Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210906142615.GA1917503@roeck-us.net> <CAHk-=wgjTePY1v_D-jszz4NrpTso0CdvB9PcdroPS=TNU1oZMQ@mail.gmail.com>
 <YTbOs13waorzamZ6@Ryzen-9-3900X.localdomain> <CAK8P3a3_Tdc-XVPXrJ69j3S9048uzmVJGrNcvi0T6yr6OrHkPw@mail.gmail.com>
In-Reply-To: <CAK8P3a3_Tdc-XVPXrJ69j3S9048uzmVJGrNcvi0T6yr6OrHkPw@mail.gmail.com>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Tue, 7 Sep 2021 10:10:10 -0700
Message-ID: <CAHk-=wgZkQ+eZ02TaCpAWo_ffiLMwA2tYNHyL+B1dQ4YB0qfmA@mail.gmail.com>
Subject: Re: [PATCH] Enable '-Werror' by default for all kernel builds
To:     Arnd Bergmann <arnd@kernel.org>
Cc:     Nathan Chancellor <nathan@kernel.org>,
        Guenter Roeck <linux@roeck-us.net>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        llvm@lists.linux.dev, Nick Desaulniers <ndesaulniers@google.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Sep 7, 2021 at 2:11 AM Arnd Bergmann <arnd@kernel.org> wrote:
>
> > x86_64-alpine.log:drivers/gpu/drm/amd/amdgpu/../display/dc/calcs/dcn_calcs.c:452:13: error: stack frame size (1800) exceeds limit (1280) in function 'dcn_bw_calc_rq_dlg_ttu' [-Werror,-Wframe-larger-than]
> > x86_64-alpine.log:drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn21/display_rq_dlg_calc_21.c:1657:6: error: stack frame size (1336) exceeds limit (1280) in function 'dml21_rq_dlg_get_dlg_reg' [-Werror,-Wframe-larger-than]
> > x86_64-alpine.log:drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn30/display_rq_dlg_calc_30.c:1831:6: error: stack frame size (1352) exceeds limit (1280) in function 'dml30_rq_dlg_get_dlg_reg' [-Werror,-Wframe-larger-than]
> > x86_64-alpine.log:drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn31/display_rq_dlg_calc_31.c:1676:6: error: stack frame size (1336) exceeds limit (1280) in function 'dml31_rq_dlg_get_dlg_reg' [-Werror,-Wframe-larger-than]
> > x86_64-alpine.log:drivers/vhost/scsi.c:1831:12: error: stack frame size (1320) exceeds limit (1280) in function 'vhost_scsi_release' [-Werror,-Wframe-larger-than]
> >
> > Another instance where distros lower CONFIG_FRAME_WARN below the 2048
> > default. Again, none look particularly scary but should still probably
> > be dealt with.
>
> I would argue that they are still scary and should be addressed in the
> code, it's just that we don't see them on build bots that use the 2048 byte default.

No, they are scary for another reason entirely: clang is clearly doing
a *HORRIBLE* job with stack usage.

To take that dml30_rq_dlg_get_dlg_reg() function as an example: yes,
it has a few structures on the stack, but gcc allocates 512-720 bytes
of stack space depending on my config.

Not 1280 bytes.

So it's not even *close* to the 1024 byte limit with gcc, much less the 2kB one.

I don't know why clang basically decides to use almost double the
stack space. Maybe it's some other config option that does it, I tried
a fairly normal one and a "almost everythign enabled" one, and
couldn't get close to the reported stack frame size with gcc.

Just to try to make things as close as possible, I tried with the
exact same normal non-debug config (apart from obvious
compiler-dependent things), and picked that dml30_rq_dlg_get_dlg_reg()
function to look at (for no real reason other than that the stack
frame was biggest above.

Gcc did a 720-byte stack frame for that case. Not great, but whatever.

clang did a 1136-byte stack frame for the same thing.

Do I know why? No. I do note that that code is disgusting.

It's passing one of those structs around by value, for example. That's
a 72-byte structure that is copied on the stack due to stupid calling
conventions. Maybe clang generates a few extra temporaries for it as
part of the function call stack setup? Who knows..

                 Linus