This fixes build on SmartOS (Joyent).
Patch cherry-picked from pkgsrc by jperkin (Joyent).
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
---
tests/Makefile.include | 2 ++
1 file changed, 2 insertions(+)
diff --git a/tests/Makefile.include b/tests/Makefile.include
index f08b7418f0..0e5e6cb9b8 100644
--- a/tests/Makefile.include
+++ b/tests/Makefile.include
@@ -810,8 +810,10 @@ tests/migration/initrd-stress.img: tests/migration/stress$(EXESUF)
rmdir $(INITRD_WORK_DIR)
ifeq ($(CONFIG_POSIX),y)
+ifneq ($(CONFIG_SOLARIS),y)
LIBS += -lutil
endif
+endif
# QTest rules
--
2.14.1
On 3 September 2017 at 17:49, Kamil Rytarowski <n54@gmx.com> wrote:
> This fixes build on SmartOS (Joyent).
>
> Patch cherry-picked from pkgsrc by jperkin (Joyent).
>
> Signed-off-by: Kamil Rytarowski <n54@gmx.com>
> ---
> tests/Makefile.include | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/tests/Makefile.include b/tests/Makefile.include
> index f08b7418f0..0e5e6cb9b8 100644
> --- a/tests/Makefile.include
> +++ b/tests/Makefile.include
> @@ -810,8 +810,10 @@ tests/migration/initrd-stress.img: tests/migration/stress$(EXESUF)
> rmdir $(INITRD_WORK_DIR)
>
> ifeq ($(CONFIG_POSIX),y)
> +ifneq ($(CONFIG_SOLARIS),y)
> LIBS += -lutil
> endif
> +endif
I don't object to this patch in principle, but I don't think
this is the best way to fix the underlying problem.
My assumption is that the issue is that Solaris doesn't
have a -lutil (it's helpful to quote the linker or
compiler error message for this kind of patch to clarify
what the failure is that we're trying to fix).
It looks like we use libutil here for openpty, which
is used in the test-char test. But the condition we're
using in tests/Makefile.include isn't the same as the
one we used to decide whether to add -lutil to the
emulator binary link line, which is in configure:
if test "$darwin" != "yes" -a "$mingw32" != "yes" -a "$solaris" != yes -a \
"$aix" != "yes" -a "$haiku" != "yes" ; then
libs_softmmu="-lutil $libs_softmmu"
fi
So I think I'd prefer it if we have:
* a configure test for "is openpty() in libutil"
* use that to determine whether to add -lutil to
libs_softmmu and to the libs for tests, rather than
doing OS-specific tests
Q: does Solaris still not have an openpty() implementation?
Currently we have a local implementation in util/qemu-openpty.c,
but if that workaround is only needed in old Solaris we don't
support any more we could perhaps drop it.
thanks
-- PMM
On 04.09.2017 11:32, Peter Maydell wrote: > On 3 September 2017 at 17:49, Kamil Rytarowski <n54@gmx.com> wrote: >> This fixes build on SmartOS (Joyent). >> >> Patch cherry-picked from pkgsrc by jperkin (Joyent). >> >> Signed-off-by: Kamil Rytarowski <n54@gmx.com> >> --- >> tests/Makefile.include | 2 ++ >> 1 file changed, 2 insertions(+) >> >> diff --git a/tests/Makefile.include b/tests/Makefile.include >> index f08b7418f0..0e5e6cb9b8 100644 >> --- a/tests/Makefile.include >> +++ b/tests/Makefile.include >> @@ -810,8 +810,10 @@ tests/migration/initrd-stress.img: tests/migration/stress$(EXESUF) >> rmdir $(INITRD_WORK_DIR) >> >> ifeq ($(CONFIG_POSIX),y) >> +ifneq ($(CONFIG_SOLARIS),y) >> LIBS += -lutil >> endif >> +endif > > I don't object to this patch in principle, but I don't think > this is the best way to fix the underlying problem. > > My assumption is that the issue is that Solaris doesn't > have a -lutil (it's helpful to quote the linker or > compiler error message for this kind of patch to clarify > what the failure