From: beth kon <eak@us.ibm.com>
To: Xen-devel@lists.xensource.com
Cc: xen-ia64-devel@lists.xensource.com, Keir.Frazer@cl.cam.ac.uk,
xen-ppc-devel@lists.xensource.com
Subject: Re: [PATCH] nr_cpus calculation problem due to incorrect sockets_per_node
Date: Mon, 29 Oct 2007 11:37:38 -0400 [thread overview]
Message-ID: <4725FE42.3000509@us.ibm.com> (raw)
In-Reply-To: <4718C36F.5090607@us.ibm.com>
Hi. Wondering if this patch has been reviewed and could be considered
for inclusion in 3.2. Sorry about the late request. You were asking for
input last week.
beth kon wrote:
> Testing on an 8-node 128-way NUMA machine has exposed a problem with
> Xen's nr_cpus calculation. In this case, since Xen cuts off recognized
> CPUs at 32, the machine appears to have 16 CPUs on the first and
> second nodes and none on the remaining nodes. Given this asymmetry,
> the calculation of sockets_per_node (which is later used to calculate
> nr_cpus) is incorrect:
>
> pi->sockets_per_node = num_online_cpus() /(num_online_nodes() *
> pi->cores_per_socket * pi->threads_per_core);
>
> The most straightforward solution is to remove sockets_per_node, and
> instead determine nr_cpus directly from num_online_cpus.
>
> This patch has been tested on x86_64 NUMA machines.
>
>------------------------------------------------------------------------
>
>diff -r b4278beaf354 docs/man/xm.pod.1
>--- a/docs/man/xm.pod.1 Wed Oct 17 13:12:03 2007 +0100
>+++ b/docs/man/xm.pod.1 Wed Oct 17 20:09:46 2007 -0700
>@@ -446,7 +446,6 @@ page more readable):
> machine : i686
> nr_cpus : 2
> nr_nodes : 1
>- sockets_per_node : 2
> cores_per_socket : 1
> threads_per_core : 1
> cpu_mhz : 696
>diff -r b4278beaf354 tools/python/xen/lowlevel/xc/xc.c
>--- a/tools/python/xen/lowlevel/xc/xc.c Wed Oct 17 13:12:03 2007 +0100
>+++ b/tools/python/xen/lowlevel/xc/xc.c Wed Oct 17 20:09:46 2007 -0700
>@@ -721,7 +721,7 @@ static PyObject *pyxc_physinfo(XcObject
> "max_cpu_id", info.max_cpu_id,
> "threads_per_core", info.threads_per_core,
> "cores_per_socket", info.cores_per_socket,
>- "sockets_per_node", info.sockets_per_node,
>+ "nr_cpus", info.nr_cpus,
> "total_memory", pages_to_kib(info.total_pages),
> "free_memory", pages_to_kib(info.free_pages),
> "scrub_memory", pages_to_kib(info.scrub_pages),
>diff -r b4278beaf354 tools/python/xen/xend/XendNode.py
>--- a/tools/python/xen/xend/XendNode.py Wed Oct 17 13:12:03 2007 +0100
>+++ b/tools/python/xen/xend/XendNode.py Wed Oct 17 20:09:46 2007 -0700
>@@ -475,7 +475,7 @@ class XendNode:
>
> cpu_info = {
> "nr_nodes": phys_info["nr_nodes"],
>- "sockets_per_node": phys_info["sockets_per_node"],
>+ "nr_cpus": phys_info["nr_cpus"],
> "cores_per_socket": phys_info["cores_per_socket"],
> "threads_per_core": phys_info["threads_per_core"]
> }
>@@ -580,17 +580,9 @@ class XendNode:
> str='none\n'
> return str[:-1];
>
>- def count_cpus(self, pinfo):
>- count=0
>- node_to_cpu=pinfo['node_to_cpu']
>- for i in range(0, pinfo['nr_nodes']):
>- count+=len(node_to_cpu[i])
>- return count;
>-
> def physinfo(self):
> info = self.xc.physinfo()
>
>- info['nr_cpus'] = self.count_cpus(info)
> info['cpu_mhz'] = info['cpu_khz'] / 1000
>
> # physinfo is in KiB, need it in MiB
>@@ -600,7 +592,6 @@ class XendNode:
>
> ITEM_ORDER = ['nr_cpus',
> 'nr_nodes',
>- 'sockets_per_node',
> 'cores_per_socket',
> 'threads_per_core',
> 'cpu_mhz',
>diff -r b4278beaf354 tools/python/xen/xm/main.py
>--- a/tools/python/xen/xm/main.py Wed Oct 17 13:12:03 2007 +0100
>+++ b/tools/python/xen/xm/main.py Wed Oct 17 20:09:46 2007 -0700
>@@ -1667,9 +1667,8 @@ def xm_info(args):
> "release": getVal(["software_version", "release"]),
> "version": getVal(["software_version", "version"]),
> "machine": getVal(["software_version", "machine"]),
>- "nr_cpus": len(getVal(["host_CPUs"], [])),
