[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[PATCH 1/2] xen/heap: Split init_heap_pages() in two



From: Julien Grall <jgrall@xxxxxxxxxx>

At the moment, init_heap_pages() will call free_heap_pages() page
by page. To reduce the time to initialize the heap, we will want
to provide multiple pages at the same time.

init_heap_pages() is now split in two parts:
    - init_heap_pages(): will break down the range in multiple set
      of contiguous pages. For now, the criteria is the pages should
      belong to the same NUMA node.
    - init_contig_pages(): will initialize a set of contiguous pages.
      For now the pages are still passed one by one to free_heap_pages().

Note that the comment before init_heap_pages() is heavily outdated and
does not reflect the current code. So update it.

This patch is a merge/rework of patches from David Woodhouse and
Hongyan Xia.

Signed-off-by: Julien Grall <jgrall@xxxxxxxxxx>

----

Interestingly, I was expecting this patch to perform worse. However,
from testing there is a small increase in perf.

That said, I split the patch because it keeps refactoring and
optimization separated.
---
 xen/common/page_alloc.c | 82 +++++++++++++++++++++++++++--------------
 1 file changed, 55 insertions(+), 27 deletions(-)

diff --git a/xen/common/page_alloc.c b/xen/common/page_alloc.c
index 3e6504283f1e..a1938df1406c 100644
--- a/xen/common/page_alloc.c
+++ b/xen/common/page_alloc.c
@@ -1778,16 +1778,55 @@ int query_page_offline(mfn_t mfn, uint32_t *status)
 }
 
 /*
- * Hand the specified arbitrary page range to the specified heap zone
- * checking the node_id of the previous page.  If they differ and the
- * latter is not on a MAX_ORDER boundary, then we reserve the page by
- * not freeing it to the buddy allocator.
+ * init_contig_heap_pages() is intended to only take pages from the same
+ * NUMA node.
  */
+static bool is_contig_page(struct page_info *pg, unsigned int nid)
+{
+    return (nid == (phys_to_nid(page_to_maddr(pg))));
+}
+
+/*
+ * This function should only be called with valid pages from the same NUMA
+ * node.
+ *
+ * Callers should use is_contig_page() first to check if all the pages
+ * in a range are contiguous.
+ */
+static void init_contig_heap_pages(struct page_info *pg, unsigned long 
nr_pages,
+                                   bool need_scrub)
+{
+    unsigned long s, e;
+    unsigned int nid = phys_to_nid(page_to_maddr(pg));
+
+    s = mfn_x(page_to_mfn(pg));
+    e = mfn_x(mfn_add(page_to_mfn(pg + nr_pages - 1), 1));
+    if ( unlikely(!avail[nid]) )
+    {
+        bool use_tail = !(s & ((1UL << MAX_ORDER) - 1)) &&
+                        (find_first_set_bit(e) <= find_first_set_bit(s));
+        unsigned long n;
+
+        n = init_node_heap(nid, s, nr_pages, &use_tail);
+        BUG_ON(n > nr_pages);
+        if ( use_tail )
+            e -= n;
+        else
+            s += n;
+    }
+
+    while ( s < e )
+    {
+        free_heap_pages(mfn_to_page(_mfn(s)), 0, need_scrub);
+        s += 1UL;
+    }
+}
+
 static void init_heap_pages(
     struct page_info *pg, unsigned long nr_pages)
 {
     unsigned long i;
-    bool idle_scrub = false;
+    bool need_scrub = scrub_debug;
 
     /*
      * Keep MFN 0 away from the buddy allocator to avoid crossing zone
@@ -1812,35 +1851,24 @@ static void init_heap_pages(
     spin_unlock(&heap_lock);
 
     if ( system_state < SYS_STATE_active && opt_bootscrub == BOOTSCRUB_IDLE )
-        idle_scrub = true;
+        need_scrub = true;
 
-    for ( i = 0; i < nr_pages; i++ )
+    for ( i = 0; i < nr_pages; )
     {
-        unsigned int nid = phys_to_nid(page_to_maddr(pg+i));
+        unsigned int nid = phys_to_nid(page_to_maddr(pg));
+        unsigned long left = nr_pages - i;
+        unsigned long contig_pages;
 
-        if ( unlikely(!avail[nid]) )
+        for ( contig_pages = 1; contig_pages < left; contig_pages++ )
         {
-            unsigned long s = mfn_x(page_to_mfn(pg + i));
-            unsigned long e = mfn_x(mfn_add(page_to_mfn(pg + nr_pages - 1), 
1));
-            bool use_tail = (nid == phys_to_nid(pfn_to_paddr(e - 1))) &&
-                            !(s & ((1UL << MAX_ORDER) - 1)) &&
-                            (find_first_set_bit(e) <= find_first_set_bit(s));
-            unsigned long n;
-
-            n = init_node_heap(nid, mfn_x(page_to_mfn(pg + i)), nr_pages - i,
-                               &use_tail);
-            BUG_ON(i + n > nr_pages);
-            if ( n && !use_tail )
-            {
-                i += n - 1;
-                continue;
-            }
-            if ( i + n == nr_pages )
+            if ( !is_contig_page(pg + contig_pages, nid) )
                 break;
-            nr_pages -= n;
         }
 
-        free_heap_pages(pg + i, 0, scrub_debug || idle_scrub);
+        init_contig_heap_pages(pg, contig_pages, need_scrub);
+
+        pg += contig_pages;
+        i += contig_pages;
     }
 }
 
-- 
2.32.0




 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.