[PATCH 1/7] UBI: fix refill_wl_user_pool()

[Richard Weinberger]

Am 03.10.2013 18:41, schrieb Artem Bityutskiy:
> On Thu, 2013-10-03 at 18:35 +0200, Richard Weinberger wrote:
>> Am 03.10.2013 18:00, schrieb Artem Bityutskiy:
>>> On Thu, 2013-10-03 at 17:53 +0200, Richard Weinberger wrote:
>>>> Am 03.10.2013 17:27, schrieb Artem Bityutskiy:
>>>>> On Thu, 2013-10-03 at 17:08 +0200, Richard Weinberger wrote:
>>>>>> Am 03.10.2013 17:00, schrieb Artem Bityutskiy:
>>>>>>> On Sat, 2013-09-28 at 15:55 +0200, Richard Weinberger wrote:
>>>>>>>> If no free PEBs are available refill_wl_user_pool() must not
>>>>>>>> return with -ENOSPC immediately.
>>>>>>>> It has to block till produce_free_peb() produced a free PEB.
>>>>>>>>
>>>>>>>> Reported-and-Tested-by: Richard Genoud <richard.genoud at gmail.com>
>>>>>>>> Signed-off-by: Richard Weinberger <richard at nod.at>
>>>>>>>
>>>>>>> What is pool size, I wonder?
>>>>>>
>>>>>> Currently it's 25 (UBI_FM_WL_POOL_SIZE).
>>>>>> If experience shows that 25 is too low/big we can change this constant.
>>>>>> Maybe it's also worth making them configurable...
>>>>>
>>>>> I if it is a possible scenario that this function will not return until
>>>>> 25 (or even 10) PEBs are erased? 
>>>>>
>>>>
>>>> Sure. It will try to gain up to 25 PEBs but return if it gets less.
>>>> That's why a pool has a current and a max size.
>>>
>>> So if erasing speed is say, 250ms, then it would take 6.25 seconds?
>>
>> Only in the very worst case if we have to call 25 times produce_free_peb().
>>
>> Of course we could add a check to return immediately if produce_free_peb()
>> got called a few times in series.
>> But I really would like to wait with such performance tweaks until fastmap
>> is more mature.
> 
> OK, but how about at least adding a comment talking about this unlikely
> scenario?

I'm fine with this. Will send a patch. :-)

Thanks,
//richard