| Message ID | 1498040401-16361-2-git-send-email-yang.zhong@intel.com |
|---|---|
| State | New |
| Headers | show |
On 06/21/2017 03:19 AM, Yang Zhong wrote: > +case "$target_name" in > + alpha|arm|armeb|aarch64|cris|hppa|lm32|m68k|microblaze|microblazeel|mips|mipsel| \ > + mipsn32|mipsn32el|mips64|mips64el|moxie|nios2|or1k|ppc|ppcemb|ppc64|ppc64le|ppc64abi32| \ > + sh4|sh4eb|sparc|sparc64|sparc32plus|s390x|tilegx|tricore|unicore32|xtensa|xtensaeb) > + if test "$target_softmmu" = "yes" ; then > + echo "CONFIG_TCG=y" >> $config_target_mak > + fi > + ;; > +esac > +case "$target_name" in > + i386|x86_64) > + if test "$tcg" = "yes" -a "$target_softmmu" = "yes" ; then > + echo "CONFIG_TCG=y" >> $config_target_mak > + fi > + ;; > +esac This is definitely the wrong way around. Test for the thing you want to special case -- x86, when running on x86, and softmmu -- and then everything else via *. I strongly suspect this test should be case "$cpu-$target-$tcg" in x86_64-i386-softmmu-no | x86_64-x86_64-softmmu-no) # CONFIG_TCG disabled. ;; *) echo "CONFIG_TCG=y" >> $config_target_mak esac This is easily extendable to allow TCG to be disabled for other KVM hosts. r~
On 21.06.2017 12:19, Yang Zhong wrote: > Add the disable-tcg option into configure and echo CONFIG_TCG=y into > $config_target_mak. The default tcg is enabled for all build. If tcg > is disabled in the build, only i386|x86_64 softmmu option can be disabled, > other softmmu of tagets and users build defaultly enabled the tcg. Why do you want to limit this disablement to x86 only? There are also other architectures that support KVM (ARM, PPC, MIPS), so disabling TCG might be possible there, too. So I think it might be better to check whether KVM is possible instead. Thomas
> On 21.06.2017 12:19, Yang Zhong wrote: > > Add the disable-tcg option into configure and echo CONFIG_TCG=y into > > $config_target_mak. The default tcg is enabled for all build. If tcg > > is disabled in the build, only i386|x86_64 softmmu option can be disabled, > > other softmmu of tagets and users build defaultly enabled the tcg. > Why do you want to limit this disablement to x86 only? There are also > other architectures that support KVM (ARM, PPC, MIPS), so disabling TCG > might be possible there, too. So I think it might be better to check > whether KVM is possible instead. You need to be careful and not use any helper from e.g. KVM or migration code. So I would be very surprised if any other architecture compiles with --disable-tcg. Paolo
On 22.06.2017 08:22, Paolo Bonzini wrote: > >> On 21.06.2017 12:19, Yang Zhong wrote: >>> Add the disable-tcg option into configure and echo CONFIG_TCG=y into >>> $config_target_mak. The default tcg is enabled for all build. If tcg >>> is disabled in the build, only i386|x86_64 softmmu option can be disabled, >>> other softmmu of tagets and users build defaultly enabled the tcg. >> Why do you want to limit this disablement to x86 only? There are also >> other architectures that support KVM (ARM, PPC, MIPS), so disabling TCG >> might be possible there, too. So I think it might be better to check >> whether KVM is possible instead. > > You need to be careful and not use any helper from e.g. KVM or migration > code. So I would be very surprised if any other architecture compiles > with --disable-tcg. OK, fair, but we finally might want to get there, so I think we should allow the parameter in the configure script for other architectures, too, and then fix the bugs once we can try it out. Thomas
