On Tue, 2025-07-22 at 00:53 +0000, Huang, Kai wrote:
> On Mon, 2025-07-21 at 17:44 -0700, H. Peter Anvin wrote:
> > On July 21, 2025 4:42:22 PM PDT, "Huang, Kai" <kai.huang@xxxxxxxxx> wrote:
> > > On Mon, 2025-07-21 at 23:29 +0000, Huang, Kai wrote:
> > > > > On July 21, 2025 2:36:48 PM PDT, "Huang, Kai" <kai.huang@xxxxxxxxx> wrote:
> > > > > > On Mon, 2025-07-21 at 16:27 -0500, Tom Lendacky wrote:
> > > > > > > > > > @@ -204,7 +202,7 @@
> > > > > SYM_CODE_START_LOCAL_NOALIGN(identity_mapped)
> > > > > > > > > > * entries that will conflict with the now unencrypted memory
> > > > > > > > > > * used by kexec. Flush the caches before copying the kernel.
> > > > > > > > > > */
> > > > > > > > > > - testq %r8, %r8
> > > > > > > > > > + testq $RELOC_KERNEL_HOST_MEM_ACTIVE, %r11
> > > > > > > > >
> > > > > > > > > Hmmm... can't both bits be set at the same time? If so, then this
> > > > > > > > > will fail. This should be doing bit tests now.
> > > > > > > >
> > > > > > > > TEST instruction performs logical AND of the two operands,
> > > > > > > > therefore the above equals to:
> > > > > > > >
> > > > > > > > set ZF if "R11 AND BIT(1) == 0".
> > > > > > > >
> > > > > > > > Whether there's any other bits set in R11 doesn't impact the above, right?
> > > > > > > >
> > > > > > >
> > > > > > > Doh! My bad, yes, not sure what I was thinking there.
> > > > > > >
> > > > > >
> > > > > > Np and thanks! I'll address your other comments but I'll see whether
> > > > > > Boris has any other comments first.
> > > > > >
> > > > >
> > > > > You can use testb in this case to save 3 bytes, too.
> > > >
> > > > Yeah I can do that, thanks for the info!
> > >
> > > I just tried. I need to do:
> > >
> > > testb $RELOC_KERNEL_HOST_MEM_ACTIVE, %r11b
> > >
> > > in order to compile, otherwise using plain %r11 generates:
> > >
> > > arch/x86/kernel/relocate_kernel_64.S:212: Error: `%r11' not allowed with
> > > `testb'
> > >
> > > I'll do some test and if there's no problem I'll switch to use this way,
> > > assuming it still saves us 3-bytes.
> > >
> >
> > That works just fine.
>
> Yeah I have now tested. Thanks :-)
Sorry for multiple emails. One minor thing related to using testb:
With this patch there's one movl to R11:
movl $RELOC_KERNEL_PRESERVE_CONTEXT, %r11d
I think it would be better to make it consistent with testb:
movb $RELOC_KERNEL_PRESERVE_CONTEXT, %r11b
Please let me know if there's any concern?
