mm: create non-atomic version of SetPageReserved for init use
authorAlexander Duyck <alexander.h.duyck@linux.intel.com>
Fri, 26 Oct 2018 22:07:48 +0000 (15:07 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 26 Oct 2018 23:26:34 +0000 (16:26 -0700)
It doesn't make much sense to use the atomic SetPageReserved at init time
when we are using memset to clear the memory and manipulating the page
flags via simple "&=" and "|=" operations in __init_single_page.

This patch adds a non-atomic version __SetPageReserved that can be used
during page init and shows about a 10% improvement in initialization times
on the systems I have available for testing.  On those systems I saw
initialization times drop from around 35 seconds to around 32 seconds to
initialize a 3TB block of persistent memory.  I believe the main advantage
of this is that it allows for more compiler optimization as the __set_bit
operation can be reordered whereas the atomic version cannot.

I tried adding a bit of documentation based on f1dd2cd13c4 ("mm,
memory_hotplug: do not associate hotadded memory to zones until online").

Ideally the reserved flag should be set earlier since there is a brief
window where the page is initialization via __init_single_page and we have
not set the PG_Reserved flag.  I'm leaving that for a future patch set as
that will require a more significant refactor.

Link: http://lkml.kernel.org/r/20180925202018.3576.11607.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: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/page-flags.h
mm/page_alloc.c

index 934f91e..50ce1bd 100644 (file)
@@ -303,6 +303,7 @@ PAGEFLAG(Foreign, foreign, PF_NO_COMPOUND);
 
 PAGEFLAG(Reserved, reserved, PF_NO_COMPOUND)
        __CLEARPAGEFLAG(Reserved, reserved, PF_NO_COMPOUND)
+       __SETPAGEFLAG(Reserved, reserved, PF_NO_COMPOUND)
 PAGEFLAG(SwapBacked, swapbacked, PF_NO_TAIL)
        __CLEARPAGEFLAG(SwapBacked, swapbacked, PF_NO_TAIL)
        __SETPAGEFLAG(SwapBacked, swapbacked, PF_NO_TAIL)
index eb6c50c..cee1abf 100644 (file)
@@ -1232,7 +1232,12 @@ void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
                        /* Avoid false-positive PageTail() */
                        INIT_LIST_HEAD(&page->lru);
 
-                       SetPageReserved(page);
+                       /*
+                        * no need for atomic set_bit because the struct
+                        * page is not visible yet so nobody should
+                        * access it yet.
+                        */
+                       __SetPageReserved(page);
                }
        }
 }
@@ -5508,7 +5513,7 @@ not_early:
                page = pfn_to_page(pfn);
                __init_single_page(page, pfn, zone, nid);
                if (context == MEMMAP_HOTPLUG)
-                       SetPageReserved(page);
+                       __SetPageReserved(page);
 
                /*
                 * Mark the block movable so that blocks are reserved for