Re: git: c0d9a07101a1 - main - MFV: zstd 1.5.7.
- Reply: Li-Wen Hsu : "Re: git: c0d9a07101a1 - main - MFV: zstd 1.5.7."
- In reply to: Michael Tuexen : "Re: git: c0d9a07101a1 - main - MFV: zstd 1.5.7."
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Sat, 11 Apr 2026 10:32:09 UTC
On Sat, Apr 11, 2026 at 3:25 PM Michael Tuexen <tuexen@freebsd.org> wrote:
> > On 11. Apr 2026, at 06:56, Xin Li <delphij@freebsd.org> wrote:
> >
> > On 2026-04-10 21:00, Adrian Chadd wrote:
> >> hi!
> >> Same; this is breaking on my Snapdragon X1E FDT BSP work.
> >> It fails super early in kernel hand-off. It's going to be a pain to
> >> figure out locally as I have a framebuffer and I'd have to go do all
> >> the early mapping to do super early (in initarm()) framebuffer work.
> >> Is this reproducible by anyone else? delphij@, do you have any arm64
> >> FDT devices? Or maybe QEMU?
> >
> > bz@ and lwhsu@ did some diagnostic work, and this should help (adding
> -DZSTD_NO_TRACE to ZSTD_C)
> >
> > diff --git a/sys/conf/kern.pre.mk b/sys/conf/kern.pre.mk
> > index 871cd1c0705a..043d3fd0927c 100644
> > --- a/sys/conf/kern.pre.mk
> > +++ b/sys/conf/kern.pre.mk
> > @@ -158,7 +158,7 @@ NORMAL_FWO= ${CC:N${CCACHE_BIN}} -c ${ASM_CFLAGS}
> ${WERROR} -o ${.TARGET} \
> > NOSAN_C= ${NORMAL_C:N-fsanitize*:N-fno-sanitize*:N-fasan-shadow-offset*}
> >
> > # for ZSTD in the kernel (include zstd/lib/freebsd before other CFLAGS)
> > -ZSTD_C= ${CC} -c -DZSTD_HEAPMODE=1 -I$S/contrib/zstd/lib/freebsd
> ${CFLAGS} \
> > +ZSTD_C= ${CC} -c -DZSTD_NO_TRACE -DZSTD_HEAPMODE=1
> -I$S/contrib/zstd/lib/freebsd ${CFLAGS} \
> > -I$S/contrib/zstd/lib -I$S/contrib/zstd/lib/common ${WERROR} \
> > -Wno-missing-prototypes -U__BMI__ -DZSTD_NO_INTRINSICS ${.IMPSRC}
> > # https://github.com/facebook/zstd/commit/812e8f2a [zstd 1.4.1]
> Should building a new kernel with this patch resolve the issue or
> is a buildworld/installworld required?
>
> I tested building a new kernel with this patch and it does not resolve the
> issue for me (VirtualBox VM on macOS and QEMU on Rasbian).
I haven't tested this one, the original patch was adding:
#define ZSTD_TRACE 0
to
sys/contrib/zstd/lib/freebsd/zstd_kfreebsd.h
This fixes my case running in QEMU.
> Best regards
> Michael
> >
> >
> >
> >> -adrian
> >> On Thu, 9 Apr 2026 at 19:31, Bjoern A. Zeeb <bz@freebsd.org> wrote:
> >>>
> >>> On Wed, 8 Apr 2026, Xin LI wrote:
> >>>
> >>>> The branch main has been updated by delphij:
> >>>>
> >>>> URL:
> https://cgit.FreeBSD.org/src/commit/?id=c0d9a07101a1e72769ee0619a583f63a078fb391
> >>>>
> >>>> commit c0d9a07101a1e72769ee0619a583f63a078fb391
> >>>> Merge: dd201e407e48 7e509d50854e
> >>>> Author: Xin LI <delphij@FreeBSD.org>
> >>>> AuthorDate: 2026-04-08 21:22:45 +0000
> >>>> Commit: Xin LI <delphij@FreeBSD.org>
> >>>> CommitDate: 2026-04-08 21:22:45 +0000
> >>>>
> >>>> MFV: zstd 1.5.7.
> >>>>
> >>>> MFC after: 2 weeks
> >>>> Relnotes: yes
> >>>
> >>> I do not know how but this makes an arm64 machine,
> >>> doing a UEFI netboot (in case it matters) fetching the
> /boot/loader.efi,
> >>> to fail.
> >>> Kernel is the GENERIC-MMCCAM and I do not do ZFS, just to be sure.
> >>>
> >>> See this thread for a "bisect" and symptoms:
> >>> https://lists.freebsd.org/archives/freebsd-arm/2026-April/005757.html
> >>>
> >>> /bz
> >>>
> >>> --
> >>> Bjoern A. Zeeb
> r15:7
> >>>
> >
> >
>
>
>