2010-07-12 08:36:09 +04:00
/*
* Procedures for maintaining information about logical memory blocks .
*
* Peter Bergner , IBM Corp . June 2001.
* Copyright ( C ) 2001 Peter Bergner .
*
* This program is free software ; you can redistribute it and / or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation ; either version
* 2 of the License , or ( at your option ) any later version .
*/
# include <linux/kernel.h>
2010-07-07 02:39:13 +04:00
# include <linux/slab.h>
2010-07-12 08:36:09 +04:00
# include <linux/init.h>
# include <linux/bitops.h>
2010-07-07 02:39:07 +04:00
# include <linux/poison.h>
2010-07-07 02:39:16 +04:00
# include <linux/pfn.h>
2010-07-07 02:39:19 +04:00
# include <linux/debugfs.h>
# include <linux/seq_file.h>
2010-07-12 08:36:09 +04:00
# include <linux/memblock.h>
2013-11-13 03:07:59 +04:00
# include <asm-generic/sections.h>
2014-01-22 03:50:19 +04:00
# include <linux/io.h>
# include "internal.h"
2013-11-13 03:07:59 +04:00
2011-12-08 22:22:07 +04:00
static struct memblock_region memblock_memory_init_regions [ INIT_MEMBLOCK_REGIONS ] __initdata_memblock ;
static struct memblock_region memblock_reserved_init_regions [ INIT_MEMBLOCK_REGIONS ] __initdata_memblock ;
2014-01-29 21:16:01 +04:00
# ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
static struct memblock_region memblock_physmem_init_regions [ INIT_PHYSMEM_REGIONS ] __initdata_memblock ;
# endif
2011-12-08 22:22:07 +04:00
struct memblock memblock __initdata_memblock = {
. memory . regions = memblock_memory_init_regions ,
. memory . cnt = 1 , /* empty dummy entry */
. memory . max = INIT_MEMBLOCK_REGIONS ,
. reserved . regions = memblock_reserved_init_regions ,
. reserved . cnt = 1 , /* empty dummy entry */
. reserved . max = INIT_MEMBLOCK_REGIONS ,
2014-01-29 21:16:01 +04:00
# ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
. physmem . regions = memblock_physmem_init_regions ,
. physmem . cnt = 1 , /* empty dummy entry */
. physmem . max = INIT_PHYSMEM_REGIONS ,
# endif
2013-11-13 03:07:59 +04:00
. bottom_up = false ,
2011-12-08 22:22:07 +04:00
. current_limit = MEMBLOCK_ALLOC_ANYWHERE ,
} ;
2010-07-12 08:36:09 +04:00
2010-07-28 09:43:02 +04:00
int memblock_debug __initdata_memblock ;
2014-01-22 03:49:35 +04:00
# ifdef CONFIG_MOVABLE_NODE
bool movable_node_enabled __initdata_memblock = false ;
# endif
2015-06-25 02:58:12 +03:00
static bool system_has_some_mirror __initdata_memblock = false ;
2011-12-08 22:22:08 +04:00
static int memblock_can_resize __initdata_memblock ;
2012-05-30 02:06:50 +04:00
static int memblock_memory_in_slab __initdata_memblock = 0 ;
static int memblock_reserved_in_slab __initdata_memblock = 0 ;
2010-07-12 08:36:09 +04:00
2015-06-25 02:58:12 +03:00
ulong __init_memblock choose_memblock_flags ( void )
{
return system_has_some_mirror ? MEMBLOCK_MIRROR : MEMBLOCK_NONE ;
}
2010-07-07 02:39:13 +04:00
/* inline so we don't get a warning when pr_debug is compiled out */
2012-10-09 03:33:55 +04:00
static __init_memblock const char *
memblock_type_name ( struct memblock_type * type )
2010-07-07 02:39:13 +04:00
{
if ( type = = & memblock . memory )
return " memory " ;
else if ( type = = & memblock . reserved )
return " reserved " ;
else
return " unknown " ;
}
2011-12-08 22:22:07 +04:00
/* adjust *@size so that (@base + *@size) doesn't overflow, return new size */
static inline phys_addr_t memblock_cap_size ( phys_addr_t base , phys_addr_t * size )
{
return * size = min ( * size , ( phys_addr_t ) ULLONG_MAX - base ) ;
}
2010-07-12 08:36:48 +04:00
/*
* Address comparison utilities
*/
2010-07-28 09:43:02 +04:00
static unsigned long __init_memblock memblock_addrs_overlap ( phys_addr_t base1 , phys_addr_t size1 ,
2010-08-04 07:34:42 +04:00
phys_addr_t base2 , phys_addr_t size2 )
2010-07-12 08:36:09 +04:00
{
return ( ( base1 < ( base2 + size2 ) ) & & ( base2 < ( base1 + size1 ) ) ) ;
}
mem-hotplug: handle node hole when initializing numa_meminfo.
When parsing SRAT, all memory ranges are added into numa_meminfo. In
numa_init(), before entering numa_cleanup_meminfo(), all possible memory
ranges are in numa_meminfo. And numa_cleanup_meminfo() removes all
ranges over max_pfn or empty.
But, this only works if the nodes are continuous. Let's have a look at
the following example:
We have an SRAT like this:
SRAT: Node 0 PXM 0 [mem 0x00000000-0x5fffffff]
SRAT: Node 0 PXM 0 [mem 0x100000000-0x1ffffffffff]
SRAT: Node 1 PXM 1 [mem 0x20000000000-0x3ffffffffff]
SRAT: Node 4 PXM 2 [mem 0x40000000000-0x5ffffffffff] hotplug
SRAT: Node 5 PXM 3 [mem 0x60000000000-0x7ffffffffff] hotplug
SRAT: Node 2 PXM 4 [mem 0x80000000000-0x9ffffffffff] hotplug
SRAT: Node 3 PXM 5 [mem 0xa0000000000-0xbffffffffff] hotplug
SRAT: Node 6 PXM 6 [mem 0xc0000000000-0xdffffffffff] hotplug
SRAT: Node 7 PXM 7 [mem 0xe0000000000-0xfffffffffff] hotplug
On boot, only node 0,1,2,3 exist.
And the numa_meminfo will look like this:
numa_meminfo.nr_blks = 9
1. on node 0: [0, 60000000]
2. on node 0: [100000000, 20000000000]
3. on node 1: [20000000000, 40000000000]
4. on node 4: [40000000000, 60000000000]
5. on node 5: [60000000000, 80000000000]
6. on node 2: [80000000000, a0000000000]
7. on node 3: [a0000000000, a0800000000]
8. on node 6: [c0000000000, a0800000000]
9. on node 7: [e0000000000, a0800000000]
And numa_cleanup_meminfo() will merge 1 and 2, and remove 8,9 because the
end address is over max_pfn, which is a0800000000. But 4 and 5 are not
removed because their end addresses are less then max_pfn. But in fact,
node 4 and 5 don't exist.
In a word, numa_cleanup_meminfo() is not able to handle holes between nodes.
Since memory ranges in node 4 and 5 are in numa_meminfo, in
numa_register_memblks(), node 4 and 5 will be mistakenly set to online.
If you run lscpu, it will show:
NUMA node0 CPU(s): 0-14,128-142
NUMA node1 CPU(s): 15-29,143-157
NUMA node2 CPU(s):
NUMA node3 CPU(s):
NUMA node4 CPU(s): 62-76,190-204
NUMA node5 CPU(s): 78-92,206-220
In this patch, we use memblock_overlaps_region() to check if ranges in
numa_meminfo overlap with ranges in memory_block. Since memory_block
contains all available memory at boot time, if they overlap, it means the
ranges exist. If not, then remove them from numa_meminfo.
After this patch, lscpu will show:
NUMA node0 CPU(s): 0-14,128-142
NUMA node1 CPU(s): 15-29,143-157
NUMA node4 CPU(s): 62-76,190-204
NUMA node5 CPU(s): 78-92,206-220
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Reviewed-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Tejun Heo <tj@kernel.org>
Cc: Luiz Capitulino <lcapitulino@redhat.com>
Cc: Xishi Qiu <qiuxishi@huawei.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Vladimir Murzin <vladimir.murzin@arm.com>
Cc: Fabian Frederick <fabf@skynet.be>
Cc: Alexander Kuleshov <kuleshovmail@gmail.com>
Cc: Baoquan He <bhe@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-09-09 01:02:03 +03:00
bool __init_memblock memblock_overlaps_region ( struct memblock_type * type ,
2011-11-01 04:09:15 +04:00
phys_addr_t base , phys_addr_t size )
2010-07-12 08:36:48 +04:00
{
unsigned long i ;
2016-01-15 02:20:39 +03:00
for ( i = 0 ; i < type - > cnt ; i + + )
if ( memblock_addrs_overlap ( base , size , type - > regions [ i ] . base ,
type - > regions [ i ] . size ) )
2010-07-12 08:36:48 +04:00
break ;
2015-09-09 01:02:00 +03:00
return i < type - > cnt ;
2010-07-12 08:36:48 +04:00
}
2013-11-13 03:07:59 +04:00
/*
* __memblock_find_range_bottom_up - find free area utility in bottom - up
* @ start : start of candidate range
* @ end : end of candidate range , can be % MEMBLOCK_ALLOC_ { ANYWHERE | ACCESSIBLE }
* @ size : size of free area to find
* @ align : alignment of free area to find
2014-01-22 03:50:16 +04:00
* @ nid : nid of the free area to find , % NUMA_NO_NODE for any node
2015-06-25 02:58:09 +03:00
* @ flags : pick from blocks based on memory attributes
2013-11-13 03:07:59 +04:00
*
* Utility called from memblock_find_in_range_node ( ) , find free area bottom - up .
*
* RETURNS :
* Found address on success , 0 on failure .
*/
static phys_addr_t __init_memblock
__memblock_find_range_bottom_up ( phys_addr_t start , phys_addr_t end ,
2015-06-25 02:58:09 +03:00
phys_addr_t size , phys_addr_t align , int nid ,
ulong flags )
2013-11-13 03:07:59 +04:00
{
phys_addr_t this_start , this_end , cand ;
u64 i ;
2015-06-25 02:58:09 +03:00
for_each_free_mem_range ( i , nid , flags , & this_start , & this_end , NULL ) {
2013-11-13 03:07:59 +04:00
this_start = clamp ( this_start , start , end ) ;
this_end = clamp ( this_end , start , end ) ;
cand = round_up ( this_start , align ) ;
if ( cand < this_end & & this_end - cand > = size )
return cand ;
}
return 0 ;
}
2011-12-08 22:22:09 +04:00
/**
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
* __memblock_find_range_top_down - find free area utility , in top - down
2011-12-08 22:22:09 +04:00
* @ start : start of candidate range
* @ end : end of candidate range , can be % MEMBLOCK_ALLOC_ { ANYWHERE | ACCESSIBLE }
* @ size : size of free area to find
* @ align : alignment of free area to find
2014-01-22 03:50:16 +04:00
* @ nid : nid of the free area to find , % NUMA_NO_NODE for any node
2015-06-25 02:58:09 +03:00
* @ flags : pick from blocks based on memory attributes
2011-12-08 22:22:09 +04:00
*
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
* Utility called from memblock_find_in_range_node ( ) , find free area top - down .
2011-12-08 22:22:09 +04:00
*
* RETURNS :
2013-11-13 03:07:59 +04:00
* Found address on success , 0 on failure .
2010-07-12 08:36:48 +04:00
*/
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
static phys_addr_t __init_memblock
__memblock_find_range_top_down ( phys_addr_t start , phys_addr_t end ,
2015-06-25 02:58:09 +03:00
phys_addr_t size , phys_addr_t align , int nid ,
ulong flags )
2013-02-23 04:33:51 +04:00
{
phys_addr_t this_start , this_end , cand ;
u64 i ;
2015-06-25 02:58:09 +03:00
for_each_free_mem_range_reverse ( i , nid , flags , & this_start , & this_end ,
NULL ) {
2013-02-23 04:33:51 +04:00
this_start = clamp ( this_start , start , end ) ;
this_end = clamp ( this_end , start , end ) ;
if ( this_end < size )
continue ;
cand = round_down ( this_end - size , align ) ;
if ( cand > = this_start )
return cand ;
}
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
2013-02-23 04:33:51 +04:00
return 0 ;
}
2010-07-12 08:36:48 +04:00
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
/**
* memblock_find_in_range_node - find free area in given range and node
* @ size : size of free area to find
* @ align : alignment of free area to find
2014-01-22 03:50:14 +04:00
* @ start : start of candidate range
* @ end : end of candidate range , can be % MEMBLOCK_ALLOC_ { ANYWHERE | ACCESSIBLE }
2014-01-22 03:50:16 +04:00
* @ nid : nid of the free area to find , % NUMA_NO_NODE for any node
2015-06-25 02:58:09 +03:00
* @ flags : pick from blocks based on memory attributes
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
*
* Find @ size free area aligned to @ align in the specified range and node .
*
2013-11-13 03:07:59 +04:00
* When allocation direction is bottom - up , the @ start should be greater
* than the end of the kernel image . Otherwise , it will be trimmed . The
* reason is that we want the bottom - up allocation just near the kernel
* image so it is highly likely that the allocated memory and the kernel
* will reside in the same node .
