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>

arch/s390/Kconfig

@@ -423,6 +423,43 @@ config NODES_SHIFT
 	  Specify the maximum number of NUMA nodes available on the target
 	  system. Increases memory reserved to accommodate various tables.
 
+menu "Select NUMA modes"
+	depends on NUMA
+
+config NUMA_EMU
+	bool "NUMA emulation"
+	default y
+	help
+	  Numa emulation mode will split the available system memory into
+	  equal chunks which then are distributed over the configured number
+	  of nodes in a round-robin manner.
+
+	  The number of fake nodes is limited by the number of available memory
+	  chunks (i.e. memory size / fake size) and the number of supported
+	  nodes in the kernel.
+
+	  The CPUs are assigned to the nodes in a way that partially respects
+	  the original machine topology (if supported by the machine).
+	  Fair distribution of the CPUs is not guaranteed.
+
+config EMU_SIZE
+	hex "NUMA emulation memory chunk size"
+	default 0x10000000
+	range 0x400000 0x100000000
+	depends on NUMA_EMU
+	help
+	  Select the default size by which the memory is chopped and then
+	  assigned to emulated NUMA nodes.
+
+	  This can be overridden by specifying
+
+	  emu_size=<n>
+
+	  on the kernel command line where also suffixes K, M, G, and T are
+	  supported.
+
+endmenu
+
 config SCHED_MC
 	def_bool n
 

arch/s390/include/asm/numa.h

@@ -26,6 +26,10 @@ extern int numa_debug_enabled;
 
 static inline void numa_setup(void) { }
 static inline void numa_update_cpu_topology(void) { }
+static inline int numa_pfn_to_nid(unsigned long pfn)
+{
+	return 0;
+}
 
 #endif /* CONFIG_NUMA */
 #endif /* _ASM_S390_NUMA_H */

arch/s390/numa/Makefile

 obj-y			+= numa.o
 obj-y			+= toptree.o
+obj-$(CONFIG_NUMA_EMU)	+= mode_emu.o

arch/s390/numa/numa.c

@@ -175,6 +175,10 @@ 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);

arch/s390/numa/numa_mode.h

@@ -19,5 +19,6 @@ struct numa_mode {
 };
 
 extern const struct numa_mode numa_mode_plain;
+extern const struct numa_mode numa_mode_emu;
 
 #endif /* __S390_NUMA_MODE_H */

drivers/s390/char/sclp_cmd.c

@@ -25,6 +25,7 @@
 #include <asm/setup.h>
 #include <asm/page.h>
 #include <asm/sclp.h>
+#include <asm/numa.h>
 
 #include "sclp.h"
 
@@ -388,11 +389,11 @@ static struct notifier_block sclp_mem_nb = {
 };
 
 static void __init align_to_block_size(unsigned long long *start,
-				       unsigned long long *size)
+				       unsigned long long *size,
+				       unsigned long long alignment)
 {
-	unsigned long long start_align, size_align, alignment;
+	unsigned long long start_align, size_align;
 
-	alignment = memory_block_size_bytes();
 	start_align = roundup(*start, alignment);
 	size_align = rounddown(*start + *size, alignment) - start_align;
 
@@ -404,8 +405,8 @@ static void __init align_to_block_size(unsigned long long *start,
 
 static void __init add_memory_merged(u16 rn)
 {
+	unsigned long long start, size, addr, block_size;
 	static u16 first_rn, num;
-	unsigned long long start, size;
 
 	if (rn && first_rn && (first_rn + num == rn)) {
 		num++;
@@ -423,9 +424,12 @@ static void __init add_memory_merged(u16 rn)
 		goto skip_add;
 	if (memory_end_set && (start + size > memory_end))
 		size = memory_end - start;
-	align_to_block_size(&start, &size);
-	if (size)
-		add_memory(0, start, size);
+	block_size = memory_block_size_bytes();
+	align_to_block_size(&start, &size, block_size);
+	if (!size)
+		goto skip_add;
+	for (addr = start; addr < start + size; addr += block_size)
+		add_memory(numa_pfn_to_nid(PFN_DOWN(addr)), addr, block_size);
 skip_add:
 	first_rn = rn;
 	num = 1;