Re: [PATCHv2] kcov: reject open when kernel not instrumented

[Mark Rutland <mark.rutland@arm.com>]

On Thu, Jun 16, 2016 at 11:09:37AM +0100, James Morse wrote:
> Hi Mark,
> 
> On 15/06/16 18:04, Mark Rutland wrote:
> > If the toolchain does not support -fsanitize-coverage=trace-pc, we blat
> > this option from CFLAGS_KCOV, and build the kernel without
> > instrumentation, even if CONFIG_KCOV was selected. However, we still
> > build the rest of the kcov infrastructure, and expose a kcov file under
> > debugfs. This can be confusing, as the kernel will appear to support
> > kcov, yet will never manage to sample any trace PC values. While we do
> > note this fact at build time, this may be missed, and a user may not
> > have access to build logs.
> > 
> > This patch ensures that CC_HAVE_SANCOV_TRACE_PC is defined when the
> > toolchain supports -fsanitize-coverage=trace-pc, and is not defined
> > otherwise. When CC_HAVE_SANCOV_TRACE_PC is not defined, the kernel will
> > return -ENOTSUPP if userspace attempts to open the kcov debugfs file,
> > indicating that kcov functionality is unavailable.
> 
> > diff --git a/Makefile b/Makefile
> > index 0f70de6..3785a63 100644
> > --- a/Makefile
> > +++ b/Makefile
> > @@ -369,7 +369,7 @@ LDFLAGS_MODULE  =
> >  CFLAGS_KERNEL	=
> >  AFLAGS_KERNEL	=
> >  CFLAGS_GCOV	= -fprofile-arcs -ftest-coverage -fno-tree-loop-im -Wno-maybe-uninitialized
> > -CFLAGS_KCOV	= -fsanitize-coverage=trace-pc
> > +CFLAGS_KCOV	= -fsanitize-coverage=trace-pc -DCC_HAVE_SANCOV_TRACE_PC
> >  
> >  
> >  # Use USERINCLUDE when you must reference the UAPI directories only.
> > diff --git a/kernel/kcov.c b/kernel/kcov.c
> > index a02f2dd..0a0b164 100644
> > --- a/kernel/kcov.c
> > +++ b/kernel/kcov.c
> > @@ -3,6 +3,7 @@
> >  #define DISABLE_BRANCH_PROFILING
> >  #include <linux/compiler.h>
> >  #include <linux/types.h>
> > +#include <linux/errno.h>
> >  #include <linux/file.h>
> >  #include <linux/fs.h>
> >  #include <linux/mm.h>
> > @@ -160,6 +161,14 @@ static int kcov_open(struct inode *inode, struct file *filep)
> >  {
> >  	struct kcov *kcov;
> >  
> > +#ifndef CC_HAVE_SANCOV_TRACE_PC
> 
> I don't think this will work as kernel/kcov.c is listed in the Makefile as:
> > # Don't self-instrument.
> > KCOV_INSTRUMENT_kcov.o := n
> 
> 
> This will cause the build machinery in scripts/Makefile.lib to not give
> kernel/kcov.c the CFLAGS_KCOV contents:
> > ifeq ($(CONFIG_KCOV),y)
> > _c_flags += $(if $(patsubst n%,, \
> > 	$(KCOV_INSTRUMENT_$(basetarget).o)$(KCOV_INSTRUMENT)y), \
> > 	$(CFLAGS_KCOV))
> > endif
> 
> ... so kernel/kcov.c will never see anything in CFLAGS_KCOV ...

Good spot!

Evidently I was trying to be overly clever here; my bad.

> An alternative would be to add the flag to the compiler test that generates the
> 'not supported' warning, but it needs to go in another CFLAGS variable.
> Something like:
> 
> -------------------%<-------------------
> diff --git a/Makefile b/Makefile
> @@ -687,6 +687,8 @@ ifdef CONFIG_KCOV
>      $(warning Cannot use CONFIG_KCOV: \
>               -fsanitize-coverage=trace-pc is not supported by compiler)
>      CFLAGS_KCOV =
> +  else
> +    KBUILD_CFLAGS += -DCC_HAVE_SANCOV_TRACE_PC
>    endif
>  endif
> -------------------%<-------------------

I'll fold the above in for v3.

Thanks,
Mark.