[Buildroot] [PATCH] package/xvisor: unconditionally disable SSP

[Arnout Vandecappelle]

On 10/11/2019 19:45, Yann E. MORIN wrote:
> Arnout, All,
> 
> This change looks a lot like what was done for kvm-unit-tests in c0ffd16e4.
> What do you think of it?

 It's not entirely the same: kvm-unit-tests uses HOSTCC, which is broken on
Arch. Since we can't fix the host compiler, that's the best we could do.

 [I notice now that that commit didn't actually fix anything... I'll send a
patch for it.]

 Here, however, the target compiler does get used, so it's Buildroot's compiler
that is broken. So I'd prefer to find a solution that fixes the compiler :-)

> 
> Regards,
> Yann E. MORIN.
> 
> On 2019-11-09 17:47 +0100, Eric Le Bihan spake thusly:
>> Fixes:
>> - http://autobuild.buildroot.net/results/5de37d3da17ca5cb17c0113d942ce8c22c0074e0
>> - http://autobuild.buildroot.net/results/41b00d95ec4eb851abceaa7919aa1c533d4f16e8
>> - http://autobuild.buildroot.net/results/119e22e8b03991ab9196e428a131f5a8ae2a2b2f
>>
>> Signed-off-by: Eric Le Bihan <eric.le.bihan.dev at free.fr>
>> ---
>>  ...MAKEFILE-Unconditionally-disable-SSP.patch | 31 +++++++++++++++++++
>>  1 file changed, 31 insertions(+)
>>  create mode 100644 package/xvisor/0001-MAKEFILE-Unconditionally-disable-SSP.patch
>>
>> diff --git a/package/xvisor/0001-MAKEFILE-Unconditionally-disable-SSP.patch b/package/xvisor/0001-MAKEFILE-Unconditionally-disable-SSP.patch
>> new file mode 100644
>> index 0000000000..b69a745462
>> --- /dev/null
>> +++ b/package/xvisor/0001-MAKEFILE-Unconditionally-disable-SSP.patch
>> @@ -0,0 +1,31 @@
>> +From f04d1b73a9060e8f09cf1173f89daab73d6c0a18 Mon Sep 17 00:00:00 2001
>> +From: Eric Le Bihan <eric.le.bihan.dev at free.fr>
>> +Date: Sat, 9 Nov 2019 17:13:46 +0100
>> +Subject: [PATCH] [MAKEFILE] Unconditionally disable SSP
>> +
>> +Though -nostdlib is passed in $(cflags), -fno-stack-protector must also be
>> +passed to avoid linking errors related to undefined references to
>> +'__stack_chk_guard' and '__stack_chk_fail' if toolchain enforces
>> +-fstack-protector.
>> +
>> +Signed-off-by: Eric Le Bihan <eric.le.bihan.dev at free.fr>
>> +---
>> + Makefile | 2 +-
>> + 1 file changed, 1 insertion(+), 1 deletion(-)
>> +
>> +diff --git a/Makefile b/Makefile
>> +index d6bcf519..de8e1043 100644
>> +--- a/Makefile
>> ++++ b/Makefile
>> +@@ -137,7 +137,7 @@ cppflags+=$(cpu-cppflags)
>> + cppflags+=$(board-cppflags)
>> + cppflags+=$(libs-cppflags-y)
>> + cc=$(CROSS_COMPILE)gcc
>> +-cflags=-g -Wall -nostdlib --sysroot=$(drivers_dir)/include -fno-builtin -D__VMM__
>> ++cflags=-g -Wall -nostdlib --sysroot=$(drivers_dir)/include -fno-builtin -D__VMM__ -fno-stack-protector

 I think we should not add -fstack-protector in our toolchain-wrapper if
-nostdlib of -ffreestanding is passed on the command line. It probably doesn't
capture all cases (e.g. it's possible to build a simple file without
-ffreestanding and then link it with -nostdlib, but the object file will already
have stack protector in that case so it will still fail).

 On the other hand, maybe the number of packages that suffer from this is so
small that it's easier to fix packages. This patch does look pretty
upstreamable. And apparently it was merged [1] though I don't see it in the repo...

 Regards,
 Arnout

[1] https://github.com/xvisor/xvisor/pull/120

>> + cflags+=$(board-cflags) 
>> + cflags+=$(cpu-cflags) 
>> + cflags+=$(libs-cflags-y) 
>> +-- 
>> +2.21.0
>> +
>> -- 
>> 2.21.0
>>
>> _______________________________________________
>> buildroot mailing list
>> buildroot at busybox.net
>> http://lists.busybox.net/mailman/listinfo/buildroot
>