numa: add support for preferred-many and weighted-interleave policies

Add support for two newer NUMA memory policies:

- MPOL_PREFERRED_MANY (Linux 5.15): like MPOL_PREFERRED but accepts
  a set of nodes instead of a single node, falling back to all nodes
  if preferred nodes cannot satisfy the allocation.

- MPOL_WEIGHTED_INTERLEAVE (Linux 6.9): like MPOL_INTERLEAVE but
  distributes pages across nodes proportionally to per-node weights
  configured via /sys/kernel/mm/mempolicy/weighted_interleave/.

On kernels that do not support the requested policy, set_mempolicy()
returns EINVAL. We convert EINVAL to EOPNOTSUPP only for the two new
policies (MPOL_PREFERRED_MANY, MPOL_WEIGHTED_INTERLEAVE), so that a
bad NUMAMask= for already-supported policies still fails the service
rather than being silently ignored.

The NUMA subsystem being absent (ENOSYS) continues to be handled
silently at debug level, as before.

Varlink serialization uses json_underscorify() on an owned copy of
the policy name string to convert hyphenated names to the underscore
form declared in the IDL enum, avoiding mutation of the read-only
static string table.

Signed-off-by: dongshengyuan <dongshengyuan@uniontech.com>

README

@@ -67,13 +67,14 @@ REQUIREMENTS:
            of systemd. Taint flag 'old-kernel' will be set. systemd will most likely
            still function, but upstream support and testing are limited.
 
-        Linux kernel ≥ 6.3 for MFD_EXEC/MFD_NOEXEC_SEAL and tmpfs noswap option
+        Linux kernel ≥ 5.15 for MPOL_PREFERRED_MANY NUMA policy
+                     ≥ 6.3 for MFD_EXEC/MFD_NOEXEC_SEAL and tmpfs noswap option
                      ≥ 6.5 for name_to_handle_at() AT_HANDLE_FID, SO_PEERPIDFD/SO_PASSPIDFD,
                                and MOVE_MOUNT_BENEATH
                      ≥ 6.6 for quota support on tmpfs
                      ≥ 6.7 for cgroup2fs memory_hugetlb_accounting option
                      ≥ 6.8 for STATX_MNT_ID_UNIQUE
-                     ≥ 6.9 for pidfs
+                     ≥ 6.9 for pidfs and MPOL_WEIGHTED_INTERLEAVE NUMA policy
                      ≥ 6.10 for fcntl(F_DUPFD_QUERY), unprivileged linkat(AT_EMPTY_PATH),
                                 and block device 'partscan' sysfs attribute
                      ≥ 6.12 for AT_HANDLE_MNT_ID_UNIQUE

man/systemd.exec.xml

@@ -1408,12 +1408,16 @@ CapabilityBoundingSet=~CAP_B CAP_C</programlisting>
         <term><varname>NUMAPolicy=</varname></term>
 
         <listitem><para>Controls the NUMA memory policy of the executed processes. Takes a policy type, one of:
-        <option>default</option>, <option>preferred</option>, <option>bind</option>, <option>interleave</option> and
-        <option>local</option>. A list of NUMA nodes that should be associated with the policy must be specified
-        in <varname>NUMAMask=</varname>. For more details on each policy please see,
+        <option>default</option>, <option>preferred</option>, <option>bind</option>, <option>interleave</option>,
+        <option>local</option>, <option>preferred-many</option> (requires Linux 5.15 or newer) and
+        <option>weighted-interleave</option> (requires Linux 6.9 or newer, weights are configured via
+        <filename>/sys/kernel/mm/mempolicy/weighted_interleave/</filename>). A list of NUMA nodes that should be
+        associated with the policy must be specified in <varname>NUMAMask=</varname>. For more details on each
+        policy please see,
         <citerefentry><refentrytitle>set_mempolicy</refentrytitle><manvolnum>2</manvolnum></citerefentry>. For overall
         overview of NUMA support in Linux see,
         <citerefentry project='man-pages'><refentrytitle>numa</refentrytitle><manvolnum>7</manvolnum></citerefentry>.
+        If the kernel does not support the requested policy, a warning is logged and the setting is ignored.
         </para>
 
         <xi:include href="version-info.xml" xpointer="v243"/></listitem>

src/core/exec-invoke.c

@@ -5646,8 +5646,10 @@ int exec_invoke(
 
         if (mpol_is_valid(numa_policy_get_type(&context->numa_policy))) {
                 r = apply_numa_policy(&context->numa_policy);
-                if (ERRNO_IS_NEG_NOT_SUPPORTED(r))
+                if (r == -ENOSYS)
                         log_debug_errno(r, "NUMA support not available, ignoring.");
+                else if (ERRNO_IS_NEG_NOT_SUPPORTED(r))
+                        log_warning_errno(r, "NUMA policy not supported by kernel, ignoring.");
                 else if (r < 0) {
                         *exit_status = EXIT_NUMA_POLICY;
                         return log_error_errno(r, "Failed to set NUMA memory policy: %m");

