Skip to contentSkip to conversation

configure / util: Auto-detect the availability of openpty()

PATCH In review Thomas Huth 5 participants 24 messages Jul 2, 2020 – Jul 4, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 configure / util: Auto-detect the availability of openpty()

Recent versions of Solaris (v11.4) now feature an openpty() function, too, causing a build failure since we ship our own implementation of openpty() for Solaris in util/qemu-openpty.c so far. Since there are now both variants available in the wild, with and without this function, let's introduce a proper HAVE_OPENPTY define for this to fix the build failure.

---
Note: Untested, since I do not have a VM with Solaris. Michele, Peter, I'd really appreciate a "Tested-by:" from you here. Thanks!

 configure           | 9 ++++++++-
 util/qemu-openpty.c | 4 +++-
 2 files changed, 11 insertions(+), 2 deletions(-)
configure +8 -1
index 4a22dcd563..0b5fd30fc4 100755
@@ -5134,10 +5134,14 @@ extern int openpty(int *am, int *as, char *name, void *termp, void *winp);
 int main(void) { return openpty(0, 0, 0, 0, 0); }
 EOF
 
-if ! compile_prog "" "" ; then
+have_openpty="no"
+if compile_prog "" "" ; then
+  have_openpty="yes"
+else
   if compile_prog "" "-lutil" ; then
     libs_softmmu="-lutil $libs_softmmu"
     libs_tools="-lutil $libs_tools"
+    have_openpty="yes"
   fi
 fi
 
@@ -7380,6 +7384,9 @@ fi
 if test "$have_broken_size_max" = "yes" ; then
     echo "HAVE_BROKEN_SIZE_MAX=y" >> $config_host_mak
 fi
+if test "$have_openpty" = "yes" ; then
+    echo "HAVE_OPENPTY=y" >> $config_host_mak
+fi
 
 # Work around a system header bug with some kernel/XFS header
 # versions where they both try to define 'struct fsxattr':
util/qemu-openpty.c +3 -1
index 2e8b43bdf5..14e5c8ae8f 100644
@@ -52,7 +52,8 @@
 #endif
 
 #ifdef __sun__
-/* Once Solaris has openpty(), this is going to be removed. */
+
+#if !defined(HAVE_OPENPTY)
 static int openpty(int *amaster, int *aslave, char *name,
                    struct termios *termp, struct winsize *winp)
 {
@@ -93,6 +94,7 @@ err:
         close(mfd);
         return -1;
 }
+#endif
 
 static void cfmakeraw (struct termios *termios_p)
 {
Signature · 1 line
2.18.1
#1 lore raw
Quoted 6 lines · Thomas Huth

Recent versions of Solaris (v11.4) now feature an openpty() function, too, causing a build failure since we ship our own implementation of openpty() for Solaris in util/qemu-openpty.c so far. Since there are now both variants available in the wild, with and without this function, let's introduce a proper HAVE_OPENPTY define for this to fix the build failure.

11.4 has been out since Aug 2018

The previous verison of solaris 11.3 was from Oct 2015, with EOL in Oct 2020 [1].

Solaris isn't an officially supported platform for QEMU, but if it was, then we'd probably consider it a long life distro, and thus consider 11.3 to be out of scope for QEMU by now.

IOW, instead of checking for openpty being missing, I think there's a decent argument to be made that we can just assume openpty exists, and delete the old solaris compat code entirely.

Regards,
Daniel

[1] https://en.wikipedia.org/wiki/Solaris_(operating_system)

Signature · 3 lines
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
#2 lore raw
Quoted 12 lines · Daniel P. Berrangé
Quoted 6 lines · Thomas Huth

Recent versions of Solaris (v11.4) now feature an openpty() function, too, causing a build failure since we ship our own implementation of openpty() for Solaris in util/qemu-openpty.c so far. Since there are now both variants available in the wild, with and without this function, let's introduce a proper HAVE_OPENPTY define for this to fix the build failure.

11.4 has been out since Aug 2018

The previous verison of solaris 11.3 was from Oct 2015, with EOL in Oct 2020 [1].

Quoted 7 lines

Solaris isn't an officially supported platform for QEMU, but if it was, then we'd probably consider it a long life distro, and thus consider 11.3 to be out of scope for QEMU by now.

IOW, instead of checking for openpty being missing, I think there's a decent argument to be made that we can just assume openpty exists, and delete the old solaris compat code entirely.

Of course there are still many people (where "many" is a relative term) running Solaris 10 for compatibility reasons and Solaris 10 does not have openpty. The only real reason to run 11.3 is because it still has support for the Sun Ray thin client and that was removed from 11.4 (which is a giant pain). So I'm stuck running Solaris 10 in an LDOM to run all my old code and have Sun Ray access, and 11.4 in the control domain because it's the latest thing.

- Michele

Quoted 6 lines

Regards,
Daniel

[1] https://en.wikipedia.org/wiki/Solaris_(operating_system)

#4 lore raw
In reply to Thomas Huth
Quoted 11 lines · Thomas Huth

Recent versions of Solaris (v11.4) now feature an openpty() function, too, causing a build failure since we ship our own implementation of openpty() for Solaris in util/qemu-openpty.c so far. Since there are now both variants available in the wild, with and without this function, let's introduce a proper HAVE_OPENPTY define for this to fix the build failure.

---
Note: Untested, since I do not have a VM with Solaris. Michele, Peter, I'd really appreciate a "Tested-by:" from you here. Thanks!

Sorry, I was off on another project (doing my taxes). Is this patch missing something? It looks like there should be more. The last line is

static void cfmakeraw (struct termios *termios_p) {

with no matching "}".

- Michele

Quoted 57 lines
  configure           | 9 ++++++++-
  util/qemu-openpty.c | 4 +++-
  2 files changed, 11 insertions(+), 2 deletions(-)
configure +8 -1
index 4a22dcd563..0b5fd30fc4 100755
@@ -5134,10 +5134,14 @@ extern int openpty(int *am, int *as, char *name, void *termp, void *winp);
  int main(void) { return openpty(0, 0, 0, 0, 0); }
  EOF
 
-if ! compile_prog "" "" ; then
+have_openpty="no"
+if compile_prog "" "" ; then
+  have_openpty="yes"
+else
    if compile_prog "" "-lutil" ; then
      libs_softmmu="-lutil $libs_softmmu"
      libs_tools="-lutil $libs_tools"
+    have_openpty="yes"
    fi
  fi
 
@@ -7380,6 +7384,9 @@ fi
  if test "$have_broken_size_max" = "yes" ; then
      echo "HAVE_BROKEN_SIZE_MAX=y">>  $config_host_mak
  fi
+if test "$have_openpty" = "yes" ; then
+    echo "HAVE_OPENPTY=y">>  $config_host_mak
+fi
 
  # Work around a system header bug with some kernel/XFS header
  # versions where they both try to define 'struct fsxattr':
util/qemu-openpty.c +3 -1
index 2e8b43bdf5..14e5c8ae8f 100644
@@ -52,7 +52,8 @@
  #endif
 
  #ifdef __sun__
-/* Once Solaris has openpty(), this is going to be removed. */
+
+#if !defined(HAVE_OPENPTY)
  static int openpty(int *amaster, int *aslave, char *name,
                     struct termios *termp, struct winsize *winp)
  {
@@ -93,6 +94,7 @@ err:
          close(mfd);
          return -1;
  }
+#endif
 
  static void cfmakeraw (struct termios *termios_p)
  {
#5 lore raw
Quoted 14 lines · Michele Denber
Quoted 11 lines · Thomas Huth

Recent versions of Solaris (v11.4) now feature an openpty() function, too, causing a build failure since we ship our own implementation of openpty() for Solaris in util/qemu-openpty.c so far. Since there are now both variants available in the wild, with and without this function, let's introduce a proper HAVE_OPENPTY define for this to fix the build failure.

---
Note: Untested, since I do not have a VM with Solaris. Michele, Peter, I'd really appreciate a "Tested-by:" from you here. Thanks!

Quoted 5 lines

Sorry, I was off on another project (doing my taxes).  Is this patch missing something?  It looks like there should be more.  The last line is

static void cfmakeraw (struct termios *termios_p) {

with no matching "}".

That's just the context, cut at three lines after the last change. Simply try to apply the patch with "patch -p1 -i ..." or "git am" in a separate git branch if you're using git. It should hopefully apply cleanly.

Thomas

#3 lore raw
In reply to Daniel P. Berrangé
Quoted 13 lines · Daniel P. Berrangé
Quoted 6 lines · Thomas Huth

Recent versions of Solaris (v11.4) now feature an openpty() function, too, causing a build failure since we ship our own implementation of openpty() for Solaris in util/qemu-openpty.c so far. Since there are now both variants available in the wild, with and without this function, let's introduce a proper HAVE_OPENPTY define for this to fix the build failure.

11.4 has been out since Aug 2018

The previous verison of solaris 11.3 was from Oct 2015, with EOL in Oct 2020 [1].

Quoted 7 lines

Solaris isn't an officially supported platform for QEMU, but if it was, then we'd probably consider it a long life distro, and thus consider 11.3 to be out of scope for QEMU by now.

IOW, instead of checking for openpty being missing, I think there's a decent argument to be made that we can just assume openpty exists, and delete the old solaris compat code entirely.

I'd normally agree, but according to Peter, illumos also does not have this function yet:

https://lists.gnu.org/archive/html/qemu-devel/2020-06/msg09276.html

... so I think we should keep it around for a little bit longer until at least the open source versions of Solaris have that function, too. (Maybe someone will still finally provide a VM based on one of the illumos distros for our regression tests...?)

Thomas

#6 lore raw
In reply to Thomas Huth
Quoted 2 lines · Thomas Huth

That's just the context, cut at three lines after the last change. Simply try to apply the patch with "patch -p1 -i ..." or "git am" in a separate git branch if you're using git. It should hopefully apply cleanly.

Looks like it partly worked:

root@hemlock:~/qemu-5.0.0# gpatch -p1 -i < configpatch.diff
gpatch: option requires an argument -- 'i'
gpatch: Try 'gpatch --help' for more information.
root@hemlock:~/qemu-5.0.0# gpatch -p1 < configpatch.diff
patching file configure
Hunk #1 succeeded at 5049 (offset -85 lines).
Hunk #2 succeeded at 7117 (offset -267 lines).
patching file util/qemu-openpty.c
Hunk #1 FAILED at 52.
Hunk #2 succeeded at 95 with fuzz 2 (offset 2 lines).
1 out of 2 hunks FAILED -- saving rejects to file util/qemu-openpty.c.rej
root@hemlock:~/qemu-5.0.0# cat util/qemu-openpty.c.rej
--- util/qemu-openpty.c
+++ util/qemu-openpty.c
@@ -52,7 +52,8 @@
 #endif

#ifdef __sun__
-/* Once Solaris has openpty(), this is going to be removed. */ +
+#if !defined(HAVE_OPENPTY)
static int openpty(int *amaster, int *aslave, char *name,
struct termios *termp, struct winsize *winp)
{
root@hemlock:~/qemu-5.0.0#

- Michele

#7 lore raw
Quoted 21 lines · Michele Denber
Quoted 2 lines · Thomas Huth

That's just the context, cut at three lines after the last change. Simply try to apply the patch with "patch -p1 -i ..." or "git am" in a separate git branch if you're using git. It should hopefully apply cleanly.

Looks like it partly worked:

root@hemlock:~/qemu-5.0.0# gpatch -p1 -i < configpatch.diff
gpatch: option requires an argument -- 'i'
gpatch: Try 'gpatch --help' for more information.
root@hemlock:~/qemu-5.0.0# gpatch -p1 < configpatch.diff
patching file configure
Hunk #1 succeeded at 5049 (offset -85 lines).
Hunk #2 succeeded at 7117 (offset -267 lines).
patching file util/qemu-openpty.c
Hunk #1 FAILED at 52.
Hunk #2 succeeded at 95 with fuzz 2 (offset 2 lines).
1 out of 2 hunks FAILED -- saving rejects to file util/qemu-openpty.c.rej
root@hemlock:~/qemu-5.0.0# cat util/qemu-openpty.c.rej

qemu-openpty.c line 58

Quoted 8 lines
qemu-openpty.c +2 -1
@@ -52,7 +52,8 @@
  #endif
 
  #ifdef __sun__
-/* Once Solaris has openpty(), this is going to be removed. */
+
+#if !defined(HAVE_OPENPTY)
  static int openpty(int *amaster, int *aslave, char *name,
                     struct termios *termp, struct winsize *winp)

Did you have a local modification to that file? If so, please revert it first before applying the patch (or add the "#if !defined(HAVE_OPENPTY)" now manually to see whether it works).

Thanks,
Thomas

#8 lore raw
Quoted 2 lines · Thomas Huth

Did you have a local modification to that file?

Oops yes, I had changed the line

static int openpty(int *amaster, int *aslave, char *name,
struct termios *termp, struct winsize *winp)

to

int openpty(int *amaster, int *aslave, char *name,
struct termios *termp, struct winsize *winp)

Quoted 1 line

If so, please revert it first before applying the patch (or add the "#if !defined(HAVE_OPENPTY)" now manually to see whether it works).

OK - looks like that did it:

root@hemlock:~/qemu-5.0.0# gpatch -p1 < configpatch.diff
patching file configure
Hunk #1 succeeded at 5049 (offset -85 lines).
Hunk #2 succeeded at 7117 (offset -267 lines).
patching file util/qemu-openpty.c
Hunk #2 succeeded at 94 with fuzz 2.
root@hemlock:~/qemu-5.0.0#

Then

root@hemlock:~/qemu-5.0.0# /opt/csw/bin/bash ./configure --cc=gcc  --extra-cflags="-m32"

succeeded with only one odd thing:

./configure: line 6326: pod2man: command not found

I fixed that with

root@hemlock:~/qemu-5.0.0# PATH=$PATH:/usr/perl5/5.26/bin

So I can confirm your patch works in Solaris 11.4.

gmake then ran successfully through qemu-openpty.c becfore stopping with:

...
CC crypto/trace.o
util/drm.c: In function \u2018qemu_drm_rendernode_open\u2019:
util/drm.c:41:16: error: \u2018struct dirent\u2019 has no member named \u2018d_type\u2019; did you mean \u2018d_name\u2019?

         if (e->d_type != DT_CHR) {
                ^~~~~~
                d_name
util/drm.c:41:26: error: \u2018DT_CHR\u2019 undeclared (first use in this function); did you mean \u2018TH_CWR\u2019?
         if (e->d_type != DT_CHR) {
                          ^~~~~~
                          TH_CWR

util/drm.c:41:26: note: each undeclared identifier is reported only once for each function it appears in
gmake: *** [/export/home/denber/qemu-5.0.0/rules.mak:69: util/drm.o] Error 1

PMM said about this:

Ah, the Haiku folks just ran into exactly this issue. Their fix should also be good for Solaris:
https://lists.gnu.org/archive/html/qemu-devel/2020-06/msg08800.html
(This whole file is for using a Linux-specific feature so there's no point even compiling it for other OSes.)

- Michele

#9 lore raw
In reply to Thomas Huth

Oops, I hit Send before I was done. I tried the Haiku patch:

util/Makefile.objs +1 -1
index cc5e37177a..faebc13fac 100644
@@ -39,7 +39,7 @@ util-obj-y += qsp.o
 util-obj-y += range.o
 util-obj-y += stats64.o
 util-obj-y += systemd.o
-util-obj-$(CONFIG_POSIX) += drm.o
+util-obj-$(CONFIG_LINUX) += drm.o
 util-obj-y += guest-random.o
 util-obj-$(CONFIG_GIO) += dbus.o
 dbus.o-cflags = $(GIO_CFLAGS)

but it didn't work:

root@hemlock:~/qemu-5.0.0# gpatch -p1 < Makefile.objs.patch.diff
patching file util/Makefile.objs
Hunk #1 FAILED at 39.
1 out of 1 hunk FAILED -- saving rejects to file util/Makefile.objs.rej
root@hemlock:~/qemu-5.0.0# cat util/Makefile.objs.rej
--- util/Makefile.objs
+++ util/Makefile.objs
@@ -39,7 +39,7 @@ util-obj-y += qsp.o
 util-obj-y += range.o
 util-obj-y += stats64.o
 util-obj-y += systemd.o
-util-obj-$(CONFIG_POSIX) += drm.o
+util-obj-$(CONFIG_LINUX) += drm.o
 util-obj-y += guest-random.o
 util-obj-$(CONFIG_GIO) += dbus.o
 dbus.o-cflags = $(GIO_CFLAGS)
root@hemlock:~/qemu-5.0.0#

