IRC channel logs

<almuhs>damo22: i don't get boot hurd with rumpkernel in upstream+apic gnumach. Have I to enable some flag in configure to enable rumpkernel?

<almuhs>i added noide to GRUB gnumach line, but only works with the default gnumach kernel

<almuhs>ext2fs: part: 1 device: wd0: No such device or address

<almuhs>seems that this gnumach is not loading rumpkernel

<youpi>you can use --disable-ide to disable gnumach's ide driver

<almuhs>ok, i will try it

<almuhs>same

<almuhs>same error

<almuhs>i am adding this in multiboot line, after gnumach file

<youpi>it's a configure option

<youpi>kernel options never have --

<almuhs>oh, now understand

<almuhs>btw, the latest commit 990f549999310abc2a47ebc6f7782d2e05ae8c96 cause a compiing error

<almuhs>vm/memory_object_user.user.c:14:10: fatal error: mach/msg_type.h: No existe el fichero o el directorio

<almuhs> 14 | #include <mach/msg_type.h>

<almuhs> | ^~~~~~~~~~~~~~~~~

<youpi>upgrade your mig

<almuhs>ok. At moment, i compiled with the previous commit using --disable-ide, and now gnumach boots

<almuhs>now i'm trying with --enable-cpus=2 . There are not driver errors, but it's really slow. It seems stuck after load "auth" module. I go to wait a time to check if is an error or simply a very slow boot

<almuhs>no progress yet

<Pellescours>almuhs: did you applied debian patches? noide is not in master

<youpi>no need for it, you can just pass --disable-ide

<Pellescours>and do and do you have eumpdisk in your grub modules boit file ?

<almuhs>with --disable-ide rumpkernel works

<almuhs>now i'm trying the smp problems which damo22 told ours some days ago

<almuhs>with 2 or 4 cpus, the boot seems freeze after load auth module

<almuhs>with 8 cpus, the boot crash after load AP tables

<almuhs>i suspect that crash could be due to a fail in memory reserve

<almuhs>maybe there are not enough space for stacks?

<almuhs>this is the situation with 4 cpus https://pasteboard.co/EEj6jCyEDorf.png

<almuhs>and this with 8 cpus https://pasteboard.co/emen9BTBMMdv.png

<almuhs>very interesting the message "no more room in gnumach"

<almuhs>it indicates an address. we need to check what is in this address

<youpi>see the corresponding printf in the code: it's the map

<youpi>here, the kernel map

<almuhs>a pagging problem?

<youpi>no

<youpi>no more room in a map means lack of addressing space

<youpi>paging comes only after that

<almuhs>true

<almuhs>then my suspect are close to situation: there are not enough memory for anything

<almuhs>but, what is this anything which demands memory?

<youpi>you can put some assert there and use kdb to get a backtrace

<youpi>then you'll know what doesn't find room

<almuhs>yes

<almuhs>which is the max amount of memory which 32-bit gnumach supports?

<youpi>see the MAX macros

<youpi>and pmap_bootstrap, etc.

<almuhs>ok

<youpi>the page table, basically

<almuhs>yes

<almuhs>i go to lunch. I return in some time

<luckyluke>almuhs: maybe you can increase VM_KERNEL_MAP_SIZE

<almuhs>luckyluke: where is this?

<youpi>in the output of grep

<almuhs>...

<youpi>aka: git grep VM_KERNEL_MAP_SIZE will tell you

<almuhs>ok...

<youpi>truth is: I have no idea where it ise

<youpi>what I know is that git grep can tell me

<youpi>so I just never remember such things

<almuhs>i386/i386/vm_param.h:82:#define VM_KERNEL_MAP_SIZE (512 * 1024 * 1024)

<almuhs>512 MB?

<almuhs>it's strange to fill all this size

<youpi>see the line just below

<youpi>it's much less than that

<almuhs>#define VM_KERNEL_MAP_SIZE (152 * 1024 * 1024)

<almuhs>152 MB

<almuhs>it's big too. But i will try to increase to 256 MB

<almuhs>with 256 MB, same problem

<almuhs>increasing to 512MB now fails in a later step

<almuhs> https://pasteboard.co/DIH6GB9NmgnX.png

<almuhs>but, anyway, it could be a memory leak. It so strange to fill so much memory during the boot

<luckyluke>how much memory do you set for the vm? I had issues only with 8G and more

<almuhs>2G

<luckyluke>also, to get the full logs, you could redirect the console to an emulated serial device, e..g with "-nographic" on qemu and adding "console=com0" to gnumach cmdline

<almuhs>it's a 32-bit installation, so 4G is the max

<almuhs>yes

<almuhs> https://pasteboard.co/34UIVXWm8mkV.png

<almuhs>with 256 MB, now there are other error

<luckyluke>maybe there is something that is allocated for each cpu configured, e.g. a pmap or some big data structure

<almuhs>yes, i think

<almuhs>must be a memory leak in some table

<luckyluke>you can also try to attach with gdb and break on Panic()

<almuhs>yes

<luckyluke>to inspect the per-cpu data structures etc.

<almuhs>even the error tell me the file and line

<almuhs>kr = vm_map_enter(parent, &addr, size,

<almuhs> (vm_offset_t) 0, TRUE,

<almuhs> vm_submap_object, (vm_offset_t) 0, FALSE,

<almuhs> VM_PROT_DEFAULT, VM_PROT_ALL, VM_INHERIT_DEFAULT);

<almuhs> if (kr != KERN_SUCCESS)

<almuhs> panic("kmem_submap");

<almuhs>the panic is this

<youpi>look at the backtrace, to know what triggered it

<luckyluke>yes but from that you only see the failed allocation

<luckyluke>with gdb you can see the local/global variables and everything

<almuhs>yes, i check this later

IRC channel logs

2023-02-22.log