Re: Silent hang in buildworld, was Re: Invoking -v for clang during buildworld

From: Mark Millard <marklmi_at_yahoo.com>
Date: Wed, 20 Jan 2021 09:51:33 -0800
On 2021-Jan-19, at 17:42, Mark Millard <marklmi at yahoo.com> wrote:

> On 2021-Jan-18, at 21:12, Mark Millard <marklmi at yahoo.com> wrote:
> 
>> On 2021-Jan-18, at 19:19, Mark Millard <marklmi at yahoo.com> wrote:
>> 
>>> . . .
>>>> FYI: I re-established my access to a RPi2B V1.1 and made
>>>> it report: "maximum recommended amount (468832 pages)"
>>>> 
>>>> (The figure can vary some from release to release.)
>>>> 
>>>> 468832*4096 == 1920335872 or a little over 1831 MiBytes
>>>> 
>>>> For the 4096 Byte pages, that means that the following from
>>>> gpart fits without complaint (size is in blocks, not pages):
>>>> 
>>>> 413140992    3686400  da0p2  freebsd-swap  (1.8G)
>>>> 
>>>> 3686400*512 is a little over 1.75 GiByte or 1800 MiByte. So
>>>> I've left some room below 1831 MiBytes, but not a lot.
>>>> 
>>>> FYI about my build experiment that is running:
>>>> 
>>>> # sysctl hw.physmem
>>>> hw.physmem: 979042304
>>>> 
>>>> which, in recent times for armv7, I can (and did) set in
>>>> /boot/loader.conf on a faster cortex-A7 SBC (that can boot
>>>> the same media but has more RAM).
>>>> 
>>>> So I tried a -j4 build, but with LDFLAGS.lld+= -Wl,--threads=1
>>>> in use and my other particular src.conf/make.conf like content
>>>> (so the builds do likely differ from yours in various ways).
>>>> My build is producing a non-debug build (but with -g symbols).
>>>> Somewhat after where your buildworld.log stops, my odd variant
>>>> of top was reporting:
>>>> 
>>>> Mem:  . . . , 753672Ki MaxObsActive, 200412Ki MaxObsWired, 892732Ki MaxObs(Act+Wir)
>>>> Swap: . . . , 145832Ki MaxObsUsed
>>>> 
>>>> and top was also showing lots of processes as having "0B" RES
>>>> in STATE "wait" or "nanslp" (so, apparently swapped out, not paging).
>>>> ("MaxObs" is short for "maximum observed".)
>>>> 
>>>> For comparison, your swapscript.log reported a maximum total of
>>>> 346192 KiBytes "Used" for swap, about 98% into the log file.
>>>> 
>>>> (Time goes by . . .)
>>>> 
>>>> It finished with building libllvm and is part way into building
>>>> libclang. This is probably well past where your hangup happened,
>>>> given that your published buildworldlog file stopped with
>>>> libllvm's Target/ARM/ARMMCInstLower.o . My odd top now shows:
>>>> 
>>>> Mem:  . . . , 753672Ki MaxObsActive, 200412Ki MaxObsWired, 892732Ki MaxObs(Act+Wir)
>>>> Swap: . . . , 392328Ki MaxObsUsed
>>>> 
>>>> The build continues to run. I'll let you know how it goes.
>>>> . . .
>>> 
>>> Just after libclang finished my odd top showed:
>>> 
>>> Mem:  . . . , 753672Ki MaxObsActive, 200412Ki MaxObsWired, 892736Ki MaxObs(Act+Wir)
>>> Swap: . . . , 537588Ki MaxObsUsed
>>> 
>>> After liblldb:
>>> 
>>> Mem:  . . . , 753672Ki MaxObsActive, 200412Ki MaxObsWired, 899276Ki MaxObs(Act+Wir)
>>> Swap: . . . , 537588Ki MaxObsUsed
>>> 
>>> Much later, after the lldb program had been built:
>>> 
>>> Mem:  . . . , 765700Ki MaxObsActive, 200412Ki MaxObsWired, 954116Ki MaxObs(Act+Wir)
>>> Swap: . . . , 537588Ki MaxObsUsed
>>> 
>>>>>> World build completed on Mon Jan 18 19:10:08 PST 2021
>>>>>> World built in 72960 seconds, ncpu: 4, make -j4
>>> 
>>> This was building from scratch what was already installed:
>>> 
>>> # ~/fbsd-based-on-what-freebsd-main.sh 
>>> merge-base: 818390ce0ca539300dd15d7a817784f1e3f7a9b8
>>> merge-base: CommitDate: 2021-01-13 21:27:44 +0000
>>> 4180404713ec (HEAD -> mm-src) mm-src snapshot for mm's patched build in git context.
>>> 818390ce0ca5 (freebsd/main, freebsd/HEAD, pure-src, main) arm64: fix early devmap assertion
>>> FreeBSD OPiP2E_RPi2v11 13.0-CURRENT FreeBSD 13.0-CURRENT mm-src-c255938-g4180404713ec GENERIC-NODBG  arm armv7 1300135 1300135
>>> 
