postinst and clang fixes

This pull fixes boost with clang
other fixes are needed for risc-v to enable offline locale and postint generation, especially after
https://patchwork.openembedded.org/patch/154363/

Now we can boot qemuriscv64 quite fast

v2:
Fix core-image-lsb-sdk build for risc-v

The following changes since commit 0d69f86e50e57afe893dade095c1dbd3d05c3b0f:

libsdl2: Fix left rotated display for RaspPi/VC4/GLES2 (2018-08-29 15:43:18 +0100)

are available in the Git repository at:

git://git.openembedded.org/openembedded-core-contrib kraj/pu-1
http://cgit.openembedded.org/openembedded-core-contrib/log/?h=kraj/pu-1

Khem Raj (7):
boost: Fix invalid const in atomic builtins
libc-package: Add risc-v specific options to cross-localedef
invocation
sysklogd: Use update-alternatives
glibc-locale: Enable riscv64 and riscv32 in BINARY_LOCALE_ARCHES
supported architectures
packagegroup-core-tools-testapps: Exclude gst-examples on riscv64
tcf-agent: Disable non-building features on riscv64
gdb: Seprate out gold and dwp into a variable

 meta/classes/libc-package.bbclass             |  2 +
 meta/recipes-core/glibc/glibc-locale.inc      |  4 +-
 .../packagegroup-core-tools-testapps.bb       |  5 +-
 meta/recipes-devtools/binutils/binutils.inc   |  5 +-
 .../tcf-agent/tcf-agent_git.bb                |  1 +
 meta/recipes-extended/sysklogd/sysklogd.inc   |  8 +-
 ...ecific-branch-for-x86-DCAS-based-loa.patch | 76 +++++++++++++++++++
 meta/recipes-support/boost/boost_1.67.0.bb    | 15 ++--
 8 files changed, 104 insertions(+), 12 deletions(-)
 create mode 100644 meta/recipes-support/boost/boost/0001-Removed-clang-specific-branch-for-x86-DCAS-based-loa.patch

2.18.0

Fixes build with clang

---
 ...ecific-branch-for-x86-DCAS-based-loa.patch | 76 +++++++++++++++++++
 meta/recipes-support/boost/boost_1.67.0.bb    | 15 ++--
 2 files changed, 84 insertions(+), 7 deletions(-)
 create mode 100644 meta/recipes-support/boost/boost/0001-Removed-clang-specific-branch-for-x86-DCAS-based-loa.patch