*
* If bottom - up allocation failed , will try to allocate memory top - down .
*
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
* RETURNS :
2013-11-13 03:07:59 +04:00
* Found address on success , 0 on failure .
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
*/
2014-01-22 03:50:14 +04:00
phys_addr_t __init_memblock memblock_find_in_range_node ( phys_addr_t size ,
phys_addr_t align , phys_addr_t start ,
2015-06-25 02:58:09 +03:00
phys_addr_t end , int nid , ulong flags )
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
{
2014-08-30 02:18:31 +04:00
phys_addr_t kernel_end , ret ;
2013-11-13 03:07:59 +04:00
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
/* pump up @end */
if ( end = = MEMBLOCK_ALLOC_ACCESSIBLE )
end = memblock . current_limit ;
/* avoid allocating the first page */
start = max_t ( phys_addr_t , start , PAGE_SIZE ) ;
end = max ( start , end ) ;
2013-11-13 03:07:59 +04:00
kernel_end = __pa_symbol ( _end ) ;
/*
* try bottom - up allocation only when bottom - up mode
* is set and @ end is above the kernel image .
*/
if ( memblock_bottom_up ( ) & & end > kernel_end ) {
phys_addr_t bottom_up_start ;
/* make sure we will allocate above the kernel */
bottom_up_start = max ( start , kernel_end ) ;
/* ok, try bottom-up allocation first */
ret = __memblock_find_range_bottom_up ( bottom_up_start , end ,
2015-06-25 02:58:09 +03:00
size , align , nid , flags ) ;
2013-11-13 03:07:59 +04:00
if ( ret )
return ret ;
/*
* we always limit bottom - up allocation above the kernel ,
* but top - down allocation doesn ' t have the limit , so
* retrying top - down allocation may succeed when bottom - up
* allocation failed .
*
* bottom - up allocation is expected to be fail very rarely ,
* so we use WARN_ONCE ( ) here to see the stack trace if
* fail happens .
*/
WARN_ONCE ( 1 , " memblock: bottom-up allocation failed, "
" memory hotunplug may be affected \n " ) ;
}
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
2015-06-25 02:58:09 +03:00
return __memblock_find_range_top_down ( start , end , size , align , nid ,
flags ) ;
mm/memblock.c: factor out of top-down allocation
[Problem]
The current Linux cannot migrate pages used by the kernel because of the
kernel direct mapping. In Linux kernel space, va = pa + PAGE_OFFSET.
When the pa is changed, we cannot simply update the pagetable and keep the
va unmodified. So the kernel pages are not migratable.
There are also some other issues will cause the kernel pages not
migratable. For example, the physical address may be cached somewhere and
will be used. It is not to update all the caches.
When doing memory hotplug in Linux, we first migrate all the pages in one
memory device somewhere else, and then remove the device. But if pages
are used by the kernel, they are not migratable. As a result, memory used
by the kernel cannot be hot-removed.
Modifying the kernel direct mapping mechanism is too difficult to do. And
it may cause the kernel performance down and unstable. So we use the
following way to do memory hotplug.
[What we are doing]
In Linux, memory in one numa node is divided into several zones. One of
the zones is ZONE_MOVABLE, which the kernel won't use.
In order to implement memory hotplug in Linux, we are going to arrange all
hotpluggable memory in ZONE_MOVABLE so that the kernel won't use these
memory. To do this, we need ACPI's help.
In ACPI, SRAT(System Resource Affinity Table) contains NUMA info. The
memory affinities in SRAT record every memory range in the system, and
also, flags specifying if the memory range is hotpluggable. (Please refer
to ACPI spec 5.0 5.2.16)
With the help of SRAT, we have to do the following two things to achieve our
goal:
1. When doing memory hot-add, allow the users arranging hotpluggable as
ZONE_MOVABLE.
(This has been done by the MOVABLE_NODE functionality in Linux.)
2. when the system is booting, prevent bootmem allocator from allocating
hotpluggable memory for the kernel before the memory initialization
finishes.
The problem 2 is the key problem we are going to solve. But before solving it,
we need some preparation. Please see below.
[Preparation]
Bootloader has to load the kernel image into memory. And this memory must
be unhotpluggable. We cannot prevent this anyway. So in a memory hotplug
system, we can assume any node the kernel resides in is not hotpluggable.
Before SRAT is parsed, we don't know which memory ranges are hotpluggable.
But memblock has already started to work. In the current kernel,
memblock allocates the following memory before SRAT is parsed:
setup_arch()
|->memblock_x86_fill() /* memblock is ready */
|......
|->early_reserve_e820_mpc_new() /* allocate memory under 1MB */
|->reserve_real_mode() /* allocate memory under 1MB */
|->init_mem_mapping() /* allocate page tables, about 2MB to map 1GB memory */
|->dma_contiguous_reserve() /* specified by user, should be low */
|->setup_log_buf() /* specified by user, several mega bytes */
|->relocate_initrd() /* could be large, but will be freed after boot, should reorder */
|->acpi_initrd_override() /* several mega bytes */
|->reserve_crashkernel() /* could be large, should reorder */
|......
|->initmem_init() /* Parse SRAT */
According to Tejun's advice, before SRAT is parsed, we should try our best
to allocate memory near the kernel image. Since the whole node the kernel
resides in won't be hotpluggable, and for a modern server, a node may have
at least 16GB memory, allocating several mega bytes memory around the
kernel image won't cross to hotpluggable memory.
[About this patchset]
So this patchset is the preparation for the problem 2 that we want to
solve. It does the following:
1. Make memblock be able to allocate memory bottom up.
1) Keep all the memblock APIs' prototype unmodified.
2) When the direction is bottom up, keep the start address greater than the
end of kernel image.
2. Improve init_mem_mapping() to support allocate page tables in
bottom up direction.
3. Introduce "movable_node" boot option to enable and disable this
functionality.
This patch (of 6):
Create a new function __memblock_find_range_top_down to factor out of
top-down allocation from memblock_find_in_range_node. This is a
preparation because we will introduce a new bottom-up allocation mode in
the following patch.
Signed-off-by: Tang Chen <tangchen@cn.fujitsu.com>
Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Toshi Kani <toshi.kani@hp.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Wanpeng Li <liwanp@linux.vnet.ibm.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: Jiang Liu <jiang.liu@huawei.com>
Cc: Wen Congyang <wency@cn.fujitsu.com>
Cc: Lai Jiangshan <laijs@cn.fujitsu.com>
Cc: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Cc: Taku Izumi <izumi.taku@jp.fujitsu.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michal Nazarewicz <mina86@mina86.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-11-13 03:07:57 +04:00
}
2011-12-08 22:22:09 +04:00
/**
* memblock_find_in_range - find free area in given range
* @ start : start of candidate range
* @ end : end of candidate range , can be % MEMBLOCK_ALLOC_ { ANYWHERE | ACCESSIBLE }
* @ size : size of free area to find
* @ align : alignment of free area to find
*
* Find @ size free area aligned to @ align in the specified range .
*
* RETURNS :
2013-11-13 03:07:59 +04:00
* Found address on success , 0 on failure .
2011-07-12 11:58:10 +04:00
*/
2011-12-08 22:22:09 +04:00
phys_addr_t __init_memblock memblock_find_in_range ( phys_addr_t start ,
phys_addr_t end , phys_addr_t size ,
phys_addr_t align )
2010-07-12 08:36:48 +04:00
{
2015-06-25 02:58:12 +03:00
phys_addr_t ret ;
ulong flags = choose_memblock_flags ( ) ;
again :
ret = memblock_find_in_range_node ( size , align , start , end ,
NUMA_NO_NODE , flags ) ;
if ( ! ret & & ( flags & MEMBLOCK_MIRROR ) ) {
pr_warn ( " Could not allocate %pap bytes of mirrored memory \n " ,
& size ) ;
flags & = ~ MEMBLOCK_MIRROR ;
goto again ;
}
return ret ;
2010-07-12 08:36:48 +04:00
}
2010-07-28 09:43:02 +04:00
static void __init_memblock memblock_remove_region ( struct memblock_type * type , unsigned long r )
2010-07-12 08:36:09 +04:00
{
2011-12-08 22:22:08 +04:00
type - > total_size - = type - > regions [ r ] . size ;
2011-07-14 13:43:42 +04:00
memmove ( & type - > regions [ r ] , & type - > regions [ r + 1 ] ,
( type - > cnt - ( r + 1 ) ) * sizeof ( type - > regions [ r ] ) ) ;
2010-08-04 08:06:41 +04:00
type - > cnt - - ;
2010-07-12 08:36:09 +04:00
mm/memblock: properly handle overlaps and fix error path
Currently memblock_reserve() or memblock_free() don't handle overlaps of
any kind. There is some special casing for coalescing exactly adjacent
regions but that's about it.
This is annoying because typically memblock_reserve() is used to mark
regions passed by the firmware as reserved and we all know how much we can
trust our firmwares...
Also, with the current code, if we do something it doesn't handle right
such as trying to memblock_reserve() a large range spanning multiple
existing smaller reserved regions for example, or doing overlapping
reservations, it can silently corrupt the internal region array, causing
odd errors much later on, such as allocations returning reserved regions
etc...
This patch rewrites the underlying functions that add or remove a region
to the arrays. The new code is a lot more robust as it fully handles
overlapping regions. It's also, imho, simpler than the previous
implementation.
In addition, while doing so, I found a bug where if we fail to double the
array while adding a region, we would remove the last region of the array
rather than the region we just allocated. This fixes it too.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Yinghai Lu <yinghai@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-23 02:33:43 +03:00
/* Special case for empty arrays */
if ( type - > cnt = = 0 ) {
2011-12-08 22:22:08 +04:00
WARN_ON ( type - > total_size ! = 0 ) ;
mm/memblock: properly handle overlaps and fix error path
Currently memblock_reserve() or memblock_free() don't handle overlaps of
any kind. There is some special casing for coalescing exactly adjacent
regions but that's about it.
This is annoying because typically memblock_reserve() is used to mark
regions passed by the firmware as reserved and we all know how much we can
trust our firmwares...
Also, with the current code, if we do something it doesn't handle right
such as trying to memblock_reserve() a large range spanning multiple
existing smaller reserved regions for example, or doing overlapping
reservations, it can silently corrupt the internal region array, causing
odd errors much later on, such as allocations returning reserved regions
etc...
This patch rewrites the underlying functions that add or remove a region
to the arrays. The new code is a lot more robust as it fully handles
overlapping regions. It's also, imho, simpler than the previous
implementation.
In addition, while doing so, I found a bug where if we fail to double the
array while adding a region, we would remove the last region of the array
rather than the region we just allocated. This fixes it too.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Yinghai Lu <yinghai@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-23 02:33:43 +03:00
type - > cnt = 1 ;
type - > regions [ 0 ] . base = 0 ;
type - > regions [ 0 ] . size = 0 ;
2014-01-22 03:49:20 +04:00
type - > regions [ 0 ] . flags = 0 ;
2011-07-14 13:43:42 +04:00
memblock_set_region_node ( & type - > regions [ 0 ] , MAX_NUMNODES ) ;
mm/memblock: properly handle overlaps and fix error path
Currently memblock_reserve() or memblock_free() don't handle overlaps of
any kind. There is some special casing for coalescing exactly adjacent
regions but that's about it.
This is annoying because typically memblock_reserve() is used to mark
regions passed by the firmware as reserved and we all know how much we can
trust our firmwares...
Also, with the current code, if we do something it doesn't handle right
such as trying to memblock_reserve() a large range spanning multiple
existing smaller reserved regions for example, or doing overlapping
reservations, it can silently corrupt the internal region array, causing
odd errors much later on, such as allocations returning reserved regions
etc...
This patch rewrites the underlying functions that add or remove a region
to the arrays. The new code is a lot more robust as it fully handles
overlapping regions. It's also, imho, simpler than the previous
implementation.
In addition, while doing so, I found a bug where if we fail to double the
array while adding a region, we would remove the last region of the array
rather than the region we just allocated. This fixes it too.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Yinghai Lu <yinghai@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-23 02:33:43 +03:00
}
2010-07-12 08:36:09 +04:00
}
2014-01-24 03:53:24 +04:00
# ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
2012-07-12 01:02:56 +04:00
phys_addr_t __init_memblock get_allocated_memblock_reserved_regions_info (
phys_addr_t * addr )
{
if ( memblock . reserved . regions = = memblock_reserved_init_regions )
return 0 ;
* addr = __pa ( memblock . reserved . regions ) ;
return PAGE_ALIGN ( sizeof ( struct memblock_region ) *
memblock . reserved . max ) ;
}
2014-01-24 03:53:11 +04:00
phys_addr_t __init_memblock get_allocated_memblock_memory_regions_info (
phys_addr_t * addr )
{
if ( memblock . memory . regions = = memblock_memory_init_regions )
return 0 ;
* addr = __pa ( memblock . memory . regions ) ;
return PAGE_ALIGN ( sizeof ( struct memblock_region ) *
memblock . memory . max ) ;
}
# endif
2012-06-20 23:53:05 +04:00
/**
* memblock_double_array - double the size of the memblock regions array
* @ type : memblock type of the regions array being doubled
* @ new_area_start : starting address of memory range to avoid overlap with
* @ new_area_size : size of memory range to avoid overlap with
*
* Double the size of the @ type regions array . If memblock is being used to
* allocate memory for a new reserved regions array and there is a previously
* allocated memory range [ @ new_area_start , @ new_area_start + @ new_area_size ]
* waiting to be reserved , ensure the memory used by the new array does
* not overlap .
*
* RETURNS :
* 0 on success , - 1 on failure .
*/
static int __init_memblock memblock_double_array ( struct memblock_type * type ,
phys_addr_t new_area_start ,
phys_addr_t new_area_size )
2010-07-07 02:39:13 +04:00
{
struct memblock_region * new_array , * old_array ;
2012-07-12 01:02:56 +04:00
phys_addr_t old_alloc_size , new_alloc_size ;
2010-07-07 02:39:13 +04:00
phys_addr_t old_size , new_size , addr ;
int use_slab = slab_is_available ( ) ;
2012-05-30 02:06:50 +04:00
int * in_slab ;
2010-07-07 02:39:13 +04:00
/* We don't allow resizing until we know about the reserved regions
* of memory that aren ' t suitable for allocation
*/
if ( ! memblock_can_resize )
return - 1 ;
/* Calculate new doubled size */
old_size = type - > max * sizeof ( struct memblock_region ) ;
new_size = old_size < < 1 ;
2012-07-12 01:02:56 +04:00
/*
* We need to allocated new one align to PAGE_SIZE ,
* so we can free them completely later .
*/
old_alloc_size = PAGE_ALIGN ( old_size ) ;
new_alloc_size = PAGE_ALIGN ( new_size ) ;
2010-07-07 02:39:13 +04:00
2012-05-30 02:06:50 +04:00
/* Retrieve the slab flag */
if ( type = = & memblock . memory )
in_slab = & memblock_memory_in_slab ;
else
in_slab = & memblock_reserved_in_slab ;
2010-07-07 02:39:13 +04:00
/* Try to find some space for it.
*
* WARNING : We assume that either slab_is_available ( ) and we use it or
2012-08-01 03:42:40 +04:00
* we use MEMBLOCK for allocations . That means that this is unsafe to
* use when bootmem is currently active ( unless bootmem itself is
* implemented on top of MEMBLOCK which isn ' t the case yet )
2010-07-07 02:39:13 +04:00
*
* This should however not be an issue for now , as we currently only
2012-08-01 03:42:40 +04:00
* call into MEMBLOCK while it ' s still active , or much later when slab
* is active for memory hotplug operations
2010-07-07 02:39:13 +04:00
*/
if ( use_slab ) {
new_array = kmalloc ( new_size , GFP_KERNEL ) ;
2011-07-12 11:58:09 +04:00
addr = new_array ? __pa ( new_array ) : 0 ;
2012-05-30 02:06:50 +04:00
} else {
2012-06-20 23:53:05 +04:00
/* only exclude range when trying to double reserved.regions */
if ( type ! = & memblock . reserved )
new_area_start = new_area_size = 0 ;
addr = memblock_find_in_range ( new_area_start + new_area_size ,
memblock . current_limit ,
2012-07-12 01:02:56 +04:00
new_alloc_size , PAGE_SIZE ) ;
2012-06-20 23:53:05 +04:00
if ( ! addr & & new_area_size )
addr = memblock_find_in_range ( 0 ,
2012-08-01 03:42:40 +04:00
min ( new_area_start , memblock . current_limit ) ,
new_alloc_size , PAGE_SIZE ) ;
2012-06-20 23:53:05 +04:00
2012-09-04 12:25:05 +04:00
new_array = addr ? __va ( addr ) : NULL ;
2012-05-30 02:06:50 +04:00
}
2011-07-12 11:58:09 +04:00
if ( ! addr ) {
2010-07-07 02:39:13 +04:00
pr_err ( " memblock: Failed to double %s array from %ld to %ld entries ! \n " ,
memblock_type_name ( type ) , type - > max , type - > max * 2 ) ;
return - 1 ;
}
2012-08-01 03:42:40 +04:00
memblock_dbg ( " memblock: %s is doubled to %ld at [%#010llx-%#010llx] " ,
memblock_type_name ( type ) , type - > max * 2 , ( u64 ) addr ,
( u64 ) addr + new_size - 1 ) ;
2010-07-28 09:13:22 +04:00
2012-08-01 03:42:40 +04:00
/*
* Found space , we now need to move the array over before we add the
* reserved region since it may be our reserved array itself that is
* full .
2010-07-07 02:39:13 +04:00
*/
memcpy ( new_array , type - > regions , old_size ) ;
memset ( new_array + type - > max , 0 , old_size ) ;
old_array = type - > regions ;
type - > regions = new_array ;
type - > max < < = 1 ;
2012-08-01 03:42:40 +04:00
/* Free old array. We needn't free it if the array is the static one */
2012-05-30 02:06:50 +04:00
if ( * in_slab )
kfree ( old_array ) ;
else if ( old_array ! = memblock_memory_init_regions & &
old_array ! = memblock_reserved_init_regions )
2012-07-12 01:02:56 +04:00
memblock_free ( __pa ( old_array ) , old_alloc_size ) ;
2010-07-07 02:39:13 +04:00
2012-08-01 03:42:40 +04:00
/*
* Reserve the new array if that comes from the memblock . Otherwise , we
* needn ' t do it
2012-05-30 02:06:50 +04:00
*/
if ( ! use_slab )
2012-07-12 01:02:56 +04:00
BUG_ON ( memblock_reserve ( addr , new_alloc_size ) ) ;
2012-05-30 02:06:50 +04:00
/* Update slab flag */
* in_slab = use_slab ;
2010-07-07 02:39:13 +04:00
return 0 ;
}
2011-07-12 13:15:55 +04:00
/**
* memblock_merge_regions - merge neighboring compatible regions
* @ type : memblock type to scan
*
* Scan @ type and merge neighboring compatible regions .
*/
static void __init_memblock memblock_merge_regions ( struct memblock_type * type )
2010-07-12 08:36:09 +04:00
{
2011-07-12 13:15:55 +04:00
int i = 0 ;
2010-07-12 08:36:09 +04:00
2011-07-12 13:15:55 +04:00
/* cnt never goes below 1 */
while ( i < type - > cnt - 1 ) {
struct memblock_region * this = & type - > regions [ i ] ;
struct memblock_region * next = & type - > regions [ i + 1 ] ;
2010-07-12 08:36:09 +04:00
2011-07-14 13:43:42 +04:00
if ( this - > base + this - > size ! = next - > base | |
memblock_get_region_node ( this ) ! =
2014-01-22 03:49:20 +04:00
memblock_get_region_node ( next ) | |
this - > flags ! = next - > flags ) {
2011-07-12 13:15:55 +04:00
BUG_ON ( this - > base + this - > size > next - > base ) ;
i + + ;
continue ;
mm/memblock: properly handle overlaps and fix error path
Currently memblock_reserve() or memblock_free() don't handle overlaps of
any kind. There is some special casing for coalescing exactly adjacent
regions but that's about it.
This is annoying because typically memblock_reserve() is used to mark
regions passed by the firmware as reserved and we all know how much we can
trust our firmwares...
Also, with the current code, if we do something it doesn't handle right
such as trying to memblock_reserve() a large range spanning multiple
existing smaller reserved regions for example, or doing overlapping
reservations, it can silently corrupt the internal region array, causing
odd errors much later on, such as allocations returning reserved regions
etc...
This patch rewrites the underlying functions that add or remove a region
to the arrays. The new code is a lot more robust as it fully handles
overlapping regions. It's also, imho, simpler than the previous
implementation.
In addition, while doing so, I found a bug where if we fail to double the
array while adding a region, we would remove the last region of the array
rather than the region we just allocated. This fixes it too.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Yinghai Lu <yinghai@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-23 02:33:43 +03:00
}
2011-07-12 13:15:55 +04:00
this - > size + = next - > size ;
2013-01-12 02:31:44 +04:00
/* move forward from next + 1, index of which is i + 2 */
memmove ( next , next + 1 , ( type - > cnt - ( i + 2 ) ) * sizeof ( * next ) ) ;
2011-07-12 13:15:55 +04:00
type - > cnt - - ;
2010-07-12 08:36:09 +04:00
}
2011-07-12 13:15:55 +04:00
}
2010-07-12 08:36:09 +04:00
2011-07-12 13:15:55 +04:00
/**
* memblock_insert_region - insert new memblock region
2013-04-30 02:08:41 +04:00
* @ type : memblock type to insert into
* @ idx : index for the insertion point
* @ base : base address of the new region
* @ size : size of the new region
* @ nid : node id of the new region
2014-01-22 03:49:20 +04:00
* @ flags : flags of the new region
2011-07-12 13:15:55 +04:00
*
* Insert new memblock region [ @ base , @ base + @ size ) into @ type at @ idx .
* @ type must already have extra room to accomodate the new region .
*/
static void __init_memblock memblock_insert_region ( struct memblock_type * type ,
int idx , phys_addr_t base ,
2014-01-22 03:49:20 +04:00
phys_addr_t size ,
int nid , unsigned long flags )
2011-07-12 13:15:55 +04:00
{
struct memblock_region * rgn = & type - > regions [ idx ] ;
BUG_ON ( type - > cnt > = type - > max ) ;
memmove ( rgn + 1 , rgn , ( type - > cnt - idx ) * sizeof ( * rgn ) ) ;
rgn - > base = base ;
rgn - > size = size ;
2014-01-22 03:49:20 +04:00
rgn - > flags = flags ;
2011-07-14 13:43:42 +04:00
memblock_set_region_node ( rgn , nid ) ;
2011-07-12 13:15:55 +04:00
type - > cnt + + ;
2011-12-08 22:22:08 +04:00
type - > total_size + = size ;
2011-07-12 13:15:55 +04:00
}
/**
2014-01-29 21:16:01 +04:00
* memblock_add_range - add new memblock region
2011-07-12 13:15:55 +04:00
* @ type : memblock type to add new region into
* @ base : base address of the new region
* @ size : size of the new region
2011-12-08 22:22:08 +04:00
* @ nid : nid of the new region
2014-01-22 03:49:20 +04:00
* @ flags : flags of the new region
2011-07-12 13:15:55 +04:00
*
* Add new memblock region [ @ base , @ base + @ size ) into @ type . The new region
* is allowed to overlap with existing ones - overlaps don ' t affect already
* existing regions . @ type is guaranteed to be minimal ( all neighbouring
* compatible regions are merged ) after the addition .
*
* RETURNS :
* 0 on success , - errno on failure .
*/
2014-01-29 21:16:01 +04:00
int __init_memblock memblock_add_range ( struct memblock_type * type ,
2014-01-22 03:49:20 +04:00
phys_addr_t base , phys_addr_t size ,
int nid , unsigned long flags )
2011-07-12 13:15:55 +04:00
{
bool insert = false ;
2011-12-08 22:22:07 +04:00
phys_addr_t obase = base ;
phys_addr_t end = base + memblock_cap_size ( base , & size ) ;
2016-01-15 02:20:42 +03:00
int idx , nr_new ;
struct memblock_region * rgn ;
2011-07-12 13:15:55 +04:00
2012-04-20 19:31:34 +04:00
if ( ! size )
return 0 ;
2011-07-12 13:15:55 +04:00
/* special case for empty array */
if ( type - > regions [ 0 ] . size = = 0 ) {
2011-12-08 22:22:08 +04:00
WARN_ON ( type - > cnt ! = 1 | | type - > total_size ) ;
mm/memblock: properly handle overlaps and fix error path
Currently memblock_reserve() or memblock_free() don't handle overlaps of
any kind. There is some special casing for coalescing exactly adjacent
regions but that's about it.
This is annoying because typically memblock_reserve() is used to mark
regions passed by the firmware as reserved and we all know how much we can
trust our firmwares...
Also, with the current code, if we do something it doesn't handle right
such as trying to memblock_reserve() a large range spanning multiple
existing smaller reserved regions for example, or doing overlapping
reservations, it can silently corrupt the internal region array, causing
odd errors much later on, such as allocations returning reserved regions
etc...
This patch rewrites the underlying functions that add or remove a region
to the arrays. The new code is a lot more robust as it fully handles
overlapping regions. It's also, imho, simpler than the previous
implementation.
In addition, while doing so, I found a bug where if we fail to double the
array while adding a region, we would remove the last region of the array
rather than the region we just allocated. This fixes it too.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Yinghai Lu <yinghai@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-23 02:33:43 +03:00
type - > regions [ 0 ] . base = base ;
type - > regions [ 0 ] . size = size ;
2014-01-22 03:49:20 +04:00
type - > regions [ 0 ] . flags = flags ;
2011-12-08 22:22:08 +04:00
memblock_set_region_node ( & type - > regions [ 0 ] , nid ) ;
2011-12-08 22:22:08 +04:00
type - > total_size = size ;
mm/memblock: properly handle overlaps and fix error path
Currently memblock_reserve() or memblock_free() don't handle overlaps of
any kind. There is some special casing for coalescing exactly adjacent
regions but that's about it.
This is annoying because typically memblock_reserve() is used to mark
regions passed by the firmware as reserved and we all know how much we can
trust our firmwares...
