xen: export NUMA topology in physinfo hcall
PATCH 4/6 xen: export NUMA topology in physinfo hcall
This patch modifies the physinfo hcall to export NUMA CPU and Memory topology information. The new physinfo hcall is integrated into libxc and xend (xm info specifically). Included in this patch is a minor tweak to xm-test's xm info testcase. The new fields in xm info are:
nr_nodes : 4mem_chunks : node0:0x0000000000000000-0x0000000190000000
node1:0x0000000190000000-0x0000000300000000
node2:0x0000000300000000-0x0000000470000000
node3:0x0000000470000000-0x0000000640000000
node_to_cpu : node0:0-7
node1:8-15
node2:16-23
node3:24-31Signature · 476 lines
Ryan Harper Software Engineer; Linux Technology Center IBM Corp., Austin, Tx (512) 838-9253 T/L: 678-9253 [email protected] diffstat output: b/xen/include/public/numa_structs.h | 26 ++++++ tools/libxc/xc_misc.c | 4 + tools/libxc/xenctrl.h | 3 tools/python/xen/lowlevel/xc/xc.c | 77 ++++++++++++++++---- tools/python/xen/xend/XendNode.py | 67 +++++++++++++++++ tools/xenmon/xenbaked.c | 3 tools/xenstat/libxenstat/src/xenstat.c | 3 tools/xentrace/xentrace.c | 3 tools/xm-test/tests/info/02_info_compiledata_pos.py | 4 - xen/arch/x86/sysctl.c | 51 ++++++++++++- xen/include/public/arch-x86_32.h | 1 xen/include/public/arch-x86_64.h | 1 xen/include/public/sysctl.h | 3 xen/include/xen/numa.h | 7 - 14 files changed, 231 insertions(+), 22 deletions(-) Signed-off-by: Ryan Harper <[email protected]> --- Export NUMA topology in physinfo hcall diff -r 6f3c6fb05af3 tools/libxc/xc_misc.c --- a/tools/libxc/xc_misc.c Mon Sep 11 13:45:23 2006 -0500 +++ b/tools/libxc/xc_misc.c Mon Sep 11 13:45:23 2006 -0500 @@ -39,6 +39,10 @@ int xc_physinfo(int xc_handle, DECLARE_SYSCTL; sysctl.cmd = XEN_SYSCTL_physinfo; + + /* set pointers to caller's so memcpy doesn't clobber them */ + sysctl.u.physinfo.memory_chunks = put_info->memory_chunks; + sysctl.u.physinfo.node_to_cpu = put_info->node_to_cpu; if ( (ret = do_sysctl(xc_handle, &sysctl)) != 0 ) return ret; diff -r 6f3c6fb05af3 tools/libxc/xenctrl.h --- a/tools/libxc/xenctrl.h Mon Sep 11 13:45:23 2006 -0500 +++ b/tools/libxc/xenctrl.h Sat Sep 16 08:17:10 2006 -0500 @@ -26,6 +26,7 @@ #include <xen/memory.h> #include <xen/acm.h> #include <xen/acm_ops.h> +#include <xen/numa_structs.h> #ifdef __ia64__ #define XC_PAGE_SHIFT 14 @@ -387,6 +388,8 @@ int xc_readconsolering(int xc_handle, int clear); typedef xen_sysctl_physinfo_t xc_physinfo_t; +typedef node_data_t xc_memory_chunk_t; +typedef uint64_t xc_node_to_cpu_t; int xc_physinfo(int xc_handle, xc_physinfo_t *info); diff -r 6f3c6fb05af3 tools/python/xen/lowlevel/xc/xc.c --- a/tools/python/xen/lowlevel/xc/xc.c Mon Sep 11 13:45:23 2006 -0500 +++ b/tools/python/xen/lowlevel/xc/xc.c Sat Sep 16 08:17:13 2006 -0500 @@ -474,8 +474,21 @@ static PyObject *pyxc_physinfo(XcObject { xc_physinfo_t info; char cpu_cap[128], *p=cpu_cap, *q=cpu_cap; - int i; - + int i,j; + PyObject *ret_obj, *memchunk_obj, *node_to_cpu_obj; + xc_memory_chunk_t *chunks; + xc_node_to_cpu_t *map; + + /* make space for mem chunks */ + chunks = (xc_memory_chunk_t *)malloc( sizeof(xc_memory_chunk_t) * + PUBLIC_MAXCHUNKS ); + set_xen_guest_handle(info.memory_chunks, chunks); + + /* make space for node_to_cpu mapping */ + map = (xc_node_to_cpu_t *)malloc( sizeof(xc_node_to_cpu_t) * + PUBLIC_MAX_NUMNODES ); + set_xen_guest_handle(info.node_to_cpu, map); + if ( xc_physinfo(self->xc_handle, &info) != 0 ) return PyErr_SetFromErrno(xc_error); @@ -489,16 +502,56 @@ static PyObject *pyxc_physinfo(XcObject if(q>cpu_cap) *(q-1)=0; - return Py_BuildValue("{s:i,s:i,s:i,s:i,s:l,s:l,s:l,s:i,s:s}", - "threads_per_core", info.threads_per_core, - "cores_per_socket", info.cores_per_socket, - "sockets_per_node", info.sockets_per_node, - "nr_nodes", info.nr_nodes, - "total_memory", pages_to_kib(info.total_pages), - "free_memory", pages_to_kib(info.free_pages), - "scrub_memory", pages_to_kib(info.scrub_pages), - "cpu_khz", info.cpu_khz, - "hw_caps", cpu_cap); + ret_obj = Py_BuildValue("{s:i,s:i,s:i,s:l,s:l,s:l,s:i,s:s}", + "threads_per_core", info.threads_per_core, + "cores_per_socket", info.cores_per_socket, + "sockets_per_node", info.sockets_per_node, + "total_memory", pages_to_kib(info.total_pages), + "free_memory", pages_to_kib(info.free_pages), + "scrub_memory", pages_to_kib(info.scrub_pages), + "cpu_khz", info.cpu_khz, + "hw_caps", cpu_cap); + /* memchunks */ + memchunk_obj = PyList_New(0); + + /* build list of each memchunk's attributes, converting pfn to paddr */ + for ( i=0; i<info.nr_nodes; i++ ) + { + PyList_Append(memchunk_obj, + Py_BuildValue("{s:i,s:K,s:K}", + "node" , chunks[i].node_id, + "start_paddr", chunks[i].node_start_pfn << XC_PAGE_SHIFT, + "end_paddr" , (chunks[i].node_start_pfn + + chunks[i].node_spanned_pages) << XC_PAGE_SHIFT )); + } + PyDict_SetItemString(ret_obj, "mem_chunks", memchunk_obj); + + /* node to cpu mappings */ + node_to_cpu_obj = PyList_New(0); + /* build list of node to cpu mappings */ + for ( i=0; i<info.nr_nodes; i++ ) + { + uint64_t cpumap = (uint64_t)map[i]; + PyObject *cpus = PyList_New(0); + + for ( j=0; cpumap != 0; j++ ) + { + if ( cpumap & 1 ) + PyList_Append(cpus, PyInt_FromLong(j)); + cpumap >>=1; + } + PyList_Append(node_to_cpu_obj, cpus); + } + /* add list of node to cpu mappings and nr_nodes to physinfo dictionary */ + PyDict_SetItemString(ret_obj, "node_to_cpu", node_to_cpu_obj); + PyDict_SetItemString(ret_obj, "nr_nodes", + Py_BuildValue("i", info.nr_nodes)); + + /* free malloc'd memory */ + free(chunks); + free(map); + + return ret_obj; } static PyObject *pyxc_xeninfo(XcObject *self) diff -r 6f3c6fb05af3 tools/python/xen/xend/XendNode.py --- a/tools/python/xen/xend/XendNode.py Mon Sep 11 13:45:23 2006 -0500 +++ b/tools/python/xen/xend/XendNode.py Sat Sep 16 08:17:13 2006 -0500 @@ -51,6 +51,69 @@ class XendNode: ['version', ver], ['machine', mch]] + def list_to_rangepairs(self,cmap): + cmap.sort() + pairs = [] + x = y = 0 + for i in range(0,len(cmap)): + try: + if ((cmap[y+1] - cmap[i]) > 1): + pairs.append((cmap[x],cmap[y])) + x = y = i+1 + else: + y = y + 1 + # if we go off the end, then just add x to y + except IndexError: + pairs.append((cmap[x],cmap[y])) + + return pairs + + def format_pairs(self,pairs): + if not pairs: + return "no cpus" + out = "" + for f,s in pairs: + if (f==s): + out += '%d'%f + else: + out += '%d-%d'%(f,s) + out += ',' + # trim trailing ',' + return out[:-1] + + def list_to_strrange(self,list): + return self.format_pairs(self.list_to_rangepairs(list)) + + def format_memchunks(self, pinfo): + str='' + whitespace='' + try: + chunk=pinfo['mem_chunks'] + for i in range(0, pinfo['nr_nodes']): + str+='%snode%d:0x%016x-0x%016x\n' % (whitespace, + chunk[i]['node'], + chunk[i]['start_paddr'], + chunk[i]['end_paddr']) + whitespace='%25s' % '' + except: + str='none\n' + return str[:-1] + + def format_node_to_cpu(self, pinfo): + str='' + whitespace='' + try: + node_to_cpu=pinfo['node_to_cpu'] + for i in range(0, pinfo['nr_nodes']): + str+='%snode%d:%s\n' % (whitespace, + i, + self.list_to_strrange(node_to_cpu[i])) + whitespace='%25s' % '' + except: + str='none\n' + return str[:-1]; + + def physinfo(self): info = self.xc.physinfo() @@ -62,6 +125,8 @@ class XendNode: # physinfo is in KiB info['total_memory'] = info['total_memory'] / 1024 info['free_memory'] = info['free_memory'] / 1024 + info['mem_chunks'] = self.format_memchunks(info) + info['node_to_cpu'] = self.format_node_to_cpu(info) ITEM_ORDER = ['nr_cpus', 'nr_nodes', @@ -72,6 +137,8 @@ class XendNode: 'hw_caps', 'total_memory', 'free_memory', + 'mem_chunks', + 'node_to_cpu' ] return [[k, info[k]] for k in ITEM_ORDER] diff -r 6f3c6fb05af3 tools/xenmon/xenbaked.c --- a/tools/xenmon/xenbaked.c Mon Sep 11 13:45:23 2006 -0500 +++ b/tools/xenmon/xenbaked.c Sat Sep 16 08:17:13 2006 -0500 @@ -448,6 +448,9 @@ unsigned int get_num_cpus(void) int xc_handle = xc_interface_open(); int ret; + /* ensure memory_chunks and node_to_cpu are NULL */ + memset(&physinfo, 0, sizeof(physinfo)); + ret = xc_physinfo(xc_handle, &physinfo); if ( ret != 0 ) diff -r 6f3c6fb05af3 tools/xenstat/libxenstat/src/xenstat.c --- a/tools/xenstat/libxenstat/src/xenstat.c Mon Sep 11 13:45:23 2006 -0500 +++ b/tools/xenstat/libxenstat/src/xenstat.c Sat Sep 16 08:17:13 2006 -0500 @@ -222,6 +222,9 @@ xenstat_node *xenstat_get_node(xenstat_h /* Store the handle in the node for later access */ node->handle = handle; + + /* ensure memory_chunks and node_to_cpu are NULL */ + memset(&physinfo, 0, sizeof(physinfo)); /* Get information about the physical system */ if (xc_physinfo(handle->xc_handle, &physinfo) < 0) { diff -r 6f3c6fb05af3 tools/xentrace/xentrace.c --- a/tools/xentrace/xentrace.c Mon Sep 11 13:45:23 2006 -0500 +++ b/tools/xentrace/xentrace.c Sat Sep 16 08:17:13 2006 -0500 @@ -259,6 +259,9 @@ unsigned int get_num_cpus(void) int xc_handle = xc_interface_open(); int ret; + /* ensure memory_chunks and node_to_cpu are NULL */ + memset(&physinfo, 0, sizeof(physinfo)); + ret = xc_physinfo(xc_handle, &physinfo); if ( ret != 0 ) diff -r 6f3c6fb05af3 tools/xm-test/tests/info/02_info_compiledata_pos.py --- a/tools/xm-test/tests/info/02_info_compiledata_pos.py Mon Sep 11 13:45:23 2006 -0500 +++ b/tools/xm-test/tests/info/02_info_compiledata_pos.py Mon Sep 11 13:45:23 2006 -0500 @@ -18,9 +18,7 @@ for line in lines: for line in lines: pieces = line.split(" : ", 1) - if len(pieces) < 2: - FAIL("Found invalid line: [%s]" % line) - else: + if len(pieces) > 1: map[pieces[0]] = pieces[1] for field in ["cores_per_socket", "threads_per_core", "cpu_mhz", diff -r 6f3c6fb05af3 xen/arch/x86/sysctl.c --- a/xen/arch/x86/sysctl.c Mon Sep 11 13:45:23 2006 -0500 +++ b/xen/arch/x86/sysctl.c Sat Sep 16 08:17:13 2006 -0500 @@ -24,6 +24,10 @@ #include <asm/hvm/hvm.h> #include <asm/hvm/support.h> #include <asm/processor.h> +#include <asm/numa.h> +#include <asm/topology.h> + +#define get_xen_guest_handle(val, hnd) do { val = (hnd).p; } while (0) long arch_do_sysctl( struct xen_sysctl *sysctl, XEN_GUEST_HANDLE(xen_sysctl_t) u_sysctl) @@ -35,6 +39,10 @@ long arch_do_sysctl( case XEN_SYSCTL_physinfo: { + int i,j; + node_data_t *chunks; + u64 *map, node_to_cpu_64[MAX_NUMNODES]; + xen_sysctl_physinfo_t *pi = &sysctl->u.physinfo; pi->threads_per_core = @@ -44,7 +52,6 @@ long arch_do_sysctl( pi->sockets_per_node = num_online_cpus() / cpus_weight(cpu_core_map[0]); - pi->nr_nodes = 1; pi->total_pages = total_pages; pi->free_pages = avail_domheap_pages(); pi->scrub_pages = avail_scrub_pages(); @@ -52,6 +59,48 @@ long arch_do_sysctl( memset(pi->hw_cap, 0, sizeof(pi->hw_cap)); memcpy(pi->hw_cap, boot_cpu_data.x86_capability, NCAPINTS*4); ret = 0; + + /* fetch memory_chunk pointer from guest*/ + get_xen_guest_handle(chunks, sysctl->u.physinfo.memory_chunks); + + /* if it is set, fill out memory chunk array */ + if ( chunks != NULL ) + for_each_online_node(i) + { + /* copy memory chunk structs to guest */ + if ( copy_to_guest_offset(sysctl->u.physinfo.memory_chunks, i, + &(node_data[i]), 1) ) { + ret = -EFAULT; + break; + } + } + + /* set number of notes */ + pi->nr_nodes = num_online_nodes(); + + /* fetch node_to_cpu pointer from guest */ + get_xen_guest_handle(map, sysctl->u.physinfo.node_to_cpu); + + /* if set, fill out node_to_cpu array */ + if ( map != NULL ) + { + /* copy cpu to node mapping to domU */ + /* converting cpumask to u64 b/c userspace doesn't + * know about cpumask_t and is accepting a u64 */ + memset(node_to_cpu_64, 0, sizeof(node_to_cpu_64)); + for ( i = 0; i < pi->nr_nodes; i++ ) { + for ( j = 0; j < num_online_cpus(); j++ ) + if ( cpu_isset(j, node_to_cpumask(i)) ) + node_to_cpu_64[i] |= (u64)1 << j; + + if ( copy_to_guest_offset(sysctl->u.physinfo.node_to_cpu, + i, &(node_to_cpu_64[i]), 1) ) { + ret = -EFAULT; + break; + } + } + } + if ( copy_to_guest(u_sysctl, sysctl, 1) ) ret = -EFAULT; } diff -r 6f3c6fb05af3 xen/include/public/arch-x86_32.h --- a/xen/include/public/arch-x86_32.h Mon Sep 11 13:45:23 2006 -0500 +++ b/xen/include/public/arch-x86_32.h Mon Sep 11 13:45:23 2006 -0500 @@ -48,6 +48,7 @@ __DEFINE_XEN_GUEST_HANDLE(uchar, unsigne __DEFINE_XEN_GUEST_HANDLE(uchar, unsigned char); __DEFINE_XEN_GUEST_HANDLE(uint, unsigned int); __DEFINE_XEN_GUEST_HANDLE(ulong, unsigned long); +__DEFINE_XEN_GUEST_HANDLE(u64, uint64_t); DEFINE_XEN_GUEST_HANDLE(char); DEFINE_XEN_GUEST_HANDLE(int); DEFINE_XEN_GUEST_HANDLE(long); diff -r 6f3c6fb05af3 xen/include/public/arch-x86_64.h --- a/xen/include/public/arch-x86_64.h Mon Sep 11 13:45:23 2006 -0500 +++ b/xen/include/public/arch-x86_64.h Mon Sep 11 13:45:23 2006 -0500 @@ -49,6 +49,7 @@ __DEFINE_XEN_GUEST_HANDLE(uchar, unsigne __DEFINE_XEN_GUEST_HANDLE(uchar, unsigned char); __DEFINE_XEN_GUEST_HANDLE(uint, unsigned int); __DEFINE_XEN_GUEST_HANDLE(ulong, unsigned long); +__DEFINE_XEN_GUEST_HANDLE(u64, uint64_t); DEFINE_XEN_GUEST_HANDLE(char); DEFINE_XEN_GUEST_HANDLE(int); DEFINE_XEN_GUEST_HANDLE(long); diff -r 6f3c6fb05af3 xen/include/public/sysctl.h --- a/xen/include/public/sysctl.h Mon Sep 11 13:45:23 2006 -0500 +++ b/xen/include/public/sysctl.h Sat Sep 16 08:17:13 2006 -0500 @@ -15,6 +15,7 @@ #include "xen.h" #include "domctl.h" +#include "numa_structs.h" #define XEN_SYSCTL_INTERFACE_VERSION 0x00000002 @@ -67,6 +68,8 @@ struct xen_sysctl_physinfo { uint64_t free_pages; uint64_t scrub_pages; uint32_t hw_cap[8]; + XEN_GUEST_HANDLE(node_data_t) memory_chunks; + XEN_GUEST_HANDLE(u64) node_to_cpu; }; typedef struct xen_sysctl_physinfo xen_sysctl_physinfo_t; DEFINE_XEN_GUEST_HANDLE(xen_sysctl_physinfo_t); diff -r 6f3c6fb05af3 xen/include/xen/numa.h --- a/xen/include/xen/numa.h Mon Sep 11 13:45:23 2006 -0500 +++ b/xen/include/xen/numa.h Mon Sep 11 13:45:23 2006 -0500 @@ -2,6 +2,7 @@ #define _XEN_NUMA_H #include <xen/config.h> +#include <public/numa_structs.h> #ifdef CONFIG_DISCONTIGMEM #include <asm/numnodes.h> @@ -26,10 +27,4 @@ extern unsigned int cpu_to_node[]; #include <xen/cpumask.h> extern cpumask_t node_to_cpumask[]; -typedef struct node_data { - unsigned long node_start_pfn; - unsigned long node_spanned_pages; - unsigned int node_id; -} node_data_t; - #endif /* _XEN_NUMA_H */ diff -r 6f3c6fb05af3 xen/include/public/numa_structs.h --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/xen/include/public/numa_structs.h Mon Sep 11 13:45:23 2006 -0500 @@ -0,0 +1,26 @@ +/* + * Ryan Grimm <[email protected]> + * Ryan Harper <[email protected]> + * Copyright (c) 2006, International Business Machines Corporation. + * + */ + +#ifndef __XEN_PUBLIC_NUMA_STRUCTS_H__ + +#define __XEN_PUBLIC_NUMA_STRUCTS_H__ + +#include "xen.h" + +/* define these for xc to use b/c MAX_NUMNODES and MAX_CHUNKS + * are not exposed in /public */ +#define PUBLIC_MAX_NUMNODES 16 +#define PUBLIC_MAXCHUNKS 32 + +typedef struct node_data { + unsigned long node_start_pfn; + unsigned long node_spanned_pages; + unsigned int node_id; +} node_data_t; +DEFINE_XEN_GUEST_HANDLE(node_data_t); + +#endif
Quoted 7 lines · Ryan Harper
This patch modifies the physinfo hcall to export NUMA CPU and Memory topology information. The new physinfo hcall is integrated into libxc and xend (xm info specifically). Included in this patch is a minor tweak to xm-test's xm info testcase. The new fields in xm info are:
nr_nodes: 4mem_chunks: node0:0x0000000000000000-0x0000000190000000
Quoted 7 lines
node1:0x0000000190000000-0x0000000300000000
node2:0x0000000300000000-0x0000000470000000
node3:0x0000000470000000-0x0000000640000000
node_to_cpu : node0:0-7
node1:8-15
node2:16-23
node3:24-31Hi,
I have successfully applied this patch on xen-ia64-unstable. It requires a small patch to fix issues.
I have tested it on a 4 node, 24 cpus system.
I have two suggestions for physinfo hcall:
* We (Bull) already sell machines with more than 64 cpus (up to 128). Unfortuantly the physinfo interface works with at most 64 cpus. May I suggest to replace the node_cpu_to maps with a cpu_to_node map ?
* On ia64 memory can be sparsly populated. There is no real relation between number of nodes and number of memory chunks. May I suggest to add a new field (nr_mem_chunks) in physinfo ? It should be a read/written field: it should return the number of mem chunks at ouput (which can be greather than the input value if the buffer was too small).
Comments ?
Tristan.
Quoted 15 lines · Tristan Gingold
Quoted 7 lines · Ryan Harper
This patch modifies the physinfo hcall to export NUMA CPU and Memory topology information. The new physinfo hcall is integrated into libxc and xend (xm info specifically). Included in this patch is a minor tweak to xm-test's xm info testcase. The new fields in xm info are:
nr_nodes: 4mem_chunks: node0:0x0000000000000000-0x0000000190000000Quoted 7 lines
node1:0x0000000190000000-0x0000000300000000 node2:0x0000000300000000-0x0000000470000000 node3:0x0000000470000000-0x0000000640000000 node_to_cpu : node0:0-7 node1:8-15 node2:16-23 node3:24-31
Quoted 4 lines
Hi,
I have successfully applied this patch on xen-ia64-unstable. It requires a small patch to fix issues.
Thanks for giving the patches a test.
Quoted 6 lines
I have tested it on a 4 node, 24 cpus system.
I have two suggestions for physinfo hcall:
* We (Bull) already sell machines with more than 64 cpus (up to 128). Unfortuantly the physinfo interface works with at most 64 cpus. May I suggest to replace thenode_cpu_tomaps with acpu_to_nodemap ?
That is fine. It shouldn't be too much trouble to pass up an array of cpu_to_node and convert to node_to_cpu (I like the brevity of the above display; based on number of nodes rather than number of cpus). Does that sound reasonable?
Quoted 6 lines
* On ia64 memory can be sparsly populated. There is no real relation between number of nodes and number of memory chunks. May I suggest to add a new field (
nr_mem_chunks) in physinfo ? It should be a read/written field: it should return the number of mem chunks at ouput (which can be greather than the input value if the buffer was too small).
Even if it sparsely populated, won't each of the chunks "belong" to a particular node? The above list of 4 entries is not hard-coded, but a result of the behavior of the srat table memory affinity parsing.
