On Mon, Feb 12, 2018 at 10:41:05AM +0100, Matthias Ferdinand wrote:
> On Mon, Feb 12, 2018 at 04:57:52PM +0800, tang.junhui@zte.com.cn wrote:
> > From: Tang Junhui <tang.junhui@zte.com.cn>
> > 
> > 
> > Hello Matthias,
> > 
> > What kind of caching mode do you use? Writeback or Writethrough?
> 
> Hi, thanks for looking into this.
> 
> This is with writeback caching (haven't tried writethrough yet).
> 
> Some more info about the disk drives used:
> 
> # ls -l /dev/disk/by-id  | grep -v -- -part | grep -v wwn-
> total 0
> lrwxrwxrwx 1 root root  9 Feb  8 00:48 ata-INTEL_SSDSC2BA200G4_BTHV7374017P200MGN -> ../../sdf
> lrwxrwxrwx 1 root root  9 Feb  8 00:48 ata-WDC_WD2005FBYZ-01YCBB2_WD-WMC6N0K0SHRT -> ../../sdc
> lrwxrwxrwx 1 root root  9 Feb  8 00:48 ata-WDC_WD2005FBYZ-01YCBB2_WD-WMC6N0K2KNK4 -> ../../sda
> lrwxrwxrwx 1 root root  9 Feb  8 00:48 ata-WDC_WD2005FBYZ-01YCBB2_WD-WMC6N0K5KHNS -> ../../sdb
> lrwxrwxrwx 1 root root  9 Feb  8 00:48 ata-WDC_WD2005FBYZ-01YCBB2_WD-WMC6N0K7212Z -> ../../sde
> lrwxrwxrwx 1 root root  9 Feb  8 00:48 usb-Samsung_Flash_Drive_0352315110002396-0:0 -> ../../sdd

More de-obfuscation:
   - Intel SSD DC S3710 SATA 200GB for caching
   - WD Gold SATA 2TB for backing device

I now also tested with alternatives writethrough and with "hdparm -W 1",
they also crash.

Perhaps more relevant: fio is started immediately after setup of md
raid and bcache, while md sync speed is set very low:

    echo "1" >/sys/block/md127/md/sync_speed_min
    echo "1" >/sys/block/md127/md/sync_speed_max


I attached the script I use for producing the crashes, hope this helps.

Regards
Matthias Ferdinand