I've been trying to wrap my head around ttm and gem these last couple of weeks. I found the nice 'drm_mm_dump_table' function and ran it on the mgag200 drm_mm struct. Instant BUG in include/drm/drm_mm.h line 100.
static inline unsigned long drm_mm_hole_node_start(struct drm_mm_node *hole_node) { BUG_ON(!hole_node->hole_follows); return __drm_mm_hole_node_start(hole_node); }
where drm_mm_hole_node_start is called from drm_mm_dump_table like so:
int drm_mm_dump_table(struct seq_file *m, struct drm_mm *mm) { struct drm_mm_node *entry; unsigned long total_used = 0, total_free = 0, total = 0; unsigned long hole_start, hole_end, hole_size;
hole_start = drm_mm_hole_node_start(&mm->head_node); ... ... ...
as far as I can tell the head_node is a list of page offsets that point to free memory chunks. It seems drm_mm_dump_table expects a free chunk at the beginning of the list. What if all the memory is used? How can we ALWAYS expect a free chunk at the first element? Is this a bug in drm_mm_dump_table? Can't we just remove the first print and let the loop do all the work that comes right after? They look the same to me.
This all started from me trying to figure out where ttm/gem is putting buffer objects in VRAM. All the variables in a ttm_buffer_object that looks like they hold variables either have address outside of the total VRAM size, or 0.
When I removed the first print in drm_mm_dump_table I get the following output: 0x00100000-0x001007e9: 0x000007e9: used 0x001007e9-0x10100000: 0x0ffff817: free total: 268435456, used 2025 free 268433431
The board only has 16M of VRAM.
thanks, Chris