Hi Neil
I saw your patch "md/raid5: avoid livelock caused by non-aligned writes".
That patch is only for RAID 6, but we are seeing similar livelock on RAID5
(mdX_raid5 is constantly looping in
raid5d/handle_active_stripes/handle_stripe/ops_run_io and some submitted
bios get stuck). The livelock happens when we issue I/O and simulate disk
failure.
Do you think that the same issue could happen on RAID5 too?
Do you think that the patch "md/raid5: fetch_block must fetch all the
blocks handle_stripe_dirtying wants." could fix it?
Mikulas
On Mon, 19 Jan 2015, Mikulas Patocka wrote:
Hi Neil
I saw your patch "md/raid5: avoid livelock caused by non-aligned writes".
That patch is only for RAID 6, but we are seeing similar livelock on RAID5
(mdX_raid5 is constantly looping in
raid5d/handle_active_stripes/handle_stripe/ops_run_io and some submitted
bios get stuck). The livelock happens when we issue I/O and simulate disk
failure.
Do you think that the same issue could happen on RAID5 too?
Do you think that the patch "md/raid5: fetch_block must fetch all the
blocks handle_stripe_dirtying wants." could fix it?
Mikulas
So, we verified that this RAID5 livelock was introduced in patch
67f455486d2ea20b2d94d6adf5b9b783d079e321. A patch
a40687ff73a5b14909d6aa522f7d778b158911c5 fixes a livelock in RAID6, but a
similar livelock was introduced in RAID5, that is still unfixed upstream.
Mikulas