- Michele

#10 lore raw
Quoted 17 lines · Michele Denber

Oops, I hit Send before I was done. I tried the Haiku patch:

util/Makefile.objs +1 -1
index cc5e37177a..faebc13fac 100644
@@ -39,7 +39,7 @@ util-obj-y += qsp.o
 util-obj-y += range.o
 util-obj-y += stats64.o
 util-obj-y += systemd.o
-util-obj-$(CONFIG_POSIX) += drm.o
+util-obj-$(CONFIG_LINUX) += drm.o
 util-obj-y += guest-random.o
 util-obj-$(CONFIG_GIO) += dbus.o
 dbus.o-cflags = $(GIO_CFLAGS)
Quoted 6 lines

but it didn't work:

root@hemlock:~/qemu-5.0.0# gpatch -p1 < Makefile.objs.patch.diff
patching file util/Makefile.objs
Hunk #1 FAILED at 39.
1 out of 1 hunk FAILED -- saving rejects to file util/Makefile.objs.rej

This is because you're trying to apply the patch to the 5.0.0 release, but it was made against the current head-of-git, and this file has changed a little between the two. Generally we do all development against head-of-git, so I would suggest that you try again with that.

As it happens, the current preferred fix for the drm.c issue is this one:
https://patchew.org/QEMU/[email protected]/[email protected]/

thanks
-- PMM

#11 lore raw

gmake in Solaris 11.4: _IOR missing

Quoted 10 lines · Peter Maydell

...
This is because you're trying to apply the patch to the 5.0.0 release, but it was made against the current head-of-git, and this file has changed a little between the two. Generally we do all development against head-of-git, so I would suggest that you try again with that.

As it happens, the current preferred fix for the drm.c issue is this one:
https://patchew.org/QEMU/[email protected]/[email protected]/

I see. I downloaded 5.0.0 mainly because I don't understand how git works. Anyway, I applied your new patch and it worked.

I tried gmake again and got further. This time it's stopping here:

root@hemlock:~/qemu-5.0.0# gmake -j16
...
CC      hw/usb/hcd-uhci.o
In file included from hw/tpm/tpm_emulator.c:37:0:
hw/tpm/tpm_ioctl.h:222:30: warning: implicit declaration of function '_IOR\'; did you mean '_IORW'? [-Wimplicit-function-declaration]
     PTM_GET_CAPABILITY     = _IOR('P', 0, ptm_cap),
                              ^~~~
                              _IORW
hw/tpm/tpm_ioctl.h:222:43: error: expected expression before \u2018ptm_cap\u2019
     PTM_GET_CAPABILITY     = _IOR('P', 0, ptm_cap),
                                           ^~~~~~~
hw/tpm/tpm_ioctl.h:223:30: warning: implicit declaration of function \u2018_IOWR\u2019; did you mean \u2018_IOWRT\u2019? [-Wimplicit-function-declaration]
     PTM_INIT               = _IOWR('P', 1, ptm_init),
                              ^~~~~
                              _IOWRT
hw/tpm/tpm_ioctl.h:223:44: error: expected expression before \u2018ptm_init\u2019
     PTM_INIT               = _IOWR('P', 1, ptm_init),
                                            ^~~~~~~~
hw/tpm/tpm_ioctl.h:224:43: error: expected expression before \u2018ptm_res\u2019
     PTM_SHUTDOWN           = _IOR('P', 2, ptm_res),
                                           ^~~~~~~
hw/tpm/tpm_ioctl.h:225:43: error: expected expression before \u2018ptm_est\u2019
     PTM_GET_TPMESTABLISHED = _IOR('P', 3, ptm_est),
                                           ^~~~~~~
hw/tpm/tpm_ioctl.h:226:44: error: expected expression before \u2018ptm_loc\u2019
     PTM_SET_LOCALITY       = _IOWR('P', 4, ptm_loc),
                                            ^~~~~~~
hw/tpm/tpm_ioctl.h:227:43: error: expected expression before \u2018ptm_res\u2019
     PTM_HASH_START         = _IOR('P', 5, ptm_res),
                                           ^~~~~~~
hw/tpm/tpm_ioctl.h:228:44: error: expected expression before \u2018ptm_hdata\u2019
     PTM_HASH_DATA          = _IOWR('P', 6, ptm_hdata),
                                            ^~~~~~~~~
hw/tpm/tpm_ioctl.h:229:43: error: expected expression before \u2018ptm_res\u2019
     PTM_HASH_END           = _IOR('P', 7, ptm_res),
                                           ^~~~~~~
hw/tpm/tpm_ioctl.h:230:43: error: expected expression before \u2018ptm_res\u2019
     PTM_CANCEL_TPM_CMD     = _IOR('P', 8, ptm_res),
                                           ^~~~~~~
hw/tpm/tpm_ioctl.h:231:43: error: expected expression before \u2018ptm_res\u2019
     PTM_STORE_VOLATILE     = _IOR('P', 9, ptm_res),
                                           ^~~~~~~
hw/tpm/tpm_ioctl.h:232:47: error: expected expression before \u2018ptm_reset_est\u2019
     PTM_RESET_TPMESTABLISHED = _IOWR('P', 10, ptm_reset_est),
                                               ^~~~~~~~~~~~~
hw/tpm/tpm_ioctl.h:233:45: error: expected expression before \u2018ptm_getstate\u2019
     PTM_GET_STATEBLOB      = _IOWR('P', 11, ptm_getstate),
                                             ^~~~~~~~~~~~
hw/tpm/tpm_ioctl.h:234:45: error: expected expression before \u2018ptm_setstate\u2019
     PTM_SET_STATEBLOB      = _IOWR('P', 12, ptm_setstate),
                                             ^~~~~~~~~~~~
  CC      hw/usb/hcd-ohci.o
hw/tpm/tpm_ioctl.h:235:44: error: expected expression before \u2018ptm_res\u2019
     PTM_STOP               = _IOR('P', 13, ptm_res),
                                            ^~~~~~~
hw/tpm/tpm_ioctl.h:236:44: error: expected expression before \u2018ptm_getconfig\u2019
     PTM_GET_CONFIG         = _IOR('P', 14, ptm_getconfig),
                                            ^~~~~~~~~~~~~
hw/tpm/tpm_ioctl.h:237:44: error: expected expression before \u2018ptm_res\u2019
     PTM_SET_DATAFD         = _IOR('P', 15, ptm_res),
                                            ^~~~~~~
hw/tpm/tpm_ioctl.h:238:45: error: expected expression before \u2018ptm_setbuffersize\u2019
     PTM_SET_BUFFERSIZE     = _IOWR('P', 16, ptm_setbuffersize),
                                             ^~~~~~~~~~~~~~~~~
  CC      hw/usb/hcd-ohci-pci.o
gmake: *** [/export/home/denber/qemu-5.0.0/rules.mak:69: hw/tpm/tpm_emulator.o] Error 1

