Building node.js for ramips

eduperez · January 5, 2024, 9:57am

Recent versions of node.js require a FPU, so packages related to node.js where deactivated on platforms that lack a FPU; so far, so good.

However (and this is a cloudy area for me), there is a kernel option (KERNEL_MIPS_FPU_EMULATOR) that could be used to enable in-kernel FPU emulation on these platforms. According to some sources, it should be possible to enable that option, and thus build node.js on ramips.

I am comfortable building and running my own modified images, I have done this in the past. But I am not sure whether this is possible, or if there are reasons to not use KERNEL_MIPS_FPU_EMULATOR at all, and this is indeed a lost cause.

In case this is doable, is it as simple as enabling this option, then configuring and building the image? Or are other tasks involved, that I should take into consideration?

slh · January 5, 2024, 1:06pm

It is doable, and that is everything needed.

The FPU emulation has been disabled on mips for two reasons:

it's not needed for typical routing related packages and avoiding those code paths does speed up all packages that do math, but don't strictly need floating point math
the mips FPU emulation is extremely slow and inefficient, things that will take a few seconds on a hardware FPU can (literally) take hours with KERNEL_MIPS_FPU_EMULATOR. This has e.g. bitten Debian in the past, with (predominantly) test suites timing out on the mips buildds, as individual test appeared to be stalled with no output for ages.

The emulations may help you to fullfill ISA requirements for packages that demand FPU support, but don't actually use it significantly, because otherwise performance would be too bad to be usable. Emulated FPU support is always bad, but it's particularly abysmal on mips.

Consider ARMv7/ ARMv8 based devices for anything that needs floating point math, there it is part of the ISA baseline and reasonably performant - and we have a number if interesting devices falling into that category (ipq40xx, ipq806x, ipq807x, mvebu, mt7622, filogic 820/ 830/ 880, sunxi, rockchip, RPi).

eduperez · January 5, 2024, 1:57pm

Many thanks for your answer!

I do not plan on doing any heavy calculations, I just want to try to run a specific node.js program, and I do not expect that this software will do much floating point, either.

I think it's worth a try, and see what happens. As a last resort, I will have to dedicate a RPi I have lying around here.

eduperez · January 14, 2024, 11:35pm

As I posted in another thread, I tried (and failed miserably)...

I first downloaded the sources:
git clone https://git.openwrt.org/openwrt/openwrt.git && cd openwrt

Then installed all the feeds:
./scripts/feeds update -a && ./scripts/feeds install -a

As I want to start with an image similar to an official build, I downloaded the configuration for my platform:
wget https://downloads.openwrt.org/snapshots/targets/ramips/mt7621/config.buildinfo -O .config

I tried to configure the kernel first:
make -j6 kernel_menuconfig

But this triggered the normal menuconfig first... I selected my device, then checked that node.js was disabled, as expected. After saving the config file, the process continued, then failed because some library was missing; I could fix this with:
export LD_LIBRARY_PATH=${PWD}/staging_dir/host/lib/

Back to kernel_menuconfig, and I could enable FPU emulation. I tried to go back to menuconfig and enable node.js, but it was still disabled. So, I edited the Makefile at feeds/packages/lang/node/, and changed the dependencies:

Before: DEPENDS:=@HAS_FPU @(i386||x86_64||arm||aarch64) \
After: DEPENDS:=@(i386||x86_64||arm||aarch64||mipssel) \

Back to menuconfig again, and this time I could enable the node.js package.
Next step was to build the image:
make -j6 download world

However, this fails (precisely on node.js) with this message:

