mm: drop meminit_pfn_in_nid as it is redundant
authorAlexander Duyck <alexander.h.duyck@linux.intel.com>
Tue, 14 May 2019 00:21:13 +0000 (17:21 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 14 May 2019 16:47:49 +0000 (09:47 -0700)
As best as I can tell the meminit_pfn_in_nid call is completely redundant.
The deferred memory initialization is already making use of
for_each_free_mem_range which in turn will call into __next_mem_range
which will only return a memory range if it matches the node ID provided
assuming it is not NUMA_NO_NODE.

I am operating on the assumption that there are no zones or pgdata_t
structures that have a NUMA node of NUMA_NO_NODE associated with them.  If
that is the case then __next_mem_range will never return a memory range
that doesn't match the zone's node ID and as such the check is redundant.

So one piece I would like to verify on this is if this works for ia64.
Technically it was using a different approach to get the node ID, but it
seems to have the node ID also encoded into the memblock.  So I am
assuming this is okay, but would like to get confirmation on that.

On my x86_64 test system with 384GB of memory per node I saw a reduction
in initialization time from 2.80s to 1.85s as a result of this patch.

Link: http://lkml.kernel.org/r/20190405221219.12227.93957.stgit@localhost.localdomain
Signed-off-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Reviewed-by: Pavel Tatashin <pavel.tatashin@microsoft.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Khalid Aziz <khalid.aziz@oracle.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: <yi.z.zhang@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/page_alloc.c

index 909adce33398181cc7761d230fff8cc89b3e3ee1..25b82be438d753652528f764b5de0e870a75f99f 100644 (file)
@@ -1416,36 +1416,22 @@ int __meminit early_pfn_to_nid(unsigned long pfn)
 #endif
 
 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
-static inline bool __meminit __maybe_unused
-meminit_pfn_in_nid(unsigned long pfn, int node,
-                  struct mminit_pfnnid_cache *state)
+/* Only safe to use early in boot when initialisation is single-threaded */
+static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
 {
        int nid;
 
-       nid = __early_pfn_to_nid(pfn, state);
+       nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
        if (nid >= 0 && nid != node)
                return false;
        return true;
 }
 
-/* Only safe to use early in boot when initialisation is single-threaded */
-static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
-{
-       return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
-}
-
 #else
-
 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
 {
        return true;
 }
-static inline bool __meminit  __maybe_unused
-meminit_pfn_in_nid(unsigned long pfn, int node,
-                  struct mminit_pfnnid_cache *state)
-{
-       return true;
-}
 #endif
 
 
@@ -1574,21 +1560,13 @@ static inline void __init pgdat_init_report_one_done(void)
  *
  * Then, we check if a current large page is valid by only checking the validity
  * of the head pfn.
- *
- * Finally, meminit_pfn_in_nid is checked on systems where pfns can interleave
- * within a node: a pfn is between start and end of a node, but does not belong
- * to this memory node.
  */
-static inline bool __init
-deferred_pfn_valid(int nid, unsigned long pfn,
-                  struct mminit_pfnnid_cache *nid_init_state)
+static inline bool __init deferred_pfn_valid(unsigned long pfn)
 {
        if (!pfn_valid_within(pfn))
                return false;
        if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn))
                return false;
-       if (!meminit_pfn_in_nid(pfn, nid, nid_init_state))
-               return false;
        return true;
 }
 
@@ -1596,15 +1574,14 @@ deferred_pfn_valid(int nid, unsigned long pfn,
  * Free pages to buddy allocator. Try to free aligned pages in
  * pageblock_nr_pages sizes.
  */
-static void __init deferred_free_pages(int nid, int zid, unsigned long pfn,
+static void __init deferred_free_pages(unsigned long pfn,
                                       unsigned long end_pfn)
 {
-       struct mminit_pfnnid_cache nid_init_state = { };
        unsigned long nr_pgmask = pageblock_nr_pages - 1;
        unsigned long nr_free = 0;
 
        for (; pfn < end_pfn; pfn++) {
-               if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) {
+               if (!deferred_pfn_valid(pfn)) {
                        deferred_free_range(pfn - nr_free, nr_free);
                        nr_free = 0;
                } else if (!(pfn & nr_pgmask)) {
@@ -1624,17 +1601,18 @@ static void __init deferred_free_pages(int nid, int zid, unsigned long pfn,
  * by performing it only once every pageblock_nr_pages.
  * Return number of pages initialized.
  */
-static unsigned long  __init deferred_init_pages(int nid, int zid,
+static unsigned long  __init deferred_init_pages(struct zone *zone,
                                                 unsigned long pfn,
                                                 unsigned long end_pfn)
 {
-       struct mminit_pfnnid_cache nid_init_state = { };
        unsigned long nr_pgmask = pageblock_nr_pages - 1;
+       int nid = zone_to_nid(zone);
        unsigned long nr_pages = 0;
+       int zid = zone_idx(zone);
        struct page *page = NULL;
 
        for (; pfn < end_pfn; pfn++) {
-               if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) {
+               if (!deferred_pfn_valid(pfn)) {
                        page = NULL;
                        continue;
                } else if (!page || !(pfn & nr_pgmask)) {
@@ -1697,12 +1675,12 @@ static int __init deferred_init_memmap(void *data)
        for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
                spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
                epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
-               nr_pages += deferred_init_pages(nid, zid, spfn, epfn);
+               nr_pages += deferred_init_pages(zone, spfn, epfn);
        }
        for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
                spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
                epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
-               deferred_free_pages(nid, zid, spfn, epfn);
+               deferred_free_pages(spfn, epfn);
        }
        pgdat_resize_unlock(pgdat, &flags);
 
@@ -1734,7 +1712,6 @@ static int __init deferred_init_memmap(void *data)
 static noinline bool __init
 deferred_grow_zone(struct zone *zone, unsigned int order)
 {
-       int zid = zone_idx(zone);
        int nid = zone_to_nid(zone);
        pg_data_t *pgdat = NODE_DATA(nid);
        unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
@@ -1784,7 +1761,7 @@ deferred_grow_zone(struct zone *zone, unsigned int order)
                while (spfn < epfn && nr_pages < nr_pages_needed) {
                        t = ALIGN(spfn + PAGES_PER_SECTION, PAGES_PER_SECTION);
                        first_deferred_pfn = min(t, epfn);
-                       nr_pages += deferred_init_pages(nid, zid, spfn,
+                       nr_pages += deferred_init_pages(zone, spfn,
                                                        first_deferred_pfn);
                        spfn = first_deferred_pfn;
                }
@@ -1796,7 +1773,7 @@ deferred_grow_zone(struct zone *zone, unsigned int order)
        for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
                spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
                epfn = min_t(unsigned long, first_deferred_pfn, PFN_DOWN(epa));
-               deferred_free_pages(nid, zid, spfn, epfn);
+               deferred_free_pages(spfn, epfn);
 
                if (first_deferred_pfn == epfn)
                        break;