iris: Improving zombie_list parsing and cleanup
After the discussions on !21460 (merged), we analysed zombie list parsing. We observed that after the mentioned fix zombie lists is not growing too big as most of the buffers are getting closed at bo_free()
itself.
In the tests which we ran, we saw that the few buffers which get added to zombie list are mostly freed in the next round of cleanup_bo_cache()
.
But I think there still can be cases, where there is a busy buffer in zombie list and the buffers after that are free. In such cases, changing the break
to continue
in cleanup_bo_cache()
zombie_list parsing logic would be beneficial.
But this seems to be a very rare case and I couldn't really find a solid proof for this that it can happen.
Note that, such cases wherein a lot of free buffers were there after a busy buffer was seen before the 21460 fix.
What is your opinion on this @kwg @llandwerlin @pzanoni?