In Linux (RHEL 8.8), I've tried a pinning the IRQ CPU affinity of certain input devices,…
- User
In Linux (RHEL 8.8), I've tried a pinning the IRQ CPU affinity of certain input devices,
for i in /proc/irq/*/; do if [ -f $i/smp_affinity_list ]; then echo "0,1,48,49" > $i/smp_affinity_list; fi; doneI am getting write error: Input/output error Note: I am logged in as root, and the CPU List is valid, and I am aware that some IRQs might have their affinity hardcoded by the kernel and cannot be changed.
Assistant
Setting IRQ CPU affinity in Linux can fail with "Input/output error" even when running as root, primarily due to two reasons:
1. Non-modifiable IRQ Controllers
Some hardware IRQ controllers don't support affinity changes. As noted in the kernel documentation[1]:- IRQ affinity cannot be modified if the controller lacks SMP affinity support
- Default affinity remains locked for certain critical interrupts (e.g., timer, IPIs)
- PCI MSI interrupts often have stricter affinity constraints
2. Invalid CPU Format for Specific IRQs
While your CPU list "0,1,48,49" appears valid:# Check available CPUs cat /sys/devices/system/cpu/possibleSome IRQs may require:
- Continuous CPU ranges (e.g., 0-3 instead of 0,1,48,49)
- Hexadecimal bitmask format in smp_affinity instead of smp_affinity_list
- Architecture-specific constraints (e.g., NUMA node boundaries)[7]
Debugging Steps
# 1. Find problematic IRQs grep -H . /proc/irq/*/smp_affinity_list 2>&1 | grep -v "Invalid argument" # 2. Check controller support for irq in /proc/irq/*; do echo "IRQ ${irq##*/}: $(cat $irq/effective_affinity)" done # 3. Test individual IRQs echo 0,1,48,49 > /proc/irq/N/smp_affinity_listRecommended Fix
Modify your script to handle edge cases:for irq_dir in /proc/irq/*/; do irq=${irq_dir%/} irq=${irq##*/} if [[ -f "$irq_dir/smp_affinity_list" ]] && [[ ! $(< /proc/irq/$irq/effective_affinity) =~ ^0,*1,*48,*49,*$ ]]; then if ! echo "0,1,48,49" > "$irq_dir/smp_affinity_list" 2>/dev/null; then echo "IRQ $irq: Failed to set affinity (controller restriction?)" >&2 fi fi doneAlternative Approaches
- Use
irqbalancewith custom policies[2][6]
# Configure irqbalance to exclude specific CPUs systemctl enable --now irqbalance echo "0,1,48,49" > /etc/sysconfig/irqbalance_EXCLUDED_CPUS- Apply CPU isolation at boot[5][7]
# Kernel command-line parameters isolcpus=0,1,48,49 nohz_full=0,1,48,49 rcu_nocbs=0,1,48,49Key Considerations
- NUMA topology affects valid affinity ranges[7]
- Real-time kernels impose additional restrictions[5]
- Some device drivers override user-space affinity settings[4]
- Always verify with
/proc/interruptsafter changes[1][3]
# Post-change verification watch -n1 "egrep 'CPU|eth0' /proc/interrupts"
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.