  /home/eperez/Downloads/OpenWrt/openwrt/staging_dir/host/bin/g++ -m32 -o /home/eperez/Downloads/OpenWrt/openwrt/build_dir/target-mipsel_24kc_musl/node-v20.11.0/out/Release/obj.host/simdutf/deps/simdutf/simdutf.o ../deps/simdutf/simdutf.cpp '-D_GLIBCXX_USE_CXX11_ABI=1' '-DNODE_OPENSSL_CONF_NAME=nodejs_conf' '-DICU_NO_USER_DATA_OVERRIDE' '-D__STDC_FORMAT_MACROS' -I/home/eperez/Downloads/OpenWrt/openwrt/staging_dir/target-mipsel_24kc_musl/usr/include -I../deps/simdutf  -pthread -Wall -Wextra -Wno-unused-parameter -m32 -O3 -fno-omit-frame-pointer -fno-rtti -fno-exceptions -std=gnu++17 -MMD -MF /home/eperez/Downloads/OpenWrt/openwrt/build_dir/target-mipsel_24kc_musl/node-v20.11.0/out/Release/.deps//home/eperez/Downloads/OpenWrt/openwrt/build_dir/target-mipsel_24kc_musl/node-v20.11.0/out/Release/obj.host/simdutf/deps/simdutf/simdutf.o.d.raw   -c
In file included from /usr/include/features.h:515,
                 from /usr/include/c++/13/x86_64-redhat-linux/32/bits/os_defines.h:39,
                 from /usr/include/c++/13/x86_64-redhat-linux/32/bits/c++config.h:679,
                 from /usr/include/c++/13/cstring:41,
                 from ../deps/simdutf/simdutf.h:5,
                 from ../deps/simdutf/simdutf.cpp:3:
/usr/include/gnu/stubs.h:7:11: fatal error: gnu/stubs-32.h: No such file or directory
    7 | # include <gnu/stubs-32.h>
      |           ^~~~~~~~~~~~~~~~
compilation terminated.

And I got stuck at this point, because I do not know whether this is related to forcing the compilation, or if the package is failing for everybody...

eduperez · January 15, 2024, 5:00pm

I am a bit confused that the compilation seems to be using includes from my OS, instead of the ones downloaded from OpenWrt... But I cannot blame the changes I made, so I decided to install the 32bit headers and try again:

/home/eperez/Downloads/OpenWrt/openwrt/staging_dir/host/bin/g++ -m32 -o /home/eperez/Downloads/OpenWrt/openwrt/build_dir/target-mipsel_24kc_musl/node-v20.11.0/out/Release/node_js2c -pthread -rdynamic -pthread -m32  -Wl,--start-group /home/eperez/Downloads/OpenWrt/openwrt/build_dir/target-mipsel_24kc_musl/node-v20.11.0/out/Release/obj.host/node_js2c/tools/js2c.o /home/eperez/Downloads/OpenWrt/openwrt/build_dir/target-mipsel_24kc_musl/node-v20.11.0/out/Release/obj.host/deps/simdutf/libsimdutf.a /home/eperez/Downloads/OpenWrt/openwrt/build_dir/target-mipsel_24kc_musl/node-v20.11.0/out/Release/obj.host/deps/uv/libuv.a -L/home/eperez/Downloads/OpenWrt/openwrt/staging_dir/target-mipsel_24kc_musl/usr/lib -lm -ldl -lrt -Wl,--end-group
/usr/bin/ld: skipping incompatible /usr/lib/gcc/x86_64-redhat-linux/13/libstdc++.so when searching for -lstdc++
/usr/bin/ld: cannot find -lstdc++: No such file or directory
/usr/bin/ld: skipping incompatible /usr/lib/gcc/x86_64-redhat-linux/13/libstdc++.so when searching for -lstdc++
collect2: error: ld returned 1 exit status

So... for some reason that I cannot understand, the build process is trying to use the headers and libraries from my OS, and that does not work at all, obviously. I have to figure out what did I do to wreck the compilation.

eduperez · January 16, 2024, 7:33pm

I the following line to "build_dir/target-mipsel_24kc_musl/node-v20.11.0/deps/simdutf/simdutf.gyp", and the compilation continued past the previous error:

      'include_dirs!': [ '<!@(echo "$STAGING_DIR"/usr/include)' ],

I'm still not sure whether I did something wrong or is a bug in the code...

However, now the compilation fails with this error (that I also cannot explain):

make[4]: Entering directory '/home/eperez/Downloads/OpenWrt/openwrt/build_dir/target-mipsel_24kc_musl/node-v20.11.0'
/home/eperez/Downloads/OpenWrt/openwrt/staging_dir/hostpkg/bin/python3.11 tools/gyp_node.py -f make
gyp: Undefined variable python in /home/eperez/Downloads/OpenWrt/openwrt/build_dir/target-mipsel_24kc_musl/node-v20.11.0/tools/v8_gypfiles/v8.gyp
make[4]: *** [Makefile:176: out/Makefile] Error 1
make[4]: Leaving directory '/home/eperez/Downloads/OpenWrt/openwrt/build_dir/target-mipsel_24kc_musl/node-v20.11.0'
make[3]: *** [Makefile:171: /home/eperez/Downloads/OpenWrt/openwrt/build_dir/target-mipsel_24kc_musl/node-v20.11.0/.built] Error 2
make[3]: Leaving directory '/home/eperez/Downloads/OpenWrt/openwrt/feeds/packages/lang/node'
time: package/feeds/packages/node/compile#0.81#0.25#0.93
    ERROR: package/feeds/packages/node failed to build.
make[2]: *** [package/Makefile:128: package/feeds/packages/node/compile] Error 1
make[2]: Leaving directory '/home/eperez/Downloads/OpenWrt/openwrt'
make[1]: *** [package/Makefile:122: /home/eperez/Downloads/OpenWrt/openwrt/staging_dir/target-mipsel_24kc_musl/stamp/.package_compile] Error 2
make[1]: Leaving directory '/home/eperez/Downloads/OpenWrt/openwrt'
make: *** [/home/eperez/Downloads/OpenWrt/openwrt/include/toplevel.mk:232: world] Error 2

nxhack · January 17, 2024, 1:16am

Recent node.js does not support mips32.

FYI:

github.com/nodejs/node

Cannot compile 18.3.0 on mipsel

opened 05:32PM - 15 Jun 22 UTC

closed 08:02AM - 17 Jun 22 UTC

kapouer

build mips

### Version 18.3.0 ### Platform SMP PREEMPT_DYNAMIC Debian 5.18.2-1 (20…22-06-06) x86_64 GNU/Linux ### Subsystem v8 ### What steps will reproduce the bug? Compile with mipsel target and `--with-mips-arch-variant=r2 --with-mips-fpu-mode=fpxx` ### How often does it reproduce? Is there a required condition? Always ### What is the expected behavior? No error ### What do you see instead? ```console $ g++ -o /<<PKGBUILDDIR>>/out/Release/obj.target/v8_base_without_compiler/deps/v8/src/execution/clobber-registers.o ../deps/v8/src/execution/clobber-registers.cc '-D_GLIBCXX_USE_CXX11_ABI=1' '-DNODE_OPENSSL_CERT_STORE' '-DV8_GYP_BUILD' '-DV8_TYPED_ARRAY_MAX_SIZE_IN_HEAP=64' '-D__STDC_FORMAT_MACROS' '-DV8_TARGET_ARCH_MIPS' '-DCAN_USE_FPU_INSTRUCTIONS' '-D__mips_hard_float=1' '-DV8_HAVE_TARGET_OS' '-DV8_TARGET_OS_LINUX' '-DV8_EMBEDDER_STRING="-node.8"' '-DENABLE_DISASSEMBLER' '-DV8_PROMISE_INTERNAL_FIELD_COUNT=1' '-DOBJECT_PRINT' '-DV8_INTL_SUPPORT' '-DV8_ATOMIC_OBJECT_FIELD_WRITES' '-DV8_ENABLE_LAZY_SOURCE_POSITIONS' '-DV8_USE_SIPHASH' '-DV8_SHARED_RO_HEAP' '-DV8_WIN64_UNWINDING_INFO' '-DV8_ENABLE_REGEXP_INTERPRETER_THREADED_DISPATCH' '-DV8_SNAPSHOT_COMPRESSION' '-DV8_ENABLE_WEBASSEMBLY' '-DV8_ENABLE_JAVASCRIPT_PROMISE_HOOKS' '-DV8_ALLOCATION_FOLDING' '-DV8_ALLOCATION_SITE_TRACKING' '-DV8_SCRIPTORMODULE_LEGACY_LIFETIME' '-DV8_ADVANCED_BIGINT_ALGORITHMS' '-DICU_UTIL_DATA_IMPL=ICU_UTIL_DATA_STATIC' '-D_MIPS_TARGET_HW' '-D_MIPS_ARCH_MIPS32R2' '-DFPU_MODE_FPXX' -I../deps/v8 -I../deps/v8/include -I/<<PKGBUILDDIR>>/out/Release/obj/gen/inspector-generated-output-root -I../deps/v8/third_party/inspector_protocol -I/<<PKGBUILDDIR>>/out/Release/obj/gen -I/<<PKGBUILDDIR>>/out/Release/obj/gen/generate-bytecode-output-root -I../deps/v8/third_party/zlib -I../deps/v8/third_party/zlib/google -pthread -Wno-unused-parameter -fPIC -Wno-return-type -fno-strict-aliasing -EL -Wno-error=array-bounds -mhard-float -mips32r2 -mfpxx -Wa,-mips32r2 -O3 -fno-omit-frame-pointer -fdata-sections -ffunction-sections -O3 -fno-rtti -fno-exceptions -std=gnu++17 -MMD -MF /<<PKGBUILDDIR>>/out/Release/.deps//<<PKGBUILDDIR>>/out/Release/obj.target/v8_base_without_compiler/deps/v8/src/execution/clobber-registers.o.d.raw -gsplit-dwarf -fPIC -g1 -gsplit-dwarf -fPIC -g1 -gsplit-dwarf -fPIC -g1 -gsplit-dwarf -fPIC -g1 -gsplit-dwarf -fPIC -g1 -c /tmp/ccD9B90Q.s: Assembler messages: /tmp/ccD9B90Q.s:55: Error: float register should be even, was 13 /tmp/ccD9B90Q.s:61: Error: float register should be even, was 15 ``` ### Additional information I have several patches on debian nodejs to make it compile and pass tests for MIPSEL already. So it might be possible that one of them is screwing up. However it is more likely that v8 has a bug, since it was compiling all right with nodejs 16.5.1. This patch fixes the issue, but honestly I did it in a dark room filled with bottles of wine: ```diff --- a/deps/v8/src/codegen/mips/register-mips.h +++ b/deps/v8/src/codegen/mips/register-mips.h @@ -32,9 +32,8 @@ // Currently, MIPS just use even float point register, except // for C function param registers. #define DOUBLE_USE_REGISTERS(V) \ - V(f0) V(f2) V(f4) V(f6) V(f8) V(f10) V(f12) V(f13) \ - V(f14) V(f15) V(f16) V(f18) V(f20) V(f22) V(f24) V(f26) \ - V(f28) V(f30) + V(f0) V(f2) V(f4) V(f6) V(f8) V(f10) V(f12) V(f14) \ + V(f16) V(f18) V(f20) V(f22) V(f24) V(f26) V(f28) V(f30) #define FLOAT_REGISTERS DOUBLE_REGISTERS #define SIMD128_REGISTERS(V) \ ``` I asked some help to debian mips porters, and will report anything new here.

https://groups.google.com/g/v8-dev/c/AU49K-sysjY

eduperez · January 17, 2024, 11:27am

Ok, thanks for the heads up... I hoped that the limitation was on the FPU, but if the CPU is no longer supported, I guess my quest is useless.

system · January 27, 2024, 11:29am

This topic was automatically closed 10 days after the last reply. New replies are no longer allowed.