Re: aarch64: lang/gcc1* build regression between Mar-28 and Apr-8 [zfs corruptions without block_cloning involved!]

From: Dima Panov <fluffy_at_FreeBSD.org> wrote on
Date: Fri, 14 Apr 2023 06:55:29 UTC :

> On 14.04.2023 01:13, Dimitry Andric wrote:
> > On 13 Apr 2023, at 23:42, Dima Panov <fluffy@freebsd.org> wrote:
> >> Somethings changed in main branch between Mar, 28 and Apr, 8 which causes permanent build error for lang/gcc1[012] on aarch64 (exactly VMWare container on M1Pro mac)
> >>
> >> during RTL pass: sched1
> >> gimple-match.cc: In function 'bool gimple_nop_atomic_bit_test_and_p(tree, tree_node**, tree_node* (*)(tree))':
> >> gimple-match.cc:39215:1: internal compiler error: Segmentation fault
> >> 39215 | }
> >> | ^
> >> mmap: Invalid argument
> >> Please submit a full bug report, with preprocessed source (by using -freport-bug).
> >> See <https://gcc.gnu.org/bugs/> for instructions.
> >> gmake[4]: *** [Makefile:1145: gimple-match.o] Error 1
> >> gmake[4]: *** Waiting for unfinished jobs....
> >> rm gcc.pod gfortran.pod
> > 
> > Are you running this build on zfs? There are apparently a bunch of
> > filesystem corruption problems which have surfaced due to recent openzfs
> > imports.
> > 
> 
> 
> Yup, pure zfs install.
> Discussion refers to block_cloning but it disabled on my pool

It is not true that it is limited to block_cloning. zfs after the
import corrupts data even without block_cloning having ever been
in use, even as the code currently is. More problems are to be
found and fixed yet, despite the several fixes now in place.

See for example (much of the reporting is on both freebsd-current
and dev-commits-src-main but some is only in one place or the
other):

https://lists.freebsd.org/archives/dev-commits-src-main/2023-April/014397.html

https://lists.freebsd.org/archives/freebsd-current/2023-April/003446.html

https://lists.freebsd.org/archives/freebsd-current/2023-April/003449.html
https://lists.freebsd.org/archives/freebsd-current/2023-April/003459.html

https://lists.freebsd.org/archives/freebsd-current/2023-April/003460.html

> DATA feature@block_cloning disabled local

You will still get corruptions but they will not be as
fundamental to problems restoring the pool to a good state.
[There is work in progress to try to have block_cloning
being enabled recoverable but effectively disabled other
than cleaning itself up. But this will not fix the other
cause(s) of corruption, which are still being worked on
as well.]

===
Mark Millard
marklmi at yahoo.com