On (31/08/06 20:08), Keith Mannthey didst pronounce:

quoted

So, do you actally expect a lot of unused mem_map to be allocated with
struct pages that are inactive until memory is hot-added in an
x86_64-specific manner? The arch-independent stuff currently will not do
that. It sets up memmap for where memory really exists. If that is not
what you expect, it will hit issues at hotadd time which is not the
current issue but one that can be fixed.

Yes. RESERVED based is a big waste of mem_map space.  The add areas
are marked as RESERVED during boot and then later onlined during add.
It might be ok.  I will play with tomorrow.  I might just need to
call add_active_range in the right spot :)

Following this mail should be two patches that may address the problem
with reserved memory hot-add. One assumption made by arch-independent
zone-sizing was that the only memory holes of interest were those before the
end of physical memory.  Another assumption was that mem_map should only be
allocated for memory that was physically present in the machine.

With MEMORY_HOTPLUG_RESERVE on x86_64, these assumptions do not hold. This
feature expects that mem_map is allocated at boot time and later activated
on a memory hot-add event. To determine if the region is usable for hot-add
in the future, holes are calculated beyond the end of physical memory.

The following two patches fix these two assumptions. They have been boot-tested
on a range of hardware (x86, ppc64, ia64 and x86_64) so there should be no
new regressions.

I don't have access to hardware that can use MEMORY_HOTPLUG_RESERVE so I'd
appreciate hearing if the patches work. I wrote a test program that simulated
the input from the machine the problem was reported on.  It registers active
memory and simulates the check made by reserve_hotadd(). push_node_boundaries()
is called to push the end of the node out by 100 pages like what SRAT would
do for reserve hot-add and it appears to do the right thing. Output is below.

mel@arnold:~/patches/brokenout/zonesizing/driver_test$ gcc driver_test.c -o
driver_test && ./driver_test | grep -v "active with" | grep -v
account_node_boundary
Stage 1: Registering active ranges
Entering add_active_range(0, 0, 152) 0 entries of 96 used
Entering add_active_range(0, 256, 524165) 1 entries of 96 used
Entering add_active_range(0, 1048576, 4653056) 2 entries of 96 used
Entering add_active_range(1, 17235968, 18219008) 3 entries of 96 used

Dumping active map
0: 0  0 -> 152
1: 0  256 -> 524165
2: 0  1048576 -> 4653056
3: 1  17235968 -> 18219008
Entering push_node_boundaries(0, 0, 4653156)

Checking reserve-hotadd
  absent_pages_in_range(4653056, 17235968) == 17235968 - 4653056 == 12582912
  absent_pages_in_range(18219008, 52428800) == 52428800 - 18219008 == 34209792

Stage 2: Calculating zone sizes and holes

Stage 3: Dumping zone sizes and holes
zone_size[0][0] =     4096 zone_holes[0][0] =      104
zone_size[0][1] =  1044480 zone_holes[0][1] =   524411
zone_size[0][2] =  3604580 zone_holes[0][2] =      100
zone_size[1][2] =   983040 zone_holes[1][2] =        0

Stage 4: Printing present pages
On node 0, 4128541 pages
 zone 0 present_pages = 3992
 zone 1 present_pages = 520069
 zone 2 present_pages = 3604480
On node 1, 983040 pages
 zone 2 present_pages = 983040

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab