On x86_64, we use the host compiler instead of the target compiler to
build kvm-unit-tests, because it is built with -m32 and our target
compiler doesn't support that.
However, the compiler on Arch Linux is broken: it *always* builds with
-fstack-protector, even when -ffreestanding is passed. However, when
-fnostdlib is passed at link time (which is normally the case when
building with -ffreestanding), it is not linked with the stack-protector
library. This leads to a link time error:
/usr/bin/ld: x86/realmode.o: in function `print_serial_u32':
.../x86/realmode.c:104: undefined reference to `__stack_chk_fail'
Since the entire package is built with -ffreestanding, it doesn't
support stack-protector at all. Therefore, simply pass
-fno-stack-protector explicitly on x86_64 to work around the bug in Arch
Linux.
Commit c0ffd16e4 tried to do this, but got the condition wrong:
-fno-stack-protector was passed in all cases *except* for x86_64. This
commit fixes that, by inverting the condition and moving the
--cross-prefix part to the else branch.
Fixes:
http://autobuild.buildroot.net/results/ca9576721214ecdce5622f2b7ec4fd4fc3699ac0/
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Cc: Fabrice Fontaine <fontaine.fabrice@gmail.com>
Cc: Matthew Weber <matthew.weber@rockwellcollins.com>
Reviewed-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Tested-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
... so we can drop all config options about it and previous versions.
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Arnout Vandecappelle <arnout@mind.be>
Cc: Peter Korsgaard <peter@korsgaard.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Cc: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
On x86_64, we use the host compiler instead of the target compiler to
build kvm-unit-tests, because it is built with -m32 and our target
compiler doesn't support that.
However, the compiler on Arch Linux is broken: it *always* builds with
-fstack-protector, even when -ffreestanding is passed. However, when
-fnostdlib is passed at link time (which is normally the case when
building with -ffreestanding), it is not linked with the stack-protector
library. This leads to a link time error:
/usr/bin/ld: x86/realmode.o: in function `print_serial_u32':
.../x86/realmode.c:104: undefined reference to `__stack_chk_fail'
Since the entire package is built with -ffreestanding, it doesn't
support stack-protector at all. Therefore, simply pass
-fno-stack-protector explicitly on x86_64 to work around the bug in Arch
Linux.
Bug reported upstream: https://bugs.archlinux.org/task/64270
Fixes:
- http://autobuild.buildroot.org/results/e6f767755ffdb5ecc014eb5ad7519814f075a60e
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
Tested-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
The BR2_GCC_TARGET_* configuration variables are copied to
corresponding GCC_TARGET_* variables which may then be optionally
modified or overwritten by architecture specific makefiles.
All makefiles must use the new GCC_TARGET_* variables instead
of the BR2_GCC_TARGET_* versions.
Signed-off-by: Mark Corbin <mark.corbin@embecosm.com>
[Thomas: simplify include of arch/arch.mk]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
The build fails when the host binutils isn't at least 2.23
(2.22.x introduced RDSEED).
The host toolchain is used for x86_64 target builds where
we need to do a 32bit build. Most other buildroot builds
are using a much newer binutils unless it's a external
older toolchain.
Fixes:
http://autobuild.buildroot.net/results/c39/c3987a3cbd2960b0ff50f872636bdfd8d1a9c820/
Upstream:
https://marc.info/?l=kvm&m=151580743523259&w=2
Signed-off-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
- 0001-Makefile-remove-Werror-to-avoid-build-failures.patch is
adjusted to apply to the new version of kvm-unit-tests
- 0001-x86-hyperv_clock-be-explicit-about-mul-instruction-d.patch is
removed, as it has been merged upstream as of commit
45276b5860522921c77fb3ccc1458ff5f223c3c6.
Signed-off-by: Matthew Weber <matthew.weber@rockwellcollins.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Since binutils commit 1a9ccd70f9a7 ("Fix the linker so that it will not
silently generate ELF binaries with invalid program headers. Fix readelf
to report such invalid binaries.") the behaviour has changed and
binutils tries to create a program header segment for which there is no
space. Thanks to Nicholas Piggin and his patch to Linux:
https://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux.git/commit/?h=next&id=ff45000fcb56b5b0f
Cedric fixed this in kvm-unit-tests commit 95062c20790d ("powerpc: add
-n to LDFLAGS") so bump to that version.
Since the patch 0001-remove-werror.patch needs to be adjusted, we take
this opportunity to reformat is as a Git patch.
Fixes:
http://autobuild.buildroot.net/results/80825b492abf61313637b109d6a0944b38d5f739
Signed-off-by: Cyril Bur <cyrilbur@gmail.com>
[Thomas: add necessary hash change, refresh patch that was not
applying anymore.]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
We use the host compiler for x86-64 builds, so ensure it is only available
on x86-64 hosts.
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Since it supports each variants individually, it should also supports
the big.LITTLE configurations.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Cyril Bur <cyrilbur@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Move all architecture options to their own symbol, so that it is easier
to add more variants in the future.
The dependency on cross-gcc >= 4.5 is only valid for i386, as we use the
host gcc for x86_64. Adapt the dependency accordingly.
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Cyril Bur <cyrilbur@gmail.com>
[Thomas:
- add comment about host gcc >= 4.5 dependency on x86-64
- remove bogus "depends on ... if ..." construct.]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
kvm-unit-tests uses the following code on x86/x86-64:
static inline u64 scale_delta(u64 delta, u64 mul_frac)
{
u64 product, unused;
__asm__ (
"mul %3"
: "=d" (product), "=a" (unused) : "1" (delta), "rm" ((u64)mul_frac) );
return product;
}
The "mul" instruction does not have a suffix that indicates the width of
the data being multiplied. When the data is passed in a register, there
is no need to specify the width, but with some gcc versions, the data is
passed as a memory reference, and therefore the assembler does not know
the width of the data to be multiplied. It causes the following build
failure:
x86/hyperv_clock.c: Assembler messages:
x86/hyperv_clock.c:21: Error: no instruction mnemonic suffix given and no register operands; can't size instruction
Since the data being multiplied is 64 bit, we explicitly specify the
instruction as being "mulq".
Fixes:
http://autobuild.buildroot.net/results/a4a65d01f049db83a93de92660f228dd18532625/
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The check-package script when ran gives warnings on ordering issues
on all of these Config files. This patch cleans up all warnings
related to the ordering in the Config files for packages starting with
the letter k in the package directory.
The appropriate ordering is: type, default, depends on, select, help
See http://nightly.buildroot.org/#_config_files for more information.
Signed-off-by: Adam Duskett <Adamduskett@outlook.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
We want to use SPDX identifier for license string as much as possible.
SPDX short identifier for LGPLv2/LGPLv2+ is LGPL-2.0/LGPL-2.0+.
This change is done using following command.
find . -name "*.mk" | xargs sed -ri '/LICENSE( )?[\+:]?=/s/LGPLv2(\+)?/LGPL-2.0\1/g'
Signed-off-by: Rahul Bedarkar <rahulbedarkar89@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The library uses the hypervisor instruction HVC, which is only
available on ARMv7-A with Hypervisor extensions. This is limited to
Cortex-A{7,12,15,17}.
In addition the build unconditionally passes -marm so it also needs
ARM instructions to be available, but that is already implied by the
CPU selection.
Implicitly fixes
http://autobuild.buildroot.net/results/53d109fd9055fd20387bb857aced5f89cf3086fd
though it is still not clear why ld doesn't accept the -Ttext=...
option there.
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
[Thomas: A12 and A17 also have the virtualization extensions.]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The configure script uses this to set the -mcpu option. If nothing is
passed, it defaults to -mcpu=cortex-a15 for arm, and -mcpu=cortex-a57
for aarch64, which may conflict with the actual architecture.
The --processor option is only really used for arm and aarch64, but it
doesn't hurt to also pass it for the other arches.
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The kvm-unit-tests package defines a KVM_UNIT_TESTS_ARCH variable, but
never used it. Due to this, on PowerPC64, we were passing an incorrect
value, causing a build failure.
In addition, once the KVM_UNIT_TESTS_ARCH is used, it appeared that
its value for x86_64 was incorrect, due a typo: it was x86_84 instead
of x86_64.
Fixes:
http://autobuild.buildroot.net/results/16836c02f4d1ab5df6b2e21a4fcdbbf4a7542f22/
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
On x86 (both i386 and x86_64), the kvm-unit-tests code uses
__builtin_reachable, which appeared in gcc 4.5. This commit adds the
relevant dependencies to kvm-unit-tests:
- on i386, we need gcc 4.5 as the target compiler
- on x86-64, we need gcc 4.5 as the host compiler, because we use the
host compiler to build kvm-unit-tests for this architecture
Fixes:
http://autobuild.buildroot.net/results/21340a7e6494e23adb2fe8dc6253ab220c380fec/
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
This version bump includes several fixes for x86 and a significant fix
to powerpc transactional memory tests.
Signed-off-by: Cyril Bur <cyrilbur@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
kvm-unit-tests requires a multilib compiler for x86_64 as it compiles
32bit boot code.
This patch uses the BR2_HOSTARCH_NEEDS_IA32_COMPILER option to enforce
this and avoids using TARGET_CROSS for x86_64 targets and uses the host
32bit (and 64bit) capable compiler.
Signed-off-by: Cyril Bur <cyrilbur@gmail.com>
[Thomas: reword comment in .mk file.]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Most importantly this version includes a change to kvm-unit-tests which
means it will stop looking attempting to include host headers. This
change should fix build breakage on i386.
"ERROR: unsafe header/library path used in cross-compilation: '-I'
'/lib/modules/3.13.0-042stab111.12/build/include'"
The new version also includes more powerpc tests.
Signed-off-by: Cyril Bur <cyrilbur@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
The unit tests are tiny guest operating systems that generally execute
only tens of lines of C and assembler test code in order to obtain its
PASS/FAIL result. Unit tests provide KVM and virt hardware functional
testing by targeting the features through minimal implementations of
their use per the hardware specification. The simplicity of unit tests
make them easy to verify they are correct, easy to maintain, and easy
to use in timing measurements. Unit tests are also often used for
quick and dirty bug reproducers. The reproducers may then be kept as
regression tests. It's strongly encouraged that patches implementing
new KVM features are submitted with accompanying unit tests.
Signed-off-by: Cyril Bur <cyrilbur@gmail.com>
[Thomas:
- order architecture dependencies in Config.in alphabetically.
- rewrap Config.in help text, lines were too long
- add an empty line between the package description and the upstream
project URL in the Config.in help text
- don't make KVM_UNIT_TESTS_ARCH default to $(ARCH). This was not
correct for i386 for example. Instead, just handle the few
architectures that the package supports.
- remove useless double quotes in variable definitions.
- remove --prefix="$(TARGET_DIR)" from CONF_OPTS. It was installing
everything in /share/ and not /usr/share/, and setting the prefix to
TARGET_DIR at configure time is not good. Instead, pass DESTDIR at
installation time.]
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>