UBI: why was UBI reboot notifier removed

UBI: why was UBI reboot notifier removed

[Jose Nimni]

hi guys,

I wanted to ask you a question regarding the change you made in UBI, when you 
removed the UBI reboot notifier. (commit "UBI: remove reboot notifier")

I have omap3530 processor, working with numonyx NOR flashe PF2800AP33EF (intel 
command set).
I am working with linux omap kernel 2.6.34, but patched it with the newest UBI 
directory (i needed the new patches for NOR PEBs).

i have seen that if i write a lot data to the NOR flash, 2 things happen:
1. UBIFS thread are flushing the data to the flash for quite some time.
2. ubi_bgt0d is working after that, cleaning up some PEBs.

if i reboot just after the write process returned, the flashed are unmounted 
(after waiting for ubifs to finish flushing).
but when the sigkills are sent,  a UBI error is occuring:
UBI error: ubi_io_read: error -5 while reading 64 bytes from PEB x:y, read 0 
bytes.
UBI error: nor_erase_prepare: cannot invalidate PEB X, write returned -5 read 
returned -5
....
ubi_thread: ubi_bgt0d: work failed with error code -5.

after some research, i saw that you removed the UBI reboot notifier, that used 
to kill ubi_bgt thread.
now, during shutdown, the cfi_cmdset_0001.c driver does not allow any 
communication with the flash - but the bgt thread is still alive, so it gets a 
read/write error (-EIO).

I am working with UBI that is compiled into the kernel - not as a module, so 
module_exit isn't being called

who should stop the bgt threads?
am i missing anything or is this a bug?

Thanks,
Jose

Re: UBI: why was UBI reboot notifier removed

[Artem Bityutskiy]

Hi,

On Wed, 2011-03-30 at 07:43 -0700, Jose Nimni wrote:
> I wanted to ask you a question regarding the change you made in UBI, when you 
> removed the UBI reboot notifier. (commit "UBI: remove reboot notifier")
> 
> I have omap3530 processor, working with numonyx NOR flashe PF2800AP33EF (intel 
> command set).
> I am working with linux omap kernel 2.6.34, but patched it with the newest UBI 
> directory (i needed the new patches for NOR PEBs).
> 
> i have seen that if i write a lot data to the NOR flash, 2 things happen:
> 1. UBIFS thread are flushing the data to the flash for quite some time.
> 2. ubi_bgt0d is working after that, cleaning up some PEBs.
> 
> if i reboot just after the write process returned, the flashed are unmounted 
> (after waiting for ubifs to finish flushing).
> but when the sigkills are sent,  a UBI error is occuring:
> UBI error: ubi_io_read: error -5 while reading 64 bytes from PEB x:y, read 0 
> bytes.
> UBI error: nor_erase_prepare: cannot invalidate PEB X, write returned -5 read 
> returned -5
> ....
> ubi_thread: ubi_bgt0d: work failed with error code -5.

OK, reboot notifiers again :-)

About the question why they were removed? I think the primary driver was
that they cause issues with suspend, see here:

http://thread.gmane.org/gmane.linux.drivers.mtd/29143/focus=29681

Kevin provided an excellent description in the corresponding NOR commit
message:

https://groups.google.com/group/fa.linux.kernel/browse_thread/thread/cbe1baeb0cc906f9?fwc=1&hl=en&pli=1

> after some research, i saw that you removed the UBI reboot notifier, that used 
> to kill ubi_bgt thread.
> now, during shutdown, the cfi_cmdset_0001.c driver does not allow any 
> communication with the flash - but the bgt thread is still alive, so it gets a 
> read/write error (-EIO).

Well, in your case these messages are harmless, but I agree that this is
not nice. I think the best way to fix this would be to make the CFI code
return a special error code in this case, e.g., -EBUSY. Then UBI thread
could wait and re-try without printing warnings.

Could you do this?

-- 
Best Regards,
Artem Bityutskiy (Артём Битюцкий)