>+ "nr_cpus": getVal(["cpu_configuration", "nr_cpus"]),
> "nr_nodes": getVal(["cpu_configuration", "nr_nodes"]),
>- "sockets_per_node": getVal(["cpu_configuration", "sockets_per_node"]),
> "cores_per_socket": getVal(["cpu_configuration", "cores_per_socket"]),
> "threads_per_core": getVal(["cpu_configuration", "threads_per_core"]),
> "cpu_mhz": getCpuMhz(),
>diff -r b4278beaf354 tools/xenmon/xenbaked.c
>--- a/tools/xenmon/xenbaked.c Wed Oct 17 13:12:03 2007 +0100
>+++ b/tools/xenmon/xenbaked.c Wed Oct 17 20:09:46 2007 -0700
>@@ -460,10 +460,7 @@ unsigned int get_num_cpus(void)
> xc_interface_close(xc_handle);
> opts.cpu_freq = (double)physinfo.cpu_khz/1000.0;
>
>- return (physinfo.threads_per_core *
>- physinfo.cores_per_socket *
>- physinfo.sockets_per_node *
>- physinfo.nr_nodes);
>+ return physinfo.nr_cpus;
> }
>
>
>diff -r b4278beaf354 tools/xenstat/libxenstat/src/xenstat.c
>--- a/tools/xenstat/libxenstat/src/xenstat.c Wed Oct 17 13:12:03 2007 +0100
>+++ b/tools/xenstat/libxenstat/src/xenstat.c Wed Oct 17 20:09:46 2007 -0700
>@@ -155,9 +155,7 @@ xenstat_node *xenstat_get_node(xenstat_h
> }
>
> node->cpu_hz = ((unsigned long long)physinfo.cpu_khz) * 1000ULL;
>- node->num_cpus =
>- (physinfo.threads_per_core * physinfo.cores_per_socket *
>- physinfo.sockets_per_node * physinfo.nr_nodes);
>+ node->num_cpus = physinfo.nr_cpus;
> node->tot_mem = ((unsigned long long)physinfo.total_pages)
> * handle->page_size;
> node->free_mem = ((unsigned long long)physinfo.free_pages)
>diff -r b4278beaf354 tools/xentrace/xentrace.c
>--- a/tools/xentrace/xentrace.c Wed Oct 17 13:12:03 2007 +0100
>+++ b/tools/xentrace/xentrace.c Wed Oct 17 20:09:46 2007 -0700
>@@ -309,10 +309,7 @@ unsigned int get_num_cpus(void)
>
> xc_interface_close(xc_handle);
>
>- return (physinfo.threads_per_core *
>- physinfo.cores_per_socket *
>- physinfo.sockets_per_node *
>- physinfo.nr_nodes);
>+ return physinfo.nr_cpus;
> }
>
>
>diff -r b4278beaf354 tools/xm-test/lib/XmTestLib/Xm.py
>--- a/tools/xm-test/lib/XmTestLib/Xm.py Wed Oct 17 13:12:03 2007 +0100
>+++ b/tools/xm-test/lib/XmTestLib/Xm.py Wed Oct 17 20:09:46 2007 -0700
>@@ -218,11 +218,9 @@ def restartXend():
> return status
>
> def smpConcurrencyLevel():
>- cores = int(getInfo("cores_per_socket"))
>- threads = int(getInfo("threads_per_core"))
>- sockets = int(getInfo("sockets_per_node"))
>-
>- return cores * sockets * threads
>+ nr_cpus = int(getInfo("nr_cpus"))
>+
>+ return nr_cpus
>
> if __name__ == "__main__":
> if isDomainRunning("0"):
>diff -r b4278beaf354 tools/xm-test/lib/XmTestReport/OSReport.py
>--- a/tools/xm-test/lib/XmTestReport/OSReport.py Wed Oct 17 13:12:03 2007 +0100
>+++ b/tools/xm-test/lib/XmTestReport/OSReport.py Wed Oct 17 20:09:46 2007 -0700
>@@ -92,7 +92,6 @@ class Machine:
>
> xenValues = {"nr_cpus" : "Unknown",
> "nr_nodes" : "Unknown",
>- "sockets_per_node" : "Unknown",
> "cores_per_socket" : "Unknown",
> "threads_per_core" : "Unknown",
> "cpu_mhz" : "Unknown",
>diff -r b4278beaf354 xen/arch/ia64/xen/dom0_ops.c
>--- a/xen/arch/ia64/xen/dom0_ops.c Wed Oct 17 13:12:03 2007 +0100
>+++ b/xen/arch/ia64/xen/dom0_ops.c Wed Oct 17 20:09:46 2007 -0700
>@@ -234,7 +234,7 @@ long arch_do_sysctl(xen_sysctl_t *op, XE
> {
> case XEN_SYSCTL_physinfo:
> {
>- int i, node_cpus = 0;
>+ int i;
> uint32_t max_array_ent;
>
> xen_sysctl_physinfo_t *pi = &op->u.physinfo;
>@@ -242,18 +242,8 @@ long arch_do_sysctl(xen_sysctl_t *op, XE
> pi->threads_per_core = cpus_weight(cpu_sibling_map[0]);
> pi->cores_per_socket =
> cpus_weight(cpu_core_map[0]) / pi->threads_per_core;
>+ pi->nr_cpus = (u32)num_online_cpus();
> pi->nr_nodes = num_online_nodes();
>-
>- /*
>- * Guess at a sockets_per_node value. Use the maximum number of
>- * CPUs per node to avoid deconfigured CPUs breaking the average.
