2019-05-19 15:08:20 +03:00
// SPDX-License-Identifier: GPL-2.0-only
2015-08-19 07:34:34 +03:00
/*
* Copyright ( c ) 2015 , Christoph Hellwig .
* Copyright ( c ) 2015 , Intel Corporation .
*/
# include <linux/platform_device.h>
2015-11-12 03:46:33 +03:00
# include <linux/memory_hotplug.h>
2015-08-19 07:34:34 +03:00
# include <linux/libnvdimm.h>
# include <linux/module.h>
libnvdimm/e820: Retrieve and populate correct 'target_node' info
Use the new phys_to_target_node() and numa_map_to_online_node() helpers
to retrieve the correct id for the 'numa_node' ("local" / online
initiator node) and 'target_node' (offline target memory node) sysfs
attributes.
Below is an example from a 4 NUMA node system where all the memory on
node2 is pmem / reserved. It should be noted that with the arrival of
the ACPI HMAT table and EFI Specific Purpose Memory the kernel will
start to see more platforms with reserved / performance differentiated
memory in its own NUMA node. Hence all the stakeholders on the Cc for
what is ostensibly a libnvdimm local patch.
=== Before ===
/* Notice no online memory on node2 at start */
# numactl --hardware
available: 3 nodes (0-1,3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 3958 MB
node 0 free: 3708 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3871 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3971 MB
node distances:
node 0 1 3
0: 10 21 21
1: 21 10 21
3: 21 21 10
/*
* Put the pmem namespace into devdax mode so it can be assigned to the
* kmem driver
*/
# ndctl create-namespace -e namespace0.0 -m devdax -f
{
"dev":"namespace0.0",
"mode":"devdax",
"map":"dev",
"size":"3.94 GiB (4.23 GB)",
"uuid":"1650af9b-9ba3-4704-acd6-10178399d9a3",
[..]
}
/* Online Persistent Memory as System RAM */
# daxctl reconfigure-device --mode=system-ram dax0.0
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
[
{
"chardev":"dax0.0",
"size":4225761280,
"target_node":0,
"mode":"system-ram"
}
]
reconfigured 1 device
/* Note that the memory is onlined by default to the wrong node, node0 */
# numactl --hardware
available: 3 nodes (0-1,3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 7926 MB
node 0 free: 7655 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3871 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3971 MB
node distances:
node 0 1 3
0: 10 21 21
1: 21 10 21
3: 21 21 10
=== After ===
/* Notice that the "phys_index" error messages are gone */
# daxctl reconfigure-device --mode=system-ram dax0.0
[
{
"chardev":"dax0.0",
"size":4225761280,
"target_node":2,
"mode":"system-ram"
}
]
reconfigured 1 device
/* Notice that node2 is now correctly populated */
# numactl --hardware
available: 4 nodes (0-3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 3958 MB
node 0 free: 3793 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3851 MB
node 2 cpus:
node 2 size: 3968 MB
node 2 free: 3968 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3908 MB
node distances:
node 0 1 2 3
0: 10 21 21 21
1: 21 10 21 21
2: 21 21 10 21
3: 21 21 21 10
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Link: https://lore.kernel.org/r/158188327614.894464.13122730362187722603.stgit@dwillia2-desk3.amr.corp.intel.com
2020-02-16 23:01:16 +03:00
# include <linux/numa.h>
2015-08-19 07:34:34 +03:00
static int e820_pmem_remove ( struct platform_device * pdev )
{
struct nvdimm_bus * nvdimm_bus = platform_get_drvdata ( pdev ) ;
nvdimm_bus_unregister ( nvdimm_bus ) ;
return 0 ;
}
2018-06-02 21:43:39 +03:00
static int e820_register_one ( struct resource * res , void * data )
{
struct nd_region_desc ndr_desc ;
struct nvdimm_bus * nvdimm_bus = data ;
libnvdimm/e820: Retrieve and populate correct 'target_node' info
Use the new phys_to_target_node() and numa_map_to_online_node() helpers
to retrieve the correct id for the 'numa_node' ("local" / online
initiator node) and 'target_node' (offline target memory node) sysfs
attributes.
