On 2019-06-04 3:36 p.m., Helge Deller wrote: > Actually, I think it makes sense to push the drop-sr1/use-lci-without-sr1 > change backward to the stable kernel series. > After that, in the second step, we could add the code to use lpa(), which > I don't think should go to stable series. > Would it be OK for you if we split it up into two patches? No problem. I didn't have an easy way to separate changes. There's a possible race in using %sr1 so doing a stable backport makes sense. Using lpa() is slightly slower than virt_to_phys() as it may generate a TLB miss. On the other hand, it handles any mapping. dave -- John David Anglin dave.anglin@xxxxxxxx