>>> This suggests that you should be able to build on the RPi2B v1.1,
>>> using -j4, with appropriate configuration for what and how to build.
>>> 
>>> 
>>> It is now building the matching kernel, my GENERIC-NODBG style.
>> 
>> Done:
>> 
>>>>> Kernel build for GENERIC-NODBG completed on Mon Jan 18 20:33:26 PST 2021
>>>>> Kernel(s)  GENERIC-NODBG built in 4998 seconds, ncpu: 4, make -j4
>> 
>> So, World+Kernel in somewhat under 22 hours.
>> 
>> The "MaxObs*" figures were unchanged, so:
>> 
>> Mem:  . . . , 765700Ki MaxObsActive, 200412Ki MaxObsWired, 954116Ki MaxObs(Act+Wir)
>> Swap: . . . , 537588Ki MaxObsUsed
>> 
>> This suggests that, for now, 800 MiByte of swap would be something
>> more than 1.5 times what it actually used and 1050 MiBytes would
>> be something like 2.0 times what it actually used, so leaving some
>> notable margin for variations in peek usage, at least when linker
>> threading is avoided.
>> 
>> 
>> 
>> As for what I used to control "what and how to build" . . .
>> 
>> # more ~/sys_build_scripts.armv7-host/make_armv7_nodebug_clang_bootstrap-armv7-host.sh 
>> kldload -n filemon && \
>> script ~/sys_typescripts/typescript_make_armv7_nodebug_clang_bootstrap-armv7-host-$(date +%Y-%m-%d:%H:%M:%S) \
>> env __MAKE_CONF="/root/src.configs/make.conf" SRCCONF="/dev/null" SRC_ENV_CONF="/root/src.configs/src.conf.armv7-clang-bootstrap.armv7-host" \
>> WITH_META_MODE=yes \
>> WORLD_FLAGS="${WORLD_FLAGS} UBLDR_LOADADDR=0x42000000" \
>> MAKEOBJDIRPREFIX="/usr/obj/armv7_clang/arm.armv7" \
>> make $*
>> 
>> (In my context, UBLDR_LOADADDR is ignored by anything that
>> can not use the figure given. So I've no bothered to be
>> more selective about having it in the armv7 builds.)
>> 
>> # more ~/src.configs/make.conf
>> LDFLAGS.lld+= -Wl,--threads=1
>> 
>> # more ~/src.configs/src.conf.armv7-clang-bootstrap.armv7-host
>> TO_TYPE=armv7
>> #
>> KERNCONF=GENERIC-NODBG
>> TARGET=arm
>> .if ${.MAKE.LEVEL} == 0
>> TARGET_ARCH=${TO_TYPE}
>> .export TARGET_ARCH
>> .endif
>> #
>> #WITH_CROSS_COMPILER=
>> WITH_SYSTEM_COMPILER=
>> WITH_SYSTEM_LINKER=
>> #
>> WITH_LIBCPLUSPLUS=
>> WITHOUT_BINUTILS_BOOTSTRAP=
>> WITH_ELFTOOLCHAIN_BOOTSTRAP=
>> #Disables avoiding bootstrap: WITHOUT_LLVM_TARGET_ALL=
>> WITHOUT_LLVM_TARGET_AARCH64=
>> WITH_LLVM_TARGET_ARM=
>> WITHOUT_LLVM_TARGET_MIPS=
>> WITHOUT_LLVM_TARGET_POWERPC=
>> WITHOUT_LLVM_TARGET_RISCV=
>> WITHOUT_LLVM_TARGET_X86=
>> WITH_CLANG=
>> WITH_CLANG_IS_CC=
>> WITH_CLANG_FULL=
>> WITH_CLANG_EXTRAS=
>> WITH_LLD=
>> WITH_LLD_IS_LD=
>> WITHOUT_BINUTILS=
>> #
>> WITH_LLDB=
>> #
>> WITH_BOOT=
>> WITHOUT_LIB32=
>> #
>> #
>> WITHOUT_WERROR=
>> #WERROR=
>> MALLOC_PRODUCTION=
>> WITH_MALLOC_PRODUCTION=
>> WITHOUT_ASSERT_DEBUG=
>> WITHOUT_LLVM_ASSERTIONS=
>> #
>> # Avoid stripping but do not control host -g status as well:
>> DEBUG_FLAGS+=
>> #
>> WITH_REPRODUCIBLE_BUILD=
>> WITH_DEBUG_FILES=
>> #
>> # Use of the .clang 's here avoids
>> # interfering with other C<?>FLAGS
>> # usage, such as ?= usage.
>> CFLAGS.clang+= -mcpu=cortex-a7
>> CXXFLAGS.clang+= -mcpu=cortex-a7
>> CPPFLAGS.clang+= -mcpu=cortex-a7
>> 
>> (I do not claim that you would want WITH_REPRODUCIBLE_BUILD .
>> I just happen to have been experimenting with it. You might
>> not want to be explicit about the cpu to target. You might
>> not want WITH_CLANG_EXTRAS .)
>> 
>> # more /usr/fbsd/mm-src/sys/arm/conf/GENERIC-NODBG
>> include "GENERIC"
>> 
>> ident   GENERIC-NODBG
>> 
>> makeoptions     DEBUG=-g                # Build kernel with gdb(1) debug symbols
>> 
>> options         AUDIT                   # Not enabled by default in armv7/v6 kernels
>>                                       # Enabled here to allow kyua test runs to
>>                                       # possibly report auditing works.
>> 
>> options         ALT_BREAK_TO_DEBUGGER
>> 
