[Bug runtime/3911] Compilation of systemtap causes the system to crash on p570 system.

["srinivasa at in dot ibm dot com" <sourceware-bugzilla@sourceware.org>]

------- Additional Comments From srinivasa at in dot ibm dot com  2007-01-30 07:17 -------
Here is my analysis of this bug
======================================
 1) Looking at the backtrace, _stp_map_init() calls kmalloc_node() with cpu as
argument. kmalloc_node() is same as kmalloc if NUMA is not configured and
kmalloc_node() calls kmem_cache_alloc_node() if NUMA is configured.

_stp_map_init is called by _stp_pmap_new() within for_each_cpu() brace.

 
=============================================
static int _stp_map_init(MAP m, unsigned max_entries, int type, int key_size,
int data_si
ze, int cpu)
{
        int size;

.....................................
.....................................
for (i = 0; i < max_entries; i++) {
                        if (cpu < 0)
                                tmp = kmalloc(size, STP_ALLOC_FLAGS);
                        else
                                tmp = kmalloc_node(size, STP_ALLOC_FLAGS, cpu);

                        if (!tmp)
                                return -1;;

                        dbug ("allocated %lx\n", (long)tmp);
=========================================================================
static PMAP _stp_pmap_new(unsigned max_entries, int type, int key_size, int
data_size)
{
        int i;
        MAP map, m;

        PMAP pmap = (PMAP) kmalloc(sizeof(struct pmap), STP_ALLOC_FLAGS);
        if (pmap == NULL)
                return NULL;
.........................
.........................
 for_each_cpu(i) {
                m = per_cpu_ptr (map, i);
                if (_stp_map_init(m, max_entries, type, key_size, data_size, i)) {
                        goto err1;
                }
        }
============================================

2) Since in my system, NUMA is configured kmalloc_node() calls
kmem_cache_alloc_node with "cpu" as nodeid
====================================================
#ifdef CONFIG_NUMA
extern void *__kmalloc_node(size_t size, gfp_t flags, int node);

static inline void *kmalloc_node(size_t size, gfp_t flags, int node)
{
        if (__builtin_constant_p(size)) {
                int i = 0;
#define CACHE(x) \
.......................
..............................
return kmem_cache_alloc_node((flags & GFP_DMA) ?
                        malloc_sizes[i].cs_dmacachep :
                        malloc_sizes[i].cs_cachep, flags, node);
        }
========================================================
3) This means systemtap code expects the number of nodes in numa should be same
as number of cpu's. 
  kmem_cache_alloc_node() inturn calls ___cache_alloc_node where
cachep->nodelists[nodeid] gives wrong address because in my system number of
nodes are less than number of cpus.
================================
Mount-cache hash table entries: 4096
Processor 1 found.
Processor 2 found.
Processor 3 found.
Processor 4 found.
Processor 5 found.
Processor 6 found.
Processor 7 found.
Brought up 8 CPUs
Node 0 CPUs: 0-7
Node 1 CPUs:
Node 2 CPUs:
Node 3 CPUs:
sizeof(vma)=176 bytes
sizeof(page)=56 bytes
sizeof(inode)=560 bytes
=================================

for example: I have 8 cpu's in my system and  4 numa nodes. nodelists[8] gives
wrong address and that is causing the oops.
============================================================
static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags,
                                int nodeid)
{
        struct list_head *entry;
        struct slab *slabp;
        struct kmem_list3 *l3;
        void *obj;
        int x;

        l3 = cachep->nodelists[nodeid];  <<<========PC is here
        BUG_ON(!l3);
============================================

Hence assuming, number of nodes equal to number of cpus in systemtap modules is
causing this bug.


Martin 
 Any ideas??

Thanks
 Srinivasa DS

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hunt at redhat dot com


http://sourceware.org/bugzilla/show_bug.cgi?id=3911

------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.