Re: Looks like the arm 20220805 snapshots are still odd, so probably kern.geom.part.mbr.enforce_chs=0 was still in use

Glen Barber <[email protected]> Sat, 13 Aug 2022 00:30:34 +0000
Newsgroups	gmane.os.freebsd.devel.hackers,gmane.os.freebsd.devel.arm
Message-ID	<[email protected]>
On Fri, Aug 12, 2022 at 05:16:12PM -0700, Mark Millard wrote:
> On 2022-Aug-9, at 11:55, Mark Millard <[email protected]> wrote:
> 
> > On 2022-Aug-9, at 11:15, Glen Barber <[email protected]> wrote:
> > 
> >> On Tue, Aug 09, 2022 at 02:06:14PM -0400, Ed Maste wrote:
> >>> On Sun, 7 Aug 2022 at 18:43, Glen Barber <[email protected]> wrote:
> >>>> 
> >>>> Will do.  I’ll commit the suggested change to main tomorrow.
> >>>> 
> >>>> Thank you for your vigilant investigation.
> >>> 
> >>> Shall I commit the enforce_chs check now?
> >>> ---
> >>> commit 6ee7d69e6b526f35789b23ba570025f1c3b39c1a
> >>> Author: Ed Maste <[email protected]>
> >>> Date:   Tue Jul 19 16:47:49 2022 -0400
> >>> 
> >>>   release: ensure enforce_chs sysctl is 0
> >>> 
> >>>   We do not want CHS-based alignment for VM or SD card release images.
> >>> 
> >>>   Sponsored by:   The FreeBSD Foundation
> >>> 
> >>> diff --git a/release/tools/arm.subr b/release/tools/arm.subr
> >>> index 6e4ae731a0b9..01c5303cd4e1 100644
> >>> --- a/release/tools/arm.subr
> >>> +++ b/release/tools/arm.subr
> >>> @@ -62,6 +62,10 @@ umount_loop() {
> >>> }
> >>> 
> >>> arm_create_disk() {
> >>> +       if [ $(sysctl -n kern.geom.part.mbr.enforce_chs) != 0 ]; then
> >>> +               return 1
> >>> +       fi
> >>> +
> >>>       # Create the target raw file and temporary work directory.
> >>>       chroot ${CHROOTDIR} gpart create -s ${PART_SCHEME} ${mddev}
> >>>       if [ "${PART_SCHEME}" = "GPT" ]; then
> >>> 
> >> 
> >> Good question.  Do we still want to ensure it is set to '0'?  I'm a bit
> >> confused from the back-and-forth on the original thread.
> >> 
> >> If we do want to ensure it is set to '0', yes, please go ahead.
> >> 
> > 
> > Hopefully this week's experiment with explicitly avoiding BSD
> > and freebsd-ufs having the same offset inside BSD (avoiding
> > both offsets being zero) will allow the UFS labeling to work
> > right: freebsd-ufs being tied to a unique offset inside BSD.
> > 
> > I really doubt that using kern.geom.part.mbr.enforce_chs=1 to
> > cause the offsets to be different is reasonable, despite that
> > it happens to make them distinct: the freebsd-ufs offset is
> > better controlled explicitly elsewhere.
> > 
> 
> The experiment with this week's snapshot is working just fine.
> It is based on the update:
> 
> QUOTE
> The branch main has been updated by gjb:
> 
> URL: https://cgit.FreeBSD.org/src/commit/?id=45add40717c24ef0b5418664fae1718b15a0422b
> 
> commit 45add40717c24ef0b5418664fae1718b15a0422b
> Author:     Glen Barber <[email protected]>
> AuthorDate: 2022-08-08 14:59:29 +0000
> Commit:     Glen Barber <[email protected]>
> CommitDate: 2022-08-08 14:59:29 +0000
> 
>     release: fix alignment for arm SoCs
>     
>     MFC after:      3 days
>     Submitted by:   Mark Millard
>     Sponsored by:   Rubicon Communications, LLC ("Netgate")
> ---
>  release/tools/arm.subr | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/release/tools/arm.subr b/release/tools/arm.subr
> index 3dea17339958..25d4640cc26b 100644
> --- a/release/tools/arm.subr
> +++ b/release/tools/arm.subr
> @@ -77,7 +77,7 @@ arm_create_disk() {
>  		chroot ${CHROOTDIR} newfs_msdos -L msdosboot -F ${FAT_TYPE} /dev/${mddev}s1
>  		chroot ${CHROOTDIR} gpart add -t freebsd ${mddev}
>  		chroot ${CHROOTDIR} gpart create -s bsd ${mddev}s2
> -		chroot ${CHROOTDIR} gpart add -t freebsd-ufs -a 64k ${mddev}s2
> +		chroot ${CHROOTDIR} gpart add -t freebsd-ufs -a 64k -b 64k ${mddev}s2
>  		chroot ${CHROOTDIR} newfs -U -L rootfs /dev/${mddev}s2a
>  	fi
> END QUOTE
> 
> This is unlike when the "-b 64k" was not present. This
> includes rebooting after the initial boot, unlike before.
> This is for dd'ing to USB3 NVMe SSD media and testing,
> for both of:
> 
> FreeBSD-14.0-CURRENT-arm64-aarch64-RPI-20220812-6a70a0c8bfa-257314.img
> FreeBSD-13.1-STABLE-arm64-aarch64-RPI-20220812-eb2a9b78586-252107.img
> 
> (The -b use implicitly also changed the size of the freebsd-ufs
> slice, making it smaller.) For the media growfs is involved in
> the initial boot.
> 
> glabel list now shows ufs/rootfs bound to da0s2a and does not
> show a ufs/rootfsa at all. There is no binding to da0s2 (BSD)
> shown.
> 
> # glabel list
> Geom name: da0s1
> Providers:
> 1. Name: msdosfs/MSDOSBOOT
>    Mediasize: 52428800 (50M)
>    Sectorsize: 512
>    Stripesize: 0
>    Stripeoffset: 1048576
>    Mode: r1w1e1
>    secoffset: 0
>    offset: 0
>    seclength: 102400
>    length: 52428800
>    index: 0
> Consumers:
> 1. Name: da0s1
>    Mediasize: 52428800 (50M)
>    Sectorsize: 512
>    Stripesize: 0
>    Stripeoffset: 1048576
>    Mode: r1w1e2
> 
> Geom name: da0s2a
> Providers:
> 1. Name: ufs/rootfs
>    Mediasize: 240003866624 (224G)
>    Sectorsize: 512
>    Stripesize: 0
>    Stripeoffset: 53542912
>    Mode: r1w1e1
>    secoffset: 0
>    offset: 0
>    seclength: 468757552
>    length: 240003866624
>    index: 0
> Consumers:
> 1. Name: da0s2a
>    Mediasize: 240003866624 (224G)
>    Sectorsize: 512
>    Stripesize: 0
>    Stripeoffset: 53542912
>    Mode: r1w1e2
> 
> # gpart show -p
> =>       63  468862065    da0  MBR  (224G)
>          63       1985         - free -  (993K)
>        2048     102400  da0s1  fat32lba  [active]  (50M)
>      104448  468757680  da0s2  freebsd  (224G)
> 
> =>        0  468757680   da0s2  BSD  (224G)
>           0        128          - free -  (64K)
>         128  468757552  da0s2a  freebsd-ufs  (224G)
> 
> A difference in behavior is that "gpart show" does not
> report the ufs/rootfs labeling. For all I know, this
> could be expected. glabel does show ufs/rootfs .
> 
> I do not know what would happen if only the size had
> been made smaller but the starting offset had been left
> at 0.
> 
> But the evidence from the "without -b" and "with -b"
> testing is that having starting offset 0 in BSD
> and the same size as BSD can be a problem for the
> freebsd-ufs slice as processed by the initial boot,
> at least when ufs labeling is in use (here ufs/rootfs).
> 
> So far, this week's snapshots look good to me for the
> issue having been avoided but ending up better aligned
> overall than when kern.geom.part.mbr.enforce_chs=1 was
> in use.
> 
> If a larger alignment is needed at some point for
> freebsd-ufs, adjusting the pair of gpart add arguments
> "-a 64k -b 64k" should deal with it.
> 
> 
> Notes:
> 
> The USB3 NNVMe SSD based testing was with a 8GiByte
> RPi4B and so does have the addition of:
> 
> #
> # Local addition that avoids USB3 SSD boot failures that look like: 
> #   uhub_reattach_port: port ? reset failed, error=USB_ERR_TIMEOUT 
> #   uhub_reattach_port: device problem (USB_ERR_TIMEOUT), disabling port ? 
> initial_turbo=60 
> 
> in the config.txt file that the RPi* firmware uses.
> (It is for a separate issue --and FreeBSD does not
> have such in place by default at this time.)
> 

Mark,

Thank you very much for the detailed information, and of course for
reporting the problem, and testing variations of the correct behavior.

Glen
signature.asc (application/pgp-signature, 833 B)