Re: [PATCH] mtd: Add simple read disturb test

[Brian Norris <computersforpeace@gmail.com>]

Resurrecting this old thread, since it was mentioned at ELCE.

On Sun, Apr 12, 2015 at 09:31:20PM +0200, Boris Brezillon wrote:
> On Thu, 02 Apr 2015 18:18:34 +0200
> Richard Weinberger <richard@nod.at> wrote:
> > Am 02.04.2015 um 18:04 schrieb Brian Norris:
> > > On Thu, Apr 02, 2015 at 04:13:46PM +0200, Richard Weinberger wrote:
> > >> This simple MTD tests allows the user to see when read disturb happens.
> > >> By reading blocks over and over it reports flipped bits.
> > >> Currently it reports only flipped bits of the worst page of a block.
> > >> If within block X page P1 has 3 bit flips and P6 4, it will report 4.
> > >> By default every 50th block is read.
> > > 
> > > Didn't read through this much yet, but why do we need another in-kernel
> > > test that coul (AFAICT) be easily replicated in userspace? The same goes
> > > for several of the other tests, I think, actually. But at least with
> > > those, we have a history of keeping them around, so it's not too much
> > > burden [1].
> > 
> > I've added the test to drivers/mtd/tests/ because it fits into.
> > As simple as that.
> > 
> > > Brian
> > > 
> > > [1] Although there are some latent issues in these tests that are still
> > > getting get worked out (e.g., bad handling of 64-bit casting; too large
> > > of stacks; uninterruptibility). The latter two would not even exist if
> > > we were in user space.
> > 
> > uninterruptibility got solved by my "[PATCH] mtd: Make MTD tests cancelable" patch.
> > 
> > But if we want to kill drivers/mtd/tests/ I'll happily help out.
> 
> I'd vote for that solution too.
> I've looked at in-kernel mtd tests, and I'm pretty sure they can all be
> done in userland.
> This would prevent any kernel crash caused by buggy test modules.  
> 
> > Where shall we move these tests into? mtd-utils?
> 
> I guess so, but I'll let Brian answer that one.
> How about dispatching them in mtd-utils' tests/ directory (some of them
> are NAND related tests, so creating a tests/nand would make sense,
> and others are more generic).

mtd-utils makes sense to me. If we're going to do this, let's make it a
policy to not add more to drivers/mtd/tests/ then. For instance, this
one [1]. Also, would we drop the in-kernel tests completely?

If we make the move, we'd need to make sure to update the documentation
(mtd-www.git).

Brian

[1] http://lists.infradead.org/pipermail/linux-mtd/2015-September/062237.html
    http://lists.infradead.org/pipermail/linux-mtd/2015-September/062236.html