On 1/9/24 13:45, Kevin Locke wrote:
On Tue, 2024-01-09 at 13:39 -0800, Bart Van Assche wrote:(cd /sys/class/scsi_host && grep -aH . */state) (cd /sys/class/scsi_device && grep -aH . */device/{device_{blocked,busy},state})Sure thing. Running the above commands produced the following output: host0/state:running host1/state:running host2/state:running host3/state:running host4/state:running host5/state:running 0:0:0:0/device/device_blocked:0 0:0:0:0/device/device_busy:1 0:0:0:0/device/state:running
So the SCSI host state (host1) is fine but the information about the SCSI devices associated with host1 is missing, most likely because sysfs information of SCSI devices is removed before a SYNCHRONIZE CACHE command is submitted. The device_del(dev) call in __scsi_remove_device() happens after scsi_device_set_state(sdev, SDEV_CANCEL) so the SCSI device should be in the SDEV_CANCEL state. scsi_device_state_check() should translate SDEV_CANCEL into BLK_STS_OFFLINE. There are several tests in the blktests suite that trigger SCSI device deletion, e.g. block/001. All blktests tests pass on my test setup. Additionally, I haven't seen any blktests failure reports recently that are related to device deletion. If I try to delete an ATA device in a VM, that works fine (kernel v6.7): # dmesg -c >/dev/null # echo 1 > /sys/class/scsi_device/3:0:0:0/device/delete # dmesg -c [ 215.533228] sd 3:0:0:0: [sdb] Synchronizing SCSI cache [ 215.543932] ata3.00: Entering standby power mode Running rescan-scsi-bus.sh -a brings this device back. I'm not sure what I'm missing but I think that it's something ATA-specific. Since I'm not an ATA expert, I hope that an ATA expert can help. There are at least two ATA experts on the Cc-list of this email. Thanks, Bart.
