Re: [Dri-devel] "DMA test failed" in mach64 branch

[Leif D. <lde...@re...>]

On 17 Aug 2002, J. Christopher Pereira wrote:

> On Sat, 2002-08-17 at 19:33, Leif Delgass wrote:
> > On 17 Aug 2002, J. Christopher Pereira wrote:
> > 
> > > On Sat, 2002-08-17 at 18:38, Leif Delgass wrote:
> > > 
> > > > Thanks.  OK, it looks like everything goes fine up until the kernel module 
> > > > initializtion, despite the BIOS problem.  So the next thing is to turn on 
> > > > debugging in the kernel module.  I wonder if you're having the same 
> > > > problem others are having there.  Could you try this:
> > > > 
> > > > Do a cold restart and then as root:
> > > > 
> > > > % modprobe mach64 drm_opts=debug
> > > > % cat /proc/kmsg > kmsg.txt
> > > > 
> > > > then start X from another console.  You just need to let X finish starting 
> > > > up and then you can log out and interrupt cat.  Then if you could post the 
> > > > log in the same place you put your X log, that would be great.
> > > 
> > > Here are is kmsg log: http://kripper.imatronix.cl/tmp/kmsg
> > 
> > Thanks.  Well, unfortunately, the part I was interested in didn't make it
> > into the log.  Sometimes stuff just gets lost.  I was looking for "Before
> > DMA Transfer" for the 3 registers.  Could you try it again once or twice
> > and see if that part shows up?
> 
> I had more luck this time, here is the new log:
> http://kripper.imatronix.cl/tmp/kmsg.2
> 
> This are the "before" registers:
> 
> <7>[drm:mach64_bm_dma_test] mach64_bm_dma_test
> <7>[drm:mach64_bm_dma_test] Allocating data memory ...
> <7>[drm:mach64_bm_dma_test] (Before DMA Transfer) reg 0 = 0x02390004
> <7>[drm:mach64_bm_dma_test] (Before DMA Transfer) reg 1 = 0x02390004
> <7>[drm:mach64_bm_dma_test] (Before DMA Transfer) reg 2 = 0x02390004

Yup.  This is what I figured, these should be all zeros, so the MMIO 
write to init. the registers fails.

> > Anyway, I suspect that the init. of the registers fails, because the
> > values are non-zero after the test.  It looks like this is probably the
> > same problem Dan and Scott are having, so you could just wait for the
> > patch to try testing again.  You all have the same family of chip (GTPro),
> > so it could be that the problem is specific to that flavor of mach64.
> 
> You are hacking on the mach64-0.0.5 branch, right?

Right.  I was planning on making a patch that tests MMIO writes to the
scratch registers (block 0, non-FIFOed), the pattern registers (block 0,
FIFO-ed), and some other block 1 registers.  The vertex registers used in
the test are aliased at two different locations, so I was also going to
try using VERTEX_1_SECONDARY_[S,T,W] for the test just in case, since they
are not aliased.  However, I can't imagine that any of the mach64's with a 
setup engine are not register compatible which each other.

Regarding the "Pseudo-DMA" mode, it won't work to enable it.  It was
broken by changes in the DMA code, and we haven't rewritten it yet.

-- 
Leif Delgass 
http://www.retinalburn.net