Oracle Linux UEK: Horrible NVMe discard performance of UEK6R3 compared to RHCK 4.18.0-348.12.2

I have a system with 10 NVMe disks á 15 TB in an MD RAID10 with LVM and xfs filesystem.

With RHCK kernel, i can run fstrim on a 15 TB filesystem in 3 minutes. With UEK kernel, it takes more than 4 minutes for 10 MB!

In lsblk --discard, I can see that with RHCK kernel, column "DISC-MAX" is set to 2T for block device, partition, md device, and lv device, whereas in UEK kernel, it is set to 64k for RAID device and LV device.

NAME                     DISC-ALN DISC-GRAN DISC-MAX DISC-ZERO

nvme0n1                         0     512B      2T        0

└─nvme0n1p1                     0     512B      2T        0

 └─md125                       0     512B     64K        0

   ├─vg_nvmedata-lv_test       0     512B     64K        0