GCC 12 warnings

GCC 12 warnings

[Jakub Kicinski]

Hi Kees,

I'm sure you're involved in a number of glorious GCC 12 conversations..

We have a handful of drivers in networking which get hit by
-Warray-bounds because they allocate partial structures (I presume 
to save memory, misguided but more than 15min of work to refactor).

Since -Warray-bounds is included by default now this is making our
lives a little hard [1]. Is there a wider effort to address this?
If not do you have a recommendation on how to deal with it?

My best idea is to try to isolate the bad files and punt -Warray-bounds
to W=1 for those, so we can prevent more of them getting in but not
break WERROR builds on GCC 12. That said, I'm not sure how to achieve
that.. This for example did not work:

--- a/drivers/net/ethernet/mediatek/Makefile
+++ b/drivers/net/ethernet/mediatek/Makefile
@@ -9,5 +9,9 @@ mtk_eth-$(CONFIG_NET_MEDIATEK_SOC_WED) += mtk_wed.o
 ifdef CONFIG_DEBUG_FS
 mtk_eth-$(CONFIG_NET_MEDIATEK_SOC_WED) += mtk_wed_debugfs.o
 endif
 obj-$(CONFIG_NET_MEDIATEK_SOC_WED) += mtk_wed_ops.o
 obj-$(CONFIG_NET_MEDIATEK_STAR_EMAC) += mtk_star_emac.o
+
+ifneq ($(findstring 1, $(KBUILD_EXTRA_WARN)),)
+CFLAGS_mtk_ppe.o += -Wno-array-bounds
+endif

[1]
https://lore.kernel.org/all/20220520012555.2262461-1-kuba@kernel.org/

Re: GCC 12 warnings

[Kees Cook]

On Thu, May 19, 2022 at 07:36:18PM -0700, Jakub Kicinski wrote:
> I'm sure you're involved in a number of glorious GCC 12 conversations..

Yeah, and I think I've even found a gcc bug. :|

> We have a handful of drivers in networking which get hit by
> -Warray-bounds because they allocate partial structures (I presume 
> to save memory, misguided but more than 15min of work to refactor).

Yeah, this idiom is pretty common I've noticed -- I fixed a few of these
in the initial work for -Warray-bounds on GCC 11 and earlier, but wow
did GCC 12 do something extra internally.

> Since -Warray-bounds is included by default now this is making our
> lives a little hard [1]. Is there a wider effort to address this?
> If not do you have a recommendation on how to deal with it?

Looks like the issue was this?
https://lore.kernel.org/all/20220520145957.1ec50e44@canb.auug.org.au/

Ah, from cf2df74e202d ("net: fix dev_fill_forward_path with pppoe + bridge")

You mean you missed this particular warning because of the other GCC
12 warnings?

> My best idea is to try to isolate the bad files and punt -Warray-bounds
> to W=1 for those, so we can prevent more of them getting in but not
> break WERROR builds on GCC 12. That said, I'm not sure how to achieve
> that.. This for example did not work:
> 
> --- a/drivers/net/ethernet/mediatek/Makefile
> +++ b/drivers/net/ethernet/mediatek/Makefile
> @@ -9,5 +9,9 @@ mtk_eth-$(CONFIG_NET_MEDIATEK_SOC_WED) += mtk_wed.o
>  ifdef CONFIG_DEBUG_FS
>  mtk_eth-$(CONFIG_NET_MEDIATEK_SOC_WED) += mtk_wed_debugfs.o
>  endif
>  obj-$(CONFIG_NET_MEDIATEK_SOC_WED) += mtk_wed_ops.o
>  obj-$(CONFIG_NET_MEDIATEK_STAR_EMAC) += mtk_star_emac.o
> +
> +ifneq ($(findstring 1, $(KBUILD_EXTRA_WARN)),)
> +CFLAGS_mtk_ppe.o += -Wno-array-bounds
> +endif

This worked for me:

diff --git a/drivers/net/can/usb/kvaser_usb/Makefile b/drivers/net/can/usb/kvaser_usb/Makefile
index cf260044f0b9..43eb921f9102 100644
--- a/drivers/net/can/usb/kvaser_usb/Makefile
+++ b/drivers/net/can/usb/kvaser_usb/Makefile
@@ -1,3 +1,7 @@
 # SPDX-License-Identifier: GPL-2.0-only
 obj-$(CONFIG_CAN_KVASER_USB) += kvaser_usb.o
 kvaser_usb-y = kvaser_usb_core.o kvaser_usb_leaf.o kvaser_usb_hydra.o
+
+ifeq ($(KBUILD_EXTRA_WARN),)
+CFLAGS_kvaser_usb_hydra.o += -Wno-array-bounds
+endif