Below is an example from a 4 NUMA node system where all the memory on
node2 is pmem / reserved. It should be noted that with the arrival of
the ACPI HMAT table and EFI Specific Purpose Memory the kernel will
start to see more platforms with reserved / performance differentiated
memory in its own NUMA node. Hence all the stakeholders on the Cc for
what is ostensibly a libnvdimm local patch.
=== Before ===
/* Notice no online memory on node2 at start */
# numactl --hardware
available: 3 nodes (0-1,3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 3958 MB
node 0 free: 3708 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3871 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3971 MB
node distances:
node 0 1 3
0: 10 21 21
1: 21 10 21
3: 21 21 10
/*
* Put the pmem namespace into devdax mode so it can be assigned to the
* kmem driver
*/
# ndctl create-namespace -e namespace0.0 -m devdax -f
{
"dev":"namespace0.0",
"mode":"devdax",
"map":"dev",
"size":"3.94 GiB (4.23 GB)",
"uuid":"1650af9b-9ba3-4704-acd6-10178399d9a3",
[..]
}
/* Online Persistent Memory as System RAM */
# daxctl reconfigure-device --mode=system-ram dax0.0
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
[
{
"chardev":"dax0.0",
"size":4225761280,
"target_node":0,
"mode":"system-ram"
}
]
reconfigured 1 device
/* Note that the memory is onlined by default to the wrong node, node0 */
# numactl --hardware
available: 3 nodes (0-1,3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 7926 MB
node 0 free: 7655 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3871 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3971 MB
node distances:
node 0 1 3
0: 10 21 21
1: 21 10 21
3: 21 21 10
=== After ===
/* Notice that the "phys_index" error messages are gone */
# daxctl reconfigure-device --mode=system-ram dax0.0
[
{
"chardev":"dax0.0",
"size":4225761280,
"target_node":2,
"mode":"system-ram"
}
]
reconfigured 1 device
/* Notice that node2 is now correctly populated */
# numactl --hardware
available: 4 nodes (0-3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 3958 MB
node 0 free: 3793 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3851 MB
node 2 cpus:
node 2 size: 3968 MB
node 2 free: 3968 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3908 MB
node distances:
node 0 1 2 3
0: 10 21 21 21
1: 21 10 21 21
2: 21 21 10 21
3: 21 21 21 10
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Link: https://lore.kernel.org/r/158188327614.894464.13122730362187722603.stgit@dwillia2-desk3.amr.corp.intel.com
2020-02-16 23:01:16 +03:00
int nid = phys_to_target_node ( res - > start ) ;
2018-06-02 21:43:39 +03:00
memset ( & ndr_desc , 0 , sizeof ( ndr_desc ) ) ;
ndr_desc . res = res ;
libnvdimm/e820: Retrieve and populate correct 'target_node' info
Use the new phys_to_target_node() and numa_map_to_online_node() helpers
to retrieve the correct id for the 'numa_node' ("local" / online
initiator node) and 'target_node' (offline target memory node) sysfs
attributes.
Below is an example from a 4 NUMA node system where all the memory on
node2 is pmem / reserved. It should be noted that with the arrival of
the ACPI HMAT table and EFI Specific Purpose Memory the kernel will
start to see more platforms with reserved / performance differentiated
memory in its own NUMA node. Hence all the stakeholders on the Cc for
what is ostensibly a libnvdimm local patch.
=== Before ===
/* Notice no online memory on node2 at start */
# numactl --hardware
available: 3 nodes (0-1,3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 3958 MB
node 0 free: 3708 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3871 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3971 MB
node distances:
node 0 1 3
0: 10 21 21
1: 21 10 21
3: 21 21 10
/*
* Put the pmem namespace into devdax mode so it can be assigned to the
* kmem driver
*/
# ndctl create-namespace -e namespace0.0 -m devdax -f
{
"dev":"namespace0.0",
"mode":"devdax",
"map":"dev",
"size":"3.94 GiB (4.23 GB)",
"uuid":"1650af9b-9ba3-4704-acd6-10178399d9a3",
[..]