It appears that _IOR & friends are defined in /usr/include/sys/ioccom.h, but I can't figure out how to point gmake to that. Do I need some sort of "-I" in the Makefile?

When I built QEMU 2.12, I ran into this too and ended up just commenting out all the references to _IOR, etc. in hw/tpm/tpm_ioctl.h but maybe we can do something more elegant this time. Thanks.

- Michele

#12 lore raw

gmake in Solaris 11.4: TFR missing

Quoted 6 lines

It appears that _IOR & friends are defined in /usr/include/sys/ioccom.h, but I can't figure out how to point gmake to that. Do I need some sort of "-I" in the Makefile?

When I built QEMU 2.12, I ran into this too and ended up just commenting out all the references to _IOR, etc. in hw/tpm/tpm_ioctl.h but maybe we can do something more elegant this time. Thanks.

OK, I fixed this by adding the line

#include </usr/include/sys/ioccom.h>

to hw/tpm/tpm_ioctl.h.

gmake now went to the end but threw a bunch of errors all complaining about something called TFR. This is the tail:

...
CC aarch64-softmmu/trace/generated-helpers.o
Undefined first referenced
symbol in file
TFR ../net/tap-solaris.o
ld: fatal: symbol referencing errors
collect2: error: ld returned 1 exit status gmake[1]: *** [Makefile:208: qemu-system-mipsel] Error 1
gmake: *** [Makefile:527: mipsel-softmmu/all] Error 2
LINK mips-softmmu/qemu-system-mips
Undefined first referenced
symbol in file
TFR ../net/tap-solaris.o
ld: fatal: symbol referencing errors
collect2: error: ld returned 1 exit status gmake[1]: *** [Makefile:208: qemu-system-mips] Error 1
gmake: *** [Makefile:527: mips-softmmu/all] Error 2
LINK arm-softmmu/qemu-system-arm
Undefined first referenced
symbol in file
TFR ../net/tap-solaris.o
ld: fatal: symbol referencing errors
collect2: error: ld returned 1 exit status gmake[1]: *** [Makefile:208: qemu-system-arm] Error 1
gmake: *** [Makefile:527: arm-softmmu/all] Error 2
LINK aarch64-softmmu/qemu-system-aarch64
Undefined first referenced
symbol in file
TFR ../net/tap-solaris.o
ld: fatal: symbol referencing errors
collect2: error: ld returned 1 exit status gmake[1]: *** [Makefile:208: qemu-system-aarch64] Error 1
gmake: *** [Makefile:527: aarch64-softmmu/all] Error 2 root@hemlock:~/qemu-5.0.0#

A web search for "solaris undefined symbol TFR" turned up nothing and I'm not seeing it on the machine. man TFR returns nothing. I didn't encounter this error while building 2.12. Here's an example of its use (in tap-solaris.c):

TFR(ip_fd = open("/dev/udp", O_RDWR, 0));

What is TFR?

- Michele

#13 lore raw

Re: gmake in Solaris 11.4: TFR missing

Quoted 55 lines · Michele Denber
Quoted 10 lines

It appears that _IOR & friends are defined in /usr/include/sys/ioccom.h, but I can't figure out how to point gmake to that.  Do I need some sort of "-I" in the Makefile?

When I built QEMU 2.12, I ran into this too and ended up just commenting out all the references to _IOR, etc. in hw/tpm/tpm_ioctl.h but maybe we can do something more elegant this time.  Thanks.

OK, I fixed this by adding the line

#include </usr/include/sys/ioccom.h>

to hw/tpm/tpm_ioctl.h.

gmake now went to the end but threw a bunch of errors all complaining about something called TFR.  This is the tail:

 ...
CC      aarch64-softmmu/trace/generated-helpers.o Undefined                       first referenced symbol                             in file
TFR                                 ../net/tap-solaris.o
ld: fatal: symbol referencing errors
collect2: error: ld returned 1 exit status gmake[1]: *** [Makefile:208: qemu-system-mipsel] Error 1
gmake: *** [Makefile:527: mipsel-softmmu/all] Error 2 LINK    mips-softmmu/qemu-system-mips Undefined                       first referenced symbol                             in file
TFR                                 ../net/tap-solaris.o
ld: fatal: symbol referencing errors
collect2: error: ld returned 1 exit status gmake[1]: *** [Makefile:208: qemu-system-mips] Error 1
gmake: *** [Makefile:527: mips-softmmu/all] Error 2 LINK    arm-softmmu/qemu-system-arm
Undefined                       first referenced symbol                             in file
TFR                                 ../net/tap-solaris.o
ld: fatal: symbol referencing errors
collect2: error: ld returned 1 exit status gmake[1]: *** [Makefile:208: qemu-system-arm] Error 1
gmake: *** [Makefile:527: arm-softmmu/all] Error 2 LINK    aarch64-softmmu/qemu-system-aarch64 Undefined                       first referenced symbol                             in file
TFR                                 ../net/tap-solaris.o
ld: fatal: symbol referencing errors
collect2: error: ld returned 1 exit status gmake[1]: *** [Makefile:208: qemu-system-aarch64] Error 1
gmake: *** [Makefile:527: aarch64-softmmu/all] Error 2 root@hemlock:~/qemu-5.0.0#

Quoted 8 lines

A web search for "solaris undefined symbol TFR" turned up nothing and I'm not seeing it on the machine.  man TFR returns nothing.  I didn't encounter this error while building 2.12.  Here's an example of its use (in tap-solaris.c):

 TFR(ip_fd = open("/dev/udp", O_RDWR, 0));

  What is TFR?

