This is a rework of this series: https://lore.kernel.org/lkml/20200921020007.35803-1-chenjun102@xxxxxxxxxx/ Originally I was investigating a percpu leak on our customer nodes and having this functionality was a huge help, which lead to this fix [1]. So probably it's a good idea to have it in mainstream too, especially as after [2] it became much easier to implement (we already have a separate tree for percpu pointers). commit 0af8c09c89681 ("netfilter: x_tables: fix percpu counter block leak on error path when creating new netns") [1] commit 39042079a0c24 ("kmemleak: avoid RCU stalls when freeing metadata for per-CPU pointers") [2] Pavel Tikhomirov (2): kmemleak: enable tracking for percpu pointers kmemleak-test: add percpu leak mm/kmemleak.c | 153 +++++++++++++++++++------------ samples/kmemleak/kmemleak-test.c | 2 + 2 files changed, 96 insertions(+), 59 deletions(-) -- 2.45.2
