ports -r484565 : qemu-arm-static fails with: (start < end), function page_set_flags, file . . . accel/tcg/translate-all.c, line 2077

[Mark Millard]

On 2018-Nov-10, at 20:00, Kyle Evans <kevans at freebsd.org> wrote:

> On Sat, Nov 10, 2018 at 4:30 PM Mark Millard <marklmi at yahoo.com> wrote:
>> 
>> On 2018-Nov-10, at 12:28, Kyle Evans <kevans at freebsd.org> wrote:
>> 
>>> On Sat, Nov 10, 2018 at 11:38 AM Mark Millard via freebsd-ports
>>> <freebsd-ports at freebsd.org> wrote:
>>>> 
>>>> Having actually installed the reverted code fist ( -r438807 ),
>>>> cmake's package stage is now well past were it was failing.
>>>> 
>>>> So it is not the pkg vintage that matters: it is the qemu-sbruno
>>>> vintage that matters.
>>>> 
>>>> (gcc8 getting that far is hours away: full bootstrap, so mostly
>>>> emulated.)
>>>> 
>>> 
>>> I find the assertion you've reported fairly bizarre, since all of the
>>> page_set_flags invocations we would've touched are generally of the
>>> form `page_set_flags(start, start + len, ...)` -- I'm working on
>>> reproducing locally, though.
>> 
>> Looking at the overall sources for the two versions ( as seen via
>> -r483807 and -r484565 ) I find a possibly-significant changed file:
>> 
>> # diff -u /wrkdirs/usr/ports/emulators/qemu-user-static/*work/qemu-bsd-user-*/bsd-user/mmap.c | more
>> --- /wrkdirs/usr/ports/emulators/qemu-user-static/483807-work/qemu-bsd-user-495fb3a/bsd-user/mmap.c     2018-05-25 07:28:13.000000000 -0700
>> +++ /wrkdirs/usr/ports/emulators/qemu-user-static/484565-work/qemu-bsd-user-2cb0cdd/bsd-user/mmap.c     2018-11-09 09:27:18.000000000 -0800
>> 
> 
> Yes, I'm familiar with this particular changeset- I wrote it. =)

I noticed that later when researching.

> Unfortunately, I can't reproduce this locally- neither with
> devel/cmake nor any of the other ports that I build.

Out of 200+ I only saw it for the two. Multiple bulk tests
for cmake. (I did not wait for lang/gcc8's full bootstrap
to finish.) Some uncommon (limiting?) condition, apparently.

> I think we'll
> have to wait until either we get more reports of this or portmgr@
> trips over it in a way that I can reproduce and dig in a bit.

If there is a change that would force a core dump or
backtrace or something to give context, I could try such.
(My normal builds are non-debug but with symbols enabled,
even for ports, a combination I had to add local support
for.) I did not find a core file when I looked in the
tar archive of the failure. (I've not yet checked
qemu-arm-static does something to prevent generating
host core files.)

Or I could substitute the old version of the one source in
with the rest being new, rebuild qemu-user-static, and try
again.

Such would be tomorrow, my time (US Pacific).

(The reason for tomorrow: I've got a bulk going for updating
to ports -r484652 --other than the reverted qemu-user-static.
It is done but for the full-bootstrap lang/gcc8 that will
take hours more.)



===
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)