is that we're trying to fix). > There is no libutil on SmartOS. I cannot provide any information about the proprietary Solaris here and in next questions. > It looks like we use libutil here for openpty, which > is used in the test-char test. But the condition we're > using in tests/Makefile.include isn't the same as the > one we used to decide whether to add -lutil to the > emulator binary link line, which is in configure: > > if test "$darwin" != "yes" -a "$mingw32" != "yes" -a "$solaris" != yes -a \ > "$aix" != "yes" -a "$haiku" != "yes" ; then > libs_softmmu="-lutil $libs_softmmu" > fi > > So I think I'd prefer it if we have: > * a configure test for "is openpty() in libutil" > * use that to determine whether to add -lutil to > libs_softmmu and to the libs for tests, rather than > doing OS-specific tests > This looks cleaner. I will go for it. > Q: does Solaris still not have an openpty() implementation? > Currently we have a local implementation in util/qemu-openpty.c, > but if that workaround is only needed in old Solaris we don't > support any more we could perhaps drop it. > SmartOS does not have openpty(). I don't see it either in the mainstream Illumos-gate repository. According to pkgsrc, openpty() is in libutil (-lutil) on all BSDs and Darwin. > thanks > -- PMM >
On 5 September 2017 at 00:46, Kamil Rytarowski <n54@gmx.com> wrote: > On 04.09.2017 11:32, Peter Maydell wrote: >> My assumption is that the issue is that Solaris doesn't >> have a -lutil (it's helpful to quote the linker or >> compiler error message for this kind of patch to clarify >> what the failure is that we're trying to fix). >> > > There is no libutil on SmartOS. > > I cannot provide any information about the proprietary Solaris here and > in next questions. Yes; I was being a bit sloppy referring to all the forks of opensolaris as 'Solaris' there. Recent news suggests we don't need to consider support for the proprietary Solaris... >> It looks like we use libutil here for openpty, which >> is used in the test-char test. But the condition we're >> using in tests/Makefile.include isn't the same as the >> one we used to decide whether to add -lutil to the >> emulator binary link line, which is in configure: >> >> if test "$darwin" != "yes" -a "$mingw32" != "yes" -a "$solaris" != yes -a \ >> "$aix" != "yes" -a "$haiku" != "yes" ; then >> libs_softmmu="-lutil $libs_softmmu" >> fi >> >> So I think I'd prefer it if we have: >> * a configure test for "is openpty() in libutil" >> * use that to determine whether to add -lutil to >> libs_softmmu and to the libs for tests, rather than >> doing OS-specific tests >> > > This looks cleaner. I will go for it. Thanks. Remember that there are three cases to handle: * has openpty, don't need -lutil to get it (OSX is like this I think) * has openpty and you need -lutil (Linux) * has no openpty at all, use QEMU internal version (smartos) thanks -- PMM
openpty(3) might:
- exists in libc (OSX)
- exists in libutil (GNU, BSD)
- does not exist (SmartOS)
Add a function to discover this function in the ./configure script.
Add new config types: CONFIG_OPENPTY_LIBC and CONFIG_OPENPTY_LIBUTIL,
respectively defined when openpts(3) links with -lc or -lutil.
Replace the condition adding -lutil in tests (for openpty(3)) from
CONFIG_POSIX to CONFIG_OPENPTY_LIBUTIL.
Replace the fallback openpty(3) impelementation comment from Solaris
to SmartOS. Solaris is EOL'ed and it's confirmed that it does not work
on Oracle Solaris.
