Hello，
In my environment, a perf_mmap softlockup syzkaller problem occurs. The
call stack is as follows:
[ 3666.984385][ C2] watchdog: BUG: soft lockup - CPU#2 stuck for 23s!
[syz-executor.2:32404]
[ 3666.986137][ C2] Modules linked in:
[ 3666.989581][ C2] CPU: 2 PID: 32404 Comm: syz-executor.2 Not
tainted 5.10.0+ #4
[ 3666.990697][ C2] Hardware name: linux,dummy-virt (DT)
[ 3666.992270][ C2] pstate: 80000005 (Nzcv daif -PAN -UAO -TCO BTYPE=--)
[ 3666.993787][ C2] pc : __kasan_check_write+0x0/0x40
[ 3666.994841][ C2] lr : perf_mmap+0x3c8/0xf80
[ 3666.995661][ C2] sp : ffff00001011f8f0
[ 3666.996598][ C2] x29: ffff00001011f8f0 x28: ffff0000cf644868
[ 3666.998488][ C2] x27: ffff000012cad2c0 x26: 0000000000000000
[ 3666.999888][ C2] x25: 0000000000000001 x24: ffff000012cad298
[ 3667.003511][ C2] x23: 0000000000000000 x22: ffff000012cad000
[ 3667.005504][ C2] x21: ffff0000cf644818 x20: ffff0000cf6d2400
[ 3667.006891][ C2] x19: ffff0000cf6d24c0 x18: 0000000000000000
[ 3667.008295][ C2] x17: 0000000000000000 x16: 0000000000000000
[ 3667.009528][ C2] x15: 0000000000000000 x14: 0000000000000000
[ 3667.010658][ C2] x13: 0000000000000000 x12: ffff800002023f17
[ 3667.012169][ C2] x11: 1fffe00002023f16 x10: ffff800002023f16
[ 3667.013780][ C2] x9 : dfffa00000000000 x8 : ffff00001011f8b7
[ 3667.015265][ C2] x7 : 0000000000000001 x6 : ffff800002023f16
[ 3667.016683][ C2] x5 : ffff0000c0f36400 x4 : 0000000000000000
[ 3667.018078][ C2] x3 : ffffa00010000000 x2 : ffffa000119a0000
[ 3667.019343][ C2] x1 : 0000000000000004 x0 : ffff0000cf6d24c0
[ 3667.021276][ C2] Call trace:
[ 3667.022598][ C2] __kasan_check_write+0x0/0x40
[ 3667.023666][ C2] __mmap_region+0x7a4/0xc90
[ 3667.024679][ C2] __do_mmap_mm+0x600/0xa20
[ 3667.025700][ C2] do_mmap+0x114/0x384
[ 3667.026583][ C2] vm_mmap_pgoff+0x138/0x230
[ 3667.027532][ C2] ksys_mmap_pgoff+0x1d8/0x570
[ 3667.028537][ C2] __arm64_sys_mmap+0xa4/0xd0
[ 3667.029597][ C2] el0_svc_common.constprop.0+0xf4/0x414
[ 3667.030682][ C2] do_el0_svc+0x50/0x11c
[ 3667.031545][ C2] el0_svc+0x20/0x30
[ 3667.032368][ C2] el0_sync_handler+0xe4/0x1e0
[ 3667.033305][ C2] el0_sync+0x148/0x180
I checked the perf_mmap code locally. When the problem occurred, the
kernel kept running the goto again command:
again:
mutex_lock(&event->mmap_mutex);
if (event->rb) {
if (data_page_nr(event->rb) != nr_pages) {
ret = -EINVAL;
goto unlock;
}
if (!atomic_inc_not_zero(&event->rb->mmap_count)) {
/*
* Raced against perf_mmap_close() through
* perf_event_set_output(). Try again, hope for
bette
* luck.
*/
mutex_unlock(&event->mmap_mutex);
goto again; ===> kept goto again
}
goto unlock;
} else {
printk("[%d]test_mmap: perf_mmap lock, NULL\n", get_cpu());
put_cpu();
}
Under what conditions would this cause?
Best Regards,
Yang