lkml.org 
[lkml]   [2015]   [Apr]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 4/4] UBI: Implement bitrot checking (linux-mtd Digest, Vol 145, Issue 24)
Il 12/04/2015 23:01, Richard Weinberger ha scritto:
> Am 12.04.2015 um 22:42 schrieb Andrea Scian:
>> Il 12/04/2015 18:55, Richard Weinberger ha scritto:
>>> Am 12.04.2015 um 18:43 schrieb Boris Brezillon:
>>>> On Sun, 12 Apr 2015 18:09:23 +0200
>>>> Richard Weinberger <richard@nod.at> wrote:
>>>>
>>>>> Am 12.04.2015 um 16:12 schrieb Boris Brezillon:
>>>>>> Hi Richard,
>>>>>>
>>>>>> Sorry for the late reply.
>>>>>>
>>>>>> On Sun, 29 Mar 2015 14:13:17 +0200
>>>>>> Richard Weinberger <richard@nod.at> wrote:
>>>>>>
>>>>>>> This patch implements bitrot checking for UBI.
>>>>>>> ubi_wl_trigger_bitrot_check() triggers a re-read of every
>>>>>>> PEB. If a bitflip is detected PEBs in use will get scrubbed
>>>>>>> and free ones erased.
>>>>>> As you'll see, I didn't have much to say about the 'UBI bitrot
>>>>>> detection' mechanism, so this review is a collection of
>>>>>> nitpicks :-).
>>>>>>
>>>>>>> Signed-off-by: Richard Weinberger <richard@nod.at>
>>>>>>> ---
>>>>>>> drivers/mtd/ubi/build.c | 39 +++++++++++++
>>>>>>> drivers/mtd/ubi/ubi.h | 4 ++
>>>>>>> drivers/mtd/ubi/wl.c | 146 ++++++++++++++++++++++++++++++++++++++++++++++++
>>>>>>> 3 files changed, 189 insertions(+)
>>>>>>>
>>>>>>> diff --git a/drivers/mtd/ubi/build.c b/drivers/mtd/ubi/build.c
>>>>>>> index 9690cf9..f58330b 100644
>>>>>>> --- a/drivers/mtd/ubi/build.c
>>>>>>> +++ b/drivers/mtd/ubi/build.c
>>>>>>> @@ -118,6 +118,9 @@ static struct class_attribute ubi_version =
>>>>>>>
>>>>>>> static ssize_t dev_attribute_show(struct device *dev,
>>>>>>> struct device_attribute *attr, char *buf);
>>>>>>> +static ssize_t trigger_bitrot_check(struct device *dev,
>>>>>>> + struct device_attribute *mattr,
>>>>>>> + const char *data, size_t count);
>>>>>>>
>>>>>>> /* UBI device attributes (correspond to files in '/<sysfs>/class/ubi/ubiX') */
>>>>>>> static struct device_attribute dev_eraseblock_size =
>>>>>>> @@ -142,6 +145,8 @@ static struct device_attribute dev_bgt_enabled =
>>>>>>> __ATTR(bgt_enabled, S_IRUGO, dev_attribute_show, NULL);
>>>>>>> static struct device_attribute dev_mtd_num =
>>>>>>> __ATTR(mtd_num, S_IRUGO, dev_attribute_show, NULL);
>>>>>>> +static struct device_attribute dev_trigger_bitrot_check =
>>>>>>> + __ATTR(trigger_bitrot_check, S_IWUSR, NULL, trigger_bitrot_check);
>>>>>> How about making this attribute a RW one, so that users could check
>>>>>> if there's a bitrot check in progress.
>>>>> As the check will be initiated only by userspace and writing to the trigger
>>>>> while a check is running will return anyway a EBUSY I don't really see
>>>>> a point why userspace would check for it.
>>>> Sometime you just want to know whether something is running or not (in
>>>> this case the bitrot check) without risking to trigger a new action...
>>> Why would they care?
>> I think is always useful to give some additional information in userspace, from both debugging and diagnostic point of view.
> The question is, why does userspace care?

Because the userspace trigger it

> Other UBI operations are also not visible...

I don't really know much about UBI internals, but I think not many
operation are triggered from userspace (apart from ubiformat and mount ;-) )


>>> But I can add this feature, no problem.
>> Thanks ;-)
>>
>> May I ask if can be useful to abort the (IMHO quite long running) operation?
>> I think it can be useful to save power, e.g. when running on batteries: smart systems will trigger the operation when charging and aborting it if on batteries (or on low batteries).
> If the system is running on low power mode just don't trigger the run...
> Userspace controls it.

It heavily depends on how long the operation takes, we may have 4 to 32
GiB devices so I think it may take more than just a few minutes to scan
the whole device (and additional time depending on how many scrub
operation are needed).
You may start the operation when power is not an issue (e.g. while
charging) and when it is (e.g. when running on batteries and you need to
save any mAh you have ;-) ) it may be still running for a while..
I know that this is some kind of advance feature, but I would like to
point this out for comments.

>
>> What happens if the system need to reboot in the middle of scanning?
> Just reboot, UBI can handle that. Work will be canceled.

Thanks for the confirm

>
>> Probably nothing at all but I think it's worth asking ;-)
>> Anyway I think it's better if we can, on runlevel 6, shutdown the operation in a clean way
>>
>> To ask a little bit more from the current implementation, can it be useful expand sysfs entry with the current status (stopped, running, completed)?
>> In this way the userspace knows whenever the operation it has triggered, it completed successfully or something interrupt it (e.g. an internal error). I will schedule a new
>> operation sooner if I have no evidence that the last one completed successfully.. WDYT?
>> But maybe all of this stuff will be implemented inside a daemon with additional ioctl() (IIRC Richard already is working on this).
> That's the plan. The interface proposed in that patch series it designed to be a simple replacement for the dd if=/dev/ubiXY of=/dev/null hack.

dd can be stopped if needed and you may also have the progress status :-)

for sure you probably don't need all of the above additional stuff but
it may be useful anyway
Maybe it's better to have an initial working implementation and add some
more (backward compatible?) features later.

> The next step is adding a more advanced ioctl() interface to support a clever deamon.

Kind Regards and thanks for you comments,

--

Andrea SCIAN

DAVE Embedded Systems



\
 
 \ /
  Last update: 2015-04-12 23:41    [W:0.107 / U:2.052 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site