Question about the ipi_raise filter usage and output

[richard clark]

Hi Steve,

On Mon, Feb 5, 2024 at 6:38 PM Steven Rostedt <rostedt at goodmis.org> wrote:
>
> On Mon, 5 Feb 2024 17:57:29 +0800
> richard clark <richard.xnu.clark at gmail.com> wrote:
>
> > I try to write below:
> > echo 'target_cpus == 11 && reason == "Function call interrupts"' >
> > events/ipi/ipi_raise/filter
>
> You mean when it is sent only to CPU 11? Not when the event is
> happening on CPU 11. Like the above example, the event was triggered on
> CPU 10, but the mask was for all the other CPUs.
>
> If you are looking for just CPU 11, you can do:
>
>   echo 'target_cpus == 0x800 && reason == "Function call interrupts"'
>

Seems both 'target_cpus == 0x800 && reason == "Function call
interrupts"' and 'target_cpus & 0x800 && reason == "Function call
interrupts"' don't work:

# cat events/ipi/ipi_raise/enable
1
# cat events/ipi/ipi_raise/filter
target_cpus == 0x800 && reason == "Function call interrupts"

The kernel module code snippet:

void ipi_func_run_cpu(void *info)
{
        pr_info("remote function runs on cpu[%d].\n", smp_processor_id());
}
static int __init ipi_send_init(void)
{
        int target = (smp_processor_id() + 1) % nr_cpu_ids;
        int ret = smp_call_function_single(target, ipi_func_run_cpu,
NULL, true);
        pr_info("ipi cpu[%d --> %d] ret = %d\n", smp_processor_id(),
target, ret);
        return 0;
}
...
module_init(ipi_send_init);
module_exit(ipi_send_exit);

$ sudo taskset -c 10 insmod ipi_send.ko
$ dmesg
...
[84931.864273] remote function runs on cpu[11].
[84931.864282] ipi cpu[10 --> 11] ret = 0

The 'cat trace' will output the below message with 'reason ==
"Function call interrupts"' filter:
...
            sudo-5726    [007] dn.h1.. 84302.833545: ipi_raise:
target_mask=00000000,00000001 (Function call interrupts)
            sudo-5726    [007] dn.h2.. 84302.837544: ipi_raise:
target_mask=00000000,00000001 (Function call interrupts)
          insmod-5727    [011] dn.h1.. 84302.841545: ipi_raise:
target_mask=00000000,00000001 (Function call interrupts)
          insmod-5727    [010] ....1.. 84302.843966: ipi_raise:
target_mask=00000000,00000bff (Function call interrupts)
          insmod-5727    [010] ....1.. 84302.843975: ipi_raise:
target_mask=00000000,00000bff (Function call interrupts)
          insmod-5727    [010] ....1.. 84302.844184: ipi_raise:
target_mask=00000000,00000800 (Function call interrupts)
...

I find that 'target_cpus == 0xfff && reason == "Function call
interrupts"' doesn't have output in the buffer, but 'target_cpus &
0xfff && reason == "Function call interrupts"' does. I also tried to
use 'target_cpus & 0xf00 && reason == "Function call interrupts"' in
my case, the trace buffer has nothing after the kmod inserted.

Any comments?

>
>
> -- Steve