Re: [PATCH FIXES/IMPROVEMENTS 0/7] block, bfq: preserve control, boost throughput, fix bugs

[Oleksandr Natalenko]

Hello.

On sobota 19. června 2021 16:09:41 CEST Paolo Valente wrote:
> Hi Jens,
> this series contains an already proposed patch by Luca, plus six new
> patches. The goals of these patches are summarized in the subject of
> this cover letter. I'm including Luca's patch here, because it enabled
> the actual use of stable merge, and, as such, triggered an otherwise
> silent bug. This series contains also the fix for that bug ("block,
> bfq: avoid delayed merge of async queues"), tested by Holger [1].
> 
> Thanks,
> Paolo
> 
> [1] https://lkml.org/lkml/2021/5/18/384
> 
> Luca Mariotti (1):
>   block, bfq: fix delayed stable merge check
> 
> Paolo Valente (5):
>   block, bfq: let also stably merged queues enjoy weight raising
>   block, bfq: consider also creation time in delayed stable merge
>   block, bfq: avoid delayed merge of async queues
>   block, bfq: check waker only for queues with no in-flight I/O
>   block, bfq: reset waker pointer with shared queues
> 
> Pietro Pedroni (1):
>   block, bfq: boost throughput by extending queue-merging times
> 
>  block/bfq-iosched.c | 68 +++++++++++++++++++++++++++++++++++----------
>  1 file changed, 53 insertions(+), 15 deletions(-)
> 
> --
> 2.20.1

Not sure everything goes fine here. After applying this series on top of the 
latest stable 5.12 kernel I got this:

```
[16730.963248] kernel BUG at block/elevator.c:236!
[16730.963254] invalid opcode: 0000 [#1] PREEMPT SMP NOPTI
[16730.963257] CPU: 11 PID: 109170 Comm: kworker/u64:5 Tainted: G        W         
5.12.0-pf7 #1
[16730.963260] Hardware name: ASUS System Product Name/Pro WS X570-ACE, BIOS 
3601 05/26/2021
[16730.963263] Workqueue: dm-thin do_worker [dm_thin_pool]
[16730.963270] RIP: 0010:elv_rqhash_find+0xcc/0xd0
[16730.963274] Code: 41 89 f0 81 e2 00 40 06 00 41 81 e0 1a 00 04 00 44 09 c2 
75 a9 be 09 00 00 00 c4 e2 4b f7 50 28 48 03 50 30 48 39 fa 75 c6 c3 <0f> 0b 
66 90 0f 1f 44 00 00 41 56 41 55 41 54 55 53 48 8b 47 68 48
[16730.963276] RSP: 0018:ffffa558d13b7af8 EFLAGS: 00010046
[16730.963279] RAX: ffff8a0007782d00 RBX: ffff8a0014b93000 RCX: ffffa558d13b7b78
[16730.963281] RDX: ffff8a0014b93000 RSI: 0000000000063082 RDI: 000000001e0fdc00
[16730.963283] RBP: ffff8a000731c770 R08: ffff8a000731c770 R09: fffffff0ffffddfb
[16730.963284] R10: 0000000000000000 R11: 0000000000000400 R12: ffff8a0330365c00
[16730.963286] R13: ffffa558d13b7b30 R14: 0000000000000000 R15: ffff8a0212fc4000
[16730.963288] FS:  0000000000000000(0000) GS:ffff8a070ecc0000(0000) knlGS:
0000000000000000
[16730.963290] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[16730.963292] CR2: 00007f1514d90f4c CR3: 0000000315952000 CR4: 
0000000000350ee0
[16730.963294] Call Trace:
[16730.963297]  elv_merge+0x96/0x120
[16730.963300]  blk_mq_sched_try_merge+0x3e/0x370
[16730.963303]  bfq_bio_merge+0xd3/0x130
[16730.963306]  blk_mq_submit_bio+0x11e/0x6c0
[16730.963309]  submit_bio_noacct+0x457/0x530
[16730.963312]  raid10_unplug+0x13f/0x1a0 [raid10]
[16730.963316]  blk_flush_plug_list+0xa9/0x110
[16730.963319]  blk_finish_plug+0x21/0x30
[16730.963322]  process_prepared_discard_passdown_pt1+0x204/0x2d0 
[dm_thin_pool]
[16730.963327]  do_worker+0x18e/0xce0 [dm_thin_pool]
[16730.963335]  process_one_work+0x217/0x3e0
[16730.963338]  worker_thread+0x4d/0x470
[16730.963343]  kthread+0x182/0x1b0
[16730.963349]  ret_from_fork+0x22/0x30
…
[16730.963419] ---[ end trace dd7e037f2028257b ]---
[16730.963524] RIP: 0010:elv_rqhash_find+0xcc/0xd0
…
[16730.963547] note: kworker/u64:5[109170] exited with preempt_count 1
[16747.948467] [drm:amdgpu_dm_atomic_commit_tail [amdgpu]] *ERROR* Waiting for 
fences timed out!
```

Which is:

```
229 struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
230 {
…
235     hash_for_each_possible_safe(e->hash, rq, next, hash, offset) {
236         BUG_ON(!ELV_ON_HASH(rq));
…
```

Yes, I carry some extra patches besides this series (the list is against v5.12 
GA):

```
block, bfq: reset waker pointer with shared queues
block, bfq: check waker only for queues with no in-flight I/O
block, bfq: avoid delayed merge of async queues
block, bfq: boost throughput by extending queue-merging times
block, bfq: consider also creation time in delayed stable merge
block, bfq: fix delayed stable merge check
block, bfq: let also stably merged queues enjoy weight raising
block: Do not pull requests from the scheduler when we cannot dispatch them
blk: Fix lock inversion between ioc lock and bfqd lock
bfq: Remove merged request already in bfq_requests_merged()
block, bfq: avoid circular stable merges
bfq: remove unnecessary BFQ_DEFAULT_GRP_IOPRIO
bfq: reset entity->prio_changed in bfq_init_entity()
bfq: optimize the calculation of bfq_weight_to_ioprio()
bfq: remove unnecessary initialization logic
bfq: keep the minimun bandwidth for CLASS_BE
bfq: limit the IO depth of CLASS_IDLE to 1
bfq: convert the type of bfq_group.bfqd to bfq_data*
bfq: introduce bfq_entity_to_bfqg helper method
bfq/mq-deadline: remove redundant check for passthrough request
blk-mq: bypass IO scheduler's limit_depth for passthrough request
block,bfq: fix the timeout calculation in bfq_bfqq_charge_time
block, bfq: merge bursts of newly-created queues
block, bfq: keep shared queues out of the waker mechanism
block, bfq: fix weight-raising resume with !low_latency
block, bfq: make shared queues inherit wakers
block, bfq: put reqs of waker and woken in dispatch list
block, bfq: always inject I/O of queues blocked by wakers
```

but nothing from there triggered this for quite some time.

Paolo, what do you think?

-- 
Oleksandr Natalenko (post-factum)