On 2021-01-27, John Ogness <john.ogness(a)linutronix.de> wrote:
I was finally able to trigger this by using my workstation
Well, I don't know how reliably I can trigger this. I think my time will
be better spent getting lkp to trigger, since that shows to be reliable.
It took me an hour to trigger this one and I have the feeling I got
lucky with it.
[ 1854.183816] INFO: rcu_tasks detected stalls on tasks:
[ 1854.184502] 0000000036ce200f: .. nvcsw: 0/0 holdout: 1 idle_cpu: -1/1
[ 1854.185234] task:dd state:R running task stack: 0 pid:19891 ppid:
[ 1854.185273] Call Trace:
[ 1854.185275] __schedule+0xfa9/0x1056
[ 1854.185283] ? firmware_map_remove+0x172/0x172
[ 1854.185300] ? ksys_write+0x147/0x180
[ 1854.185304] schedule+0x170/0x1e3
[ 1854.185307] exit_to_user_mode_prepare+0x27/0xac
[ 1854.185312] syscall_exit_to_user_mode+0x41/0x4f
[ 1854.185316] entry_SYSCALL_64_after_hwframe+0x44/0xa9
[ 1854.185321] RIP: 0033:0x7f99bf52fc00
[ 1854.185324] RSP: 002b:00007ffd75722bf8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
[ 1854.185328] RAX: 0000000000000001 RBX: 00007f99bfa1b690 RCX: 00007f99bf52fc00
[ 1854.185331] RDX: 0000000000000001 RSI: 0000560056ce3010 RDI: 0000000000000001
[ 1854.185335] RBP: 0000000000000001 R08: 0000000000000000 R09: 0000000000000000
[ 1854.185338] R10: 00007ffd757229c0 R11: 0000000000000246 R12: 0000560056ce3010
[ 1854.185341] R13: 0000000000000001 R14: 0000560056ce3010 R15: 0000560056ce3010
To generate syscalls and interrupts, "dd" was doing:
dd if=/dev/zero bs=1 count=100000000 | nc 10.0.2.2 12345
I'm not even sure how to go about debugging this. The only printk load I
can really see are the:
"tasks-torture:torture_onoff task: online 0 failed: errno -5"
error messages. But they are only occurring at a rate of 40-50 per
second. They are coming from the pr_alert() in kernel/torture.c:174, so
it should still be hitting spinlocks (either from the console driver or
I will re-focus on reproducing this with lkp.