[kvm-unit-tests PATCH v2 0/4] Support cross compiling with clang

[kvm-unit-tests PATCH v2 0/4] Support cross compiling with clang

[Andrew Jones]

Modify configure to allow --cc=clang and a cross-prefix to be specified
together (as well as --cflags). This allows compiling with clang, but
using cross binutils for everything else, including the linker. So far
tested on riscv 32- and 64-bit and aarch64 (with some hacks to the code
to get it to compile - which is why there's no gitlab-ci patch for aarch64
in this series). I suspect it should work for other architectures too.

v2:
 - fix building with clang and --config-efi by suppressing a warning
 - added riscv clang efi build to CI
 - picked up Thomas's tags

Andrew Jones (4):
  riscv: Drop mstrict-align
  Makefile: Prepare for clang EFI builds
  configure: Support cross compiling with clang
  riscv: gitlab-ci: Add clang build tests

 .gitlab-ci.yml | 43 +++++++++++++++++++++++++++++++++++++++++++
 Makefile       |  2 ++
 configure      | 11 ++++++++---
 riscv/Makefile |  2 +-
 4 files changed, 54 insertions(+), 4 deletions(-)

-- 
2.46.0

[kvm-unit-tests PATCH v2 0/4] Support cross compiling with clang

[Andrew Jones]

Modify configure to allow --cc=clang and a cross-prefix to be specified
together (as well as --cflags). This allows compiling with clang, but
using cross binutils for everything else, including the linker. So far
tested on riscv 32- and 64-bit and aarch64 (with some hacks to the code
to get it to compile - which is why there's no gitlab-ci patch for aarch64
in this series). I suspect it should work for other architectures too.

v2:
 - fix building with clang and --config-efi by suppressing a warning
 - added riscv clang efi build to CI
 - picked up Thomas's tags

Andrew Jones (4):
  riscv: Drop mstrict-align
  Makefile: Prepare for clang EFI builds
  configure: Support cross compiling with clang
  riscv: gitlab-ci: Add clang build tests

 .gitlab-ci.yml | 43 +++++++++++++++++++++++++++++++++++++++++++
 Makefile       |  2 ++
 configure      | 11 ++++++++---
 riscv/Makefile |  2 +-
 4 files changed, 54 insertions(+), 4 deletions(-)

-- 
2.46.0

[kvm-unit-tests PATCH v2 1/4] riscv: Drop mstrict-align

[Andrew Jones]

The spec says unaligned accesses are supported, so this isn't required
and clang doesn't support it. A platform might have slow unaligned
accesses, but kvm-unit-tests isn't about speed anyway.

Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
---
 riscv/Makefile | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/riscv/Makefile b/riscv/Makefile
index 179a373dbacf..2ee7c5bb5ad8 100644
--- a/riscv/Makefile
+++ b/riscv/Makefile
@@ -76,7 +76,7 @@ LDFLAGS += -melf32lriscv
 endif
 CFLAGS += -DCONFIG_RELOC
 CFLAGS += -mcmodel=medany
-CFLAGS += -mstrict-align
+#CFLAGS += -mstrict-align
 CFLAGS += -std=gnu99
 CFLAGS += -ffreestanding
 CFLAGS += -O2
-- 
2.46.0

[kvm-unit-tests PATCH v2 1/4] riscv: Drop mstrict-align

[Andrew Jones]

The spec says unaligned accesses are supported, so this isn't required
and clang doesn't support it. A platform might have slow unaligned
accesses, but kvm-unit-tests isn't about speed anyway.

Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
---
 riscv/Makefile | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/riscv/Makefile b/riscv/Makefile
index 179a373dbacf..2ee7c5bb5ad8 100644
--- a/riscv/Makefile
+++ b/riscv/Makefile
@@ -76,7 +76,7 @@ LDFLAGS += -melf32lriscv
 endif
 CFLAGS += -DCONFIG_RELOC
 CFLAGS += -mcmodel=medany
-CFLAGS += -mstrict-align
+#CFLAGS += -mstrict-align
 CFLAGS += -std=gnu99
 CFLAGS += -ffreestanding
 CFLAGS += -O2
-- 
2.46.0

[kvm-unit-tests PATCH v2 1/4] riscv: Drop mstrict-align

[Nicholas Piggin]

On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> The spec says unaligned accesses are supported, so this isn't required
> and clang doesn't support it. A platform might have slow unaligned
> accesses, but kvm-unit-tests isn't about speed anyway.
>
> Reviewed-by: Thomas Huth <thuth@redhat.com>
> Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> ---
>  riscv/Makefile | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/riscv/Makefile b/riscv/Makefile
> index 179a373dbacf..2ee7c5bb5ad8 100644
> --- a/riscv/Makefile
> +++ b/riscv/Makefile
> @@ -76,7 +76,7 @@ LDFLAGS += -melf32lriscv
>  endif
>  CFLAGS += -DCONFIG_RELOC
>  CFLAGS += -mcmodel=medany
> -CFLAGS += -mstrict-align
> +#CFLAGS += -mstrict-align

Just remove the line?

Or put a comment there instead to explain.

