Linux-Next discussions
 help / color / mirror / Atom feed
From: Nathan Chancellor <nathan@kernel.org>
To: Thorsten Leemhuis <regressions@leemhuis.info>
Cc: kernelci-results@groups.io, regressions@lists.linux.dev,
	kernelci@lists.linux.dev, gus@collabora.com,
	linux-next@vger.kernel.org, Randy Dunlap <rdunlap@infradead.org>,
	amd-gfx@lists.freedesktop.org,
	Gaghik Khachatrian <gaghik.khachatrian@amd.com>,
	James Lin <pinglei.lin@amd.com>,
	Alex Deucher <alexander.deucher@amd.com>,
	llvm@lists.linux.dev
Subject: Re: [REGRESSION] next/master: (build) stack frame size (2088) exceeds limit (2048) in 'dml31_ModeSupport...
Date: Mon, 8 Jun 2026 17:32:38 -0700	[thread overview]
Message-ID: <20260609003238.GA3576576@ax162> (raw)
In-Reply-To: <5f2d4707-ae11-44f3-ad09-1c0580abde91@leemhuis.info>

On Mon, Jun 08, 2026 at 01:24:32PM +0200, Thorsten Leemhuis wrote:
> On 6/5/26 06:23, Randy Dunlap wrote:
> > [adding amd-gfx]
> 
> [+Nathan -- I'm also wondering if we should CC the amdgpu maintainers,
> but lets wait with that for a moment]

Thanks for the CC. Bringing in more folks based on my bisect below.

> Nathan afaics saw this problem a few weeks ago already and filed and
> issue for tracking: https://github.com/ClangBuiltLinux/linux/issues/2161
> 
> To quote from there: ""After LLVM commit 8ac9461e48e0 ("[SLP] Reject
> 2-element vectorization when vector inst count exceeds scalar",
> 2026-04-11), I am seeing two new -Wframe-larger-than instances from
> drivers/gpu/drm/amd/display […]""
> 
> Side note: I wonder if that LLVM commit somehow made it to the KernelCI
> env, or if something in the amdgpu code in -next changed and made the
> problem worse.

So I think these are two separate but perhaps related issues (i.e., the
latter of the "side note" comment), as that LLVM commit is only in LLVM
main (currently 23.0.0) but this instance that KernelCI flags is visible
with clang-21.

With clang-22 and a distribution configuration, I see:

  drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn30/display_mode_vba_30.c:3381:6: warning: stack frame size (2312) exceeds limit (2048) in 'dml30_ModeSupportAndSystemConfigurationFull' [-Wframe-larger-than]
   3381 | void dml30_ModeSupportAndSystemConfigurationFull(struct display_mode_lib *mode_lib)
        |      ^
  1 warning generated.
  drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn31/display_mode_vba_31.c:3800:6: warning: stack frame size (2128) exceeds limit (2048) in 'dml31_ModeSupportAndSystemConfigurationFull' [-Wframe-larger-than]
   3800 | void dml31_ModeSupportAndSystemConfigurationFull(struct display_mode_lib *mode_lib)
        |      ^
  1 warning generated.
  drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn314/display_mode_vba_314.c:3892:6: warning: stack frame size (2136) exceeds limit (2048) in 'dml314_ModeSupportAndSystemConfigurationFull' [-Wframe-larger-than]
   3892 | void dml314_ModeSupportAndSystemConfigurationFull(struct display_mode_lib *mode_lib)
        |      ^
  1 warning generated.

My bisect lands on commit faaeeecef94a ("drm/amd/display: Fix type
mismatches in DML and normalize loop bounds") in -next, which makes
sense. I have not looked into this super closely but I suspect that
change makes it easier for LLVM to inline or unroll loops, which can
result in more stack usage (especially as this code has some functions
with many tens of parameters). It is rather hard to look at and modify
this drm/amd/display code so I have not prioritized it...

I wonder if

  https://github.com/llvm/llvm-project/issues/115862
  https://github.com/llvm/llvm-project/issues/143908

are related, even if KASAN is not enabled here.

> > On 6/4/26 10:59 AM, KernelCI bot wrote:
> >> Hello,
> >>
> >> New build issue found on next/master:
> >>
> >> ---
> >>  stack frame size (2088) exceeds limit (2048) in 'dml31_ModeSupportAndSystemConfigurationFull' [-Werror,-Wframe-larger-than] in drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn31/display_mode_vba_31.o (drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn31/display_mode_vba_31.c) [logspec:kbuild,kbuild.compiler.error]
> >> ---
> >>
> >> - dashboard: https://d.kernelci.org/i/maestro:96417cd30041ab8f3153128d0120c6cd70782d99
> >> - giturl: https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git
> >> - commit HEAD:  b99ae45861eccff1e1d8c7b05a13650be805d437
> >> - tags: next-20260604
> >>
> >> Please include the KernelCI tag when submitting a fix:
> >>
> >> Reported-by: kernelci.org bot <bot@kernelci.org>
> >>
> >>
> >> Log excerpt:
> >> =====================================================
> >> /tmp/kci/linux/drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn31/display_mode_vba_31.c:3872:6: error: stack frame size (2088) exceeds limit (2048) in 'dml31_ModeSupportAndSystemConfigurationFull' [-Werror,-Wframe-larger-than]
> >>  3872 | void dml31_ModeSupportAndSystemConfigurationFull(struct display_mode_lib *mode_lib)
> >>       |      ^
> >> 1 error generated.
> >>
> >> =====================================================
> >>
> >>
> >> # Builds where the incident occurred:
> >>
> >> ## x86_64_defconfig+kselftest+x86-board on (x86_64):
> >> - compiler: clang-21
> >> - config: None
> >> - dashboard: https://d.kernelci.org/build/maestro:6a21abf22cc72b6e94bf3c60
> >>
> >>
> >> #kernelci issue maestro:96417cd30041ab8f3153128d0120c6cd70782d99
> >>
> >> --
> >> This is an experimental report format. Please send feedback in!
> >> Talk to us at kernelci@lists.linux.dev
> >>
> >> Made with love by the KernelCI team - https://kernelci.org
> >>
> > 
> 

-- 
Cheers,
Nathan

      reply	other threads:[~2026-06-09  0:32 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-06-04 17:59 [REGRESSION] next/master: (build) stack frame size (2088) exceeds limit (2048) in 'dml31_ModeSupport KernelCI bot
2026-06-05  4:23 ` Randy Dunlap
2026-06-08 11:24   ` Thorsten Leemhuis
2026-06-09  0:32     ` Nathan Chancellor [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260609003238.GA3576576@ax162 \
    --to=nathan@kernel.org \
    --cc=alexander.deucher@amd.com \
    --cc=amd-gfx@lists.freedesktop.org \
    --cc=gaghik.khachatrian@amd.com \
    --cc=gus@collabora.com \
    --cc=kernelci-results@groups.io \
    --cc=kernelci@lists.linux.dev \
    --cc=linux-next@vger.kernel.org \
    --cc=llvm@lists.linux.dev \
    --cc=pinglei.lin@amd.com \
    --cc=rdunlap@infradead.org \
    --cc=regressions@leemhuis.info \
    --cc=regressions@lists.linux.dev \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox