#1154436 linux-image-6.1.0-56-amd64: NULL pointer dereference in xas_load (filemap_map_pages) still locks up system with 6.1.192-2

#1154436#5
Date:
2026-11-19 07:20:46 UTC
From:
To:
Dear Maintainer,

After the ext4 regression in 6.1.192-1 (#1154101, #1154151), I upgraded
to 6.1.192-2 (linux-image-6.1.0-56-amd64). The ext4 mballoc crash is gone
(the create files / drop_caches / delete reproducer passes), but the system
still locked up about 9 minutes after boot, with a different oops in the
page cache path.

* Hardware / setup
  - Intel DN2800MT (Atom N2800), 2 GB RAM, single SATA HDD
  - ext4 root filesystem, swap partition
  - Docker (overlay2) workload: MariaDB, PostgreSQL, Apache/PHP,
    reverse proxy, file sync service, WireGuard

* Kernel history on this machine
  - 6.1.0-54-amd64 (6.1.190-1): stable, 8+ hours with the same
    workload, currently used as workaround
  - 6.1.0-55-amd64 (6.1.192-1): lockups from a few minutes to ~4.5 hours
    after boot (consistent with #1154101)
  - 6.1.0-56-amd64 (6.1.192-2): oops below ~9 minutes after boot

* Symptoms on 6.1.192-2
  - Oops in xas_load called from filemap_map_pages, in a page fault from
    an apache2 process
  - The task dies inside an RCU read-side critical section
    ("Voluntary context switch within RCU read-side critical section!")
    and then blocks in exit_mmap on the mmap rwsem
  - Followed by repeated RCU expedited stalls on that task, every minute
  - All disk writes then hang, SSH login hangs right after authentication,
    the machine still answers ping; a hard reset is required

My guess (not verified) is another NULL vs ERR_PTR mismatch introduced by
the folio backports, similar to the ext4 one, this time in the page cache
fault-around path.

* Oops (hostname removed)

BUG: kernel NULL pointer dereference, address: 0000000000000000
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 5fd4067 P4D 5fd4067 PUD 0
Oops: 0000 [#1] PREEMPT SMP NOPTI
CPU: 2 PID: 2537 Comm: apache2 Not tainted 6.1.0-56-amd64 #1  Debian
6.1.192-2
Hardware name:  /DN2800MT, BIOS MTCDT10N.86A.0158.2012.0808.1446 08/08/2012
RIP: 0010:xas_load+0x26/0x40
Code: 00 00 00 00 e8 3b ff ff ff 48 89 c2 83 e2 03 48 83 fa 02 75 08 48 3d
00 10 00 00 77 05 c3 cc cc cc cc 0f b6 4f 10 48 8d 70 fe <38> 48 fe 72 ee
e8 90 fd ff ff 80 3e 00 75 d0 c3 cc cc cc cc 66 0f
RSP: 0000:ffffd2d0c133bce8 EFLAGS: 00010202
RAX: 0000000000000002 RBX: ffffd2d0c133bd40 RCX: 0000000000000048
RDX: 0000000000000003 RSI: 0000000000000000 RDI: fffff68d806f1f80
RBP: ffff8cc9cac91778 R08: 000000004056f000 R09: 000000004856efff
R10: 00000000000008c0 R11: 00000000000008b1 R12: 00000000000008c0
R13: fffff68d806f1fb4 R14: ffff8cc9cac91778 R15: ffff8cc9c5fd4008
FS:  00007ff390e38c00(0000) GS:ffff8cca3c100000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 000000000612c000 CR4: 00000000000006e0
Call Trace:
 <TASK>
 ? filemap_map_pages+0xb0/0x720
 ? do_fault+0x1a4/0x410
 ? __handle_mm_fault+0x660/0xfa0
 ? handle_mm_fault+0xdb/0x2d0
 ? do_user_addr_fault+0x191/0x570
 ? exc_page_fault+0x70/0x170
 ? asm_exc_page_fault+0x22/0x30
 </TASK>
Modules linked in: wireguard libchacha20poly1305 chacha_x86_64
poly1305_x86_64 curve25519_x86_64 libcurve25519_generic libchacha
ip6_udp_tunnel udp_tunnel nf_conntrack_netlink xt_nat xt_tcpudp veth
xt_conntrack xt_MASQUERADE bridge stp llc xfrm_user xfrm_algo xt_set ip_set
nft_chain_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 xt_addrtype
ipt_REJECT nf_reject_ipv4 xt_multiport nft_compat nf_tables nfnetlink
overlay cfg80211 rfkill isofs cdrom binfmt_misc intel_powerclamp
drm_kms_helper coretemp evdev sha512_ssse3 sha512_generic ppdev iTCO_wdt
sha256_ssse3 sha1_ssse3 i2c_algo_bit at24 intel_pmc_bxt iTCO_vendor_support
parport_pc watchdog parport sg button acpi_cpufreq drm dm_mod efi_pstore
fuse loop configfs ip_tables x_tables autofs4 ext4 crc16 mbcache jbd2 btrfs
blake2b_generic zstd_compress raid10 raid456 async_raid6_recov async_memcpy
async_pq async_xor async_tx xor raid6_pq libcrc32c crc32c_generic raid1
raid0 multipath linear md_mod sd_mod t10_pi crc64_rocksoft crc64 crc_t10dif
crct10dif_generic crct10dif_common ahci libahci libata scsi_mod e1000e
uhci_hcd ehci_pci ehci_hcd usbcore scsi_common i2c_i801 i2c_smbus lpc_ich
usb_common video wmi
CR2: 0000000000000000
---[ end trace 0000000000000000 ]---
------------[ cut here ]------------
Voluntary context switch within RCU read-side critical section!
WARNING: CPU: 2 PID: 2537 at kernel/rcu/tree_plugin.h:320
rcu_note_context_switch+0x4ee/0x650
CPU: 2 PID: 2537 Comm: apache2 Tainted: G      D            6.1.0-56-amd64
#1  Debian 6.1.192-2
Hardware name:  /DN2800MT, BIOS MTCDT10N.86A.0158.2012.0808.1446 08/08/2012
RIP: 0010:rcu_note_context_switch+0x4ee/0x650
Call Trace:
 <TASK>
 __schedule+0xae/0xa00
 schedule+0x5a/0xd0
 rwsem_down_write_slowpath+0x315/0x6f0
 exit_mmap+0xf6/0x310
 __mmput+0x3e/0x130
 do_exit+0x326/0xb10
 ? do_user_addr_fault+0x191/0x570
 make_task_dead+0x8d/0x90
 rewind_stack_and_make_dead+0x17/0x20
 </TASK>