Also, with the current code, if we do something it doesn't handle right
such as trying to memblock_reserve() a large range spanning multiple
existing smaller reserved regions for example, or doing overlapping
reservations, it can silently corrupt the internal region array, causing
odd errors much later on, such as allocations returning reserved regions
etc...
This patch rewrites the underlying functions that add or remove a region
to the arrays. The new code is a lot more robust as it fully handles
overlapping regions. It's also, imho, simpler than the previous
implementation.
In addition, while doing so, I found a bug where if we fail to double the
array while adding a region, we would remove the last region of the array
rather than the region we just allocated. This fixes it too.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Yinghai Lu <yinghai@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-23 02:33:43 +03:00
return 0 ;
2010-07-12 08:36:09 +04:00
}
2011-07-12 13:15:55 +04:00
repeat :
/*
* The following is executed twice . Once with % false @ insert and
* then with % true . The first counts the number of regions needed
* to accomodate the new area . The second actually inserts them .
2010-07-07 02:39:13 +04:00
*/
2011-07-12 13:15:55 +04:00
base = obase ;
nr_new = 0 ;
2010-07-12 08:36:09 +04:00
2016-01-15 02:20:42 +03:00
for_each_memblock_type ( type , rgn ) {
2011-07-12 13:15:55 +04:00
phys_addr_t rbase = rgn - > base ;
phys_addr_t rend = rbase + rgn - > size ;
if ( rbase > = end )
2010-07-12 08:36:09 +04:00
break ;
2011-07-12 13:15:55 +04:00
if ( rend < = base )
continue ;
/*
* @ rgn overlaps . If it separates the lower part of new
* area , insert that portion .
*/
if ( rbase > base ) {
2015-09-05 01:47:38 +03:00
# ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
WARN_ON ( nid ! = memblock_get_region_node ( rgn ) ) ;
# endif
2015-09-09 00:59:53 +03:00
WARN_ON ( flags ! = rgn - > flags ) ;
2011-07-12 13:15:55 +04:00
nr_new + + ;
if ( insert )
2016-01-15 02:20:42 +03:00
memblock_insert_region ( type , idx + + , base ,
2014-01-22 03:49:20 +04:00
rbase - base , nid ,
flags ) ;
2010-07-12 08:36:09 +04:00
}
2011-07-12 13:15:55 +04:00
/* area below @rend is dealt with, forget about it */
base = min ( rend , end ) ;
2010-07-12 08:36:09 +04:00
}
2011-07-12 13:15:55 +04:00
/* insert the remaining portion */
if ( base < end ) {
nr_new + + ;
if ( insert )
2016-01-15 02:20:42 +03:00
memblock_insert_region ( type , idx , base , end - base ,
2014-01-22 03:49:20 +04:00
nid , flags ) ;
2010-07-12 08:36:09 +04:00
}
2011-07-12 13:15:55 +04:00
/*
* If this was the first round , resize array and repeat for actual
* insertions ; otherwise , merge and return .
2010-07-07 02:39:13 +04:00
*/
2011-07-12 13:15:55 +04:00
if ( ! insert ) {
while ( type - > cnt + nr_new > type - > max )
2012-06-20 23:53:05 +04:00
if ( memblock_double_array ( type , obase , size ) < 0 )
2011-07-12 13:15:55 +04:00
return - ENOMEM ;
insert = true ;
goto repeat ;
} else {
memblock_merge_regions ( type ) ;
return 0 ;
2010-07-07 02:39:13 +04:00
}
2010-07-12 08:36:09 +04:00
}
2011-12-08 22:22:08 +04:00
int __init_memblock memblock_add_node ( phys_addr_t base , phys_addr_t size ,
int nid )
{
2014-01-29 21:16:01 +04:00
return memblock_add_range ( & memblock . memory , base , size , nid , 0 ) ;
2011-12-08 22:22:08 +04:00
}
2015-04-16 02:14:44 +03:00
static int __init_memblock memblock_add_region ( phys_addr_t base ,
phys_addr_t size ,
int nid ,
unsigned long flags )
{
memblock_dbg ( " memblock_add: [%#016llx-%#016llx] flags %#02lx %pF \n " ,
( unsigned long long ) base ,
( unsigned long long ) base + size - 1 ,
flags , ( void * ) _RET_IP_ ) ;
2016-03-16 00:57:42 +03:00
return memblock_add_range ( & memblock . memory , base , size , nid , flags ) ;
2015-04-16 02:14:44 +03:00
}
2011-12-08 22:22:06 +04:00
int __init_memblock memblock_add ( phys_addr_t base , phys_addr_t size )
2010-07-12 08:36:09 +04:00
{
2015-04-16 02:14:44 +03:00
return memblock_add_region ( base , size , MAX_NUMNODES , 0 ) ;
2010-07-12 08:36:09 +04:00
}
2011-12-08 22:22:07 +04:00
/**
* memblock_isolate_range - isolate given range into disjoint memblocks
* @ type : memblock type to isolate range for
* @ base : base of range to isolate
* @ size : size of range to isolate
* @ start_rgn : out parameter for the start of isolated region
* @ end_rgn : out parameter for the end of isolated region
*
* Walk @ type and ensure that regions don ' t cross the boundaries defined by
* [ @ base , @ base + @ size ) . Crossing regions are split at the boundaries ,
* which may create at most two more regions . The index of the first
* region inside the range is returned in * @ start_rgn and end in * @ end_rgn .
*
* RETURNS :
* 0 on success , - errno on failure .
*/
static int __init_memblock memblock_isolate_range ( struct memblock_type * type ,
phys_addr_t base , phys_addr_t size ,
int * start_rgn , int * end_rgn )
{
2011-12-08 22:22:07 +04:00
phys_addr_t end = base + memblock_cap_size ( base , & size ) ;
2016-01-15 02:20:42 +03:00
int idx ;
struct memblock_region * rgn ;
2011-12-08 22:22:07 +04:00
* start_rgn = * end_rgn = 0 ;
2012-04-20 19:31:34 +04:00
if ( ! size )
return 0 ;
2011-12-08 22:22:07 +04:00
/* we'll create at most two more regions */
while ( type - > cnt + 2 > type - > max )
2012-06-20 23:53:05 +04:00
if ( memblock_double_array ( type , base , size ) < 0 )
2011-12-08 22:22:07 +04:00
return - ENOMEM ;
2016-01-15 02:20:42 +03:00
for_each_memblock_type ( type , rgn ) {
2011-12-08 22:22:07 +04:00
phys_addr_t rbase = rgn - > base ;
phys_addr_t rend = rbase + rgn - > size ;
if ( rbase > = end )
break ;
if ( rend < = base )
continue ;
if ( rbase < base ) {
/*
* @ rgn intersects from below . Split and continue
* to process the next region - the new top half .
*/
rgn - > base = base ;
2011-12-08 22:22:08 +04:00
rgn - > size - = base - rbase ;
type - > total_size - = base - rbase ;
2016-01-15 02:20:42 +03:00
memblock_insert_region ( type , idx , rbase , base - rbase ,
2014-01-22 03:49:20 +04:00
memblock_get_region_node ( rgn ) ,
rgn - > flags ) ;
2011-12-08 22:22:07 +04:00
} else if ( rend > end ) {
/*
* @ rgn intersects from above . Split and redo the
* current region - the new bottom half .
*/
rgn - > base = end ;
2011-12-08 22:22:08 +04:00
rgn - > size - = end - rbase ;
type - > total_size - = end - rbase ;
2016-01-15 02:20:42 +03:00
memblock_insert_region ( type , idx - - , rbase , end - rbase ,
2014-01-22 03:49:20 +04:00
memblock_get_region_node ( rgn ) ,
rgn - > flags ) ;
2011-12-08 22:22:07 +04:00
} else {
/* @rgn is fully contained, record it */
if ( ! * end_rgn )
2016-01-15 02:20:42 +03:00
* start_rgn = idx ;
* end_rgn = idx + 1 ;
2011-12-08 22:22:07 +04:00
}
}
return 0 ;
}
2015-11-06 05:47:00 +03:00
static int __init_memblock memblock_remove_range ( struct memblock_type * type ,
2014-01-29 21:16:01 +04:00
phys_addr_t base , phys_addr_t size )
2010-07-12 08:36:09 +04:00
{
2011-12-08 22:22:07 +04:00
int start_rgn , end_rgn ;
int i , ret ;
2010-07-12 08:36:09 +04:00
2011-12-08 22:22:07 +04:00
ret = memblock_isolate_range ( type , base , size , & start_rgn , & end_rgn ) ;
if ( ret )
return ret ;
2010-07-12 08:36:09 +04:00
2011-12-08 22:22:07 +04:00
for ( i = end_rgn - 1 ; i > = start_rgn ; i - - )
memblock_remove_region ( type , i ) ;
mm/memblock: properly handle overlaps and fix error path
Currently memblock_reserve() or memblock_free() don't handle overlaps of
any kind. There is some special casing for coalescing exactly adjacent
regions but that's about it.
This is annoying because typically memblock_reserve() is used to mark
regions passed by the firmware as reserved and we all know how much we can
trust our firmwares...
Also, with the current code, if we do something it doesn't handle right
such as trying to memblock_reserve() a large range spanning multiple
existing smaller reserved regions for example, or doing overlapping
reservations, it can silently corrupt the internal region array, causing
odd errors much later on, such as allocations returning reserved regions
etc...
This patch rewrites the underlying functions that add or remove a region
to the arrays. The new code is a lot more robust as it fully handles
overlapping regions. It's also, imho, simpler than the previous
implementation.
In addition, while doing so, I found a bug where if we fail to double the
array while adding a region, we would remove the last region of the array
rather than the region we just allocated. This fixes it too.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Yinghai Lu <yinghai@kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2011-03-23 02:33:43 +03:00
return 0 ;
2010-07-12 08:36:09 +04:00
}
2011-12-08 22:22:06 +04:00
int __init_memblock memblock_remove ( phys_addr_t base , phys_addr_t size )
2010-07-12 08:36:09 +04:00
{
2014-01-29 21:16:01 +04:00
return memblock_remove_range ( & memblock . memory , base , size ) ;
2010-07-12 08:36:09 +04:00
}
2014-01-29 21:16:01 +04:00
2011-12-08 22:22:06 +04:00
int __init_memblock memblock_free ( phys_addr_t base , phys_addr_t size )
2010-07-12 08:36:09 +04:00
{
2011-07-12 13:16:06 +04:00
memblock_dbg ( " memblock_free: [%#016llx-%#016llx] %pF \n " ,
2011-07-14 22:57:10 +04:00
( unsigned long long ) base ,
2014-01-22 03:49:17 +04:00
( unsigned long long ) base + size - 1 ,
2011-07-14 22:57:10 +04:00
( void * ) _RET_IP_ ) ;
2011-07-12 13:16:06 +04:00
2014-06-07 01:38:20 +04:00
kmemleak_free_part ( __va ( base ) , size ) ;
2014-01-29 21:16:01 +04:00
return memblock_remove_range ( & memblock . reserved , base , size ) ;
2010-07-12 08:36:09 +04:00
}
2014-01-22 03:49:20 +04:00
static int __init_memblock memblock_reserve_region ( phys_addr_t base ,
phys_addr_t size ,
int nid ,
unsigned long flags )
2010-07-12 08:36:09 +04:00
{
2014-01-22 03:49:20 +04:00
memblock_dbg ( " memblock_reserve: [%#016llx-%#016llx] flags %#02lx %pF \n " ,
2011-07-14 22:57:10 +04:00
( unsigned long long ) base ,
2014-01-22 03:49:17 +04:00
( unsigned long long ) base + size - 1 ,
2014-01-22 03:49:20 +04:00
flags , ( void * ) _RET_IP_ ) ;
2016-03-16 00:57:42 +03:00
return memblock_add_range ( & memblock . reserved , base , size , nid , flags ) ;
2014-01-22 03:49:20 +04:00
}
2010-07-12 08:36:09 +04:00
2014-01-22 03:49:20 +04:00
int __init_memblock memblock_reserve ( phys_addr_t base , phys_addr_t size )
{
return memblock_reserve_region ( base , size , MAX_NUMNODES , 0 ) ;
2010-07-12 08:36:09 +04:00
}
2014-01-22 03:49:23 +04:00
/**
*
2014-12-13 03:54:59 +03:00
* This function isolates region [ @ base , @ base + @ size ) , and sets / clears flag
2014-01-22 03:49:23 +04:00
*
2015-09-09 01:04:02 +03:00
* Return 0 on success , - errno on failure .
2014-01-22 03:49:23 +04:00
*/
2014-12-13 03:54:59 +03:00
static int __init_memblock memblock_setclr_flag ( phys_addr_t base ,
phys_addr_t size , int set , int flag )
2014-01-22 03:49:23 +04:00
{
struct memblock_type * type = & memblock . memory ;
int i , ret , start_rgn , end_rgn ;
ret = memblock_isolate_range ( type , base , size , & start_rgn , & end_rgn ) ;
if ( ret )
return ret ;
for ( i = start_rgn ; i < end_rgn ; i + + )
2014-12-13 03:54:59 +03:00
if ( set )
memblock_set_region_flags ( & type - > regions [ i ] , flag ) ;
else
memblock_clear_region_flags ( & type - > regions [ i ] , flag ) ;
2014-01-22 03:49:23 +04:00
memblock_merge_regions ( type ) ;
return 0 ;
}
/**
2014-12-13 03:54:59 +03:00
* memblock_mark_hotplug - Mark hotpluggable memory with flag MEMBLOCK_HOTPLUG .
