Messages in this thread | | | Date | Thu, 16 Oct 2008 23:18:24 +0200 | From | Brice Goglin <> | Subject | Re: [PATCH 4/5] mm: rework do_pages_move() to work on page_sized chunks |
| |
Christoph Lameter wrote: >> + err = -ENOMEM; >> + pm = kmalloc(PAGE_SIZE, GFP_KERNEL); >> + if (!pm) >> > > ok.... But if you need a page sized chunk then you can also do > get_zeroed_page(GFP_KERNEL). Why bother the slab allocator for page sized > allocations? >
Right. But why get_zeroed_page()? I don't think I need anything zeroed (and I needed so, I would have to zero again between each chunk).
alloc_pages(order=0)+__free_pages() is probably good.
>> + /* fill the chunk pm with addrs and nodes from user-space */ >> + for (j = 0; j < chunk_nr_pages; j++) { >> > > j? So the chunk_start used to be i? >
The original "i" is somehow "chunk_start+j" now.
Thanks Christoph, I'll send an updated "4/5" patch in the next days.
Brice
| |