graphics on G4

[Peter Grehan]

Hi Justin,

>>  The framebuffer uses a large chunk of kernel VM when mmap'd.
>> Are you able to dump some vm stats while doing this e.g.
>> sysctl vm.kvm_free ?

 I shouldn't hit send before the morning coffee :( mmap'ing
uses process VM, not kernel, except for the extra alloc's
needed for this but these should all come from the non-KVA UMA
small malloc area.

>There is no vm.kvm_free sysctl on my install.

 That's probably i386/amd64 only. Worth adding for ppc though.

>vm.vmtotal: 
>System wide totals computed every five seconds: (values in
kilobytes)
>===============================================
>Processes:              (RUNQ: 1 Disk Wait: 0 Page Wait: 0
Sleep: 74)
>Virtual Memory:         (Total: 1391388K, Active 288188K)
>Real Memory:            (Total: 142548K Active 130496K)
>Shared Virtual Memory:  (Total: 53356K Active: 51276K)
>Shared Real Memory:     (Total: 23100K Active: 22912K)
>Free Memory Pages:      58896K
>
>vm.loadavg: { 0.15 0.10 0.07 }
>
>What other stats are relevant?
>
>I neglected to mention before, but it happens even when I
exit out of X, or the
>program mmap()ing the video card quits.  I have a hunch that
it's overwriting
>some piece of memory needed by the process list.
 
 That sounds like a good place to start. For the case when the
program mmap's the video, does it write to all the pages ? One
thing to try would be to write a known value to screen memory
e.g. 0xdeadcode, and on exit see if that triggers a panic
rather than a hang.

later,

Peter.