diff mbox

Re: [PATCH] gcc-plugins: disable under COMPILE_TEST

Message ID 20160613001244.b4b3c675d59e3ad3d8d656a4@gmail.com (mailing list archive)
State New, archived
Headers show

Commit Message

Emese Revfy June 12, 2016, 10:12 p.m. UTC
On Sat, 11 Jun 2016 12:29:26 -0400
Paul Gortmaker <paul.gortmaker@windriver.com> wrote:

> [[PATCH] gcc-plugins: disable under COMPILE_TEST] On 11/06/2016 (Sat 09:12) Kees Cook wrote:
> 
> > Since adding the gcc plugin development headers is required for the
> > gcc plugin support, we should ease into this new kernel build dependency
> > more slowly. For now, disable the gcc plugins under COMPILE_TEST so that
> > all*config builds will skip it.
> 
> Wouldn't it be better to test compile a one line program that tries to
> source the header(s) and then react accordingly?

The scripts/gcc-plugin.sh script does exactly that.
 
> Then at least you would get the test coverage from people who have the
> headers installed who are doing all[yes|mod]config.  This "for now"
> solution doesn't really have a path forward other than assuming all
> distros install the plugin headers sometime in the future.
> 
> Either way, this is an improvement over the current situation, so thanks
> for that.

If it is not too late I think this patch would be better:


When there is no gcc plugin support then don't compile the plugins
(but still print a warning). This allows building allyes/allmod configs
until the gcc plugin headers get installed.
    
Signed-off-by: Emese Revfy <re.emese@gmail.com>
---
 Makefile                     | 6 +++---
 scripts/Makefile.gcc-plugins | 8 ++++----
 2 files changed, 7 insertions(+), 7 deletions(-)


--
2.8.1

Comments

Kees Cook June 12, 2016, 10:25 p.m. UTC | #1
On Sun, Jun 12, 2016 at 3:12 PM, Emese Revfy <re.emese@gmail.com> wrote:
> On Sat, 11 Jun 2016 12:29:26 -0400
> Paul Gortmaker <paul.gortmaker@windriver.com> wrote:
>
>> [[PATCH] gcc-plugins: disable under COMPILE_TEST] On 11/06/2016 (Sat 09:12) Kees Cook wrote:
>>
>> > Since adding the gcc plugin development headers is required for the
>> > gcc plugin support, we should ease into this new kernel build dependency
>> > more slowly. For now, disable the gcc plugins under COMPILE_TEST so that
>> > all*config builds will skip it.
>>
>> Wouldn't it be better to test compile a one line program that tries to
>> source the header(s) and then react accordingly?
>
> The scripts/gcc-plugin.sh script does exactly that.
>
>> Then at least you would get the test coverage from people who have the
>> headers installed who are doing all[yes|mod]config.  This "for now"
>> solution doesn't really have a path forward other than assuming all
>> distros install the plugin headers sometime in the future.
>>
>> Either way, this is an improvement over the current situation, so thanks
>> for that.
>
> If it is not too late I think this patch would be better:

I don't like this because it means if someone specifically selects
some plugins in their .config, and the headers are missing, the kernel
will successfully compile. For many plugins, this results in a kernel
that lacks the requested security features, and that I really do not
want to have happening. I'm okay leaving these disabled for compile
tests for now. We can revisit this once more distros have plugins
enabled by default.