TFR() is defined in include/qemu-common.h:

#define TFR(expr) do { if ((expr) != -1) break; } while (errno == EINTR)

Quoted 4 lines

            - Michele

#14 lore raw

Re: gmake in Solaris 11.4: TFR missing

Quoted 5 lines · Philippe Mathieu-Daudé
Quoted 1 line · Michele Denber

What is TFR?

TFR() is defined in include/qemu-common.h:

Yep; adding #include "qemu-common.h" to tap-solaris.c should fix this (I think we missed tap-solaris.c in the header cleanup of commit a8d2532645cf5ce4).

thanks
-- PMM

#15 lore raw

Re: gmake in Solaris 11.4: TFR missing

+Eric for macros & errno handling

Quoted 10 lines · Peter Maydell
Quoted 5 lines · Philippe Mathieu-Daudé
Quoted 1 line · Michele Denber

What is TFR?

TFR() is defined in include/qemu-common.h:

Yep; adding #include "qemu-common.h" to tap-solaris.c should fix this (I think we missed tap-solaris.c in the header cleanup of commit a8d2532645cf5ce4).

Any idea what means TFR? I understand it keeps retrying while interrupted, but can't find the origin of that abbreviation.

I'm not sure what we gain by using this macro, it seems dangerous as there is no guaranty we 'expr' is a single libc call updating errno.

We have:

include/qemu-common.h:13:#define TFR(expr) do { if ((expr) != -1) break; } while (errno == EINTR)

block/file-posix.c:1515: } while (errno == EINTR);
block/file-posix.c:1539: } while (errno == EINTR);
block/file-posix.c:1701: } while (errno == EINTR);

Good, de-macro-ified form.

chardev/char-fd.c:122: TFR(fd = qemu_open(src, flags, 0666));
chardev/char-pipe.c:135: TFR(fd_in = qemu_open(filename_in, O_RDWR |
O_BINARY));
chardev/char-pipe.c:136: TFR(fd_out = qemu_open(filename_out, O_RDWR
| O_BINARY));
chardev/char-pipe.c:146: TFR(fd_in = fd_out = qemu_open(filename,
O_RDWR | O_BINARY));

^ few cases, we need to check if qemu_open() is errno safe. Maybe better to inline instead?

os-posix.c:297: TFR(fd = qemu_open("/dev/null", O_RDWR));

^ I don't really understand this one.

net/tap-bsd.c:68: TFR(fd = open(dname, O_RDWR));
net/tap-bsd.c:123: TFR(fd = open(PATH_NET_TAP, O_RDWR));
net/tap-bsd.c:171: TFR(fd = open(dname, O_RDWR));
net/tap-linux.c:49: TFR(fd = open(PATH_NET_TUN, O_RDWR));
net/tap-solaris.c:87: TFR(ip_fd = open("/dev/udp", O_RDWR, 0));
net/tap-solaris.c:93: TFR(tap_fd = open("/dev/tap", O_RDWR, 0));
net/tap-solaris.c:107: TFR(if_fd = open("/dev/tap", O_RDWR, 0));
net/tap-solaris.c:140: TFR(arp_fd = open ("/dev/tap", O_RDWR, 0));
net/tap.c:625: TFR(fd = tap_open(ifname, ifname_sz, vnet_hdr,
vnet_hdr_required,

Ditto, is tap_open() errno safe? If this code prefers to keep the macro, we could move it to net/tap_int.h.

#18 lore raw

Re: gmake in Solaris 11.4: TFR missing

In reply to Michele Denber
Quoted 17 lines · Michele Denber
Quoted 10 lines

It appears that _IOR & friends are defined in /usr/include/sys/ioccom.h, but I can't figure out how to point gmake to that.  Do I need some sort of "-I" in the Makefile?

When I built QEMU 2.12, I ran into this too and ended up just commenting out all the references to _IOR, etc. in hw/tpm/tpm_ioctl.h but maybe we can do something more elegant this time.  Thanks.

OK, I fixed this by adding the line

#include </usr/include/sys/ioccom.h>

to hw/tpm/tpm_ioctl.h.

Quoted 8 lines

gmake now went to the end but threw a bunch of errors all complaining about something called TFR.  This is the tail:

 ...
CC      aarch64-softmmu/trace/generated-helpers.o Undefined                       first referenced symbol                             in file
TFR                                 ../net/tap-solaris.o

TFR is a macro that is defined in our qemu-common.h header ... does it work if you add a #include "qemu-common.h" somewhere at the beginning of net/tap-solaris.c ?

Thomas

#16 lore raw

Re: gmake in Solaris 11.4: TFR missing

In reply to Philippe Mathieu-Daudé
Quoted 2 lines · Philippe Mathieu-Daudé

Any idea what means TFR? I understand it keeps retrying while interrupted, but can't find the origin of that abbreviation.

Probably it stands for "temporary failure retry" -- glibc provides essentially the same primitive in unistd.h as TEMP_FAILURE_RETRY:
https://www.gnu.org/software/libc/manual/html_node/Interrupted-Primitives.html

TFR is pretty old as QEMU code goes: added in commit aeb30be60a92148e38d in 2007.