new file mode 100644
index 0000000000..182693079a
@@ -0,0 +1,76 @@
+From 39b027171e0a619d49b9dd2e8471d10b6c41bc25 Mon Sep 17 00:00:00 2001
+From: Andrey Semashev <[email protected]>
+Date: Tue, 17 Jul 2018 12:37:29 +0300
+Subject: [PATCH] Removed clang-specific branch for x86 DCAS-based loads.
+
+The storage to load from is const-qualified and DCAS via compiler intrinsics
+require an unqualified pointer. Use asm implementation instead, which should be
+as efficient as intrinsics, if not better, in this case.
+
+Fixes https://github.com/boostorg/atomic/issues/15.
+
+Upstream-Status: Backport [https://github.com/boostorg/atomic/commit/6e14ca24dab50ad4c1fa8c27c7dd6f1cb791b534]
+Signed-off-by: Khem Raj <[email protected]>
+---
+ boost/atomic/detail/ops_gcc_x86_dcas.hpp | 23 ++++++++++++-----------
+ 1 file changed, 12 insertions(+), 11 deletions(-)
+
+diff --git a/boost/atomic/detail/ops_gcc_x86_dcas.hpp b/boost/atomic/detail/ops_gcc_x86_dcas.hpp
+index 4dacc66f..b43ef23a 100644
+--- a/boost/atomic/detail/ops_gcc_x86_dcas.hpp
++++ b/boost/atomic/detail/ops_gcc_x86_dcas.hpp
+@@ -158,11 +158,13 @@ struct gcc_dcas_x86
+         }
+         else
+         {
+-#if defined(__clang__)
+-            // Clang cannot allocate eax:edx register pairs but it has sync intrinsics
+-            value = __sync_val_compare_and_swap(&storage, (storage_type)0, (storage_type)0);
+-#elif defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS)
++            // Note that despite const qualification cmpxchg8b below may issue a store to the storage. The storage value
++            // will not change, but this prevents the storage to reside in read-only memory.
++
++#if defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS)
++
+             uint32_t value_bits[2];
++
+             // We don't care for comparison result here; the previous value will be stored into value anyway.
+             // Also we don't care for ebx and ecx values, they just have to be equal to eax and edx before cmpxchg8b.
+             __asm__ __volatile__
+@@ -175,7 +177,9 @@ struct gcc_dcas_x86
+                 : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
+             );
+             BOOST_ATOMIC_DETAIL_MEMCPY(&value, value_bits, sizeof(value));
++
+ #else // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS)
++
+             // We don't care for comparison result here; the previous value will be stored into value anyway.
+             // Also we don't care for ebx and ecx values, they just have to be equal to eax and edx before cmpxchg8b.
+             __asm__ __volatile__
+@@ -187,6 +191,7 @@ struct gcc_dcas_x86
+                 : [storage] "m" (storage)
+                 : BOOST_ATOMIC_DETAIL_ASM_CLOBBER_CC_COMMA "memory"
+             );
++
+ #endif // defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS)
+         }
+ 
+@@ -401,15 +406,11 @@ struct gcc_dcas_x86_64
+ 
+     static BOOST_FORCEINLINE storage_type load(storage_type const volatile& storage, memory_order) BOOST_NOEXCEPT
+     {
+-#if defined(__clang__)
++        // Note that despite const qualification cmpxchg16b below may issue a store to the storage. The storage value
++        // will not change, but this prevents the storage to reside in read-only memory.
+ 
+-        // Clang cannot allocate rax:rdx register pairs but it has sync intrinsics
+-        storage_type value = storage_type();
+-        return __sync_val_compare_and_swap(&storage, value, value);
+-
+-#elif defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS)
++#if defined(BOOST_ATOMIC_DETAIL_X86_NO_ASM_AX_DX_PAIRS)
+ 
+-        // Some compilers can't allocate rax:rdx register pair either and also don't support 128-bit __sync_val_compare_and_swap
+         uint64_t value_bits[2];
+ 
+         // We don't care for comparison result here; the previous value will be stored into value anyway.

index 7bb451166c..ef178edc0c 100644
@@ -2,10 +2,11 @@ require boost-${PV}.inc
 require boost.inc
 
 SRC_URI += "\
-    file://arm-intrinsics.patch \
-    file://boost-CVE-2012-2677.patch \
-    file://boost-math-disable-pch-for-gcc.patch \
-    file://0001-Apply-boost-1.62.0-no-forced-flags.patch.patch \
-    file://0003-Don-t-set-up-arch-instruction-set-flags-we-do-that-o.patch \
-    file://0001-make_x86_64_sysv_elf_gas.S-set-.file-section.patch \
-"
+           file://arm-intrinsics.patch \
+           file://boost-CVE-2012-2677.patch \
+           file://boost-math-disable-pch-for-gcc.patch \
+           file://0001-Apply-boost-1.62.0-no-forced-flags.patch.patch \
+           file://0003-Don-t-set-up-arch-instruction-set-flags-we-do-that-o.patch \
+           file://0001-make_x86_64_sysv_elf_gas.S-set-.file-section.patch \
+           file://0001-Removed-clang-specific-branch-for-x86-DCAS-based-loa.patch \
+           "

2.18.0

It was otherwise being deferred until boot time

---
 meta/classes/libc-package.bbclass | 2 ++
 1 file changed, 2 insertions(+)

index 2e7cd25297..9d09c7be6a 100644
@@ -257,6 +257,8 @@ python package_do_split_gconvs () {
                 "mipsisa32r6el":  " --uint32-align=4 --little-endian ", \
                 "mips64el":" --uint32-align=4 --little-endian ", \
                 "mipsisa64r6el":" --uint32-align=4 --little-endian ", \
+                "riscv64": " --uint32-align=4 --little-endian ", \
+                "riscv32": " --uint32-align=4 --little-endian ", \
                 "i586":    " --uint32-align=4 --little-endian ", \
                 "i686":    " --uint32-align=4 --little-endian ", \
                 "x86_64":  " --uint32-align=4 --little-endian "  }

