https://bugzilla.kernel.org/show_bug.cgi?id=201631 --- Comment #11 from Benjamin Herrenschmidt (benh@xxxxxxxxxxxxxxxxxxx) --- Jan, I don't know for sure what problems triggers it here, but there definitely is a HW bug on POWER9 with the "NestMMU" (the MMU that services coprocessors and the nvlink), which can cause a dirty bit to be set on a read-only PTE if it races with the PTE update & invalidation. It was "assumed" this was ok. If not I don't know how to work around it. However, at least in the above cases, I don't see how we would be setting dirty if we don't have RW access permission. A PowerMac G5 uses the hash MMU, we set dirty atomically with checking the PTE permission when hashing a PTE in. -- You are receiving this mail because: You are watching the assignee of the bug.
