Re: Slab panic on 2.6.19-rc3-git5 (-git4 was OK)

[Giridhar Pemmasani <giri@lmc.cs.sunysb.edu>]

Nick Piggin wrote:

> Andrew Morton wrote:
>> --- a/mm/vmalloc.c~__vmalloc_area_node-fix
>> +++ a/mm/vmalloc.c
>> @@ -428,7 +428,8 @@ void *__vmalloc_area_node(struct vm_stru
>>  area->nr_pages = nr_pages;
>>  /* Please note that the recursion is strictly bounded. */
>>  if (array_size > PAGE_SIZE) {
>> -            pages = __vmalloc_node(array_size, gfp_mask, PAGE_KERNEL, node);
>> +            pages = __vmalloc_node(array_size, gfp_mask & ~__GFP_HIGHMEM,
>> +                                    PAGE_KERNEL, node);
>>  area->flags |= VM_VPAGES;
>>  } else {
>>  pages = kmalloc_node(array_size,
> 
> Don't you actually *want* the page array to be allocated from highmem? So
> the gfp mask here should be just for whether we're allowed to sleep /
> reclaim (ie gfp_mask & ~(__GFP_DMA|__GFP_DMA32) | (__GFP_HIGHMEM))?
> 
> Slab allocations should be (gfp_mask &
> ~(__GFP_DMA|__GFP_DMA32|__GFP_HIGHMEM)), which you could mask in
> __get_vm_area_node
> 

Since gfp_mask there would also have GFP_ZERO, we need to mask off that too.
How about my earlier suggestion of masking off flags in __get_vm_area_node
with GFP_LEVEL_MASK?

Giri

PS: I am not sure if this mail gets to all recipients in the original
thread - I am not subscribed to lkml and I haven't found a way to reply to
all people and the group.