2.18.0

busybox also provides klogd and syslogd, this change makes it coexist peacefully. Currently rootfs fails in situations where both of them are providing these binaries and busybox postinsts fail

update-alternatives: Error: not linking /mnt/a/oe/build/tmp/work/qemuriscv64-bec-linux/core-image-full-cmdline/1.0-r0/rootfs/sbin/klogd to /bin/busybox.nosuid since /mnt/a/oe/build/tmp/work/qemuriscv64-bec-linux/core-image-full-cmdline/1.0-r0/rootfs/sbin/klogd exists and is not a link

---
 meta/recipes-extended/sysklogd/sysklogd.inc | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

index fc4e67c18e..2a8bed00f3 100644
@@ -11,7 +11,7 @@ LIC_FILES_CHKSUM = "file://COPYING;md5=8ca43cbc842c2336e835926c2166c28b \
                     file://klogd.c;beginline=2;endline=19;md5=7e87ed0ae6142de079bce738c10c899d \
                    "
 
-inherit update-rc.d systemd
+inherit update-rc.d systemd update-alternatives
 
 SRC_URI = "http://www.infodrom.org/projects/sysklogd/download/sysklogd-${PV}.tar.gz \
            file://no-strip-install.patch \

@@ -70,3 +70,9 @@ python () {
     if not bb.utils.contains('DISTRO_FEATURES', 'sysvinit', True, False, d):
         d.setVar("INHIBIT_UPDATERCD_BBCLASS", "1")
 }
+
+ALTERNATIVE_PRIORITY = "100"
+ALTERNATIVE_${PN} = "klogd syslogd"
+ALTERNATIVE_LINK_NAME[klogd] = "${base_sbindir}/klogd"
+ALTERNATIVE_LINK_NAME[syslogd] = "${base_sbindir}/syslogd"
+

2.18.0

This support is there using cross-localedef but was not enabled, this caused the post-install for locales to be emitted which delays the boot significantly.

emitted postints for locales contain exit 1 which is flagged now and was causing image generation to fail.

---
 meta/recipes-core/glibc/glibc-locale.inc | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

index e50e5cf5e3..1b676dc26e 100644
@@ -21,13 +21,13 @@ RDEPENDS_localedef += "glibc"
 # to decrease initial boot time and avoid localedef being killed by the OOM
 # killer which used to effectively break i18n on machines with < 128MB RAM.
 
-# default to disabled 
+# default to disabled
 ENABLE_BINARY_LOCALE_GENERATION ?= "0"
 ENABLE_BINARY_LOCALE_GENERATION_pn-nativesdk-glibc-locale = "1"
 
 #enable locale generation on these arches
 # BINARY_LOCALE_ARCHES is a space separated list of regular expressions
-BINARY_LOCALE_ARCHES ?= "arm.* aarch64 i[3-6]86 x86_64 powerpc mips mips64"
+BINARY_LOCALE_ARCHES ?= "arm.* aarch64 i[3-6]86 x86_64 powerpc mips mips64 riscv32 riscv64"
 
 # set "1" to use cross-localedef for locale generation
 # set "0" for qemu emulation of native localedef for locale generation

2.18.0

---
 .../packagegroups/packagegroup-core-tools-testapps.bb        | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

index a35f15e6ce..643219852d 100644
@@ -17,6 +17,9 @@ KEXECTOOLS_microblaze ?= ""
 KEXECTOOLS_nios2 ?= ""
 KEXECTOOLS_riscv64 ?= ""
 
+GSTEXAMPLES ?= "gst-examples"
+GSTEXAMPLES_riscv64 = ""
+
 X11GLTOOLS = "\
     mesa-demos \
     "

@@ -26,7 +29,7 @@ X11GLTOOLS = "\
     "
 
