x86-64, NUMA: Introduce struct numa_meminfo
authorTejun Heo <tj@kernel.org>
Wed, 16 Feb 2011 16:11:08 +0000 (17:11 +0100)
committerTejun Heo <tj@kernel.org>
Wed, 16 Feb 2011 16:11:08 +0000 (17:11 +0100)
Arrays for memblks and nodeids and their length lived in separate
variables making things unnecessarily cumbersome.  Introduce struct
numa_meminfo which contains all memory configuration info.  This patch
doesn't cause any behavior change.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: Shaohui Zheng <shaohui.zheng@intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: H. Peter Anvin <hpa@linux.intel.com>
arch/x86/mm/numa_64.c

index 243d18d..c3496e2 100644 (file)
 #include <asm/acpi.h>
 #include <asm/amd_nb.h>
 
+struct numa_memblk {
+       u64                     start;
+       u64                     end;
+       int                     nid;
+};
+
+struct numa_meminfo {
+       int                     nr_blks;
+       struct numa_memblk      blk[NR_NODE_MEMBLKS];
+};
+
 struct pglist_data *node_data[MAX_NUMNODES] __read_mostly;
 EXPORT_SYMBOL(node_data);
 
@@ -33,9 +44,7 @@ struct memnode memnode;
 static unsigned long __initdata nodemap_addr;
 static unsigned long __initdata nodemap_size;
 
-static int num_node_memblks __initdata;
-static struct bootnode node_memblk_range[NR_NODE_MEMBLKS] __initdata;
-static int memblk_nodeid[NR_NODE_MEMBLKS] __initdata;
+static struct numa_meminfo numa_meminfo __initdata;
 
 struct bootnode numa_nodes[MAX_NUMNODES] __initdata;
 
@@ -46,16 +55,15 @@ struct bootnode numa_nodes[MAX_NUMNODES] __initdata;
  * 0 if memnodmap[] too small (of shift too small)
  * -1 if node overlap or lost ram (shift too big)
  */
-static int __init populate_memnodemap(const struct bootnode *nodes,
-                                     int numnodes, int shift, int *nodeids)
+static int __init populate_memnodemap(const struct numa_meminfo *mi, int shift)
 {
        unsigned long addr, end;
        int i, res = -1;
 
        memset(memnodemap, 0xff, sizeof(s16)*memnodemapsize);
-       for (i = 0; i < numnodes; i++) {
-               addr = nodes[i].start;
-               end = nodes[i].end;
+       for (i = 0; i < mi->nr_blks; i++) {
+               addr = mi->blk[i].start;
+               end = mi->blk[i].end;
                if (addr >= end)
                        continue;
                if ((end >> shift) >= memnodemapsize)
@@ -63,7 +71,7 @@ static int __init populate_memnodemap(const struct bootnode *nodes,
                do {
                        if (memnodemap[addr >> shift] != NUMA_NO_NODE)
                                return -1;
-                       memnodemap[addr >> shift] = nodeids[i];
+                       memnodemap[addr >> shift] = mi->blk[i].nid;
                        addr += (1UL << shift);
                } while (addr < end);
                res = 1;
@@ -101,16 +109,15 @@ static int __init allocate_cachealigned_memnodemap(void)
  * The LSB of all start and end addresses in the node map is the value of the
  * maximum possible shift.
  */
-static int __init extract_lsb_from_nodes(const struct bootnode *nodes,
-                                        int numnodes)
+static int __init extract_lsb_from_nodes(const struct numa_meminfo *mi)
 {
        int i, nodes_used = 0;
        unsigned long start, end;
        unsigned long bitfield = 0, memtop = 0;
 
-       for (i = 0; i < numnodes; i++) {
-               start = nodes[i].start;
-               end = nodes[i].end;
+       for (i = 0; i < mi->nr_blks; i++) {
+               start = mi->blk[i].start;
+               end = mi->blk[i].end;
                if (start >= end)
                        continue;
                bitfield |= start;
@@ -126,18 +133,17 @@ static int __init extract_lsb_from_nodes(const struct bootnode *nodes,
        return i;
 }
 
-static int __init compute_hash_shift(struct bootnode *nodes, int numnodes,
-                                    int *nodeids)
+static int __init compute_hash_shift(const struct numa_meminfo *mi)
 {
        int shift;
 
