summaryrefslogtreecommitdiff
path: root/include/uapi/linux/loop.h
diff options
context:
space:
mode:
authorIngrid Gallardo <ingridg@codeaurora.org>2014-12-18 14:21:28 -0800
committerDavid Keitel <dkeitel@codeaurora.org>2016-03-23 20:39:06 -0700
commit8100cb00f3032389e1fdfe7950d60a7ad998eb15 (patch)
tree3202388963583b2f643999e7ce1185a0a6172425 /include/uapi/linux/loop.h
parent19f0ce55204822aea4898f416b1966f7cca138d4 (diff)
msm: mdss: debugfs: add support to serialize wait4pingpong
For command mode panels, current code configures the HW to transfer a frame and returns the commit context without waiting for transfer to finish. As an example, after kicking off frame N transfer, commit context doesn't wait for Frame N's transfer to finish. By doing so, it allows SW to configure/queue frame (N+1) and improves the performance. But before frame (N+1) is kicked off, SW needs to make sure that frame N is finished otherwise it may lead to tearing and/or unknown behavior. This is achieved by calling wait4pingpong routine. But this optimization has a side effect when debugging critical issues. If for some reason, frame N had a bad HW configuration and it had HW hang then wait4pingpong during frame (N+1) will timeout. This timeout will lead to HW register dumps but this register dump is related frame (N+1) and not frame N. This gives incorrect debug information. To overcome this side effect, allow wait4pingpong's timing to change using debugfs. Setting this node to true will make the driver to run this processing in a synchronous way, setting it to false will return the driver to normal asynchronous operation: echo 1 > /<debugfs>/mdp/serialize_wait4pp -> sync echo 0 > /<debugfs>/mdp/serialize_wait4pp -> async (default) Change-Id: Ief25232b65cf3dfec7c2147120dac57d0f54d22f Signed-off-by: Ingrid Gallardo <ingridg@codeaurora.org> [cip@codeaurora.org: Use debugfs_create_u32 for wait4pp] Signed-off-by: Clarence Ip <cip@codeaurora.org>
Diffstat (limited to 'include/uapi/linux/loop.h')
0 files changed, 0 insertions, 0 deletions