lkml.org 
[lkml]   [2019]   [May]   [22]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] mtd: onenand_base: Avoid fall-through warnings
On Wed, 22 May 2019 14:30:11 -0700
Kees Cook <keescook@chromium.org> wrote:

> Sorry for being late to speaking up on this. I missed something in the
> code the first time I read the thread, that now stood out to me. Notes
> below...
>
> On Wed, May 22, 2019 at 01:04:46PM -0500, Gustavo A. R. Silva wrote:
> > diff --git a/drivers/mtd/nand/onenand/onenand_base.c b/drivers/mtd/nand/onenand/onenand_base.c
> > index f41d76248550..6cf4df9f8c01 100644
> > --- a/drivers/mtd/nand/onenand/onenand_base.c
> > +++ b/drivers/mtd/nand/onenand/onenand_base.c
> > @@ -3280,12 +3280,14 @@ static void onenand_check_features(struct mtd_info *mtd)
>
> Reverse-order review, second hunk first:
>
> > case ONENAND_DEVICE_DENSITY_2Gb:
> > /* 2Gb DDP does not have 2 plane */
> > if (!ONENAND_IS_DDP(this))
> > this->options |= ONENAND_HAS_2PLANE;
> > this->options |= ONENAND_HAS_UNLOCK_ALL;
> > + /* Fall through - ? */
> >
> > case ONENAND_DEVICE_DENSITY_1Gb:
> > /* A-Die has all block unlock */
>
> So, I think the ONENAND_DEVICE_DENSITY_2Gb should be a "break". Though,
> actually, it doesn't matter:
>
> case ONENAND_DEVICE_DENSITY_2Gb:
> /* 2Gb DDP does not have 2 plane */
> if (!ONENAND_IS_DDP(this))
> this->options |= ONENAND_HAS_2PLANE;
> this->options |= ONENAND_HAS_UNLOCK_ALL;
>
> case ONENAND_DEVICE_DENSITY_1Gb:
> /* A-Die has all block unlock */
> if (process)
> this->options |= ONENAND_HAS_UNLOCK_ALL;
> break;
>
> Falling through from ONENAND_DEVICE_DENSITY_2Gb to
> ONENAND_DEVICE_DENSITY_1Gb will actually have no side-effects:
> ONENAND_HAS_UNLOCK_ALL was unconditionally set in ..._2Gb, so there is
> no reason to fall through to ..._1Gb. (But falling through is harmless.)
>
> Now the first hunk:
>
> > if ((this->version_id & 0xf) == 0xe)
> > this->options |= ONENAND_HAS_NOP_1;
> > }
> > + /* Fall through - ? */
> >
>
> case ONENAND_DEVICE_DENSITY_4Gb:
> if (ONENAND_IS_DDP(this))
> this->options |= ONENAND_HAS_2PLANE;
> else if (numbufs == 1) {
> this->options |= ONENAND_HAS_4KB_PAGE;
> this->options |= ONENAND_HAS_CACHE_PROGRAM;
> /*
> * There are two different 4KiB pagesize chips
> * and no way to detect it by H/W config values.
> *
> * To detect the correct NOP for each chips,
> * It should check the version ID as workaround.
> *
> * Now it has as following
> * KFM4G16Q4M has NOP 4 with version ID 0x0131
> * KFM4G16Q5M has NOP 1 with versoin ID 0x013e
> */
> if ((this->version_id & 0xf) == 0xe)
> this->options |= ONENAND_HAS_NOP_1;
> }
>
> Falling through from ONENAND_DEVICE_DENSITY_4Gb to
> ONENAND_DEVICE_DENSITY_2Gb looks like it would mean that
> ONENAND_HAS_2PLANE would be unconditionally set for ...4Gb, which seems
> very strange to expect:
>
> if (ONENAND_IS_DDP(this))
> this->options |= ONENAND_HAS_2PLANE;
> ...
> if (!ONENAND_IS_DDP(this))
> this->options |= ONENAND_HAS_2PLANE;

Oops, didn't notice the ! on the second test.

>
> However! This happens later:
>
> if (ONENAND_IS_4KB_PAGE(this))
> this->options &= ~ONENAND_HAS_2PLANE;
>
> i.e. falling through to ...2Gb (which sets ONENAND_HAS_2PLANE) has no
> effect because when ONENAND_HAS_2PLANE isn't set (numbufs == 1), it gets
> _cleared_ by the above code due to ONENAND_HAS_4KB_PAGE getting set:

Are you sure !DDP implies num_bufs == 1?


>
> #define ONENAND_IS_4KB_PAGE(this) \
> (this->options & ONENAND_HAS_4KB_PAGE)
>
>
> Unfortunately, though, it's less clear about ONENAND_HAS_UNLOCK_ALL,
> which is getting set unconditionally for ...4Gb currently (due to the
> fallthrough to ...2Gb). However, this happens later:
>
> if (FLEXONENAND(this)) {
> this->options &= ~ONENAND_HAS_CONT_LOCK;
> this->options |= ONENAND_HAS_UNLOCK_ALL;
> }
> ...
> #define FLEXONENAND(this) \
> (this->device_id & DEVICE_IS_FLEXONENAND)
>
> So it's possible this fall through has no effect (are all 4Gb density
> devices also FLEXONENAND devices?)
>

All this look suspicious, and even if the fall through logic
has no side effects in practice (which I'm still not sure is the case),
I think it'd be better to explicitly set the flags that have
to be set in each case statement and add breaks.

\
 
 \ /
  Last update: 2019-05-22 23:58    [W:0.071 / U:0.216 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site