>> options         KDB                     # Enable kernel debugger support
>> 
>> # For minimum debugger support (stable branch) use:
>> options         KDB_TRACE               # Print a stack trace for a panic
>> options         DDB                     # Enable the kernel debugger
>> 
>> # Extra stuff:
>> #options        VERBOSE_SYSINIT=0       # Enable verbose sysinit messages
>> #options        BOOTVERBOSE=1
>> #options        BOOTHOWTO=RB_VERBOSE
>> options         ALT_BREAK_TO_DEBUGGER   # Enter debugger on keyboard escape sequence
>> options         KLD_DEBUG
>> #options        KTR
>> #options        KTR_MASK=KTR_TRAP
>> ##options       KTR_CPUMASK=0xF
>> #options        KTR_VERBOSE
>> 
>> # Disable any extra checking for. . .
>> nooptions       INVARIANTS              # Enable calls of extra sanity checking
>> nooptions       INVARIANT_SUPPORT       # Extra sanity checks of internal structures, required by INVARIANTS
>> nooptions       WITNESS                 # Enable checks to detect deadlocks and cycles
>> nooptions       WITNESS_SKIPSPIN        # Don't run witness on spinlocks for speed
>> nooptions       DEADLKRES               # Enable the deadlock resolver
>> nooptions       MALLOC_DEBUG_MAXZONES   # Separate malloc(9) zones
>> nooptions       DIAGNOSTIC
>> nooptions       BUF_TRACKING
>> nooptions       FULL_BUF_TRACKING
>> nooptions       USB_DEBUG
>> nooptions       USB_REQ_DEBUG
>> nooptions       USB_VERBOSE
>> 
>> The /boot/loader.conf file and the /etc/sysctl.conf files
>> both contained:
>> 
>> vm.pageout_oom_seq=120
>> vm.pfault_oom_attempts=-1
>> 
>> (The hw.physmem=979042304 in /boot/loader.conf was very-special,
>> to better approximate your environment. I also controlled the
>> cpu frequency used via a line in /etc/sysctl.conf . I do not
>> bother with such non-default frequency usage [or related settings]
>> for RPi*'s with the pre-RPi4B style power connections but do
>> control the frequency for the OPi+2E.)
> 
> The following had been left implicit about my context and
> how it manages memory space use.
> 
> I'll note that I do not use tmpfs or other such memory based
> file system techniques that could compete for RAM/swap. What
> is in use for the only file system involved is just the
> root file system:
> 
> # df -m
> Filesystem            1M-blocks  Used  Avail Capacity  Mounted on
> /dev/gpt/BPIM3root       195378 63940 115808    36%    /
> devfs                         0     0      0   100%    /dev
> 
> It is a USB SSD. The swap partition is also on that same
> media. (The BPIM3 based name dates back to before the
> BPI-M3 power connection failed and I switched to the
> OPi+2E.)
> 
> I'll note that I've started a new from-scratch build without
> LDFLAGS.lld+= -Wl,--threads=1 . So at some point I'll have
> information about how much of a difference (+/-) in swap
> usage it actually made for with vs. without, if any.

Looks like, for such 4-core contexts, that bothering
with LDFLAGS.lld+= -Wl,--threads=1 is typically a
waste of effort for both swap usage and time . . .

With LDFLAGS.lld+= -Wl,--threads=1 :

Mem:  . . . , 765700Ki MaxObsActive, 200412Ki MaxObsWired, 954116Ki MaxObs(Act+Wir)
Swap: . . . , 537588Ki MaxObsUsed

without:

Mem:  . . ., 715756Ki MaxObsActive, 194816Ki MaxObsWired, 903132Ki MaxObs(Act+Wir)
Swap: . . ., 557208Ki MaxObsUsed


With LDFLAGS.lld+= -Wl,--threads=1 :

World built in 72960 seconds, ncpu: 4, make -j4
Kernel(s)  GENERIC-NODBG built in 4998 seconds, ncpu: 4, make -j4

without:

World built in 72804 seconds, ncpu: 4, make -j4
Kernel(s)  GENERIC-NODBG built in 4824 seconds, ncpu: 4, make -j4


So, just not that much of a difference compared to the overall
sizes or times involved.

===
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)
Received on Wed Jan 20 2021 - 16:51:42 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:41:26 UTC