Re: ipv4/tcp.c:4234:1: error: the frame size of 1152 bytes is larger than 1024 bytes [-Werror=frame-larger-than=]

[Brendan Higgins]

On Tue, Sep 14, 2021 at 3:04 PM Arnd Bergmann <arnd@xxxxxxxx> wrote:
>
> On Tue, Sep 14, 2021 at 10:48 PM Brendan Higgins
> <brendanhiggins@xxxxxxxxxx> wrote:
> >
> > On Mon, Sep 13, 2021 at 1:55 PM Shuah Khan <skhan@xxxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > On 9/8/21 3:24 PM, Brendan Higgins wrote:
> > > Brendan,
> > >
> > > Would you like to send me the fix with Suggested-by for Arnd or Kees?
> >
> > So it looks like Arnd's fix was accepted (whether by him or someone
> > else) for property-entry-test and Linus already fixed thunderbolt, so
> > the only remaining of Arnd's patches is for the bitfield test, so I'll
> > resend that one in a bit.
> >
> > Also, I haven't actually tried Linus' suggestion yet, but the logic is
> > sound and the change *should* be fairly unintrusive - I am going to
> > give that a try and report back (but I will get the bitfield
> > structleak disable patch out first since I already got that applying).
>
> Looking at my randconfig tree, I find these six instances:
>
> $ git grep -w DISABLE_STRUCTLEAK_PLUGIN
> drivers/base/test/Makefile:CFLAGS_property-entry-test.o +=
> $(DISABLE_STRUCTLEAK_PLUGIN)
> drivers/iio/test/Makefile:CFLAGS_iio-test-format.o +=
> $(DISABLE_STRUCTLEAK_PLUGIN)
> drivers/mmc/host/Makefile:CFLAGS_sdhci-of-aspeed.o              +=
> $(DISABLE_STRUCTLEAK_PLUGIN)
> drivers/thunderbolt/Makefile:CFLAGS_test.o += $(DISABLE_STRUCTLEAK_PLUGIN)
> lib/Makefile:CFLAGS_test_scanf.o += $(DISABLE_STRUCTLEAK_PLUGIN)
> lib/Makefile:CFLAGS_bitfield_kunit.o += $(DISABLE_STRUCTLEAK_PLUGIN)
> scripts/Makefile.gcc-plugins:    DISABLE_STRUCTLEAK_PLUGIN +=
> -fplugin-arg-structleak_plugin-disable
> scripts/Makefile.gcc-plugins:export DISABLE_STRUCTLEAK_PLUGIN

Alright, I incorporated all the above into a patchset that I think is
ready to send out, but I had a couple of issues with the above
suggestions:

- I could not find a config which causes a stacksize warning for
sdhci-of-aspeed.
- test_scanf is not a KUnit test.
- Linus already fixed the thunderbolt test by breaking up the test cases.

I am going to send out patches for the thunderbolt test and for the
sdhci-of-aspeed test for the sake of completeness, but I am not sure
if we should merge those two. I'll let y'all decide on the patch
review.

I only based the thunderbolt and bitfield test fixes on actual patches
from Arnd, but I think Arnd pretty much did all the work here so I am
crediting him with a Co-developed-by on all the other patches, so
Arnd: please follow up on the other patches with a signed-off-by,
unless you would rather me credit you in some other way.

> Sorry for failing to submit these as a proper patch. If you send a new version,
> I think you need to make sure you cover all of the above, using whichever
> change you like best.

I am still going to try to get Linus' suggestion working since it
actually solves the problem, but I would rather get the above
suggested fix out there since it is quick and I know it works.

Cheers