[PATCH] x86/kexec: Disable KCOV instrumentation after load_segments()

[PATCH] x86/kexec: Disable KCOV instrumentation after load_segments()

[Aleksandr Nogikh]

The load_segments() function changes segment registers, invalidating
GS base (which KCOV relies on for per-cpu data). When CONFIG_KCOV is
enabled, any subsequent instrumented C code call (e.g.
native_gdt_invalidate()) begins crashing the kernel in an
endless loop.

To reproduce the problem, it's sufficient to do kexec on a
KCOV-instrumented kernel:
$ kexec -l /boot/otherKernel
$ kexec -e

(additional problems arise when the kernel is booting into a crash
kernel)

Disabling instrumentation for the individual functions would be too
fragile, so let's fix the bug by disabling KCOV instrumentation for
the whole machine_kexec_64.c and physaddr.c.

The problem is not relevant for 32 bit kernels as CONFIG_KCOV is not
supported there.

Signed-off-by: Aleksandr Nogikh <nogikh@google.com>
Cc: stable@vger.kernel.org
---
 arch/x86/kernel/Makefile | 4 ++++
 arch/x86/mm/Makefile     | 4 ++++
 2 files changed, 8 insertions(+)

diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index e9aeeeafad173..5703fa6027866 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -43,6 +43,10 @@ KCOV_INSTRUMENT_dumpstack_$(BITS).o			:= n
 KCOV_INSTRUMENT_unwind_orc.o				:= n
 KCOV_INSTRUMENT_unwind_frame.o				:= n
 KCOV_INSTRUMENT_unwind_guess.o				:= n
+# When a kexec kernel is loaded, calling load_segments() breaks all
+# subsequent KCOV instrumentation until new kernel takes control.
+# Keep KCOV instrumentation disabled to prevent kernel crashes.
+KCOV_INSTRUMENT_machine_kexec_64.o			:= n
 
 CFLAGS_head32.o := -fno-stack-protector
 CFLAGS_head64.o := -fno-stack-protector
diff --git a/arch/x86/mm/Makefile b/arch/x86/mm/Makefile
index 5b9908f13dcfd..a678a38a40266 100644
--- a/arch/x86/mm/Makefile
+++ b/arch/x86/mm/Makefile
@@ -4,6 +4,10 @@ KCOV_INSTRUMENT_tlb.o			:= n
 KCOV_INSTRUMENT_mem_encrypt.o		:= n
 KCOV_INSTRUMENT_mem_encrypt_amd.o	:= n
 KCOV_INSTRUMENT_pgprot.o		:= n
+# When a kexec kernel is loaded, calling load_segments() breaks all
+# subsequent KCOV instrumentation until new kernel takes control.
+# Keep KCOV instrumentation disabled to prevent kernel crashes.
+KCOV_INSTRUMENT_physaddr.o		:= n
 
 KASAN_SANITIZE_mem_encrypt.o		:= n
 KASAN_SANITIZE_mem_encrypt_amd.o	:= n
-- 
2.53.0.273.g2a3d683680-goog

Re: [PATCH] x86/kexec: Disable KCOV instrumentation after load_segments()

[Aleksandr Nogikh]

On Mon, Feb 16, 2026 at 6:37 PM Aleksandr Nogikh <nogikh@google.com> wrote:
>
> The load_segments() function changes segment registers, invalidating
> GS base (which KCOV relies on for per-cpu data). When CONFIG_KCOV is
> enabled, any subsequent instrumented C code call (e.g.
> native_gdt_invalidate()) begins crashing the kernel in an
> endless loop.
>
> To reproduce the problem, it's sufficient to do kexec on a
> KCOV-instrumented kernel:
> $ kexec -l /boot/otherKernel
> $ kexec -e
>
> (additional problems arise when the kernel is booting into a crash
> kernel)
>
> Disabling instrumentation for the individual functions would be too
> fragile, so let's fix the bug by disabling KCOV instrumentation for
> the whole machine_kexec_64.c and physaddr.c.
>
> The problem is not relevant for 32 bit kernels as CONFIG_KCOV is not
> supported there.
>
> Signed-off-by: Aleksandr Nogikh <nogikh@google.com>
> Cc: stable@vger.kernel.org
> ---
>  arch/x86/kernel/Makefile | 4 ++++
>  arch/x86/mm/Makefile     | 4 ++++
>  2 files changed, 8 insertions(+)

A gentle ping on this patch.

Should it go through the x86 tree or the mm tree like other kcov patches?

>
> diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
> index e9aeeeafad173..5703fa6027866 100644
> --- a/arch/x86/kernel/Makefile
> +++ b/arch/x86/kernel/Makefile
> @@ -43,6 +43,10 @@ KCOV_INSTRUMENT_dumpstack_$(BITS).o                  := n
>  KCOV_INSTRUMENT_unwind_orc.o                           := n
>  KCOV_INSTRUMENT_unwind_frame.o                         := n
>  KCOV_INSTRUMENT_unwind_guess.o                         := n
> +# When a kexec kernel is loaded, calling load_segments() breaks all
> +# subsequent KCOV instrumentation until new kernel takes control.
> +# Keep KCOV instrumentation disabled to prevent kernel crashes.
> +KCOV_INSTRUMENT_machine_kexec_64.o                     := n
>
>  CFLAGS_head32.o := -fno-stack-protector
>  CFLAGS_head64.o := -fno-stack-protector
> diff --git a/arch/x86/mm/Makefile b/arch/x86/mm/Makefile
> index 5b9908f13dcfd..a678a38a40266 100644
> --- a/arch/x86/mm/Makefile
> +++ b/arch/x86/mm/Makefile
> @@ -4,6 +4,10 @@ KCOV_INSTRUMENT_tlb.o                  := n
>  KCOV_INSTRUMENT_mem_encrypt.o          := n
>  KCOV_INSTRUMENT_mem_encrypt_amd.o      := n
>  KCOV_INSTRUMENT_pgprot.o               := n
> +# When a kexec kernel is loaded, calling load_segments() breaks all
> +# subsequent KCOV instrumentation until new kernel takes control.
> +# Keep KCOV instrumentation disabled to prevent kernel crashes.
> +KCOV_INSTRUMENT_physaddr.o             := n
>
>  KASAN_SANITIZE_mem_encrypt.o           := n
>  KASAN_SANITIZE_mem_encrypt_amd.o       := n
> --
> 2.53.0.273.g2a3d683680-goog
>

Re: [PATCH] x86/kexec: Disable KCOV instrumentation after load_segments()

[Dmitry Vyukov]

On Mon, 16 Feb 2026 at 18:37, Aleksandr Nogikh <nogikh@google.com> wrote:
>
> The load_segments() function changes segment registers, invalidating
> GS base (which KCOV relies on for per-cpu data). When CONFIG_KCOV is
> enabled, any subsequent instrumented C code call (e.g.
> native_gdt_invalidate()) begins crashing the kernel in an
> endless loop.
>
> To reproduce the problem, it's sufficient to do kexec on a
> KCOV-instrumented kernel:
> $ kexec -l /boot/otherKernel
> $ kexec -e
>
> (additional problems arise when the kernel is booting into a crash
> kernel)
>
> Disabling instrumentation for the individual functions would be too
> fragile, so let's fix the bug by disabling KCOV instrumentation for
> the whole machine_kexec_64.c and physaddr.c.
>
> The problem is not relevant for 32 bit kernels as CONFIG_KCOV is not
> supported there.
>
> Signed-off-by: Aleksandr Nogikh <nogikh@google.com>
> Cc: stable@vger.kernel.org

