When building a kernel with CONFIG_PARAVIRT_XXL the paravirt
infrastructure will always use functions for reading or writing MSRs,
even when running on bare metal.

Switch to inline RDMSR/WRMSR instructions in this case, reducing the
paravirt overhead.

The first patch is a prerequisite fix for alternative patching. Its
is needed due to the initial indirect call needs to be padded with
NOPs in some cases with the following patches.

In order to make this less intrusive, some further reorganization of
the MSR access helpers is done in the patches 2-5.

The next 5 patches are converting the non-paravirt case to use direct
inlining of the MSR access instructions, including the WRMSRNS
instruction and the immediate variants of RDMSR and WRMSR if possible.

Patches 11-13 are some further preparations for making the real switch
to directly patch in the native MSR instructions easier.

Patch 14 is switching the paravirt MSR function interface from normal
call ABI to one more similar to the native MSR instructions.

Patch 15 is a little cleanup patch.

Patch 16 is the final step for patching in the native MSR instructions
when not running as a Xen PV guest.

Patch 17 converts the rest of the MSR helpers to __always_inline.

This series has been tested to work with Xen PV and on bare metal.

Based on [1] and [2].

Changes since V4:
- Rebase
- dropped patch 3 of V4, as already covered by [1]

Changes since V3:
- Rebase
- wrmsrns() related changes (patches 9+10)

Changes since V2:
- switch back to the paravirt approach

Changes since V1:
- Use Xin Li's approach for inlining
- Several new patches

[1]: https://lore.kernel.org/lkml/20260911074530.3140830-1-jgross@suse.com/T/#t
[2]: https://lore.kernel.org/lkml/20260911075216.3142309-1-jgross@suse.com/T/#t

Juergen Gross (17):
  x86/alternative: Support alt_replace_call() with instructions after
    call
  coco/tdx: Rename MSR access helpers
  x86/msr: Minimize usage of native_*() msr access functions
  x86/msr: Move MSR trace calls one function level up
  x86/hyperv: Switch from __rdmsr() to native_rdmsrq()
  x86/opcode: Add immediate form MSR instructions
  x86/extable: Add support for immediate form MSR instructions
  x86/msr: Make wrmsrns() a first class citizen
  x86/msr: Introduce sync_cpu_after_wrmsrns()
  x86/msr: Use the alternatives mechanism for RDMSR
  x86/alternatives: Add ALTERNATIVE_4()
  x86/paravirt: Split off MSR related hooks into new header
  x86/paravirt: Prepare support of MSR instruction interfaces
  x86/paravirt: Switch MSR access pv_ops functions to instruction
    interfaces
  x86/msr: Reduce number of low level MSR access helpers
  x86/paravirt: Use alternatives for MSR access with paravirt
  x86/msr: Make all MSR access functions __always_inline

 arch/x86/coco/tdx/tdx.c                   |   8 +-
 arch/x86/hyperv/hv_crash.c                |   6 +-
 arch/x86/hyperv/ivm.c                     |   2 +-
 arch/x86/include/asm/alternative.h        |   6 +
 arch/x86/include/asm/fred.h               |   2 +-
 arch/x86/include/asm/msr.h                | 340 +++++++++++++++++-----
 arch/x86/include/asm/paravirt-msr.h       | 180 ++++++++++++
 arch/x86/include/asm/paravirt.h           |  45 ---
 arch/x86/include/asm/paravirt_types.h     |  57 ++--
 arch/x86/include/asm/qspinlock_paravirt.h |   4 +-
 arch/x86/kernel/alternative.c             |   5 +-
 arch/x86/kernel/cpu/mshyperv.c            |   4 +-
 arch/x86/kernel/kvmclock.c                |   2 +-
 arch/x86/kernel/paravirt.c                |  42 ++-
 arch/x86/kvm/svm/svm.c                    |  16 +-
 arch/x86/lib/x86-opcode-map.txt           |   5 +-
 arch/x86/mm/extable.c                     |  41 ++-
 arch/x86/xen/enlighten_pv.c               |  52 +++-
 arch/x86/xen/pmu.c                        |   4 +-
 tools/arch/x86/lib/x86-opcode-map.txt     |   5 +-
 tools/objtool/check.c                     |   1 +
 21 files changed, 632 insertions(+), 195 deletions(-)
 create mode 100644 arch/x86/include/asm/paravirt-msr.h

-- 
2.55.0

On 11.09.26 10:41, Juergen Gross wrote:
> When building a kernel with CONFIG_PARAVIRT_XXL the paravirt
> infrastructure will always use functions for reading or writing MSRs,
> even when running on bare metal.