Then, every minute until the hard reset:
rcu: INFO: rcu_preempt detected expedited stalls on CPUs/tasks: { P2537 }
85122 jiffies s: 1505 root: 0x0/T

So far this happened once on 6.1.192-2. I can provide the full kernel log
of that boot or test patches/kernels if needed.

Thanks for your work.

#1154436#10
Date:
2026-11-19 07:40:34 UTC
From:
To:
Hi,

Is it reproducible? Yes please provide the full log of from where the
problem occured.

If it is reproducible I will need to ask you to bisect between the
last good, and the 6.1.192 upstream kernel, I can sent you
instructions for that.

Thanks already,

Regards,
Salvatore

#1154436#17
Date:
2026-11-19 08:15:01 UTC
From:
To:
Hi Salvatore,

Thanks for the quick reply.

- Full log: attached is the complete kernel log of the affected boot
  on 6.1.192-2 (only the hostname and MAC address are masked).

- Reproducibility: so far it happened once on 6.1.192-2, about 9 minutes
  after boot, with the usual Docker workload. I went back to 6.1.190-1
  right after, which is stable with the same workload. I will boot
  6.1.192-2 again to check whether it reproduces, and report back.

- Bisect: unfortunately I cannot build kernels myself (the affected
  machine is an Atom N2800 with 2 GB RAM and I have no other suitable
  build environment). However, I am happy to install and test prebuilt
  kernel packages if you can provide them (for example test builds with
  suspect commits reverted), and report the results here.

Thanks for your work,

Le jeu. 19 nov. 2026 à 08:40, Salvatore Bonaccorso <carnil@debian.org> a
écrit :

#1154436#22
Date:
2026-11-19 19:42:59 UTC
From:
To:
Hi Salvatore,

Follow-up on reproducibility:

- I booted the affected machine again on 6.1.192-2 with the same Docker
  workload. It has now been running for about 9h40 without any problem.
- During that run I added a 20-minute stress test aimed at the page cache
  fault path: stress-ng (--mmap 2 --mmap-file --vm 1 --vm-bytes 512M)
  plus "sync; echo 3 > /proc/sys/vm/drop_caches" every 20 seconds, and
  some HTTP load on the Apache container.
- netconsole was active towards another host during the whole run:
  no BUG/Oops, no RCU stall, no hung task.
- A second machine (Xeon E3-1245 v2, 32 GB RAM, Docker workload) has been
  running 6.1.192-2 since 2026-11-18 02:21 UTC (about 40 hours) without
  any issue.

So far the xas_load oops happened only once, on the first boot into
6.1.192-2 (right after several hard resets caused by the 6.1.192-1
regression). I am keeping 6.1.192-2 in production with netconsole
enabled for the next days, and will report here immediately with a full
trace if it happens again.

Feel free to adjust the severity accordingly.

Thanks,

Le jeu. 19 nov. 2026 à 09:15, MAXIME CARON <maxinfo59@gmail.com> a écrit :