The current srat code from Linux x86_64 (specifically, acpi_numa_memory_affinity_init(), merges each memory entry from the srat table based on the entries proximity value (a.k.a node number).
It will grow the node's memory range either down, or up if the new entry's start or end is outside the nodes current range:
if (!node_test_and_set(node, nodes_parsed)) {
nd->start = start;
nd->end = end;
} else {
if (start < nd->start)
nd->start = start;
if (nd->end < end)
nd->end = end;}
The end result will be a mapping of any number of memory chunks to the number of nodes in the system as each chunk must belong to one node.
One of the goal for the NUMA patches was to not re-invent this parsing and data structures all over, but to reuse what is available in Linux. It may be that the x86_64 srat table parsing in Linux differs from ia64 in Linux. Is there something that needs fixing here?
Signature · 5 lines
Ryan Harper Software Engineer; Linux Technology Center IBM Corp., Austin, Tx (512) 838-9253 T/L: 678-9253 [email protected]
Quoted 31 lines · Ryan Harper
Quoted 16 lines · Tristan Gingold
Quoted 7 lines · Ryan Harper
This patch modifies the physinfo hcall to export NUMA CPU and Memory topology information. The new physinfo hcall is integrated into libxc and xend (xm info specifically). Included in this patch is a minor tweak to xm-test's xm info testcase. The new fields in xm info are:
nr_nodes: 4mem_chunks: node0:0x0000000000000000-0x0000000190000000Quoted 7 lines
node1:0x0000000190000000-0x0000000300000000 node2:0x0000000300000000-0x0000000470000000 node3:0x0000000470000000-0x0000000640000000 node_to_cpu : node0:0-7 node1:8-15 node2:16-23 node3:24-31Quoted 4 lines
Hi,
I have successfully applied this patch on xen-ia64-unstable. It requires a small patch to fix issues.
Thanks for giving the patches a test.
Quoted 6 lines
I have tested it on a 4 node, 24 cpus system.
I have two suggestions for physinfo hcall:
* We (Bull) already sell machines with more than 64 cpus (up to 128). Unfortuantly the physinfo interface works with at most 64 cpus. May I suggest to replace thenode_cpu_tomaps with acpu_to_nodemap ?
Quoted 4 lines
That is fine. It shouldn't be too much trouble to pass up an array of
cpu_to_nodeand convert tonode_to_cpu(I like the brevity of the above display; based on number of nodes rather than number of cpus). Does that sound reasonable?
I like the current display, and yes it sounds reasonable.
Quoted 28 lines
Quoted 5 lines
* On ia64 memory can be sparsly populated. There is no real relation between number of nodes and number of memory chunks. May I suggest to add a new field (
nr_mem_chunks) in physinfo ? It should be a read/written field: it should return the number of mem chunks at ouput (which can be greather than the input value if the buffer was too small).Even if it sparsely populated, won't each of the chunks "belong" to a particular node? The above list of 4 entries is not hard-coded, but a result of the behavior of the srat table memory affinity parsing.
The current srat code from Linux
x86_64(specifically,acpi_numa_memory_affinity_init(), merges each memory entry from the srat table based on the entries proximity value (a.k.a node number).It will grow the node's memory range either down, or up if the new entry's start or end is outside the nodes current range:
if (!node_test_and_set(node, nodes_parsed)) { nd->start = start; nd->end = end; } else { if (start < nd->start) nd->start = start; if (nd->end < end) nd->end = end;}
Quoted 7 lines
The end result will be a mapping of any number of memory chunks to the number of nodes in the system as each chunk must belong to one node.
One of the goal for the NUMA patches was to not re-invent this parsing and data structures all over, but to reuse what is available in Linux. It may be that the
x86_64srat table parsing in Linux differs from ia64 in Linux. Is there something that needs fixing here?
On ia64 we have reused ia64 code from linux. Therefore we don't share all the srat parsing code.
I know on my 4 nodes system there are 5 srat entries. I have to check if the entries can be merged.
Stay tune!
Tristan.