FB even if FB handle is null.
This is already done for external FB.
Change-Id: I5c405f933ce0fd99df6813acea45169802c54226
This change introduces the following changes:
- Frames having 2 yuv layer can be bypassed
- if not handled by MDP comp, video path will not
handle any such frames and will be FB rendered
Change-Id: Id211dbc185608c482badbca00a2708fa6138a8e5
Android has scenarios where it calls HWC::prepare and HWC::set
without any application layers. In such cases, for power
optimization, user space avoids calling PLAY
(on FB - since there is no new content drawn) and COMMIT ioctls.
But we might have called UNSET on mdp pipes used for the previous
frame. As per the current MDP design, these freed pipes will not
be unstaged unless atleast a COMMIT ioctl is called on that mixer.
Change-Id: I7f7574b2e589e80c4926533066e35784bb29b4a5
CRs-Fixed: 437736
- Add CopybitEngine to hwc_context
- draw the layer on the HWC_FRAMEBUFFER_TARGET using
copybit(c2d) if it qualifies for C2D composition.
- use fence returned from the copybit during c2d
composition
Change-Id: I052da015cd031f7abd6411d83d7944c335caeff7
- When the device is suspended, the overlay pipes will be unset,
but they wont be unstaged as there is no DISPLAY_COMMIT ioctl
called for ext display, as a result the fence wont be signalled
by the MDP.
- This is needed as we dont turn off the External display(HDMI)
Change-Id: I9794ad2ae7c9d7b4ce3f9787f50ea9da87966e41
CRs-fixed: 425483
This change introduces hdmi pending state, where we allow
frames to fall back to FB, closing all MDP pipes.
Pipes could be used by primary (mixer 0) in which case moving
them to external (mixer 1) involves unsetting them and setting
them in the same round which is not recommended by the driver.
It will fall back to GPU till HDMI is configured on a connect
event.
CRs-Fixed: 430057
Change-Id: I4b6677883c499f6d15656c3578190be1115ee6b4
Restructure MDPComp class for 4kx2k displays.
Cleanup MDPComp class to remove unused stuff and formatting.
Add a MDPCompLowRes class for current functionality.
Add a hollow MDPCompHighRes class for future.
Change-Id: I343540e712d31ca907257a73b49567578f49d5f7
Add support for 4kx2k FB for primary and external panels.
Change class design to create appropriate version of FBUpdate on boot up based
on the panel resolution.
Change-Id: I216d815d9b81c610aa39e351f7b55736dfa48b43
When MDP comp is done the layer cache
is not reset. Calling the layer cache updated
after MDP comp is configured ensures that the
cache is reset correctly.
Change-Id: I9fc3ddd713ceac4be708e33782c3661e758dfcb9
This commit has the following changes:
1. Use hdmi switch node for hdmi hotplug
- uses /devices/virtual/switch/hdmi switch_state to determine
the hdmi connect/disconnect
- this switch will be set to 1 on connect and 0 on disconnect
- on connect, open framebuffer device, will trigger the hdcp
authentication.
- the driver will set the hdmi_audio switch to indicate that
the HDMI sink supports audio
2. Enable HDP on unblank
- this will resolve the device freeze during the bootup with hdmi
connected
Change-Id: Id7a656d548a31178e31dcb9aea8acf38a56c9e0c
(cherry picked from commit 953b04b3cbba2f91d25795e5b0560c13d47844b0)
Does a basic implementation of the dump function
called by SurfaceFlinger in dumpsys.
Further logs in different areas can be added later on
using the dumpsys_log utility.
Change-Id: I2194b4ed7f10947272b7b824ef802ce2ee3c8a48
Also make sure layer cache is used correctly when
we're not using MDP composition
Bug: 7643563
Change-Id: I4dcd70c2da128b2dcbe0b72dc14031983649719d
Signed-off-by: Iliyan Malchev <malchev@google.com>
Signed-off-by: Naseer Ahmed <naseer@codeaurora.org>
This change extends MDP composition to support
upto 4 layers by making border fill as base pipe and
using all MDP h/w pipes in overlay mode.
Conflicts:
libhwcomposer/hwc.cpp
libhwcomposer/hwc_utils.cpp
libhwcomposer/hwc_utils.h
Bug: 7626586
Change-Id: I01a0e53ddfbcf3ed46734f6f3bb0ef7d912ceac6
Signed-off-by: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Iliyan Malchev <malchev@google.com>
Remove state based pipe handling and make pipes tracked dynamically.
Add the configStart, configDone, nextPipe availablePipes APIs.
Remove setState API.
-configStart marks all pipes available.
