Fix cross for tagaspi

m/fox/configuration.nix

@@ -93,4 +93,20 @@
     wantedBy = [ "multi-user.target" ];
     serviceConfig.ExecStart = script;
   };
+
+  # Only allow SSH connections from users who have a SLURM allocation
+  # See: https://slurm.schedmd.com/pam_slurm_adopt.html
+  security.pam.services.sshd.rules.account.slurm = {
+    control = "required";
+    enable = true;
+    modulePath = "${pkgs.slurm}/lib/security/pam_slurm_adopt.so";
+    args = [ "log_level=debug5" ];
+    order = 999999; # Make it last one
+  };
+
+  # Disable systemd session (pam_systemd.so) as it will conflict with the
+  # pam_slurm_adopt.so module. What happens is that the shell is first adopted
+  # into the slurmstepd task and then into the systemd session, which is not
+  # what we want, otherwise it will linger even if all jobs are gone.
+  security.pam.services.sshd.startSession = lib.mkForce false;
 }

m/module/slurm-client.nix

@@ -1,4 +1,4 @@
-{ lib, pkgs, ... }:
+{ lib, ... }:
 
 {
   imports = [
@@ -21,20 +21,4 @@
   };
 
   services.slurm.client.enable = true;
-
-  # Only allow SSH connections from users who have a SLURM allocation
-  # See: https://slurm.schedmd.com/pam_slurm_adopt.html
-  security.pam.services.sshd.rules.account.slurm = {
-    control = "required";
-    enable = true;
-    modulePath = "${pkgs.slurm}/lib/security/pam_slurm_adopt.so";
-    args = [ "log_level=debug5" ];
-    order = 999999; # Make it last one
-  };
-
-  # Disable systemd session (pam_systemd.so) as it will conflict with the
-  # pam_slurm_adopt.so module. What happens is that the shell is first adopted
-  # into the slurmstepd task and then into the systemd session, which is not
-  # what we want, otherwise it will linger even if all jobs are gone.
-  security.pam.services.sshd.startSession = lib.mkForce false;
 }

m/module/slurm-common.nix

@@ -1,6 +1,31 @@
 { config, pkgs, ... }:
 
-{
+let
+  suspendProgram = pkgs.writeShellScript "suspend.sh" ''
+    exec 1>>/var/log/power_save.log 2>>/var/log/power_save.log
+    set -x
+    export "PATH=/run/current-system/sw/bin:$PATH"
+    echo "$(date) Suspend invoked $0 $*" >> /var/log/power_save.log
+    hosts=$(scontrol show hostnames $1)
+    for host in $hosts; do
+      echo Shutting down host: $host
+      ipmitool -I lanplus -H ''${host}-ipmi -P "" -U "" chassis power off
+    done
+  '';
+
+  resumeProgram = pkgs.writeShellScript "resume.sh" ''
+    exec 1>>/var/log/power_save.log 2>>/var/log/power_save.log
+    set -x
+    export "PATH=/run/current-system/sw/bin:$PATH"
+    echo "$(date) Suspend invoked $0 $*" >> /var/log/power_save.log
+    hosts=$(scontrol show hostnames $1)
+    for host in $hosts; do
+      echo Starting host: $host
+      ipmitool -I lanplus -H ''${host}-ipmi -P "" -U "" chassis power on
+    done
+  '';
+
+in {
   services.slurm = {
     controlMachine = "apex";
     clusterName = "jungle";
@@ -34,6 +59,16 @@
       # the resources. Use the task/cgroup plugin to enable process containment.
       TaskPlugin=task/affinity,task/cgroup
 
+      # Power off unused nodes until they are requested
+      SuspendProgram=${suspendProgram}
+      SuspendTimeout=60
+      ResumeProgram=${resumeProgram}
+      ResumeTimeout=300
+      SuspendExcNodes=fox
+
+      # Turn the nodes off after 1 hour of inactivity
+      SuspendTime=3600
+
       # Reduce port range so we can allow only this range in the firewall
       SrunPortRange=60000-61000
 

overlay.nix

@@ -101,16 +101,14 @@ let
   pkgsTopLevel = filterAttrs (_: isDerivation) bscPkgs;
 
   # Native build in that platform doesn't imply cross build works
-  canCrossCompile = platform: default: pkg:
+  canCrossCompile = platform: pkg:
     (isDerivation pkg) &&
-    # If meta.cross is undefined, use default
-    (pkg.meta.cross or default) &&
-    (meta.availableOn final.pkgsCross.${platform}.stdenv.hostPlatform pkg);
+    # Must be defined explicitly
+    (pkg.meta.cross or false) &&
+    (meta.availableOn platform pkg);
 