---
configure | 25 +++++++++++++++++++++++++
tests/Makefile.include | 2 +-
util/qemu-openpty.c | 4 ++--
3 files changed, 28 insertions(+), 3 deletions(-)
diff --git a/configure b/configure
index fd7e3a5e81..a614adcd29 100755
--- a/configure
+++ b/configure
@@ -3819,6 +3819,25 @@ EOF
fi
fi
+##########################################
+# openpty probe
+openpty_libc=no
+openpty_libutil=no
+cat > $TMPC << EOF
+extern int openpty(int *amaster, int *aslave, char *name, void *termp, void *winp);
+
+int main(void)
+{
+ int master_fd, slave_fd;
+ return openpty(&master_fd, &slave_fd, 0, 0, 0) == 0;
+}
+EOF
+if compile_prog "" "" ; then
+ openpty_libc=yes
+elif compile_prog "" "-lutil" ; then
+ openpty_libutil=yes
+fi
+
##########################################
# signalfd probe
signalfd="no"
@@ -5788,6 +5807,12 @@ fi
if test "$fdt" = "yes" ; then
echo "CONFIG_FDT=y" >> $config_host_mak
fi
+if test "$openpty_libc" = "yes" ; then
+ echo "CONFIG_OPENPTY_LIBC=y" >> $config_host_mak
+fi
+if test "$openpty_libutil" = "yes" ; then
+ echo "CONFIG_OPENPTY_LIBUTIL=y" >> $config_host_mak
+fi
if test "$signalfd" = "yes" ; then
echo "CONFIG_SIGNALFD=y" >> $config_host_mak
fi
diff --git a/tests/Makefile.include b/tests/Makefile.include
index fae5715e9c..e7e0bc2724 100644
--- a/tests/Makefile.include
+++ b/tests/Makefile.include
@@ -814,7 +814,7 @@ tests/migration/initrd-stress.img: tests/migration/stress$(EXESUF)
rm $(INITRD_WORK_DIR)/init
rmdir $(INITRD_WORK_DIR)
-ifeq ($(CONFIG_POSIX),y)
+ifeq ($(CONFIG_OPENPTY_LIBUTIL),y)
LIBS += -lutil
endif
diff --git a/util/qemu-openpty.c b/util/qemu-openpty.c
index 2e8b43bdf5..62c87e5563 100644
--- a/util/qemu-openpty.c
+++ b/util/qemu-openpty.c
@@ -51,8 +51,8 @@
# include <termios.h>
#endif
-#ifdef __sun__
-/* Once Solaris has openpty(), this is going to be removed. */
+/* The fallback implementation is needed at least on SmartOS. */
+#if !defined(CONFIG_OPENPTY_LIBC) && !defined(CONFIG_OPENPTY_LIBUTIL)
static int openpty(int *amaster, int *aslave, char *name,
struct termios *termp, struct winsize *winp)
{
--
2.14.1
On 11.09.2017 19:16, Kamil Rytarowski wrote:
> openpty(3) might:
> - exists in libc (OSX)
> - exists in libutil (GNU, BSD)
> - does not exist (SmartOS)
>
> Add a function to discover this function in the ./configure script.
> Add new config types: CONFIG_OPENPTY_LIBC and CONFIG_OPENPTY_LIBUTIL,
> respectively defined when openpts(3) links with -lc or -lutil.
>
> Replace the condition adding -lutil in tests (for openpty(3)) from
> CONFIG_POSIX to CONFIG_OPENPTY_LIBUTIL.
>
> Replace the fallback openpty(3) impelementation comment from Solaris
> to SmartOS. Solaris is EOL'ed and it's confirmed that it does not work
> on Oracle Solaris.
