[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] [RFC][PATCH 1/2] export NUMA topology from xen



For post-3.0.5 inclusion:

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               : 4
mem_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
                         node3:24-31

I've also reworked the the physinfo call to contain an array of
cpu_to_node elements rather than node_to_cpu to support machines larger
than 64-way.  I convert the array back to node_to_cpu for brevity in
xm info display.

-- 
Ryan Harper
Software Engineer; Linux Technology Center
IBM Corp., Austin, Tx
(512) 838-9253   T/L: 678-9253
ryanh@xxxxxxxxxx


diffstat output:
 b/xen/include/public/numa_structs.h                 |   27 ++++++
 tools/libxc/xc_misc.c                               |    4 
 tools/libxc/xenctrl.h                               |    3 
 tools/python/xen/lowlevel/xc/xc.c                   |   81 +++++++++++++++++---
 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                               |   47 +++++++++++
 xen/include/asm-x86/numa.h                          |    7 -
 xen/include/public/arch-x86/xen.h                   |    1 
 xen/include/public/sysctl.h                         |    3 
 13 files changed, 232 insertions(+), 21 deletions(-)

Signed-off-by: Ryan Harper <ryanh@xxxxxxxxxx>
---
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               : 4
mem_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
                         node3:24-31

I've also reworked the the physinfo call to contain an array of
cpu_to_node elements rather than node_to_cpu to support machines larger
than 64-way.  I convert the array back to node_to_cpu for brevity in
xm info display.

Signed-off-by: Ryan Harper <ryanh@xxxxxxxxxx>

diff -r 8f9ca49175ce tools/libxc/xc_misc.c
--- a/tools/libxc/xc_misc.c     Sat Mar 31 19:02:09 2007 +0100
+++ b/tools/libxc/xc_misc.c     Mon Apr 02 16:33:10 2007 -0500
@@ -59,6 +59,10 @@ 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.cpu_to_node = put_info->cpu_to_node;
 
     if ( (ret = do_sysctl(xc_handle, &sysctl)) != 0 )
         return ret;
diff -r 8f9ca49175ce tools/libxc/xenctrl.h
--- a/tools/libxc/xenctrl.h     Sat Mar 31 19:02:09 2007 +0100
+++ b/tools/libxc/xenctrl.h     Mon Apr 02 16:33:10 2007 -0500
@@ -28,6 +28,7 @@
 #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
@@ -473,6 +474,8 @@ int xc_send_debug_keys(int xc_handle, ch
 int xc_send_debug_keys(int xc_handle, char *keys);
 
 typedef xen_sysctl_physinfo_t xc_physinfo_t;
+typedef node_data_t xc_memory_chunk_t;
+typedef uint32_t xc_cpu_to_node_t;
 int xc_physinfo(int xc_handle,
                 xc_physinfo_t *info);
 
diff -r 8f9ca49175ce tools/python/xen/lowlevel/xc/xc.c
--- a/tools/python/xen/lowlevel/xc/xc.c Sat Mar 31 19:02:09 2007 +0100
+++ b/tools/python/xen/lowlevel/xc/xc.c Mon Apr 02 16:33:10 2007 -0500
@@ -644,10 +644,27 @@ static PyObject *pyxc_physinfo(XcObject 
 {
     xc_physinfo_t info;
     char cpu_cap[128], *p=cpu_cap, *q=cpu_cap;
-    int i;
+    int i,j, nr_cpus;
+    PyObject *ret_obj, *memchunk_obj, *node_to_cpu_obj;
+    xc_memory_chunk_t *chunks;
+    xc_cpu_to_node_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 cpu_to_node mapping */
+    map = (xc_cpu_to_node_t *)malloc( sizeof(xc_cpu_to_node_t) *
+                                      PUBLIC_MAX_CPUS ); 
+    set_xen_guest_handle(info.cpu_to_node, map);
     
     if ( xc_physinfo(self->xc_handle, &info) != 0 )
         return pyxc_error_to_exception();
+
+    /* calc number of cpus, ignore nr_nodes since sockets_per node is bogus */
+    nr_cpus = info.threads_per_core * info.cores_per_socket *
+              info.sockets_per_node;
 
