On Thu, Feb 10, 2022 at 11:09:55AM +1100, Alistair Popple wrote: > Yes, that's the cause. The fix[1] from SeongJae looks correct. What's easier > for you - taking that fix or me posting a v3 of my series including that now? It isn't. Your patch uses migrate_vma_* for device coherent pages, so we need to build that code even if support for private pages is no enabled. The right fix is something like this: http://git.infradead.org/users/hch/misc.git/commitdiff/f22a8437b590f943cdde1d56b2de0ec214de79e4 I'll repost this as part of the series to redo the pgmap page refcounting.