 X11TOOLS = "\
-    gst-examples \
+    ${GSTEXAMPLES} \
     x11perf \
     xrestop \
     xwininfo \

2.18.0

---
 meta/recipes-devtools/tcf-agent/tcf-agent_git.bb | 1 +
 1 file changed, 1 insertion(+)

index 20cb9a844d..d901c18c39 100644
@@ -45,6 +45,7 @@ CFLAGS_append_mips = " ${LCL_STOP_SERVICES}"
 CFLAGS_append_mips64 = " ${LCL_STOP_SERVICES}"
 CFLAGS_append_libc-musl = " ${LCL_STOP_SERVICES}"
 CFLAGS_append_powerpc64 = " ${LCL_STOP_SERVICES}"
+CFLAGS_append_riscv64 = " ${LCL_STOP_SERVICES}"
 
 do_install() {
 	oe_runmake install INSTALLROOT=${D}

2.18.0

This is needed for riscv64 to build target binutils fixes alternatives creation

---
 meta/recipes-devtools/binutils/binutils.inc | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

index 37813dd864..b56fcd4cd0 100644
@@ -38,6 +38,9 @@ FILES_${PN}-dev = " \
 # Rather than duplicating multiple entries for these, make one
 # list and reuse it.
 
+LDGOLD ?= "ld.gold dwp"
+LDGOLD_riscv64 = ""
+
 USE_ALTERNATIVES_FOR = " \
 	addr2line \
 	ar \

@@ -47,7 +50,7 @@ USE_ALTERNATIVES_FOR = " \
 	gprof \
 	ld \
 	ld.bfd \
-	ld.gold dwp \
+	${LDGOLD} \
 	nm \
 	objcopy \
 	objdump \

2.18.0

This is a (partial) revert of commit 988aad01b2 (sysklogd: don't use update-alternatives). Can you come to an agreement regarding which is the correct solution?

//Peter

I think the previous commit (syslogd: don't use update-alternatives) is made because syslog daemon conflict with each other. I guess the author assumed that the 'syslogd' and 'klogd' alternatives entries are handled by busybox-syslog package.

On the other hand, I think the patch is trying to solve the problem of busybox being installed after sysklogd.
We are currently not seeing errors because busybox is likely to be installed before sysklogd. Even in this situation, the result is not all correct, because the links busybox's postinstall creates are covered by the real binaries from sysklogd.

I think the problem is about busybox's handling of alternatives.

Khem, I've sent out a patch to fix busybox's alternatives logic. Could you please help review it?

Best Regards,
Chen Qi

Hi,

IIRC, the problem was that all the syslog packages were providing ${sysconfdir}/init.d/syslog which caused problems. And I don't think that has changed.

Why would you want to install two syslog daemons on the system? You should be able to install busybox after syslog as busybox-syslog is packaged in separate binary rpm.

Cheers,
Markus

On 30/08/2018, 12.41, "ChenQi" <[email protected]> wrote:

    On 08/30/2018 03:44 PM, Peter Kjellerstedt wrote:
    >> -----Original Message-----
    >> From: [email protected] <openembedded-
    >> [email protected]> On Behalf Of Khem Raj
    >> Sent: den 30 augusti 2018 05:56
    >> To: [email protected]
    >> Subject: [OE-core] [PATCH 3/7] sysklogd: Use update-alternatives
    >>
    >> busybox also provides klogd and syslogd, this change makes it coexist
    >> peacefully. Currently rootfs fails in situations where both of them are
    >> providing these binaries and busybox postinsts fail
    >>
    >> update-alternatives: Error: not linking
    >> /mnt/a/oe/build/tmp/work/qemuriscv64-bec-linux/core-image-full-
    >> cmdline/1.0-r0/rootfs/sbin/klogd
    >> to /bin/busybox.nosuid since
    >> /mnt/a/oe/build/tmp/work/qemuriscv64-bec-linux/core-image-full-
    >> cmdline/1.0-r0/rootfs/sbin/klogd
    >> exists and is not a link
    >>
    >> Signed-off-by: Khem Raj <[email protected]>
    >> ---
    >>   meta/recipes-extended/sysklogd/sysklogd.inc | 8 +++++++-
    >>   1 file changed, 7 insertions(+), 1 deletion(-)
    >>
    >> diff --git a/meta/recipes-extended/sysklogd/sysklogd.inc
    >> b/meta/recipes-extended/sysklogd/sysklogd.inc
    >> index fc4e67c18e..2a8bed00f3 100644
    >> --- a/meta/recipes-extended/sysklogd/sysklogd.inc
    >> +++ b/meta/recipes-extended/sysklogd/sysklogd.inc
    >> @@ -11,7 +11,7 @@ LIC_FILES_CHKSUM = "file://COPYING;md5=8ca43cbc842c2336e835926c2166c28b \
    >>                       file://klogd.c;beginline=2;endline=19;md5=7e87ed0ae6142de079bce738c10c899d \
    >>                      "
    >>
    >> -inherit update-rc.d systemd
    >> +inherit update-rc.d systemd update-alternatives
    >>
    >>   SRC_URI = "http://www.infodrom.org/projects/sysklogd/download/sysklogd-${PV}.tar.gz \
    >>              file://no-strip-install.patch \
    >> @@ -70,3 +70,9 @@ python () {
    >>       if not bb.utils.contains('DISTRO_FEATURES', 'sysvinit', True, False, d):
    >>           d.setVar("INHIBIT_UPDATERCD_BBCLASS", "1")
    >>   }
    >> +
    >> +ALTERNATIVE_PRIORITY = "100"
    >> +ALTERNATIVE_${PN} = "klogd syslogd"
    >> +ALTERNATIVE_LINK_NAME[klogd] = "${base_sbindir}/klogd"
    >> +ALTERNATIVE_LINK_NAME[syslogd] = "${base_sbindir}/syslogd"
    >> +
    >> --
    >> 2.18.0
    > This is a (partial) revert of commit 988aad01b2 (sysklogd: don't use
    > update-alternatives). Can you come to an agreement regarding which is
    > the correct solution?
    >
    > //Peter
    >
    
    
    I think the previous commit (syslogd: don't use update-alternatives) is 
    made because syslog daemon conflict with each other. I guess the author 
    assumed that the 'syslogd' and 'klogd' alternatives entries are handled 
    by busybox-syslog package.
    
    On the other hand, I think the patch is trying to solve the problem of 
    busybox being installed after sysklogd.
    We are currently not seeing errors because busybox is likely to be 
    installed before sysklogd. Even in this situation, the result is not all 
    correct, because the links busybox's postinstall creates are covered by 
    the real binaries from sysklogd.
    
    I think the problem is about busybox's handling of alternatives.
    
    Khem, I've sent out a patch to fix busybox's alternatives logic. Could 
    you please help review it?
    
    Best Regards,
    Chen Qi

thats fine, we need to have a RCONFLICTS statement to catch these issues explicitly during packaging.

Hmm,

There should be rconflicts in place, in both way – i.e. busybox-syslog rconflicts with sysklogd and the other way around. Isn't there(?)
- Markus

On 03/09/2018, 19.34, "Khem Raj" <[email protected]> wrote:

    On Sun, Sep 2, 2018 at 11:49 PM Markus Lehtonen
    <[email protected]> wrote:
    >
    > Hi,
    >
    > IIRC, the problem was that all the syslog packages were providing ${sysconfdir}/init.d/syslog which caused problems. And I don't think that has changed.
    >
    > Why would you want to install two syslog daemons on the system? You should be able to install busybox after syslog as busybox-syslog is packaged in separate binary rpm.
    >
    
    thats fine, we need to have a RCONFLICTS statement to catch these
    issues explicitly during packaging.
    
    > Cheers,
    >   Markus
    >
    > On 30/08/2018, 12.41, "ChenQi" <[email protected]> wrote:
    >
    >     On 08/30/2018 03:44 PM, Peter Kjellerstedt wrote:
    >     >> -----Original Message-----
    >     >> From: [email protected] <openembedded-
    >     >> [email protected]> On Behalf Of Khem Raj
    >     >> Sent: den 30 augusti 2018 05:56
    >     >> To: [email protected]
    >     >> Subject: [OE-core] [PATCH 3/7] sysklogd: Use update-alternatives
    >     >>
    >     >> busybox also provides klogd and syslogd, this change makes it coexist
    >     >> peacefully. Currently rootfs fails in situations where both of them are
    >     >> providing these binaries and busybox postinsts fail
    >     >>
    >     >> update-alternatives: Error: not linking
    >     >> /mnt/a/oe/build/tmp/work/qemuriscv64-bec-linux/core-image-full-
    >     >> cmdline/1.0-r0/rootfs/sbin/klogd
    >     >> to /bin/busybox.nosuid since
    >     >> /mnt/a/oe/build/tmp/work/qemuriscv64-bec-linux/core-image-full-
    >     >> cmdline/1.0-r0/rootfs/sbin/klogd
    >     >> exists and is not a link
    >     >>
    >     >> Signed-off-by: Khem Raj <[email protected]>
    >     >> ---
    >     >>   meta/recipes-extended/sysklogd/sysklogd.inc | 8 +++++++-
    >     >>   1 file changed, 7 insertions(+), 1 deletion(-)
    >     >>
    >     >> diff --git a/meta/recipes-extended/sysklogd/sysklogd.inc
    >     >> b/meta/recipes-extended/sysklogd/sysklogd.inc
    >     >> index fc4e67c18e..2a8bed00f3 100644
    >     >> --- a/meta/recipes-extended/sysklogd/sysklogd.inc
    >     >> +++ b/meta/recipes-extended/sysklogd/sysklogd.inc
    >     >> @@ -11,7 +11,7 @@ LIC_FILES_CHKSUM = "file://COPYING;md5=8ca43cbc842c2336e835926c2166c28b \
    >     >>                       file://klogd.c;beginline=2;endline=19;md5=7e87ed0ae6142de079bce738c10c899d \
    >     >>                      "
    >     >>
    >     >> -inherit update-rc.d systemd
    >     >> +inherit update-rc.d systemd update-alternatives
    >     >>
    >     >>   SRC_URI = "http://www.infodrom.org/projects/sysklogd/download/sysklogd-${PV}.tar.gz \
    >     >>              file://no-strip-install.patch \
    >     >> @@ -70,3 +70,9 @@ python () {
    >     >>       if not bb.utils.contains('DISTRO_FEATURES', 'sysvinit', True, False, d):
    >     >>           d.setVar("INHIBIT_UPDATERCD_BBCLASS", "1")
    >     >>   }
    >     >> +
    >     >> +ALTERNATIVE_PRIORITY = "100"
    >     >> +ALTERNATIVE_${PN} = "klogd syslogd"
    >     >> +ALTERNATIVE_LINK_NAME[klogd] = "${base_sbindir}/klogd"
    >     >> +ALTERNATIVE_LINK_NAME[syslogd] = "${base_sbindir}/syslogd"
    >     >> +
    >     >> --
    >     >> 2.18.0
    >     > This is a (partial) revert of commit 988aad01b2 (sysklogd: don't use
    >     > update-alternatives). Can you come to an agreement regarding which is
    >     > the correct solution?
    >     >
    >     > //Peter
    >     >
    >
    >
    >     I think the previous commit (syslogd: don't use update-alternatives) is
    >     made because syslog daemon conflict with each other. I guess the author
    >     assumed that the 'syslogd' and 'klogd' alternatives entries are handled
    >     by busybox-syslog package.
    >
    >     On the other hand, I think the patch is trying to solve the problem of
    >     busybox being installed after sysklogd.
    >     We are currently not seeing errors because busybox is likely to be
    >     installed before sysklogd. Even in this situation, the result is not all
    >     correct, because the links busybox's postinstall creates are covered by
    >     the real binaries from sysklogd.
    >
    >     I think the problem is about busybox's handling of alternatives.
    >
    >     Khem, I've sent out a patch to fix busybox's alternatives logic. Could
    >     you please help review it?
    >
    >     Best Regards,
    >     Chen Qi
    >
    >
    >