Migrating from -current to stable/12 on RPI2B (ARMv7)

Tue Dec 29 21:25:00 UTC 2020

On 2020-Dec-29, at 02:15, Mark Millard <marklmi at yahoo.com> wrote:

> On 2020-Dec-29, at 00:16, Mark Millard <marklmi at yahoo.com> wrote:
> 
>> On 2020-Dec-28, at 17:02, bob prohaska <fbsd at www.zefox.net> wrote:
>> 
>>> On Mon, Dec 28, 2020 at 04:14:51PM -0800, Mark Millard wrote:
>>>> [I get the problem as well! I report a backtrace of the failure
>>>> and some more.]
>>>> 
>>> 
>>> Relieved to see it's reproducible.  
>> . . .
>> 
>> I tried reducing the size of things to be produced via building
>> the target based on the likes of using:
>> 
>> WITHOUT_LLVM_TARGET_AARCH64=
>> WITH_LLVM_TARGET_ARM=
>> WITHOUT_LLVM_TARGET_MIPS=
>> WITHOUT_LLVM_TARGET_POWERPC=
>> WITHOUT_LLVM_TARGET_RISCV=
>> WITHOUT_LLVM_TARGET_X86=
>> MALLOC_PRODUCTION=
>> WITH_MALLOC_PRODUCTION=
>> WITHOUT_ASSERT_DEBUG=
>> WITHOUT_LLVM_ASSERTIONS=
>> WITHOUT_DEBUG_FILES=
>> 
>> (But the host 13 was unchanged.)
>> 
>> It still failed, but at a different memory allocation, of a
>> different size:
>> 
>> r4             0x8000              32768
>> 
>> (gdb) bt
>> #0  thr_kill () at thr_kill.S:4
>> #1  0x4227d998 in __raise (s=6) at /usr/src/lib/libc/gen/raise.c:52
>> #2  0x42332284 in abort () at /usr/src/lib/libc/stdlib/abort.c:67
>> #3  0x00da5e4c in report_bad_alloc_error () at /usr/src/contrib/llvm-project/llvm/lib/Support/ErrorHandling.cpp:174
>> #4  0x00da61c8 in out_of_memory_new_handler() () at /usr/src/contrib/llvm-project/llvm/lib/Support/ErrorHandling.cpp:187
>> #5  0x420f5d24 in operator new (size=<optimized out>) at /usr/src/contrib/llvm-project/libcxx/src/new.cpp:73
>> #6  0x00e16878 in SetBufferSize () at /usr/src/contrib/llvm-project/llvm/include/llvm/Support/raw_ostream.h:131
>> #7  SetBuffered () at /usr/src/contrib/llvm-project/llvm/lib/Support/raw_ostream.cpp:97
>> #8  0x00e17368 in write () at /usr/src/contrib/llvm-project/llvm/lib/Support/raw_ostream.cpp:251
>> #9  0x00ddfe20 in operator<< () at /usr/src/contrib/llvm-project/llvm/include/llvm/Support/raw_ostream.h:192
>> #10 operator<< () at /usr/src/contrib/llvm-project/llvm/include/llvm/Support/raw_ostream.h:205
>> #11 printSymbolizedStackTrace () at /usr/src/contrib/llvm-project/llvm/lib/Support/Signals.cpp:154
>> #12 0x00de163c in PrintStackTrace () at /usr/src/contrib/llvm-project/llvm/lib/Support/Unix/Signals.inc:575
>> #13 0x00ddf604 in RunSignalHandlers () at /usr/src/contrib/llvm-project/llvm/lib/Support/Signals.cpp:67
>> #14 0x00de1f3c in SignalHandler () at /usr/src/contrib/llvm-project/llvm/lib/Support/Unix/Signals.inc:396
>> #15 0x42021db4 in handle_signal (actp=actp at entry=0x424df7b0, sig=<optimized out>, info=info at entry=0x424df7f0, ucp=<optimized out>) at /usr/src/lib/libthr/thread/thr_sig.c:303
>> #16 0x420213f8 in thr_sighandler (sig=0, info=0x424df7f0, _ucp=0x424df830) at /usr/src/lib/libthr/thread/thr_sig.c:246
>> #17 0xffffe190 in ?? ()
>> Backtrace stopped: previous frame identical to this frame (corrupt stack?)
> 
> 
> Updating the host 13 made no significant difference in
> behavior.
> 
> Using:
> 