src/core/main.c

@@ -2029,8 +2029,10 @@ static void update_numa_policy(bool skip_setup) {
         }
 
         r = apply_numa_policy(&arg_numa_policy);
-        if (r == -EOPNOTSUPP)
+        if (r == -ENOSYS)
                 log_debug_errno(r, "NUMA support not available, ignoring.");
+        else if (ERRNO_IS_NEG_NOT_SUPPORTED(r))
+                log_warning_errno(r, "NUMA policy not supported by kernel, ignoring.");
         else if (r < 0)
                 log_warning_errno(r, "Failed to set NUMA memory policy, ignoring: %m");
 }

src/core/varlink-execute.c

@@ -315,7 +315,11 @@ static int numa_policy_build_json(sd_json_variant **ret, const char *name, void
                 return 0;
         }
 
-        return sd_json_variant_new_string(ret, mpol_to_string(t));
+        _cleanup_free_ char *s = strdup(mpol_to_string(t));
+        if (!s)
+                return -ENOMEM;
+
+        return sd_json_variant_new_string(ret, json_underscorify(s));
 }
 
 static int numa_mask_build_json(sd_json_variant **ret, const char *name, void *userdata) {

src/shared/numa-util.c

@@ -73,7 +73,11 @@ int apply_numa_policy(const NUMAPolicy *policy) {
         assert(policy);
 
         if (get_mempolicy(NULL, NULL, 0, NULL, 0) < 0 && errno == ENOSYS)
-                return -EOPNOTSUPP;
+                /* NUMA syscall interface not available (kernel compiled without NUMA support).
+                 * Return -ENOSYS so callers can distinguish this from -EOPNOTSUPP, which we
+                 * return below when the syscall interface exists but the requested policy is
+                 * not supported by this kernel version. */
+                return -ENOSYS;
 
         if (!numa_policy_is_valid(policy))
                 return -EINVAL;
@@ -83,8 +87,14 @@ int apply_numa_policy(const NUMAPolicy *policy) {
                 return r;
 
         r = set_mempolicy(numa_policy_get_type(policy), nodes, maxnode);
-        if (r < 0)
+        if (r < 0) {
+                // FIXME: This compatibility code path shall be removed once kernel 6.9
+                //        becomes the new minimal baseline (MPOL_WEIGHTED_INTERLEAVE).
+                if (errno == EINVAL && IN_SET(numa_policy_get_type(policy),
+                                             MPOL_PREFERRED_MANY, MPOL_WEIGHTED_INTERLEAVE))
+                        return -EOPNOTSUPP;
                 return -errno;
+        }
 
         return 0;
 }
@@ -241,11 +251,13 @@ int numa_mask_add_all(CPUSet *mask) {
 }
 
 static const char* const mpol_table[] = {
-        [MPOL_DEFAULT]    = "default",
-        [MPOL_PREFERRED]  = "preferred",
-        [MPOL_BIND]       = "bind",
-        [MPOL_INTERLEAVE] = "interleave",
-        [MPOL_LOCAL]      = "local",
+        [MPOL_DEFAULT]             = "default",
+        [MPOL_PREFERRED]           = "preferred",
+        [MPOL_BIND]                = "bind",
+        [MPOL_INTERLEAVE]          = "interleave",
+        [MPOL_LOCAL]               = "local",
+        [MPOL_PREFERRED_MANY]      = "preferred-many",
+        [MPOL_WEIGHTED_INTERLEAVE] = "weighted-interleave",
 };
 
 DEFINE_STRING_TABLE_LOOKUP(mpol, int);

src/shared/numa-util.h

@@ -7,7 +7,7 @@
 #include "shared-forward.h"
 
 static inline bool mpol_is_valid(int t) {
-        return t >= MPOL_DEFAULT && t <= MPOL_LOCAL;
+        return t >= MPOL_DEFAULT && t <= MPOL_WEIGHTED_INTERLEAVE;
 }
 
 typedef struct NUMAPolicy {

src/shared/varlink-io.systemd.Unit.c

@@ -173,7 +173,9 @@ SD_VARLINK_DEFINE_ENUM_TYPE(
                 SD_VARLINK_DEFINE_ENUM_VALUE(preferred),
                 SD_VARLINK_DEFINE_ENUM_VALUE(bind),
                 SD_VARLINK_DEFINE_ENUM_VALUE(interleave),
-                SD_VARLINK_DEFINE_ENUM_VALUE(local));
+                SD_VARLINK_DEFINE_ENUM_VALUE(local),
+                SD_VARLINK_DEFINE_ENUM_VALUE(preferred_many),
+                SD_VARLINK_DEFINE_ENUM_VALUE(weighted_interleave));
 
 SD_VARLINK_DEFINE_ENUM_TYPE(
                 MountPropagationFlag,

test/units/TEST-36-NUMAPOLICY.sh

@@ -227,6 +227,26 @@ else
     pid1ReloadWithStrace
     grep -E "set_mempolicy\((MPOL_LOCAL|0x4 [^,]*), NULL" "$straceLog"
 
