code runs as expected but kernel shows "no enough memory for the allocation"

[laurento]

I compiled my small tool using tinygo because I'm targeting a Raspberry Pi first generation with 256MB of RAM, so every byte counts :) It works just fine. However I just noticed that it generates the following kernel error every time it runs.

[33719.335172] __vm_enough_memory: pid: 7816, comm: mystuff, no enough memory for the allocation

If I compile it using GOOS=linux GOARCH=arm GOARM=6 tinygo build mystuff.go. If I run the same code but compiled it using go the kernel doesn't seem to complain.

For my tests I'm using:

Linux raspi 6.1.0-17-rpi #1 Debian 6.1.69-1 (2023-12-30) armv6l GNU/Linux
tinygo version 0.30.0 linux/amd64 (using go version go1.21.7 and LLVM version 16.0.1)

[peritonow]

I am new to tinygo as well, but there is no much to go on with the information you provided.

From my understanding this go code is translate to c and recompiled with "gcc" ...
https://forum.arduino.cc/t/not-enough-memory-coding-troubleshooting/417437 and the error is from c compiler.

so this might be a bug withing c library and nothing to do with tinygo... also might be related to go stack escape that embedded programming might not tolerate and you need to find out the problem lines within your go code. Hard to say.

last time over 30 yrs ago I was doing embedded in c ... program running just fine on linux would not work in embedded since hardware or embedded library coded for the specific chip set would not allow to exceed physical hardware limitation that library was coded for.

[aykevl]

It's probably this code:

tinygo/src/runtime/runtime_unix.go

Lines 220 to 237 in 6384eca

	// Allocate a large chunk of virtual memory. Because it is virtual, it won't
	// really be allocated in RAM. Memory will only be allocated when it is
	// first touched.
	heapMaxSize = 1 * 1024 * 1024 * 1024 // 1GB for the entire heap
	for {
	addr := mmap(nil, heapMaxSize, flag_PROT_READ|flag_PROT_WRITE, flag_MAP_PRIVATE|flag_MAP_ANONYMOUS, -1, 0)
	if addr == unsafe.Pointer(^uintptr(0)) {
	// Heap was too big to be mapped by mmap. Reduce the maximum size.
	// We might want to make this a bit smarter than simply halving the
	// heap size.
	// This can happen on 32-bit systems.
	heapMaxSize /= 2
	continue
	}
	heapStart = uintptr(addr)
	heapEnd = heapStart + heapSize
	break
	}

I'm surprised it prints a warning. Maybe the kernel really doesn't like an 1GB allocation? (It should just return an error if it is too big).

From my understanding this go code is translate to c and recompiled with "gcc" ...
https://forum.arduino.cc/t/not-enough-memory-coding-troubleshooting/417437 and the error is from c compiler.

It is not, we use LLVM as a compiler toolchain (and don't use GCC anywhere in the process).

[laurento]

Perhaps this kernel patch is relevant https://lore.kernel.org/linux-mm/18dc0807-5596-3aad-350d-3673a033bca8@redhat.com/T/

Edit: This thread too https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1031520

To my understanding, if brk triggers a 'no enough memory for the allocation'
condition, it should simply return an error code to the user space
process, and not generate any kernel level difficulties. brk just
reserves the memory, and does not actually allocate it until it is
written, so no oom machinery should be involved.