1 ==============================
2 IRQ affinity on IA64 platforms
3 ==============================
5 07.01.2002, Erich Focht <efocht@ess.nec.de>
8 By writing to /proc/irq/IRQ#/smp_affinity the interrupt routing can be
9 controlled. The behavior on IA64 platforms is slightly different from
10 that described in Documentation/core-api/irq/irq-affinity.rst for i386 systems.
12 Because of the usage of SAPIC mode and physical destination mode the
13 IRQ target is one particular CPU and cannot be a mask of several
14 CPUs. Only the first non-zero bit is taken into account.
20 The target CPU has to be specified as a hexadecimal CPU mask. The
21 first non-zero bit is the selected CPU. This format has been kept for
22 compatibility reasons with i386.
24 Set the delivery mode of interrupt 41 to fixed and route the
25 interrupts to CPU #3 (logical CPU number) (2^3=0x08)::
27 echo "8" >/proc/irq/41/smp_affinity
29 Set the default route for IRQ number 41 to CPU 6 in lowest priority
30 delivery mode (redirectable)::
32 echo "r 40" >/proc/irq/41/smp_affinity
34 The output of the command::
36 cat /proc/irq/IRQ#/smp_affinity
38 gives the target CPU mask for the specified interrupt vector. If the CPU
39 mask is preceded by the character "r", the interrupt is redirectable
40 (i.e. lowest priority mode routing is used), otherwise its route is
45 Initialization and default behavior
46 ===================================
48 If the platform features IRQ redirection (info provided by SAL) all
49 IO-SAPIC interrupts are initialized with CPU#0 as their default target
50 and the routing is the so called "lowest priority mode" (actually
51 fixed SAPIC mode with hint). The XTP chipset registers are used as hints
52 for the IRQ routing. Currently in Linux XTP registers can have three
55 - minimal for an idle task,
56 - normal if any other task runs,
57 - maximal if the CPU is going to be switched off.
59 The IRQ is routed to the CPU with lowest XTP register value, the
60 search begins at the default CPU. Therefore most of the interrupts
61 will be handled by CPU #0.
63 If the platform doesn't feature interrupt redirection IOSAPIC fixed
64 routing is used. The target CPUs are distributed in a round robin
65 manner. IRQs will be routed only to the selected target CPUs. Check
75 On large (multi-node) systems it is recommended to route the IRQs to
76 the node to which the corresponding device is connected.
77 For systems like the NEC AzusA we get IRQ node-affinity for free. This
78 is because usually the chipsets on each node redirect the interrupts
79 only to their own CPUs (as they cannot see the XTP registers on the