[cfarm-users] Does cfarm421 need more RAM?

[Pierre Muller]

Hi,

Le 27/03/2024 à 02:58, Jing Luo via cfarm-users a écrit :
> Hi,
> 
> I'm the maintainer of cfarm{420,421,422} and I noticed that cfarm421
> spits out these in the kernel log message buffer:
> 
> ================8<=============8<===============
> [407044.868840] __vm_enough_memory: pid: 1547477, comm: qemu-aarch64,
> not enough memory for the allocation
> [407044.873072] __vm_enough_memory: pid: 1547477, comm: qemu-aarch64,
> not enough memory for the allocation
> [407793.227963] __vm_enough_memory: pid: 1629843, comm: qemu-loongarch6,
> not enough memory for the allocation
> [407793.252490] __vm_enough_memory: pid: 1629843, comm: qemu-loongarch6,
> not enough memory for the allocation
> [408951.450373] __vm_enough_memory: pid: 1782006, comm: qemu-ppc64, not
> enough memory for the allocation
> [408951.455591] __vm_enough_memory: pid: 1782006, comm: qemu-ppc64, not
> enough memory for the allocation
> [409186.357363] __vm_enough_memory: pid: 1813298, comm: qemu-ppc64le,
> not enough memory for the allocation
> [409186.361280] __vm_enough_memory: pid: 1813298, comm: qemu-ppc64le,
> not enough memory for the allocation
> [409864.075096] __vm_enough_memory: pid: 1873876, comm: qemu-riscv64,
> not enough memory for the allocation
> [409864.084699] __vm_enough_memory: pid: 1873876, comm: qemu-riscv64,
> not enough memory for the allocation
> [410768.150482] __vm_enough_memory: pid: 1934927, comm: qemu-sparc64,
> not enough memory for the allocation
> [410768.156067] __vm_enough_memory: pid: 1934927, comm: qemu-sparc64,
> not enough memory for the allocation
> [411410.679436] __vm_enough_memory: pid: 1964674, comm: qemu-x86_64, not
> enough memory for the allocation
> [411410.684506] __vm_enough_memory: pid: 1964674, comm: qemu-x86_64, not
> enough memory for the allocation
> [430615.068394] __vm_enough_memory: pid: 2110282, comm: qemu-aarch64,
> not enough memory for the allocation
> [430615.083721] __vm_enough_memory: pid: 2110282, comm: qemu-aarch64,
> not enough memory for the allocation
> [433503.627242] __vm_enough_memory: pid: 2347231, comm: qemu-ppc64, not
> enough memory for the allocation
> [433503.631411] __vm_enough_memory: pid: 2347231, comm: qemu-ppc64, not
> enough memory for the allocation
> [433935.278186] __vm_enough_memory: pid: 2383232, comm: qemu-ppc64le,
> not enough memory for the allocation
> [433935.283613] __vm_enough_memory: pid: 2383232, comm: qemu-ppc64le,
> not enough memory for the allocation
> [435267.994912] __vm_enough_memory: pid: 2456518, comm: qemu-sparc64,
> not enough memory for the allocation
> [435268.002203] __vm_enough_memory: pid: 2456518, comm: qemu-sparc64,
> not enough memory for the allocation
> [436006.040243] __vm_enough_memory: pid: 2491131, comm: qemu-x86_64, not
> enough memory for the allocation
> [436006.061130] __vm_enough_memory: pid: 2491131, comm: qemu-x86_64, not
> enough memory for the allocation
> [449685.944334] __vm_enough_memory: pid: 3055522, comm: tw2494, not
> enough memory for the allocation
> [449685.948704] __vm_enough_memory: pid: 3055522, comm: tw2494, not
> enough memory for the allocation
> [450558.392384] __vm_enough_memory: pid: 3091900, comm: tw2494, not
> enough memory for the allocation
> [450558.396205] __vm_enough_memory: pid: 3091900, comm: tw2494, not
> enough memory for the allocation
> ================8<=============8<===============
> root at cfarm421:~# free -h
>                  total        used        free      shared  buff/cache
> available
> Mem:            64Gi       4.5Gi        20Gi       1.6Mi        40Gi
>      60Gi
> Swap:             0B          0B          0B
> ================8<=============8<===============
> 
> There is no log about OOM killer. Does cfarm421 need more RAM? Doesn't
> look like it though, the messages troubled me. I may be able to add
> another 64GB to cfarm421 (via qemu VM guest RAM hotplug), hopefully
> without the host crashes and burns.
> 
> PS: if anyone needs a dump of dmesg on cfarm420~422, PM me anytime. I
> see lines like "dosbox-x[pid]: memfd_create() called without MFD_EXEC or
> MFD_NOEXEC_SEAL set" or "segfault at ... likely on CPU ... (core ...,
> socket ...)"

   These failures are all corresponding to several types of tests I am running,
mainly to test Free Pascal Compiler.

   One of these tests consists in using linux-user mode qemu to
run binaries cross-compiled using Free Pascal compiler for another CPU.

   Test tw2494 is specifically a test that checks what is happening
if the program tries to allocate 4096 Tb (=2Gb*2Gb).
It is really just here to check that the OS delivers an appropriate failure.
   I suspect that the failures of the linux-user mode qemu
are due to trying to run the same tw2494 executable cross-compiled for another CPU.

   Another part of the test is to use dosbox-x emulator to run cross-compiled
Free Pascal programs under msdos environment.

   All in all, I would say that there is no need to increase the RAM size.

   Is it really annoying for you to get this kernel message?
   Should I remove those tests from the testsuite on that machine, or is
it OK to have these dmesg entries?

   Sorry for the extra burden.

Pierre Muller
Member of the Free Pascal core developer team