     *q=0;
     for(i=0;i<sizeof(info.hw_cap)/4;i++)
@@ -659,16 +676,58 @@ 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);
+
+    /* make a list for each node */
+    for ( i=0; i<info.nr_nodes; i++)
+    {
+        PyObject *cpus = PyList_New(0);
+ 
+        /* walk the cpu_to_node array, for each cpu
+           which maps to node i, add to cpus list */
+        for ( j=0; j<nr_cpus; j++)
+        {
+            /* this cpu j maps to node i */
+            if ( i == (uint32_t)map[j])
+                PyList_Append(cpus, PyInt_FromLong(j));
+        }
+        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 8f9ca49175ce tools/python/xen/xend/XendNode.py
--- a/tools/python/xen/xend/XendNode.py Sat Mar 31 19:02:09 2007 +0100
+++ b/tools/python/xen/xend/XendNode.py Mon Apr 02 11:48:17 2007 -0500
@@ -534,6 +534,69 @@ 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()
 
@@ -546,6 +609,8 @@ class XendNode:
         # physinfo is in KiB, need it in MiB
         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',
@@ -556,6 +621,8 @@ class XendNode:
                       'hw_caps',
                       'total_memory',
                       'free_memory',
+                      'mem_chunks',
+                      'node_to_cpu'
                       ]
 
         return [[k, info[k]] for k in ITEM_ORDER]
diff -r 8f9ca49175ce tools/xenmon/xenbaked.c
--- a/tools/xenmon/xenbaked.c   Sat Mar 31 19:02:09 2007 +0100
+++ b/tools/xenmon/xenbaked.c   Mon Apr 02 16:33:10 2007 -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 8f9ca49175ce tools/xenstat/libxenstat/src/xenstat.c
--- a/tools/xenstat/libxenstat/src/xenstat.c    Sat Mar 31 19:02:09 2007 +0100
+++ b/tools/xenstat/libxenstat/src/xenstat.c    Mon Apr 02 16:33:10 2007 -0500
@@ -147,6 +147,9 @@ 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 8f9ca49175ce tools/xentrace/xentrace.c
--- a/tools/xentrace/xentrace.c Sat Mar 31 19:02:09 2007 +0100
+++ b/tools/xentrace/xentrace.c Mon Apr 02 16:33:10 2007 -0500
@@ -260,6 +260,9 @@ 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 8f9ca49175ce tools/xm-test/tests/info/02_info_compiledata_pos.py
--- a/tools/xm-test/tests/info/02_info_compiledata_pos.py       Sat Mar 31 
19:02:09 2007 +0100
+++ b/tools/xm-test/tests/info/02_info_compiledata_pos.py       Mon Apr 02 
11:48:17 2007 -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 8f9ca49175ce xen/arch/x86/sysctl.c
--- a/xen/arch/x86/sysctl.c     Sat Mar 31 19:02:09 2007 +0100
+++ b/xen/arch/x86/sysctl.c     Mon Apr 02 16:33:10 2007 -0500
@@ -23,6 +23,10 @@
 #include <asm/hvm/hvm.h>
 #include <asm/hvm/support.h>
 #include <asm/processor.h>
+#include <asm/numa.h>
+#include <xen/nodemask.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)
@@ -34,6 +38,10 @@ long arch_do_sysctl(
 
     case XEN_SYSCTL_physinfo:
     {
+        int i;
+        node_data_t *chunks;
+        uint32_t *map, cpu_to_node_map[NR_CPUS];
+
         xen_sysctl_physinfo_t *pi = &sysctl->u.physinfo;
 
         pi->threads_per_core =
@@ -43,7 +51,6 @@ 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();
@@ -51,6 +58,44 @@ 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 cpu_to_node pointer from guest */
+        get_xen_guest_handle(map, sysctl->u.physinfo.cpu_to_node);
+
+        /* if set, fill out cpu_to_node array */
+        if ( map != NULL )
+        {
+            /* copy cpu to node mapping to domU */
+            memset(cpu_to_node_map, 0, sizeof(cpu_to_node_map));
+            for ( i = 0; i < num_online_cpus(); i++)
+            {
+                cpu_to_node_map[i]=cpu_to_node(i);
+                if ( copy_to_guest_offset(sysctl->u.physinfo.cpu_to_node,
+                                          i, &(cpu_to_node_map[i]), 1) ) {
+                    ret = -EFAULT;
+                    break;
+                }
+            }
+        }
+
         if ( copy_to_guest(u_sysctl, sysctl, 1) )
             ret = -EFAULT;
     }
diff -r 8f9ca49175ce xen/include/asm-x86/numa.h
--- a/xen/include/asm-x86/numa.h        Sat Mar 31 19:02:09 2007 +0100
+++ b/xen/include/asm-x86/numa.h        Mon Apr 02 11:48:17 2007 -0500
@@ -2,6 +2,7 @@
 #define _ASM_X8664_NUMA_H 1
 
 #include <xen/cpumask.h>
+#include <public/numa_structs.h>
 
 #define NODES_SHIFT 6
 
@@ -44,12 +45,6 @@ extern int memnode_shift;
 extern int memnode_shift; 
 extern u8  memnodemap[NODEMAPSIZE]; 
 
-struct node_data {
-    unsigned long node_start_pfn;
-    unsigned long node_spanned_pages;
-    unsigned int  node_id;
-};
-
 extern struct node_data node_data[];
 
 static inline __attribute__((pure)) int phys_to_nid(unsigned long addr) 
diff -r 8f9ca49175ce xen/include/public/arch-x86/xen.h
--- a/xen/include/public/arch-x86/xen.h Sat Mar 31 19:02:09 2007 +0100
+++ b/xen/include/public/arch-x86/xen.h Mon Apr 02 11:48:17 2007 -0500
@@ -54,6 +54,7 @@ __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 8f9ca49175ce xen/include/public/numa_structs.h
--- /dev/null   Thu Jan 01 00:00:00 1970 +0000
+++ b/xen/include/public/numa_structs.h Mon Apr 02 11:48:17 2007 -0500
@@ -0,0 +1,27 @@
+/*
+ * Ryan Grimm  <grimm@xxxxxxxxxx>
+ * Ryan Harper <ryanh@xxxxxxxxxx>
+ * Copyright 2006, 2007 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
+#define PUBLIC_MAX_CPUS 256
+
+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
diff -r 8f9ca49175ce xen/include/public/sysctl.h
--- a/xen/include/public/sysctl.h       Sat Mar 31 19:02:09 2007 +0100
+++ b/xen/include/public/sysctl.h       Mon Apr 02 16:33:10 2007 -0500
@@ -33,6 +33,7 @@
 
 #include "xen.h"
 #include "domctl.h"
+#include "numa_structs.h"
 
 #define XEN_SYSCTL_INTERFACE_VERSION 0x00000003
 
@@ -85,6 +86,8 @@ struct xen_sysctl_physinfo {
     uint64_aligned_t free_pages;
     uint64_aligned_t scrub_pages;
     uint32_t hw_cap[8];
+    XEN_GUEST_HANDLE(node_data_t) memory_chunks;
+    XEN_GUEST_HANDLE(uint32_t) cpu_to_node;
 };
 typedef struct xen_sysctl_physinfo xen_sysctl_physinfo_t;
 DEFINE_XEN_GUEST_HANDLE(xen_sysctl_physinfo_t);

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.