+    echo "PID1 NUMAPolicy support - Preferred-many policy w/o mask"
+    writePID1NUMAPolicy "preferred-many"
+    pid1ReloadWithJournal
+    grep "Failed to set NUMA memory policy, ignoring: Invalid argument" "$journalLog"
+
+    echo "PID1 NUMAPolicy support - Preferred-many policy w/ mask"
+    writePID1NUMAPolicy "preferred-many" "0"
+    pid1ReloadWithStrace
+    grep -E "set_mempolicy\((MPOL_PREFERRED_MANY|0x5 [^,]*), \[0x0*1\]" "$straceLog"
+
+    echo "PID1 NUMAPolicy support - Weighted-interleave policy w/o mask"
+    writePID1NUMAPolicy "weighted-interleave"
+    pid1ReloadWithJournal
+    grep "Failed to set NUMA memory policy, ignoring: Invalid argument" "$journalLog"
+
+    echo "PID1 NUMAPolicy support - Weighted-interleave policy w/ mask"
+    writePID1NUMAPolicy "weighted-interleave" "0"
+    pid1ReloadWithStrace
+    grep -E "set_mempolicy\((MPOL_WEIGHTED_INTERLEAVE|0x6 [^,]*), \[0x0*1\]" "$straceLog"
+
     echo "Unit file NUMAPolicy support - Default policy w/o mask"
     writeTestUnitNUMAPolicy "default"
     pid1StartUnitWithStrace "$testUnit"
@@ -297,6 +317,34 @@ else
     # Mask must be ignored
     grep -E "set_mempolicy\((MPOL_LOCAL|0x4 [^,]*), NULL" "$straceLog"
 
+    echo "Unit file NUMAPolicy support - Preferred-many policy w/o mask"
+    writeTestUnitNUMAPolicy "preferred-many"
+    pid1StartUnitWithStrace "$testUnit"
+    pid1StopUnit "$testUnit"
+    [[ $(systemctl show "$testUnit" -P ExecMainStatus) == "242" ]]
+
+    echo "Unit file NUMAPolicy support - Preferred-many policy w/ mask"
+    writeTestUnitNUMAPolicy "preferred-many" "0"
+    pid1StartUnitWithStrace "$testUnit"
+    systemctlCheckNUMAProperties "$testUnit" "preferred-many" "0"
+    varlinkctl call /run/systemd/io.systemd.Manager io.systemd.Unit.List "{\"name\":\"$testUnit\"}" | jq -e '.context.Exec.NUMAPolicy == "preferred_many"'
+    pid1StopUnit "$testUnit"
+    grep -E "set_mempolicy\((MPOL_PREFERRED_MANY|0x5 [^,]*), \[0x0*1\]" "$straceLog"
+
+    echo "Unit file NUMAPolicy support - Weighted-interleave policy w/o mask"
+    writeTestUnitNUMAPolicy "weighted-interleave"
+    pid1StartUnitWithStrace "$testUnit"
+    pid1StopUnit "$testUnit"
+    [[ $(systemctl show "$testUnit" -P ExecMainStatus) == "242" ]]
+
+    echo "Unit file NUMAPolicy support - Weighted-interleave policy w/ mask"
+    writeTestUnitNUMAPolicy "weighted-interleave" "0"
+    pid1StartUnitWithStrace "$testUnit"
+    systemctlCheckNUMAProperties "$testUnit" "weighted-interleave" "0"
+    varlinkctl call /run/systemd/io.systemd.Manager io.systemd.Unit.List "{\"name\":\"$testUnit\"}" | jq -e '.context.Exec.NUMAPolicy == "weighted_interleave"'
+    pid1StopUnit "$testUnit"
+    grep -E "set_mempolicy\((MPOL_WEIGHTED_INTERLEAVE|0x6 [^,]*), \[0x0*1\]" "$straceLog"
+
     echo "Unit file CPUAffinity=NUMA support"
     writeTestUnitNUMAPolicy "bind" "0"
     echo "CPUAffinity=numa" >>"$testUnitNUMAConf"
@@ -343,6 +391,14 @@ else
     systemctlCheckNUMAProperties "$runUnit" "local" ""
     systemctl cat "$runUnit" | grep 'CPUAffinity=numa' >/dev/null
     pid1StopUnit "$runUnit"
+
+    systemd-run -p NUMAPolicy=preferred-many -p NUMAMask=0 --unit "$runUnit" sleep 1000
+    systemctlCheckNUMAProperties "$runUnit" "preferred-many" "0"
+    pid1StopUnit "$runUnit"
+
+    systemd-run -p NUMAPolicy=weighted-interleave -p NUMAMask=0 --unit "$runUnit" sleep 1000
+    systemctlCheckNUMAProperties "$runUnit" "weighted-interleave" "0"
+    pid1StopUnit "$runUnit"
 fi
 
 # Cleanup