Re: [PATCH] x86/bugs: Use fixed addressing for VERW operand

[Pawan Gupta]

On Thu, Feb 29, 2024 at 10:19:09AM +0100, Jiri Slaby wrote:
> On 29. 02. 24, 10:14, Pawan Gupta wrote:
> > On Wed, Feb 28, 2024 at 05:39:27PM -0800, Pawan Gupta wrote:
> > > On Tue, Feb 27, 2024 at 10:43:53AM +0100, Jiri Slaby wrote:
> > > > On 27. 02. 24, 9:47, Nikolay Borisov wrote:
> > > > > 
> > > > > 
> > > > > On 27.02.24 г. 1:52 ч., Pawan Gupta wrote:
> > > > > > Macro used for MDS mitigation executes VERW with relative addressing for
> > > > > > the operand. This is unnecessary and creates a problem for backports on
> > > > > > older kernels that don't support relocations in alternatives. Relocation
> > > > > > support was added by commit 270a69c4485d ("x86/alternative: Support
> > > > > > relocations in alternatives"). Also asm for fixed addressing is much
> > > > > > more cleaner than relative RIP addressing.
> > > > > > 
> > > > > > Simplify the asm by using fixed addressing for VERW operand.
> > > > > > 
> > > > > > Fixes: baf8361e5455 ("x86/bugs: Add asm helpers for executing VERW")
> > > > > > Reported-by: Nikolay Borisov <nik.borisov@xxxxxxxx>
> > > > > > Closes: https://lore.kernel.org/lkml/20558f89-299b-472e-9a96-171403a83bd6@xxxxxxxx/
> > > > > > Signed-off-by: Pawan Gupta <pawan.kumar.gupta@xxxxxxxxxxxxxxx>
> > > > > > ---
> > > > > >    arch/x86/include/asm/nospec-branch.h | 2 +-
> > > > > >    1 file changed, 1 insertion(+), 1 deletion(-)
> > > > > > 
> > > > > > diff --git a/arch/x86/include/asm/nospec-branch.h
> > > > > > b/arch/x86/include/asm/nospec-branch.h
> > > > > > index 2aa52cab1e46..ab19c7f1167b 100644
> > > > > > --- a/arch/x86/include/asm/nospec-branch.h
> > > > > > +++ b/arch/x86/include/asm/nospec-branch.h
> > > > > > @@ -323,7 +323,7 @@
> > > > > >     * Note: Only the memory operand variant of VERW clears the CPU
> > > > > > buffers.
> > > > > >     */
> > > > > >    .macro CLEAR_CPU_BUFFERS
> > > > > > -    ALTERNATIVE "", __stringify(verw _ASM_RIP(mds_verw_sel)),
> > > > > > X86_FEATURE_CLEAR_CPU_BUF
> > > > > > +    ALTERNATIVE "", __stringify(verw mds_verw_sel),
> > > > > > X86_FEATURE_CLEAR_CPU_BUF
> > > > > 
> > > > > Actually thinking about it more and discussing with Jiri (cc'ed), will
> > > > > this work with KASLR enabled?
> > > > 
> > > > I might of course be wrong. We appear to rely on the asm+linker here.
> > > 
> > > You were right, with KASLR enabled, instructions with fixed addressing
> > > in alternatives don't get relocated. I guess we will have to keep
> > > rip-relative as is. Thanks for catching that.
> > 
> > Looks like this is not settled yet, it was naive of me to trust gdb on
> > /proc/kcore earlier with KASLR enabled.
> > 
> > With the below debug patch it appears the relocation with fixed
> > addresses is working as expected with KASLR enabled.
> 
> As I wrote already, asm+linker converts the fixed address to rip-rela
> anyway:
> 
> https://lore.kernel.org/all/fd8f2df0-563e-4f5c-aca4-bc92a14e9426@xxxxxxxxxx/
> 
> I also raised questions in there:
> ====
> The assembler generates a relocation for the fixed address anyway. And
> the linker resolves it as rip-relative. At least the pair from my
> binutils-2.42.
> 
> But if it generates a rip-relative address, is < 6.5 with no support of
> rip-rel in alternatives still fine?

In that case backports < 6.5 can do:

+.macro CLEAR_CPU_BUFFERS
+	ALTERNATIVE "jmp .Lskip_verw_\@", "", X86_FEATURE_CLEAR_CPU_BUF
+	verw _ASM_RIP(mds_verw_sel)
+.Lskip_verw_\@:
+.endm

As done in Nikolay's 5.4 backport:

https://lore.kernel.org/stable/20240226122237.198921-3-nik.borisov@xxxxxxxx/

> Another question: can we rely on the assembler to generate a relocation
> and on the linker to resolve it as rip-relative?

This is definitely not my area of expertise, but with the above approach
VERW should be inlined always, and rip-relative should be resolved as
with any other instruction.