On 22/06/2017 08:33, Thomas Huth wrote: > On 22.06.2017 08:22, Paolo Bonzini wrote: >> >>> On 21.06.2017 12:19, Yang Zhong wrote: >>>> Add the disable-tcg option into configure and echo CONFIG_TCG=y into >>>> $config_target_mak. The default tcg is enabled for all build. If tcg >>>> is disabled in the build, only i386|x86_64 softmmu option can be disabled, >>>> other softmmu of tagets and users build defaultly enabled the tcg. >>> Why do you want to limit this disablement to x86 only? There are also >>> other architectures that support KVM (ARM, PPC, MIPS), so disabling TCG >>> might be possible there, too. So I think it might be better to check >>> whether KVM is possible instead. >> >> You need to be careful and not use any helper from e.g. KVM or migration >> code. So I would be very surprised if any other architecture compiles >> with --disable-tcg. > > OK, fair, but we finally might want to get there, so I think we should > allow the parameter in the configure script for other architectures, > too, and then fix the bugs once we can try it out. I think it's the other way round---when someone wants to fix it, they should add their architecture to the list of targets that support --disable-tcg. "./configure <anything>", as a goal, should fail immediately if make won't succeed; compilation errors are always a worse experience. Paolo
On 22.06.2017 11:26, Paolo Bonzini wrote: > > > On 22/06/2017 08:33, Thomas Huth wrote: >> On 22.06.2017 08:22, Paolo Bonzini wrote: >>> >>>> On 21.06.2017 12:19, Yang Zhong wrote: >>>>> Add the disable-tcg option into configure and echo CONFIG_TCG=y into >>>>> $config_target_mak. The default tcg is enabled for all build. If tcg >>>>> is disabled in the build, only i386|x86_64 softmmu option can be disabled, >>>>> other softmmu of tagets and users build defaultly enabled the tcg. >>>> Why do you want to limit this disablement to x86 only? There are also >>>> other architectures that support KVM (ARM, PPC, MIPS), so disabling TCG >>>> might be possible there, too. So I think it might be better to check >>>> whether KVM is possible instead. >>> >>> You need to be careful and not use any helper from e.g. KVM or migration >>> code. So I would be very surprised if any other architecture compiles >>> with --disable-tcg. >> >> OK, fair, but we finally might want to get there, so I think we should >> allow the parameter in the configure script for other architectures, >> too, and then fix the bugs once we can try it out. > > I think it's the other way round---when someone wants to fix it, they > should add their architecture to the list of targets that support > --disable-tcg. "./configure <anything>", as a goal, should fail > immediately if make won't succeed; compilation errors are always a worse > experience. Fine for me, too, but then the configure script should exit with a proper error message when the user tried to specify --disable-tcg on a non-x86 architecture. If I get the current patch right, it silently falls back to CONFIG_TCG=y instead, which is IMHO also a bad experience. Thomas
On 22/06/2017 11:30, Thomas Huth wrote: > On 22.06.2017 11:26, Paolo Bonzini wrote: >> >> >> On 22/06/2017 08:33, Thomas Huth wrote: >>> On 22.06.2017 08:22, Paolo Bonzini wrote: >>>> >>>>> On 21.06.2017 12:19, Yang Zhong wrote: >>>>>> Add the disable-tcg option into configure and echo CONFIG_TCG=y into >>>>>> $config_target_mak. The default tcg is enabled for all build. If tcg >>>>>> is disabled in the build, only i386|x86_64 softmmu option can be disabled, >>>>>> other softmmu of tagets and users build defaultly enabled the tcg. >>>>> Why do you want to limit this disablement to x86 only? There are also >>>>> other architectures that support KVM (ARM, PPC, MIPS), so disabling TCG >>>>> might be possible there, too. So I think it might be better to check >>>>> whether KVM is possible instead. >>>> >>>> You need to be careful and not use any helper from e.g. KVM or migration >>>> code. So I would be very surprised if any other architecture compiles >>>> with --disable-tcg. >>> >>> OK, fair, but we