> On Oct 5, 2019, at 5:22 PM, Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote: > > Apparently some console drivers can do memory allocation on the printk() > path. > > This behavior is daft, IMO. Have we identified which ones and looked > into fixing them? Not necessary that simple. It is more of 2+ CPUs required to trigger the deadlock. Please see my v2 patch I sent which has all the information. Especially, the thread link included there which contains a few lockdep splat traces and the s390 one in the patch description.
