Re: rump_server, "panic: uvm_km_alloc failed" on module load

Thank you for replying, I am new to the system so any direction
means a lot.

I've tracked down uvm_km_alloc failed to PaX mprotect restrictions
for underlying mmap. t_modautoload passes as it has +a +m paxctl
flags.

Now I am hitting following problem: any module which calls functions
from modcmd segfaults. For instance des.kmod wihtout any calls from
modcmd loads fine, while examples/hello.kmod crashes.

Here is back trace. gdb doesn't see symbols for loaded module.


--------------------------------
./rump_server -s -v -m 
/stand/amd64/10.99.14/modules/examples/hello/hello.kmod   unix://sock

(gdb) bt
#0  0x00000000680d696c in ?? ()
#1  0x000000007fffe039 in ?? ()
#2  0x00007f7fffebb300 in ?? ()
#3  0x0000796168125c87 in module_do_load (name=0x79616814f602 
<mutex_exit+108> 
"\220H\211\354]\303UH\211\345H\203\354\bH\211}\370\270\001", isdep=127, 
flags=-149928913, props=0x796168151d54 <uvm_km_protect>,
     modp=0x7f7ff7109a55 <_rtld_shared_exit+51>, modclass=31073, 
autoload=152) at 
/home/nemanja/source/src/lib/librump/../../sys/rump/../kern/kern_module.c:1420
Backtrace stopped: frame did not save the PC

...

1417            module_load_evcnt(mod);         /* Attach any static 
evcnt needed */
(gdb) bt
#0  module_do_load (name=0x7a53c3b525c0 "/module.mod", isdep=false, 
flags=0, props=0x0, modp=0x0, modclass=MODULE_CLASS_ANY, autoload=false) 
at 
/home/nemanja/source/src/lib/librump/../../sys/rump/../kern/kern_module.c:1417
#1  0x00007a53c42b3187 in module_load (filename=0x7a53c3b525c0 
"/module.mod", flags=0, props=0x0, modclass=MODULE_CLASS_ANY) at 
/home/nemanja/source/src/lib/librump/../../sys/rump/../kern/kern_module.c:713
#2  0x00007a53c4228146 in handle_modctl_load (ml_filename=0xf1fb73 
"/module.mod", ml_flags=0, ml_props=0x0, ml_propslen=0) at 
/home/nemanja/source/src/lib/librump/../../sys/rump/../kern/sys_module.c:99
#3  0x00007a53c42288ec in sys_modctl (l=0x7a53c3b52180, 
uap=0x7f7fff7bb878, retval=0x7f7fff7bb868) at 
/home/nemanja/source/src/lib/librump/../../sys/rump/../kern/sys_module.c:285
#4  0x00007a53c42e8f88 in sy_call (sy=0x7a53c433c750 
<rumpns_sysent+5904>, l=0x7a53c3b52180, uap=0x7f7fff7bb878, 
rval=0x7f7fff7bb868) at 
/home/nemanja/source/src/lib/librump/../../sys/rump/../sys/syscallvar.h:65
#5  0x00007a53c42e9057 in sy_invoke (sy=0x7a53c433c750 
<rumpns_sysent+5904>, l=0x7a53c3b52180, uap=0x7f7fff7bb878, 
rval=0x7f7fff7bb868, code=246) at 
/home/nemanja/source/src/lib/librump/../../sys/rump/../sys/syscallvar.h:94
#6  0x00007a53c42ea601 in rump_syscall (num=246, data=0x7f7fff7bb878, 
dlen=16, retval=0x7f7fff7bb868) at 
/home/nemanja/source/src/lib/librump/../../sys/rump/librump/rumpkern/rump.c:782
#7  0x00007a53c42d8a08 in rump___sysimpl_modctl (cmd=0, 
arg=0x7f7fff7bb920) at 
/home/nemanja/source/src/lib/librump/../../sys/rump/librump/rumpkern/rump_syscalls.c:2716
#8  0x0000000000f1e5b4 in main (argc=<optimized out>, argv=<optimized 
out>) at 
/home/nemanja/source/src/usr.bin/rump_server/../rump_allserver/rump_allserver.c:420
--------------------------------



On 7/10/25 23:43, Christoph Badura wrote:
> On Thu, Jul 10, 2025 at 06:27:59PM +0200, Nemanja Simonovic wrote:
>> I tried asking this on users list but got no reply.
> 
> I saw it but didn't have time to reply yet.
> 
>> When trying to load modules to rump_server I am getting kernel
>> "panic: uvm_km_alloc failed". Both with -m parameter and with
>> rump.modload.
> 
> That message comes from:
> https://nxr.netbsd.org/xref/src/sys/rump/librump/rumpkern/vm.c#888
> 
>> I not sure if I am doing something wrong or hitting a bug?
> 
> We do have 3 ATF tests that deal with rump module loading:
> 
> src/tests/rump/modautoload/t_modautoload.c
> src/tests/rump/rumpkern/t_modcmd.c
> src/tests/rump/rumpkern/t_modlinkset.c
> 
> These tests pass OK:
> https://releng.netbsd.org/b5reports/amd64/2025/2025.07.10.11.44.12/test.log
> 
> But the use neither -m nor rump.modload.  Could be the implementation for
> -m or rump.modload has bitrotted.  I don't know.
> Could be you are doing something wrong.
> 
> Have you tried debugging this with gdb?
> 
> Set a break point in rumpkern/vmc line 878, the call to rumpuser_anonmap(),
> and step through that function and see where it fails.
> 
> --chris