Measurements: Massive Display-related power savings
andy at openmoko.com
Tue Oct 28 02:38:24 CET 2008
-----BEGIN PGP SIGNED MESSAGE-----
Somebody in the thread at some point said:
| The real problem is in glamofb_blank() code (which I have written
| as part of the patchset regarding the measurements in this thread).
| All that we do in glamofb_blank() is to disable DCLK of the display
| controller. If I uncomment this code, the device properly comes back
| the blanking.
Thanks for looking into it further. Currently on stable-tracking we
have what seems to be quite stable suspend/resume now but always it
resumes with a WSOD, I found yesterday I was also looking in the wrong
place about jbt6k74 init because with a 'scope I saw no video data is
coming from Glamo and it's the DCLK issue you found too. However, I
differentially dumped the glamo registers before and after then to try
to identify the bad boy and I didn't see any relevant difference :-(
- -0040: 05db 51f2 0aba b000 0003 0000 0000 0000
- -0050: 000f 111e c443 111e 000f 0001 050f 020f
+0040: 05db 518d 0aba aedf 0003 0000 0000 0000
+0050: 000f 111e ccc3 111e 000f 0001 050f 020f
- -1180: 8023 0000 0000 0000 0000 0000 0000 0000
+1180: 81ce 0025 2b00 0003 0000 0000 0000 0000
This compared all the regs covered in
(note changed path on stable-tracking since Glamo is a child of PMU now).
Also, MMC is up and working fine so there is no memory- or pll- level
issue in Glamo when it is in this mode.
| It really puzzles me what could be the cause of this. Only few GTA02
| seem to show this behaviour. Mickey's device exhibiting this problem is
| a MP GTA02v6, just like other devices not exhibiting the problem. It
| reliable on a given device. Either it does always come back with a
| or it does always work. So it's unlikely timing related.
| There would be two questions for Smedia (who's the contact inside
| talk to them?):
| 1) Is there a certain condition when we can disable DCLK? Maybe stopping
| this clock is only valid in a certain state e.g. during vertical blank?
| What we want to do is to stop the glamo from producing any video output
| 2) Did Openmoko ever use different silicon revisions of the Glamo 3362 for
| different production runs?
I think we probably solve it ourselves before S-Media will solve it.
One thing to bear in mind, framebuffer notifcation callback action on
jbt6k74 driver is asynchronous to resume on that driver, if I understood
the situation. It would be one of a very shrunken number of races left
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org
-----END PGP SIGNATURE-----
More information about the openmoko-kernel