Reviewed-by: Dmitry Vyukov <dvyukov@google.com>

> ---
>  arch/x86/kernel/Makefile | 4 ++++
>  arch/x86/mm/Makefile     | 4 ++++
>  2 files changed, 8 insertions(+)
>
> diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
> index e9aeeeafad173..5703fa6027866 100644
> --- a/arch/x86/kernel/Makefile
> +++ b/arch/x86/kernel/Makefile
> @@ -43,6 +43,10 @@ KCOV_INSTRUMENT_dumpstack_$(BITS).o                  := n
>  KCOV_INSTRUMENT_unwind_orc.o                           := n
>  KCOV_INSTRUMENT_unwind_frame.o                         := n
>  KCOV_INSTRUMENT_unwind_guess.o                         := n
> +# When a kexec kernel is loaded, calling load_segments() breaks all
> +# subsequent KCOV instrumentation until new kernel takes control.
> +# Keep KCOV instrumentation disabled to prevent kernel crashes.
> +KCOV_INSTRUMENT_machine_kexec_64.o                     := n
>
>  CFLAGS_head32.o := -fno-stack-protector
>  CFLAGS_head64.o := -fno-stack-protector
> diff --git a/arch/x86/mm/Makefile b/arch/x86/mm/Makefile
> index 5b9908f13dcfd..a678a38a40266 100644
> --- a/arch/x86/mm/Makefile
> +++ b/arch/x86/mm/Makefile
> @@ -4,6 +4,10 @@ KCOV_INSTRUMENT_tlb.o                  := n
>  KCOV_INSTRUMENT_mem_encrypt.o          := n
>  KCOV_INSTRUMENT_mem_encrypt_amd.o      := n
>  KCOV_INSTRUMENT_pgprot.o               := n
> +# When a kexec kernel is loaded, calling load_segments() breaks all
> +# subsequent KCOV instrumentation until new kernel takes control.
> +# Keep KCOV instrumentation disabled to prevent kernel crashes.
> +KCOV_INSTRUMENT_physaddr.o             := n
>
>  KASAN_SANITIZE_mem_encrypt.o           := n
>  KASAN_SANITIZE_mem_encrypt_amd.o       := n
> --
> 2.53.0.273.g2a3d683680-goog
>

Re: [PATCH] x86/kexec: Disable KCOV instrumentation after load_segments()

[Aleksandr Nogikh]

+Cc linux-mm

On Fri, Feb 27, 2026 at 2:26 PM Dmitry Vyukov <dvyukov@google.com> wrote:
>
> On Mon, 16 Feb 2026 at 18:37, Aleksandr Nogikh <nogikh@google.com> wrote:
> >
> > The load_segments() function changes segment registers, invalidating
> > GS base (which KCOV relies on for per-cpu data). When CONFIG_KCOV is
> > enabled, any subsequent instrumented C code call (e.g.
> > native_gdt_invalidate()) begins crashing the kernel in an
> > endless loop.
> >
> > To reproduce the problem, it's sufficient to do kexec on a
> > KCOV-instrumented kernel:
> > $ kexec -l /boot/otherKernel
> > $ kexec -e
> >
> > (additional problems arise when the kernel is booting into a crash
> > kernel)
> >
> > Disabling instrumentation for the individual functions would be too
> > fragile, so let's fix the bug by disabling KCOV instrumentation for
> > the whole machine_kexec_64.c and physaddr.c.
> >
> > The problem is not relevant for 32 bit kernels as CONFIG_KCOV is not
> > supported there.
> >
> > Signed-off-by: Aleksandr Nogikh <nogikh@google.com>
> > Cc: stable@vger.kernel.org
>
> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
>
> > ---
> >  arch/x86/kernel/Makefile | 4 ++++
> >  arch/x86/mm/Makefile     | 4 ++++
> >  2 files changed, 8 insertions(+)
> >
> > diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
> > index e9aeeeafad173..5703fa6027866 100644
> > --- a/arch/x86/kernel/Makefile
> > +++ b/arch/x86/kernel/Makefile
> > @@ -43,6 +43,10 @@ KCOV_INSTRUMENT_dumpstack_$(BITS).o                  := n
> >  KCOV_INSTRUMENT_unwind_orc.o                           := n
> >  KCOV_INSTRUMENT_unwind_frame.o                         := n
> >  KCOV_INSTRUMENT_unwind_guess.o                         := n
> > +# When a kexec kernel is loaded, calling load_segments() breaks all
> > +# subsequent KCOV instrumentation until new kernel takes control.
> > +# Keep KCOV instrumentation disabled to prevent kernel crashes.
> > +KCOV_INSTRUMENT_machine_kexec_64.o                     := n
> >
> >  CFLAGS_head32.o := -fno-stack-protector
> >  CFLAGS_head64.o := -fno-stack-protector
> > diff --git a/arch/x86/mm/Makefile b/arch/x86/mm/Makefile
> > index 5b9908f13dcfd..a678a38a40266 100644
> > --- a/arch/x86/mm/Makefile
> > +++ b/arch/x86/mm/Makefile
> > @@ -4,6 +4,10 @@ KCOV_INSTRUMENT_tlb.o                  := n
> >  KCOV_INSTRUMENT_mem_encrypt.o          := n
> >  KCOV_INSTRUMENT_mem_encrypt_amd.o      := n
> >  KCOV_INSTRUMENT_pgprot.o               := n
> > +# When a kexec kernel is loaded, calling load_segments() breaks all
> > +# subsequent KCOV instrumentation until new kernel takes control.
> > +# Keep KCOV instrumentation disabled to prevent kernel crashes.
> > +KCOV_INSTRUMENT_physaddr.o             := n
> >
> >  KASAN_SANITIZE_mem_encrypt.o           := n
> >  KASAN_SANITIZE_mem_encrypt_amd.o       := n
> > --
> > 2.53.0.273.g2a3d683680-goog
> >

