Re: ppp/pppoe, still panic 4.15.3 in ppp_push

On Wed, Feb 14, 2018 at 06:49:19PM +0200, Denys Fedoryshchenko wrote:

On 2018-02-14 18:47, Guillaume Nault wrote:

quoted

On Wed, Feb 14, 2018 at 06:29:34PM +0200, Denys Fedoryshchenko wrote:

quoted

On 2018-02-14 18:07, Guillaume Nault wrote:

quoted

On Wed, Feb 14, 2018 at 03:17:23PM +0200, Denys Fedoryshchenko wrote:

quoted

Hi,

Upgraded kernel to 4.15.3, still it crashes after while (several
hours,
cannot do bisect, as it is production server).

dev ppp # gdb ppp_generic.o
GNU gdb (Gentoo 7.12.1 vanilla) 7.12.1
<<skipped>>
Reading symbols from ppp_generic.o...done.
(gdb) list *ppp_push+0x73
0x681 is in ppp_push (drivers/net/ppp/ppp_generic.c:1663).
1658			list = list->next;
1659			pch = list_entry(list, struct channel, clist);
1660
1661			spin_lock(&pch->downl);
1662			if (pch->chan) {
1663				if (pch->chan->ops->start_xmit(pch->chan, skb))
1664					ppp->xmit_pending = NULL;
1665			} else {
1666				/* channel got unregistered */
1667				kfree_skb(skb);

I expect a memory corruption. Do you have the possibility to run with
KASAN by any chance?

I will try to enable it tonight. For now i reverted "drivers, net,
ppp:
convert ppp_file.refcnt from atomic_t to refcount_t" for test.

This commit looks good to me. Do you have doubts about it because it's
new in 4.15? Does it mean that your last known-good kernel is 4.14?

I am just doing "manual" bisect, checking all possibilities, and picking
patch to revert randomly.

Must be a painful process. Are all of your networking modules required?
With luck, you might be able to isolate a faulty module in fewer steps.

Yes, correct, my known-good is 4.14.2.

Good to know.

Let me know if you can get a KASAN trace.