Skip to contentSkip to conversation

restarting tests/sleep

DISCUSSION Serge E. Hallyn 2 participants 3 messages Jun 9, 2008 5 messages not in the archive

lore
Share & tools
lore
#4 lore raw
In reply to a message not in the archive
Quoted 12 lines · Nadia Derbey
Quoted 4 lines · Serge E. Hallyn

Can you check your /var/log/messages for a 'protection fault' message? Then, if there is one for 'sleep', run gdb on sleep, take the IP in the protection fault msg, and do 'x/10 0x8005cd0' or whatever the address is?

No, I'm not getting a protection fault, and I've got nothing generated in the /var/log/messages. But when I reran the test to check what you're asking me to, I didn't reach the same point: I'm not reaching the [vdso] restoring. I'm getting an EFAULT. Here are the last lines from the output:

Quoted 9 lines

=====
DEBUG (cr.c::923) next memseg_t is: start bfd99000 end bfdae000 prot 3 flag 50 offset 3221139456 fnam [stack]
DEBUG (cr.c::969) Delete segment bfc96000 - bfcab000
DEBUG (cr.c::971) Restore segment bfd99000 - bfdae000
[1 cr.c:972 restore_mem()] inj_func(ic, inj_readbuf, 3, remfd, memseg->start, memseg->end - memseg->start) : -14
ERROR (cr.c:926) unexpected item name: '' (size=0)
=====

Yeah that's insurmountable - notice the stack in the process which was fork()ed to be the restarted process topped at bfcab000, while the checkpointed stack topped at bfdae000. You're not allowed to write above the stack. So the only things to do are

1. keep trying the restart in the hopes you get a task with stack topping at or above bfdae000
2. if the checkpointed stack is too high to be likely to be restartable, generate a new checkpoint image and you should get a lower stack top.

(Dave, maybe you had other ideas I haven't considered)

Good thing cryo is just a proof of concept to exploit the kernel code :)

thanks,
-serge

#6 lore raw
In reply to a message not in the archive
Quoted 7 lines · Serge E. Hallyn

Yeah that's insurmountable - notice the stack in the process which was fork()ed to be the restarted process topped at bfcab000, while the checkpointed stack topped at bfdae000. You're not allowed to write above the stack. So the only things to do are

Quoted 7 lines

1. keep trying the restart in the hopes you get a task with stack topping at or above bfdae000
2. if the checkpointed stack is too high to be likely to be restartable, generate a new checkpoint image and you should get a lower stack top.

(Dave, maybe you had other ideas I haven't considered)

Have you tried turning of stack randomization? It should make the stack more dependable at exec. We could also provide a hint somewhere on exec() where to place the stack.

-- Dave

#7 lore raw
Quoted 14 lines · Dave Hansen
Quoted 13 lines · Serge E. Hallyn

Yeah that's insurmountable - notice the stack in the process which was fork()ed to be the restarted process topped at bfcab000, while the checkpointed stack topped at bfdae000. You're not allowed to write above the stack. So the only things to do are

1. keep trying the restart in the hopes you get a task with stack topping at or above bfdae000
2. if the checkpointed stack is too high to be likely to be restartable, generate a new checkpoint image and you should get a lower stack top.

Quoted 5 lines
Quoted 1 line

(Dave, maybe you had other ideas I haven't considered)

Have you tried turning of stack randomization? It should make the stack more dependable at exec. We could also provide a hint somewhere on exec() where to place the stack.

Yup, that should prevent the spurious -EFAULTS. Suka, you were getting these too last week, so I recommend doing

echo 0 > /proc/sys/kernel/randomize_va_space

on startup.

-serge