Re: [patch] mm: node-setup agnostic free_bootmem()

[Johannes Weiner <hannes@saeurebad.de>]

Hi,

"Yinghai Lu" <yhlu.kernel@gmail.com> writes:

> On Wed, Apr 30, 2008 at 3:50 AM, Johannes Weiner <hannes@saeurebad.de> wrote:
>>
>> Hi,
>>
>>  "Yinghai Lu" <yhlu.kernel@gmail.com> writes:
>>
>>  > On Mon, Apr 28, 2008 at 12:11 PM, Yinghai Lu <yhlu.kernel@gmail.com> wrote:
>>  >>
>>  >> On Mon, Apr 28, 2008 at 9:54 AM, Johannes Weiner <hannes@saeurebad.de> wrote:
>>  >>  > Hi Yinghai,
>>  >>  >
>>  >>  >
>>  >>  >
>>  >>  >  "Yinghai Lu" <yhlu.kernel@gmail.com> writes:
>>  >>  >
>>  >>  >  > On Sun, Apr 27, 2008 at 5:40 PM, Ingo Molnar <mingo@elte.hu> wrote:
>>  >>  >  >>
>>  >>  >  >>  * Johannes Weiner <hannes@saeurebad.de> wrote:
>>  >>  >  >>
>>  >>  >  >>  > > so i very much agree that your changes are cleaner, i just wanted to
>>  >>  >  >>  > > have one that has all the fixes included.
>>  >>  >  >>  >
>>  >>  >  >>  > I had planned this to be another patch because there are more then one
>>  >>  >  >>  > boundary check I wanted to tighten.  I can merge them though if you
>>  >>  >  >>  > like.
>>  >>  >  >>
>>  >>  >  >>  no, better to have them in separate patches.
>>  >>  >  >>
>>  >>  >  >>  > > Would you like to post a patch against current -git or should i
>>  >>  >  >>  > > extract the cleaner reserve_bootmem() from your previous patch?
>>  >>  >  >>  >
>>  >>  >  >>  > I just moved and have only sporadic internet access and free time
>>  >>  >  >>  > slots available.  Would be nice if you could do it!
>>  >>  >  >>
>>  >>  >  >>  sure, find the merged patch below, against latest -git, boot-tested on
>>  >>  >  >>  x86. Is this what you had in mind?
>>  >>  >  >>
>>  >>  >  >>         Ingo
>>  >>  >  >>
>>  >>  >  >>  ---------------->
>>  >>  >  >>  Subject: mm: node-setup agnostic free_bootmem()
>>  >>  >  >>  From: Johannes Weiner <hannes@saeurebad.de>
>>  >>  >  >>  Date: Wed, 16 Apr 2008 13:36:31 +0200
>>  >>  >  >>
>>  >>  >  >>  Make free_bootmem() look up the node holding the specified address
>>  >>  >  >>  range which lets it work transparently on single-node and multi-node
>>  >>  >  >>  configurations.
>>  >>  >  >>
>>  >>  >  >>  If the address range exceeds the node range, it well be marked free
>>  >>  >  >>  across node boundaries, too.
>>  >>  >  >>
>>  >>  >  >>  Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
>>  >>  >  >>  CC: Andi Kleen <andi@firstfloor.org>
>>  >>  >  >>  CC: Yinghai Lu <yhlu.kernel@gmail.com>
>>  >>  >  >>  CC: Yasunori Goto <y-goto@jp.fujitsu.com>
>>  >>  >  >>  CC: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
>>  >>  >  >>  CC: Christoph Lameter <clameter@sgi.com>
>>  >>  >  >>  CC: Andrew Morton <akpm@linux-foundation.org>
>>  >>  >  >>  Signed-off-by: Ingo Molnar <mingo@elte.hu>
>>  >>  >  >>  ---
>>  >>  >  >>   mm/bootmem.c |   27 +++++++++++++++++++++++++--
>>  >>  >  >>   1 file changed, 25 insertions(+), 2 deletions(-)
>>  >>  >  >>
>>  >>  >  >>  Index: linux-x86.q/mm/bootmem.c
>>  >>  >  >>  ===================================================================
>>  >>  >  >>  --- linux-x86.q.orig/mm/bootmem.c
>>  >>  >  >>  +++ linux-x86.q/mm/bootmem.c
>>  >>  >  >>  @@ -493,8 +493,31 @@ int __init reserve_bootmem(unsigned long
>>  >>  >  >>   void __init free_bootmem(unsigned long addr, unsigned long size)
>>  >>  >  >>   {
>>  >>  >  >>         bootmem_data_t *bdata;
>>  >>  >  >>  -       list_for_each_entry(bdata, &bdata_list, list)
>>  >>  >  >>  -               free_bootmem_core(bdata, addr, size);
>>  >>  >  >>  +       unsigned long pos = addr;
>>  >>  >  >>  +       unsigned long partsize = size;
>>  >>  >  >>  +
>>  >>  >  >>  +       list_for_each_entry(bdata, &bdata_list, list) {
>>  >>  >  >>  +               unsigned long remainder = 0;
>>  >>  >  >>  +
>>  >>  >  >>  +               if (pos < bdata->node_boot_start)
>>  >>  >  >>  +                       continue;
>>  >>  >  >>  +
>>  >>  >  >>  +               if (PFN_DOWN(pos + partsize) > bdata->node_low_pfn) {
>>  >>  >  >>  +                       remainder = PFN_DOWN(pos + partsize) - bdata->node_low_pfn;
>>  >>  >  >>  +                       partsize -= remainder;
>>  >>  >  >>  +               }
>>  >>  >  >>  +
>>  >>  >  >>  +               free_bootmem_core(bdata, pos, partsize);
>>  >>  >  >>  +
>>  >>  >  >>  +               if (!remainder)
>>  >>  >  >>  +                       return;
>>  >>  >  >>  +
>>  >>  >  >>  +               pos = PFN_PHYS(bdata->node_low_pfn + 1);
>>  >>  >  >>  +       }
>>  >>  >  >>  +       printk(KERN_ERR "free_bootmem: request: addr=%lx, size=%lx, "
>>  >>  >  >>  +                       "state: pos=%lx, partsize=%lx\n", addr, size,
>>  >>  >  >>  +                       pos, partsize);
>>  >>  >  >>  +       BUG();
>>  >>  >  >>   }
>>  >>  >  >>
>>  >>  >  >>   unsigned long __init free_all_bootmem(void)
>>  >>  >  >>
>>  >>  >  >
>>  >>  >  > it will not work with cross nodes.
>>  >>  >  >
>>  >>  >  > for example: node 0: 0-2g, 4-6g, node1: 2-4g, 6-8g.
>>  >>  >  > and if ramdisk sit cross 2G boundary. you will only free the range
>>  >>  >  > before 2g.
>>  >>  >
>>  >>  >  Yes, you stated that several times but this is not a technical argument:
>>  >>  >  These setups are afaik not yet supported by the kernel at all.  And you
>>  >>  >  could not explain the node layout with the patch that implements support
>>  >>  >  for these configurations.
>>  >>
>>  >>  I looked at Suresh's patch, and it still only has one bdata for one node.
>>  >
>>  > Suresh's patch already in the Linus tree.
>>  > commit 6ec6e0d9f2fd7cb6ca6bc3bfab5ae7b5cdd8c36f
>>  > Author: Suresh Siddha <suresh.b.siddha@intel.com>
>>  > Date:   Tue Mar 25 10:14:35 2008 -0700
>>  >
>>  >     srat, x86: add support for nodes spanning other nodes
>>  >
>>  >     For example, If the physical address layout on a two node system with 8 GB
>>  >     memory is something like:
>>  >     node 0: 0-2GB, 4-6GB
>>  >     node 1: 2-4GB, 6-8GB
>>  >
>>  >     Current kernels fail to boot/detect this NUMA topology.
>>  >
>>  >     ACPI SRAT tables can expose such a topology which needs to be supported.
>>  >
>>  >     Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
>>  >     Signed-off-by: Ingo Molnar <mingo@elte.hu>
>>  >     Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
>>
>>  Okay, so we have one bdata for node 0 and one for node 1. Does that mean
>>  that both have overlapping pfn ranges?
>>
>>  [1    |||||     ]
>>      [2    |||||     ]
>>
>>  Like this?  How are the ||||| represented in the bootmem maps of each bdata?
>
> Yes.

Okay.  So they share the same PFNs.  Now imagine the following scenario:

node0: 0-2GB, 4-6GB
node1: 2-4GB, 6-8GB

/* Marks the range on node0 and node1 */
free_bootmem(1.5G, 2G);

/* Frees all bootmem on both nodes */
free_all_bootmem_node(NODE_DATA(0));
free_all_bootmem_node(NODE_DATA(1));

Aren't the same page descriptors send to __free_bootmem_pages() twice?

	Hannes