Re: Tracing NetBSD sun4v issues

[Sad Clouds <cryintothebluesky%gmail.com@localhost>]

On Thu, 15 Oct 2026 10:57:16 +0100
Sad Clouds <cryintothebluesky%gmail.com@localhost> wrote:

> On Thu, 15 Oct 2026 08:47:29 +0100
> Sad Clouds <cryintothebluesky%gmail.com@localhost> wrote:
> 
> > On Wed, 14 Oct 2026 19:42:39 +0000 (UTC)
> > Palle Lyckegaard <palle%netbsd.org@localhost> wrote:
> > 
> > > I fixed an issue with spilling register window contents about a year ago 
> > > (locore.s rev. 1.438).
> > 
> > I applied this revision to NetBSD-11 kernel and the test program still
> > dies with SEGV. I think the original issue is still there.
> 
> Actually I take it back. It looks like the new locore.o was not
> combined properly into the kernel image. I removed all object files and
> rebuilt from scratch and no longer can reproduce the SEGV crash.
> 
> I do see other crashes when booting into multi-user mode, so looks like
> some issues still remain. Can you remember any other relevant fixes you
> made apart from locore.s? I can try and backport those to 11 until the
> interrupt issue is resolved in -current.

I think the same issue is still there, with the patched locore.s
(rev 1.439) it has shifted to a different location.

This command runs fine:
# ps -j

This command ("j" instead of "-j") crashes with SEGV:
# ps j

And I get the same backtrace in jemalloc

Program terminated with signal SIGSEGV, Segmentation fault.
#0  0x0000000040b89790 in ecache_alloc_grow (tsdn=0xffffffffffffc1f5, pac=0xffffffffffffc1f5, 
    ehooks=0x40e05c00 <ehooks_default_extent_hooks>, ecache=<optimized out>, expand_edata=<optimized out>, size=1, 
    alignment=<optimized out>, zero=<optimized out>, guarded=<optimized out>)
    at /usr/src/external/bsd/jemalloc/lib/../dist/src/extent.c:126

(gdb) disas
Dump of assembler code for function ecache_alloc_grow:
...
   0x0000000040b89784 <+804>:   call  0x40b87940 <je_extent_alloc_wrapper>
   0x0000000040b89788 <+808>:   stx  %l4, [ %sp + 0x8af ]
   0x0000000040b8978c <+812>:   ret 
=> 0x0000000040b89790 <+816>:   restore  %g0, %o0, %o0
   0x0000000040b89794 <+820>:   call  0x40b7e4c0 <je_malloc_mutex_lock_slow>