Comment 60 for bug 1217189

Hence why I said I/O errors on *write*. If the block dev layer reports
an error to the VM on page-out, it knows the data hasn't been correctly
written.

I'm not sure zram actually allocates the whole amount of memory it's
configured to use, initially:

steved@xubuntu:~$ free
             total used free shared buffers cached
Mem: 2050796 1473456 577340 0 254232 696144
-/+ buffers/cache: 523080 1527716
Swap: 2975088 20848 2954240
steved@xubuntu:~$ swapon -s
Filename Type Size Used Priority
/dev/zram0 partition 1025396 20848 100
/dev/mapper/lvg2-swap partition 1949692 0 -1

At the very least, free seems to know it's cache memory. Don't know
whether that's merely a presentation thing, need to look at the code.

Anyway, the SNR here is getting awful, and I don't want to contribute to
it. Anyone who wants to can contact me privately via LP.

S.

On Fri, Oct 04, 2013 at 03:28:18PM -0000, Kenneth Parker wrote:
> Mel, the reason I do NOT believe this is a duplicate of # 1215513 is
> that many get error messages (including I/O errors in some cases)
> WITHOUT a hang. But any time I see I/O error on any disk, whether
> "hardware, ram or simulated [i.e. network]", there's a chance of data
> corruption, even without crashes. Please reconsider.
>
> Steve, you also mention I/O error messages. Once again, that could
> corrupt data, the WORST TYPE of error, in my NOT SO humble opinion.
> (Think about it: If SWAP data comes back in, and your "bank balance"
> changes from 5000 to 500,000, you might THINK you have more in your
> account than you do, causing "mishaps" in International Finance.
> [Example used, due to how badly your life could be hurt].
>
> My personal solution is to come up with ONLY a Root Text console [yes,
> Ubuntu users aren't SUPPOSED to become "real Root", but I've been doing
> this sort of thing since the 1990's], "rmmod zram" [with prejudice, if
> possible], and I'm prepared to do a "permanent rename" of the correct
> name of the module, as soon as I "get around" to it. [Visions of Round
> Tuits...]
>
> A point I made earlier: Even if a SWAP partition is NOT made out of
> /dev/zram0, part of RAM is being "reserved for nothing".
>
> Thank you and best regards,
>
> Kenneth Parker, Seattle, WA
>
> --
> You received this bug notification because you are subscribed to the bug
> report.
> https://bugs.launchpad.net/bugs/1217189
>
> Title:
> Buffer I/O error on device zram0
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1217189/+subscriptions