linux-stable/arch/s390/numa/numa.c
Michael Holzheu c29a7baf09 s390/numa: add emulation support
NUMA emulation (aka fake NUMA) distributes the available memory to nodes
without using real topology information about the physical memory of the
machine.

Splitting the system memory into nodes replicates the memory management
structures for each node. Particularly each node has its own "mm locks"
and its own "kswapd" task.

For large systems, under certain conditions, this results in improved
system performance and/or latency based on reduced pressure on the mm
locks and the kswapd tasks.

NUMA emulation distributes CPUs to nodes while respecting the original
machine topology information. This is done by trying to avoid to separate
CPUs which reside on the same book or even on the same MC. Because the
current Linux scheduler code requires a stable cpu to node mapping, cores
are pinned to nodes when the first CPU thread is set online.

This patch is based on the initial implementation from Philipp Hachtmann.

Signed-off-by: Michael Holzheu <holzheu@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
2015-08-04 14:06:53 +02:00

184 lines
4.1 KiB
C

/*
* NUMA support for s390
*
* Implement NUMA core code.
*
* Copyright IBM Corp. 2015
*/
#define KMSG_COMPONENT "numa"
#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
#include <linux/kernel.h>
#include <linux/mmzone.h>
#include <linux/cpumask.h>
#include <linux/bootmem.h>
#include <linux/memblock.h>
#include <linux/slab.h>
#include <linux/node.h>
#include <asm/numa.h>
#include "numa_mode.h"
pg_data_t *node_data[MAX_NUMNODES];
EXPORT_SYMBOL(node_data);
cpumask_var_t node_to_cpumask_map[MAX_NUMNODES];
EXPORT_SYMBOL(node_to_cpumask_map);
const struct numa_mode numa_mode_plain = {
.name = "plain",
};
static const struct numa_mode *mode = &numa_mode_plain;
int numa_pfn_to_nid(unsigned long pfn)
{
return mode->__pfn_to_nid ? mode->__pfn_to_nid(pfn) : 0;
}
void numa_update_cpu_topology(void)
{
if (mode->update_cpu_topology)
mode->update_cpu_topology();
}
int __node_distance(int a, int b)
{
return mode->distance ? mode->distance(a, b) : 0;
}
int numa_debug_enabled;
/*
* alloc_node_data() - Allocate node data
*/
static __init pg_data_t *alloc_node_data(void)
{
pg_data_t *res;
res = (pg_data_t *) memblock_alloc(sizeof(pg_data_t), 1);
if (!res)
panic("Could not allocate memory for node data!\n");
memset(res, 0, sizeof(pg_data_t));
return res;
}
/*
* numa_setup_memory() - Assign bootmem to nodes
*
* The memory is first added to memblock without any respect to nodes.
* This is fixed before remaining memblock memory is handed over to the
* buddy allocator.
* An important side effect is that large bootmem allocations might easily
* cross node boundaries, which can be needed for large allocations with
* smaller memory stripes in each node (i.e. when using NUMA emulation).
*
* Memory defines nodes:
* Therefore this routine also sets the nodes online with memory.
*/
static void __init numa_setup_memory(void)
{
unsigned long cur_base, align, end_of_dram;
int nid = 0;
end_of_dram = memblock_end_of_DRAM();
align = mode->align ? mode->align() : ULONG_MAX;
/*
* Step through all available memory and assign it to the nodes
* indicated by the mode implementation.
* All nodes which are seen here will be set online.
*/
cur_base = 0;
do {
nid = numa_pfn_to_nid(PFN_DOWN(cur_base));
node_set_online(nid);
memblock_set_node(cur_base, align, &memblock.memory, nid);
cur_base += align;
} while (cur_base < end_of_dram);
/* Allocate and fill out node_data */
for (nid = 0; nid < MAX_NUMNODES; nid++)
NODE_DATA(nid) = alloc_node_data();
for_each_online_node(nid) {
unsigned long start_pfn, end_pfn;
unsigned long t_start, t_end;
int i;
start_pfn = ULONG_MAX;
end_pfn = 0;
for_each_mem_pfn_range(i, nid, &t_start, &t_end, NULL) {
if (t_start < start_pfn)
start_pfn = t_start;
if (t_end > end_pfn)
end_pfn = t_end;
}
NODE_DATA(nid)->node_spanned_pages = end_pfn - start_pfn;
NODE_DATA(nid)->node_id = nid;
}
}
/*
* numa_setup() - Earliest initialization
*
* Assign the mode and call the mode's setup routine.
*/
void __init numa_setup(void)
{
pr_info("NUMA mode: %s\n", mode->name);
if (mode->setup)
mode->setup();
numa_setup_memory();
memblock_dump_all();
}
/*
* numa_init_early() - Initialization initcall
*
* This runs when only one CPU is online and before the first
* topology update is called for by the scheduler.
*/
static int __init numa_init_early(void)
{
/* Attach all possible CPUs to node 0 for now. */
cpumask_copy(node_to_cpumask_map[0], cpu_possible_mask);
return 0;
}
early_initcall(numa_init_early);
/*
* numa_init_late() - Initialization initcall
*
* Register NUMA nodes.
*/
static int __init numa_init_late(void)
{
int nid;
for_each_online_node(nid)
register_one_node(nid);
return 0;
}
device_initcall(numa_init_late);
static int __init parse_debug(char *parm)
{
numa_debug_enabled = 1;
return 0;
}
early_param("numa_debug", parse_debug);
static int __init parse_numa(char *parm)
{
if (strcmp(parm, numa_mode_plain.name) == 0)
mode = &numa_mode_plain;
#ifdef CONFIG_NUMA_EMU
if (strcmp(parm, numa_mode_emu.name) == 0)
mode = &numa_mode_emu;
#endif
return 0;
}
early_param("numa", parse_numa);