On 22/05/18 11:25 AM, Dan Williams wrote: > As far as I can see by then it's too late, or we need to expose > release details to the caller which defeats the purpose of devm > semantics. In the dax/pmem case, I *think* it should be fine... devm_add_action_or_reset() only calls the action it is passed on failure not the entire devm chain. In which case, it should drop all the references to the percpu counter. Then, if on an error return from devm_memremap_pages() we call dax_pmem_percpu_kill(), the rest of the devm chain should be called when we return from a failed probe and it should proceed as usual. I think dax_pmem_percpu_kill() also must be called on any error return from devm_memremap_pages and it is currently not... Logan
