[kernel ext4 ] bug starting from 4.8.4-1

[Sergeanter]

This bug has been fixed by moonman in kernel 4.10.12-2
Reading block with address above 2^31 causes short read error starting from linux-kirkwood-dt-4.8.4-1. The bug is still present in 4.10.6
This has been checked rolling back to linux-kirkwood-dt-4.8.1-1. Also, I have verified that mdadm and e2fsprogs are not related to this issue. Please take a look at the code and I will try to go over this myself too when I figure how

[moonman]

Hmm can you check the gcc version the good and the bad versions were compiled with? It tells you in the first few lines of boot up.

[Sergeanter]

Both were compiled with gcc 6.2.1 datecode 20160830.
All I see in 4.8.1-1 to 4.8.4-1 history is the changes related to bfq scheduler that I do not use. Also, I have just checked the latest 4.10.8-1 that supposedly does not have bfq anymore and the issue is still there.
I saw this in posts related to ext4 and xfs:
"The Large Block Device (LBD and HAVE_SECTOR_T) patch" that is needed to avoid wraparound at 8TB.
this sets sector_t to long long , otherwise it would be "long" and overflow.
Can you list variables when you compile the latest kernel and check what is the type of :

ext4_fsblk_t (should be unsigned long long )
sector_t (should be unsigned long long )