-Kees

>
>
> When there is no gcc plugin support then don't compile the plugins
> (but still print a warning). This allows building allyes/allmod configs
> until the gcc plugin headers get installed.
>
> Signed-off-by: Emese Revfy <re.emese@gmail.com>
> ---
>  Makefile                     | 6 +++---
>  scripts/Makefile.gcc-plugins | 8 ++++----
>  2 files changed, 7 insertions(+), 7 deletions(-)
>
> diff --git a/Makefile b/Makefile
> index a49c075..715210c 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -623,15 +623,15 @@ endif
>  # Tell gcc to never replace conditional load with a non-conditional one
>  KBUILD_CFLAGS  += $(call cc-option,--param=allow-store-data-races=0)
>
> +include scripts/Makefile.gcc-plugins
> +
>  PHONY += gcc-plugins
>  gcc-plugins: scripts_basic
> -ifdef CONFIG_GCC_PLUGINS
> +ifneq ($(GCC_PLUGINS_CFLAGS),)
>         $(Q)$(MAKE) $(build)=scripts/gcc-plugins
>  endif
>         @:
>
> -include scripts/Makefile.gcc-plugins
> -
>  ifdef CONFIG_READABLE_ASM
>  # Disable optimizations that make assembler listings hard to read.
>  # reorder blocks reorders the control in the function
> diff --git a/scripts/Makefile.gcc-plugins b/scripts/Makefile.gcc-plugins
> index c7372cb..2f101ea 100644
> --- a/scripts/Makefile.gcc-plugins
> +++ b/scripts/Makefile.gcc-plugins
> @@ -21,6 +21,7 @@ ifdef CONFIG_GCC_PLUGINS
>          CFLAGS_KCOV := $(SANCOV_PLUGIN)
>        else
>          $(warning warning: cannot use CONFIG_KCOV: -fsanitize-coverage=trace-pc is not supported by compiler)
> +        CFLAGS_KCOV =
>        endif
>      endif
>    endif
> @@ -37,13 +38,12 @@ ifdef CONFIG_GCC_PLUGINS
>        else
>          $(warning warning: your gcc version does not support plugins, you should upgrade it to gcc 4.5 at least)
>        endif
> +      GCC_PLUGINS_CFLAGS =
>      endif
> -  else
> -    # SANCOV_PLUGIN can be only in CFLAGS_KCOV because avoid duplication.
> -    GCC_PLUGINS_CFLAGS := $(filter-out $(SANCOV_PLUGIN), $(GCC_PLUGINS_CFLAGS))
>    endif
>
> -  KBUILD_CFLAGS += $(GCC_PLUGINS_CFLAGS)
> +  # SANCOV_PLUGIN can be only in CFLAGS_KCOV because avoid duplication.
> +  KBUILD_CFLAGS += $(filter-out $(SANCOV_PLUGIN), $(GCC_PLUGINS_CFLAGS))
>    GCC_PLUGIN := $(gcc-plugin-y)
>
>  endif
>
> --
> 2.8.1
Emese Revfy June 13, 2016, 12:18 a.m. UTC | #2
On Sun, 12 Jun 2016 15:25:39 -0700
Kees Cook <keescook@chromium.org> wrote:

> I don't like this because it means if someone specifically selects
> some plugins in their .config, and the headers are missing, the kernel
> will successfully compile. For many plugins, this results in a kernel
> that lacks the requested security features, and that I really do not
> want to have happening. I'm okay leaving these disabled for compile
> tests for now. We can revisit this once more distros have plugins
> enabled by default.

You are right. Your patch is safer.
Austin S. Hemmelgarn June 13, 2016, 6:32 p.m. UTC | #3
On 2016-06-12 20:18, Emese Revfy wrote:
> On Sun, 12 Jun 2016 15:25:39 -0700
> Kees Cook <keescook@chromium.org> wrote:
>
>> I don't like this because it means if someone specifically selects
>> some plugins in their .config, and the headers are missing, the kernel
>> will successfully compile. For many plugins, this results in a kernel
>> that lacks the requested security features, and that I really do not
>> want to have happening. I'm okay leaving these disabled for compile
>> tests for now. We can revisit this once more distros have plugins
>> enabled by default.
>
> You are right. Your patch is safer.
>
Why not make it so that if COMPILE_TEST is enabled, the build warns if 
it can't find the headers, otherwise it fails?  That way, people who are 
doing all*config builds but don't have the headers will still get some 
build coverage, and the people who are enabling it as a security feature 
will still get build failures.
Kees Cook June 13, 2016, 8:11 p.m. UTC | #4
On Mon, Jun 13, 2016 at 11:32 AM, Austin S. Hemmelgarn
<ahferroin7@gmail.com> wrote:
> On 2016-06-12 20:18, Emese Revfy wrote:
>>
>> On Sun, 12 Jun 2016 15:25:39 -0700
>> Kees Cook <keescook@chromium.org> wrote:
>>
>>> I don't like this because it means if someone specifically selects
>>> some plugins in their .config, and the headers are missing, the kernel
>>> will successfully compile. For many plugins, this results in a kernel
>>> that lacks the requested security features, and that I really do not
>>> want to have happening. I'm okay leaving these disabled for compile
>>> tests for now. We can revisit this once more distros have plugins
>>> enabled by default.
>>
>>
>> You are right. Your patch is safer.
>>
> Why not make it so that if COMPILE_TEST is enabled, the build warns if it
> can't find the headers, otherwise it fails?  That way, people who are doing
> all*config builds but don't have the headers will still get some build
> coverage, and the people who are enabling it as a security feature will
> still get build failures.

