On Wed, 2024-04-10 at 21:32 +0200, Cyril Brulebois wrote:
Hi,
Munin uses the following command to get sensor-type information out of SMART-aware disks (e.g. temperature):
/usr/sbin/smartctl -A --nocheck=standby -d ata /dev/sda
This broke following an upgrade from v6.1.76 (as found in Debian 12) to v6.1.82 (as currently found in the proposed-updates repository for the next point release of Debian 12), with smartctl's now reporting:
smartctl 7.3 2022-02-28 r5338 [x86_64-linux-6.1.0-19-amd64] (local build) Copyright (C) 2002-22, Bruce Allen, Christian Franke, www.smartmontools.org Device is in SLEEP mode, exit(2)
This happens on baremetal with 2 pairs of disks: - 2×ST4000VN008-2DR1 (sda, sdb) - 2×ST8000VN004-2M21 (sdc, sdd)
and that's an obvious lie with one pair doing system stuff and the other one doing media stuff.
This also happens within a Debian 12 QEMU VM running on a Debian 12 libvirt host, when using a SATA disk, which is what I've used to test various builds from the stable/linux-6.1.y branch and associated tags.
Building stable releases, I pinpointed it as a regression between v6.1.80 and v6.1.81, then pinpointed it to commit cf33e6ca12d8.
#regzbot introduced: v6.1.80..v6.1.81 #regzbot introduced: cf33e6ca12d8
This is also affecting v6.1.84 and v6.1.85 (released during my git bisect session).
Reported in Debian via: https://bugs.debian.org/1068675%C2%A0(which included a trace with the distribution-provided v6.1.82 package).
Most recent trace, with v6.1.85 (mainline, using the distribution's config but without any patches):
[ 30.547027] ------------[ cut here ]------------ [ 30.547034] WARNING: CPU: 0 PID: 697 at drivers/scsi/scsi_lib.c:214 scsi_execute_cmd+0x42/0x2c0 [scsi_mod] [ 30.547082] Modules linked in: tls tun intel_rapl_msr intel_rapl_common kvm_intel kvm irqbypass ghash_clmulni_intel sha512_ssse3 sha512_generic sha256_ssse3 sha1_ssse3 snd_hda_codec_generic ledtrig_audio snd_hda_intel snd_intel_dspcfg snd_intel_sdw_acpi aesni_intel snd_hda_codec crypto_simd cryptd rapl snd_hda_core snd_hwdep bochs drm_vram_helper pcspkr drm_ttm_helper snd_pcm iTCO_wdt snd_timer intel_pmc_bxt ttm iTCO_vendor_support snd watchdog soundcore virtio_console virtio_balloon drm_kms_helper button joydev evdev serio_raw sg binfmt_misc fuse loop drm efi_pstore dm_mod configfs qemu_fw_cfg virtio_rng ip_tables x_tables autofs4 ext4 crc32c_generic crc16 mbcache jbd2 hid_generic usbhid hid sd_mod t10_pi crc64_rocksoft crc64 crc_t10dif crct10dif_generic ahci libahci virtio_scsi virtio_blk virtio_net net_failover failover xhci_pci crct10dif_pclmul crct10dif_common crc32_pclmul libata crc32c_intel xhci_hcd psmouse i2c_i801 i2c_smbus scsi_mod scsi_common lpc_ich virtio_pci [ 30.547194] virtio_pci_legacy_dev virtio_pci_modern_dev usbcore usb_common virtio virtio_ring [ 30.547205] CPU: 0 PID: 697 Comm: smartctl Not tainted 6.1.85 #1 [ 30.547210] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.2-debian-1.16.2-1 04/01/2014 [ 30.547217] RIP: 0010:scsi_execute_cmd+0x42/0x2c0 [scsi_mod]
This is a different manifestation of the same bug in stable that was introduced by a backport of scsi_execute_cmd. The proposed fix for the domain validation problem here will also sort out this problem:
https://lore.kernel.org/linux-scsi/yq1frvvpymp.fsf@ca-mkp.ca.oracle.com/
James