Date:   Fri, 3 Aug 2018 14:30:56 +0200
From:   Matias Bjørling <mb@...htnvm.io>
To:     javier@...xlabs.com
Cc:     hans.holmberg@...xlabs.com, linux-block@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH] lightnvm: pblk: recover chunk state on 1.2 devices

On 08/03/2018 02:02 PM, Javier Gonzalez wrote:
>> On 3 Aug 2018, at 13.57, Matias Bjørling <mb@...htnvm.io> wrote:
>>
>> On 07/24/2018 09:54 AM, Javier Gonzalez wrote:
>>>> On 29 Jun 2018, at 13.28, Matias Bjørling <mb@...htnvm.io> wrote:
>>>>
>>>> On 06/29/2018 01:22 PM, Javier Gonzalez wrote:
>>>>>> On 29 Jun 2018, at 13.14, Matias Bjørling <mb@...htnvm.io> wrote:
>>>>>>
>>>>>> On 06/28/2018 11:12 AM, Javier González wrote:
>>>>>>> The Open-Channel 1.2 spec does not define a mechanism for the host to
>>>>>>> recover the block (chunk) state. As a consequence, a newly format device
>>>>>>> will need to reconstruct the state. Currently, pblk assumes that blocks
>>>>>>> are not erased, which might cause double-erases in case that the device
>>>>>>> does not protect itself against them (which is not specified in the spec
>>>>>>> either).
>>>>>>
>>>>>> It should not be specified in the spec. It is up to the device to handle
>>>>>> double erases and not do it.
>>>>>>
>>>>>>> This patch, reconstructs the state based on read errors. If the first
>>>>>>> sector of a block returns and empty page (NVM_RSP_ERR_EMPTYPAGE), then
>>>>>>> the block s marked free, i.e., erased and ready to be used
>>>>>>> (NVM_CHK_ST_FREE). Otherwise, the block is marked as closed
>>>>>>> (NVM_CHK_ST_CLOSED). Note that even if a block is open and not fully
>>>>>>> written, it has to be erased in order to be used again.
>>>>>>
>>>>>> Should we extend it to do the scan, and update the write pointer as
>>>>>> well? I think this kind of feature already is baked into pblk?
>>>>> This is already in place: we scan until empty page and take it from
>>>>> there. This patch is only for the case in which we start a pblk instance
>>>>> form scratch. On a device already owned by pblk, we would not have the
>>>>> problem we are trying to solve here because we know the state.
>>>>
>>>> Agree. What I meant was that when we anyway are recovering the state,
>>>> we could just as well update ->wp and set to NVM_CHK_ST_OPEN and so
>>>> forth for the initialization phase.
>>> In 1.2 the use of chunk metadata is purely fictional. We respect the
>>> chunk state machine as we transition lines, but all the write pointers
>>> are ignored. Instead, we use the line bitmap to point to the next
>>> writable entry. This is BTW the same way we it in open lines on 2.0 too.
>>
>> Now I understand where you are coming from. I had the understanding
>> that we where using the write pointer now that we moved to 2.0,
>> looking through the code, that wasn't the case. :) Which means that
>> pblk doesn't work with a devices that implements 2.0. Yikes... I knew
>> I had forgot a detail when support was added into pblk.
>>
> 
> I think you misunderstood; pblk does support 2.0 devices. What happens
> is that we transform the per chunk WP in 2.0 into the line bitmap to
> simplify the lookup. The point being that we do not need to create a
> fictional chunk for 1.2 devices since we do the translation to the
> bitmap directly. Does this make sense?

The chunk->wp isn't used anywhere. So it can't take wp into account. It 
uses the EMPTYPAGE marker from 1.2 instead. See pblk-core and 
pblk-recovery.

> 
>> There are no empty sector marker in the 2.0 spec, since it uses the
>> write pointer to know where it is in the chunk. So there is a bit of
>> work to do there.
>>
> 
> Yes. And for 2.0 devices we go and look at the WP, but for 1.2 devices we
> need to scan.
> 
>> Since this properly is a bit more work to do, I'll look into it after FMS.
>>
> 
> Look the comments above. All we need for 2.0 support is in place. We can
> talk about it f2f.
> 
>> I'm also moving the explicit coding of 1.2/2.0 chunk / bad block
>> fixing into core, so pblk can be simplfied, and doesn't have to think
>> to manage each version separately.
>>
> 
> Good. I have a patch I was expecting to send after FMS for moving chunk
> / bad block out of pblk for the same reason. If you're doing the same
> thing I can stop looking into it...

I am, will post when done.

> 
>>
>>> Chunk metadata is only used to setup the bitmaps on init/recovery. From
>>> here on, we use the bitmap to find the next writable sector, without
>>> worrying about the specific per-chunk write pointer. Thus, updating
>>> chunk metadata here has no effect.
>>> Does this make sense to you?
>>> Javier

Powered by blists - more mailing lists