Closed Bug 982968 Opened 11 years ago Closed 10 years ago

[buri] run gralloc framebuffer thread at RT priority to minimize fb_lockBuffer()

Categories

(Firefox OS Graveyard :: GonkIntegration, defect, P3)

ARM
Gonk (Firefox OS)
defect

Tracking

(Not tracked)

RESOLVED INCOMPLETE

People

(Reporter: bkelly, Unassigned)

References

Details

(Keywords: perf, Whiteboard: [c=uniformity p= s= u=])

Attachments

(1 file)

As discussed in bug 980027 we would like the ability to possibly run the gecko compositor thread at RT priority. Currently, however, the buri gonk firmware interacts poorly with this schedule policy. In particular it forces us into a state where we hit the fb_lockBuffer() issue described in bug 980171. For example, see this profile: http://people.mozilla.org/~bgirard/cleopatra/#report=3948400a8133b559f3850e164f7e28b27c57d9ff This is quite bad with composites consistently taking 30ms. If we update the disp_loop() thread priority in the gralloc framebuffer to the same RT priority, however, things get dramatically better. Not only do we avoid the 30ms composites, but we see a reduction of the delays currently experienced in bug 980171. http://people.mozilla.org/~bgirard/cleopatra/#report=af28b72e2a28a81637691799f5d7b0149367a369 Whats more, it appears that the framebuffer thread at RT1 and the compositor thread without RT priority also works very well: http://people.mozilla.org/~bgirard/cleopatra/#report=1dbfd12c36c0b7d6a6223a8b24d4dff8965ef13a Therefore I think we should investigate ways to change the gralloc framebuffer's disp_loop() thread to RT priority 1. The attached patch shows the change I made to the gralloc library.
Comment on attachment 8390193 [details] [diff] [review] fb_rt_priority.patch Sushil, I know buri is not a priority at the moment, but what is the possibility of getting a patch like this upstreamed? Or is there another way for us to get its thread ID and manually adjust its priority from gecko? Thanks!
Attachment #8390193 - Flags: feedback?(sushilchauhan)
Blocks: 982972
Ben, If you are not seeing any issues in any other IOCTL by bumping up the priority, it should be OK to have this. Regarding patch up-streaming, Diego should comment on this because I do not know if we officially support ICS HAL based devices currently.
Flags: needinfo?(dwilson)
Attachment #8390193 - Flags: feedback?(sushilchauhan)
Are you planning on doing this in v1.3 or v1.4? Maybe mvines can comment on the state of CAF ICS support for those baselines.
Flags: needinfo?(dwilson) → needinfo?(mvines)
Our current focus is not on this baseline.
Flags: needinfo?(mvines)
Priority: -- → P3
Whiteboard: [c=uniformity p= s= u=]
Pretty sure this is not going to get fixed at this point if it's buri-only.
Status: NEW → RESOLVED
Closed: 10 years ago
Resolution: --- → INCOMPLETE
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: