diff mbox series

[v2,2/7] Overhaul how Argo is built and packged

Message ID 20250414101843.2348330-3-andrew.cooper3@citrix.com (mailing list archive)
State New
Headers show
Series Rootfs/argo cleanup and microcode support | expand

Commit Message

Andrew Cooper April 14, 2025, 10:18 a.m. UTC 
Right now, the argo artefacts are a pile of files which the test has to turn
back into something which resembles a filesystem.  Furthermore, because we do
not build modules for the main kernel, it is extra important to make sure that
xen-argo.ko doesn't get out of sync.

Build argo conditionally as part of the linux artefact.  It's ~100kb all
together, compared to ~14M for the kernel.

Produce a single argo.cpio.gz with xen-argo.ko in the standard location.
Prune userspace down to just the executables and libraries.

This is cribbed from the existing scripts/x86_64-linux-argo.sh, which stays in
place in the short term until Xen can be updated to use the new scheme.

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
---
CC: Anthony PERARD <anthony.perard@vates.tech>
CC: Stefano Stabellini <sstabellini@kernel.org>
CC: Michal Orzel <michal.orzel@amd.com>
CC: Doug Goldstein <cardoe@cardoe.com>
CC: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
CC: Jason Andryuk <jason.andryuk@amd.com>
CC: Daniel P. Smith <dpsmith@apertussolutions.com>

v2:
 * Only build conditionally.  Argo is bust with Linux 6.12, which is needed
   for new hardware runners.
 * Parallel build of xen-argo.ko and libargo.
 * Use -print0

I tried to make MODPOST work properly, but we don't build enough of it for the
kernel, and I didn't feel like adding an extra 10 mins to the build (all
modules) just to get the metadata right.
---
 .gitlab-ci.yml         |  2 ++
 scripts/build-argo.sh  | 67 ++++++++++++++++++++++++++++++++++++++++++
 scripts/build-linux.sh |  8 ++++-
 3 files changed, 76 insertions(+), 1 deletion(-)
 create mode 100755 scripts/build-argo.sh

Comments

Marek Marczykowski-Górecki April 14, 2025, 10:35 a.m. UTC | #1 
On Mon, Apr 14, 2025 at 11:18:38AM +0100, Andrew Cooper wrote:
> --- a/scripts/build-linux.sh
> +++ b/scripts/build-linux.sh
> @@ -8,7 +8,7 @@ fi
>  set -ex -o pipefail
>  
>  WORKDIR="${PWD}"
> -COPYDIR="${WORKDIR}/binaries/"
> +COPYDIR="${WORKDIR}/binaries"

Is this change intentional? It has worse failure mode if "binaries" dir
wouldn't exist for some reason...
Andrew Cooper April 14, 2025, 10:38 a.m. UTC | #2 
On 14/04/2025 11:35 am, Marek Marczykowski-Górecki wrote:
> On Mon, Apr 14, 2025 at 11:18:38AM +0100, Andrew Cooper wrote:
>> --- a/scripts/build-linux.sh
>> +++ b/scripts/build-linux.sh
>> @@ -8,7 +8,7 @@ fi
>>  set -ex -o pipefail
>>  
>>  WORKDIR="${PWD}"
>> -COPYDIR="${WORKDIR}/binaries/"
>> +COPYDIR="${WORKDIR}/binaries"
> Is this change intentional? It has worse failure mode if "binaries" dir
> wouldn't exist for some reason...

Yes it is intentional.  It causes problems when we derive new variables
from COPYDIR.

binaries/ always exists.  It's in the base repo.

~Andrew
Daniel P. Smith April 14, 2025, 10:55 a.m. UTC | #3 
On 4/14/25 06:18, Andrew Cooper wrote:
> Right now, the argo artefacts are a pile of files which the test has to turn
> back into something which resembles a filesystem.  Furthermore, because we do
> not build modules for the main kernel, it is extra important to make sure that
> xen-argo.ko doesn't get out of sync.
> 
> Build argo conditionally as part of the linux artefact.  It's ~100kb all
> together, compared to ~14M for the kernel.
> 
> Produce a single argo.cpio.gz with xen-argo.ko in the standard location.
> Prune userspace down to just the executables and libraries.
> 
> This is cribbed from the existing scripts/x86_64-linux-argo.sh, which stays in
> place in the short term until Xen can be updated to use the new scheme.
> 
> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>

Reviewed-by: Daniel P. Smith <dpsmith@apertussolutions.com>
Stefano Stabellini April 15, 2025, 11:48 p.m. UTC | #4 
On Mon, 14 Apr 2025, Andrew Cooper wrote:
> Right now, the argo artefacts are a pile of files which the test has to turn
> back into something which resembles a filesystem.  Furthermore, because we do
> not build modules for the main kernel, it is extra important to make sure that
> xen-argo.ko doesn't get out of sync.
> 
> Build argo conditionally as part of the linux artefact.  It's ~100kb all
> together, compared to ~14M for the kernel.
> 
> Produce a single argo.cpio.gz with xen-argo.ko in the standard location.
> Prune userspace down to just the executables and libraries.
> 
> This is cribbed from the existing scripts/x86_64-linux-argo.sh, which stays in
> place in the short term until Xen can be updated to use the new scheme.
> 
> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>

Do we need to remove x86_64-argo-linux.sh?
Andrew Cooper April 16, 2025, 10:15 a.m. UTC | #5 
On 16/04/2025 12:48 am, Stefano Stabellini wrote:
> On Mon, 14 Apr 2025, Andrew Cooper wrote:
>> Right now, the argo artefacts are a pile of files which the test has to turn
>> back into something which resembles a filesystem.  Furthermore, because we do
>> not build modules for the main kernel, it is extra important to make sure that
>> xen-argo.ko doesn't get out of sync.
>>
>> Build argo conditionally as part of the linux artefact.  It's ~100kb all
>> together, compared to ~14M for the kernel.
>>
>> Produce a single argo.cpio.gz with xen-argo.ko in the standard location.
>> Prune userspace down to just the executables and libraries.
>>
>> This is cribbed from the existing scripts/x86_64-linux-argo.sh, which stays in
>> place in the short term until Xen can be updated to use the new scheme.
>>
>> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
> Do we need to remove x86_64-argo-linux.sh?

Yes, but 4.20 still uses it.

One thing I'm not sure about is whether "keep latest artefact" is just
artefacts from the latest run, or the latest of each named artefact.

Now is as good a time to experiment as any.

~Andrew
Marek Marczykowski-Górecki April 16, 2025, 10:21 a.m. UTC | #6 
On Wed, Apr 16, 2025 at 11:15:21AM +0100, Andrew Cooper wrote:
> On 16/04/2025 12:48 am, Stefano Stabellini wrote:
> > On Mon, 14 Apr 2025, Andrew Cooper wrote:
> >> Right now, the argo artefacts are a pile of files which the test has to turn
> >> back into something which resembles a filesystem.  Furthermore, because we do
> >> not build modules for the main kernel, it is extra important to make sure that
> >> xen-argo.ko doesn't get out of sync.
> >>
> >> Build argo conditionally as part of the linux artefact.  It's ~100kb all
> >> together, compared to ~14M for the kernel.
> >>
> >> Produce a single argo.cpio.gz with xen-argo.ko in the standard location.
> >> Prune userspace down to just the executables and libraries.
> >>
> >> This is cribbed from the existing scripts/x86_64-linux-argo.sh, which stays in
> >> place in the short term until Xen can be updated to use the new scheme.
> >>
> >> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
> > Do we need to remove x86_64-argo-linux.sh?
> 
> Yes, but 4.20 still uses it.
> 
> One thing I'm not sure about is whether "keep latest artefact" is just
> artefacts from the latest run, or the latest of each named artefact.

I think it's latest run for a branch. If newer job (on a branch) doesn't
include an artifact anymore, it will be gone. Jobs referencing it won't
see it immediately, regardless of expire time.
diff mbox series

Patch

diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index fb997cc62162..790a6d9f9896 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -47,6 +47,8 @@  linux-6.6.56-x86_64:
   script: ./scripts/build-linux.sh
   variables:
     LINUX_VERSION: 6.6.56
+    ARGO_SHA: "705a7a8a624b42e13e655d3042059b8a85cdf6a3"
+    ARGOEXEC_SHA: "d900429f6640acc6f68a3d3a4c945d7da60625d8"
 