Signed-off-by: Kamil Rytarowski <n54@gmx.com>
> ---
> configure | 25 +++++++++++++++++++++++++
> tests/Makefile.include | 2 +-
> util/qemu-openpty.c | 4 ++--
> 3 files changed, 28 insertions(+), 3 deletions(-)
>
> diff --git a/configure b/configure
> index fd7e3a5e81..a614adcd29 100755
> --- a/configure
> +++ b/configure
> @@ -3819,6 +3819,25 @@ EOF
> fi
> fi
>
> +##########################################
> +# openpty probe
> +openpty_libc=no
> +openpty_libutil=no
> +cat > $TMPC << EOF
> +extern int openpty(int *amaster, int *aslave, char *name, void *termp, void *winp);
> +
> +int main(void)
> +{
> + int master_fd, slave_fd;
> + return openpty(&master_fd, &slave_fd, 0, 0, 0) == 0;
> +}
> +EOF
> +if compile_prog "" "" ; then
> + openpty_libc=yes
> +elif compile_prog "" "-lutil" ; then
> + openpty_libutil=yes
> +fi
> +
> ##########################################
> # signalfd probe
> signalfd="no"
> @@ -5788,6 +5807,12 @@ fi
> if test "$fdt" = "yes" ; then
> echo "CONFIG_FDT=y" >> $config_host_mak
> fi
> +if test "$openpty_libc" = "yes" ; then
> + echo "CONFIG_OPENPTY_LIBC=y" >> $config_host_mak
> +fi
> +if test "$openpty_libutil" = "yes" ; then
> + echo "CONFIG_OPENPTY_LIBUTIL=y" >> $config_host_mak
> +fi
> if test "$signalfd" = "yes" ; then
> echo "CONFIG_SIGNALFD=y" >> $config_host_mak
> fi
> diff --git a/tests/Makefile.include b/tests/Makefile.include
> index fae5715e9c..e7e0bc2724 100644
> --- a/tests/Makefile.include
> +++ b/tests/Makefile.include
> @@ -814,7 +814,7 @@ tests/migration/initrd-stress.img: tests/migration/stress$(EXESUF)
> rm $(INITRD_WORK_DIR)/init
> rmdir $(INITRD_WORK_DIR)
>
> -ifeq ($(CONFIG_POSIX),y)
> +ifeq ($(CONFIG_OPENPTY_LIBUTIL),y)
> LIBS += -lutil
> endif
>
> diff --git a/util/qemu-openpty.c b/util/qemu-openpty.c
> index 2e8b43bdf5..62c87e5563 100644
> --- a/util/qemu-openpty.c
> +++ b/util/qemu-openpty.c
> @@ -51,8 +51,8 @@
> # include <termios.h>
> #endif
>
> -#ifdef __sun__
> -/* Once Solaris has openpty(), this is going to be removed. */
> +/* The fallback implementation is needed at least on SmartOS. */
> +#if !defined(CONFIG_OPENPTY_LIBC) && !defined(CONFIG_OPENPTY_LIBUTIL)
> static int openpty(int *amaster, int *aslave, char *name,
> struct termios *termp, struct winsize *winp)
> {
>
On 11 September 2017 at 18:16, Kamil Rytarowski <n54@gmx.com> wrote:
> openpty(3) might:
> - exists in libc (OSX)
> - exists in libutil (GNU, BSD)
> - does not exist (SmartOS)
>
> Add a function to discover this function in the ./configure script.
> Add new config types: CONFIG_OPENPTY_LIBC and CONFIG_OPENPTY_LIBUTIL,
> respectively defined when openpts(3) links with -lc or -lutil.
>
> Replace the condition adding -lutil in tests (for openpty(3)) from
> CONFIG_POSIX to CONFIG_OPENPTY_LIBUTIL.
>
> Replace the fallback openpty(3) impelementation comment from Solaris
> to SmartOS. Solaris is EOL'ed and it's confirmed that it does not work
> on Oracle Solaris.
> ---
> configure | 25 +++++++++++++++++++++++++
> tests/Makefile.include | 2 +-
> util/qemu-openpty.c | 4 ++--
> 3 files changed, 28 insertions(+), 3 deletions(-)
>
> diff --git a/configure b/configure
> index fd7e3a5e81..a614adcd29 100755
> --- a/configure
> +++ b/configure
> @@ -3819,6 +3819,25 @@ EOF
> fi
> fi
>
> +##########################################
> +# openpty probe
> +openpty_libc=no
> +openpty_libutil=no
> +cat > $TMPC << EOF
> +extern int openpty(int *amaster, int *aslave, char *name, void *termp, void *winp);
I think the need to provide a prototype here rather than
using the system header to define it is revealing that we
also need a configure test for
* openpty() in <pty.h>
* openpty() in <libutil.h>
* openpty() in <util.h>
Something like this untested code ought to do:
cat > $TMPC << EOF
#if defined(CONFIG_OPENPTY_IN_PTY_H)
#include <pty.h>
#elif defined(CONFIG_OPENPTY_IN_LIBUTIL_H)
#include <libutil.h>
#elif defined(CONFIG_OPENPTY_IN_UTIL_H)
#include <util.h>
#endif
int main(void) { return openpty(0, 0, 0, 0, 0); }
EOF
# Different platforms put openpty() in different headers,
# and may or may not need us to link against -lutil
if compile_prog -DCONFIG_OPENPTY_IN_PTY_H ""; then
openpty_in_pty_h=yes
elif compile_prog -D_CONFIG_OPENPTY_IN_PTY_H -lutil; then
openpty_in_pty_h=yes
openpty_libutil=yes
elif compile_object -DCONFIG_OPENPTY_IN_LIBUTIL_H; then
openpty_in_libutil_h=yes
elif compile_prog -D_CONFIG_OPENPTY_IN_LIBUTIL_H -lutil; then
openpty_in_libutil_h=yes
openpty_libutil=yes
elif compile_object -DCONFIG_OPENPTY_IN_UTIL_H; then
openpty_in_util_h=yes
elif compile_prog -D_CONFIG_OPENPTY_IN_UTIL_H -lutil; then
openpty_in_util_h=yes
openpty_libutil=yes
fi
Then in qemu-openpty.c we can do
#include <termios.h>
#if defined(CONFIG_OPENPTY_IN_PTY_H)
#include <pty.h>
#elif defined(CONFIG_OPENPTY_IN_LIBUTIL_H)
#include <libutil.h>
#elif defined(CONFIG_OPENPTY_IN_UTIL_H)
#include <util.h>
#else
/* sunos fallback code */
#endif
> +if test "$openpty_libc" = "yes" ; then
> + echo "CONFIG_OPENPTY_LIBC=y" >> $config_host_mak
> +fi
If we use the CONFIG_OPENPTY_IN_*_H constants as above
we don't need CONFIG_OPENPTY_LIBC any more.
thanks
-- PMM
On 14.09.2017 15:52, Peter Maydell wrote:
> On 11 September 2017 at 18:16, Kamil Rytarowski <n54@gmx.com> wrote:
>> openpty(3) might:
>> - exists in libc (OSX)
>> - exists in libutil (GNU, BSD)
>> - does not exist (SmartOS)
>>
>> Add a function to discover this function in the ./configure script.
>> Add new config types: CONFIG_OPENPTY_LIBC and CONFIG_OPENPTY_LIBUTIL,
>> respectively defined when openpts(3) links with -lc or -lutil.
>>
>> Replace the condition adding -lutil in tests (for openpty(3)) from
>> CONFIG_POSIX to CONFIG_OPENPTY_LIBUTIL.
>>
>> Replace the fallback openpty(3) impelementation comment from Solaris
>> to SmartOS. Solaris is EOL'ed and it's confirmed that it does not work
>> on Oracle Solaris.
>> ---
>> configure | 25 +++++++++++++++++++++++++
>> tests/Makefile.include | 2 +-
>> util/qemu-openpty.c | 4 ++--
>> 3 files changed, 28 insertions(+), 3 deletions(-)
>>
>> diff --git a/configure b/configure
>> index fd7e3a5e81..a614adcd29 100755
>> --- a/configure
>> +++ b/configure
>> @@ -3819,6 +3819,25 @@ EOF
>> fi
>> fi
>>
>> +##########################################
>> +# openpty probe
>> +openpty_libc=no
>> +openpty_libutil=no
>> +cat > $TMPC << EOF
>> +extern int openpty(int *amaster, int *aslave, char *name, void *termp, void *winp);
>
> I think the need to provide a prototype here rather than
> using the system header to define it is revealing that we
> also need a configure test for
> * openpty() in <pty.h>
> * openpty() in <libutil.h>
> * openpty() in <util.h>
>
> Something like this untested code ought to do:
>
> cat > $TMPC << EOF
> #if defined(CONFIG_OPENPTY_IN_PTY_H)
> #include <pty.h>
> #elif defined(CONFIG_OPENPTY_IN_LIBUTIL_H)
> #include <libutil.h>
> #elif defined(CONFIG_OPENPTY_IN_UTIL_H)
> #include <util.h>
> #endif
> int main(void) { return openpty(0, 0, 0, 0, 0); }
> EOF
>
> # Different platforms put openpty() in different headers,
> # and may or may not need us to link against -lutil
> if compile_prog -DCONFIG_OPENPTY_IN_PTY_H ""; then
> openpty_in_pty_h=yes
> elif compile_prog -D_CONFIG_OPENPTY_IN_PTY_H -lutil; then
> openpty_in_pty_h=yes
> openpty_libutil=yes
> elif compile_object -DCONFIG_OPENPTY_IN_LIBUTIL_H; then
> openpty_in_libutil_h=yes
> elif compile_prog -D_CONFIG_OPENPTY_IN_LIBUTIL_H -lutil; then
> openpty_in_libutil_h=yes
> openpty_libutil=yes
> elif compile_object -DCONFIG_OPENPTY_IN_UTIL_H; then
> openpty_in_util_h=yes
> elif compile_prog -D_CONFIG_OPENPTY_IN_UTIL_H -lutil; then
> openpty_in_util_h=yes
> openpty_libutil=yes
> fi
>
> Then in qemu-openpty.c we can do
>
> #include <termios.h>
>
> #if defined(CONFIG_OPENPTY_IN_PTY_H)
> #include <pty.h>
> #elif defined(CONFIG_OPENPTY_IN_LIBUTIL_H)
> #include <libutil.h>
> #elif defined(CONFIG_OPENPTY_IN_UTIL_H)
> #include <util.h>
> #else
> /* sunos fallback code */
> #endif
>
>
>> +if test "$openpty_libc" = "yes" ; then
>> + echo "CONFIG_OPENPTY_LIBC=y" >> $config_host_mak
>> +fi
>
>
> If we use the CONFIG_OPENPTY_IN_*_H constants as above
> we don't need CONFIG_OPENPTY_LIBC any more.
>
> thanks
> -- PMM
>
Hello,
I've naively expected that once a build on a certain OS will be fixed,
it will be functional for longer period (at least few weeks). In one
month qemu has been broken on DragonFly and SmartOS so they are not just
broken in tests, but also in the basic build.
SmartOS has new compatibility problems with grep(1) invocations and tpm
emulator fatal build failure. Few old problems with conflicts with
system headers like conflicts with symbols CS and REG_SP.
There are certainly other small incompatibility problems, after the
breakage. Just a remind that SmartOS was not capable to execute tests
because of incompatible diff(1) invocations and unclear gmake(1)
misbehavior. And this was just the beginning of the problems.
DragonFlyBSD has problems with PAGE_SIZE, PAGE_MASK clash with headers
in s390x code. PAGE_MASK, PAGE_SHIFT and PAGE_SIZE in nand. No single
GTESTER invocation passes.
Because there is need to fix the SmartOS build in order to improve this
openpty(3) patch, I'm resigning from it. I also resign from helping
DragonFly port.
I will focus entirely on the bits that I registered as a maintainer.
I leave the decisions and steps about support of these platforms to
other volunteers and qemu core maintainers.
+ CC Antonio and Peter
On 28 October 2017 at 18:07, Kamil Rytarowski <n54@gmx.com> wrote: > I've naively expected that once a build on a certain OS will be fixed, > it will be functional for longer period (at least few weeks). In one > month qemu has been broken on DragonFly and SmartOS so they are not just > broken in tests, but also in the basic build. Yep; things do break if they're not tested, which is why I think we have to have a host system being in our regular build/test process in order to be able to call it supported. > Because there is need to fix the SmartOS build in order to improve this > openpty(3) patch, I'm resigning from it. I also resign from helping > DragonFly port. > > I will focus entirely on the bits that I registered as a maintainer. > > I leave the decisions and steps about support of these platforms to > other volunteers and qemu core maintainers. Thanks for your efforts with the BSD porting work; we appreciate it. thanks -- PMM
© 2016 - 2025 Red Hat, Inc.