On Thu, Aug 04, 2016 at 09:12:07AM +0300, Joonas Lahtinen wrote:
> On ke, 2016-08-03 at 20:38 +0100, Chris Wilson wrote:
> > In the next merge, we can build support for kcov at the individual file,
> > or driver level. This is useful to filter out the noise when doing
> > coverage test, i.e. we do get edges through code outside of i915.ko.
> >
> > Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
>
> Umm, is not KCOV enabled by selecting KCOV and then instrumentation is
> only disabled on select objects? "KCOV_INSTRUMENT_foo.o := n"?
commit a4691deabf284a601149a067525759939cc563b2
Author: Vegard Nossum <vegard.nossum@xxxxxxxxxx>
Date: Tue Aug 2 14:07:30 2016 -0700
kcov: allow more fine-grained coverage instrumentation
For more targeted fuzzing, it's better to disable kernel-wide
instrumentation and instead enable it on a per-subsystem basis. This
follows the pattern of UBSAN and allows you to compile in the kcov
driver without instrumenting the whole kernel.
To instrument a part of the kernel, you can use either
# for a single file in the current directory
KCOV_INSTRUMENT_filename.o := y
or
# for all the files in the current directory (excluding subdirectories)
KCOV_INSTRUMENT := y
or
# (same as above)
ccflags-y += $(CFLAGS_KCOV)
or
# for all the files in the current directory (including subdirectories)
subdir-ccflags-y += $(CFLAGS_KCOV)
The commit itself adds KCOV_INSTRUMENT_ALL config target to enable
kernel-wide coverage, with KCOV then just enabling the debug interface.
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
