Re: [Dri-devel] Some Mobile Radeon problems

[Keith W. <ke...@tu...>]

> I've done some more digging on this.  Enabling RADEON_DEBUG=sanity doesn't
> seem to do anything.  However, I did add some DRM_INFO messages in
> radeon_state.c, and I got the following log messages:
> 
> [drm:radeon_cp_cmdbuf] RADEON_CMD_PACKET
> [drm:radeon_emit_packets] cmdsz/bufsz = 4/48
> [drm:radeon_emit_packets] cmdsz/bufsz = 4/44
> [drm:radeon_cp_cmdbuf] RADEON_CMD_PACKET
> [drm:radeon_emit_packets] cmdsz/bufsz = 12/40
> [drm:radeon_emit_packets] cmdsz/bufsz = 12/28
> [drm:radeon_cp_cmdbuf] RADEON_CMD_PACKET
> [drm:radeon_emit_packets] cmdsz/bufsz = 4/24
> [drm:radeon_emit_packets] cmdsz/bufsz = 4/20
> [drm:radeon_cp_cmdbuf] RADEON_CMD_PACKET
> [drm:radeon_emit_packets] cmdsz/bufsz = 8/16
> [drm:radeon_emit_packets] cmdsz/bufsz = 8/8
> [drm:radeon_cp_cmdbuf] RADEON_CMD_PACKET
> [drm:radeon_emit_packets] cmdsz/bufsz = 4/4
> [drm:radeon_emit_packets] cmdsz/bufsz = 4/0
> [drm:radeon_cp_cmdbuf] RADEON_CMD_PACKET3_CLIP
> [drm:radeon_emit_packet3_cliprect] nbox/cmdsz/bufsz/orig_nbox = 1/764/50332413/1
> [drm:radeon_emit_packet3_cliprect] nbox/cmdsz/bufsz = 0/764/50331649
> [drm:radeon_cp_cmdbuf] *ERROR* bad cmd_type 63 at 088a8c3c
> 
> So, the buffer size slowly drains until it rolls over from 0 bytes free to
> ~50MB free.  This seems a bit suspect to me. :)  I'm still digging, but
> perhaps the will jog somebody's brain.  Ideas?

There should never be more than about 16k in those buffers, so this is a big 
smoking gun!

Keith