Hi Juergen,

This doesn't build with CONFIG_PARAVIRT_XXL=y. 16/17 and 17/17 are where
it breaks, but the cause is the .byte fallbacks: ASM_WRMSRNS_IMM from
08/17 and ASM_RDMSR_IMM from 10/17 don't end in a separator, unlike the
.insn variants above them.

  #define ASM_RDMSR_IMM				\
	" .byte 0xc4,0xe7,0x7b,0xf6,0xc0; .long %c[msr]"

That worked while they were only ever the last argument of an
ALTERNATIVE(), which appends its own newline. 16/17 concatenates them
with ASM_CLRERR:

	ASM_RDMSR_IMM ASM_CLRERR, X86_FEATURE_MSR_IMM,	\

so the .long operand runs into the xor. From
make arch/x86/kernel/cpu/common.s:

	.byte 0xc4,0xe7,0x7b,0xf6,0xc0; .long 266xor %rdx,%rdx

  paravirt-msr.h:165: Error: junk at end of line, first unrecognized character is `x'

clang reports "error: unexpected token" in the same place. 71 objects
fail, the same 71 either way, no vmlinux.

msr.h chooses between the .insn form and the .byte fallback with:

	#if defined(CONFIG_AS_IS_GNU) && CONFIG_AS_VERSION >= 24100

Two kinds of toolchain end up on the .byte side of that test:

  - GNU as older than 2.41. RHEL 9 and CentOS Stream 9 ship 2.35, and
    Documentation/process/changes.rst sets the minimum at 2.30, so this
    is a supported configuration rather than an old outlier.
  - clang, any version. CONFIG_AS_IS_GNU is never set for clang, so the
    && short-circuits and the version comparison is never reached. Your
    08/17 comment already notes that clang has no .insn support.

gcc with binutils 2.41 or newer takes the .insn path, where both macros
do end in a separator, and is unaffected.

Reproduced on v7.3-rc2 with your v3 00/13, v2 0/5 and v5 00/17 applied in
that order, x86_64 defconfig plus HYPERVISOR_GUEST, PARAVIRT, XEN and
XEN_PV. gcc 11.5.0 with GNU as 2.35.2, and clang 21.1.7.

Terminating both fallbacks fixes it, and both toolchains then build
vmlinux with no errors or warnings:

diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h
index eba325ecfe4c..529c13553c63 100644
--- a/arch/x86/include/asm/msr.h
+++ b/arch/x86/include/asm/msr.h
@@ -78,9 +78,9 @@ static inline void do_trace_rdpmc(u32 msr, u64 val, int failed) {}
  * form MSR access instructions reference %rax as the register operand.
  */
 #define ASM_RDMSR_IMM				\
-	" .byte 0xc4,0xe7,0x7b,0xf6,0xc0; .long %c[msr]"
+	" .byte 0xc4,0xe7,0x7b,0xf6,0xc0; .long %c[msr]\n\t"
 #define ASM_WRMSRNS_IMM				\
-	" .byte 0xc4,0xe7,0x7a,0xf6,0xc0; .long %c[msr]"
+	" .byte 0xc4,0xe7,0x7a,0xf6,0xc0; .long %c[msr]\n\t"
 #endif

 #define RDMSR_AND_SAVE_RESULT			\

ASM_WRMSRNS needs no change, _ASM_BYTES() already emits a semicolon.

The WRMSRNS line belongs in 08/17 and the RDMSR line in 10/17.

Thanks,
Shreshth

On Wed, Sep 23, 2026 at 12:36 PM Shreshth Srivastava
<shreshth.srivastava@intel.com> wrote:
>
> On 11.09.26 10:41, Juergen Gross wrote:
> > When building a kernel with CONFIG_PARAVIRT_XXL the paravirt
> > infrastructure will always use functions for reading or writing MSRs,
> > even when running on bare metal.
>
> Hi Juergen,
>
> This doesn't build with CONFIG_PARAVIRT_XXL=y. 16/17 and 17/17 are where
> it breaks, but the cause is the .byte fallbacks: ASM_WRMSRNS_IMM from
> 08/17 and ASM_RDMSR_IMM from 10/17 don't end in a separator, unlike the
> .insn variants above them.
>
>   #define ASM_RDMSR_IMM                         \
>         " .byte 0xc4,0xe7,0x7b,0xf6,0xc0; .long %c[msr]"
>
> That worked while they were only ever the last argument of an
> ALTERNATIVE(), which appends its own newline. 16/17 concatenates them
> with ASM_CLRERR:
>
>         ASM_RDMSR_IMM ASM_CLRERR, X86_FEATURE_MSR_IMM,  \
>
> so the .long operand runs into the xor. From
> make arch/x86/kernel/cpu/common.s:
>
>         .byte 0xc4,0xe7,0x7b,0xf6,0xc0; .long 266xor %rdx,%rdx
>
>   paravirt-msr.h:165: Error: junk at end of line, first unrecognized character is `x'
>
> clang reports "error: unexpected token" in the same place. 71 objects
> fail, the same 71 either way, no vmlinux.
>
> msr.h chooses between the .insn form and the .byte fallback with:
>
>         #if defined(CONFIG_AS_IS_GNU) && CONFIG_AS_VERSION >= 24100
>
> Two kinds of toolchain end up on the .byte side of that test:
>
>   - GNU as older than 2.41. RHEL 9 and CentOS Stream 9 ship 2.35, and
>     Documentation/process/changes.rst sets the minimum at 2.30, so this
>     is a supported configuration rather than an old outlier.
>   - clang, any version. CONFIG_AS_IS_GNU is never set for clang, so the
>     && short-circuits and the version comparison is never reached. Your
>     08/17 comment already notes that clang has no .insn support.

Indeed, looks like we're missing support for .insn for x86.
Filed https://github.com/llvm/llvm-project/issues/225916.
(Please do file bugs against the toolchain when you encounter issues
like this, and cc someone from kernel development).

>
> gcc with binutils 2.41 or newer takes the .insn path, where both macros
> do end in a separator, and is unaffected.
>
> Reproduced on v7.3-rc2 with your v3 00/13, v2 0/5 and v5 00/17 applied in
> that order, x86_64 defconfig plus HYPERVISOR_GUEST, PARAVIRT, XEN and
> XEN_PV. gcc 11.5.0 with GNU as 2.35.2, and clang 21.1.7.
>
> Terminating both fallbacks fixes it, and both toolchains then build
> vmlinux with no errors or warnings:
>
> diff --git a/arch/x86/include/asm/msr.h b/arch/x86/include/asm/msr.h
> index eba325ecfe4c..529c13553c63 100644
> --- a/arch/x86/include/asm/msr.h
> +++ b/arch/x86/include/asm/msr.h
> @@ -78,9 +78,9 @@ static inline void do_trace_rdpmc(u32 msr, u64 val, int failed) {}
>   * form MSR access instructions reference %rax as the register operand.
>   */
>  #define ASM_RDMSR_IMM                          \
> -       " .byte 0xc4,0xe7,0x7b,0xf6,0xc0; .long %c[msr]"
> +       " .byte 0xc4,0xe7,0x7b,0xf6,0xc0; .long %c[msr]\n\t"
>  #define ASM_WRMSRNS_IMM                                \
> -       " .byte 0xc4,0xe7,0x7a,0xf6,0xc0; .long %c[msr]"
> +       " .byte 0xc4,0xe7,0x7a,0xf6,0xc0; .long %c[msr]\n\t"
>  #endif
>
>  #define RDMSR_AND_SAVE_RESULT                  \
>
> ASM_WRMSRNS needs no change, _ASM_BYTES() already emits a semicolon.
>
> The WRMSRNS line belongs in 08/17 and the RDMSR line in 10/17.
>
> Thanks,
> Shreshth



-- 
Thanks,
~Nick Desaulniers