Messages in this thread | | | Subject | Re: 2.6.28-rc9 panics with crashkernel=256M while booting | From | Dave Hansen <> | Date | Wed, 21 Jan 2009 16:29:39 -0800 |
| |
On Mon, 2009-01-19 at 17:00 +0530, Chandru wrote: > --- linux-2.6.29-rc2/arch/powerpc/mm/numa.c.orig 2009-01-19 16:14:49.000000000 +0530 > +++ linux-2.6.29-rc2/arch/powerpc/mm/numa.c 2009-01-19 16:36:38.000000000 +0530 > @@ -901,7 +901,8 @@ static void mark_reserved_regions_for_ni > get_node_active_region(start_pfn, &node_ar); > while (start_pfn < end_pfn && > node_ar.start_pfn < node_ar.end_pfn) { > - unsigned long reserve_size = size; > + unsigned long reserve_size = (size >> PAGE_SHIFT) << > + PAGE_SHIFT; > /* > * if reserved region extends past active region > * then trim size to active region > @@ -917,7 +918,8 @@ static void mark_reserved_regions_for_ni > dbg("reserve_bootmem %lx %lx nid=%d\n", > physbase, reserve_size, node_ar.nid); > reserve_bootmem_node(NODE_DATA(node_ar.nid), > - physbase, reserve_size, > + (start_pfn << PAGE_SHIFT), > + reserve_size, > BOOTMEM_DEFAULT); > } > /*
Chandru, I don't mean to keep ragging on your patches, but I really don't think this is right, yet.
Let's take, for instance, a 1-byte reservation. With this code, you've suddenly turned that into a 0-byte reservation, and that *can't* be right. The same thing happens if you have a reservation that spans two pages. If you unconditionally round it down, then you might miss the part that spans a portion of the second page.
It needs to be rounded down like you are suggesting here, but only in the case where we've gone over the *CURRENT* node's boundary. This is kinda what that "if (end_pfn > node_ar.end_pfn)" check is doing. But, it evidently screws it up if the overlap isn't by an entire page or something.
Please also, for pete's sake, use masks (a la PAGE_MASK) or macros if you're going to page-align something. Don't shift down and up like that.
-- Dave
| |