Re: [PATCH] mm: Remove VM_BUG_ON in __alloc_pages_node
From: Bharath Vedartham
Date: Wed Jun 05 2019 - 09:11:39 EST
[Not replying inline as my mail is bouncing back]
This patch is based on reading the code rather than a kernel crash. My
thought process was that if an invalid node id was passed to
__alloc_pages_node, it would be better to add a VM_WARN_ON and fail the
allocation rather than crashing the kernel.
I feel it would be better to fail the allocation early in the hot path
if an invalid node id is passed. This is irrespective of whether the
VM_[BUG|WARN]_*s are enabled or not. I do not see any checks in the hot
path for the node id, which in turn may cause NODE_DATA(nid) to fail to
get the pglist_data pointer for the node id.
We can optimise the branch by wrapping it around in unlikely(), if
performance is the issue?
What are your thoughts on this?