Re: ont out of 6 bcache devices does not register automatically

[Stefan Priebe - Profihost AG <s.priebe@profihost.ag>]

Am 22.11.2017 um 16:51 schrieb Coly Li:
> On 22/11/2017 10:16 PM, Stefan Priebe - Profihost AG wrote:
>> Hi Coly,
>> Am 22.11.2017 um 14:29 schrieb Coly Li:
>>> On 22/11/2017 9:14 PM, Stefan Priebe - Profihost AG wrote:
>>>> Am 22.11.2017 um 13:57 schrieb Coly Li:
>>>>> On 22/11/2017 8:26 PM, Stefan Priebe - Profihost AG wrote:
>>>>>>
>>>>>> Am 22.11.2017 um 13:16 schrieb Coly Li:
>>>>>>> On 22/11/2017 7:23 PM, Stefan Priebe - Profihost AG wrote:
>>>>>>>> Hello,
>>>>>>>>
>>>>>>>> i've 6 bcache blk devices attached to 3 caching ssds (2 each). One fails
>>>>>>>> to register automatically at boot time.
>>>>>>>>
>>>>>>>> After reboot i always need to execute:
>>>>>>>> echo /dev/sdf1 >/sys/fs/bcache/register
>>>>>>>>
>>>>>>>> to bring up the bcache device.
>>>>>>>>
>>>>>>>> Any idea?
>>>>>>>
>>>>>>> Hi Stefan,
>>>>>>>
>>>>>>> Is there any clue from kernel message ?
>>>>>>
>>>>>> Sadly not.
>>>>>>
>>>>>> Working one:
>>>>>> ]# dmesg | grep sdi
>>>>>> [    1.060377] sd 5:0:0:0: [sdi] 1953525168 512-byte logical blocks:
>>>>>> (1.00 TB/932 GiB)
>>>>>> [    1.060393] sd 5:0:0:0: [sdi] Write Protect is off
>>>>>> [    1.060396] sd 5:0:0:0: [sdi] Mode Sense: 00 3a 00 00
>>>>>> [    1.060425] sd 5:0:0:0: [sdi] Write cache: enabled, read cache:
>>>>>> enabled, doesn't support DPO or FUA
>>>>>> [    1.104341]  sdi: sdi1
>>>>>> [    1.104644] sd 5:0:0:0: [sdi] Attached SCSI disk
>>>>>> [    2.005452] bcache: register_bdev() registered backing device sdi1
>>>>>> [    2.045211] bcache: bch_cached_dev_attach() Caching sdi1 as bcache0
>>>>>> on set 76b95bf8-9cc7-407f-9f9e-a42b6d1bcb27
>>>>>>
>>>>>>
>>>>>> not working one:
>>>>>> # dmesg | grep sdf
>>>>>> [    0.946107] sd 0:0:5:0: [sdf] 1953525168 512-byte logical blocks:
>>>>>> (1.00 TB/932 GiB)
>>>>>> [    1.400267] sd 0:0:5:0: [sdf] Write Protect is off
>>>>>> [    1.400267] sd 0:0:5:0: [sdf] Mode Sense: 00 00 00 00
>>>>>> [    1.401347] sd 0:0:5:0: [sdf] Write cache: enabled, read cache:
>>>>>> enabled, supports DPO and FUA
>>>>>> [    1.902910]  sdf: sdf1
>>>>>> [    2.341289] sd 0:0:5:0: [sdf] Attached SCSI disk
>>>>>> [  295.458804] bcache: register_bdev() registered backing device sdf1
>>>>>> [  295.506656] bcache: bch_cached_dev_attach() Caching sdf1 as bcache5
>>>>>> on set 76b95bf8-9cc7-407f-9f9e-a42b6d1bcb27
>>>>>>
>>>>>> At 295s i registeeed it manually.
>>>>>
>>>>> Hi Stefan,
>>>>>
>>>>> Hmm, I don't have idea here. Anyway, is the cache mode set as writeback?
>>>>> Recently I post a patch to fix a potential deadlock between writeback
>>>>> rate update kworker and register code, I am not sure whether it is
>>>>> relative to your issue, but at least we can have a try. the patch title is:
>>>>>  [RFC] bcache: fix a circular dead locking with dc->writeback_lock and
>>>>> bch_register_lock
>>>>>
>>>>> Just for your information.
>>>>
>>>> i can try that one - can you please resend? I can't find that mail and
>>>> don't know how to grab that one our of the web html archives to apply
>>>> correctly.
>>>
>>> Hi Stefan,
>>>
>>> Can you download the patch from following URL:
>>> https://git.kernel.org/pub/scm/linux/kernel/git/colyli/bcache-patches.git/plain/for-test/0001-bcache-fix-a-circular-dead-locking-with-dc-writeback.patch
>>
>> thanks, sadly it does not help. More ideas? Is there an easy way to
>> debug the calls to probe-bcache and bcache-register from udev?
> 
> Hmm, udev is a miracle to me yet...
> 
> I guess maybe I happen to have 6 hard drives (from 120G to 2TB) as
> backing device, and 3 PCIe SSDs as cache device. Can you give me a
> detail configuration that how the bcache devices are configured and how
> the udev rules are deployed. I can also try to reproduce on my server.
> 
> Currently the hardware is occupied by other task, I can take a try for
> the registration issue if the hardware can be free next Monday.

Thanks Coly but there's no need to test this. I've another bunch of
servers running at least the same amount of bcache devices without any
problems. Just this one does not register.

Stefan

> 
> Thanks.
> 
> Coly Li
> 
>