-- 
Kees Cook

Re: GCC 12 warnings

[Jakub Kicinski]

On Fri, 20 May 2022 09:46:36 -0700 Kees Cook wrote:
> > Since -Warray-bounds is included by default now this is making our
> > lives a little hard [1]. Is there a wider effort to address this?
> > If not do you have a recommendation on how to deal with it?  
> 
> Looks like the issue was this?
> https://lore.kernel.org/all/20220520145957.1ec50e44@canb.auug.org.au/
> 
> Ah, from cf2df74e202d ("net: fix dev_fill_forward_path with pppoe + bridge")
> 
> You mean you missed this particular warning because of the other GCC
> 12 warnings?

Yup :(

> > My best idea is to try to isolate the bad files and punt -Warray-bounds
> > to W=1 for those, so we can prevent more of them getting in but not
> > break WERROR builds on GCC 12. That said, I'm not sure how to achieve
> > that.. This for example did not work:
> > 
> > --- a/drivers/net/ethernet/mediatek/Makefile
> > +++ b/drivers/net/ethernet/mediatek/Makefile
> > @@ -9,5 +9,9 @@ mtk_eth-$(CONFIG_NET_MEDIATEK_SOC_WED) += mtk_wed.o
> >  ifdef CONFIG_DEBUG_FS
> >  mtk_eth-$(CONFIG_NET_MEDIATEK_SOC_WED) += mtk_wed_debugfs.o
> >  endif
> >  obj-$(CONFIG_NET_MEDIATEK_SOC_WED) += mtk_wed_ops.o
> >  obj-$(CONFIG_NET_MEDIATEK_STAR_EMAC) += mtk_star_emac.o
> > +
> > +ifneq ($(findstring 1, $(KBUILD_EXTRA_WARN)),)
> > +CFLAGS_mtk_ppe.o += -Wno-array-bounds
> > +endif  
> 
> This worked for me:
> 
> diff --git a/drivers/net/can/usb/kvaser_usb/Makefile b/drivers/net/can/usb/kvaser_usb/Makefile
> index cf260044f0b9..43eb921f9102 100644
> --- a/drivers/net/can/usb/kvaser_usb/Makefile
> +++ b/drivers/net/can/usb/kvaser_usb/Makefile
> @@ -1,3 +1,7 @@
>  # SPDX-License-Identifier: GPL-2.0-only
>  obj-$(CONFIG_CAN_KVASER_USB) += kvaser_usb.o
>  kvaser_usb-y = kvaser_usb_core.o kvaser_usb_leaf.o kvaser_usb_hydra.o
> +
> +ifeq ($(KBUILD_EXTRA_WARN),)
> +CFLAGS_kvaser_usb_hydra.o += -Wno-array-bounds
> +endif

Ah, thanks, I must have tried -Wno-array-bounds before I figured out
the condition and reverted back to full $(call cc-disable-warning, ..)
Let me redo the patches.

Thanks!

RE: GCC 12 warnings

[David Laight]

From: Jakub Kicinski
> Sent: 20 May 2022 18:24
...
> > +ifeq ($(KBUILD_EXTRA_WARN),)
> > +CFLAGS_kvaser_usb_hydra.o += -Wno-array-bounds
> > +endif
> 
> Ah, thanks, I must have tried -Wno-array-bounds before I figured out
> the condition and reverted back to full $(call cc-disable-warning, ..)
> Let me redo the patches.

I think you need a check that the compiler actually supports
-Wno-array-bounds.
But that only needs be done once, not for every file
that needs the option.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

Re: GCC 12 warnings

[Kees Cook]

On Fri, May 20, 2022 at 09:43:03PM +0000, David Laight wrote:
> From: Jakub Kicinski
> > Sent: 20 May 2022 18:24
> ...
> > > +ifeq ($(KBUILD_EXTRA_WARN),)
> > > +CFLAGS_kvaser_usb_hydra.o += -Wno-array-bounds
> > > +endif
> > 
> > Ah, thanks, I must have tried -Wno-array-bounds before I figured out
> > the condition and reverted back to full $(call cc-disable-warning, ..)
> > Let me redo the patches.
> 
> I think you need a check that the compiler actually supports
> -Wno-array-bounds.
> But that only needs be done once, not for every file
> that needs the option.

godbolt tells me that all kernel supported versions of GCC and Clang
handle -Wno-array-bounds. Let me know if there is a combination that
doesn't?

-- 
Kees Cook