radeon driver: checkpoint

[Matthieu Herrb]

Benjamin Herrenschmidt wrote:
>> Here are some updates. Sorry for taking so long to do that.
>>
>> 1. With the current ati-1-0-branch, the mergedfb problem is gone. The 
>> screen comes up nicely. The traditional 2-screens mode still doesn't 
>> produce any output on the 2nd head.
>>
>> 2. I installed Linux on a disk for this machine in order to be able to 
>> run radeontool and get register dumps. The behaviour is the same as 
>> under OpenBSD: mergedfb works well with the current ati-1-0-branch 
>> driver, but no signal on 2nd head with the traditional 2-screen mode.
>> X.Org 6.8.2 works in this mode (but mergedfb displays garbage).
>>
>> I've attached register dumps.
> 
> Thanks !
> 
> I spotted a couple of "interesting" ones (at the very least)
> 
> -BUS_CNTL (0030)        0x5133a3b0
> +BUS_CNTL (0030)        0x00000000
> 
> ARGH ! That is bad bad bad !
> 
> Can you try modifying RADEONSaveCommonRegisters()
> 
> And add:
> 
>     save->bus_cntl           = INREG(RADEON_BUS_CNTL);
> 
> to it ?

Done, but it doesn't help.

> 
> Also, I spotted:
> 
> -CRTC_EXT_CNTL (0054)   0x00008048
> +CRTC_EXT_CNTL (0054)   0x00000048
> 
> RADEON_CRTC_CRT_ON is clear here, that means that the CRT is off, is
> that expected ?  (That is, your primary head is a flat panel ?

My primary head is a flat panel connected to the DVI connector of the 
card, the 2nd head is connected to the analog 15 pins VGA connector.

> Does it
> "fixes" your problem if you use radeontool to dynamically "fix" the
> value to 0x8048 (radeontool regset CRTC_EXT_CNTL 0x8048)

With the above patch in place, this activates the video signal on the 
2nd head, and it displays screen 1 as expected.

> 
> Also, you seem to have tiling enabled in the 7.0 dump and not the 6.82
> dump, is that actually the case ?

The log file for 7.0 says it's enabled on primary head and disabled on 
the secondary head. The log file for 6.8.2 doesn't mention if it's 
enabled or not.  All it says is

Setting up tile and stipple cache:
     32 128x128 slots
     32 256x256 slots
     16 512x512 slots

-- 
Matthieu