>- */
>- for_each_online_node(i)
>- node_cpus = max(node_cpus, cpus_weight(node_to_cpumask(i)));
>-
>- pi->sockets_per_node = node_cpus /
>- (pi->cores_per_socket * pi->threads_per_core);
>-
> pi->total_pages = total_pages;
> pi->free_pages = avail_domheap_pages();
> pi->scrub_pages = avail_scrub_pages();
>diff -r b4278beaf354 xen/arch/powerpc/sysctl.c
>--- a/xen/arch/powerpc/sysctl.c Wed Oct 17 13:12:03 2007 +0100
>+++ b/xen/arch/powerpc/sysctl.c Wed Oct 17 20:09:46 2007 -0700
>@@ -45,9 +45,7 @@ long arch_do_sysctl(struct xen_sysctl *s
> cpus_weight(cpu_sibling_map[0]);
> pi->cores_per_socket =
> cpus_weight(cpu_core_map[0]) / pi->threads_per_core;
>- pi->sockets_per_node = num_online_cpus() /
>- (num_online_nodes() * pi->cores_per_socket * pi->threads_per_core);
>-
>+ pi->nr_cpus = (u32)num_online_cpus();
> pi->nr_nodes = num_online_nodes();
> pi->total_pages = total_pages;
> pi->free_pages = avail_domheap_pages();
>diff -r b4278beaf354 xen/arch/x86/sysctl.c
>--- a/xen/arch/x86/sysctl.c Wed Oct 17 13:12:03 2007 +0100
>+++ b/xen/arch/x86/sysctl.c Wed Oct 17 20:09:46 2007 -0700
>@@ -51,10 +51,8 @@ long arch_do_sysctl(
> cpus_weight(cpu_sibling_map[0]);
> pi->cores_per_socket =
> cpus_weight(cpu_core_map[0]) / pi->threads_per_core;
>+ pi->nr_cpus = (u32)num_online_cpus();
> pi->nr_nodes = num_online_nodes();
>- pi->sockets_per_node = num_online_cpus() /
>- (pi->nr_nodes * pi->cores_per_socket * pi->threads_per_core);
>-
> pi->total_pages = total_pages;
> pi->free_pages = avail_domheap_pages();
> pi->scrub_pages = avail_scrub_pages();
>diff -r b4278beaf354 xen/include/public/sysctl.h
>--- a/xen/include/public/sysctl.h Wed Oct 17 13:12:03 2007 +0100
>+++ b/xen/include/public/sysctl.h Wed Oct 17 20:09:46 2007 -0700
>@@ -79,7 +79,7 @@ struct xen_sysctl_physinfo {
> /* IN variables. */
> uint32_t threads_per_core;
> uint32_t cores_per_socket;
>- uint32_t sockets_per_node;
>+ uint32_t nr_cpus;
> uint32_t nr_nodes;
> uint32_t cpu_khz;
> uint64_aligned_t total_pages;
>
>
>------------------------------------------------------------------------
>
>_______________________________________________
>Xen-devel mailing list
>Xen-devel@lists.xensource.com
>http://lists.xensource.com/xen-devel
>
>
--
Elizabeth Kon (Beth)
IBM Linux Technology Center
Open Hypervisor Team
email: eak@us.ibm.com
next prev parent reply other threads:[~2007-10-29 15:37 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2007-10-19 14:47 [PATCH] nr_cpus calculation problem due to incorrect sockets_per_node beth kon
2007-10-19 17:22 ` [PATCH] nr_cpus calculation problem due to incorrectsockets_per_node Puthiyaparambil, Aravindh
2007-10-19 18:27 ` beth kon
2007-10-29 15:37 ` beth kon [this message]
2007-10-29 15:46 ` [PATCH] nr_cpus calculation problem due to incorrect sockets_per_node Keir Fraser
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=4725FE42.3000509@us.ibm.com \
--to=eak@us.ibm.com \
--cc=Keir.Frazer@cl.cam.ac.uk \
--cc=Xen-devel@lists.xensource.com \
--cc=xen-ia64-devel@lists.xensource.com \
--cc=xen-ppc-devel@lists.xensource.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.