[PATCH] sparsemem hotplug base
Andy Whitcroft [Thu, 23 Jun 2005 07:08:00 +0000 (00:08 -0700)]
Make sparse's initalization be accessible at runtime.  This allows sparse
mappings to be created after boot in a hotplug situation.

This patch is separated from the previous one just to give an indication how
much of the sparse infrastructure is *just* for hotplug memory.

The section_mem_map doesn't really store a pointer.  It stores something that
is convenient to do some math against to get a pointer.  It isn't valid to
just do *section_mem_map, so I don't think it should be stored as a pointer.

There are a couple of things I'd like to store about a section.  First of all,
the fact that it is !NULL does not mean that it is present.  There could be
such a combination where section_mem_map *is* NULL, but the math gets you
properly to a real mem_map.  So, I don't think that check is safe.

Since we're storing 32-bit-aligned structures, we have a few bits in the
bottom of the pointer to play with.  Use one bit to encode whether there's
really a mem_map there, and the other one to tell whether there's a valid
section there.  We need to distinguish between the two because sometimes
there's a gap between when a section is discovered to be present and when we
can get the mem_map for it.

Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Bob Picco <bob.picco@hp.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

include/linux/mmzone.h
mm/page_alloc.c
mm/sparse.c

index 746b57e..6c90461 100644 (file)
@@ -476,11 +476,56 @@ extern struct pglist_data contig_page_data;
 
 struct page;
 struct mem_section {
-       struct page *section_mem_map;
+       /*
+        * This is, logically, a pointer to an array of struct
+        * pages.  However, it is stored with some other magic.
+        * (see sparse.c::sparse_init_one_section())
+        *
+        * Making it a UL at least makes someone do a cast
+        * before using it wrong.
+        */
+       unsigned long section_mem_map;
 };
 
 extern struct mem_section mem_section[NR_MEM_SECTIONS];
 
+static inline struct mem_section *__nr_to_section(unsigned long nr)
+{
+       return &mem_section[nr];
+}
+
+/*
+ * We use the lower bits of the mem_map pointer to store
+ * a little bit of information.  There should be at least
+ * 3 bits here due to 32-bit alignment.
+ */
+#define        SECTION_MARKED_PRESENT  (1UL<<0)
+#define SECTION_HAS_MEM_MAP    (1UL<<1)
+#define SECTION_MAP_LAST_BIT   (1UL<<2)
+#define SECTION_MAP_MASK       (~(SECTION_MAP_LAST_BIT-1))
+
+static inline struct page *__section_mem_map_addr(struct mem_section *section)
+{
+       unsigned long map = section->section_mem_map;
+       map &= SECTION_MAP_MASK;
+       return (struct page *)map;
+}
+
+static inline int valid_section(struct mem_section *section)
+{
+       return (section->section_mem_map & SECTION_MARKED_PRESENT);
+}
+
+static inline int section_has_mem_map(struct mem_section *section)
+{
+       return (section->section_mem_map & SECTION_HAS_MEM_MAP);
+}
+
+static inline int valid_section_nr(unsigned long nr)
+{
+       return valid_section(__nr_to_section(nr));
+}
+
 /*
  * Given a kernel address, find the home node of the underlying memory.
  */
@@ -488,24 +533,25 @@ extern struct mem_section mem_section[NR_MEM_SECTIONS];
 
 static inline struct mem_section *__pfn_to_section(unsigned long pfn)
 {
-       return &mem_section[pfn_to_section_nr(pfn)];
+       return __nr_to_section(pfn_to_section_nr(pfn));
 }
 
 #define pfn_to_page(pfn)                                               \
 ({                                                                     \
        unsigned long __pfn = (pfn);                                    \
-       __pfn_to_section(__pfn)->section_mem_map + __pfn;               \
+       __section_mem_map_addr(__pfn_to_section(__pfn)) + __pfn;        \
 })
 #define page_to_pfn(page)                                              \
 ({                                                                     \
-       page - mem_section[page_to_section(page)].section_mem_map;      \
+       page - __section_mem_map_addr(__nr_to_section(                  \
+               page_to_section(page)));                                \
 })
 
 static inline int pfn_valid(unsigned long pfn)
 {
        if (pfn_to_section_nr(pfn) >= NR_MEM_SECTIONS)
                return 0;
-       return mem_section[pfn_to_section_nr(pfn)].section_mem_map != 0;
+       return valid_section(__nr_to_section(pfn_to_section_nr(pfn)));
 }
 
 /*
index 1eb683f..7ee675a 100644 (file)
@@ -1650,8 +1650,8 @@ void __init memmap_init_zone(unsigned long size, int nid, unsigned long zone,
                unsigned long start_pfn)
 {
        struct page *page;
-       int end_pfn = start_pfn + size;
-       int pfn;
+       unsigned long end_pfn = start_pfn + size;
+       unsigned long pfn;
 
        for (pfn = start_pfn; pfn < end_pfn; pfn++, page++) {
                if (!early_pfn_valid(pfn))
index f888385..b54e304 100644 (file)
@@ -25,7 +25,7 @@ void memory_present(int nid, unsigned long start, unsigned long end)
        for (pfn = start; pfn < end; pfn += PAGES_PER_SECTION) {
                unsigned long section = pfn_to_section_nr(pfn);
                if (!mem_section[section].section_mem_map)
-                       mem_section[section].section_mem_map = (void *) -1;
+                       mem_section[section].section_mem_map = SECTION_MARKED_PRESENT;
        }
 }
 
@@ -51,6 +51,56 @@ unsigned long __init node_memmap_size_bytes(int nid, unsigned long start_pfn,
 }
 
 /*
+ * Subtle, we encode the real pfn into the mem_map such that
+ * the identity pfn - section_mem_map will return the actual
+ * physical page frame number.
+ */
+static unsigned long sparse_encode_mem_map(struct page *mem_map, unsigned long pnum)
+{
+       return (unsigned long)(mem_map - (section_nr_to_pfn(pnum)));
+}
+
+/*
+ * We need this if we ever free the mem_maps.  While not implemented yet,
+ * this function is included for parity with its sibling.
+ */
+static __attribute((unused))
+struct page *sparse_decode_mem_map(unsigned long coded_mem_map, unsigned long pnum)
+{
+       return ((struct page *)coded_mem_map) + section_nr_to_pfn(pnum);
+}
+
+static int sparse_init_one_section(struct mem_section *ms,
+               unsigned long pnum, struct page *mem_map)
+{
+       if (!valid_section(ms))
+               return -EINVAL;
+
+       ms->section_mem_map |= sparse_encode_mem_map(mem_map, pnum);
+
+       return 1;
+}
+
+static struct page *sparse_early_mem_map_alloc(unsigned long pnum)
+{
+       struct page *map;
+       int nid = early_pfn_to_nid(section_nr_to_pfn(pnum));
+
+       map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
+       if (map)
+               return map;
+
+       map = alloc_bootmem_node(NODE_DATA(nid),
+                       sizeof(struct page) * PAGES_PER_SECTION);
+       if (map)
+               return map;
+
+       printk(KERN_WARNING "%s: allocation failed\n", __FUNCTION__);
+       mem_section[pnum].section_mem_map = 0;
+       return NULL;
+}
+
+/*
  * Allocate the accumulated non-linear sections, allocate a mem_map
  * for each and record the physical to section mapping.
  */
@@ -58,28 +108,30 @@ void sparse_init(void)
 {
        unsigned long pnum;
        struct page *map;
-       int nid;
 
        for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
-               if (!mem_section[pnum].section_mem_map)
+               if (!valid_section_nr(pnum))
                        continue;
 
-               nid = early_pfn_to_nid(section_nr_to_pfn(pnum));
-               map = alloc_remap(nid, sizeof(struct page) * PAGES_PER_SECTION);
-               if (!map)
-                       map = alloc_bootmem_node(NODE_DATA(nid),
-                               sizeof(struct page) * PAGES_PER_SECTION);
-               if (!map) {
-                       mem_section[pnum].section_mem_map = 0;
-                       continue;
-               }
-
-               /*
-                * Subtle, we encode the real pfn into the mem_map such that
-                * the identity pfn - section_mem_map will return the actual
-                * physical page frame number.
-                */
-               mem_section[pnum].section_mem_map = map -
-                                               section_nr_to_pfn(pnum);
+               map = sparse_early_mem_map_alloc(pnum);
+               if (map)
+                       sparse_init_one_section(&mem_section[pnum], pnum, map);
        }
 }
+
+/*
+ * returns the number of sections whose mem_maps were properly
+ * set.  If this is <=0, then that means that the passed-in
+ * map was not consumed and must be freed.
+ */
+int sparse_add_one_section(unsigned long start_pfn, int nr_pages, struct page *map)
+{
+       struct mem_section *ms = __pfn_to_section(start_pfn);
+
+       if (ms->section_mem_map & SECTION_MARKED_PRESENT)
+               return -EEXIST;
+
+       ms->section_mem_map |= SECTION_MARKED_PRESENT;
+
+       return sparse_init_one_section(ms, pfn_to_section_nr(start_pfn), map);
+}