> time -l /usr/bin/ld --eh-frame-hdr -Bstatic -o clang /usr/lib/crt1.o /usr/lib/crti.o /usr/lib/crtbeginT.o -L/usr/obj/rpi2_clang/arm.armv7/usr/fbsd/stable-12-src/arm.armv7/tmp/legacy/usr/lib -L/usr/obj/rpi2_clang/arm.armv7/usr/fbsd/stable-12-src/arm.armv7/tmp/obj-tools/lib/libz -L/usr/obj/rpi2_clang/arm.armv7/usr/fbsd/stable-12-src/arm.armv7/tmp/obj-tools/lib/libexecinfo -L/usr/obj/rpi2_clang/arm.armv7/usr/fbsd/stable-12-src/arm.armv7/tmp/obj-tools/lib/libelf -L/usr/obj/rpi2_clang/arm.armv7/usr/fbsd/stable-12-src/arm.armv7/tmp/obj-tools/lib/ncurses/ncursesw -L/usr/obj/rpi2_clang/arm.armv7/usr/fbsd/stable-12-src/arm.armv7/tmp/obj-tools/lib/libthr -L/usr/lib --gc-sections cc1_main.o cc1as_main.o cc1gen_reproducer_main.o driver.o /usr/obj/rpi2_clang/arm.armv7/usr/fbsd/stable-12-src/arm.armv7/tmp/obj-tools/lib/clang/libclang/libclang.a /usr/obj/rpi2_clang/arm.armv7/usr/fbsd/stable-12-src/arm.armv7/tmp/obj-tools/lib/clang/libllvm/libllvm.a -lz -lexecinfo -lelf -lncursesw -lpthread -legacy -lc++ -lm -lgcc -lgcc_eh -lc -lgcc -lgcc_eh /usr/lib/crtend.o /usr/lib/crtn.o
> 
> it reported after the failure output (an example):
> 
> time: command terminated abnormally
>       26.59 real         9.40 user        12.48 sys
>   2052032  maximum resident set size
>     21793  average shared memory size
>       218  average unshared data size
>       127  average unshared stack size
>    703238  page reclaims
>         0  page faults
>         0  swaps
>        55  block input operations
>      3104  block output operations
>         0  messages sent
>         0  messages received
>         1  signals received
>     41126  voluntary context switches
>       447  involuntary context switches
> 
> where looking up getrusage reports that:
> 
> ru_maxrss    the maximum resident set size utilized (in kilobytes).
> 
> which means: 44.0625 MiByte below 2 GiBytes for the resident set.
> 
> Adding --threads=1 made little difference:
> 
> time: command terminated abnormally
>       28.84 real         9.28 user        12.78 sys
>   2065016  maximum resident set size
>     21872  average shared memory size
>       219  average unshared data size
>       127  average unshared stack size
>    709661  page reclaims
>         0  page faults
>         0  swaps
>        48  block input operations
>      3050  block output operations
>         0  messages sent
>         0  messages received
>         1  signals received
>     36717  voluntary context switches
>       114  involuntary context switches
> Abort trap


I found another difference between how 13 is built and
stable/12 is built, so I'm trying another build.

The below shows what I've changed in stable/12 , but 13
eliminated even having the conditional logic, always
using -O2 -pipe :

# git diff | more

diff --git a/share/mk/sys.mk b/share/mk/sys.mk
index 9099b63a61a0..53d1e30b1d56 100644
--- a/share/mk/sys.mk
+++ b/share/mk/sys.mk
@@ -167,7 +167,7 @@ CFLAGS              ?=      -O
 .else
 CC             ?=      cc
 .if ${MACHINE_CPUARCH} == "arm" || ${MACHINE_CPUARCH} == "mips"
-CFLAGS         ?=      -O -pipe
+CFLAGS         ?=      -O2 -pipe
 .else
 CFLAGS         ?=      -O2 -pipe
 .endif

I'll report if the use of ld to produce clang.full still
fails or not.

I've left the other things attempting to lead to less memory
use in place.


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