-configDone garbage-collects unused pipe objects.
-nextPipe returns the index of next available pipe and create a corresponding
pipe object if not present
-availablePipes returns total of "unallocated" pipes. (Pipes could be allocated
but unused)
Changes in hwc adapt to the new overlay APIs.
Compile out MDP comp.
Remove unused files in overlay and hwc.
Update licenses.
Conflicts:
libhwcomposer/hwc.cpp
libhwcomposer/hwc_utils.cpp
Bug: 7626586
Change-Id: Id8e56901d34e5dc0fd088260d05e3e46f68ff090
Signed-off-by: Saurabh Shah <saurshah@codeaurora.org>
Signed-off-by: Iliyan Malchev <malchev@google.com>
In the cases where all the layers that are marked as HWC_FRAMEBUFFER did not
change, we can mark them as HWC_OVERLAY since their contents are already on the
framebuffer and do not need any additional processing.
This optimization shows power improvement for video playback.
Bug: 7623741
Change-Id: Ia178a926e6f56a3ec9291250a22a66f212c30b14
Signed-off-by: Iliyan Malchev <malchev@google.com>
Check for errors while drawing the layers in hwc_set.
Return -1 in the error case so that framework doesn't wait on fences.
Bug: 7284105
Change-Id: I024f9ca07faf64d6b90d4f483164dc15c0d7992c
Signed-off-by: Iliyan Malchev <malchev@google.com>
Cleanup overlay for external from the draw thread. If done from the uevent
thread, its possible that the object being used by draw thread is deleted by
uevent thread.
This also removes unnecessary side-effects where libexternal sets states in
hwc, whereas, it could be set from hwc itself. There should be no need for
libexternal to modify states in hwc.
Bug: 7335863 (partial fix)
Change-Id: If07483e640abae2ced2418e0d5c8f278f8c6ec33
Signed-off-by: Iliyan Malchev <malchev@google.com>
Wait for hwc_set completion if already in progress before processing
the async events.
Bug: 7288769
Change-Id: Ibdd0ee82f079df566b899c94df19f6bb1031d686
Signed-off-by: Iliyan Malchev <malchev@google.com>
We are seeing errors where a device's hwcVsyncThread is not present for some
reason. The surface flinger has not crashed, so either the thread never got
created, or it exited withouth throwing an error. This patch adds some more
verbose on-error logging to the HWC as an attempt to verify the theory that the
thread does not get created, or fails in pthread_create. While we're at it, we
add the same logging at pthread_create() for hwcUeventThread, as well.
Also, replace the lseek()+read() with a pread() combo in the vsync thread.
Change-Id: I555d786a7d66ff4ef1dbfd95947a7d9341e56f11
related-to-bug: 7305728
Signed-off-by: Iliyan Malchev <malchev@google.com>
If the FB is not used, because all the app layers are overlays, then set the
flag to have the driver not fetch from the FB pipe.
Bug: 7183992
Change-Id: I96ecc1610ac052e8512158d189b0614ecf82d139
Signed-off-by: Iliyan Malchev <malchev@google.com>
-- numHwLayers==1 means that there is only the FB_TARGET which HWC shouldn't
draw. Also check for blank in set. Checking for this in prepare prevents us
from acquiring pipes.
-- add null check for private handle
-- print out vsync timestamp data on error
bug: 7274417
related-to-bug: 7288769
Change-Id: I8dae0d0748c5226fdc33ed8f5cc5259835dec2de
Signed-off-by: Iliyan Malchev <malchev@google.com>
It had a glFinish which is not needed anymore on HWC 1.1
Bug: 7274951
Change-Id: I1389802a0ed3f0199d7d7ba0110a343b2f571106
Acked-by: Naseer Ahmed <naseer@codeaurora.org>
Signed-off-by: Iliyan Malchev <malchev@google.com>
Fix some race conditions between hwc_eventControl and the hwcVsyncThread.
Bug: 7274951
Change-Id: Ic71d65918303b4cef6a379ff5397702b2b578373
Signed-off-by: Iliyan Malchev <malchev@google.com>
* Read sysfs entries to get vsync timestamp from kernel.
* External display continues to use uevents for hotplug events
* A new thread, vsyncThread is created to read and send vsync
timestamps to HAL.
* Disable H/W VSYNC for MDP 5.x targets until kernel changes are
complete.
* Synchronization is needed to make sure read() is not called
by hwcVsyncThread when VSYNC is disabled.
Change-Id: Iab0a94a3cfce9142b8867859f2a7d9bcaecb0996
Acked-by: Neti Ravi Kumar <ravineti@codeaurora.org>
