> But are you sure i386 is broken as well? The "&= PG_FRAME" there > doesn't do any damage on i386 and i386 doesn't have the VA space hole > that cause more grief on amd64 beyond that.
I am not; I am not able to test the troublesome change on i386. But due to the pmap code being similar with no particular attention to pmap_kernel(), I wouldn't be surprised if it doesn't work correctly, even if the masking operation does not damage the addresses too much.