   # For now only RISC-V
-  crossSet = genAttrs [ "riscv64" ] (platform:
-    filterAttrs (_: canCrossCompile platform true)
-      final.pkgsCross.${platform}.bsc.pkgsTopLevel);
+  crossSet = { riscv64 = final.pkgsCross.riscv64.bsc.pkgsTopLevel; };
 
   buildList = name: paths:
     final.runCommandLocal name { } ''
@@ -130,7 +128,7 @@ let
   # For now only RISC-V
   crossList = buildList "ci-cross"
     (filter
-      (canCrossCompile "riscv64" false) # opt-in (pkgs with: meta.cross = true)
+      (canCrossCompile final.pkgsCross.riscv64.stdenv.hostPlatform)
         (builtins.attrValues crossSet.riscv64));
 
 in bscPkgs // {

pkgs/amd-uprof/default.nix

@@ -90,7 +90,7 @@ in
     meta = {
       description = "Performance analysis tool-suite for x86 based applications";
       homepage = "https://www.amd.com/es/developer/uprof.html";
-      platforms = [ "x86_64-linux" ];
+      platforms = lib.platforms.linux;
       license = lib.licenses.unfree;
       maintainers = with lib.maintainers.bsc; [ rarias varcila ];
     };

pkgs/cudainfo/default.nix

@@ -1,6 +1,5 @@
 {
   stdenv
-, lib
 , cudatoolkit
 , cudaPackages
 , autoAddDriverRunpath
@@ -41,9 +40,4 @@ stdenv.mkDerivation (finalAttrs: {
     '';
     installPhase = "touch $out";
   };
-
-  meta = {
-    platforms = [ "x86_64-linux" ];
-    maintainers = with lib.maintainers.bsc; [ rarias ];
-  };
 })

pkgs/gpi-2/default.nix

@@ -9,7 +9,6 @@
 , automake
 , libtool
 , mpi
-, rsync
 , gfortran
 }:
 
@@ -50,7 +49,18 @@ stdenv.mkDerivation rec {
     "CXXFLAGS=-fPIC"
   ];
 
-  buildInputs = [ slurm mpiAll rdma-core-all autoconf automake libtool rsync gfortran ];
+  nativeBuildInputs = [
+    autoconf
+    automake
+    gfortran
+    libtool
+  ];
+
+  buildInputs = [
+    slurm
+    mpiAll
+    rdma-core-all
+  ];
 
   hardeningDisable = [ "all" ];
 

pkgs/mpich/default.nix

@@ -6,6 +6,12 @@
 , pmix
 , gfortran
 , symlinkJoin
+# Disabled when cross-compiling
+# To fix cross compilation, we should fill the values in:
+# https://github.com/pmodels/mpich/blob/main/maint/fcrosscompile/cross_values.txt.in
+# For each arch
+, enableFortran ? stdenv.hostPlatform == stdenv.buildPlatform
+, perl
 }:
 
 let
@@ -15,10 +21,13 @@ let
     paths = [ pmix.dev pmix.out ];
   };
 in mpich.overrideAttrs (old: {
-  buildInput = old.buildInputs ++ [
+  buildInputs = old.buildInputs ++ [
     libfabric
     pmixAll
   ];
+  nativeBuildInputs = old.nativeBuildInputs ++ [
+    perl
+  ];
   configureFlags = [
     "--enable-shared"
     "--enable-sharedlib"
@@ -31,7 +40,18 @@ in mpich.overrideAttrs (old: {
   ] ++ lib.optionals (lib.versionAtLeast gfortran.version "10") [
     "FFLAGS=-fallow-argument-mismatch" # https://github.com/pmodels/mpich/issues/4300
     "FCFLAGS=-fallow-argument-mismatch"
+  ] ++ lib.optionals (!enableFortran) [
+    "--disable-fortran"
   ];
+
+  preFixup = ''
+    # Ensure the default compilers are the ones mpich was built with
+    sed -i 's:CC="gcc":CC=${stdenv.cc}/bin/gcc:' $out/bin/mpicc
+    sed -i 's:CXX="g++":CXX=${stdenv.cc}/bin/g++:' $out/bin/mpicxx
+  '' + lib.optionalString enableFortran ''
+    sed -i 's:FC="gfortran":FC=${gfortran}/bin/gfortran:' $out/bin/mpifort
+  '';
+
   hardeningDisable = [ "all" ];
 
   meta = old.meta // {