Quoted 3 lines

I'm not sure what we gain by using this macro, it seems dangerous as there is no guaranty we 'expr' is a single libc call updating errno.

Well, the design of the macro is that it is for use to wrap expressions which are a single thing that update errno...

The gain overall is that "we need to retry practically everything to handle EINTR" results in lots of awkward open-coded loops, and in theory putting them behind macros makes the normal-case handling more obvious. But this only works if the codebase is going to consistently adopt TFR() as its EINTR-handling strategy.

(EINTR is one of the big failures of the Unix/POSIX API, in my view.)

thanks
-- PMM

#17 lore raw

Re: gmake in Solaris 11.4: TFR missing

In reply to Philippe Mathieu-Daudé
Quoted 1 line · Philippe Mathieu-Daudé

TFR() is defined in include/qemu-common.h: #define TFR(expr) do { if ((expr) != -1) break; } while (errno == EINTR)

Ah ha, thank you. It figures, the one place I didn't look. I assumed that QEMU would know how to find its own things so TFR must be something external.

- Michele

#19 lore raw

Re: gmake in Solaris 11.4: TFR missing

In reply to Thomas Huth
Quoted 3 lines · Thomas Huth

TFR is a macro that is defined in our qemu-common.h header ... does it work if you add a #include "qemu-common.h" somewhere at the beginning of net/tap-solaris.c ?

Yes, thank you, that fixed it:

root@hemlock:~/qemu-5.0.0# gmake -j16
gmake[1]: Entering directory '/export/home/denber/qemu-5.0.0/slirp'
grep: illegal option -- m
Usage: grep [-c|-l|-q] -bhinsvw pattern file . . .
gmake[1]: Nothing to be done for 'all'.
gmake[1]: Leaving directory '/export/home/denber/qemu-5.0.0/slirp'
        CHK version_gen.h
  CC      net/tap-solaris.o
  LINK    moxie-softmmu/qemu-system-moxie
  LINK    cris-softmmu/qemu-system-cris
  LINK    microblaze-softmmu/qemu-system-microblaze
  LINK    microblazeel-softmmu/qemu-system-microblazeel
  LINK    nios2-softmmu/qemu-system-nios2
  LINK    lm32-softmmu/qemu-system-lm32
  LINK    m68k-softmmu/qemu-system-m68k
  LINK    hppa-softmmu/qemu-system-hppa
  LINK    alpha-softmmu/qemu-system-alpha
  LINK    mips64-softmmu/qemu-system-mips64
  LINK    mips64el-softmmu/qemu-system-mips64el
  LINK    mipsel-softmmu/qemu-system-mipsel
  LINK    mips-softmmu/qemu-system-mips
  LINK    i386-softmmu/qemu-system-i386
  LINK    aarch64-softmmu/qemu-system-aarch64
  LINK    arm-softmmu/qemu-system-arm
  LINK    or1k-softmmu/qemu-system-or1k
  CC      s390x-softmmu/gen-features
/bin/sh: cc: not found
gmake[1]: *** [/export/home/denber/qemu-5.0.0/target/s390x/Makefile.objs:25: /export/home/denber/qemu-5.0.0/s390x-softmmu/gen-features] Error 127
gmake: *** [Makefile:527: s390x-softmmu/all] Error 2
gmake: *** Waiting for unfinished jobs....
  LINK    rx-softmmu/qemu-system-rx
  LINK    riscv32-softmmu/qemu-system-riscv32
  LINK    riscv64-softmmu/qemu-system-riscv64
  LINK    ppc64-softmmu/qemu-system-ppc64
  LINK    ppc-softmmu/qemu-system-ppc
root@hemlock:~/qemu-5.0.0#

I still have one grep/ggrep issue and one "wrong sh" issue. And it doesn't look like it really completed correctly because for example I see:

root@hemlock:~/qemu-5.0.0# cd x86_64-softmmu
root@hemlock:~/qemu-5.0.0/x86_64-softmmu# ls
config-devices.mak      config-devices.mak.old  Makefile
config-devices.mak.d    config-target.mak
root@hemlock:~/qemu-5.0.0/x86_64-softmmu#

So even though gmake finished, it didn't make at least one VM. Is that because of the two remaining errors?

- Michele

#20 lore raw

Re: gmake in Solaris 11.4: TFR missing

Quoted 25 lines · Michele Denber
Quoted 3 lines · Thomas Huth

TFR is a macro that is defined in our qemu-common.h header ... does it work if you add a #include "qemu-common.h" somewhere at the beginning of net/tap-solaris.c ?

Yes, thank you, that fixed it:

root@hemlock:~/qemu-5.0.0# gmake -j16
gmake[1]: Entering directory '/export/home/denber/qemu-5.0.0/slirp'
grep: illegal option -- m
Usage: grep [-c|-l|-q] -bhinsvw pattern file . . .
gmake[1]: Nothing to be done for 'all'.
gmake[1]: Leaving directory '/export/home/denber/qemu-5.0.0/slirp'
        CHK version_gen.h
  CC      net/tap-solaris.o
  LINK    moxie-softmmu/qemu-system-moxie
  LINK    cris-softmmu/qemu-system-cris
  LINK    microblaze-softmmu/qemu-system-microblaze
  LINK    microblazeel-softmmu/qemu-system-microblazeel
  LINK    nios2-softmmu/qemu-system-nios2
  LINK    lm32-softmmu/qemu-system-lm32
  LINK    m68k-softmmu/qemu-system-m68k
  LINK    hppa-softmmu/qemu-system-hppa
  LINK    alpha-softmmu/qemu-system-alpha
  LINK    mips64-softmmu/qemu-system-mips64
  LINK    mips64el-softmmu/qemu-system-mips64el
Quoted 8 lines
  LINK    mipsel-softmmu/qemu-system-mipsel
  LINK    mips-softmmu/qemu-system-mips
  LINK    i386-softmmu/qemu-system-i386
  LINK    aarch64-softmmu/qemu-system-aarch64
  LINK    arm-softmmu/qemu-system-arm
  LINK    or1k-softmmu/qemu-system-or1k
  CC      s390x-softmmu/gen-features
/bin/sh: cc: not found

Odd...

