[Crash-utility] "zero-size memory allocation!" is back for Linux 3.1

[Dave Anderson]

----- Original Message -----
> In include/linux/slab_def.h circa linux 3.0, this def for field
> nodelists:
> 
> struct kmem_cache {
> /* 1) per-cpu data, touched during every alloc/free */
>         struct array_cache *array[NR_CPUS];
> 
>         ...
> 
>         struct kmem_list3 *nodelists[MAX_NUMNODES];
>         /*
>          * Do not add fields after nodelists[]
>          */
> };
> 
> Became this in 3.1:
> 
> struct kmem_cache {
> ...
> 
> /* 6) per-cpu/per-node data, touched during every alloc/free */
>       /*
>        * We put array[] at the end of kmem_cache, because we want to size
>        * this array to nr_cpu_ids slots instead of NR_CPUS
>        * (see kmem_cache_init())
>        * We still use [NR_CPUS] and not [1] or [0] because
>        cache_cache
>        * is statically defined, so we reserve the max number of cpus.
>        */
>       struct kmem_list3 **nodelists;
>       struct array_cache *array[NR_CPUS];
>       /*
>        * Do not add fields after array[]
>        */
> };
> 
> 
> Which causes this in crash/memory.c:vm_init()
> 
>        ARRAY_LENGTH_INIT(vt->kmem_cache_len_nodes, NULL,
>             "kmem_cache.nodelists", NULL, 0);
> 
> to set vt->kmem_cache_len_nodes to 0, and leads to the initialization
> failure when max_cpudata_limit calls getbuf with a size of 0.
> 
> Got a fix in the works yet?
> 
> Thanks,
> Bob Montgomery

No, afraid not.  Fedora uses slub instead of slab, so I haven't 
noticed it.  I wonder why kmem_cache_downsize() doesn't recalculate
vt->kmem_cache_len_nodes based upon "nr_node_ids"?:

        if (buffer_size < SIZE(kmem_cache_s)) {

                if (kernel_symbol_exists("nr_node_ids")) {
                        get_symbol_data("nr_node_ids", sizeof(int),
                                &nr_node_ids);
                        vt->kmem_cache_len_nodes = nr_node_ids;

                } else
                        vt->kmem_cache_len_nodes = 1;

Dave