RE: [RFC PATCH] mm: move xa forward when run across zombie page

From: "Pulavarty, Badari" <badari.pulavarty@intel.com>
To: Matthew Wilcox <willy@infradead.org>
Cc: "david@fromorbit.com" <david@fromorbit.com>,
	"akpm@linux-foundation.org" <akpm@linux-foundation.org>,
	"bfoster@redhat.com" <bfoster@redhat.com>,
	"huangzhaoyang@gmail.com" <huangzhaoyang@gmail.com>,
	"ke.wang@unisoc.com" <ke.wang@unisoc.com>,
	"linux-fsdevel@vger.kernel.org" <linux-fsdevel@vger.kernel.org>,
	"linux-mm@kvack.org" <linux-mm@kvack.org>,
	"zhaoyang.huang@unisoc.com" <zhaoyang.huang@unisoc.com>,
	"Shutemov, Kirill" <kirill.shutemov@intel.com>,
	"Tang, Feng" <feng.tang@intel.com>,
	"Huang, Ying" <ying.huang@intel.com>,
	"Yin, Fengwei" <fengwei.yin@intel.com>,
	"Hansen, Dave" <dave.hansen@intel.com>,
	"Zanussi, Tom" <tom.zanussi@intel.com>
Subject: RE: [RFC PATCH] mm: move xa forward when run across zombie page
Date: Fri, 21 Oct 2022 22:40:05 +0000	[thread overview]
Message-ID: <DM6PR11MB397838A94F24C1F4CCE306929C2D9@DM6PR11MB3978.namprd11.prod.outlook.com> (raw)
In-Reply-To: <Y1Md0hzhkqzik/WA@casper.infradead.org>

-----Original Message-----
From: Matthew Wilcox <willy@infradead.org> 
Sent: Friday, October 21, 2022 3:32 PM
To: Pulavarty, Badari <badari.pulavarty@intel.com>
Cc: david@fromorbit.com; akpm@linux-foundation.org; bfoster@redhat.com; huangzhaoyang@gmail.com; ke.wang@unisoc.com; linux-fsdevel@vger.kernel.org; inux-kernel@vger.kernel.org; linux-mm@kvack.org; zhaoyang.huang@unisoc.com; Shutemov, Kirill <kirill.shutemov@intel.com>; Tang, Feng <feng.tang@intel.com>; Huang, Ying <ying.huang@intel.com>; Yin, Fengwei <fengwei.yin@intel.com>; Hansen, Dave <dave.hansen@intel.com>; Zanussi, Tom <tom.zanussi@intel.com>
Subject: Re: [RFC PATCH] mm: move xa forward when run across zombie page

On Fri, Oct 21, 2022 at 09:37:36PM +0000, Pulavarty, Badari wrote:
> I have been tracking similar issue(s) with soft lockup or panics on my system consistently with my workload.
> Tried multiple kernel versions. Issue seem to happen consistently on 
> 6.1-rc1 (while it seem to happen on 5.17, 5.19, 6.0.X)
> 
> PANIC: "Kernel panic - not syncing: softlockup: hung tasks"
> 
>     RIP: 0000000000000001  RSP: ff3d8e7f0d9978ea  RFLAGS: ff3d8e7f0d9978e8
>     RAX: 0000000000000000  RBX: 0000000000000000  RCX: 0000000000000000
>     RDX: 000000006b9c66f1  RSI: ff506ca15ff33c20  RDI: 0000000000000000
>     RBP: ffffffff84bc64cc   R8: ff3d8e412cabdff0   R9: ffffffff84c00e8b
>     R10: ff506ca15ff33b69  R11: 0000000000000000  R12: ff506ca15ff33b58
>     R13: ffffffff84bc79a3  R14: ff506ca15ff33b38  R15: 0000000000000000
>     ORIG_RAX: ff506ca15ff33a80  CS: ff506ca15ff33c78  SS: 0000
> #9 [ff506ca15ff33c18] xas_load at ffffffff84b49a7f
> #10 [ff506ca15ff33c28] __filemap_get_folio at ffffffff840985da
> #11 [ff506ca15ff33ce8] swap_cache_get_folio at ffffffff841119db

Oh, this is interesting.  It's the swapper address_space.
I bet that 0xffffffff85044560 (the value of a_ops) is the address of swap_ops in your kernel?

I don't know if it will help, but it's an interesting data point.

> Looking at the crash dump, mapping->host became NULL. Not sure what exactly is happening.

That's always true for the swapper_spaces, AIUI.

>   a_ops = 0xffffffff85044560,

Correct. Its swap_ops. (I am using zswap).

In my scenario - I run the workload in a container and use DAMON or PSI to squeeze the cold pages out to zswap.

Thanks,
Badari