On 04/08/2020 16:37, Johannes Thumshirn wrote:

quoted

On 04/08/2020 16:34, Coly Li wrote:

quoted

On 2020/8/4 22:31, Johannes Thumshirn wrote:

quoted

On 04/08/2020 16:23, Coly Li wrote:

quoted

This is the procedure to reproduce the panic,
  # modprobe scsi_debug delay=0 dev_size_mb=2048 max_queue=1
  # losetup -f /dev/nvme0n1 --direct-io=on
  # blkdiscard /dev/loop0 -o 0 -l 0x200

losetup -f /dev/sdX isn't it?

In my case, I use a NVMe SSD as the backing device of the loop device.
Because I don't have a scsi lun.

And loading scsi_debug module seems necessary, otherwise the discard
process just hang and I cannot see the kernel panic (I don't know why yet).

OK, now that's highly interesting. Does it also happen if you back loop with
a file? loop_config_discard() has different cases for the different backing devices/files. S

Damn I didn't want to hit sent....

Does this (untested) change make a difference:

diff --git a/drivers/block/loop.c b/drivers/block/loop.c
index 475e1a738560..8a07a89d702e 100644
--- a/drivers/block/loop.c
+++ b/drivers/block/loop.c

@@ -895,6 +895,9 @@ static void loop_config_discard(struct loop_device *lo)
                blk_queue_max_write_zeroes_sectors(q,
                        backingq->limits.max_write_zeroes_sectors);
 
+               q->limits.discard_granularity =
+                       backingq->limits.discard_granularity;
+
        /*
         * We use punch hole to reclaim the free space used by the
         * image a.k.a. discard. However we do not support discard if