Re: [RFC][PATCH 7/7] MTD: UBI: wire up checkpointing

[Richard Weinberger <rw@linutronix.de>]

Am 19.02.2012 15:40, schrieb Shmulik Ladkani:
> On Sun, 19 Feb 2012 15:08:44 +0100 Richard Weinberger<rw@linutronix.de>  wrote:
>> Am 19.02.2012 14:57, schrieb Shmulik Ladkani:
>>> On Tue, 14 Feb 2012 21:06:46 +0100 Richard Weinberger<rw@linutronix.de>   wrote:
>>>> +#ifdef CONFIG_MTD_UBI_CHECKPOINT
>>>> +static int attach_by_checkpointing(struct ubi_device *ubi)
>>>> +{
>>>> +	int cp_start, err;
>>>> +	struct ubi_scan_info *si;
>>>> +
>>>> +	cp_start = ubi_find_checkpoint(ubi);
>>>> +	if (cp_start<   0)
>>>> +		return -ENOENT;
>>>> +
>>>> +	si = ubi_read_checkpoint(ubi, cp_start);
>>>> +	if (IS_ERR(si))
>>>> +		return PTR_ERR(si);
>>>> +
>>>> +	ubi->bad_peb_count = 0;
>>>> +	ubi->good_peb_count = ubi->peb_count;
>>>
>>> Zero reported bad PEBs when checkpointing.
>>> Seems that checkpointing does not remember number/location of bad PEBs.
>>
>> Currently checkpointing cares only about used and free PEBs.
>> Bad PEBs are no longer visible to UBI after recovering from a checkpoint.
>
> Ok.
> However it is still reported to the log in 'ubi_attach_mtd_dev'
> and as a sysfs attribute.
> BTW, the counter is still incremented by WL subsystem, though.
> Hence, reported value will be bad PEBs encountered since last attach
> (where formerly, it was absolute total bad PEBs in the ubi device).
> Maybe remove 'bad_peb_count' altogether.
>
> Also, "ubi->good_peb_count = ubi->peb_count" results in different
> 'beb_rsvd_level' caculation, see 'ubi_calculate_reserved'.

I can add these counters into the checkpoint.

>>> Are we fine with that?
>>
>> This patch is a RFC. :-)
>
> So I noticed :-)
> Just trying to point out things which cause ubi system to behave
> differently.
>

Thanks a lot for reviewing this feature!
//richard