finally might want to get there, so I think we should >>> allow the parameter in the configure script for other architectures, >>> too, and then fix the bugs once we can try it out. >> >> I think it's the other way round---when someone wants to fix it, they >> should add their architecture to the list of targets that support >> --disable-tcg. "./configure <anything>", as a goal, should fail >> immediately if make won't succeed; compilation errors are always a worse >> experience. > > Fine for me, too, but then the configure script should exit with a > proper error message when the user tried to specify --disable-tcg on a > non-x86 architecture. If I get the current patch right, it silently > falls back to CONFIG_TCG=y instead, which is IMHO also a bad experience. Yes, I agree. Thanks for reviewing the patch. Paolo
diff --git a/configure b/configure index ff0f8b9..b4d80e5 100755 --- a/configure +++ b/configure @@ -224,6 +224,7 @@ cap_ng="" attr="" libattr="" xfs="" +tcg="yes" vhost_net="no" vhost_scsi="no" @@ -953,6 +954,10 @@ for opt do ;; --enable-hax) hax="yes" ;; + --disable-tcg) tcg="no" + ;; + --enable-tcg) tcg="yes" + ;; --disable-tcg-interpreter) tcg_interpreter="no" ;; --enable-tcg-interpreter) tcg_interpreter="yes" @@ -5117,7 +5122,6 @@ echo "module support $modules" echo "host CPU $cpu" echo "host big endian $bigendian" echo "target list $target_list" -echo "tcg debug enabled $debug_tcg" echo "gprof enabled $gprof" echo "sparse enabled $sparse" echo "strip binaries $strip_opt" @@ -5171,6 +5175,11 @@ echo "Linux AIO support $linux_aio" echo "ATTR/XATTR support $attr" echo "Install blobs $blobs" echo "KVM support $kvm" +echo "TCG support $tcg" +if test "$tcg" = "yes" ; then + echo "TCG debug enabled $debug_tcg" + echo "TCG interpreter $tcg_interpreter" +fi echo "HAX support $hax" echo "RDMA support $rdma" echo "TCG interpreter $tcg_interpreter" @@ -6229,6 +6238,7 @@ fi if test "$target_user_only" = "yes" ; then echo "CONFIG_USER_ONLY=y" >> $config_target_mak echo "CONFIG_QEMU_INTERP_PREFIX=\"$interp_prefix1\"" >> $config_target_mak + echo "CONFIG_TCG=y" >> $config_target_mak fi if test "$target_linux_user" = "yes" ; then echo "CONFIG_LINUX_USER=y" >> $config_target_mak @@ -6248,6 +6258,22 @@ if test "$target_bsd_user" = "yes" ; then echo "CONFIG_BSD_USER=y" >> $config_target_mak fi +case "$target_name" in + alpha|arm|armeb|aarch64|cris|hppa|lm32|m68k|microblaze|microblazeel|mips|mipsel| \ + mipsn32|mipsn32el|mips64|mips64el|moxie|nios2|or1k|ppc|ppcemb|ppc64|ppc64le|ppc64abi32| \ + sh4|sh4eb|sparc|sparc64|sparc32plus|s390x|tilegx|tricore|unicore32|xtensa|xtensaeb) + if test "$target_softmmu" = "yes" ; then + echo "CONFIG_TCG=y" >> $config_target_mak + fi + ;; +esac +case "$target_name" in + i386|x86_64) + if test "$tcg" = "yes" -a "$target_softmmu" = "yes" ; then + echo "CONFIG_TCG=y" >> $config_target_mak + fi + ;; +esac # generate QEMU_CFLAGS/LDFLAGS for targets cflags=""
Add the disable-tcg option into configure and echo CONFIG_TCG=y into $config_target_mak. The default tcg is enabled for all build. If tcg is disabled in the build, only i386|x86_64 softmmu option can be disabled, other softmmu of tagets and users build defaultly enabled the tcg. This operation do not make big change with the older build command. The new configure build command like below (1)./configure tcg is defaultly enabled (2)./configure --disable-tcg tcg is disabled in i386 and x86_64 softmmu build tcg is enabled in others softmmu and all users build (3)./configure --disable-tcg --target-list=x86_64-softmmu, x86_64-linux-user tcg is disabled in x86_64 softmmu build tcg is enabled in x86_64 linux user build Signed-off-by: Yang Zhong <yang.zhong@intel.com> --- configure | 28 +++++++++++++++++++++++++++- 1 file changed, 27 insertions(+), 1 deletion(-)