x86/mm/hotplug: Modify PGD entry when removing memory
authorYasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Fri, 22 Aug 2014 20:27:34 +0000 (13:27 -0700)
committerIngo Molnar <mingo@kernel.org>
Tue, 16 Sep 2014 06:55:09 +0000 (08:55 +0200)
When hot-adding/removing memory, sync_global_pgds() is called
for synchronizing PGD to PGD entries of all processes MM.  But
when hot-removing memory, sync_global_pgds() does not work
correctly.

At first, sync_global_pgds() checks whether target PGD is none
or not.  And if PGD is none, the PGD is skipped.  But when
hot-removing memory, PGD may be none since PGD may be cleared by
free_pud_table().  So when sync_global_pgds() is called after
hot-removing memory, sync_global_pgds() should not skip PGD even
if the PGD is none.  And sync_global_pgds() must clear PGD
entries of all processes MM.

Currently sync_global_pgds() does not clear PGD entries of all
processes MM when hot-removing memory.  So when hot adding
memory which is same memory range as removed memory after
hot-removing memory, following call traces are shown:

 kernel BUG at arch/x86/mm/init_64.c:206!
 ...
 [<ffffffff815e0c80>] kernel_physical_mapping_init+0x1b2/0x1d2
 [<ffffffff815ced94>] init_memory_mapping+0x1d4/0x380
 [<ffffffff8104aebd>] arch_add_memory+0x3d/0xd0
 [<ffffffff815d03d9>] add_memory+0xb9/0x1b0
 [<ffffffff81352415>] acpi_memory_device_add+0x1af/0x28e
 [<ffffffff81325dc4>] acpi_bus_device_attach+0x8c/0xf0
 [<ffffffff813413b9>] acpi_ns_walk_namespace+0xc8/0x17f
 [<ffffffff81325d38>] ? acpi_bus_type_and_status+0xb7/0xb7
 [<ffffffff81325d38>] ? acpi_bus_type_and_status+0xb7/0xb7
 [<ffffffff813418ed>] acpi_walk_namespace+0x95/0xc5
 [<ffffffff81326b4c>] acpi_bus_scan+0x9a/0xc2
 [<ffffffff81326bff>] acpi_scan_bus_device_check+0x8b/0x12e
 [<ffffffff81326cb5>] acpi_scan_device_check+0x13/0x15
 [<ffffffff81320122>] acpi_os_execute_deferred+0x25/0x32
 [<ffffffff8107e02b>] process_one_work+0x17b/0x460
 [<ffffffff8107edfb>] worker_thread+0x11b/0x400
 [<ffffffff8107ece0>] ? rescuer_thread+0x400/0x400
 [<ffffffff81085aef>] kthread+0xcf/0xe0
 [<ffffffff81085a20>] ? kthread_create_on_node+0x140/0x140
 [<ffffffff815fc76c>] ret_from_fork+0x7c/0xb0
 [<ffffffff81085a20>] ? kthread_create_on_node+0x140/0x140

This patch clears PGD entries of all processes MM when
sync_global_pgds() is called after hot-removing memory

Signed-off-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Tang Chen <tangchen@cn.fujitsu.com>
Cc: Gu Zheng <guz.fnst@cn.fujitsu.com>
Cc: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86/include/asm/pgtable_64.h
arch/x86/mm/fault.c
arch/x86/mm/init_64.c

index 5be9063545d278bd978d8c14e67fb7be08cc937b..809abb335627a6735a0b6bc994c5e335369fe704 100644 (file)
@@ -115,7 +115,8 @@ static inline void native_pgd_clear(pgd_t *pgd)
        native_set_pgd(pgd, native_make_pgd(0));
 }
 
-extern void sync_global_pgds(unsigned long start, unsigned long end);
+extern void sync_global_pgds(unsigned long start, unsigned long end,
+                            int removed);
 
 /*
  * Conversion functions: convert a page and protection to a page entry,
index a241946815131904498ae6ecd95d87c4e28bf2f9..d393ac669cc0fe6419e3368624e2b94b06f411d0 100644 (file)
@@ -350,7 +350,7 @@ out:
 
 void vmalloc_sync_all(void)
 {
-       sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END);
+       sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END, 0);
 }
 
 /*
index 0e996c0a7effc84bdd0848f5dbd060a4bb5dd19e..529625118ff6f8b2bc9cdd6fe01f98fffb12c40a 100644 (file)
@@ -178,7 +178,7 @@ __setup("noexec32=", nonx32_setup);
  * When memory was added/removed make sure all the processes MM have
  * suitable PGD entries in the local PGD level page.
  */
-void sync_global_pgds(unsigned long start, unsigned long end)
+void sync_global_pgds(unsigned long start, unsigned long end, int removed)
 {
        unsigned long address;
 
@@ -186,7 +186,12 @@ void sync_global_pgds(unsigned long start, unsigned long end)
                const pgd_t *pgd_ref = pgd_offset_k(address);
                struct page *page;
 
-               if (pgd_none(*pgd_ref))
+               /*
+                * When it is called after memory hot remove, pgd_none()
+                * returns true. In this case (removed == 1), we must clear
+                * the PGD entries in the local PGD level page.
+                */
+               if (pgd_none(*pgd_ref) && !removed)
                        continue;
 
                spin_lock(&pgd_lock);
@@ -199,12 +204,18 @@ void sync_global_pgds(unsigned long start, unsigned long end)
                        pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
                        spin_lock(pgt_lock);
 
-                       if (pgd_none(*pgd))
-                               set_pgd(pgd, *pgd_ref);
-                       else
+                       if (!pgd_none(*pgd_ref) && !pgd_none(*pgd))
                                BUG_ON(pgd_page_vaddr(*pgd)
                                       != pgd_page_vaddr(*pgd_ref));
 
+                       if (removed) {
+                               if (pgd_none(*pgd_ref) && !pgd_none(*pgd))
+                                       pgd_clear(pgd);
+                       } else {
+                               if (pgd_none(*pgd))
+                                       set_pgd(pgd, *pgd_ref);
+                       }
+
                        spin_unlock(pgt_lock);
                }
                spin_unlock(&pgd_lock);
@@ -633,7 +644,7 @@ kernel_physical_mapping_init(unsigned long start,
        }
 
        if (pgd_changed)
-               sync_global_pgds(addr, end - 1);
+               sync_global_pgds(addr, end - 1, 0);
 
        __flush_tlb_all();
 
@@ -995,7 +1006,7 @@ remove_pagetable(unsigned long start, unsigned long end, bool direct)
        }
 
        if (pgd_changed)
-               sync_global_pgds(start, end - 1);
+               sync_global_pgds(start, end - 1, 1);
 
        flush_tlb_all();
 }
@@ -1342,7 +1353,7 @@ int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
        else
                err = vmemmap_populate_basepages(start, end, node);
        if (!err)
-               sync_global_pgds(start, end - 1);
+               sync_global_pgds(start, end - 1, 0);
        return err;
 }
 
This page took 0.083631 seconds and 5 git commands to generate.