2014-01-22 03:49:23 +04:00
* @ base : the base phys addr of the region
* @ size : the size of the region
*
2015-09-09 01:04:02 +03:00
* Return 0 on success , - errno on failure .
2014-12-13 03:54:59 +03:00
*/
int __init_memblock memblock_mark_hotplug ( phys_addr_t base , phys_addr_t size )
{
return memblock_setclr_flag ( base , size , 1 , MEMBLOCK_HOTPLUG ) ;
}
/**
* memblock_clear_hotplug - Clear flag MEMBLOCK_HOTPLUG for a specified region .
* @ base : the base phys addr of the region
* @ size : the size of the region
2014-01-22 03:49:23 +04:00
*
2015-09-09 01:04:02 +03:00
* Return 0 on success , - errno on failure .
2014-01-22 03:49:23 +04:00
*/
int __init_memblock memblock_clear_hotplug ( phys_addr_t base , phys_addr_t size )
{
2014-12-13 03:54:59 +03:00
return memblock_setclr_flag ( base , size , 0 , MEMBLOCK_HOTPLUG ) ;
2014-01-22 03:49:23 +04:00
}
2015-06-25 02:58:12 +03:00
/**
* memblock_mark_mirror - Mark mirrored memory with flag MEMBLOCK_MIRROR .
* @ base : the base phys addr of the region
* @ size : the size of the region
*
2015-09-09 01:04:02 +03:00
* Return 0 on success , - errno on failure .
2015-06-25 02:58:12 +03:00
*/
int __init_memblock memblock_mark_mirror ( phys_addr_t base , phys_addr_t size )
{
system_has_some_mirror = true ;
return memblock_setclr_flag ( base , size , 1 , MEMBLOCK_MIRROR ) ;
}
2015-11-30 15:28:15 +03:00
/**
* memblock_mark_nomap - Mark a memory region with flag MEMBLOCK_NOMAP .
* @ base : the base phys addr of the region
* @ size : the size of the region
*
* Return 0 on success , - errno on failure .
*/
int __init_memblock memblock_mark_nomap ( phys_addr_t base , phys_addr_t size )
{
return memblock_setclr_flag ( base , size , 1 , MEMBLOCK_NOMAP ) ;
}
2015-06-25 02:58:12 +03:00
memblock: introduce a for_each_reserved_mem_region iterator
Struct page initialisation had been identified as one of the reasons why
large machines take a long time to boot. Patches were posted a long time ago
to defer initialisation until they were first used. This was rejected on
the grounds it should not be necessary to hurt the fast paths. This series
reuses much of the work from that time but defers the initialisation of
memory to kswapd so that one thread per node initialises memory local to
that node.
After applying the series and setting the appropriate Kconfig variable I
see this in the boot log on a 64G machine
[ 7.383764] kswapd 0 initialised deferred memory in 188ms
[ 7.404253] kswapd 1 initialised deferred memory in 208ms
[ 7.411044] kswapd 3 initialised deferred memory in 216ms
[ 7.411551] kswapd 2 initialised deferred memory in 216ms
On a 1TB machine, I see
[ 8.406511] kswapd 3 initialised deferred memory in 1116ms
[ 8.428518] kswapd 1 initialised deferred memory in 1140ms
[ 8.435977] kswapd 0 initialised deferred memory in 1148ms
[ 8.437416] kswapd 2 initialised deferred memory in 1148ms
Once booted the machine appears to work as normal. Boot times were measured
from the time shutdown was called until ssh was available again. In the
64G case, the boot time savings are negligible. On the 1TB machine, the
savings were 16 seconds.
Nate Zimmer said:
: On an older 8 TB box with lots and lots of cpus the boot time, as
: measure from grub to login prompt, the boot time improved from 1484
: seconds to exactly 1000 seconds.
Waiman Long said:
: I ran a bootup timing test on a 12-TB 16-socket IvyBridge-EX system. From
: grub menu to ssh login, the bootup time was 453s before the patch and 265s
: after the patch - a saving of 188s (42%).
Daniel Blueman said:
: On a 7TB, 1728-core NumaConnect system with 108 NUMA nodes, we're seeing
: stock 4.0 boot in 7136s. This drops to 2159s, or a 70% reduction with
: this patchset. Non-temporal PMD init (https://lkml.org/lkml/2015/4/23/350)
: drops this to 1045s.
This patch (of 13):
As part of initializing struct page's in 2MiB chunks, we noticed that at
the end of free_all_bootmem(), there was nothing which had forced the
reserved/allocated 4KiB pages to be initialized.
This helper function will be used for that expansion.
Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Nate Zimmer <nzimmer@sgi.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Tested-by: Nate Zimmer <nzimmer@sgi.com>
Tested-by: Waiman Long <waiman.long@hp.com>
Tested-by: Daniel J Blueman <daniel@numascale.com>
Acked-by: Pekka Enberg <penberg@kernel.org>
Cc: Robin Holt <robinmholt@gmail.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Waiman Long <waiman.long@hp.com>
Cc: Scott Norton <scott.norton@hp.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-07-01 00:56:41 +03:00
/**
* __next_reserved_mem_region - next function for for_each_reserved_region ( )
* @ idx : pointer to u64 loop variable
* @ out_start : ptr to phys_addr_t for start address of the region , can be % NULL
* @ out_end : ptr to phys_addr_t for end address of the region , can be % NULL
*
* Iterate over all reserved memory regions .
*/
void __init_memblock __next_reserved_mem_region ( u64 * idx ,
phys_addr_t * out_start ,
phys_addr_t * out_end )
{
2015-09-09 01:03:33 +03:00
struct memblock_type * type = & memblock . reserved ;
memblock: introduce a for_each_reserved_mem_region iterator
Struct page initialisation had been identified as one of the reasons why
large machines take a long time to boot. Patches were posted a long time ago
to defer initialisation until they were first used. This was rejected on
the grounds it should not be necessary to hurt the fast paths. This series
reuses much of the work from that time but defers the initialisation of
memory to kswapd so that one thread per node initialises memory local to
that node.
After applying the series and setting the appropriate Kconfig variable I
see this in the boot log on a 64G machine
[ 7.383764] kswapd 0 initialised deferred memory in 188ms
[ 7.404253] kswapd 1 initialised deferred memory in 208ms
[ 7.411044] kswapd 3 initialised deferred memory in 216ms
[ 7.411551] kswapd 2 initialised deferred memory in 216ms
On a 1TB machine, I see
[ 8.406511] kswapd 3 initialised deferred memory in 1116ms
[ 8.428518] kswapd 1 initialised deferred memory in 1140ms
[ 8.435977] kswapd 0 initialised deferred memory in 1148ms
[ 8.437416] kswapd 2 initialised deferred memory in 1148ms
Once booted the machine appears to work as normal. Boot times were measured
from the time shutdown was called until ssh was available again. In the
64G case, the boot time savings are negligible. On the 1TB machine, the
savings were 16 seconds.
Nate Zimmer said:
: On an older 8 TB box with lots and lots of cpus the boot time, as
: measure from grub to login prompt, the boot time improved from 1484
: seconds to exactly 1000 seconds.
Waiman Long said:
: I ran a bootup timing test on a 12-TB 16-socket IvyBridge-EX system. From
: grub menu to ssh login, the bootup time was 453s before the patch and 265s
: after the patch - a saving of 188s (42%).
Daniel Blueman said:
: On a 7TB, 1728-core NumaConnect system with 108 NUMA nodes, we're seeing
: stock 4.0 boot in 7136s. This drops to 2159s, or a 70% reduction with
: this patchset. Non-temporal PMD init (https://lkml.org/lkml/2015/4/23/350)
: drops this to 1045s.
This patch (of 13):
As part of initializing struct page's in 2MiB chunks, we noticed that at
the end of free_all_bootmem(), there was nothing which had forced the
reserved/allocated 4KiB pages to be initialized.
This helper function will be used for that expansion.
Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Nate Zimmer <nzimmer@sgi.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Tested-by: Nate Zimmer <nzimmer@sgi.com>
Tested-by: Waiman Long <waiman.long@hp.com>
Tested-by: Daniel J Blueman <daniel@numascale.com>
Acked-by: Pekka Enberg <penberg@kernel.org>
Cc: Robin Holt <robinmholt@gmail.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Waiman Long <waiman.long@hp.com>
Cc: Scott Norton <scott.norton@hp.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-07-01 00:56:41 +03:00
2015-09-09 01:03:33 +03:00
if ( * idx > = 0 & & * idx < type - > cnt ) {
struct memblock_region * r = & type - > regions [ * idx ] ;
memblock: introduce a for_each_reserved_mem_region iterator
Struct page initialisation had been identified as one of the reasons why
large machines take a long time to boot. Patches were posted a long time ago
to defer initialisation until they were first used. This was rejected on
the grounds it should not be necessary to hurt the fast paths. This series
reuses much of the work from that time but defers the initialisation of
memory to kswapd so that one thread per node initialises memory local to
that node.
After applying the series and setting the appropriate Kconfig variable I
see this in the boot log on a 64G machine
[ 7.383764] kswapd 0 initialised deferred memory in 188ms
[ 7.404253] kswapd 1 initialised deferred memory in 208ms
[ 7.411044] kswapd 3 initialised deferred memory in 216ms
[ 7.411551] kswapd 2 initialised deferred memory in 216ms
On a 1TB machine, I see
[ 8.406511] kswapd 3 initialised deferred memory in 1116ms
[ 8.428518] kswapd 1 initialised deferred memory in 1140ms
[ 8.435977] kswapd 0 initialised deferred memory in 1148ms
[ 8.437416] kswapd 2 initialised deferred memory in 1148ms
Once booted the machine appears to work as normal. Boot times were measured
from the time shutdown was called until ssh was available again. In the
64G case, the boot time savings are negligible. On the 1TB machine, the
savings were 16 seconds.
Nate Zimmer said:
: On an older 8 TB box with lots and lots of cpus the boot time, as
: measure from grub to login prompt, the boot time improved from 1484
: seconds to exactly 1000 seconds.
Waiman Long said:
: I ran a bootup timing test on a 12-TB 16-socket IvyBridge-EX system. From
: grub menu to ssh login, the bootup time was 453s before the patch and 265s
: after the patch - a saving of 188s (42%).
Daniel Blueman said:
: On a 7TB, 1728-core NumaConnect system with 108 NUMA nodes, we're seeing
: stock 4.0 boot in 7136s. This drops to 2159s, or a 70% reduction with
: this patchset. Non-temporal PMD init (https://lkml.org/lkml/2015/4/23/350)
: drops this to 1045s.
This patch (of 13):
As part of initializing struct page's in 2MiB chunks, we noticed that at
the end of free_all_bootmem(), there was nothing which had forced the
reserved/allocated 4KiB pages to be initialized.
This helper function will be used for that expansion.
Signed-off-by: Robin Holt <holt@sgi.com>
Signed-off-by: Nate Zimmer <nzimmer@sgi.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Tested-by: Nate Zimmer <nzimmer@sgi.com>
Tested-by: Waiman Long <waiman.long@hp.com>
Tested-by: Daniel J Blueman <daniel@numascale.com>
Acked-by: Pekka Enberg <penberg@kernel.org>
Cc: Robin Holt <robinmholt@gmail.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Waiman Long <waiman.long@hp.com>
Cc: Scott Norton <scott.norton@hp.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-07-01 00:56:41 +03:00
phys_addr_t base = r - > base ;
phys_addr_t size = r - > size ;
if ( out_start )
* out_start = base ;
if ( out_end )
* out_end = base + size - 1 ;
* idx + = 1 ;
return ;
}
/* signal end of iteration */
* idx = ULLONG_MAX ;
}
2011-07-12 13:15:59 +04:00
/**
2014-01-29 21:16:01 +04:00
* __next__mem_range - next function for for_each_free_mem_range ( ) etc .
2011-07-12 13:15:59 +04:00
* @ idx : pointer to u64 loop variable
2014-01-22 03:50:16 +04:00
* @ nid : node selector , % NUMA_NO_NODE for all nodes
2015-06-25 02:58:09 +03:00
* @ flags : pick from blocks based on memory attributes
2014-01-29 21:16:01 +04:00
* @ type_a : pointer to memblock_type from where the range is taken
* @ type_b : pointer to memblock_type which excludes memory from being taken
2012-06-20 23:53:01 +04:00
* @ out_start : ptr to phys_addr_t for start address of the range , can be % NULL
* @ out_end : ptr to phys_addr_t for end address of the range , can be % NULL
* @ out_nid : ptr to int for nid of the range , can be % NULL
2011-07-12 13:15:59 +04:00
*
2014-01-29 21:16:01 +04:00
* Find the first area from * @ idx which matches @ nid , fill the out
2011-07-12 13:15:59 +04:00
* parameters , and update * @ idx for the next iteration . The lower 32 bit of
2014-01-29 21:16:01 +04:00
* * @ idx contains index into type_a and the upper 32 bit indexes the
* areas before each region in type_b . For example , if type_b regions
2011-07-12 13:15:59 +04:00
* look like the following ,
*
* 0 : [ 0 - 16 ) , 1 : [ 32 - 48 ) , 2 : [ 128 - 130 )
*
* The upper 32 bit indexes the following regions .
