Re: [PATCH] mm/zsmalloc: don't fail if can't create debugfs info

From: Minchan Kim
Date: Fri Apr 29 2016 - 11:33:23 EST


On Fri, Apr 29, 2016 at 10:50:13AM -0400, Dan Streetman wrote:
> On Fri, Apr 29, 2016 at 1:37 AM, Minchan Kim <minchan@xxxxxxxxxx> wrote:
> > On Fri, Apr 29, 2016 at 09:38:24AM +0900, Sergey Senozhatsky wrote:
> >> On (04/28/16 15:07), Andrew Morton wrote:
> >> > Needed a bit of tweaking due to
> >> > http://ozlabs.org/~akpm/mmotm/broken-out/zsmalloc-reordering-function-parameter.patch
> >>
> >> Thanks.
> >>
> >> > From: Dan Streetman <ddstreet@xxxxxxxx>
> >> > Subject: mm/zsmalloc: don't fail if can't create debugfs info
> >> >
> >> > Change the return type of zs_pool_stat_create() to void, and
> >> > remove the logic to abort pool creation if the stat debugfs
> >> > dir/file could not be created.
> >> >
> >> > The debugfs stat file is for debugging/information only, and doesn't
> >> > affect operation of zsmalloc; there is no reason to abort creating
> >> > the pool if the stat file can't be created. This was seen with
> >> > zswap, which used the same name for all pool creations, which caused
> >> > zsmalloc to fail to create a second pool for zswap if
> >> > CONFIG_ZSMALLOC_STAT was enabled.
> >>
> >> no real objections from me. given that both zram and zswap now provide
> >> unique names for zsmalloc stats dir, this patch does not fix any "real"
> >> (observed) problem /* ENOMEM in debugfs_create_dir() is a different
> >> case */. so it's more of a cosmetic patch.
> >>
> >
> > Logically, I agree with Dan that debugfs is just optional so it
> > shouldn't affect the module running *but* practically, debugfs_create_dir
> > failure with no memory would be rare. Rather than it, we would see
> > error from same entry naming like Dan's case.
> >
> > If we removes such error propagation logic in case of same naming,
> > how do zsmalloc user can notice that debugfs entry was not created
> > although zs_creation was successful returns success?
>
> Does it actually matter to the caller?
>
> Since there's no way for zsmalloc to know if the stats dir/file
> creation failed because of EEXIST or because of ENOMEM, there's no way
> for it to let the caller know why it failed, either. Thus all
> zsmalloc can do is return a generic error, or possibly-wrong ENOMEM.
> In that case what will the caller do? Change the name and try again?
> How does the caller know what name to change it to, maybe the new name
> is taken too?
>
> The point of debugfs is to provide debug; failures should be ignored,
> because it's just debug. It should never prevent actual operation of
> the driver.
>
> >
> > Otherwise, future user of zsmalloc can miss it easily if they repeates
> > same mistakes. So, what's the gain with this patch in real practice?
>
> Well as far as future users, zs_create_pool doesn't document 'name' at
> all, and certainly doesn't clarify that 'name' should be unique across
> *all* zs pools that exist. And zsmalloc behavior should not be
> different depending on whether the ZSMALLOC_STAT param - which appears
> to be a debug/info only param - is enabled or not.

Fair enough.

Then, could you apply it to zs_stat_init?
We could make zs_stat_init to return void to not affect module loading,
too. As well, we should be okay in zs_stat_root failue, too.

I hope your patch handles them all in this chance.

Thanks for the looking this.

>
> But after any future driver using zsmalloc is created, if it did use
> an already-existing name - either because it was not coded to use
> unique names, or because of a bug that reused an existing name - which
> is worse?
> 1) driver suddenly stops working because new zs pools can't be created?
> 2) statistics information isn't available for some of the pools created?
>
> And, even though zswap is now patched to provide a unique name, why
> does zswap have to bear the burden of that? zswap doesn't care at all
> about the pool name, and there's no way for users to tell which
> zsmalloc pool corresponds to which zswap pool parameters. Future
> users of zsmalloc (from a code point of view, not person) probably
> will also not care about the zsmalloc pool name. And zsmalloc still
> logs the failure - so anyone looking for the stats and not finding it
> can easily check the logs to see the reason.
>
> The other alternative that I mentioned before, is for zsmalloc to take
> care of the problem itself. If the debugfs dir creation fails, it
> should change the name and retry; or zsmalloc can keep a list of
> active pool names so it knows if a new pool's name exists already or
> not. But neither expecting the calling code to retry with a different
> name, nor failing pool creation, seem like a good response when the
> only failure is providing debug/stats information.
>