Message ID | 20190522180446.GA30082@embeddedor |
---|---|
State | Accepted |
Delegated to: | Miquel Raynal |
Headers | show |
Series | mtd: onenand_base: Avoid fall-through warnings | expand |
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; 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: #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?) Setting a "break" after 4Gb may remove ONENAND_HAS_UNLOCK_ALL in the !FLEXONENAND(this) case. Does anyone have real hardware to test with? Thoughts?
On Wed, 22 May 2019 13:04:46 -0500 "Gustavo A. R. Silva" <gustavo@embeddedor.com> wrote: > NOTICE THAT: > > "...we don't know whether we need fallthroughs or breaks there and this > is just a change to avoid having new warnings when switching to > -Wimplicit-fallthrough but this change might be entirely wrong."[1] > > See the original thread of discussion here: > > https://lore.kernel.org/patchwork/patch/1036251/ > > So, in preparation to enabling -Wimplicit-fallthrough, this patch silences > the following warnings: > > drivers/mtd/nand/onenand/onenand_base.c: In function ‘onenand_check_features’: > drivers/mtd/nand/onenand/onenand_base.c:3264:6: warning: this statement may fall through [-Wimplicit-fallthrough=] > if (ONENAND_IS_DDP(this)) > ^ > drivers/mtd/nand/onenand/onenand_base.c:3284:2: note: here > case ONENAND_DEVICE_DENSITY_2Gb: > ^~~~ > drivers/mtd/nand/onenand/onenand_base.c:3288:17: warning: this statement may fall through [-Wimplicit-fallthrough=] > this->options |= ONENAND_HAS_UNLOCK_ALL; > drivers/mtd/nand/onenand/onenand_base.c:3290:2: note: here > case ONENAND_DEVICE_DENSITY_1Gb: > ^~~~ > > Warning level 3 was used: -Wimplicit-fallthrough=3 > > This patch is part of the ongoing efforts to enable > -Wimplicit-fallthrough. > > [1] https://lore.kernel.org/lkml/20190509085318.34a9d4be@xps13/ > > Cc: Miquel Raynal <miquel.raynal@bootlin.com> > Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com> > --- > drivers/mtd/nand/onenand/onenand_base.c | 2 ++ > 1 file changed, 2 insertions(+) > > 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) > if ((this->version_id & 0xf) == 0xe) > this->options |= ONENAND_HAS_NOP_1; > } > + /* Fall through - ? */ So, the only thing that you'll re-use by falling through the next case is the '->options |= ONENAND_HAS_UNLOCK_ALL' operation. I find it easier to follow with an explicit copy of this line + a break. > > 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 - ? */ This fall through certainly doesn't make sense, as the only thing that might be done in the 1Gb case is conditionally adding the HAS_UNLOCK_ALL flag, and this flag is already unconditionally set. Please add a break here. > > case ONENAND_DEVICE_DENSITY_1Gb: > /* A-Die has all block unlock */
On Wed, May 22, 2019 at 11:37:05PM +0200, Boris Brezillon wrote: > > @@ -3280,12 +3280,14 @@ static void onenand_check_features(struct mtd_info *mtd) > > if ((this->version_id & 0xf) == 0xe) > > this->options |= ONENAND_HAS_NOP_1; > > } > > + /* Fall through - ? */ > > So, the only thing that you'll re-use by falling through the next case > is the '->options |= ONENAND_HAS_UNLOCK_ALL' operation. I find it easier > to follow with an explicit copy of this line + a break. > > > > > 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 - ? */ > > This fall through certainly doesn't make sense, as the only thing that > might be done in the 1Gb case is conditionally adding the > HAS_UNLOCK_ALL flag, and this flag is already unconditionally set. > Please add a break here. > > > > > case ONENAND_DEVICE_DENSITY_1Gb: > > /* A-Die has all block unlock */ > Your reply was much more to-the-point than mine. :) I'd agree: retain existing behavior (ONENAND_HAS_UNLOCK_ALL) and add breaks.
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.
On Wed, May 22, 2019 at 11:57:38PM +0200, Boris Brezillon wrote: > 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. Yeah, totally agreed. :)
On 5/22/19 5:20 PM, Kees Cook wrote: > On Wed, May 22, 2019 at 11:57:38PM +0200, Boris Brezillon wrote: >> 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. > > Yeah, totally agreed. :) > Thank you, Kees and Boris for the feedback. Some external opinions were certainly much needed here. I'll just wait for any comments from the MTD guys. Thanks -- Gustavo
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) if ((this->version_id & 0xf) == 0xe) this->options |= ONENAND_HAS_NOP_1; } + /* Fall through - ? */ 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 */
NOTICE THAT: "...we don't know whether we need fallthroughs or breaks there and this is just a change to avoid having new warnings when switching to -Wimplicit-fallthrough but this change might be entirely wrong."[1] See the original thread of discussion here: https://lore.kernel.org/patchwork/patch/1036251/ So, in preparation to enabling -Wimplicit-fallthrough, this patch silences the following warnings: drivers/mtd/nand/onenand/onenand_base.c: In function ‘onenand_check_features’: drivers/mtd/nand/onenand/onenand_base.c:3264:6: warning: this statement may fall through [-Wimplicit-fallthrough=] if (ONENAND_IS_DDP(this)) ^ drivers/mtd/nand/onenand/onenand_base.c:3284:2: note: here case ONENAND_DEVICE_DENSITY_2Gb: ^~~~ drivers/mtd/nand/onenand/onenand_base.c:3288:17: warning: this statement may fall through [-Wimplicit-fallthrough=] this->options |= ONENAND_HAS_UNLOCK_ALL; drivers/mtd/nand/onenand/onenand_base.c:3290:2: note: here case ONENAND_DEVICE_DENSITY_1Gb: ^~~~ Warning level 3 was used: -Wimplicit-fallthrough=3 This patch is part of the ongoing efforts to enable -Wimplicit-fallthrough. [1] https://lore.kernel.org/lkml/20190509085318.34a9d4be@xps13/ Cc: Miquel Raynal <miquel.raynal@bootlin.com> Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com> --- drivers/mtd/nand/onenand/onenand_base.c | 2 ++ 1 file changed, 2 insertions(+)