Re: Collection of small Alpha issues

[Jason Thorpe <thorpej%me.com@localhost>]

> On Jun 23, 2022, at 8:30 AM, John Klos <john%ziaspace.com@localhost> wrote:
> 
> Hi,
> 
> NetBSD/alpha has been running wonderfully on my AlphaServer DS25 for a while now since thorpej fixed lots of things. However, there are still a few things that are a bit buggy.
> 
> First, I'm seeing an issue where multiple runnable processes stay on just one processor, like so:
> 
> load averages:  2.04,  2.09,  1.93;               up 34+10:47:53   14:10:
> 84 processes: 2 runnable, 79 sleeping, 1 stopped, 1 zombie, 1 on CPU
> CPU0: 59.5% user, 33.7% nice,  6.4% system,  0.4% interrupt,  0.0% idle
> CPU1:  0.4% user,  0.0% nice,  0.7% system,  0.5% interrupt, 98.4% idle
> Memory: 6073M Act, 2966M Inact, 12M Wired, 123M Exec, 7814M File, 3696K F
> Swap: 24G Total, 24G Free

Hm, interesting.  The MD code isn’t really responsible for choosing which processor on which to run a given LWP, but perhaps there’s some tunable that’s not quite right on Alpha that is causing the MI scheduling code to decide migrating to a different CPU is not worth the cost.  (The Alpha has a very large value of HZ compared to most ports, and that number is a factor in making scheduling decisions…)

>  PID USERNAME PRI NICE   SIZE   RES STATE       TIME   WCPU    CPU COMMA
> 14054 john      26    0    15M 9712K RUN/0      11:05 58.98% 58.98% mini-
> 9576 pbulk     24    4   258M  240M RUN/0       0:12 33.91% 28.03% cc1pl
> 
> cpuctl shows nothing out of order:
> 
> Num  HwId Unbound LWPs Interrupts Last change              #Intr
> ---- ---- ------------ ---------- ------------------------ -----
> 0    0    online       intr       Fri May 20 03:23:51 2022 6
> 1    1    online       intr       Fri May 20 03:23:51 2022 5
> 
> 
> Second, several pkgsrc builds fail similarly:
> 
> In file included from vm.c:360:0:
> vm_eval.c: In function 'rb_eval_cmd_kw':
> vm_eval.c:1929:1: internal compiler error: in move_insn, at haifa-sched.c:5444
> }
> ^
> no stack trace because unwind library not available
> Please submit a full bug report,
> with preprocessed source if appropriate.
> See <http://www.NetBSD.org/support/send-pr.html> for instructions.
> 
> from http://ftp.ziaspace.com/pub/pkgsrc/reports/9.0_2022Q1/20220522.0008/ruby30-base-3.0.4nb1/build.log
> 
> Any idea how to fix these?

First step here is going to be to try and isolate a reduced test case.  Do you think you could try to do that?

> Third, is there any reason to care about unaligned accesses from sendmail?
> 
> Starting sendmail.
> [ 2975706.6549787] pid 24869 (sendmail): unaligned access: va=0x3fffd6261b2 pc=0x12001d114 ra=0x12001ce44 sp=0x1ffffd1d0 op=ldl
> [ 2975706.6559539] pid 24869 (sendmail): unaligned access: va=0x3fffd626242 pc=0x12001d114 ra=0x12001ce44 sp=0x1ffffd1d0 op=ldl
> ...

It’s annoying, of course, but the kernel fixes them up.  Might be instructive to know the code that’s generating these.

> Fourth, firefox52 compiles, but it doesn't run:
> 
> (No debugging symbols found in firefox52)
> (gdb) run
> Starting program: /usr/local/bin/firefox52
> MESA-LOADER: failed to open swrast (search paths /usr/X11R7/lib/modules/dri)
> libGL error: failed to load driver: swrast
> [New LWP 20128 of process 17827]
> [New process 17827]
> [New LWP 25245 of process 17827]
> [New LWP 21530 of process 17827]
> [New LWP 24961 of process 17827]
> [New LWP 25914 of process 17827]
> [New LWP 29448 of process 17827]
> [New LWP 5257 of process 17827]
> [New LWP 6672 of process 17827]
> [New LWP 13889 of process 17827]
> [ 2975802.9964870] pid 17827 (firefox52): unaligned access: va=0x3fffa591b91 pc=0x3fff9c264b0 ra=0x3fff7e5bc5c sp=0x1ffffe560 op=ldq
> 
> Thread 1 "" received signal SIGSEGV, Segmentation fault.
> 0x000003fff9c264b4 in ?? () from /usr/local/lib/firefox52/libxul.so
> 
> I can rebuild with symbols, or I can give a shell to anyone who wants to have a gander.

I’m gonna pass on this one.  :-)

-- thorpej