Thanks,
Nick

Re: [kvm-unit-tests PATCH v2 1/4] riscv: Drop mstrict-align

[Nicholas Piggin]

On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> The spec says unaligned accesses are supported, so this isn't required
> and clang doesn't support it. A platform might have slow unaligned
> accesses, but kvm-unit-tests isn't about speed anyway.
>
> Reviewed-by: Thomas Huth <thuth@redhat.com>
> Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> ---
>  riscv/Makefile | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/riscv/Makefile b/riscv/Makefile
> index 179a373dbacf..2ee7c5bb5ad8 100644
> --- a/riscv/Makefile
> +++ b/riscv/Makefile
> @@ -76,7 +76,7 @@ LDFLAGS += -melf32lriscv
>  endif
>  CFLAGS += -DCONFIG_RELOC
>  CFLAGS += -mcmodel=medany
> -CFLAGS += -mstrict-align
> +#CFLAGS += -mstrict-align

Just remove the line?

Or put a comment there instead to explain.

Thanks,
Nick

[kvm-unit-tests PATCH v2 1/4] riscv: Drop mstrict-align

[Andrew Jones]

On Wed, Sep 11, 2024 at 10:08:23AM GMT, Nicholas Piggin wrote:
> On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> > The spec says unaligned accesses are supported, so this isn't required
> > and clang doesn't support it. A platform might have slow unaligned
> > accesses, but kvm-unit-tests isn't about speed anyway.
> >
> > Reviewed-by: Thomas Huth <thuth@redhat.com>
> > Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> > ---
> >  riscv/Makefile | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/riscv/Makefile b/riscv/Makefile
> > index 179a373dbacf..2ee7c5bb5ad8 100644
> > --- a/riscv/Makefile
> > +++ b/riscv/Makefile
> > @@ -76,7 +76,7 @@ LDFLAGS += -melf32lriscv
> >  endif
> >  CFLAGS += -DCONFIG_RELOC
> >  CFLAGS += -mcmodel=medany
> > -CFLAGS += -mstrict-align
> > +#CFLAGS += -mstrict-align
> 
> Just remove the line?
> 
> Or put a comment there instead to explain.

I'll add a comment.

Thanks,
drew

Re: [kvm-unit-tests PATCH v2 1/4] riscv: Drop mstrict-align

[Andrew Jones]

On Wed, Sep 11, 2024 at 10:08:23AM GMT, Nicholas Piggin wrote:
> On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> > The spec says unaligned accesses are supported, so this isn't required
> > and clang doesn't support it. A platform might have slow unaligned
> > accesses, but kvm-unit-tests isn't about speed anyway.
> >
> > Reviewed-by: Thomas Huth <thuth@redhat.com>
> > Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> > ---
> >  riscv/Makefile | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/riscv/Makefile b/riscv/Makefile
> > index 179a373dbacf..2ee7c5bb5ad8 100644
> > --- a/riscv/Makefile
> > +++ b/riscv/Makefile
> > @@ -76,7 +76,7 @@ LDFLAGS += -melf32lriscv
> >  endif
> >  CFLAGS += -DCONFIG_RELOC
> >  CFLAGS += -mcmodel=medany
> > -CFLAGS += -mstrict-align
> > +#CFLAGS += -mstrict-align
> 
> Just remove the line?
> 
> Or put a comment there instead to explain.

I'll add a comment.

Thanks,
drew

[kvm-unit-tests PATCH v2 2/4] Makefile: Prepare for clang EFI builds

[Andrew Jones]

clang complains about GNU extensions such as variable sized types not
being at the end of structs unless -Wno-gnu is used. We may
eventually want -Wno-gnu, but for now let's just handle the warnings
as they come. Add -Wno-gnu-variable-sized-type-not-at-end to avoid
the warning issued for the initrd_dev_path struct.

Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
---
 Makefile | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/Makefile b/Makefile
index 3d51cb726120..7471f7285b78 100644
--- a/Makefile
+++ b/Makefile
@@ -50,6 +50,8 @@ EFI_CFLAGS += -fshort-wchar
 # EFI applications use PIC as they are loaded to dynamic addresses, not a fixed
 # starting address
 EFI_CFLAGS += -fPIC
+# Avoid error with the initrd_dev_path struct
+EFI_CFLAGS += -Wno-gnu-variable-sized-type-not-at-end
 # Create shared library
 EFI_LDFLAGS := -Bsymbolic -shared -nostdlib
 endif
-- 
2.46.0

[kvm-unit-tests PATCH v2 2/4] Makefile: Prepare for clang EFI builds

[Andrew Jones]

clang complains about GNU extensions such as variable sized types not
being at the end of structs unless -Wno-gnu is used. We may
eventually want -Wno-gnu, but for now let's just handle the warnings
as they come. Add -Wno-gnu-variable-sized-type-not-at-end to avoid
the warning issued for the initrd_dev_path struct.

Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
---
 Makefile | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/Makefile b/Makefile
index 3d51cb726120..7471f7285b78 100644
--- a/Makefile
+++ b/Makefile
@@ -50,6 +50,8 @@ EFI_CFLAGS += -fshort-wchar
 # EFI applications use PIC as they are loaded to dynamic addresses, not a fixed
 # starting address
 EFI_CFLAGS += -fPIC
+# Avoid error with the initrd_dev_path struct
+EFI_CFLAGS += -Wno-gnu-variable-sized-type-not-at-end
 # Create shared library
 EFI_LDFLAGS := -Bsymbolic -shared -nostdlib
 endif
-- 
2.46.0

[kvm-unit-tests PATCH v2 2/4] Makefile: Prepare for clang EFI builds

[Thomas Huth]

On 04/09/2024 12.50, Andrew Jones wrote:
> clang complains about GNU extensions such as variable sized types not
> being at the end of structs unless -Wno-gnu is used. We may
> eventually want -Wno-gnu, but for now let's just handle the warnings
> as they come. Add -Wno-gnu-variable-sized-type-not-at-end to avoid
> the warning issued for the initrd_dev_path struct.
> 
> Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> ---
>   Makefile | 2 ++
>   1 file changed, 2 insertions(+)
> 
> diff --git a/Makefile b/Makefile
> index 3d51cb726120..7471f7285b78 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -50,6 +50,8 @@ EFI_CFLAGS += -fshort-wchar
>   # EFI applications use PIC as they are loaded to dynamic addresses, not a fixed
>   # starting address
>   EFI_CFLAGS += -fPIC
> +# Avoid error with the initrd_dev_path struct
> +EFI_CFLAGS += -Wno-gnu-variable-sized-type-not-at-end
>   # Create shared library
>   EFI_LDFLAGS := -Bsymbolic -shared -nostdlib
>   endif

Reviewed-by: Thomas Huth <thuth@redhat.com>

Re: [kvm-unit-tests PATCH v2 2/4] Makefile: Prepare for clang EFI builds

[Thomas Huth]

On 04/09/2024 12.50, Andrew Jones wrote:
> clang complains about GNU extensions such as variable sized types not
> being at the end of structs unless -Wno-gnu is used. We may
> eventually want -Wno-gnu, but for now let's just handle the warnings
> as they come. Add -Wno-gnu-variable-sized-type-not-at-end to avoid
> the warning issued for the initrd_dev_path struct.
> 
> Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> ---
>   Makefile | 2 ++
>   1 file changed, 2 insertions(+)
> 
> diff --git a/Makefile b/Makefile
> index 3d51cb726120..7471f7285b78 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -50,6 +50,8 @@ EFI_CFLAGS += -fshort-wchar
>   # EFI applications use PIC as they are loaded to dynamic addresses, not a fixed
>   # starting address
>   EFI_CFLAGS += -fPIC
> +# Avoid error with the initrd_dev_path struct
> +EFI_CFLAGS += -Wno-gnu-variable-sized-type-not-at-end
>   # Create shared library
>   EFI_LDFLAGS := -Bsymbolic -shared -nostdlib
>   endif

Reviewed-by: Thomas Huth <thuth@redhat.com>

[kvm-unit-tests PATCH v2 2/4] Makefile: Prepare for clang EFI builds

[Nicholas Piggin]

On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> clang complains about GNU extensions such as variable sized types not
> being at the end of structs unless -Wno-gnu is used. We may
> eventually want -Wno-gnu, but for now let's just handle the warnings
> as they come. Add -Wno-gnu-variable-sized-type-not-at-end to avoid
> the warning issued for the initrd_dev_path struct.

You could also make a variant of struct efi_vendor_dev_path with no
vendordata just for initrd_dev_path?

It's taken from Linux or some efi upstream though so maybe it's annoying
to make such changes here. Okay in that case since it's limited to EFI.

Reviewed-by: Nicholas Piggin <npiggin@gmail.com>


>
> Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> ---
>  Makefile | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/Makefile b/Makefile
> index 3d51cb726120..7471f7285b78 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -50,6 +50,8 @@ EFI_CFLAGS += -fshort-wchar
>  # EFI applications use PIC as they are loaded to dynamic addresses, not a fixed
>  # starting address
>  EFI_CFLAGS += -fPIC
> +# Avoid error with the initrd_dev_path struct
> +EFI_CFLAGS += -Wno-gnu-variable-sized-type-not-at-end
>  # Create shared library
>  EFI_LDFLAGS := -Bsymbolic -shared -nostdlib
>  endif

Re: [kvm-unit-tests PATCH v2 2/4] Makefile: Prepare for clang EFI builds

[Nicholas Piggin]

On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> clang complains about GNU extensions such as variable sized types not
> being at the end of structs unless -Wno-gnu is used. We may
> eventually want -Wno-gnu, but for now let's just handle the warnings
> as they come. Add -Wno-gnu-variable-sized-type-not-at-end to avoid
> the warning issued for the initrd_dev_path struct.

You could also make a variant of struct efi_vendor_dev_path with no
vendordata just for initrd_dev_path?

It's taken from Linux or some efi upstream though so maybe it's annoying
to make such changes here. Okay in that case since it's limited to EFI.

Reviewed-by: Nicholas Piggin <npiggin@gmail.com>


>
> Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> ---
>  Makefile | 2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/Makefile b/Makefile
> index 3d51cb726120..7471f7285b78 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -50,6 +50,8 @@ EFI_CFLAGS += -fshort-wchar
>  # EFI applications use PIC as they are loaded to dynamic addresses, not a fixed
>  # starting address
>  EFI_CFLAGS += -fPIC
> +# Avoid error with the initrd_dev_path struct
> +EFI_CFLAGS += -Wno-gnu-variable-sized-type-not-at-end
>  # Create shared library
>  EFI_LDFLAGS := -Bsymbolic -shared -nostdlib
>  endif

[kvm-unit-tests PATCH v2 2/4] Makefile: Prepare for clang EFI builds

[Andrew Jones]

On Wed, Sep 11, 2024 at 10:21:39AM GMT, Nicholas Piggin wrote:
> On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> > clang complains about GNU extensions such as variable sized types not
> > being at the end of structs unless -Wno-gnu is used. We may
> > eventually want -Wno-gnu, but for now let's just handle the warnings
> > as they come. Add -Wno-gnu-variable-sized-type-not-at-end to avoid
> > the warning issued for the initrd_dev_path struct.
> 
> You could also make a variant of struct efi_vendor_dev_path with no
> vendordata just for initrd_dev_path?
> 
> It's taken from Linux or some efi upstream though so maybe it's annoying
> to make such changes here. Okay in that case since it's limited to EFI.

I should have mentioned in the commit message that efi_vendor_dev_path is
a direct import of the Linux code and that we'd rather keep it that way.
I can do that for v2.

> 
> Reviewed-by: Nicholas Piggin <npiggin@gmail.com>

Thanks,
drew

> 
> 
> >
> > Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> > ---
> >  Makefile | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/Makefile b/Makefile
> > index 3d51cb726120..7471f7285b78 100644
> > --- a/Makefile
> > +++ b/Makefile
> > @@ -50,6 +50,8 @@ EFI_CFLAGS += -fshort-wchar
> >  # EFI applications use PIC as they are loaded to dynamic addresses, not a fixed
> >  # starting address
> >  EFI_CFLAGS += -fPIC
> > +# Avoid error with the initrd_dev_path struct
> > +EFI_CFLAGS += -Wno-gnu-variable-sized-type-not-at-end
> >  # Create shared library
> >  EFI_LDFLAGS := -Bsymbolic -shared -nostdlib
> >  endif
> 
> 
> -- 
> kvm-riscv mailing list
> kvm-riscv at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/kvm-riscv

Re: [kvm-unit-tests PATCH v2 2/4] Makefile: Prepare for clang EFI builds

[Andrew Jones]

On Wed, Sep 11, 2024 at 10:21:39AM GMT, Nicholas Piggin wrote:
> On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> > clang complains about GNU extensions such as variable sized types not
> > being at the end of structs unless -Wno-gnu is used. We may
> > eventually want -Wno-gnu, but for now let's just handle the warnings
> > as they come. Add -Wno-gnu-variable-sized-type-not-at-end to avoid
> > the warning issued for the initrd_dev_path struct.
> 
> You could also make a variant of struct efi_vendor_dev_path with no
> vendordata just for initrd_dev_path?
> 
> It's taken from Linux or some efi upstream though so maybe it's annoying
> to make such changes here. Okay in that case since it's limited to EFI.

I should have mentioned in the commit message that efi_vendor_dev_path is
a direct import of the Linux code and that we'd rather keep it that way.
I can do that for v2.

> 
> Reviewed-by: Nicholas Piggin <npiggin@gmail.com>

Thanks,
drew

> 
> 
> >
> > Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> > ---
> >  Makefile | 2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/Makefile b/Makefile
> > index 3d51cb726120..7471f7285b78 100644
> > --- a/Makefile
> > +++ b/Makefile
> > @@ -50,6 +50,8 @@ EFI_CFLAGS += -fshort-wchar
> >  # EFI applications use PIC as they are loaded to dynamic addresses, not a fixed
> >  # starting address
> >  EFI_CFLAGS += -fPIC
> > +# Avoid error with the initrd_dev_path struct
> > +EFI_CFLAGS += -Wno-gnu-variable-sized-type-not-at-end
> >  # Create shared library
> >  EFI_LDFLAGS := -Bsymbolic -shared -nostdlib
> >  endif
> 
> 
> -- 
> kvm-riscv mailing list
> kvm-riscv@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/kvm-riscv

[kvm-unit-tests PATCH v2 3/4] configure: Support cross compiling with clang

[Andrew Jones]

When a user specifies the compiler with --cc assume it's already
fully named, even if the user also specifies a cross-prefix. This
allows clang to be selected for the compiler, which doesn't use
prefixes, but also still provide a cross prefix for binutils. If
a user needs a prefix on the compiler that they specify with --cc,
then they'll just have to specify it with the prefix prepended.

Also ensure user provided cflags are used when testing the compiler,
since the flags may drastically change behavior, such as the --target
flag for clang.

With these changes it's possible to cross compile for riscv with
clang after configuring with

 ./configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' \
             --cross-prefix=riscv64-linux-gnu-

Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
---
 configure | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/configure b/configure
index 27ae9cc89657..337af07374df 100755
--- a/configure
+++ b/configure
@@ -130,6 +130,7 @@ while [[ "$1" = -* ]]; do
 	    ;;
 	--cc)
 	    cc="$arg"
+	    cc_selected=yes
 	    ;;
 	--cflags)
 	    cflags="$arg"
@@ -200,6 +201,10 @@ while [[ "$1" = -* ]]; do
     esac
 done
 
+if [ -z "$cc_selected" ] && [ "$cross_prefix" ]; then
+    cc="$cross_prefix$cc"
+fi
+
 if [ -z "$efi" ] || [ "$efi" = "n" ]; then
     [ "$efi_direct" = "y" ] && efi_direct=
 fi
@@ -370,7 +375,7 @@ fi
 cat << EOF > lib-test.c
 __UINT32_TYPE__
 EOF
-u32_long=$("$cross_prefix$cc" -E lib-test.c | grep -v '^#' | grep -q long && echo yes)
+u32_long=$("$cc" $cflags -E lib-test.c | grep -v '^#' | grep -q long && echo yes)
 rm -f lib-test.c
 
 # check if slash can be used for division
@@ -379,7 +384,7 @@ if [ "$arch" = "i386" ] || [ "$arch" = "x86_64" ]; then
 foo:
     movl (8 / 2), %eax
 EOF
-  wa_divide=$("$cross_prefix$cc" -c lib-test.S >/dev/null 2>&1 || echo yes)
+  wa_divide=$("$cc" $cflags -c lib-test.S >/dev/null 2>&1 || echo yes)
   rm -f lib-test.{o,S}
 fi
 
@@ -442,7 +447,7 @@ ARCH=$arch
 ARCH_NAME=$arch_name
 ARCH_LIBDIR=$arch_libdir
 PROCESSOR=$processor
-CC=$cross_prefix$cc
+CC=$cc
 CFLAGS=$cflags
 LD=$cross_prefix$ld
 OBJCOPY=$cross_prefix$objcopy
-- 
2.46.0

[kvm-unit-tests PATCH v2 3/4] configure: Support cross compiling with clang

[Andrew Jones]

When a user specifies the compiler with --cc assume it's already
fully named, even if the user also specifies a cross-prefix. This
allows clang to be selected for the compiler, which doesn't use
prefixes, but also still provide a cross prefix for binutils. If
a user needs a prefix on the compiler that they specify with --cc,
then they'll just have to specify it with the prefix prepended.

Also ensure user provided cflags are used when testing the compiler,
since the flags may drastically change behavior, such as the --target
flag for clang.

With these changes it's possible to cross compile for riscv with
clang after configuring with

 ./configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' \
             --cross-prefix=riscv64-linux-gnu-

Reviewed-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
---
 configure | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/configure b/configure
index 27ae9cc89657..337af07374df 100755
--- a/configure
+++ b/configure
@@ -130,6 +130,7 @@ while [[ "$1" = -* ]]; do
 	    ;;
 	--cc)
 	    cc="$arg"
+	    cc_selected=yes
 	    ;;
 	--cflags)
 	    cflags="$arg"
@@ -200,6 +201,10 @@ while [[ "$1" = -* ]]; do
     esac
 done
 
+if [ -z "$cc_selected" ] && [ "$cross_prefix" ]; then
+    cc="$cross_prefix$cc"
+fi
+
 if [ -z "$efi" ] || [ "$efi" = "n" ]; then
     [ "$efi_direct" = "y" ] && efi_direct=
 fi
@@ -370,7 +375,7 @@ fi
 cat << EOF > lib-test.c
 __UINT32_TYPE__
 EOF
-u32_long=$("$cross_prefix$cc" -E lib-test.c | grep -v '^#' | grep -q long && echo yes)
+u32_long=$("$cc" $cflags -E lib-test.c | grep -v '^#' | grep -q long && echo yes)
 rm -f lib-test.c
 
 # check if slash can be used for division
@@ -379,7 +384,7 @@ if [ "$arch" = "i386" ] || [ "$arch" = "x86_64" ]; then
 foo:
     movl (8 / 2), %eax
 EOF
-  wa_divide=$("$cross_prefix$cc" -c lib-test.S >/dev/null 2>&1 || echo yes)
+  wa_divide=$("$cc" $cflags -c lib-test.S >/dev/null 2>&1 || echo yes)
   rm -f lib-test.{o,S}
 fi
 
@@ -442,7 +447,7 @@ ARCH=$arch
 ARCH_NAME=$arch_name
 ARCH_LIBDIR=$arch_libdir
 PROCESSOR=$processor
-CC=$cross_prefix$cc
+CC=$cc
 CFLAGS=$cflags
 LD=$cross_prefix$ld
 OBJCOPY=$cross_prefix$objcopy
-- 
2.46.0

[kvm-unit-tests PATCH v2 3/4] configure: Support cross compiling with clang

[Nicholas Piggin]

On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> When a user specifies the compiler with --cc assume it's already
> fully named, even if the user also specifies a cross-prefix. This
> allows clang to be selected for the compiler, which doesn't use
> prefixes, but also still provide a cross prefix for binutils. If
> a user needs a prefix on the compiler that they specify with --cc,
> then they'll just have to specify it with the prefix prepended.

Makes sense.

> Also ensure user provided cflags are used when testing the compiler,
> since the flags may drastically change behavior, such as the --target
> flag for clang.

Could be a separate patch but no big deal.

>
> With these changes it's possible to cross compile for riscv with
> clang after configuring with
>
>  ./configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' \
>              --cross-prefix=riscv64-linux-gnu-

Nice. Perhaps add a recipe to README?

Reviewed-by: Nicholas Piggin <npiggin@gmail.com>

Thanks,
Nick

>
> Reviewed-by: Thomas Huth <thuth@redhat.com>
> Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> ---
>  configure | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)
>
> diff --git a/configure b/configure
> index 27ae9cc89657..337af07374df 100755
> --- a/configure
> +++ b/configure
> @@ -130,6 +130,7 @@ while [[ "$1" = -* ]]; do
>  	    ;;
>  	--cc)
>  	    cc="$arg"
> +	    cc_selected=yes
>  	    ;;
>  	--cflags)
>  	    cflags="$arg"
> @@ -200,6 +201,10 @@ while [[ "$1" = -* ]]; do
>      esac
>  done
>  
> +if [ -z "$cc_selected" ] && [ "$cross_prefix" ]; then
> +    cc="$cross_prefix$cc"
> +fi
> +
>  if [ -z "$efi" ] || [ "$efi" = "n" ]; then
>      [ "$efi_direct" = "y" ] && efi_direct=
>  fi
> @@ -370,7 +375,7 @@ fi
>  cat << EOF > lib-test.c
>  __UINT32_TYPE__
>  EOF
> -u32_long=$("$cross_prefix$cc" -E lib-test.c | grep -v '^#' | grep -q long && echo yes)
> +u32_long=$("$cc" $cflags -E lib-test.c | grep -v '^#' | grep -q long && echo yes)
>  rm -f lib-test.c
>  
>  # check if slash can be used for division
> @@ -379,7 +384,7 @@ if [ "$arch" = "i386" ] || [ "$arch" = "x86_64" ]; then
>  foo:
>      movl (8 / 2), %eax
>  EOF
> -  wa_divide=$("$cross_prefix$cc" -c lib-test.S >/dev/null 2>&1 || echo yes)
> +  wa_divide=$("$cc" $cflags -c lib-test.S >/dev/null 2>&1 || echo yes)
>    rm -f lib-test.{o,S}
>  fi
>  
> @@ -442,7 +447,7 @@ ARCH=$arch
>  ARCH_NAME=$arch_name
>  ARCH_LIBDIR=$arch_libdir
>  PROCESSOR=$processor
> -CC=$cross_prefix$cc
> +CC=$cc
>  CFLAGS=$cflags
>  LD=$cross_prefix$ld
>  OBJCOPY=$cross_prefix$objcopy

Re: [kvm-unit-tests PATCH v2 3/4] configure: Support cross compiling with clang

[Nicholas Piggin]

On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> When a user specifies the compiler with --cc assume it's already
> fully named, even if the user also specifies a cross-prefix. This
> allows clang to be selected for the compiler, which doesn't use
> prefixes, but also still provide a cross prefix for binutils. If
> a user needs a prefix on the compiler that they specify with --cc,
> then they'll just have to specify it with the prefix prepended.

Makes sense.

> Also ensure user provided cflags are used when testing the compiler,
> since the flags may drastically change behavior, such as the --target
> flag for clang.

Could be a separate patch but no big deal.

>
> With these changes it's possible to cross compile for riscv with
> clang after configuring with
>
>  ./configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' \
>              --cross-prefix=riscv64-linux-gnu-

Nice. Perhaps add a recipe to README?

Reviewed-by: Nicholas Piggin <npiggin@gmail.com>

Thanks,
Nick

>
> Reviewed-by: Thomas Huth <thuth@redhat.com>
> Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> ---
>  configure | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)
>
> diff --git a/configure b/configure
> index 27ae9cc89657..337af07374df 100755
> --- a/configure
> +++ b/configure
> @@ -130,6 +130,7 @@ while [[ "$1" = -* ]]; do
>  	    ;;
>  	--cc)
>  	    cc="$arg"
> +	    cc_selected=yes
>  	    ;;
>  	--cflags)
>  	    cflags="$arg"
> @@ -200,6 +201,10 @@ while [[ "$1" = -* ]]; do
>      esac
>  done
>  
> +if [ -z "$cc_selected" ] && [ "$cross_prefix" ]; then
> +    cc="$cross_prefix$cc"
> +fi
> +
>  if [ -z "$efi" ] || [ "$efi" = "n" ]; then
>      [ "$efi_direct" = "y" ] && efi_direct=
>  fi
> @@ -370,7 +375,7 @@ fi
>  cat << EOF > lib-test.c
>  __UINT32_TYPE__
>  EOF
> -u32_long=$("$cross_prefix$cc" -E lib-test.c | grep -v '^#' | grep -q long && echo yes)
> +u32_long=$("$cc" $cflags -E lib-test.c | grep -v '^#' | grep -q long && echo yes)
>  rm -f lib-test.c
>  
>  # check if slash can be used for division
> @@ -379,7 +384,7 @@ if [ "$arch" = "i386" ] || [ "$arch" = "x86_64" ]; then
>  foo:
>      movl (8 / 2), %eax
>  EOF
> -  wa_divide=$("$cross_prefix$cc" -c lib-test.S >/dev/null 2>&1 || echo yes)
> +  wa_divide=$("$cc" $cflags -c lib-test.S >/dev/null 2>&1 || echo yes)
>    rm -f lib-test.{o,S}
>  fi
>  
> @@ -442,7 +447,7 @@ ARCH=$arch
>  ARCH_NAME=$arch_name
>  ARCH_LIBDIR=$arch_libdir
>  PROCESSOR=$processor
> -CC=$cross_prefix$cc
> +CC=$cc
>  CFLAGS=$cflags
>  LD=$cross_prefix$ld
>  OBJCOPY=$cross_prefix$objcopy

[kvm-unit-tests PATCH v2 3/4] configure: Support cross compiling with clang

[Andrew Jones]

On Wed, Sep 11, 2024 at 10:24:34AM GMT, Nicholas Piggin wrote:
> On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> > When a user specifies the compiler with --cc assume it's already
> > fully named, even if the user also specifies a cross-prefix. This
> > allows clang to be selected for the compiler, which doesn't use
> > prefixes, but also still provide a cross prefix for binutils. If
> > a user needs a prefix on the compiler that they specify with --cc,
> > then they'll just have to specify it with the prefix prepended.
> 
> Makes sense.
> 
> > Also ensure user provided cflags are used when testing the compiler,
> > since the flags may drastically change behavior, such as the --target
> > flag for clang.
> 
> Could be a separate patch but no big deal.
> 
> >
> > With these changes it's possible to cross compile for riscv with
> > clang after configuring with
> >
> >  ./configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' \
> >              --cross-prefix=riscv64-linux-gnu-
> 
> Nice. Perhaps add a recipe to README?

Sure.

> 
> Reviewed-by: Nicholas Piggin <npiggin@gmail.com>

Thanks,
drew

Re: [kvm-unit-tests PATCH v2 3/4] configure: Support cross compiling with clang

[Andrew Jones]

On Wed, Sep 11, 2024 at 10:24:34AM GMT, Nicholas Piggin wrote:
> On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> > When a user specifies the compiler with --cc assume it's already
> > fully named, even if the user also specifies a cross-prefix. This
> > allows clang to be selected for the compiler, which doesn't use
> > prefixes, but also still provide a cross prefix for binutils. If
> > a user needs a prefix on the compiler that they specify with --cc,
> > then they'll just have to specify it with the prefix prepended.
> 
> Makes sense.
> 
> > Also ensure user provided cflags are used when testing the compiler,
> > since the flags may drastically change behavior, such as the --target
> > flag for clang.
> 
> Could be a separate patch but no big deal.
> 
> >
> > With these changes it's possible to cross compile for riscv with
> > clang after configuring with
> >
> >  ./configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' \
> >              --cross-prefix=riscv64-linux-gnu-
> 
> Nice. Perhaps add a recipe to README?

Sure.

> 
> Reviewed-by: Nicholas Piggin <npiggin@gmail.com>

Thanks,
drew

[kvm-unit-tests PATCH v2 4/4] riscv: gitlab-ci: Add clang build tests

[Andrew Jones]

Test building 32 and 64-bit with clang. Throw a test of in- and out-
of-tree building in too by swapping which is done to which (32-bit
vs. 64-bit) with respect to the gcc build tests.

Acked-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
---
 .gitlab-ci.yml | 43 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 43 insertions(+)

diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index 67a9a15733f1..b7ad99870e5a 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -176,6 +176,49 @@ build-riscv64-efi:
       | tee results.txt
  - grep -q PASS results.txt && ! grep -q FAIL results.txt
 
+build-riscv32-clang:
+ extends: .intree_template
+ script:
+ - dnf install -y qemu-system-riscv gcc-riscv64-linux-gnu clang
+ - ./configure --arch=riscv32 --cc=clang --cflags='--target=riscv32' --cross-prefix=riscv64-linux-gnu-
+ - make -j2
+ - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
+ - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
+      selftest
+      sbi
+      | tee results.txt
+ - grep -q PASS results.txt && ! grep -q FAIL results.txt
+
+build-riscv64-clang:
+ extends: .outoftree_template
+ script:
+ - dnf install -y qemu-system-riscv gcc-riscv64-linux-gnu clang
+ - mkdir build
+ - cd build
+ - ../configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' --cross-prefix=riscv64-linux-gnu-
+ - make -j2
+ - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
+ - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
+      selftest
+      sbi
+      | tee results.txt
+ - grep -q PASS results.txt && ! grep -q FAIL results.txt
+
+build-riscv64-clang-efi:
+ extends: .intree_template
+ script:
+ - dnf install -y edk2-riscv64 qemu-system-riscv gcc-riscv64-linux-gnu clang
+ - cp /usr/share/edk2/riscv/RISCV_VIRT_CODE.fd .
+ - truncate -s 32M RISCV_VIRT_CODE.fd
+ - ./configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' --cross-prefix=riscv64-linux-gnu- --enable-efi
+ - make -j2
+ - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
+ - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
+      selftest
+      sbi
+      | tee results.txt
+ - grep -q PASS results.txt && ! grep -q FAIL results.txt
+
 build-s390x:
  extends: .outoftree_template
  script:
-- 
2.46.0

[kvm-unit-tests PATCH v2 4/4] riscv: gitlab-ci: Add clang build tests

[Andrew Jones]

Test building 32 and 64-bit with clang. Throw a test of in- and out-
of-tree building in too by swapping which is done to which (32-bit
vs. 64-bit) with respect to the gcc build tests.

Acked-by: Thomas Huth <thuth@redhat.com>
Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
---
 .gitlab-ci.yml | 43 +++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 43 insertions(+)

diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index 67a9a15733f1..b7ad99870e5a 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -176,6 +176,49 @@ build-riscv64-efi:
       | tee results.txt
  - grep -q PASS results.txt && ! grep -q FAIL results.txt
 
+build-riscv32-clang:
+ extends: .intree_template
+ script:
+ - dnf install -y qemu-system-riscv gcc-riscv64-linux-gnu clang
+ - ./configure --arch=riscv32 --cc=clang --cflags='--target=riscv32' --cross-prefix=riscv64-linux-gnu-
+ - make -j2
+ - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
+ - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
+      selftest
+      sbi
+      | tee results.txt
+ - grep -q PASS results.txt && ! grep -q FAIL results.txt
+
+build-riscv64-clang:
+ extends: .outoftree_template
+ script:
+ - dnf install -y qemu-system-riscv gcc-riscv64-linux-gnu clang
+ - mkdir build
+ - cd build
+ - ../configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' --cross-prefix=riscv64-linux-gnu-
+ - make -j2
+ - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
+ - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
+      selftest
+      sbi
+      | tee results.txt
+ - grep -q PASS results.txt && ! grep -q FAIL results.txt
+
+build-riscv64-clang-efi:
+ extends: .intree_template
+ script:
+ - dnf install -y edk2-riscv64 qemu-system-riscv gcc-riscv64-linux-gnu clang
+ - cp /usr/share/edk2/riscv/RISCV_VIRT_CODE.fd .
+ - truncate -s 32M RISCV_VIRT_CODE.fd
+ - ./configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' --cross-prefix=riscv64-linux-gnu- --enable-efi
+ - make -j2
+ - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
+ - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
+      selftest
+      sbi
+      | tee results.txt
+ - grep -q PASS results.txt && ! grep -q FAIL results.txt
+
 build-s390x:
  extends: .outoftree_template
  script:
-- 
2.46.0

[kvm-unit-tests PATCH v2 4/4] riscv: gitlab-ci: Add clang build tests

[Nicholas Piggin]

On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> Test building 32 and 64-bit with clang. Throw a test of in- and out-
> of-tree building in too by swapping which is done to which (32-bit
> vs. 64-bit) with respect to the gcc build tests.
>
Acked-by: Nicholas Piggin <npiggin@gmail.com>

> Acked-by: Thomas Huth <thuth@redhat.com>
> Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> ---
>  .gitlab-ci.yml | 43 +++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 43 insertions(+)
>
> diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
> index 67a9a15733f1..b7ad99870e5a 100644
> --- a/.gitlab-ci.yml
> +++ b/.gitlab-ci.yml
> @@ -176,6 +176,49 @@ build-riscv64-efi:
>        | tee results.txt
>   - grep -q PASS results.txt && ! grep -q FAIL results.txt
>  
> +build-riscv32-clang:
> + extends: .intree_template
> + script:
> + - dnf install -y qemu-system-riscv gcc-riscv64-linux-gnu clang
> + - ./configure --arch=riscv32 --cc=clang --cflags='--target=riscv32' --cross-prefix=riscv64-linux-gnu-
> + - make -j2
> + - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
> + - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
> +      selftest
> +      sbi
> +      | tee results.txt
> + - grep -q PASS results.txt && ! grep -q FAIL results.txt
> +
> +build-riscv64-clang:
> + extends: .outoftree_template
> + script:
> + - dnf install -y qemu-system-riscv gcc-riscv64-linux-gnu clang
> + - mkdir build
> + - cd build
> + - ../configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' --cross-prefix=riscv64-linux-gnu-
> + - make -j2
> + - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
> + - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
> +      selftest
> +      sbi
> +      | tee results.txt
> + - grep -q PASS results.txt && ! grep -q FAIL results.txt
> +
> +build-riscv64-clang-efi:
> + extends: .intree_template
> + script:
> + - dnf install -y edk2-riscv64 qemu-system-riscv gcc-riscv64-linux-gnu clang
> + - cp /usr/share/edk2/riscv/RISCV_VIRT_CODE.fd .
> + - truncate -s 32M RISCV_VIRT_CODE.fd
> + - ./configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' --cross-prefix=riscv64-linux-gnu- --enable-efi
> + - make -j2
> + - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
> + - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
> +      selftest
> +      sbi
> +      | tee results.txt
> + - grep -q PASS results.txt && ! grep -q FAIL results.txt
> +
>  build-s390x:
>   extends: .outoftree_template
>   script:

Re: [kvm-unit-tests PATCH v2 4/4] riscv: gitlab-ci: Add clang build tests

[Nicholas Piggin]

On Wed Sep 4, 2024 at 8:50 PM AEST, Andrew Jones wrote:
> Test building 32 and 64-bit with clang. Throw a test of in- and out-
> of-tree building in too by swapping which is done to which (32-bit
> vs. 64-bit) with respect to the gcc build tests.
>
Acked-by: Nicholas Piggin <npiggin@gmail.com>

> Acked-by: Thomas Huth <thuth@redhat.com>
> Signed-off-by: Andrew Jones <andrew.jones@linux.dev>
> ---
>  .gitlab-ci.yml | 43 +++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 43 insertions(+)
>
> diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
> index 67a9a15733f1..b7ad99870e5a 100644
> --- a/.gitlab-ci.yml
> +++ b/.gitlab-ci.yml
> @@ -176,6 +176,49 @@ build-riscv64-efi:
>        | tee results.txt
>   - grep -q PASS results.txt && ! grep -q FAIL results.txt
>  
> +build-riscv32-clang:
> + extends: .intree_template
> + script:
> + - dnf install -y qemu-system-riscv gcc-riscv64-linux-gnu clang
> + - ./configure --arch=riscv32 --cc=clang --cflags='--target=riscv32' --cross-prefix=riscv64-linux-gnu-
> + - make -j2
> + - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
> + - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
> +      selftest
> +      sbi
> +      | tee results.txt
> + - grep -q PASS results.txt && ! grep -q FAIL results.txt
> +
> +build-riscv64-clang:
> + extends: .outoftree_template
> + script:
> + - dnf install -y qemu-system-riscv gcc-riscv64-linux-gnu clang
> + - mkdir build
> + - cd build
> + - ../configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' --cross-prefix=riscv64-linux-gnu-
> + - make -j2
> + - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
> + - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
> +      selftest
> +      sbi
> +      | tee results.txt
> + - grep -q PASS results.txt && ! grep -q FAIL results.txt
> +
> +build-riscv64-clang-efi:
> + extends: .intree_template
> + script:
> + - dnf install -y edk2-riscv64 qemu-system-riscv gcc-riscv64-linux-gnu clang
> + - cp /usr/share/edk2/riscv/RISCV_VIRT_CODE.fd .
> + - truncate -s 32M RISCV_VIRT_CODE.fd
> + - ./configure --arch=riscv64 --cc=clang --cflags='--target=riscv64' --cross-prefix=riscv64-linux-gnu- --enable-efi
> + - make -j2
> + - printf "FOO=foo\nBAR=bar\nBAZ=baz\nMVENDORID=0\nMARCHID=0\nMIMPID=0\n" >test-env
> + - ACCEL=tcg KVM_UNIT_TESTS_ENV=test-env ./run_tests.sh
> +      selftest
> +      sbi
> +      | tee results.txt
> + - grep -q PASS results.txt && ! grep -q FAIL results.txt
> +
>  build-s390x:
>   extends: .outoftree_template
>   script: