From the stack trace, zfs seems to be setting a property in a newly created zfs filesystem (since i'm mostly creating containers at this specific point in time). Following the code, zfs_ioctl_init sets as a callback to ZFS_IOC_SET_PROP, the function zfs_ioc_set_prop() which will trigger the code what gets soft locked by 120 sec (spinning on txg_wait_synced). The txg kthread, likely the one responsible for sync to be finished, is waiting on completion for zio that were submitted. So, yes, I do agree that this might be an overload because of the amount of spindles/throughput I have. I have to check now if 4.4 is also in this same step because it looked dead locked to me, for hours. Upstream kernel/zfs/spl is, at least, in good conditions after I/O is finished and sync is done.
@Colin
----
1. its a really simple one since this is my devel machine: one ssd disk only
zpool: https:/ /pastebin. ubuntu. com/p/gZPtrxRDn F/ /pastebin. ubuntu. com/p/JYyGwQ7cV r/
hdparm: https:/
2. 8 x 4.0GHz AMD cores and 32GB RAM
meminfo: https:/ /pastebin. ubuntu. com/p/kSjXSC6R3 q/ (at the time of soft lockup)
3. likely it does since its a single spindle for zpool
iostat: https:/ /pastebin. ubuntu. com/p/vbxXZKzmB p/
----
From the stack trace, zfs seems to be setting a property in a newly created zfs filesystem (since i'm mostly creating containers at this specific point in time). Following the code, zfs_ioctl_init sets as a callback to ZFS_IOC_SET_PROP, the function zfs_ioc_set_prop() which will trigger the code what gets soft locked by 120 sec (spinning on txg_wait_synced). The txg kthread, likely the one responsible for sync to be finished, is waiting on completion for zio that were submitted. So, yes, I do agree that this might be an overload because of the amount of spindles/throughput I have. I have to check now if 4.4 is also in this same step because it looked dead locked to me, for hours. Upstream kernel/zfs/spl is, at least, in good conditions after I/O is finished and sync is done.