clvmd coredump issue

clvmd coredump issue

[Liuhua Wang]

Hi, list

I have encountered a coredump when start clvmd on ppc64 as followed:

#p $pc
$1 = (void (*)()) 0x10070c18
#bt
#0  0x0000000010070c18 in _allocate_memory () at mm/memlock.c:130
#1  0x0000000010070d38 in _lock_mem (cmd=<optimized out>) at mm/memlock.c:310
#2  _lock_mem_if_needed (cmd=0x10104f40) at mm/memlock.c:387
#3  0x0000000010011358 in init_clvm (excl_uuid=<optimized out>) at
lvm-functions.c:919
#4  0x000000001000f47c in lvm_thread_fn (arg=0xfffcaf4e530) at clvmd.c:2020
#5  0x00000fff9df6c67c in .start_thread () from /lib64/power6/libpthread.so.0
#6  0x00000fff9de7643c in .__clone () from /lib64/power6/libc.so.6


When the followed four lines are commented then the problem is ok.
I suppose the problem is probably because of alloca.

static void _allocate_memory(void)
{
	void *stack_mem, *temp_malloc_mem;
	struct rlimit limit;

	/* Check if we could preallocate requested stack */
-	if ((getrlimit (RLIMIT_STACK, &limit) == 0) &&
-	    ((_size_stack * 2) < limit.rlim_cur) &&
-         ((stack_mem = alloca(_size_stack))))
-	   _touch_memory(stack_mem, _size_stack);
	/* FIXME else warn user setting got ignored */

	if ((temp_malloc_mem = malloc(_size_malloc_tmp)))
		_touch_memory(temp_malloc_mem, _size_malloc_tmp);

	if ((_malloc_mem = malloc(_size_malloc)))
		_touch_memory(_malloc_mem, _size_malloc);

	free(temp_malloc_mem);
} 

My questions are:
1. Can we delete the four lines, since it is just PREallocate?
2. What is the purpose of _touch_memory()
3. why do first malloc temp_malloc_mem but then free it?

Any information is very appreciated.

Best regards,
Melody 

clvmd coredump issue

[Zdenek Kabelac]

Dne 24.4.2014 08:58, Liuhua Wang napsal(a):
> Hi, list
>
> I have encountered a coredump when start clvmd on ppc64 as followed:
>
> #p $pc
> $1 = (void (*)()) 0x10070c18
> #bt
> #0  0x0000000010070c18 in _allocate_memory () at mm/memlock.c:130
> #1  0x0000000010070d38 in _lock_mem (cmd=<optimized out>) at mm/memlock.c:310
> #2  _lock_mem_if_needed (cmd=0x10104f40) at mm/memlock.c:387
> #3  0x0000000010011358 in init_clvm (excl_uuid=<optimized out>) at
> lvm-functions.c:919
> #4  0x000000001000f47c in lvm_thread_fn (arg=0xfffcaf4e530) at clvmd.c:2020
> #5  0x00000fff9df6c67c in .start_thread () from /lib64/power6/libpthread.so.0
> #6  0x00000fff9de7643c in .__clone () from /lib64/power6/libc.so.6
>
>
> When the followed four lines are commented then the problem is ok.
> I suppose the problem is probably because of alloca.
>
> static void _allocate_memory(void)
> {
> 	void *stack_mem, *temp_malloc_mem;
> 	struct rlimit limit;
>
> 	/* Check if we could preallocate requested stack */
> -	if ((getrlimit (RLIMIT_STACK, &limit) == 0) &&
> -	    ((_size_stack * 2) < limit.rlim_cur) &&
> -         ((stack_mem = alloca(_size_stack))))
> -	   _touch_memory(stack_mem, _size_stack);
> 	/* FIXME else warn user setting got ignored */
>
> 	if ((temp_malloc_mem = malloc(_size_malloc_tmp)))
> 		_touch_memory(temp_malloc_mem, _size_malloc_tmp);
>
> 	if ((_malloc_mem = malloc(_size_malloc)))
> 		_touch_memory(_malloc_mem, _size_malloc);
>
> 	free(temp_malloc_mem);
> }
>
> My questions are:
> 1. Can we delete the four lines, since it is just PREallocate?
> 2. What is the purpose of _touch_memory()
> 3. why do first malloc temp_malloc_mem but then free it?
>
> Any information is very appreciated.

The memory use in lvm2 has deeper logic.

The idea behind preallocation and touching memory is - to preallocate memory 
pages and enforce their physical allocation for stack & heap - before the 
memory is mlock()-ed into RAM.

Since the suspend may free i.e. swap device - all the code parts which run in 
suspend needs to be locked in RAM.

Now back to powerpc issue - check why it's crashing -  is the rlim_cur  wrong?
Stack size is too big ?

Code will not be removed it needs to be fixed for powerpc.

Zdenek

clvmd coredump issue

[Liuhua Wang]

Hi, Zdenek

Thank you very much for the reply. It really helps.

>>> On 4/24/2014 at 04:08 PM, in message <5358C694.4060807@redhat.com>, Zdenek
Kabelac <zkabelac@redhat.com> wrote: 
> Dne 24.4.2014 08:58, Liuhua Wang napsal(a):
> > Hi, list
> >
> > I have encountered a coredump when start clvmd on ppc64 as followed:
> >
> > #p $pc
> > $1 = (void (*)()) 0x10070c18
> > #bt
> > #0  0x0000000010070c18 in _allocate_memory () at mm/memlock.c:130
> > #1  0x0000000010070d38 in _lock_mem (cmd=<optimized out>) at mm/memlock.c:310
> > #2  _lock_mem_if_needed (cmd=0x10104f40) at mm/memlock.c:387
> > #3  0x0000000010011358 in init_clvm (excl_uuid=<optimized out>) at
> > lvm-functions.c:919
> > #4  0x000000001000f47c in lvm_thread_fn (arg=0xfffcaf4e530) at clvmd.c:2020
> > #5  0x00000fff9df6c67c in .start_thread () from 
> /lib64/power6/libpthread.so.0
> > #6  0x00000fff9de7643c in .__clone () from /lib64/power6/libc.so.6
> >
> >
> > When the followed four lines are commented then the problem is ok.
> > I suppose the problem is probably because of alloca.
> >
> > static void _allocate_memory(void)
> > {
> > 	void *stack_mem, *temp_malloc_mem;
> > 	struct rlimit limit;
> >
> > 	/* Check if we could preallocate requested stack */
> > -	if ((getrlimit (RLIMIT_STACK, &limit) == 0) &&
> > -	    ((_size_stack * 2) < limit.rlim_cur) &&
> > -         ((stack_mem = alloca(_size_stack))))
> > -	   _touch_memory(stack_mem, _size_stack);
> > 	/* FIXME else warn user setting got ignored */
> >
> > 	if ((temp_malloc_mem = malloc(_size_malloc_tmp)))
> > 		_touch_memory(temp_malloc_mem, _size_malloc_tmp);
> >
> > 	if ((_malloc_mem = malloc(_size_malloc)))
> > 		_touch_memory(_malloc_mem, _size_malloc);
> >
> > 	free(temp_malloc_mem);
> > }
> >
> > My questions are:
> > 1. Can we delete the four lines, since it is just PREallocate?
> > 2. What is the purpose of _touch_memory()
> > 3. why do first malloc temp_malloc_mem but then free it?
> >
> > Any information is very appreciated.
> 
> The memory use in lvm2 has deeper logic.
> 
> The idea behind preallocation and touching memory is - to preallocate memory 
> pages and enforce their physical allocation for stack & heap - before the 
> memory is mlock()-ed into RAM.
> 
> Since the suspend may free i.e. swap device - all the code parts which run in 
> 
> suspend needs to be locked in RAM.
> 
> Now back to powerpc issue - check why it's crashing -  is the rlim_cur  wrong?
> Stack size is too big ?
> 
I displayed the values in the coredump as followed:
 #p limit
  $1 = {rlim_cur = 8388608, rlim_max = 18446744073709551615}
 #p _size_stack
  $2 = 65536
 #p stack_mem
  $3 = <optimized out>

It seems that should not fail since we alloca the much smaller space from
a bigger stack. There would be probable that the stack size of the thread
will get much smaller? (I cannot prove it because I don't have ppc64 machine now)

Best regards,
Melody
> Code will not be removed it needs to be fixed for powerpc.
> 
> Zdenek
> 
> 
> --
> lvm-devel mailing list
> lvm-devel at redhat.com 
> https://www.redhat.com/mailman/listinfo/lvm-devel 
> 
>