Date: Tue, 10 Mar 2015 10:07:19 -0600 From: Alan Somers <asomers@freebsd.org> To: Freddie Cash <fjwcash@gmail.com> Cc: "freebsd-hackers@freebsd.org" <freebsd-hackers@freebsd.org>, "lokadamus@gmx.de" <lokadamus@gmx.de>, "Pokala, Ravi" <rpokala@panasas.com>, Rui Paulo <rpaulo@me.com> Subject: Re: detecting hyperthreading Message-ID: <CAOtMX2i6MOyHKc08GgyMgsxWcCWUUCTd2qHSnY=9WaZhUm2xdA@mail.gmail.com> In-Reply-To: <CAOjFWZ594OKAZ95ybegKiu%2BKr4-YeEZfJGVmv1obhxeZjVVbNA@mail.gmail.com> References: <D12358C5.12F652%rpokala@panasas.com> <9F2E1411-B517-4BC8-AF61-BB15EE35083C@me.com> <D123957A.12F6CA%rpokala@panasas.com> <54FF1343.1020705@gmx.de> <D1246191.12F7D6%rpokala@panasas.com> <CAOjFWZ594OKAZ95ybegKiu%2BKr4-YeEZfJGVmv1obhxeZjVVbNA@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Mar 10, 2015 at 10:02 AM, Freddie Cash <fjwcash@gmail.com> wrote:
> On Tue, Mar 10, 2015 at 8:56 AM, Pokala, Ravi <rpokala@panasas.com> wrote:
>
>> -----Original Message-----
>> From: "lokadamus@gmx.de" <lokadamus@gmx.de>
>> Date: 2015-03-10, Tuesday at 08:52
>> To: Ravi Pokala <rpokala@panasas.com>, Rui Paulo <rpaulo@me.com>
>> Cc: "freebsd-hackers@freebsd.org" <freebsd-hackers@freebsd.org>
>> Subject: Re: detecting hyperthreading
>>
>> >Have you look at dmesg?
>> >My system is a P4 with HTT.
>> >dmesg |more
>> [...]
>> >CPU: Intel(R) Pentium(R) 4 CPU 3.00GHz (3000.00-MHz 686-class CPU)
>> > Origin = "GenuineIntel" Id = 0xf29 Family = 0xf Model = 0x2
>> >Stepping = 9
>> >
>> >Features=0xbfebfbff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE,M
>> >CA,CMOV,PAT,PSE36,CLFLUSH,DTS,ACPI,MMX,FXSR,SSE,SSE2,SS,HTT,TM,PBE>
>> > Features2=0x4400<CNXT-ID,xTPR>
>>
>> Of course. :-)
>>
>> But there are two problems:
>>
>> (1) That just tells me HTT is supported by the CPU, not that if kernel is
>> using it.
>> (2) It's difficult to parse.
>>
>> Of the two, (1) is the bigger concern for my use-case.
>>
>
> 6 lines or so below the Features line shows the kernel loading "cpu0
> (BSP)", and then "cpu1 (AP/HT)".
>
> Compare that to a system without HTT, where any extra cpus only show "(AP)".
>
> It's not perfect, but one could grep through /var/run/dmesg.boot looking
> for "cpu" lines and checking for "(AP)" or "(AP/HT)".
>
> --
> Freddie Cash
> fjwcash@gmail.com
I always look at "sysctl kern.sched.topology_spec" to tell if
hyperthreading is enabled. It's overkill, but it works. Here's some
Ruby code that can parse it:
require 'rexml/document'
# Parses the output of "sysctl kern.sched.topology_spec and returns a triple
# of [number of sockets, cores/socket, threads/core]
def _parse_topo(topology)
cpu_xpath = "groups/group/children/group/cpu"
xmldoc = REXML::Document.new(topology)
_punt_topo(topology) unless xmldoc.get_elements("groups/group").count == 1
sockets = xmldoc.get_elements("groups/group/children/group").count
all_cores_per_socket = Set.new([])
all_threads_per_core = Set.new([])
xmldoc.get_elements("groups/group/children/group").each do |group2|
cores = 0
children = group2.get_elements("children/group")
if children.empty?
# No hyperthreading
cpu_entities = group2.get_elements("cpu")
_punt_topo(topology) unless cpu_entities.count == 1
core_count = cpu_entities.first.attribute("count").value
cores += core_count.to_i
else
children.each do |group3|
if group3.get_elements("flags/flag[@name='SMT']").empty? && \
group3.get_elements("flags/flag[@name='HTT']").empty? && \
group3.get_elements("flags/flag[@name='THREAD']").empty?
else
# This cpu group represents a single hyperthreaded core
cores += 1
cpu_entities = group3.get_elements("cpu")
_punt_topo(topology) unless cpu_entities.count == 1
core_count = cpu_entities.first.attribute("count").value
all_threads_per_core.add(core_count.to_i)
end
end
end
all_cores_per_socket.add cores
end
if all_cores_per_socket.size > 1
cores_per_socket = "mixed"
else
cores_per_socket = all_cores_per_socket.first.to_i
end
if all_threads_per_core.size > 1
threads_per_core = "mixed"
elsif all_threads_per_core.size == 0
# No SMT sections means no hyperthreading
threads_per_core = 1
else
threads_per_core = all_threads_per_core.first.to_i
end
[sockets, cores_per_socket, threads_per_core]
end
# Helper method for _parse_topo
def _punt_topo(topo)
STDERR.puts topology
raise IOError.new "Output of kern.sched.topology is not understood"
end
-Alan
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAOtMX2i6MOyHKc08GgyMgsxWcCWUUCTd2qHSnY=9WaZhUm2xdA>