}
/* Online Persistent Memory as System RAM */
# daxctl reconfigure-device --mode=system-ram dax0.0
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
libdaxctl: memblock_in_dev: dax0.0: memory0: Unable to determine phys_index: Success
[
{
"chardev":"dax0.0",
"size":4225761280,
"target_node":0,
"mode":"system-ram"
}
]
reconfigured 1 device
/* Note that the memory is onlined by default to the wrong node, node0 */
# numactl --hardware
available: 3 nodes (0-1,3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 7926 MB
node 0 free: 7655 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3871 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3971 MB
node distances:
node 0 1 3
0: 10 21 21
1: 21 10 21
3: 21 21 10
=== After ===
/* Notice that the "phys_index" error messages are gone */
# daxctl reconfigure-device --mode=system-ram dax0.0
[
{
"chardev":"dax0.0",
"size":4225761280,
"target_node":2,
"mode":"system-ram"
}
]
reconfigured 1 device
/* Notice that node2 is now correctly populated */
# numactl --hardware
available: 4 nodes (0-3)
node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
node 0 size: 3958 MB
node 0 free: 3793 MB
node 1 cpus: 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
node 1 size: 4027 MB
node 1 free: 3851 MB
node 2 cpus:
node 2 size: 3968 MB
node 2 free: 3968 MB
node 3 cpus:
node 3 size: 3994 MB
node 3 free: 3908 MB
node distances:
node 0 1 2 3
0: 10 21 21 21
1: 21 10 21 21
2: 21 21 10 21
3: 21 21 21 10
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Ira Weiny <ira.weiny@intel.com>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Link: https://lore.kernel.org/r/158188327614.894464.13122730362187722603.stgit@dwillia2-desk3.amr.corp.intel.com
2020-02-16 23:01:16 +03:00
ndr_desc . numa_node = numa_map_to_online_node ( nid ) ;
ndr_desc . target_node = nid ;
2018-06-02 21:43:39 +03:00
set_bit ( ND_REGION_PAGEMAP , & ndr_desc . flags ) ;
if ( ! nvdimm_pmem_region_create ( nvdimm_bus , & ndr_desc ) )
return - ENXIO ;
return 0 ;
}
2015-08-19 07:34:34 +03:00
static int e820_pmem_probe ( struct platform_device * pdev )
{
static struct nvdimm_bus_descriptor nd_desc ;
struct device * dev = & pdev - > dev ;
struct nvdimm_bus * nvdimm_bus ;
2018-06-02 21:43:39 +03:00
int rc = - ENXIO ;
2015-08-19 07:34:34 +03:00
nd_desc . provider_name = " e820 " ;
2016-07-22 06:03:19 +03:00
nd_desc . module = THIS_MODULE ;
2015-08-19 07:34:34 +03:00
nvdimm_bus = nvdimm_bus_register ( dev , & nd_desc ) ;
if ( ! nvdimm_bus )
goto err ;
platform_set_drvdata ( pdev , nvdimm_bus ) ;
2018-06-02 21:43:39 +03:00
rc = walk_iomem_res_desc ( IORES_DESC_PERSISTENT_MEMORY_LEGACY ,
IORESOURCE_MEM , 0 , - 1 , nvdimm_bus , e820_register_one ) ;
if ( rc )
goto err ;
2015-08-19 07:34:34 +03:00
return 0 ;
2018-06-02 21:43:39 +03:00
err :
2015-08-19 07:34:34 +03:00
nvdimm_bus_unregister ( nvdimm_bus ) ;
dev_err ( dev , " failed to register legacy persistent memory ranges \n " ) ;
2018-06-02 21:43:39 +03:00
return rc ;
2015-08-19 07:34:34 +03:00
}
static struct platform_driver e820_pmem_driver = {
. probe = e820_pmem_probe ,
. remove = e820_pmem_remove ,
. driver = {
. name = " e820_pmem " ,
} ,
} ;
2016-12-05 11:23:20 +03:00
module_platform_driver ( e820_pmem_driver ) ;
2015-08-19 07:34:34 +03:00
MODULE_ALIAS ( " platform:e820_pmem* " ) ;
MODULE_LICENSE ( " GPL v2 " ) ;
MODULE_AUTHOR ( " Intel Corporation " ) ;