I don't see a clear way to do this, but if you can find a way to make
that happen, please send a patch! :)

-Kees
Michael Ellerman June 14, 2016, 2:01 a.m. UTC | #5
On Mon, 2016-06-13 at 13:11 -0700, Kees Cook wrote:
> On Mon, Jun 13, 2016 at 11:32 AM, Austin S. Hemmelgarn
> <ahferroin7@gmail.com> wrote:
> > On 2016-06-12 20:18, Emese Revfy wrote:
> > > 
> > > On Sun, 12 Jun 2016 15:25:39 -0700
> > > Kees Cook <keescook@chromium.org> wrote:
> > > 
> > > > I don't like this because it means if someone specifically selects
> > > > some plugins in their .config, and the headers are missing, the kernel
> > > > will successfully compile. For many plugins, this results in a kernel
> > > > that lacks the requested security features, and that I really do not
> > > > want to have happening. I'm okay leaving these disabled for compile
> > > > tests for now. We can revisit this once more distros have plugins
> > > > enabled by default.
> > > 
> > > You are right. Your patch is safer.
> > > 
> > Why not make it so that if COMPILE_TEST is enabled, the build warns if it
> > can't find the headers, otherwise it fails?  That way, people who are doing
> > all*config builds but don't have the headers will still get some build
> > coverage, and the people who are enabling it as a security feature will
> > still get build failures.
> 
> I don't see a clear way to do this, but if you can find a way to make
> that happen, please send a patch! :)

Another option is to make the top-level option negative, that way when it's
enabled by allmod/yes the plugins are turned off.

So eg. you would have:

config DISABLE_GCC_PLUGINS
	bool "Disable building GCC plugins"
	default y
	...
	
This makes all the problems with allmod/yes go away, and means you always honor
the users intent - when DISABLE_GCC_PLUGINS=n you can fail the build if you
can't build the plugins.

The downside is the logic's a bit awkward, ie. to enable the plugins you have to
disable the option which disables them.

cheers
diff mbox

Patch

diff --git a/Makefile b/Makefile
index a49c075..715210c 100644
--- a/Makefile
+++ b/Makefile
@@ -623,15 +623,15 @@  endif
 # Tell gcc to never replace conditional load with a non-conditional one
 KBUILD_CFLAGS	+= $(call cc-option,--param=allow-store-data-races=0)
 
+include scripts/Makefile.gcc-plugins
+
 PHONY += gcc-plugins
 gcc-plugins: scripts_basic
-ifdef CONFIG_GCC_PLUGINS
+ifneq ($(GCC_PLUGINS_CFLAGS),)
 	$(Q)$(MAKE) $(build)=scripts/gcc-plugins
 endif
 	@:
 
-include scripts/Makefile.gcc-plugins
-
 ifdef CONFIG_READABLE_ASM
 # Disable optimizations that make assembler listings hard to read.
 # reorder blocks reorders the control in the function
diff --git a/scripts/Makefile.gcc-plugins b/scripts/Makefile.gcc-plugins
index c7372cb..2f101ea 100644
--- a/scripts/Makefile.gcc-plugins
+++ b/scripts/Makefile.gcc-plugins
@@ -21,6 +21,7 @@  ifdef CONFIG_GCC_PLUGINS
         CFLAGS_KCOV := $(SANCOV_PLUGIN)
       else
         $(warning warning: cannot use CONFIG_KCOV: -fsanitize-coverage=trace-pc is not supported by compiler)
+        CFLAGS_KCOV =
       endif
     endif
   endif
@@ -37,13 +38,12 @@  ifdef CONFIG_GCC_PLUGINS
       else
         $(warning warning: your gcc version does not support plugins, you should upgrade it to gcc 4.5 at least)
       endif
+      GCC_PLUGINS_CFLAGS =
     endif
-  else
-    # SANCOV_PLUGIN can be only in CFLAGS_KCOV because avoid duplication.
-    GCC_PLUGINS_CFLAGS := $(filter-out $(SANCOV_PLUGIN), $(GCC_PLUGINS_CFLAGS))
   endif
 
-  KBUILD_CFLAGS += $(GCC_PLUGINS_CFLAGS)
+  # SANCOV_PLUGIN can be only in CFLAGS_KCOV because avoid duplication.
+  KBUILD_CFLAGS += $(filter-out $(SANCOV_PLUGIN), $(GCC_PLUGINS_CFLAGS))
   GCC_PLUGIN := $(gcc-plugin-y)
 
 endif