 #
 # The jobs below here are legacy and being phased out.
diff --git a/scripts/build-argo.sh b/scripts/build-argo.sh
new file mode 100755
index 000000000000..ef7057d847d4
--- /dev/null
+++ b/scripts/build-argo.sh
@@ -0,0 +1,67 @@ 
+#
+# This is a partial script, sourced by build-linux.sh
+# It has expectations about the environment
+#
+
+cd "${WORKDIR}"
+
+#
+# We're going to collect everything in argo.cpio.gz.  Construct it under
+# $ARGODIR as we go.
+#
+ARGODIR="${WORKDIR}/argo-root"
+
+git clone https://github.com/OpenXT/linux-xen-argo.git --depth=1
+git -C "${WORKDIR}/linux-xen-argo" fetch origin "${ARGO_SHA}"
+git -C "${WORKDIR}/linux-xen-argo" switch --detach FETCH_HEAD
+
+# Build xen-argo.ko against the target kernel, and install it.  Install
+# linux/argo.h too because userspace needs it.
+make -j$(nproc) -C "linux-${LINUX_VERSION}" \
+     M="${WORKDIR}/linux-xen-argo/argo-linux" \
+     KBUILD_MODPOST_WARN=1 \
+     CFLAGS_MODULE="-Wno-error" \
+     modules
+install -D -m644 "${WORKDIR}/linux-xen-argo/argo-linux/xen-argo.ko" \
+     "${ARGODIR}/lib/modules/${LINUX_VERSION}/updates/xen-argo.ko"
+install -D -m644 "${WORKDIR}/linux-xen-argo/argo-linux/include/linux/argo.h" \
+     "${ARGODIR}/usr/include/linux/argo.h"
+
+# Build and install libargo, applying fixes to build in Alpine Linux
+cd "${WORKDIR}/linux-xen-argo/libargo"
+sed -e "s|AM_INIT_AUTOMAKE|AC_CONFIG_AUX_DIR(.)\nAM_INIT_AUTOMAKE|" \
+    -i configure.ac
+sed -e "s/__SOCKADDR_COMMON (sxenargo_)/sa_family_t sxenargo_family/" \
+    -e "s/__SOCKADDR_COMMON_SIZE/(sizeof (unsigned short int))/" \
+    -i src/libargo.h
+
+autoreconf --install
+./configure --prefix=/usr CPPFLAGS="-I${PWD}/../argo-linux/include"
+make -j$(nproc)
+make install DESTDIR="${ARGODIR}"
+
+# Build and install argo-exec, modifying for xilinx argo test
+cd "${WORKDIR}"
+curl -fsSLO \
+    https://raw.githubusercontent.com/OpenXT/xenclient-oe/${ARGOEXEC_SHA}/recipes-openxt/argo-exec/argo-exec/argo-exec.c
+sed -e "/#include <xen\/xen.h>/d" \
+    -e "s|ret = shuffle(s, fds\[0\], fds\[1\]);|ret = shuffle(s, 0, 1);|" \
+    -i argo-exec.c
+
+gcc -I"${ARGODIR}/usr/include" -L"${ARGODIR}/usr/lib/" \
+    argo-exec.c -o "${ARGODIR}/usr/bin/argo-exec" -largo
+
+#
+# Building is now complete.  Strip the devel components and the nointerposer
+# lib, which we don't care to deploy to the test system.
+#
+cd $ARGODIR
+rm -r usr/include usr/lib/pkgconfig
+find usr/lib -name \*nointerposer\* -delete
+find usr/lib \( -name \*.a -o -name \*.so -o -name \*.la \) -delete
+
+# Package everything up
+find . -print0 | cpio -0 -R 0:0 -H newc -o | gzip > "$COPYDIR/argo.cpio.gz"
+
+# Print the contents for the build log
+zcat "${COPYDIR}/argo.cpio.gz" | cpio -tv
diff --git a/scripts/build-linux.sh b/scripts/build-linux.sh
index 652fdba7b9d1..441b8721a490 100755
--- a/scripts/build-linux.sh
+++ b/scripts/build-linux.sh
@@ -8,7 +8,7 @@  fi
 set -ex -o pipefail
 
 WORKDIR="${PWD}"
-COPYDIR="${WORKDIR}/binaries/"
+COPYDIR="${WORKDIR}/binaries"
 UNAME=$(uname -m)
 
 # Build Linux
@@ -45,6 +45,12 @@  case $UNAME in
     x86_64)
         make -j$(nproc) bzImage
         cp arch/x86/boot/bzImage "${COPYDIR}"
+
+        # Build argo if requested
+        if [[ -n "${ARGO_SHA}" ]]; then
+            make modules_prepare
+            . "${WORKDIR}/scripts/build-argo.sh"
+        fi
         ;;
 
     aarch64)