Assuming you are using gcc, have you tried 'configure --host-cc=gcc'?

Quoted 15 lines

gmake[1]: *** [/export/home/denber/qemu-5.0.0/target/s390x/Makefile.objs:25:
/export/home/denber/qemu-5.0.0/s390x-softmmu/gen-features] Error 127
gmake: *** [Makefile:527: s390x-softmmu/all] Error 2
gmake: *** Waiting for unfinished jobs....
  LINK    rx-softmmu/qemu-system-rx
  LINK    riscv32-softmmu/qemu-system-riscv32
  LINK    riscv64-softmmu/qemu-system-riscv64
  LINK    ppc64-softmmu/qemu-system-ppc64
  LINK    ppc-softmmu/qemu-system-ppc root@hemlock:~/qemu-5.0.0#

I still have one grep/ggrep issue and one "wrong sh" issue.  And it doesn't look like it really completed correctly because for example I see:

Quoted 8 lines
root@hemlock:~/qemu-5.0.0# cd x86_64-softmmu
root@hemlock:~/qemu-5.0.0/x86_64-softmmu# ls
config-devices.mak      config-devices.mak.old  Makefile
config-devices.mak.d    config-target.mak
root@hemlock:~/qemu-5.0.0/x86_64-softmmu#

So even though gmake finished, it didn't make at least one VM.  Is that because of the two remaining errors?

From the LINK error message you pasted, you successfully built various QEMU, but the s390x-softmmu one failed.

Try running for example the QEMU targetting ppc64-softmmu, by listing the available machines:

$ ppc64-softmmu/qemu-system-ppc64 -machine help
Supported machines are:
40p                  IBM RS/6000 7020 (40p)
bamboo               bamboo
g3beige              Heathrow based PowerMAC
mac99                Mac99 based PowerMAC
mpc8544ds            mpc8544ds
...
Quoted 3 lines

            - Michele

#21 lore raw

Re: gmake in Solaris 11.4: TFR missing

Quoted 3 lines · Philippe Mathieu-Daudé

Odd...

Assuming you are using gcc, have you tried 'configure --host-cc=gcc'?

OK, so I changed it from

# /opt/csw/bin/bash ./configure --cc=gcc  --extra-cflags="-m32"

to

# /opt/csw/bin/bash ./configure --cc=gcc  --extra-cflags="-m32" --host-cc=gcc

and then reran gmake -j16. That seems to have fixed it. ppc64 is there now:

root@hemlock:~/qemu-5.0.0# ppc64-softmmu/qemu-system-ppc64 -machine help
Supported machines are:
40p                  IBM RS/6000 7020 (40p)
bamboo               bamboo
g3beige              Heathrow based PowerMAC
mac99                Mac99 based PowerMAC
mpc8544ds            mpc8544ds
... and a bunch more ...
root@hemlock:~/qemu-5.0.0#

And I also have the right stuff in x86_64-softmmu. So I tried running my Win XP image and it started up OK, opening the QEMU window and displaying the XP logo. But unfortunately it's really really slow, like 20 minutes just to get to the XP desktop appearing. That takes about 2 minutes in QEMU 2.12 on the same host hardware. I had been hoping 5.0 would be faster. Is there a reason for it to be an order of magnitude slower? I'm saying:

root@hemlock:~/qemu-5.0.0/x86_64-softmmu# ./qemu-system-x86_64  -m 2047  -usb -device usb-tablet -smp 3 -device rtl8139,netdev=net0 -netdev user,id=net0 -boot d -hda  /bkpool/qemuimages/XP.img -cdrom /bkpool/qemuimages/xp.iso &

But in any case, I will now cautiously claim that I have succeeded in building QEMU 5.0 in Solaris 11.4 SPARC. Thanks very much to everyone for all your help.

- Michele

#23 lore raw

Re: gmake in Solaris 11.4: TFR missing

In reply to Philippe Mathieu-Daudé

Update: actually, after using it for a bit, it appears that XP runs at about the same speed in 5.0 as it did in 2.12. It's just the initial boot that's very slow. Does this message (which I discovered hiding under the QEMU window) have anything to do with it?

(qemu:27712): Gtk-WARNING **: Could not load a pixbuf from /org/gtk/libgtk/theme/Adwaita/assets/bullet-symbolic.svg. This may indicate that pixbuf loaders or the mime database could not be found.

- Michele

#22 lore raw

Re: gmake in Solaris 11.4: TFR missing

In reply to Michele Denber
Quoted 7 lines · Michele Denber

On 07/04/20 11:57, Philippe Mathieu-Daudé wrote:

Odd...

Assuming you are using gcc, have you tried 'configure --host-cc=gcc'?

Quoted 7 lines

OK, so I changed it from

# /opt/csw/bin/bash ./configure --cc=gcc  --extra-cflags="-m32"

to

# /opt/csw/bin/bash ./configure --cc=gcc  --extra-cflags="-m32" --host-cc=gcc

It would theoretically be possible to make configure a bit smarter about --host-cc. It defaults to "cc" even if you passed a --cc option, which is probably necessary for cross-compile setups, but we could make the default look in more places. We also ideally ought to check that whatever it is actually exists and builds host binaries, rather than just passing it through to make and letting the build process fail later.

thanks
-- PMM