SymbX/systemd
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

udev/net: document IRQAffinityPolicy=, IRQAffinity=, and

Browse Source 
IRQAffinityNUMA=
This commit is contained in:
Quentin Deslandes
2026-05-20 11:39:11 +02:00
parent 7dfbbfb90d
commit 1662ab6683
1 changed files with 56 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
man/systemd.link.xml
+56
View File
@@ -1080,6 +1080,62 @@
<xi:include href="version-info.xml" xpointer="v256"/>
</listitem>
</varlistentry>
<varlistentry>
<term><varname>IRQAffinityPolicy=</varname></term>
<listitem>
<para>Specifies the IRQ distribution strategy for network interface MSI/MSI-X interrupts. Takes one
of <literal>spread</literal> or <literal>single</literal>. When set to <literal>spread</literal>,
queue IRQs are distributed across CPUs using a topology-aware maximum distance algorithm that
prefers CPUs on different NUMA nodes, then different physical cores, then different hyperthreads.
When set to <literal>single</literal>, all IRQs are pinned to a single CPU (CPU 0 by default, or
the first CPU in the set specified by <varname>IRQAffinity=</varname>). When there are more IRQs
than available CPUs, queues wrap around using round-robin assignment. When unset, no affinity
management is performed. Setting this to an empty string explicitly disables affinity management.</para>
<para>This option only applies to devices with MSI/MSI-X interrupts discoverable via
<filename>/sys/class/net/<replaceable>iface</replaceable>/device/msi_irqs/</filename>. Virtual
devices (veth, tap) and legacy INTx devices are skipped with a notice logged.</para>
<para>Note that if <citerefentry><refentrytitle>irqbalance</refentrytitle><manvolnum>1</manvolnum>
</citerefentry> or similar IRQ management daemons are running, they may override the configured
affinity. Consider disabling such daemons or configuring them to exclude managed interfaces.</para>
<xi:include href="version-info.xml" xpointer="v260"/>
</listitem>
</varlistentry>
<varlistentry>
<term><varname>IRQAffinity=</varname></term>
<listitem>
<para>Filters the set of CPUs eligible for IRQ placement. Takes a list of CPU indices or ranges
separated by either whitespace or commas (e.g., <literal>0-3</literal>, <literal>0,2,4,6</literal>,
<literal>0-3,8-11</literal>). This option works in conjunction with
<varname>IRQAffinityPolicy=</varname> and <varname>IRQAffinityNUMA=</varname> to constrain which
CPUs receive network IRQs. When specified with <literal>spread</literal> policy, only the listed
CPUs are considered for IRQ distribution. When specified with <literal>single</literal> policy,
IRQs are pinned to the first CPU in the allowed set instead of CPU 0. Has no effect if
<varname>IRQAffinityPolicy=</varname> is not set.</para>
<xi:include href="version-info.xml" xpointer="v260"/>
</listitem>
</varlistentry>
<varlistentry>
<term><varname>IRQAffinityNUMA=</varname></term>
<listitem>
<para>Filters CPUs to those belonging to the specified NUMA node. Takes either
<literal>local</literal> or an explicit NUMA node number (0, 1, 2, ...). When set to
<literal>local</literal>, the NUMA node local to the NIC's PCIe slot is used (determined from
<filename>/sys/class/net/<replaceable>iface</replaceable>/device/numa_node</filename>). If the
device's NUMA node cannot be determined (e.g., non-NUMA system), a warning is logged and IRQ
affinity configuration is skipped.</para>
<para>When both <varname>IRQAffinity=</varname> and <varname>IRQAffinityNUMA=</varname> are
specified, their intersection is used. If the intersection results in an empty set, an error is
logged and no affinity is applied. Has no effect if <varname>IRQAffinityPolicy=</varname> is not
set.</para>
<xi:include href="version-info.xml" xpointer="v260"/>
</listitem>
</varlistentry>
<varlistentry>
<term><varname>ReceiveVLANCTAGHardwareAcceleration=</varname></term>
<listitem>