fix memory less node allocation

Tony:

The original memory less node allocation attempted to use NODEDATA_ALIGN for alignment. The bootmem allocator only allows a power of two alignments. This causes a BUG_ON for some nodes. For cpu only nodes just allocate with a PERCPU_PAGE_SIZE alignment.

Some older firmware reports SLIT distances of 0xff and results in bestnode not being computed. This is now treated correctly.

The failed allocation check was removed because it's redundant. The bootmem allocator already makes this check.

This fix has been boot tested on 4 node machine which has 4 cpu only nodes and 1 memory node. Thanks to Pete Keilty for reporting this and helping me test it.

thanks,

bob

 arch/ia64/mm/discontig.c |   19 +++++++++----------
 1 files changed, 9 insertions(+), 10 deletions(-)

Index: linux-2.6.14-rc3/arch/ia64/mm/discontig.c
=================================--- linux-2.6.14-rc3.orig/arch/ia64/mm/discontig.c 2005-10-26 17:44:36.000000000 -0400
+++ linux-2.6.14-rc3/arch/ia64/mm/discontig.c 2005-10-27 16:44:29.000000000 -0400
@@ -350,14 +350,12 @@ static void __init initialize_pernode_da
* for best.
* @nid: node id
* @pernodesize: size of this node's pernode data
- * @align: alignment to use for this node's pernode data
*/
-static void __init *memory_less_node_alloc(int nid, unsigned long pernodesize,
- unsigned long align)
+static void __init *memory_less_node_alloc(int nid, unsigned long pernodesize)
{
void *ptr = NULL;
u8 best = 0xff;
- int bestnode = -1, node;
+ int bestnode = -1, node, anynode = 0;

for_each_online_node(node) {
if (node_isset(node, memory_less_mask))
@@ -366,13 +364,15 @@ static void __init *memory_less_node_all
best = node_distance(nid, node); bestnode = node;
}
+ anynode = node;
}

- ptr = __alloc_bootmem_node(mem_data[bestnode].pgdat,
- pernodesize, align, __pa(MAX_DMA_ADDRESS));
+ if (bestnode = -1)
+ bestnode = anynode;
+
+ ptr = __alloc_bootmem_node(mem_data[bestnode].pgdat, pernodesize,
+ PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS));

- if (!ptr)
- panic("NO memory for memory less node\n");
return ptr;
}

@@ -413,8 +413,7 @@ static void __init memory_less_nodes(voi

for_each_node_mask(node, memory_less_mask) {
pernodesize = compute_pernodesize(node);
- pernode = memory_less_node_alloc(node, pernodesize,
- (node) ? (node * PERCPU_PAGE_SIZE) : (1024*1024));
+ pernode = memory_less_node_alloc(node, pernodesize);
fill_pernode(node, __pa(pernode), pernodesize);
}

- if (!ptr)
- panic("NO memory for memory less node\n");

Why did you drop this panic? Although your patch now makes sure to never try to alloc from a memory less node ... surely it is still possible (though unlikely) that the call to __alloc_bootmem_node() might fail. A panic here might be easier to debug than letting it run until fill_pernode() does a memset(__va(0), 0, ...).

-Tony

luck wrote: [Thu Nov 03 2005, 05:33:53PM EST]

I dropped it because it seemed redundant. __alloc_bootmem_node calls __alloc_bootmem_node_limit who calls __alloc_bootmem_core for this node. Should __alloc_bootmem_core core fail, then call __alloc_bootmem_limit. __alloc_bootmem_limit attempts to allocate from any node. Should __alloc_bootmem_limit (old name __alloc_bootmem) fail, then it panics.

bob