*
* 0 : [ 0 - 0 ) , 1 : [ 16 - 32 ) , 2 : [ 48 - 128 ) , 3 : [ 130 - MAX )
*
* As both region arrays are sorted , the function advances the two indices
* in lockstep and returns each intersection .
*/
2015-06-25 02:58:09 +03:00
void __init_memblock __next_mem_range ( u64 * idx , int nid , ulong flags ,
2014-01-29 21:16:01 +04:00
struct memblock_type * type_a ,
struct memblock_type * type_b ,
phys_addr_t * out_start ,
phys_addr_t * out_end , int * out_nid )
2011-07-12 13:15:59 +04:00
{
2014-01-29 21:16:01 +04:00
int idx_a = * idx & 0xffffffff ;
int idx_b = * idx > > 32 ;
2014-01-22 03:50:16 +04:00
2014-01-29 21:16:01 +04:00
if ( WARN_ONCE ( nid = = MAX_NUMNODES ,
" Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead \n " ) )
2014-01-22 03:50:55 +04:00
nid = NUMA_NO_NODE ;
2011-07-12 13:15:59 +04:00
2014-01-29 21:16:01 +04:00
for ( ; idx_a < type_a - > cnt ; idx_a + + ) {
struct memblock_region * m = & type_a - > regions [ idx_a ] ;
2011-07-12 13:15:59 +04:00
phys_addr_t m_start = m - > base ;
phys_addr_t m_end = m - > base + m - > size ;
2014-01-29 21:16:01 +04:00
int m_nid = memblock_get_region_node ( m ) ;
2011-07-12 13:15:59 +04:00
/* only memory regions are associated with nodes, check it */
2014-01-29 21:16:01 +04:00
if ( nid ! = NUMA_NO_NODE & & nid ! = m_nid )
2011-07-12 13:15:59 +04:00
continue ;
2014-09-10 01:50:46 +04:00
/* skip hotpluggable memory regions if needed */
if ( movable_node_is_enabled ( ) & & memblock_is_hotpluggable ( m ) )
continue ;
2015-06-25 02:58:12 +03:00
/* if we want mirror memory skip non-mirror memory regions */
if ( ( flags & MEMBLOCK_MIRROR ) & & ! memblock_is_mirror ( m ) )
continue ;
2015-11-30 15:28:15 +03:00
/* skip nomap memory unless we were asked for it explicitly */
if ( ! ( flags & MEMBLOCK_NOMAP ) & & memblock_is_nomap ( m ) )
continue ;
2014-01-29 21:16:01 +04:00
if ( ! type_b ) {
if ( out_start )
* out_start = m_start ;
if ( out_end )
* out_end = m_end ;
if ( out_nid )
* out_nid = m_nid ;
idx_a + + ;
* idx = ( u32 ) idx_a | ( u64 ) idx_b < < 32 ;
return ;
}
/* scan areas before each reservation */
for ( ; idx_b < type_b - > cnt + 1 ; idx_b + + ) {
struct memblock_region * r ;
phys_addr_t r_start ;
phys_addr_t r_end ;
r = & type_b - > regions [ idx_b ] ;
r_start = idx_b ? r [ - 1 ] . base + r [ - 1 ] . size : 0 ;
r_end = idx_b < type_b - > cnt ?
r - > base : ULLONG_MAX ;
2011-07-12 13:15:59 +04:00
2014-01-29 21:16:01 +04:00
/*
* if idx_b advanced past idx_a ,
* break out to advance idx_a
*/
2011-07-12 13:15:59 +04:00
if ( r_start > = m_end )
break ;
/* if the two regions intersect, we're done */
if ( m_start < r_end ) {
if ( out_start )
2014-01-29 21:16:01 +04:00
* out_start =
max ( m_start , r_start ) ;
2011-07-12 13:15:59 +04:00
if ( out_end )
* out_end = min ( m_end , r_end ) ;
if ( out_nid )
2014-01-29 21:16:01 +04:00
* out_nid = m_nid ;
2011-07-12 13:15:59 +04:00
/*
2014-01-29 21:16:01 +04:00
* The region which ends first is
* advanced for the next iteration .
2011-07-12 13:15:59 +04:00
*/
if ( m_end < = r_end )
2014-01-29 21:16:01 +04:00
idx_a + + ;
2011-07-12 13:15:59 +04:00
else
2014-01-29 21:16:01 +04:00
idx_b + + ;
* idx = ( u32 ) idx_a | ( u64 ) idx_b < < 32 ;
2011-07-12 13:15:59 +04:00
return ;
}
}
}
/* signal end of iteration */
* idx = ULLONG_MAX ;
}
2011-12-08 22:22:09 +04:00
/**
2014-01-29 21:16:01 +04:00
* __next_mem_range_rev - generic next function for for_each_ * _range_rev ( )
*
* Finds the next range from type_a which is not marked as unsuitable
* in type_b .
*
2011-12-08 22:22:09 +04:00
* @ idx : pointer to u64 loop variable
2015-09-09 01:04:22 +03:00
* @ nid : node selector , % NUMA_NO_NODE for all nodes
2015-06-25 02:58:09 +03:00
* @ flags : pick from blocks based on memory attributes
2014-01-29 21:16:01 +04:00
* @ type_a : pointer to memblock_type from where the range is taken
* @ type_b : pointer to memblock_type which excludes memory from being taken
2012-06-20 23:53:01 +04:00
* @ out_start : ptr to phys_addr_t for start address of the range , can be % NULL
* @ out_end : ptr to phys_addr_t for end address of the range , can be % NULL
* @ out_nid : ptr to int for nid of the range , can be % NULL
2011-12-08 22:22:09 +04:00
*
2014-01-29 21:16:01 +04:00
* Reverse of __next_mem_range ( ) .
2011-12-08 22:22:09 +04:00
*/
2015-06-25 02:58:09 +03:00
void __init_memblock __next_mem_range_rev ( u64 * idx , int nid , ulong flags ,
2014-01-29 21:16:01 +04:00
struct memblock_type * type_a ,
struct memblock_type * type_b ,
phys_addr_t * out_start ,
phys_addr_t * out_end , int * out_nid )
2011-12-08 22:22:09 +04:00
{
2014-01-29 21:16:01 +04:00
int idx_a = * idx & 0xffffffff ;
int idx_b = * idx > > 32 ;
2014-01-22 03:50:16 +04:00
2014-01-22 03:50:55 +04:00
if ( WARN_ONCE ( nid = = MAX_NUMNODES , " Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead \n " ) )
nid = NUMA_NO_NODE ;
2011-12-08 22:22:09 +04:00
if ( * idx = = ( u64 ) ULLONG_MAX ) {
2014-01-29 21:16:01 +04:00
idx_a = type_a - > cnt - 1 ;
idx_b = type_b - > cnt ;
2011-12-08 22:22:09 +04:00
}
2014-01-29 21:16:01 +04:00
for ( ; idx_a > = 0 ; idx_a - - ) {
struct memblock_region * m = & type_a - > regions [ idx_a ] ;
2011-12-08 22:22:09 +04:00
phys_addr_t m_start = m - > base ;
phys_addr_t m_end = m - > base + m - > size ;
2014-01-29 21:16:01 +04:00
int m_nid = memblock_get_region_node ( m ) ;
2011-12-08 22:22:09 +04:00
/* only memory regions are associated with nodes, check it */
2014-01-29 21:16:01 +04:00
if ( nid ! = NUMA_NO_NODE & & nid ! = m_nid )
2011-12-08 22:22:09 +04:00
continue ;
2014-01-22 03:49:35 +04:00
/* skip hotpluggable memory regions if needed */
if ( movable_node_is_enabled ( ) & & memblock_is_hotpluggable ( m ) )
continue ;
2015-06-25 02:58:12 +03:00
/* if we want mirror memory skip non-mirror memory regions */
if ( ( flags & MEMBLOCK_MIRROR ) & & ! memblock_is_mirror ( m ) )
continue ;
2015-11-30 15:28:15 +03:00
/* skip nomap memory unless we were asked for it explicitly */
if ( ! ( flags & MEMBLOCK_NOMAP ) & & memblock_is_nomap ( m ) )
continue ;
2014-01-29 21:16:01 +04:00
if ( ! type_b ) {
if ( out_start )
* out_start = m_start ;
if ( out_end )
* out_end = m_end ;
if ( out_nid )
* out_nid = m_nid ;
idx_a + + ;
* idx = ( u32 ) idx_a | ( u64 ) idx_b < < 32 ;
return ;
}
/* scan areas before each reservation */
for ( ; idx_b > = 0 ; idx_b - - ) {
struct memblock_region * r ;
phys_addr_t r_start ;
phys_addr_t r_end ;
r = & type_b - > regions [ idx_b ] ;
r_start = idx_b ? r [ - 1 ] . base + r [ - 1 ] . size : 0 ;
r_end = idx_b < type_b - > cnt ?
r - > base : ULLONG_MAX ;
/*
* if idx_b advanced past idx_a ,
* break out to advance idx_a
*/
2011-12-08 22:22:09 +04:00
if ( r_end < = m_start )
break ;
/* if the two regions intersect, we're done */
if ( m_end > r_start ) {
if ( out_start )
* out_start = max ( m_start , r_start ) ;
if ( out_end )
* out_end = min ( m_end , r_end ) ;
if ( out_nid )
2014-01-29 21:16:01 +04:00
* out_nid = m_nid ;
2011-12-08 22:22:09 +04:00
if ( m_start > = r_start )
2014-01-29 21:16:01 +04:00
idx_a - - ;
2011-12-08 22:22:09 +04:00
else
2014-01-29 21:16:01 +04:00
idx_b - - ;
* idx = ( u32 ) idx_a | ( u64 ) idx_b < < 32 ;
2011-12-08 22:22:09 +04:00
return ;
}
}
}
2014-01-29 21:16:01 +04:00
/* signal end of iteration */
2011-12-08 22:22:09 +04:00
* idx = ULLONG_MAX ;
}
2011-07-14 13:43:42 +04:00
# ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
/*
* Common iterator interface used to define for_each_mem_range ( ) .
*/
void __init_memblock __next_mem_pfn_range ( int * idx , int nid ,
unsigned long * out_start_pfn ,
unsigned long * out_end_pfn , int * out_nid )
{
struct memblock_type * type = & memblock . memory ;
struct memblock_region * r ;
while ( + + * idx < type - > cnt ) {
r = & type - > regions [ * idx ] ;
if ( PFN_UP ( r - > base ) > = PFN_DOWN ( r - > base + r - > size ) )
continue ;
if ( nid = = MAX_NUMNODES | | nid = = r - > nid )
break ;
}
if ( * idx > = type - > cnt ) {
* idx = - 1 ;
return ;
}
if ( out_start_pfn )
* out_start_pfn = PFN_UP ( r - > base ) ;
if ( out_end_pfn )
* out_end_pfn = PFN_DOWN ( r - > base + r - > size ) ;
if ( out_nid )
* out_nid = r - > nid ;
}
/**
* memblock_set_node - set node ID on memblock regions
* @ base : base of area to set node ID for
* @ size : size of area to set node ID for
2014-01-22 03:49:26 +04:00
* @ type : memblock type to set node ID for
2011-07-14 13:43:42 +04:00
* @ nid : node ID to set
*
2014-01-22 03:49:26 +04:00
* Set the nid of memblock @ type regions in [ @ base , @ base + @ size ) to @ nid .
2011-07-14 13:43:42 +04:00
* Regions which cross the area boundaries are split as necessary .
*
* RETURNS :
* 0 on success , - errno on failure .
*/
int __init_memblock memblock_set_node ( phys_addr_t base , phys_addr_t size ,
2014-01-22 03:49:26 +04:00
struct memblock_type * type , int nid )
2011-07-14 13:43:42 +04:00
{
2011-12-08 22:22:07 +04:00
int start_rgn , end_rgn ;
int i , ret ;
2011-07-14 13:43:42 +04:00
2011-12-08 22:22:07 +04:00
ret = memblock_isolate_range ( type , base , size , & start_rgn , & end_rgn ) ;
if ( ret )
return ret ;
2011-07-14 13:43:42 +04:00
2011-12-08 22:22:07 +04:00
for ( i = start_rgn ; i < end_rgn ; i + + )
2012-10-09 03:32:21 +04:00
memblock_set_region_node ( & type - > regions [ i ] , nid ) ;
2011-07-14 13:43:42 +04:00
memblock_merge_regions ( type ) ;
return 0 ;
}
# endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
2014-06-05 03:06:53 +04:00
static phys_addr_t __init memblock_alloc_range_nid ( phys_addr_t size ,
phys_addr_t align , phys_addr_t start ,
2015-06-25 02:58:09 +03:00
phys_addr_t end , int nid , ulong flags )
2010-07-12 08:36:09 +04:00
{
2010-07-12 08:36:48 +04:00
phys_addr_t found ;
2010-07-12 08:36:09 +04:00
2014-01-22 03:50:12 +04:00
if ( ! align )
align = SMP_CACHE_BYTES ;
2013-04-30 02:06:15 +04:00
2015-06-25 02:58:09 +03:00
found = memblock_find_in_range_node ( size , align , start , end , nid ,
flags ) ;
2014-06-07 01:38:20 +04:00
if ( found & & ! memblock_reserve ( found , size ) ) {
/*
* The min_count is set to 0 so that memblock allocations are
* never reported as leaks .
*/
kmemleak_alloc ( __va ( found ) , size , 0 , 0 ) ;
2010-07-12 08:36:48 +04:00
return found ;
2014-06-07 01:38:20 +04:00
}
2010-07-12 08:36:48 +04:00
return 0 ;
2010-07-12 08:36:09 +04:00
}
2014-06-05 03:06:53 +04:00
phys_addr_t __init memblock_alloc_range ( phys_addr_t size , phys_addr_t align ,
2015-06-25 02:58:09 +03:00
phys_addr_t start , phys_addr_t end ,
ulong flags )
2014-06-05 03:06:53 +04:00
{
2015-06-25 02:58:09 +03:00
return memblock_alloc_range_nid ( size , align , start , end , NUMA_NO_NODE ,
flags ) ;
2014-06-05 03:06:53 +04:00
}
static phys_addr_t __init memblock_alloc_base_nid ( phys_addr_t size ,
phys_addr_t align , phys_addr_t max_addr ,
2015-06-25 02:58:09 +03:00
int nid , ulong flags )
2014-06-05 03:06:53 +04:00
{
2015-06-25 02:58:09 +03:00
return memblock_alloc_range_nid ( size , align , 0 , max_addr , nid , flags ) ;
2014-06-05 03:06:53 +04:00
}
2011-12-08 22:22:09 +04:00
phys_addr_t __init memblock_alloc_nid ( phys_addr_t size , phys_addr_t align , int nid )
{
2015-06-25 02:58:12 +03:00
ulong flags = choose_memblock_flags ( ) ;
phys_addr_t ret ;
again :
ret = memblock_alloc_base_nid ( size , align , MEMBLOCK_ALLOC_ACCESSIBLE ,
nid , flags ) ;
if ( ! ret & & ( flags & MEMBLOCK_MIRROR ) ) {
flags & = ~ MEMBLOCK_MIRROR ;
goto again ;
}
return ret ;
2011-12-08 22:22:09 +04:00
}
phys_addr_t __init __memblock_alloc_base ( phys_addr_t size , phys_addr_t align , phys_addr_t max_addr )
{
2015-06-25 02:58:09 +03:00
return memblock_alloc_base_nid ( size , align , max_addr , NUMA_NO_NODE ,
MEMBLOCK_NONE ) ;
2011-12-08 22:22:09 +04:00
}
2010-07-12 08:36:48 +04:00
phys_addr_t __init memblock_alloc_base ( phys_addr_t size , phys_addr_t align , phys_addr_t max_addr )
2010-07-12 08:36:09 +04:00
{
2010-07-12 08:36:48 +04:00
phys_addr_t alloc ;
alloc = __memblock_alloc_base ( size , align , max_addr ) ;
if ( alloc = = 0 )
panic ( " ERROR: Failed to allocate 0x%llx bytes below 0x%llx. \n " ,
( unsigned long long ) size , ( unsigned long long ) max_addr ) ;
return alloc ;
2010-07-12 08:36:09 +04:00
}
2010-07-12 08:36:48 +04:00
phys_addr_t __init memblock_alloc ( phys_addr_t size , phys_addr_t align )
2010-07-12 08:36:09 +04:00
{
2010-07-12 08:36:48 +04:00
return memblock_alloc_base ( size , align , MEMBLOCK_ALLOC_ACCESSIBLE ) ;
}
2010-07-12 08:36:09 +04:00
2010-07-07 02:39:17 +04:00
phys_addr_t __init memblock_alloc_try_nid ( phys_addr_t size , phys_addr_t align , int nid )
{
phys_addr_t res = memblock_alloc_nid ( size , align , nid ) ;
if ( res )
return res ;
2011-07-12 11:58:07 +04:00
return memblock_alloc_base ( size , align , MEMBLOCK_ALLOC_ACCESSIBLE ) ;
2010-07-12 08:36:09 +04:00
}
2014-01-22 03:50:19 +04:00
/**
* memblock_virt_alloc_internal - allocate boot memory block
* @ size : size of memory block to be allocated in bytes
* @ align : alignment of the region and block ' s size
* @ min_addr : the lower bound of the memory region to allocate ( phys address )
* @ max_addr : the upper bound of the memory region to allocate ( phys address )
* @ nid : nid of the free area to find , % NUMA_NO_NODE for any node
*
* The @ min_addr limit is dropped if it can not be satisfied and the allocation
* will fall back to memory below @ min_addr . Also , allocation may fall back
* to any node in the system if the specified node can not
* hold the requested memory .
*
* The allocation is performed from memory region limited by
* memblock . current_limit if @ max_addr = = % BOOTMEM_ALLOC_ACCESSIBLE .
*
* The memory block is aligned on SMP_CACHE_BYTES if @ align = = 0.
*
* The phys address of allocated boot memory block is converted to virtual and
* allocated memory is reset to 0.
*
* In addition , function sets the min_count to 0 using kmemleak_alloc for
* allocated boot memory block , so that it is never reported as leaks .
*
* RETURNS :
* Virtual address of allocated memory block on success , NULL on failure .
*/
static void * __init memblock_virt_alloc_internal (
phys_addr_t size , phys_addr_t align ,
phys_addr_t min_addr , phys_addr_t max_addr ,
int nid )
{
phys_addr_t alloc ;
void * ptr ;
2015-06-25 02:58:12 +03:00
ulong flags = choose_memblock_flags ( ) ;
2014-01-22 03:50:19 +04:00
2014-01-22 03:50:55 +04:00
if ( WARN_ONCE ( nid = = MAX_NUMNODES , " Usage of MAX_NUMNODES is deprecated. Use NUMA_NO_NODE instead \n " ) )
nid = NUMA_NO_NODE ;
2014-01-22 03:50:19 +04:00
/*
* Detect any accidental use of these APIs after slab is ready , as at
* this moment memblock may be deinitialized already and its
* internal data may be destroyed ( after execution of free_all_bootmem )
*/
if ( WARN_ON_ONCE ( slab_is_available ( ) ) )
return kzalloc_node ( size , GFP_NOWAIT , nid ) ;
if ( ! align )
align = SMP_CACHE_BYTES ;
2014-01-30 02:05:52 +04:00
if ( max_addr > memblock . current_limit )
max_addr = memblock . current_limit ;
2014-01-22 03:50:19 +04:00
again :
alloc = memblock_find_in_range_node ( size , align , min_addr , max_addr ,
2015-06-25 02:58:12 +03:00
nid , flags ) ;
2014-01-22 03:50:19 +04:00
if ( alloc )
goto done ;
if ( nid ! = NUMA_NO_NODE ) {
alloc = memblock_find_in_range_node ( size , align , min_addr ,
2015-06-25 02:58:09 +03:00
max_addr , NUMA_NO_NODE ,
2015-06-25 02:58:12 +03:00
flags ) ;
2014-01-22 03:50:19 +04:00
if ( alloc )
goto done ;
}
if ( min_addr ) {
min_addr = 0 ;
goto again ;
}
2015-06-25 02:58:12 +03:00
if ( flags & MEMBLOCK_MIRROR ) {
flags & = ~ MEMBLOCK_MIRROR ;
pr_warn ( " Could not allocate %pap bytes of mirrored memory \n " ,
& size ) ;
goto again ;
}
return NULL ;
2014-01-22 03:50:19 +04:00
done :
memblock_reserve ( alloc , size ) ;
ptr = phys_to_virt ( alloc ) ;
memset ( ptr , 0 , size ) ;
/*
* The min_count is set to 0 so that bootmem allocated blocks
* are never reported as leaks . This is because many of these blocks
* are only referred via the physical address which is not
* looked up by kmemleak .
*/
kmemleak_alloc ( ptr , size , 0 , 0 ) ;
return ptr ;
}
/**
* memblock_virt_alloc_try_nid_nopanic - allocate boot memory block
* @ size : size of memory block to be allocated in bytes
* @ align : alignment of the region and block ' s size
* @ min_addr : the lower bound of the memory region from where the allocation
* is preferred ( phys address )
* @ max_addr : the upper bound of the memory region from where the allocation
* is preferred ( phys address ) , or % BOOTMEM_ALLOC_ACCESSIBLE to
* allocate only from memory limited by memblock . current_limit value
* @ nid : nid of the free area to find , % NUMA_NO_NODE for any node
*
* Public version of _memblock_virt_alloc_try_nid_nopanic ( ) which provides
* additional debug information ( including caller info ) , if enabled .
*
* RETURNS :
* Virtual address of allocated memory block on success , NULL on failure .
*/
void * __init memblock_virt_alloc_try_nid_nopanic (
phys_addr_t size , phys_addr_t align ,
phys_addr_t min_addr , phys_addr_t max_addr ,
int nid )
{
memblock_dbg ( " %s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF \n " ,
__func__ , ( u64 ) size , ( u64 ) align , nid , ( u64 ) min_addr ,
( u64 ) max_addr , ( void * ) _RET_IP_ ) ;
return memblock_virt_alloc_internal ( size , align , min_addr ,
max_addr , nid ) ;
}
/**
* memblock_virt_alloc_try_nid - allocate boot memory block with panicking
* @ size : size of memory block to be allocated in bytes
* @ align : alignment of the region and block ' s size
* @ min_addr : the lower bound of the memory region from where the allocation
* is preferred ( phys address )
* @ max_addr : the upper bound of the memory region from where the allocation
* is preferred ( phys address ) , or % BOOTMEM_ALLOC_ACCESSIBLE to
* allocate only from memory limited by memblock . current_limit value
* @ nid : nid of the free area to find , % NUMA_NO_NODE for any node
*
* Public panicking version of _memblock_virt_alloc_try_nid_nopanic ( )
* which provides debug information ( including caller info ) , if enabled ,
* and panics if the request can not be satisfied .
*
* RETURNS :
* Virtual address of allocated memory block on success , NULL on failure .
*/
void * __init memblock_virt_alloc_try_nid (
phys_addr_t size , phys_addr_t align ,
phys_addr_t min_addr , phys_addr_t max_addr ,
int nid )
{
void * ptr ;
memblock_dbg ( " %s: %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx %pF \n " ,
__func__ , ( u64 ) size , ( u64 ) align , nid , ( u64 ) min_addr ,
( u64 ) max_addr , ( void * ) _RET_IP_ ) ;
ptr = memblock_virt_alloc_internal ( size , align ,
min_addr , max_addr , nid ) ;
if ( ptr )
return ptr ;
panic ( " %s: Failed to allocate %llu bytes align=0x%llx nid=%d from=0x%llx max_addr=0x%llx \n " ,
__func__ , ( u64 ) size , ( u64 ) align , nid , ( u64 ) min_addr ,
( u64 ) max_addr ) ;
return NULL ;
}
/**
* __memblock_free_early - free boot memory block
* @ base : phys starting address of the boot memory block
* @ size : size of the boot memory block in bytes
*
* Free boot memory block previously allocated by memblock_virt_alloc_xx ( ) API .
* The freeing memory will not be released to the buddy allocator .
*/
void __init __memblock_free_early ( phys_addr_t base , phys_addr_t size )
{
memblock_dbg ( " %s: [%#016llx-%#016llx] %pF \n " ,
__func__ , ( u64 ) base , ( u64 ) base + size - 1 ,
( void * ) _RET_IP_ ) ;
kmemleak_free_part ( __va ( base ) , size ) ;
2014-01-29 21:16:01 +04:00
memblock_remove_range ( & memblock . reserved , base , size ) ;
2014-01-22 03:50:19 +04:00
}
/*
* __memblock_free_late - free bootmem block pages directly to buddy allocator
* @ addr : phys starting address of the boot memory block
* @ size : size of the boot memory block in bytes
*
* This is only useful when the bootmem allocator has already been torn
* down , but we are still initializing the system . Pages are released directly
* to the buddy allocator , no bootmem metadata is updated because it is gone .