Re: [PATCH] x86/kexec: Disable KCOV instrumentation after load_segments()

[Borislav Petkov]

On Mon, Feb 16, 2026 at 06:37:16PM +0100, Aleksandr Nogikh wrote:
> Disabling instrumentation for the individual functions would be too
> fragile, so let's fix the bug by disabling KCOV instrumentation for
> the whole machine_kexec_64.c and physaddr.c.

Seems like a whack-a-mole thing to me. Why not make KEXEC depend on !KCOV?

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette

Re: [PATCH] x86/kexec: Disable KCOV instrumentation after load_segments()

[Aleksandr Nogikh]

On Sun, Mar 15, 2026 at 1:19 PM Borislav Petkov <bp@alien8.de> wrote:
>
> On Mon, Feb 16, 2026 at 06:37:16PM +0100, Aleksandr Nogikh wrote:
> > Disabling instrumentation for the individual functions would be too
> > fragile, so let's fix the bug by disabling KCOV instrumentation for
> > the whole machine_kexec_64.c and physaddr.c.
>
> Seems like a whack-a-mole thing to me. Why not make KEXEC depend on !KCOV?

Some more context:
The problem I am trying to solve is enabling crash dump collection in
syzkaller. For this, the tool loads a panic kernel before fuzzing and
then calls makedumpfile after the panic (which fails due to the bug I
mentioned in the patch). It requires both KEXEC and KCOV.

The most whack-a-mole solution was to disable instrumentation for
several functions called after load_segments(); this particular patch
is more generic, but yes, it can still be fragile. Another approach
would be to add more checks to
__sanitizer_cov_trace_pc()/check_kcov_mode(), but this would also be
somewhat undesirable as it would slow KCOV down even further.

-- 
Aleksandr




>
> --
> Regards/Gruss,
>     Boris.
>
> https://people.kernel.org/tglx/notes-about-netiquette

Re: [PATCH] x86/kexec: Disable KCOV instrumentation after load_segments()

[Borislav Petkov]

On Mon, Mar 16, 2026 at 11:07:15AM +0100, Aleksandr Nogikh wrote:
> Some more context:
> The problem I am trying to solve is enabling crash dump collection in
> syzkaller. For this, the tool loads a panic kernel before fuzzing and
> then calls makedumpfile after the panic (which fails due to the bug I
> mentioned in the patch). It requires both KEXEC and KCOV.
> 
> The most whack-a-mole solution was to disable instrumentation for
> several functions called after load_segments(); this particular patch
> is more generic, but yes, it can still be fragile. Another approach
> would be to add more checks to
> __sanitizer_cov_trace_pc()/check_kcov_mode(), but this would also be
> somewhat undesirable as it would slow KCOV down even further.

I guess...

So I'd like the *real* justification  - the context you just gave - for this
to be put somewhere over the code, I guess in both Makefiles so that it is
clear why we're doing this.

I guess those two compilation units will be excluded from KCOV fuzzing and if
someone wants to do coverage-guided fuzzing for them, then someone would have
to come up with a different solution, like, I dunno, putting only the
relevant, simple functions into a separate compilation unit which would be the
only one excluded from KCOV instrumentation or something more clever and less
whack-a-mole-y...

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette