NUMA API for 2.5.12 (2/4)

PATCH New Matthew Dobson 1 participant 1 message May 2, 2002

Subsystems (6)

Memory management MM core Scheduler exec and binfmt +2 more subsystems

lore

Share & tools

lore

Matthew Dobson

author

2002-05-02 22:53 UTC #0 lore raw

PATCH 1/1 NUMA API for 2.5.12 (2/4)

Ok all,
I'm going to go ahead and assume (hope) that the no response on the last
posting was because the patch was really large. We'll try this again with 4 smaller patches and see what happens.

This patch implements the NUMA API specified at:
http://lse.sourceforge.net/numa/numa_api.html for the 2.5.12 version of the kernel. The API implements such features as binding processes to CPUs (similar to Robert Love's recent patch), binding to memory blocks, setting launch policies for processes, and rudimentary topology features. The patch is currently used via a prctl() interface with a possible /proc interface in the future. I also have a syscall version if that is preferred.

Again, this is a slightly cleaned-up, and (more) bite-sized version of the
previous patch sent out...

Enjoy, and please send me any comments on the patch, or the API it implements!

-Matt
[email protected]

kernel/Makefile +1 -1

diff -Nur linux-2.5.8-vanilla/kernel/Makefile linux-2.5.8-api/kernel/Makefile
@@ -15,7 +15,7 @@
 obj-y     = sched.o dma.o fork.o exec_domain.o panic.o printk.o \
 	    module.o exit.o itimer.o info.o time.o softirq.o resource.o \
 	    sysctl.o capability.o ptrace.o timer.o user.o \
-	    signal.o sys.o kmod.o context.o futex.o platform.o
+	    signal.o sys.o kmod.o context.o futex.o platform.o numa.o
 
 obj-$(CONFIG_UID16) += uid16.o
 obj-$(CONFIG_MODULES) += ksyms.o

kernel/fork.c +14 -0

diff -Nur linux-2.5.8-vanilla/kernel/fork.c linux-2.5.8-api/kernel/fork.c
@@ -707,6 +707,20 @@
 		spin_lock_init(&p->sigmask_lock);
 	}
 #endif
+	if (!null_restrict(&p->numa_launch_policy)){
+		p->numa_binding = p->numa_launch_policy;
+		p->cpus_allowed = p->numa_binding.cpus.list & p->numa_restrict.cpus.list;
+		if (!(p->cpus_allowed & cpu_online_map))
+			BUG();
+		if (p->cpus_allowed & (1UL << smp_processor_id()))
+			p->thread_info->cpu = smp_processor_id();
+		else
+			p->thread_info->cpu = __ffs(p->cpus_allowed & cpu_online_map);
+	} else
+		p->thread_info->cpu = smp_processor_id();
+	numa_set_init(&p->numa_launch_policy);
+	rwlock_init(&p->numa_api_lock);
+
 	p->array = NULL;
 	p->lock_depth = -1;		/* -1 = no lock */
 	p->start_time = jiffies;

kernel/numa.c +378 -0

378 lines not on this page · raw message

mm/numa.c +76 ~~-13~~

137 lines not on this page · raw message

mm/page_alloc.c +7 -0

diff -Nur linux-2.5.8-vanilla/mm/page_alloc.c linux-2.5.8-api/mm/page_alloc.c
@@ -39,6 +39,9 @@
 static int zone_balance_min[MAX_NR_ZONES] __initdata = { 20 , 20, 20, };
 static int zone_balance_max[MAX_NR_ZONES] __initdata = { 255 , 255, 255, };
 
+extern int nummemblks;
+extern unsigned long memblk_online_map;
+
 /*
  * Free_page() adds the page to the free lists. This is optimized for
  * fast normal cases (no error jumps taken normally).

@@ -936,6 +939,10 @@
 void __init free_area_init(unsigned long *zones_size)
 {
 	free_area_init_core(0, &contig_page_data, &mem_map, zones_size, 0, 0, 0);
+	contig_page_data.node_id = 0;
+	contig_page_data.memblk_id = 0;
+	nummemblks = 1;
+	memblk_online_map = 1UL;
 }
 
 static int __init setup_mem_frac(char *str)