-       shift = extract_lsb_from_nodes(nodes, numnodes);
+       shift = extract_lsb_from_nodes(mi);
        if (allocate_cachealigned_memnodemap())
                return -1;
        printk(KERN_DEBUG "NUMA: Using %d for the hash shift.\n",
                shift);
 
-       if (populate_memnodemap(nodes, numnodes, shift, nodeids) != 1) {
+       if (populate_memnodemap(mi, shift) != 1) {
                printk(KERN_INFO "Your memory is not aligned you need to "
                       "rebuild your kernel with a bigger NODEMAPSIZE "
                       "shift=%d\n", shift);
@@ -185,21 +191,25 @@ static void * __init early_node_mem(int nodeid, unsigned long start,
 
 static __init int conflicting_memblks(unsigned long start, unsigned long end)
 {
+       struct numa_meminfo *mi = &numa_meminfo;
        int i;
-       for (i = 0; i < num_node_memblks; i++) {
-               struct bootnode *nd = &node_memblk_range[i];
-               if (nd->start == nd->end)
+
+       for (i = 0; i < mi->nr_blks; i++) {
+               struct numa_memblk *blk = &mi->blk[i];
+
+               if (blk->start == blk->end)
                        continue;
-               if (nd->end > start && nd->start < end)
-                       return memblk_nodeid[i];
-               if (nd->end == end && nd->start == start)
-                       return memblk_nodeid[i];
+               if (blk->end > start && blk->start < end)
+                       return blk->nid;
+               if (blk->end == end && blk->start == start)
+                       return blk->nid;
        }
        return -1;
 }
 
 int __init numa_add_memblk(int nid, u64 start, u64 end)
 {
+       struct numa_meminfo *mi = &numa_meminfo;
        int i;
 
        i = conflicting_memblks(start, end);
@@ -213,10 +223,10 @@ int __init numa_add_memblk(int nid, u64 start, u64 end)
                return -EINVAL;
        }
 
-       node_memblk_range[num_node_memblks].start = start;
-       node_memblk_range[num_node_memblks].end = end;
-       memblk_nodeid[num_node_memblks] = nid;
-       num_node_memblks++;
+       mi->blk[mi->nr_blks].start = start;
+       mi->blk[mi->nr_blks].end = end;
+       mi->blk[mi->nr_blks].nid = nid;
+       mi->nr_blks++;
        return 0;
 }
 
@@ -315,66 +325,59 @@ static int __init nodes_cover_memory(const struct bootnode *nodes)
 
 static int __init numa_register_memblks(void)
 {
+       struct numa_meminfo *mi = &numa_meminfo;
        int i;
 
        /*
         * Join together blocks on the same node, holes between
         * which don't overlap with memory on other nodes.
         */
-       for (i = 0; i < num_node_memblks; ++i) {
+       for (i = 0; i < mi->nr_blks; ++i) {
+               struct numa_memblk *bi = &mi->blk[i];
                int j, k;
 
-               for (j = i + 1; j < num_node_memblks; ++j) {
+               for (j = i + 1; j < mi->nr_blks; ++j) {
+                       struct numa_memblk *bj = &mi->blk[j];
                        unsigned long start, end;
 
-                       if (memblk_nodeid[i] != memblk_nodeid[j])
+                       if (bi->nid != bj->nid)
                                continue;
-                       start = min(node_memblk_range[i].end,
-                                   node_memblk_range[j].end);
-                       end = max(node_memblk_range[i].start,
-                                 node_memblk_range[j].start);
-                       for (k = 0; k < num_node_memblks; ++k) {
-                               if (memblk_nodeid[i] == memblk_nodeid[k])
+                       start = min(bi->end, bj->end);
+                       end = max(bi->start, bj->start);
+                       for (k = 0; k < mi->nr_blks; ++k) {
+                               struct numa_memblk *bk = &mi->blk[k];
+
+                               if (bi->nid == bk->nid)
                                        continue;
-                               if (start < node_memblk_range[k].end &&
-                                   end > node_memblk_range[k].start)
+                               if (start < bk->end && end > bk->start)
                                        break;
                        }
-                       if (k < num_node_memblks)
+                       if (k < mi->nr_blks)
                                continue;
-                       start = min(node_memblk_range[i].start,
-                                   node_memblk_range[j].start);
-                       end = max(node_memblk_range[i].end,
-                                 node_memblk_range[j].end);
+                       start = min(bi->start, bj->start);
+                       end = max(bi->end, bj->end);
                        printk(KERN_INFO "NUMA: Node %d [%Lx,%Lx) + [%Lx,%Lx) -> [%lx,%lx)\n",
-                              memblk_nodeid[i],
-                              node_memblk_range[i].start,
-                              node_memblk_range[i].end,
-                              node_memblk_range[j].start,
-                              node_memblk_range[j].end,
+                              bi->nid, bi->start, bi->end, bj->start, bj->end,
                               start, end);
-                       node_memblk_range[i].start = start;
-                       node_memblk_range[i].end = end;
-                       k = --num_node_memblks - j;
-                       memmove(memblk_nodeid + j, memblk_nodeid + j+1,
-                               k * sizeof(*memblk_nodeid));
-                       memmove(node_memblk_range + j, node_memblk_range + j+1,
-                               k * sizeof(*node_memblk_range));
+                       bi->start = start;
+                       bi->end = end;
+                       k = --mi->nr_blks - j;
+                       memmove(mi->blk + j, mi->blk + j + 1,
+                               k * sizeof(mi->blk[0]));
                        --j;
                }
        }
 
-       memnode_shift = compute_hash_shift(node_memblk_range, num_node_memblks,
-                                          memblk_nodeid);
+       memnode_shift = compute_hash_shift(mi);
        if (memnode_shift < 0) {
                printk(KERN_ERR "NUMA: No NUMA node hash function found. Contact maintainer\n");
                return -EINVAL;
        }
 
-       for (i = 0; i < num_node_memblks; i++)
-               memblock_x86_register_active_regions(memblk_nodeid[i],
-                               node_memblk_range[i].start >> PAGE_SHIFT,
-                               node_memblk_range[i].end >> PAGE_SHIFT);
+       for (i = 0; i < mi->nr_blks; i++)
+               memblock_x86_register_active_regions(mi->blk[i].nid,
+                                       mi->blk[i].start >> PAGE_SHIFT,
+                                       mi->blk[i].end >> PAGE_SHIFT);
 
        /* for out of order entries */
        sort_node_map();
@@ -701,7 +704,7 @@ static int __init split_nodes_size_interleave(u64 addr, u64 max_addr, u64 size)
 static int __init numa_emulation(unsigned long start_pfn,
                        unsigned long last_pfn, int acpi, int amd)
 {
-       static int nodeid[NR_NODE_MEMBLKS] __initdata;
+       static struct numa_meminfo ei __initdata;
        u64 addr = start_pfn << PAGE_SHIFT;
        u64 max_addr = last_pfn << PAGE_SHIFT;
        int num_nodes;
@@ -727,10 +730,14 @@ static int __init numa_emulation(unsigned long start_pfn,
        if (num_nodes < 0)
                return num_nodes;
 
-       for (i = 0; i < ARRAY_SIZE(nodeid); i++)
-               nodeid[i] = i;
+       ei.nr_blks = num_nodes;
+       for (i = 0; i < ei.nr_blks; i++) {
+               ei.blk[i].start = nodes[i].start;
+               ei.blk[i].end = nodes[i].end;
+               ei.blk[i].nid = i;
+       }
 
-       memnode_shift = compute_hash_shift(nodes, num_nodes, nodeid);
+       memnode_shift = compute_hash_shift(&ei);
        if (memnode_shift < 0) {
                memnode_shift = 0;
                printk(KERN_ERR "No NUMA hash function found.  NUMA emulation "
@@ -797,9 +804,7 @@ void __init initmem_init(void)
                nodes_clear(mem_nodes_parsed);
                nodes_clear(node_possible_map);
                nodes_clear(node_online_map);
-               num_node_memblks = 0;
-               memset(node_memblk_range, 0, sizeof(node_memblk_range));
-               memset(memblk_nodeid, 0, sizeof(memblk_nodeid));
+               memset(&numa_meminfo, 0, sizeof(numa_meminfo));
                memset(numa_nodes, 0, sizeof(numa_nodes));
                remove_all_active_ranges();