*/
void __init __memblock_free_late ( phys_addr_t base , phys_addr_t size )
{
u64 cursor , end ;
memblock_dbg ( " %s: [%#016llx-%#016llx] %pF \n " ,
__func__ , ( u64 ) base , ( u64 ) base + size - 1 ,
( void * ) _RET_IP_ ) ;
kmemleak_free_part ( __va ( base ) , size ) ;
cursor = PFN_UP ( base ) ;
end = PFN_DOWN ( base + size ) ;
for ( ; cursor < end ; cursor + + ) {
2015-07-01 00:56:52 +03:00
__free_pages_bootmem ( pfn_to_page ( cursor ) , cursor , 0 ) ;
2014-01-22 03:50:19 +04:00
totalram_pages + + ;
}
}
2010-07-07 02:39:17 +04:00
/*
* Remaining API functions
*/
2016-02-06 02:36:19 +03:00
phys_addr_t __init_memblock memblock_phys_mem_size ( void )
2010-07-12 08:36:09 +04:00
{
2011-12-08 22:22:08 +04:00
return memblock . memory . total_size ;
2010-07-12 08:36:09 +04:00
}
2013-01-25 00:20:09 +04:00
phys_addr_t __init memblock_mem_size ( unsigned long limit_pfn )
{
unsigned long pages = 0 ;
struct memblock_region * r ;
unsigned long start_pfn , end_pfn ;
for_each_memblock ( memory , r ) {
start_pfn = memblock_region_memory_base_pfn ( r ) ;
end_pfn = memblock_region_memory_end_pfn ( r ) ;
start_pfn = min_t ( unsigned long , start_pfn , limit_pfn ) ;
end_pfn = min_t ( unsigned long , end_pfn , limit_pfn ) ;
pages + = end_pfn - start_pfn ;
}
2014-04-08 02:37:53 +04:00
return PFN_PHYS ( pages ) ;
2013-01-25 00:20:09 +04:00
}
2011-11-01 04:08:16 +04:00
/* lowest address */
phys_addr_t __init_memblock memblock_start_of_DRAM ( void )
{
return memblock . memory . regions [ 0 ] . base ;
}
2010-07-28 09:43:02 +04:00
phys_addr_t __init_memblock memblock_end_of_DRAM ( void )
2010-07-12 08:36:09 +04:00
{
int idx = memblock . memory . cnt - 1 ;
2010-08-04 08:06:41 +04:00
return ( memblock . memory . regions [ idx ] . base + memblock . memory . regions [ idx ] . size ) ;
2010-07-12 08:36:09 +04:00
}
2011-12-08 22:22:07 +04:00
void __init memblock_enforce_memory_limit ( phys_addr_t limit )
2010-07-12 08:36:09 +04:00
{
2011-12-08 22:22:07 +04:00
phys_addr_t max_addr = ( phys_addr_t ) ULLONG_MAX ;
2014-04-08 02:37:52 +04:00
struct memblock_region * r ;
2010-07-12 08:36:09 +04:00
2011-12-08 22:22:07 +04:00
if ( ! limit )
2010-07-12 08:36:09 +04:00
return ;
2011-12-08 22:22:07 +04:00
/* find out max address */
2014-04-08 02:37:52 +04:00
for_each_memblock ( memory , r ) {
2011-12-08 22:22:07 +04:00
if ( limit < = r - > size ) {
max_addr = r - > base + limit ;
break ;
2010-07-12 08:36:09 +04:00
}
2011-12-08 22:22:07 +04:00
limit - = r - > size ;
2010-07-12 08:36:09 +04:00
}
2011-12-08 22:22:07 +04:00
/* truncate both memory and reserved regions */
2014-01-29 21:16:01 +04:00
memblock_remove_range ( & memblock . memory , max_addr ,
( phys_addr_t ) ULLONG_MAX ) ;
memblock_remove_range ( & memblock . reserved , max_addr ,
( phys_addr_t ) ULLONG_MAX ) ;
2010-07-12 08:36:09 +04:00
}
2010-10-11 23:34:09 +04:00
static int __init_memblock memblock_search ( struct memblock_type * type , phys_addr_t addr )
2010-08-04 08:38:47 +04:00
{
unsigned int left = 0 , right = type - > cnt ;
do {
unsigned int mid = ( right + left ) / 2 ;
if ( addr < type - > regions [ mid ] . base )
right = mid ;
else if ( addr > = ( type - > regions [ mid ] . base +
type - > regions [ mid ] . size ) )
left = mid + 1 ;
else
return mid ;
} while ( left < right ) ;
return - 1 ;
}
2016-01-15 02:18:54 +03:00
bool __init memblock_is_reserved ( phys_addr_t addr )
2010-07-12 08:36:09 +04:00
{
2010-08-04 08:38:47 +04:00
return memblock_search ( & memblock . reserved , addr ) ! = - 1 ;
}
2010-07-12 08:36:09 +04:00
2016-01-15 02:18:54 +03:00
bool __init_memblock memblock_is_memory ( phys_addr_t addr )
2010-08-04 08:38:47 +04:00
{
return memblock_search ( & memblock . memory , addr ) ! = - 1 ;
}
2015-11-30 15:28:15 +03:00
int __init_memblock memblock_is_map_memory ( phys_addr_t addr )
{
int i = memblock_search ( & memblock . memory , addr ) ;
if ( i = = - 1 )
return false ;
return ! memblock_is_nomap ( & memblock . memory . regions [ i ] ) ;
}
2013-09-12 01:22:17 +04:00
# ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
int __init_memblock memblock_search_pfn_nid ( unsigned long pfn ,
unsigned long * start_pfn , unsigned long * end_pfn )
{
struct memblock_type * type = & memblock . memory ;
2014-04-08 02:37:53 +04:00
int mid = memblock_search ( type , PFN_PHYS ( pfn ) ) ;
2013-09-12 01:22:17 +04:00
if ( mid = = - 1 )
return - 1 ;
2014-06-05 03:07:51 +04:00
* start_pfn = PFN_DOWN ( type - > regions [ mid ] . base ) ;
* end_pfn = PFN_DOWN ( type - > regions [ mid ] . base + type - > regions [ mid ] . size ) ;
2013-09-12 01:22:17 +04:00
return type - > regions [ mid ] . nid ;
}
# endif
2012-05-24 11:45:21 +04:00
/**
* memblock_is_region_memory - check if a region is a subset of memory
* @ base : base of region to check
* @ size : size of region to check
*
* Check if the region [ @ base , @ base + @ size ) is a subset of a memory block .
*
* RETURNS :
* 0 if false , non - zero if true
*/
2010-09-16 00:05:29 +04:00
int __init_memblock memblock_is_region_memory ( phys_addr_t base , phys_addr_t size )
2010-08-04 08:38:47 +04:00
{
2011-01-21 01:44:20 +03:00
int idx = memblock_search ( & memblock . memory , base ) ;
2011-12-08 22:22:07 +04:00
phys_addr_t end = base + memblock_cap_size ( base , & size ) ;
2010-08-04 08:38:47 +04:00
if ( idx = = - 1 )
return 0 ;
2011-01-21 01:44:20 +03:00
return memblock . memory . regions [ idx ] . base < = base & &
( memblock . memory . regions [ idx ] . base +
2011-12-08 22:22:07 +04:00
memblock . memory . regions [ idx ] . size ) > = end ;
2010-07-12 08:36:09 +04:00
}
2012-05-24 11:45:21 +04:00
/**
* memblock_is_region_reserved - check if a region intersects reserved memory
* @ base : base of region to check
* @ size : size of region to check
*
* Check if the region [ @ base , @ base + @ size ) intersects a reserved memory block .
*
* RETURNS :
2015-09-09 01:02:00 +03:00
* True if they intersect , false if not .
2012-05-24 11:45:21 +04:00
*/
2015-09-09 01:02:00 +03:00
bool __init_memblock memblock_is_region_reserved ( phys_addr_t base , phys_addr_t size )
2010-07-12 08:36:09 +04:00
{
2011-12-08 22:22:07 +04:00
memblock_cap_size ( base , & size ) ;
2015-09-09 01:02:00 +03:00
return memblock_overlaps_region ( & memblock . reserved , base , size ) ;
2010-07-12 08:36:09 +04:00
}
2012-10-23 03:35:18 +04:00
void __init_memblock memblock_trim_memory ( phys_addr_t align )
{
phys_addr_t start , end , orig_start , orig_end ;
2014-04-08 02:37:52 +04:00
struct memblock_region * r ;
2012-10-23 03:35:18 +04:00
2014-04-08 02:37:52 +04:00
for_each_memblock ( memory , r ) {
orig_start = r - > base ;
orig_end = r - > base + r - > size ;
2012-10-23 03:35:18 +04:00
start = round_up ( orig_start , align ) ;
end = round_down ( orig_end , align ) ;
if ( start = = orig_start & & end = = orig_end )
continue ;
if ( start < end ) {
2014-04-08 02:37:52 +04:00
r - > base = start ;
r - > size = end - start ;
2012-10-23 03:35:18 +04:00
} else {
2014-04-08 02:37:52 +04:00
memblock_remove_region ( & memblock . memory ,
r - memblock . memory . regions ) ;
r - - ;
2012-10-23 03:35:18 +04:00
}
}
}
2010-07-07 02:39:01 +04:00
2010-09-16 00:05:29 +04:00
void __init_memblock memblock_set_current_limit ( phys_addr_t limit )
2010-07-07 02:39:01 +04:00
{
memblock . current_limit = limit ;
}
2014-02-27 04:23:43 +04:00
phys_addr_t __init_memblock memblock_get_current_limit ( void )
{
return memblock . current_limit ;
}
2011-07-14 13:43:42 +04:00
static void __init_memblock memblock_dump ( struct memblock_type * type , char * name )
2010-07-12 08:36:48 +04:00
{
unsigned long long base , size ;
2014-01-22 03:49:20 +04:00
unsigned long flags ;
2016-01-15 02:20:42 +03:00
int idx ;
struct memblock_region * rgn ;
2010-07-12 08:36:48 +04:00
2011-07-14 13:43:42 +04:00
pr_info ( " %s.cnt = 0x%lx \n " , name , type - > cnt ) ;
2010-07-12 08:36:48 +04:00
2016-01-15 02:20:42 +03:00
for_each_memblock_type ( type , rgn ) {
2011-07-14 13:43:42 +04:00
char nid_buf [ 32 ] = " " ;
base = rgn - > base ;
size = rgn - > size ;
2014-01-22 03:49:20 +04:00
flags = rgn - > flags ;
2011-07-14 13:43:42 +04:00
# ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
if ( memblock_get_region_node ( rgn ) ! = MAX_NUMNODES )
snprintf ( nid_buf , sizeof ( nid_buf ) , " on node %d " ,
memblock_get_region_node ( rgn ) ) ;
# endif
2014-01-22 03:49:20 +04:00
pr_info ( " %s[%#x] \t [%#016llx-%#016llx], %#llx bytes%s flags: %#lx \n " ,
2016-01-15 02:20:42 +03:00
name , idx , base , base + size - 1 , size , nid_buf , flags ) ;
2010-07-12 08:36:48 +04:00
}
}
2011-12-08 22:22:06 +04:00
void __init_memblock __memblock_dump_all ( void )
2010-07-12 08:36:48 +04:00
{
pr_info ( " MEMBLOCK configuration: \n " ) ;
2011-12-08 22:22:08 +04:00
pr_info ( " memory size = %#llx reserved size = %#llx \n " ,
( unsigned long long ) memblock . memory . total_size ,
( unsigned long long ) memblock . reserved . total_size ) ;
2010-07-12 08:36:48 +04:00
memblock_dump ( & memblock . memory , " memory " ) ;
memblock_dump ( & memblock . reserved , " reserved " ) ;
}
2011-12-08 22:22:08 +04:00
void __init memblock_allow_resize ( void )
2010-07-12 08:36:48 +04:00
{
2010-07-07 02:39:13 +04:00
memblock_can_resize = 1 ;
2010-07-12 08:36:48 +04:00
}
static int __init early_memblock ( char * p )
{
if ( p & & strstr ( p , " debug " ) )
memblock_debug = 1 ;
return 0 ;
}
early_param ( " memblock " , early_memblock ) ;
2011-07-14 13:46:03 +04:00
# if defined(CONFIG_DEBUG_FS) && !defined(CONFIG_ARCH_DISCARD_MEMBLOCK)
2010-07-07 02:39:19 +04:00
static int memblock_debug_show ( struct seq_file * m , void * private )
{
struct memblock_type * type = m - > private ;
struct memblock_region * reg ;
int i ;
for ( i = 0 ; i < type - > cnt ; i + + ) {
reg = & type - > regions [ i ] ;
seq_printf ( m , " %4d: " , i ) ;
if ( sizeof ( phys_addr_t ) = = 4 )
seq_printf ( m , " 0x%08lx..0x%08lx \n " ,
( unsigned long ) reg - > base ,
( unsigned long ) ( reg - > base + reg - > size - 1 ) ) ;
else
seq_printf ( m , " 0x%016llx..0x%016llx \n " ,
( unsigned long long ) reg - > base ,
( unsigned long long ) ( reg - > base + reg - > size - 1 ) ) ;
}
return 0 ;
}
static int memblock_debug_open ( struct inode * inode , struct file * file )
{
return single_open ( file , memblock_debug_show , inode - > i_private ) ;
}
static const struct file_operations memblock_debug_fops = {
. open = memblock_debug_open ,
. read = seq_read ,
. llseek = seq_lseek ,
. release = single_release ,
} ;
static int __init memblock_init_debugfs ( void )
{
struct dentry * root = debugfs_create_dir ( " memblock " , NULL ) ;
if ( ! root )
return - ENXIO ;
debugfs_create_file ( " memory " , S_IRUGO , root , & memblock . memory , & memblock_debug_fops ) ;
debugfs_create_file ( " reserved " , S_IRUGO , root , & memblock . reserved , & memblock_debug_fops ) ;
2014-01-29 21:16:01 +04:00
# ifdef CONFIG_HAVE_MEMBLOCK_PHYS_MAP
debugfs_create_file ( " physmem " , S_IRUGO , root , & memblock . physmem , & memblock_debug_fops ) ;
# endif
2010-07-07 02:39:19 +04:00
return 0 ;
}
__initcall ( memblock_init_debugfs ) ;
# endif /* CONFIG_DEBUG_FS */