Problem with initial sync of a RAID1 with 4Kn drives

Problem with initial sync of a RAID1 with 4Kn drives

[Tobias Jakobi]

Hello folks,

I recently got two new Seagate drives, Seagate Exos X X16, which are 
able to switch into native 4K mode. I used the Seagate tools to trigger 
the switch, and now want to use the two drives in a RAID1.

I proceeded to create the md device:
mdadm --create --verbose --level=1 --metadata=1.2 --raid-devices=2 
/dev/md/onyx-new /dev/sda /dev/sdb

As soon as the initial sync starts the kernel log begins to fill up with 
errors. Here's part of the kernel log:
https://gist.github.com/tobiasjakobi/d5b5f1638d23765d5e1fecf0dfbcea3e

This is a 5.10.11 vanilla kernel, but it also happens with the 5.9.y 
series. As you can see I already patched the kernel to include 
additional information:
DEBUG: 40801 1 7

This is just a simple print() in sd_setup_read_write_cmnd().
printk("DEBUG: %x %x %x\n", blk_rq_pos(rq), blk_rq_sectors(rq), mask);

So it seems that neither the position nor the number of sectors is 
aligned to eight. I assume that the md subsystem is supposed to handle this?

I then double-checked for sizes for the underlying block devices and the 
md device itself:
cat 
/sys/block/{sda,sdb,md126}/queue/{hw_sector_size,logical_block_size,physical_block_size}
4096

So it's 4K for all of them.

Here's the hdparm output for one of the drives:
https://gist.github.com/tobiasjakobi/a52cf5d4461db35c1f474c6a69b0782a

The drives themselves are fine, I can use them properly as standalone 
devices, so I rule out a hardware error.

Any hints how to further triage this? Also would it help to open a bug 
on the kernel bugtracker?

Thanks in advance!

With best wishes,
Tobias Jakobi