arch/x86/kernel/os/memnode.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
  23  * Use is subject to license terms.
  24  */
  25
  26 #include <sys/systm.h>
  27 #include <sys/sysmacros.h>
  28 #include <sys/bootconf.h>
  29 #include <sys/atomic.h>
  30 #include <sys/lgrp.h>
  31 #include <sys/memlist.h>
  32 #include <sys/memnode.h>
  33 #include <sys/platform_module.h>
  34 #include <vm/vm_dep.h>
  35
  36 int     max_mem_nodes = 1;
  37
  38 struct mem_node_conf mem_node_config[MAX_MEM_NODES];
  39 int mem_node_pfn_shift;
  40 /*
  41  * num_memnodes should be updated atomically and always >=
  42  * the number of bits in memnodes_mask or the algorithm may fail.
  43  */
  44 uint16_t num_memnodes;
  45 mnodeset_t memnodes_mask; /* assumes 8*(sizeof(mnodeset_t)) >= MAX_MEM_NODES */
  46
  47 /*
  48  * If set, mem_node_physalign should be a power of two, and
  49  * should reflect the minimum address alignment of each node.
  50  */
  51 uint64_t mem_node_physalign;
  52
  53 /*
  54  * Platform hooks we will need.
  55  */
  56
  57 #pragma weak plat_build_mem_nodes
  58 #pragma weak plat_slice_add
  59 #pragma weak plat_slice_del
  60
  61 /*
  62  * Adjust the memnode config after a DR operation.
  63  *
  64  * It is rather tricky to do these updates since we can't
  65  * protect the memnode structures with locks, so we must
  66  * be mindful of the order in which updates and reads to
  67  * these values can occur.
  68  */
  69
  70 void
  71 mem_node_add_slice(pfn_t start, pfn_t end)
  72 {
  73         int mnode;
  74
  75         /*
  76          * DR will pass us the first pfn that is allocatable.
  77          * We need to round down to get the real start of
  78          * the slice.
  79          */
  80         if (mem_node_physalign) {
  81                 start &= ~(btop(mem_node_physalign) - 1);
  82                 end = roundup(end, btop(mem_node_physalign)) - 1;
  83         }
  84
  85         mnode = PFN_2_MEM_NODE(start);
  86         ASSERT(mnode >= 0 && mnode < max_mem_nodes);
  87
  88         if (atomic_cas_32((uint32_t *)&mem_node_config[mnode].exists, 0, 1)) {
  89                 /*
  90                  * Add slice to existing node.
  91                  */
  92                 if (start < mem_node_config[mnode].physbase)
  93                         mem_node_config[mnode].physbase = start;
  94                 if (end > mem_node_config[mnode].physmax)
  95                         mem_node_config[mnode].physmax = end;
  96         } else {
  97                 mem_node_config[mnode].physbase = start;
  98                 mem_node_config[mnode].physmax = end;
  99                 atomic_inc_16(&num_memnodes);
 100                 atomic_or_64(&memnodes_mask, 1ull << mnode);
 101         }
 102
 103         /*
 104          * Inform the common lgrp framework about the new memory
 105          */
 106         lgrp_config(LGRP_CONFIG_MEM_ADD, mnode, MEM_NODE_2_LGRPHAND(mnode));
 107 }
 108
 109 /*
 110  * Remove a PFN range from a memnode.  On some platforms,
 111  * the memnode will be created with physbase at the first
 112  * allocatable PFN, but later deleted with the MC slice
 113  * base address converted to a PFN, in which case we need
 114  * to assume physbase and up.
 115  */
 116 void
 117 mem_node_del_slice(pfn_t start, pfn_t end)
 118 {
 119         int mnode;
 120         pgcnt_t delta_pgcnt, node_size;
 121
 122         if (mem_node_physalign) {
 123                 start &= ~(btop(mem_node_physalign) - 1);
 124                 end = roundup(end, btop(mem_node_physalign)) - 1;
 125         }
 126         mnode = PFN_2_MEM_NODE(start);
 127
 128         ASSERT(mnode >= 0 && mnode < max_mem_nodes);
 129         ASSERT(mem_node_config[mnode].exists == 1);
 130
 131         delta_pgcnt = end - start;
 132         node_size = mem_node_config[mnode].physmax -
 133             mem_node_config[mnode].physbase;
 134
 135         if (node_size > delta_pgcnt) {
 136                 /*
 137                  * Subtract the slice from the memnode.
 138                  */
 139                 if (start <= mem_node_config[mnode].physbase)
 140                         mem_node_config[mnode].physbase = end + 1;
 141                 ASSERT(end <= mem_node_config[mnode].physmax);
 142                 if (end == mem_node_config[mnode].physmax)
 143                         mem_node_config[mnode].physmax = start - 1;
 144         } else {
 145                 /*
 146                  * Let the common lgrp framework know this mnode is
 147                  * leaving
 148                  */
 149                 lgrp_config(LGRP_CONFIG_MEM_DEL,
 150                     mnode, MEM_NODE_2_LGRPHAND(mnode));
 151
 152                 /*
 153                  * Delete the whole node.
 154                  */
 155                 ASSERT(MNODE_PGCNT(mnode) == 0);
 156                 atomic_and_64(&memnodes_mask, ~(1ull << mnode));
 157                 atomic_dec_16(&num_memnodes);
 158                 mem_node_config[mnode].exists = 0;
 159         }
 160 }
 161
 162 void
 163 mem_node_add_range(pfn_t start, pfn_t end)
 164 {
 165         if (&plat_slice_add)
 166                 plat_slice_add(start, end);
 167         else
 168                 mem_node_add_slice(start, end);
 169 }
 170
 171 void
 172 mem_node_del_range(pfn_t start, pfn_t end)
 173 {
 174         if (&plat_slice_del)
 175                 plat_slice_del(start, end);
 176         else
 177                 mem_node_del_slice(start, end);
 178 }
 179
 180 void
 181 startup_build_mem_nodes(struct memlist *list)
 182 {
 183         pfn_t   start, end;
 184
 185         /* LINTED: ASSERT will always true or false */
 186         ASSERT(NBBY * sizeof (mnodeset_t) >= max_mem_nodes);
 187
 188         if (&plat_build_mem_nodes) {
 189                 plat_build_mem_nodes(list);
 190         } else {
 191                 /*
 192                  * Boot install lists are arranged <addr, len>, ...
 193                  */
 194                 while (list) {
 195                         start = list->ml_address >> PAGESHIFT;
 196                         if (start > physmax)
 197                                 continue;
 198                         end =
 199                             (list->ml_address + list->ml_size - 1) >> PAGESHIFT;
 200                         if (end > physmax)
 201                                 end = physmax;
 202                         mem_node_add_range(start, end);
 203                         list = list->ml_next;
 204                 }
 205                 mem_node_physalign = 0;
 206                 mem_node_pfn_shift = 0;
 207         }
 208 }
 209
 210 /*
 211  * Allocate an unassigned memnode.
 212  */
 213 int
 214 mem_node_alloc()
 215 {
 216         int mnode;
 217
 218         /*
 219          * Find an unused memnode.  Update it atomically to prevent
 220          * a first time memnode creation race.
 221          */
 222         for (mnode = 0; mnode < max_mem_nodes; mnode++)
 223                 if (atomic_cas_32((uint32_t *)&mem_node_config[mnode].exists,
 224                     0, 1) == 0)
 225                         break;
 226
 227         if (mnode >= max_mem_nodes)
 228                 panic("Out of free memnodes\n");
 229
 230         mem_node_config[mnode].physbase = (pfn_t)-1l;
 231         mem_node_config[mnode].physmax = 0;
 232         atomic_inc_16(&num_memnodes);
 233         atomic_or_64(&memnodes_mask, 1ull << mnode);
 234
 235         return (mnode);
 236 }
 237
 238 /*
 239  * Find the intersection between a memnode and a memlist
 240  * and returns the number of pages that overlap.
 241  *
 242  * Assumes the list is protected from DR operations by
 243  * the memlist lock.
 244  */
 245 pgcnt_t
 246 mem_node_memlist_pages(int mnode, struct memlist *mlist)
 247 {
 248         pfn_t           base, end;
 249         pfn_t           cur_base, cur_end;
 250         pgcnt_t         npgs;
 251         struct memlist  *pmem;
 252
 253         base = mem_node_config[mnode].physbase;
 254         end = mem_node_config[mnode].physmax;
 255         npgs = 0;
 256
 257         memlist_read_lock();
 258
 259         for (pmem = mlist; pmem; pmem = pmem->ml_next) {
 260                 cur_base = btop(pmem->ml_address);
 261                 cur_end = cur_base + btop(pmem->ml_size) - 1;
 262                 if (end < cur_base || base > cur_end)
 263                         continue;
 264                 npgs = npgs + (MIN(cur_end, end) -
 265                     MAX(cur_base, base)) + 1;
 266         }
 267
 268         memlist_read_unlock();
 269
 270         return (npgs);
 271 }