Re: n_tty_write() going into schedule but NOT coming out

[Vineet Gupta <Vineet.Gupta1@synopsys.com>]

On 04/06/2013 01:22 AM, Peter Hurley wrote:
> I'll see if I can reproduce this over the weekend on an old single-core laptop I
> still have.

TIA for doing this.

> There were some race conditions in the N_TTY line discipline which I
> recently fixed. Those changes are in linux-next. Can you test if this is
> reproducible on linux-next?

I tried today's -next and I see the same issue :-(

> Assuming I don't reproduce this on the laptop, the only other
> explanation I can think of right now is that ARCLinux is not properly
> handling signal-driven i/o (assuming the BusyBox /bin/sh uses SIGIO). Do
> you know if there is anything special about the way ARCLinux handle
> signals?

No, it's pretty standard stuff - we are uClibc based though - as opposed to glibc
so there might be some subtleties - but we do run LTP open posix regularly. Also
the test setup is a slowish 80MHz FPGA so this is not something many people have
in their regular test setups.

I'll start reading the relevant code myself - and will be willing to take any
debug/test patches which help with troubleshooting of this issue.

Just to re-iterate, my test setup has a minimal busybox based rootfs, 3 telnet
sessions, each running a
while true; do find / -name "*" ; done

I'm running out of ramfs, no external disk/nfs mounts to reduce the peripheral I/O
slowing down the find (although NFS stuff will be caches anyways after first fetch).

Also please make sure you have CONFIG_PREEMPT_COUNT otherwise there's a
possibility (atleast on ARC builds) that a stale register causes timer list to be
corrupted in mod_timer().

Thx,
-Vineet