IRQ smp_affinity reverting

Issue

Attempting to steer interrupt from Solarflare NIC to core0 using:

# echo 00000001 >/proc/irq/192/smp_affinity

Initial inspection shows value changed:

[root@lx-cnscitcffexb-fut02 ~]# cat /proc/irq/192/smp_affinity
000000,00000001

However reverts after an interrupt arrives on this queue:

[root@lx-cnscitcffexb-fut02 ~]# cat /proc/irq/192/smp_affinity
000000,08000000

Used SystemTap script:

probe begin
{
        printf("tracing...\n")
}

probe kernel.function("irq_set_affinity")
{
        printf("irq_set_affinity called:\n")
        print_stack(backtrace())
        print_usyms(ubacktrace())
}

probe kernel.function("setup_affinity")
{
        printf("setup_affinity called:\n")
        print_stack(backtrace())
        print_usyms(ubacktrace())
}

This was to check if some other process was doing the reset i.e. irqbalance. This is not the case. No other user space process is shown. The trace is:

irq_set_affinity called:
 0xffffffff810e8970 : irq_set_affinity+0x0/0x140 [kernel]
 0xffffffff810eabe5 : write_irq_affinity+0x135/0x140 [kernel]
 0xffffffff810eac3c : irq_affinity_proc_write+0x1c/0x20 [kernel]
 0xffffffff811f3ade : proc_reg_write+0x7e/0xc0 [kernel]
 0xffffffff81188df8 : vfs_write+0xb8/0x1a0 [kernel]
 0xffffffff811896f1 : sys_write+0x51/0x90 [kernel]
 0xffffffff8100b288 : tracesys+0xd9/0xde [kernel]
 0x3b05cdb490 : __write_nocancel+0x7/0x57 [/lib64/libc-2.12.so]
 0x3b05c71ae3 : _IO_file_write@@GLIBC_2.2.5+0x43/0xa0 [/lib64/libc-2.12.so]
 0x3b05c73095 : _IO_do_write@@GLIBC_2.2.5+0x95/0x12a [/lib64/libc-2.12.so]
 0x3b05c723ef : _IO_file_overflow@@GLIBC_2.2.5+0x14f/0x240 [/lib64/libc-2.12.so]
 0x3b05c6f147 : putc+0xb7/0xfd [/lib64/libc-2.12.so]
 0x469ace : 0x469ace [/bin/bash+0x69ace/0xd4000]

Which looks good and we can see the kernel routine "irq_set_affinity" being called.

Server is a HP DL380 Gen 9 with Haswell processors.