On Mon, Apr 16, 2018 at 9:01 AM, Joerg Roedel <j...@8bytes.org> wrote:
> Okay, I verify if there are any global bits left in the page-tables.
> According to the PTDUMP_X86 the cpu_entry_area is mapped with G=1 (which
> should be fine?) and another 4M range in the kernel mapping. I need to
> check what that is.
All the kernel entry code that is both in the user mapping and the
kernel mapping should be marked G.
We had missed a lot of it (and the impact is very small with PCID),
but if you rebased on top of 4.17-rc1 you should have it fixed at
least on 64-bit.
See for example commit 8c06c7740d19 ("x86/pti: Leave kernel text
global for !PCID") and in particular the performance numbers (that's
an Atom microserver, but it was chosen due to lack of PCID).